gcc/ChangeLog

   1 2018-04-10  Martin Liska  <mliska@suse.cz>
   2
   3         * gdbhooks.py: Add pretty-printers for varpool_node, symtab_node,
   4         cgraph_edge and ipa_ref.
   5
   6 2018-04-10  Jakub Jelinek  <jakub@redhat.com>
   7
   8         PR target/85177
   9         PR target/85255
  10         * config/i386/sse.md
  11         (<extract_type>_vinsert<shuffletype><extract_suf>_mask): Fix
  12         computation of the VEC_MERGE selector from mask.
  13         (<extract_type>_vinsert<shuffletype><extract_suf>_1<mask_name>):
  14         Fix decoding of the VEC_MERGE selector into mask.
  15
  16 2018-04-10  Richard Sandiford  <richard.sandiford@linaro.org>
  17
  18         PR tree-optimization/85286
  19         * tree-vect-data-refs.c (vect_get_smallest_scalar_type):
  20
  21 2018-04-10  Richard Sandiford  <richard.sandiford@linaro.org>
  22
  23         * final.c (final_1): Set insn_last_address as well as
  24         insn_current_address.
  25
  26 2018-04-10  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
  27
  28         PR target/85173
  29         * explow.c (emit_stack_probe): Call validize_mem on memory location
  30         before passing it to gen_probe_stack.  Create address operand and
  31         legitimize it for the probe_stack_address case.
  32
  33 2018-04-09  Jan Hubicka  <jh@suse.cz>
  34
  35         PR lto/85078
  36         * ipa-devirt.c (rebuild_type_inheritance-hash): New.
  37         * ipa-utils.h (rebuild_type_inheritance-hash): Declare.
  38         * tree.c (free_lang_data_in_type): Fix handling of binfos;
  39         walk basetypes.
  40         (free_lang_data): Rebuild type inheritance graph.
  41
  42 2018-04-09  Martin Sebor  <msebor@redhat.com>
  43
  44         * invoke.texi (-finline-small-functions): Mention other optimization
  45         options.
  46         (-findirect-inlining, -fpartial-inlining): Same.
  47         (-finline-functions-called-once): Same.
  48         (-freorder-blocks-and-partition): Same.
  49
  50 2018-04-09  Jan Hubicka  <jh@suse.cz>
  51
  52         PR rtl/84058
  53         * cfgcleanup.c (try_forward_edges): Do not give up on crossing
  54         jumps; choose last target that matches the criteria (i.e.
  55         no partition changes for non-crossing jumps).
  56         * cfgrtl.c (cfg_layout_redirect_edge_and_branch): Add basic
  57         support for redirecting crossing jumps to non-crossing.
  58
  59 2018-04-09  Alexey Brodkin  <abrodkin@synopsys.com>
  60
  61         * config/arc/arc.c (arc_expand_prologue): Set stack usage info
  62         also for naked functions.
  63
  64 2018-04-09  Claudiu Zissulescu  <claziss@synopsys.com>
  65
  66         * config/arc/arc.md (add_shift): New pattern.
  67         (add_shift2): Likewise.
  68         (sub_shift): Likewise.
  69         (sub_shift_cmp0_noout): Likewise.
  70         (compare_si_ashiftsi): Likewise.
  71         (xbfu_cmp0_noout): New combine pattern.
  72         (xbfu_cmp0"): Likewise.
  73         (movsi_set_cc_insn): Place the predicable variant first.
  74         (commutative_binary_cmp0_noout): Remove clobber.
  75         (commutative_binary_cmp0): New pattern.
  76         (noncommutative_binary_cmp0): Likewise.
  77         (noncommutative_binary_cmp0_noout): Likewise.
  78         (noncommutative_binary_comparison_result_used): Removed.
  79         (rsub_cmp0): New pattern.
  80         (rsub_cmp0_noout): Likewise.
  81         (extzvsi): Changed, keep only meaningful variants.
  82         (SQH, SEZ): New iterators.
  83         (SQH_postfix): New mode attribute.
  84         (SEZ_prefix): New code attribute.
  85         (<SEZ_prefix>xt<SQH_postfix>_cmp0_noout): New instruction pattern.
  86         (<SEZ_prefix>xt<SQH_postfix>_cmp0): Likewise.
  87         * config/arc/predicates.md (cc_set_register): Use CC_REG instead
  88         of numerical value.
  89         (noncommutative_operator): Check the availability of barrel
  90         shifter option.
  91
  92 2018-04-09  Richard Biener  <rguenther@suse.de>
  93
  94         PR tree-optimization/85284
  95         * tree-ssa-loop-niter.c (number_of_iterations_exit_assumptions):
  96         Only use the niter constraining form of simple_iv when the exit
  97         is always executed.
  98
  99 2018-04-09  Tom de Vries  <tom@codesourcery.com>
 100
 101         PR target/84041
 102         * config/nvptx/nvptx.md (define_c_enum "unspecv"): Add UNSPECV_MEMBAR.
 103         (define_expand "*memory_barrier"): New define_expand.
 104         (define_insn "memory_barrier"): New insn.
 105
 106 2018-04-09  Andrey Belevantsev  <abel@ispras.ru>
 107
 108         PR rtl-optimization/80463
 109         PR rtl-optimization/83972
 110         PR rtl-optimization/83480
 111
 112         * sel-sched-ir.c (has_dependence_note_mem_dep): Take into account the
 113         correct producer for the insn.
 114         (tidy_control_flow): Fixup seqnos in case of debug insns.
 115
 116 2018-04-09  Andrey Belevantsev  <abel@ispras.ru>
 117
 118         PR rtl-optimization/83913
 119
 120         * sel-sched-ir.c (merge_expr_data): Choose the middle between two
 121         different sched-times when merging exprs.
 122
 123 2018-04-09  Andrey Belevantsev  <abel@ispras.ru>
 124
 125         PR rtl-optimization/83962
 126
 127         * sel-sched-ir.c (tidy_control_flow): Correct the order in which we call
 128         tidy_fallthru_edge and tidy_control_flow.
 129
 130 2018-04-09  Andrey Belevantsev  <abel@ispras.ru>
 131
 132         PR rtl-optimization/83530
 133
 134         * sel-sched.c (force_next_insn): New global variable.
 135         (remove_insn_for_debug): When force_next_insn is true, also leave only
 136         next insn in the ready list.
 137         (sel_sched_region): When the region wasn't scheduled, make another pass
 138         over it with force_next_insn set to 1.
 139
 140 2018-04-08  Monk Chiang  <sh.chiang04@gmail.com>
 141
 142         * config.gcc (nds32le-*-*, nds32be-*-*): Add nds32/nds32_intrinsic.h
 143         into tm_file.
 144         * config/nds32/constants.md (unspec_volatile_element): Add enum values
 145         for interrupt control.
 146         * config/nds32/nds32-intrinsic.c: Implementation of intrinsic
 147         functions for interrupt control.
 148         * config/nds32/nds32-intrinsic.md: Likewise.
 149         * config/nds32/nds32_intrinsic.h: Likewise.
 150         * config/nds32/nds32.h (nds32_builtins): Likewise.
 151
 152 2018-04-08  Chung-Ju Wu  <jasonwucj@gmail.com>
 153
 154         * config/nds32/nds32.c (nds32_init_machine_status,
 155         nds32_legitimate_index_p, nds32_legitimate_address_p): Consider
 156         strict_aligned_p field.
 157         (nds32_expand_to_rtl_hook): New function.
 158         (TARGET_EXPAND_TO_RTL_HOOK): Define.
 159         * config/nds32/nds32.h (machine_function): Add strict_aligned_p field.
 160
 161 2018-04-08  Kito Cheng  <kito.cheng@gmail.com>
 162             Chung-Ju Wu  <jasonwucj@gmail.com>
 163
 164         * config.gcc (nds32*-*-*): Check that n7 is valid to --with-cpu.
 165         * config/nds32/nds32-n7.md: New file.
 166         * config/nds32/nds32-opts.h (nds32_cpu_type): Add CPU_N7.
 167         * config/nds32/nds32-pipelines-auxiliary.c: Implementation for n7
 168         pipeline.
 169         * config/nds32/nds32-protos.h: More declarations for n7 pipeline.
 170         * config/nds32/nds32.md (pipeline_model): Add n7.
 171         * config/nds32/nds32.opt (mcpu): Support n7 pipeline cpus.
 172         * config/nds32/pipelines.md: Include n7 settings.
 173
 174 2018-04-08  Kito Cheng  <kito.cheng@gmail.com>
 175             Chung-Ju Wu  <jasonwucj@gmail.com>
 176
 177         * config.gcc (nds32*-*-*): Check that e8 is valid to --with-cpu.
 178         * config/nds32/nds32-e8.md: New file.
 179         * config/nds32/nds32-opts.h (nds32-cpu_type): Add CPU_E8.
 180         * config/nds32/nds32-pipelines-auxiliary.c: Implementation for e8
 181         pipeline.
 182         * config/nds32/nds32-protos.h: More declarations for e8 pipeline.
 183         * config/nds32/nds32.md (pipeline_model): Add e8.
 184         * config/nds32/nds32.opt (mcpu): Support e8 pipeline cpus.
 185         * config/nds32/pipelines.md: Include e8 settings.
 186
 187 2018-04-08  Kito Cheng  <kito.cheng@gmail.com>
 188             Chung-Ju Wu  <jasonwucj@gmail.com>
 189
 190         * config.gcc (nds32*-*-*): Check that n6/n8/s8 are valid to --with-cpu.
 191         * config/nds32/nds32-n8.md: New file.
 192         * config/nds32/nds32-opts.h (nds32_cpu_type): Add CPU_N6 and CPU_N8.
 193         * config/nds32/nds32-pipelines-auxiliary.c: Implementation for n8
 194         pipeline.
 195         * config/nds32/nds32-protos.h: More declarations for n8 pipeline.
 196         * config/nds32/nds32-utils.c: More implementations for n8 pipeline.
 197         * config/nds32/nds32.md (pipeline_model): Add n8.
 198         * config/nds32/nds32.opt (mcpu): Support n8 pipeline cpus.
 199         * config/nds32/pipelines.md: Include n8 settings.
 200
 201 2018-04-08  Kito Cheng  <kito.cheng@gmail.com>
 202             Chung-Ju Wu  <jasonwucj@gmail.com>
 203
 204         * config.gcc (nds32*): Add nds32-utils.o into extra_objs.
 205         * config/nds32/nds32-n9-2r1w.md: New file.
 206         * config/nds32/nds32-n9-3r2w.md: New file.
 207         * config/nds32/nds32-opts.h (nds32_cpu_type, nds32_mul_type,
 208         nds32_register_ports): New or modify for cpu n9.
 209         * config/nds32/nds32-pipelines-auxiliary.c: Implementation for n9
 210         pipeline.
 211         * config/nds32/nds32-protos.h: More declarations for n9 pipeline.
 212         * config/nds32/nds32-utils.c: New file.
 213         * config/nds32/nds32.h (TARGET_PIPELINE_N9, TARGET_PIPELINE_SIMPLE,
 214         TARGET_MUL_SLOW): Define.
 215         * config/nds32/nds32.md (pipeline_model): New attribute.
 216         * config/nds32/nds32.opt (mcpu, mconfig-mul, mconfig-register-ports):
 217         New options that support cpu n9.
 218         * config/nds32/pipelines.md: Include n9 settings.
 219         * config/nds32/t-nds32 (nds32-utils.o): Add dependency.
 220
 221 2018-04-08  Chung-Ju Wu  <jasonwucj@gmail.com>
 222
 223         * config/nds32/nds32-md-auxiliary.c (output_cond_branch): Output align
 224         information if necessary.
 225         (output_cond_branch_compare_zero): Likewise.
 226         * config/nds32/nds32.c (nds32_adjust_insn_length): Consider align case.
 227         (nds32_target_alignment): Refine for alignment.
 228         * config/nds32/nds32.h (NDS32_ALIGN_P): Define.
 229         (FUNCTION_BOUNDARY): Modify.
 230         * config/nds32/nds32.md (call_internal, call_value_internal): Consider
 231         align case.
 232         * config/nds32/nds32.opt (malways-align, malign-functions): New.
 233
 234 2018-04-08  Monk Chiang  <sh.chiang04@gmail.com>
 235
 236         * config/nds32/constants.md (unspec_volatile_element): Add values for
 237         TLB operation and data prefetch.
 238         * config/nds32/nds32-intrinsic.c: Implementation of intrinsic
 239         functions for TLB operation and data prefetch.
 240         * config/nds32/nds32-intrinsic.md: Likewise.
 241         * config/nds32/nds32_intrinsic.h: Likewise.
 242         * config/nds32/nds32.c (nds32_dpref_names): Likewise.
 243         (nds32_print_operand): Likewise.
 244         * config/nds32/nds32.h (nds32_builtins): Likewise.
 245
 246 2018-04-07  Thomas Koenig  <tkoenig@gcc.gnu.org>
 247         Andrew Pinski <pinsika@gcc.gnu.org>
 248
 249         PR middle-end/82976
 250         * match.pd: Use constant_boolean_node of correct type instead of
 251         boolean_true_node or boolean_false_node for simplifying
 252         pointer comparisons to zero.
 253
 254 2018-04-07  Jakub Jelinek  <jakub@redhat.com>
 255
 256         PR tree-optimization/80021
 257         * tree.c (verify_type_variant): Make error call in verify_variant_match
 258         translatable and remove final full stop.
 259
 260 2018-04-07  Chung-Ju Wu  <jasonwucj@gmail.com>
 261
 262         * config/nds32/constants.md (unspec_volatile_element): Add
 263         UNSPEC_VOLATILE_EH_RETURN.
 264         * config/nds32/nds32-md-auxiliary.c (nds32_output_stack_push,
 265         nds32_output_stack_pop): Support dwarf exception handling process.
 266         * config/nds32/nds32-protos.h (nds32_dynamic_chain_address): Declare.
 267         * config/nds32/nds32.c (nds32_init_machine_status): Support dwarf
 268         exception handling process.
 269         (nds32_compute_stack_frame): Likewise.
 270         (nds32_return_addr_rtx): Likewise.
 271         (nds32_initial_elimination_offset): Likewise.
 272         (nds32_expand_prologue): Likewise.
 273         (nds32_expand_epilogue): Likewise.
 274         (nds32_dynamic_chain_address): New function.
 275         * config/nds32/nds32.h (machine_function): Add fields for dwarf
 276         exception handling.
 277         (DYNAMIC_CHAIN_ADDRESS): Define.
 278         (EH_RETURN_DATA_REGNO): Define.
 279         (EH_RETURN_STACKADJ_RTX): Define.
 280         * config/nds32/nds32.md (eh_return, nds32_eh_return): Implement
 281         patterns for dwarf exception handling.
 282
 283 2018-04-07  Chung-Ju Wu  <jasonwucj@gmail.com>
 284
 285         * config/nds32/nds32.h: Clean up obsolete macros.
 286
 287 2018-04-07  Monk Chiang  <sh.chiang04@gmail.com>
 288
 289         * config/nds32/constants.md (unspec_element, unspec_volatile_element):
 290         Add enum values for particular instructions.
 291         * config/nds32/nds32-intrinsic.c: Implementation of expanding
 292         particular intrinsic functions.
 293         * config/nds32/nds32-intrinsic.md: Likewise.
 294         * config/nds32/nds32_intrinsic.h: Likewise.
 295         * config/nds32/nds32.h (nds32_builtins): Likewise.
 296         * config/nds32/nds32.md (type): Add pbsad and pbsada.
 297         (btst, ave): New patterns for particular instructions.
 298
 299 2018-04-07  Monk Chiang  <sh.chiang04@gmail.com>
 300
 301         * config/nds32/constants.md (unspec_element, unspec_volatile_element):
 302         Add enum values for atomic load/store and memory sync.
 303         * config/nds32/nds32-intrinsic.c: Implementation for atomic load/store
 304         and memory sync.
 305         * config/nds32/nds32-intrinsic.md: Likewise.
 306         * config/nds32/nds32_intrinsic.h: Likewise.
 307         * config/nds32/nds32.h (nds32_builtins): Likewise.
 308
 309 2018-04-07  Jakub Jelinek  <jakub@redhat.com>
 310
 311         PR tree-optimization/85257
 312         * fold-const.c (native_encode_vector): If not all elts could fit
 313         and off is -1, return 0 rather than offset.
 314         * tree-ssa-sccvn.c (vn_reference_lookup_3): Pass
 315         (offseti - offset2) / BITS_PER_UNIT as 4th argument to
 316         native_encode_expr.  Verify len * BITS_PER_UNIT >= maxsizei.  Don't
 317         adjust buffer in native_interpret_expr call.
 318
 319 2018-04-07  Monk Chiang  <sh.chiang04@gmail.com>
 320
 321         * config/nds32/constants.md (unspec_volatile_element): Add cache
 322         control enum values.
 323         * config/nds32/nds32-intrinsic.c: Add cache control expand functions.
 324         * config/nds32/nds32-intrinsic.md: Add cache control patterns.
 325         * config/nds32/nds32.c (nds32_cctl_names): New.
 326         (nds32_print_operand): Handle cache control register names.
 327         * config/nds32/nds32.h (nds32_builtins): New enum values.
 328         * config/nds32/nds32_intrinsic.h: Add cache control enum types and
 329         macros.
 330         * config/nds32/nds32.md (type): Add mmu.
 331         * config/nds32/pipelines.md (simple_insn): Add mmu.
 332
 333 2018-04-07  Chung-Ju Wu  <jasonwucj@gmail.com>
 334
 335         * config/nds32/nds32.md (type): Remove call.
 336         * config/nds32/pipelines.md (simple_insn): Likewise.
 337
 338 2018-04-07  Monk Chiang  <sh.chiang04@gmail.com>
 339
 340         * config/nds32/constants.md (unspec_volatile_element): Add
 341         UNSPEC_VOLATILE_FMFCSR, UNSPEC_VOLATILE_FMTCSR and
 342         UNSPEC_VOLATILE_FMFCFG.
 343         * config/nds32/nds32-intrinsic.c (bdesc_noarg): New builtin
 344         description for fmfcfg and fmfcsr.
 345         (bdesc_1arg): Add fmtcsr.
 346         (bdesc_2arg): Add fcpynss, fcpyss, fcpynsd and fcpysd.
 347         (nds32_expand_builtin_impl): Deal with FPU intrinsic functions.
 348         * config/nds32/nds32-intrinsic.md (unspec_fcpynsd, unspec_fcpysd,
 349         unspec_fcpynss, unspec_fcpysd, unspec_fcpyss, unspec_fmfcsr,
 350         unspec_fmfcfg): New patterns.
 351         * config/nds32/nds32.h (nds32_builtins): Add NDS32_BUILTIN_FMFCFG,
 352         NDS32_BUILTIN_FMFCSR, NDS32_BUILTIN_FMTCSR, NDS32_BUILTIN_FCPYNSS,
 353         NDS32_BUILTIN_FCPYSS,NDS32_BUILTIN_FCPYNSD and NDS32_BUILTIN_FCPYSD.
 354         * config/nds32/nds32_intrinsic.h (__nds32__fcpynsd, __nds32__fcpynss,
 355         __nds32__fcpysd, __nds32__fcpyss, __nds32__fmfcsr, __nds32__fmtcsr,
 356         __nds32__fmfcfg): Define.
 357
 358 2018-04-07  Monk Chiang  <sh.chiang04@gmail.com>
 359
 360         * config/nds32/nds32.c (nds32_intrinsic_register_names): Add more
 361         intrinsic register names.
 362         * config/nds32/nds32_intrinsic.h (nds32_intrinsic_registers): Add more
 363         intrinsic register enum values and macros.
 364
 365 2018-04-07  Chung-Ju Wu  <jasonwucj@gmail.com>
 366
 367         * config/nds32/nds32.c (nds32_legitimate_index_p): Modify condition
 368         for load/store addressing form.
 369         (nds32_print_operand_address): Likewise.
 370
 371 2018-04-06  Eric Botcazou  <ebotcazou@adacore.com>
 372
 373         PR target/85196
 374         * config/sparc/sparc.c (sparc_expand_move): Deal with symbolic operands
 375         based on LABEL_REF.  Remove useless assertion.
 376         (pic_address_needs_scratch): Fix formatting.
 377         (sparc_legitimize_pic_address): Minor tweaks.
 378         (sparc_delegitimize_address): Adjust assertion accordingly.
 379         * config/sparc/sparc.md (movsi_pic_label_ref): Change label_ref_operand
 380         into symbolic_operand.
 381         (movsi_high_pic_label_ref): Likewise.
 382         (movsi_lo_sum_pic_label_ref): Likewise.
 383         (movdi_pic_label_ref): Likewise.
 384         (movdi_high_pic_label_ref): Likewise.
 385         (movdi_lo_sum_pic_label_ref): Likewise.
 386
 387 2018-04-06  Amaan Cheval  <amaan.cheval@gmail.com>
 388
 389         * config.gcc (x86_64-*-rtems*): Add rtems.h to tm_file for
 390         custom LIB_SPEC setup.
 391
 392 2018-04-06  Ruslan Bukin  <br@bsdpad.com>
 393             Kito Cheng  <kito.cheng@gmail.com>
 394
 395         * config.gcc (riscv*-*-freebsd*): Add RISC-V FreeBSD support.
 396         * config/riscv/freebsd.h: New.
 397
 398 2018-04-06  Chung-Ju Wu  <jasonwucj@gmail.com>
 399
 400         * config/nds32/nds32.c (nds32_adjust_insn_length): Refine.
 401         * config/nds32/nds32.h (ADJUST_INSN_LENGTH): Change the location in
 402         file.
 403
 404 2018-04-06  Chung-Ju Wu  <jasonwucj@gmail.com>
 405             Kito Cheng  <kito.cheng@gmail.com>
 406
 407         * config/nds32/nds32-md-auxiliary.c (nds32_output_return,
 408         nds32_output_call, nds32_symbol_binds_local_p): New functions.
 409         * config/nds32/nds32-protos.h (nds32_output_call,
 410         nds32_output_return): Declare.
 411         * config/nds32/nds32.md: Refine all the call and return patterns.
 412
 413 2018-04-06  Jakub Jelinek  <jakub@redhat.com>
 414
 415         PR debug/85252
 416         * dwarf2out.c (rtl_for_decl_init): For STRING_CST initializer only
 417         build CONST_STRING if TYPE_MAX_VALUE is non-NULL and is INTEGER_CST.
 418
 419         PR rtl-optimization/84872
 420         * cfgloopmanip.c (create_preheader): Use make_forwarder_block even if
 421         nentry == 1 when CP_FALLTHRU_PREHEADERS and single_entry is
 422         EDGE_CROSSING edge.
 423
 424 2018-04-06  Tamar Christina  <tamar.christina@arm.com>
 425
 426         * expr.c (copy_blkmode_to_reg): Revert 254862.
 427         * doc/sourcebuild.texi (word_mode_no_slow_unalign): Likewise.
 428
 429 2018-04-06  Richard Biener  <rguenther@suse.de>
 430
 431         PR middle-end/85244
 432         * tree-dfa.c (get_ref_base_and_extent): Reset seen_variable_array_ref
 433         after seeing a component reference with an adjacent field.  Treat
 434         refs to arrays at struct end of external decls similar to
 435         refs to unconstrained commons.
 436
 437 2018-04-06  Jakub Jelinek  <jakub@redhat.com>
 438
 439         PR sanitizer/85213
 440         * fold-const.c (twoval_comparison_p): Remove SAVE_P argument and don't
 441         look through SAVE_EXPRs with non-side-effects argument.  Adjust
 442         recursive calls.
 443         (fold_comparison): Adjust twoval_comparison_p caller, don't handle
 444         save_p here.
 445
 446 2018-04-06  Richard Biener  <rguenther@suse.de>
 447
 448         PR middle-end/85180
 449         * alias.c (find_base_term): New wrapper around find_base_term
 450         unwinding CSELIB_VAL_PTR changes.
 451         (find_base_term): Do not restore CSELIB_VAL_PTR during the
 452         recursion.
 453
 454 2018-04-06  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>
 455
 456         * config/s390/s390.c (s390_z10_optimize_cmp): Expand dedicated NOP
 457         instructions.
 458         * config/s390/s390.md (UNSPECV_NOP_LR_0, UNSPECV_NOP_LR_1): New
 459         constant definitions.
 460         ("nop"): lr 0,0 -> nopr r0
 461         ("nop_lr0", "nop_lr1"): New insn definitions.
 462
 463 2018-04-06  Chung-Ju Wu  <jasonwucj@gmail.com>
 464
 465         * config/nds32/nds32.md (*stack_push, *stack_pop): Use
 466         NDS32_V3PUSH_AVAILABLE_P macro.
 467
 468 2018-04-06  Monk Chiang  <sh.chiang04@gmail.com>
 469             Chung-Ju Wu  <jasonwucj@gmail.com>
 470
 471         * config.gcc (nds32*-*-*): Add v2j v3f v3s checking.
 472         (nds32*-*-*): Add float and fpu_config into supported_defaults.
 473         * common/config/nds32/nds32-common.c (TARGET_DEFAULT_TARGET_FLAGS):
 474         Include TARGET_DEFAULT_FPU_ISA and TARGET_DEFAULT_FPU_FMA.
 475         * config/nds32/constants.md (unspec_element): Add UNSPEC_COPYSIGN,
 476         UNSPEC_FCPYNSD, UNSPEC_FCPYNSS, UNSPEC_FCPYSD and UNSPEC_FCPYSS.
 477         * config/nds32/constraints.md: New constraints and checking for hard
 478         float configuration.
 479         * config/nds32/iterators.md: New mode iterator and attribute for hard
 480         float configuration.
 481         * config/nds32/nds32-doubleword.md: Use hard float alternatives and
 482         patterns.
 483         * config/nds32/nds32-fpu.md: New file.
 484         * config/nds32/nds32-md-auxiliary.c: New functions and checkings to
 485         deal with hard float code generation.
 486         * config/nds32/nds32-opts.h (nds32_arch_type): Add ARCH_V3F and
 487         ARCH_V3S.
 488         (abi_type, float_reg_number): New enum type.
 489         * config/nds32/nds32-predicates.c: New predicates for hard float.
 490         * config/nds32/nds32-protos.h: Declare functions for hard float.
 491         * config/nds32/nds32.c: Implementation for hard float configuration.
 492         * config/nds32/nds32.h: Definitions for hard float configuration.
 493         * config/nds32/nds32.md: Include hard float machine description and
 494         modify patterns for hard float configuration.
 495         * config/nds32/nds32.opt: New options for hard float configuration.
 496         * config/nds32/predicates.md: New predicates for hard float
 497         configuration.
 498
 499 2018-04-06  Kuan-Lin Chen  <kuanlinchentw@gmail.com>
 500
 501         * common/config/nds32/nds32-common.c
 502         (nds32_option_optimization_table): Enable -mreleax-hint by default.
 503
 504 2018-04-05  Jakub Jelinek  <jakub@redhat.com>
 505
 506         PR middle-end/85195
 507         * match.pd (BIT_FIELD_REF CONSTRUCTOR@0 @1 @2): Use view_convert around
 508         CONSTRUCTOR_ELT (ctor, ...)->value.
 509
 510 2018-04-05  Uros Bizjak  <ubizjak@gmail.com>
 511
 512         PR target/85193
 513         * config/i386/i386.md (define_attr "memory"): Handle rotate1 type.
 514
 515 2018-04-05  Tom de Vries  <tom@codesourcery.com>
 516
 517         PR target/85204
 518         * config/nvptx/nvptx.c (nvptx_single): Fix neutering of bb with only
 519         cond jump.
 520
 521 2018-04-05  Shiva Chen  <shiva0217@gmail.com>
 522             Kito Cheng  <kito.cheng@gmail.com>
 523
 524         * config/nds32/constraints.md (U33): Fine-tune checking condition.
 525         * config/nds32/nds32-md-auxiliary.c (nds32_mem_format): Ditto.
 526         * config/nds32/nds32.h (nds32_16bit_address_type): Add
 527         ADDRESS_POST_MODIFY_LO_REG_IMM3U.
 528
 529 2018-04-05  Shiva Chen  <shiva0217@gmail.com>
 530             Kito Cheng  <kito.cheng@gmail.com>
 531
 532         * config/nds32/constraints.md (Ufe): New memory constraint.
 533         * config/nds32/nds32-md-auxiliary.c (nds32_mem_format,
 534         nds32_output_16bit_load): Consider r8 register for lwi45.fe format.
 535         * config/nds32/nds32.c (nds32_print_operand): Output lwi45.fe
 536         operands.
 537         * config/nds32/nds32.h (nds32_16bit_address_type): Add ADDRESS_R8_IMM7U.
 538         * config/nds32/nds32.md (*mov<mode>): Adjust pattern.
 539
 540 2018-04-05  Chung-Ju Wu  <jasonwucj@gmail.com>
 541
 542         * config/nds32/nds32.md: Use optimize_size in the condition for
 543         alu-shift instructions.
 544
 545 2018-04-05  Chung-Ju Wu  <jasonwucj@gmail.com>
 546
 547         * config/nds32/nds32.md (divsi4, udivsi4): New patterns.
 548
 549 2018-04-05  Chung-Ju Wu  <jasonwucj@gmail.com>
 550
 551         * config/nds32/nds32.md (negsi2): Refine pattern.
 552
 553 2018-04-05  Kito Cheng  <kito.cheng@gmail.com>
 554             Chung-Ju Wu  <jasonwucj@gmail.com>
 555
 556         * config/nds32/iterators.md (shift_rotate): New code iterator.
 557         (shift): New code attribute.
 558         * config/nds32/nds32-md-auxiliary.c (nds32_expand_constant): New.
 559         * config/nds32/nds32-protos.h (nds32_expand_constant): Declare.
 560         * config/nds32/nds32.c (nds32_print_operand): Deal with more cases.
 561         * config/nds32/nds32.md (addsi3, *add_srli): Refine implementation for
 562         bit-wise operations.
 563         (andsi3, *andsi3): Ditto.
 564         (iorsi3, *iorsi3, *or_slli, *or_srli): Ditto.
 565         (xorsi3, *xorsi3, *xor_slli, *xor_srli): Ditto.
 566         (<shift>si3, *ashlsi3, *ashrsi3, *lshrsi3, *rotrsi3): Ditto.
 567         * config/nds32/predicates.md (nds32_rimm5u_operand, nds32_and_operand,
 568         nds32_ior_operand, nds32_xor_operand): New predicates.
 569
 570 2018-04-05  Chung-Ju Wu  <jasonwucj@gmail.com>
 571
 572         * config/nds32/nds32.md (add<mode>3, sub<mode>3): Rename to ...
 573         (addsi3, subsi3): ... this.
 574
 575 2018-04-05  Chung-Ju Wu  <jasonwucj@gmail.com>
 576
 577         * config/nds32/nds32.md (*sub_srli, *and_slli): Fine-tune predicator.
 578
 579 2018-04-05  Chung-Ju Wu  <jasonwucj@gmail.com>
 580
 581         * config/nds32/nds32.md: Adjust indention.
 582
 583 2018-04-05  Kito Cheng  <kito.cheng@gmail.com>
 584
 585         * config/nds32/nds32.md (feature): New attribute.
 586
 587 2018-04-05  Chung-Ju Wu  <jasonwucj@gmail.com>
 588
 589         * config/nds32/nds32.md (subtype): New attribute.
 590
 591 2018-04-04  Thomas Preud'homme  <thomas.preudhomme@arm.com>
 592
 593         PR target/85203
 594         * config/arm/arm-builtins.c (arm_expand_builtin): Change
 595         expansion to perform a bitwise AND of the argument followed by a
 596         boolean negation of the result.
 597
 598 2018-04-04  Peter Bergner  <bergner@vnet.ibm.com>
 599
 600         PR rtl-optimization/84878
 601         * ddg.c (add_cross_iteration_register_deps): Use DF_REF_BB to determine
 602         the basic block.  Assert the use reference is not artificial and that
 603         it has an associated insn.
 604
 605 2018-04-04  Michael Matz  <matz@suse.de>
 606
 607         * builtins.c (compute_objsize): Pass correct operand
 608         to array_at_struct_end_p.
 609
 610 2018-04-04  Richard Biener  <rguenther@suse.de>
 611
 612         PR lto/85176
 613         * dwarf2out.c (dwarf2out_register_external_die): Peel namespaces
 614         from contexts for DINFO_LEVEL_TERSE and below.
 615
 616 2018-04-04  Kito Cheng  <kito.cheng@gmail.com>
 617
 618         * config/nds32/nds32-doubleword.md (move_<mode>): Require
 619         resiter_operand condition.
 620         * config/nds32/nds32.md (*move<mode>): Ditto.
 621
 622 2018-04-04  Kito Cheng  <kito.cheng@gmail.com>
 623             Monk Chiang  <sh.chiang04@gmail.com>
 624
 625         * config/nds32/nds32.md (movmisalign<mode>): New pattern.
 626
 627 2018-04-04  Chung-Ju Wu  <jasonwucj@gmail.com>
 628
 629         * config/nds32/nds32.md (movqi, movhi): Merge into mov<mode>.
 630
 631 2018-04-04  Chung-Ju Wu  <jasonwucj@gmail.com>
 632             Kito Cheng  <kito.cheng@gmail.com>
 633
 634         * config/nds32/nds32-md-auxiliary.c (nds32_inverse_cond_code,
 635         nds32_cond_code_str, output_cond_branch,
 636         output_cond_branch_compare_zero, nds32_expand_cbranch,
 637         nds32_expand_cstore, nds32_expand_movcc,
 638         nds32_output_cbranchsi4_equality_zero,
 639         nds32_output_cbranchsi4_equality_reg,
 640         nds32_output_cbranchsi4_equality_reg_or_const_int,
 641         nds32_output_cbranchsi4_greater_less_zero: New functions.
 642         * config/nds32/nds32-protos.h (nds32_expand_cbranch,
 643         nds32_expand_cstore, nds32_expand_movcc,
 644         nds32_output_cbranchsi4_equality_zero,
 645         nds32_output_cbranchsi4_equality_reg,
 646         nds32_output_cbranchsi4_equality_reg_or_const_int,
 647         nds32_output_cbranchsi4_greater_less_zero): Declare.
 648         * config/nds32/predicates.md (nds32_movecc_comparison_operator,
 649         nds32_rimm11s_operand): New predicates.
 650         * config/nds32/nds32.h (nds32_expand_result_type): New enum type.
 651         * config/nds32/nds32.md: Rewrite all the branch and conditional move
 652         patterns.
 653
 654 2018-04-04  Kito Cheng  <kito.cheng@gmail.com>
 655
 656         * config/nds32/nds32-doubleword.md: Refine all the instruction type.
 657         * config/nds32/nds32.md: Ditto.
 658         * config/nds32/pipelines.md: Ditto.
 659
 660 2018-04-04  Richard Biener  <rguenther@suse.de>
 661
 662         PR tree-optimization/85168
 663         * tree-ssa-sccvn.c (vn_reference_maybe_forwprop_address): Avoid
 664         propagating abnormals.
 665
 666 2018-04-04  Chung-Ju Wu  <jasonwucj@gmail.com>
 667
 668         * config/nds32/nds32.md (enabled): Use yes/no for this attribute.
 669
 670 2018-04-04  Chung-Ju Wu  <jasonwucj@gmail.com>
 671             Kito Cheng  <kito.cheng@gmail.com>
 672
 673         * config/nds32/nds32-md-auxiliary.c (nds32_long_call_p): New function.
 674         * config/nds32/nds32-protos.h (nds32_long_call_p): Declare.
 675         * config/nds32/nds32.c (nds32_function_ok_for_sibcall): New function.
 676         (TARGET_FUNCTION_OK_FOR_SIBCALL): Define.
 677         * config/nds32/nds32.md (sibcall_internal): New.
 678         (sibcall_register): Remove.
 679         (sibcall_immediate): Remove.
 680         (sibcall_value_internal): New.
 681         (sibcall_value_register): Remove.
 682         (sibcall_value_immediate): Remove.
 683         * config/nds32/predicates.md (nds32_general_register_operand): New.
 684         (nds32_call_address_operand): New.
 685
 686 2018-04-03  Jakub Jelinek  <jakub@redhat.com>
 687
 688         PR rtl-optimization/85167
 689         * shrink-wrap.c (move_insn_for_shrink_wrap): Don't set bb_uses and
 690         bb_defs if *split_p, instead preinitialize it to NULL.
 691
 692         PR tree-optimization/85156
 693         * builtins.c (fold_builtin_expect): Use save_expr on arg1 to avoid
 694         evaluating the argument multiple times.
 695
 696 2018-04-03  Bill Schmidt  <wschmidt@linux.ibm.com>
 697
 698         * config/rs6000/emmintrin.h (_mm_cvtpd_epi32): Use __vector rather
 699         than vector.
 700         (_mm_cvtpd_ps): Likewise.
 701         (_mm_cvttpd_epi32): Likewise.
 702         * config/rs6000/mmintrin.h (_mm_unpacklo_pi8): Likewise.
 703         * config/rs6000/xmmintrin.h: For strict-ANSI C++ or C11, undefine
 704         vector, pixel, and bool following altivec.h include.
 705
 706 2018-04-03  Martin Sebor  <msebor@redhat.com>
 707
 708         * doc/extend.texi (Common Function Attributes): Clarify.
 709         (const attribute): Likewise.
 710         (pure attribute): Likewise.
 711
 712 2018-04-03  Jakub Jelinek  <jakub@redhat.com>
 713
 714         PR target/85169
 715         * config/i386/i386.c (ix86_expand_vector_set): Use
 716         HOST_WIDE_INT_1U << elt instead of 1 << elt.  Formatting fix.
 717
 718 2018-04-03  Uros Bizjak  <ubizjak@gmail.com>
 719
 720         * config/i386/i386.c (emit_i387_cw_initialization): Always use logic
 721         instructions when changing rounding bits to preserve precision bits
 722         in the x87 control word.
 723
 724 2018-04-03  Martin Liska  <mliska@suse.cz>
 725
 726         PR tree-optimization/82491
 727         * rtl.h (strip_offset_and_add): Replace += suboffset with
 728         poly_uint64 () + suboffset.
 729
 730 2018-03-29  Martin Liska  <mliska@suse.cz>
 731             Martin Jambor  <mjambor@suse.cz>
 732
 733         PR ipa/84947
 734         * ipa-cp.c (propagate_bits_across_jump_function): Bail out if
 735         param_type is not an integral or pointer type.
 736
 737 2018-04-03  Richard Biener  <rguenther@suse.de>
 738
 739         * sese.h (recompute_all_dominators): Remove.
 740
 741 2018-04-02  Martin Sebor  <msebor@redhat.com>
 742
 743         * doc/invoke.texi (-Wrestrict): Fix typos.
 744
 745 2018-04-02  Jim Wilson  <jimw@sifive.com>
 746
 747         * config/riscv/riscv.h (SHIFT_COUNT_TRUNCATED): Set to zero.
 748         * config/riscv/riscv.md (<optab>si3): Use QImode shift count.
 749         (<optab>di3, <optab>si3_extend): Likewise.
 750         (<optab>si3_mask, <optab>si3_mask_1): New.
 751         (<optab>di3_mask, <optab>di3_mask_1): New.
 752         (<optab>si3_extend_mask, <optab>si3_extend_mask_1): New.
 753         (lshrsi3_zero_extend_1): Use VOIDmode shift count.
 754         * config/riscv/sync.md (atomic_test_and_set): Emit QImode shift count.
 755
 756 2018-04-02  Gerald Pfeifer  <gerald@pfeifer.com>
 757
 758         * doc/cpp.texi (Variadic Macros): Fix line continuation in an
 759         example.
 760
 761 2018-04-02  Chung-Ju Wu  <jasonwucj@gmail.com>
 762
 763         * config/nds32/nds32.c (TARGET_CANONICALIZE_COMPARISON): Define.
 764         (nds32_canonicalize_comparison): New function.
 765
 766 2018-04-01  Chung-Ju Wu  <jasonwucj@gmail.com>
 767             Kito Cheng  <kito.cheng@gmail.com>
 768             Kuan-Lin Chen  <kuanlinchentw@gmail.com>
 769
 770         * config.gcc (nds32): Add nds32-relax-opt.o into extra_objs.
 771         * config/nds32/constants.md (unspec_volatile_element): Add
 772         UNSPEC_VOLATILE_RELAX_GROUP.
 773         * config/nds32/nds32-relax-opt.c: New file.
 774         * config/nds32/nds32-predicates.c
 775         (nds32_symbol_load_store_p): New function.
 776         * config/nds32/nds32-protos.h
 777         (nds32_symbol_load_store_p): Declare function.
 778         (make_pass_nds32_relax_opt): Declare new rtl pass function.
 779         * config/nds32/nds32.c
 780         (nds32_register_pass): New function to register pass.
 781         (nds32_register_passes): New function to register passes.
 782         * config/nds32/nds32.md (relax_group): New pattern.
 783         * config/nds32/nds32.opt (mrelax-hint): New option.
 784         * config/nds32/t-nds32 (nds32-relax-opt.o): New dependency.
 785
 786 2018-04-01  Kito Cheng  <kito.cheng@gmail.com>
 787
 788         * config/nds32/t-nds32: Modify files dependency.
 789
 790 2018-04-01  Chung-Ju Wu  <jasonwucj@gmail.com>
 791
 792         * config/nds32/nds32.h (FUNCTION_PROFILER): Output newline character.
 793         (PROFILE_HOOK): Define its implementation.
 794
 795 2018-04-01  Chung-Ju Wu  <jasonwucj@gmail.com>
 796
 797         * config/nds32/nds32.h (WCHAR_TYPE, WCHAR_TYPE_SIZE): Use unsigned int
 798         type and 32-bit size.
 799
 800 2018-04-01  Jakub Jelinek  <jakub@redhat.com>
 801
 802         PR middle-end/85090
 803         * config/i386/sse.md (V): Add V64QI and V32HI for TARGET_AVX512F.
 804         (V_128_256): New mode iterator.
 805         (*avx512dq_vextract<shuffletype>64x2_1 splitter): New define_split.
 806         (*avx512f_vextract<shuffletype>32x4_1 splitter): Likewise.
 807         (xop_pcmov_<mode><avxsizesuffix>): Use V_128_256 mode iterator instead
 808         of V.
 809         * config/i386/i386.c (ix86_expand_vector_set): Improve V32HImode and
 810         V64QImode expansion for !TARGET_AVX512BW && TARGET_AVX512F.
 811
 812 2018-03-31  Segher Boessenkool  <segher@kernel.crashing.org>
 813
 814         PR target/83315
 815         * config/rs6000/xmmintrin.h (_mm_set_ps, _mm_max_ps): Handle (quiet)
 816         NaN inputs correctly.
 817
 818 2018-03-30  Peter Bergner  <bergner@vnet.ibm.com>
 819
 820         PR target/80546
 821         * config/rs6000/vsx.md (??r): New mode attribute.
 822         (*vsx_mov<mode>_64bit): Use it.
 823         (*vsx_mov<mode>_32bit): Likewise.
 824
 825 2018-03-30  Martin Sebor  <msebor@redhat.com>
 826
 827         PR tree-optimization/84818
 828         * builtins.c (check_access): Use warning_n.
 829
 830 2018-03-30  Aaron Sawdey  <acsawdey@linux.vnet.ibm.com>
 831
 832         PR target/83822
 833         * config/rs6000/rs6000-string.c (expand_compare_loop): Fix redundant
 834         condition.
 835         * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): Fix redundant
 836         condition.
 837
 838 2018-03-30  Julia Koval  <julia.koval@intel.com>
 839
 840         PR target/84413
 841         * x86-tune.def (movx, partial_reg_dependency): Enable for
 842         m_SKYLAKE_AVX512.
 843
 844 2018-03-29  Vladimir Makarov  <vmakarov@redhat.com>
 845
 846         PR inline-asm/84985
 847         * lra-constraints.c (process_alt_operands): Move setting
 848         this_alternative_matches below.
 849
 850 2018-03-29  Martin Liska  <mliska@suse.cz>
 851
 852         PR lto/84995.
 853         * doc/invoke.texi: Document how LTO works with debug info.
 854         Describe auto-load support of binutils.  Mention 'x86-64'
 855         as valid option value of -march option.
 856
 857 2018-03-29  Jakub Jelinek  <jakub@redhat.com>
 858
 859         * config/i386/sse.md (<avx512>_blendm<mode>): Use <sseintprefix>.
 860
 861         PR c/85094
 862         * fold-const.c (operand_equal_p): Handle DEBUG_BEGIN_STMT.
 863         For STATEMENT_LIST, pass down OEP_LEXICOGRAPHIC and maybe
 864         OEP_NO_HASH_CHECK for recursive call, to avoid exponential
 865         checking.
 866
 867 2018-03-28  Peter Bergner  <bergner@vnet.ibm.com>
 868
 869         PR target/84912
 870         * config/rs6000/rs6000.h (RS6000_BTM_POWERPC64): New define.
 871         (RS6000_BTM_COMMON): Add RS6000_BTM_POWERPC64.
 872         * config/rs6000/rs6000.c (rs6000_builtin_mask_calculate): Add support
 873         for RS6000_BTM_POWERPC64.
 874         (rs6000_invalid_builtin): Add handling for RS6000_BTM_POWERPC64
 875         (rs6000_builtin_mask_names): Add RS6000_BTM_POWERPC64.
 876         * config/rs6000/rs6000-builtin.def (BU_P7_POWERPC64_MISC_2): New macro
 877         definition.
 878         (DIVDE): Use it.
 879         (DIVDEU): Likewise.
 880
 881 2018-03-28 Carl Love  <cel@us.ibm.com>
 882
 883         Revert
 884         2017-09-27  Carl Love  <cel@us.ibm.com>
 885
 886         * config/rs6000/rs6000-builtin.def (BU_FP_1MISC_1): Add define macro.
 887         (FCTID, FCTIW): Add BU_FP_MISC_1 macro expansion for builtins.
 888         * config/rs6000/rs6000.md (lrintsfsi2): Add define_insn for the
 889         fctiw instruction.
 890
 891 2018-03-28  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
 892
 893         * config/rs6000/xmmintrin.h (_mm_max_pi16): Use __vector __bool
 894         instead of __vector bool.
 895         (_mm_max_pu8): Likewise.
 896         (_mm_min_pi16): Likewise.
 897
 898 2018-03-28  Peter Bergner  <bergner@vnet.ibm.com>
 899
 900         PR target/84912
 901         * config/rs6000/rs6000-builtin.def (DIVWEO): Delete macro expansion.
 902         (DIVWEUO): Likewise.
 903         (DIVDEO): Likewise.
 904         (DIVDEUO): Likewise.
 905         * config/rs6000/rs6000.c (builtin_function_type): Remove support for
 906         DIVWEUO and DIVDEUO.
 907         * config/rs6000/rs6000.md (UNSPEC_DIVEO, UNSPEC_DIVEUO): Delete unspecs.
 908         (UNSPEC_DIV_EXTEND): Remove deleted unspecs.
 909         (div_extend): Likewise.
 910         * doc/extend.texi (__builtin_divweo): Remove documentation for deleted
 911         builtin function.
 912         (__builtin_divweuo): Likewise.
 913         (__builtin_divdeo): Likewise.
 914         (__builtin_divdeuo): Likewise.
 915
 916 2018-03-28  Jakub Jelinek  <jakub@redhat.com>
 917
 918         PR target/85095
 919         * config/i386/i386.md (*add<mode>3_carry_0, *addsi3_carry_zext_0,
 920         *sub<mode>3_carry_0, *subsi3_carry_zext_0): New patterns.
 921
 922         PR tree-optimization/82004
 923         * gimple-match-head.c (optimize_pow_to_exp): New function.
 924         * match.pd (pow(C,x) -> exp(log(C)*x)): Wrap with #if GIMPLE.
 925         Don't fold to exp if optimize_pow_to_exp is false.
 926
 927 2018-03-28  Martin Liska  <mliska@suse.cz>
 928
 929         PR other/84819
 930         * calls.c (initialize_argument_information): Fix trailing space.
 931         * common.opt: Fix typo and provide better explanation for
 932         -fsanitize-coverage option.
 933         * config/i386/i386.opt: Fix typo.
 934
 935 2018-03-28  Jakub Jelinek  <jakub@redhat.com>
 936             Martin Liska  <mliska@suse.cz>
 937
 938         PR sanitizer/85081
 939         * gimplify.c (asan_poison_variable): Don't do the check for
 940         gimplify_omp_ctxp here.
 941         (gimplify_decl_expr): Do it here.
 942         (gimplify_target_expr): Likewise.
 943
 944 2018-03-28  Martin Liska  <mliska@suse.cz>
 945
 946         PR target/84988
 947         * config/i386/i386.c (ix86_function_arg_advance): Do not call
 948         chkp_type_bounds_count if MPX is not enabled.
 949
 950 2018-03-27  Chung-Ju Wu  <jasonwucj@gmail.com>
 951
 952         * config/nds32/nds32.h (BRANCH_COST): Adjust cost.
 953
 954 2018-03-27  Michael Meissner  <meissner@linux.vnet.ibm.com>
 955
 956         PR target/84914
 957         * config/rs6000/rs6000.c (create_complex_muldiv): New helper
 958         function to create the function decl for complex long double
 959         multiply and divide for -mabi=ieeelongdouble.
 960         (init_float128_ieee): Call it.
 961
 962 2018-03-27  H.J. Lu  <hongjiu.lu@intel.com>
 963
 964         PR target/85044
 965         * config/i386/i386.c (ix86_trampoline_init): Insert ENDBR for
 966         -fcf-protection=branch -mibt.
 967         * config/i386/i386.h (TRAMPOLINE_SIZE): Increased by 4 bytes.
 968
 969 2018-03-27  Ramana Radhakrishnan  <ramana.radhakrishnan@arm.com>
 970
 971         PR target/81863
 972         * config/arm/arm.c (arm_valid_symbolic_address): Handle arm_word_relocations
 973
 974 2018-03-27  Cesar Philippidis  <cesar@codesourcery.com>
 975
 976         PR target/85056
 977         * config/nvptx/nvptx.c (nvptx_assemble_decl_begin): Add '[]' to
 978         extern array declarations.
 979
 980 2018-03-27  Richard Biener  <rguenther@suse.de>
 981
 982         PR middle-end/84067
 983         * match.pd ((A * C) +- (B * C) -> (A+-B) * C): Guard with
 984         explicit single_use checks.
 985
 986 2018-03-27  Richard Biener  <rguenther@suse.de>
 987
 988         PR tree-optimization/85082
 989         * tree-ssa-sccvn.c (vn_reference_lookup_or_insert_for_pieces):
 990         Valueize the VUSE.
 991
 992 2018-03-27  Ramana Radhakrishnan  <ramana.radhakrishnan@arm.com>
 993
 994         * config.gcc (aarch64*-*-linux*): New TARGET_DEFAULT_ASYNC_UNWIND_TABLES
 995         * common/config/aarch64/aarch64-common.c (aarch64_optimization_table[]):
 996         Turn on fasynchronous-unwind-tables and funwind-tables.
 997
 998 2018-03-26  Uros Bizjak  <ubizjak@gmail.com>
 999
1000         PR target/85073
1001         * config/i386/i386.md (*bmi_blsr_<mode>_cmp): New insn pattern.
1002         (*bmi_blsr_<mode>_ccz): Ditto.
1003
1004 2018-03-26  Tom de Vries  <tom@codesourcery.com>
1005
1006         PR tree-optimization/85063
1007         * omp-general.c (offloading_function_p): New function.  Factor out
1008         of ...
1009         * omp-offload.c (pass_omp_target_link::gate): ... here.
1010         * omp-general.h (offloading_function_p): Declare.
1011         * tree-switch-conversion.c (build_one_array): Mark CSWTCH.x variable
1012         with attribute omp declare target for offloading functions.
1013
1014 2018-03-24  Richard Sandiford  <richard.sandiford@linaro.org>
1015
1016         PR tree-optimization/84005
1017         * tree-data-ref.h (get_base_for_alignment): Declare.
1018         * tree-data-ref.c (get_base_for_alignment_1): New function.
1019         (get_base_for_alignment): Likewise.
1020         * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Use
1021         get_base_for_alignment to find a suitable base object, instead
1022         of always using drb->base_address.
1023
1024 2018-03-23  Jakub Jelinek  <jakub@redhat.com>
1025
1026         PR inline-asm/85022
1027         * emit-rtl.c (init_emit_regs): Indicate that VOIDmode MEMs don't have
1028         known size by default.
1029
1030 2018-03-23  Vladimir Makarov  <vmakarov@redhat.com>
1031
1032         PR inline-asm/85030
1033         * lra-constraints.c (process_alt_operands): Don't match BLKmode
1034         and non BLKmode operands.
1035
1036 2018-03-23  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
1037
1038         PR target/85026
1039         * config/arm/arm.md (unaligned_loadhis): Remove first alternative.
1040         Clean up attributes.
1041
1042 2018-03-23  Richard Biener  <rguenther@suse.de>
1043
1044         PR debug/85020
1045         * dwarf2out.c (rtl_for_decl_location): Do not generate RTL early when
1046         we are going to emit early debug for LTO.
1047
1048 2018-03-23  Jakub Jelinek  <jakub@redhat.com>
1049
1050         PR inline-asm/85034
1051         * function.c (match_asm_constraints_1): Don't optimize if input
1052         doesn't satisfy general_operand predicate for output's mode.
1053
1054         PR inline-asm/85022
1055         * alias.c (write_dependence_p): Don't require for x_canonicalized
1056         non-VOIDmode if x has VOIDmode.
1057
1058         PR sanitizer/85029
1059         * sanopt.c (maybe_optimize_ubsan_ptr_ifn): If DECL_REGISTER (base),
1060         just don't try to optimize it rather than assert it never happens.
1061
1062 2018-03-22  Kelvin Nilsen  <kelvin@gcc.gnu.org>
1063
1064         * config/rs6000/rs6000-builtin.def: Remove various BU_ALTIVEC_X
1065         macro expansions for definition of ST_INTERNAL_<mode> and
1066         LD_INTERNAL_<mode> builtins.
1067         * config/rs6000/rs6000-protos.h (rs6000_address_for_altivec):
1068         Remove prototype.
1069         * config/rs6000/rs6000.c (altivec_expand_ld_builtin): Delete this
1070         function.
1071         (altivec_expand_st_builtin): Likewise.
1072         (altivec_expand_builtin): Remove calls to deleted functions.
1073         (rs6000_address_for_altivec): Delete this function.
1074         * config/rs6000/vector.md: Remove expands for
1075         vector_altivec_load_<mode> and vector_altivec_store_<mode>.
1076
1077 2018-03-22  Sudakshina Das  <sudi.das@arm.com>
1078
1079         PR target/84826
1080         * config/arm/arm.h (machine_function): Add static_chain_stack_bytes.
1081         * config/arm/arm.c (arm_compute_static_chain_stack_bytes): Avoid
1082         re-computing once computed.
1083         (arm_expand_prologue): Compute machine->static_chain_stack_bytes.
1084         (arm_init_machine_status): Initialize
1085         machine->static_chain_stack_bytes.
1086
1087 2018-03-22  Kelvin Nilsen  <kelvin@gcc.gnu.org>
1088
1089         PR target/84760
1090         * doc/extend.texi: Add four new prototypes for vec_ld.
1091         * config/rs6000/rs6000-builtin.def (LVX_V1TI): Reorder symbol
1092         definitions for more logical presentation.
1093         * config/rs6000/rs6000-c.c: (altivec_overloaded_builtins): Add
1094         entries for V1TI variants of __builtin_altivec_ld builtin.
1095         * config/rs6000/rs6000.c: (altivec_expand_lv_builtin): Add test and
1096         handling of V1TI variant of LVX icode pattern.
1097         (altivec_expand_builtin): Add case for ALTIVEC_BUILTIN_LVX_V1TI.
1098         (rs6000_gimple_fold_builtin): Likewise.
1099         (altivec_init_builtins): Add code to define
1100         __builtin_altivec_lvx_v1ti function.
1101
1102 2018-03-22  Jakub Jelinek  <jakub@redhat.com>
1103
1104         PR inline-asm/84941
1105         * function.c (match_asm_constraints_1): Don't do the optimization
1106         if input isn't a REG, SUBREG, MEM or constant.
1107
1108 2018-03-22  Tom de Vries  <tom@codesourcery.com>
1109
1110         PR tree-optimization/84956
1111         * tree-ssa-tail-merge.c (find_clusters_1): Skip bbs with
1112         bb_has_abnormal_pred.
1113
1114 2018-03-22  Jakub Jelinek  <jakub@redhat.com>
1115
1116         PR sanitizer/85018
1117         * dwarf2asm.c (dw2_output_indirect_constant_1): Set
1118         DECL_INITIAL (decl) to decl at the end.
1119         * varasm.c (use_blocks_for_decl_p): Revert the 2018-03-20 change,
1120         adjust the comment.
1121
1122 2018-03-21  Joseph Myers  <joseph@codesourcery.com>
1123
1124         * doc/extend.texi (__builtin_tgmath): Document when complex
1125         integer types are treated as _Complex _Float64.
1126
1127 2018-03-21  Tom de Vries  <tom@codesourcery.com>
1128
1129         * doc/extend.texi (__builtin_extend_pointer): Remove pasto.
1130
1131 2018-03-21  Jakub Jelinek  <jakub@redhat.com>
1132
1133         PR tree-optimization/84960
1134         * tree-cfg.c (remove_bb): Don't move forced labels into bb->prev_bb
1135         if it is ENTRY block, move them into single succ of ENTRY in that case.
1136
1137 2018-03-21  Richard Sandiford  <richard.sandiford@linaro.org>
1138
1139         PR tree-optimization/84811
1140         * poly-int.h (poly_span_traits): Remove the T3 parameter and
1141         promote HOST_WIDE_INT T2 - T1 results to unsigned HOST_WIDE_INT.
1142         (maybe_in_range_p, known_in_range_p, ranges_known_overlap_p):
1143         (known_subrange_p): Update accordingly.  Cast each value involved
1144         in the size comparison, rather than casting the result of the
1145         subtraction.
1146
1147 2018-03-21  Jakub Jelinek  <jakub@redhat.com>
1148
1149         PR tree-optimization/84982
1150         * gimple-ssa-store-merging.c (invert_op): Handle boolean inversion
1151         by flipping the least significant bit rather than all bits from
1152         bitpos to bitpos + bitsize - 1.
1153
1154 2018-03-21  Nathan Sidwell  <nathan@acm.org>
1155
1156         * doc/extend.texi (Deprecated Features): Remove mention of
1157         long-deleted deprecations.
1158
1159 2018-03-21  Rainer Orth  <ro@CeBiTec.Uni-Bielefeld.DE>
1160
1161         PR jit/84288
1162         * configure.ac (gcc_cv_ld_soname) <*-*-solaris2*>: Set.
1163         * configure: Regenerate.
1164
1165 2018-03-21  Tom de Vries  <tom@codesourcery.com>
1166
1167         PR tree-optimization/83126
1168         * tree-parloops.c (num_phis): New function.
1169         (gen_parallel_loop): Detect and handle canonicalize_loop_ivs failure.
1170
1171 2018-03-21  Nathan Sidwell  <nathan@acm.org>
1172
1173         * doc/extend.texi (Deprecated Features): Update deprecated flags,
1174         mention anon-struct/union members and trailing attributes.
1175
1176 2018-03-21  Bin Cheng  <bin.cheng@arm.com>
1177
1178         PR tree-optimization/84969
1179         * tree-loop-distribution.c (fuse_memset_builtins): Don't reorder
1180         builtin memset partitions if they set different rhs values.
1181
1182 2018-03-21  Jakub Jelinek  <jakub@redhat.com>
1183
1184         PR rtl-optimization/84989
1185         * simplify-rtx.c (simplify_unary_operation_1): Don't try to simplify
1186         VEC_DUPLICATE with scalar result mode.
1187
1188 2018-03-21  Martin Liska  <mliska@suse.cz>
1189
1190         PR ipa/84963
1191         * ipa-icf.c (sem_item_optimizer::fixup_points_to_sets): Remove
1192         not intended return statement.
1193
1194 2018-03-21  Martin Liska  <mliska@suse.cz>
1195
1196         PR target/84988
1197         * tree-chkp.c (CHKP_ARRAY_MAX_CHECK_STEPS): Define a new macro.
1198         (chkp_find_bound_slots_1): Limit number of iterations.
1199
1200 2018-03-20  David H. Gutteridge  <dhgutteridge@sympatico.ca>
1201
1202         PR target/84838
1203         * Minor grammar fixes for x86 options.
1204
1205 2018-03-20  Jakub Jelinek  <jakub@redhat.com>
1206
1207         PR debug/84875
1208         * dce.c (delete_unmarked_insns): Don't remove frame related noop moves
1209         holding REG_CFA_RESTORE notes, instead turn them into a USE.
1210
1211 2018-03-20  Peter Bergner  <bergner@vnet.ibm.com>
1212
1213         PR target/83789
1214         * config/rs6000/altivec.md (altivec_lvx_<mode>_2op): Delete define_insn.
1215         (altivec_lvx_<mode>_1op): Likewise.
1216         (altivec_stvx_<mode>_2op): Likewise.
1217         (altivec_stvx_<mode>_1op): Likewise.
1218         (altivec_lvx_<VM2:mode>): New define_expand.
1219         (altivec_stvx_<VM2:mode>): Likewise.
1220         (altivec_lvx_<VM2:mode>_2op_<P:mptrsize>): New define_insn.
1221         (altivec_lvx_<VM2:mode>_1op_<P:mptrsize>): Likewise.
1222         (altivec_stvx_<VM2:mode>_2op_<P:mptrsize>): Likewise.
1223         (altivec_stvx_<VM2:mode>_1op_<P:mptrsize>): Likewise.
1224         * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Use new expanders.
1225         (rs6000_gen_lvx): Likewise.
1226         * config/rs6000/rs6000.c (altivec_expand_lv_builtin): Likewise.
1227         (altivec_expand_stv_builtin): Likewise.
1228         (altivec_expand_builtin): Likewise.
1229         * config/rs6000/vector.md: Likewise.
1230
1231 2018-03-20  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
1232
1233         PR target/82518
1234         * config/arm/arm.c (arm_array_mode_supported_p): Return false for
1235         BYTES_BIG_ENDIAN.
1236
1237 2018-03-20  Richard Biener  <rguenther@suse.de>
1238
1239         PR target/84986
1240         * config/i386/i386.c (ix86_add_stmt_cost): Only cost
1241         sign-conversions as zero, fall back to standard scalar_stmt
1242         cost for the rest.
1243
1244 2018-03-20  Martin Liska  <mliska@suse.cz>
1245
1246         PR ipa/84825
1247         * predict.c (rebuild_frequencies): Handle case when we have
1248         PROFILE_ABSENT, but flag_guess_branch_prob is false.
1249
1250 2018-03-20  Jakub Jelinek  <jakub@redhat.com>
1251
1252         PR target/84990
1253         * dwarf2asm.c (dw2_output_indirect_constant_1): Temporarily turn off
1254         flag_section_anchors.
1255         * varasm.c (use_blocks_for_decl_p): Remove hack for
1256         dw2_force_const_mem.
1257
1258         PR target/84845
1259         * config/aarch64/aarch64.md (*aarch64_reg_<mode>3_neg_mask2): Rename
1260         to ...
1261         (*aarch64_<optab>_reg_<mode>3_neg_mask2): ... this.  If pseudos can't
1262         be created, use lowpart_subreg of operands[0] rather than operands[0]
1263         itself.
1264         (*aarch64_reg_<mode>3_minus_mask): Rename to ...
1265         (*aarch64_ashl_reg_<mode>3_minus_mask): ... this.
1266         (*aarch64_<optab>_reg_di3_mask2): Use const_int_operand predicate
1267         and n constraint instead of aarch64_shift_imm_di and Usd.
1268         (*aarch64_reg_<optab>_minus<mode>3): Rename to ...
1269         (*aarch64_<optab>_reg_minus<mode>3): ... this.
1270
1271 2018-03-20  Sudakshina Das  <sudi.das@arm.com>
1272
1273         PR target/82989
1274         * config/arm/neon.md (ashldi3_neon): Update ?s for constraints
1275         to favor GPR over NEON registers.
1276         (<shift>di3_neon): Likewise.
1277
1278 2018-03-20  Tom de Vries  <tom@codesourcery.com>
1279
1280         PR target/84952
1281         * config/nvptx/nvptx.c (nvptx_single): Don't neuter bar.sync.
1282         (nvptx_process_pars): Emit bar.sync asap and alap.
1283
1284 2018-03-20  Tom de Vries  <tom@codesourcery.com>
1285
1286         PR target/84954
1287         * config/nvptx/nvptx.c (prevent_branch_around_nothing): Also update
1288         seen_label if seen_label is already set.
1289
1290 2018-03-20  Jakub Jelinek  <jakub@redhat.com>
1291
1292         PR target/84945
1293         * config/i386/i386.c (fold_builtin_cpu): For features above 31
1294         use __cpu_features2 variable instead of __cpu_model.__cpu_features[0].
1295         Use 1U instead of 1.  Formatting fixes.
1296
1297         PR c/84953
1298         * builtins.c (fold_builtin_strpbrk): For strpbrk(x, "") use type
1299         instead of TREE_TYPE (s1) for the return value.
1300
1301 2018-03-19  Jakub Jelinek  <jakub@redhat.com>
1302
1303         PR tree-optimization/84946
1304         * gimple-ssa-store-merging.c (mem_valid_for_store_merging): Compute
1305         bitsize + bitsize in poly_uint64 rather than poly_int64.
1306
1307         PR sanitizer/78651
1308         * dwarf2asm.c: Include fold-const.c.
1309         (dw2_output_indirect_constant_1): Set DECL_INITIAL (decl) to ADDR_EXPR
1310         of decl rather than decl itself.
1311
1312         PR rtl-optimization/84643
1313         * memmodel.h (enum memmodel): Add MEMMODEL_MAX enumerator.
1314
1315 2018-03-19  Maxim Ostapenko  <m.ostapenko@samsung.com>
1316
1317         PR sanitizer/78651
1318         * dwarf2asm.c (dw2_output_indirect_constant_1): Disable ASan before
1319         calling assemble_variable.
1320
1321 2018-03-19  Sudakshina Das  <sudi.das@arm.com>
1322
1323         PR target/81647
1324         * config/aarch64/aarch64-simd.md (vec_cmp<mode><v_int_equiv>): Modify
1325         instructions for UNLT, UNLE, UNGT, UNGE, UNEQ, UNORDERED and ORDERED.
1326
1327 2018-03-19  Jim Wilson  <jimw@sifive.com>
1328
1329         PR bootstrap/84856
1330         * config/riscv/riscv.c (riscv_function_arg_boundary): Use
1331         PREFERRED_STACK_BOUNDARY instead of STACK_BOUNDARY.
1332         (riscv_first_stack_step): Likewise.
1333         (riscv_option_override): Use STACK_BOUNDARY instead of
1334         MIN_STACK_BOUNDARY.
1335         * config/riscv/riscv.h (STACK_BOUNDARY): Renamed from
1336         MIN_STACK_BOUNDARY.
1337         (BIGGEST_ALIGNMENT): Set to 128.
1338         (PREFERRED_STACK_BOUNDARY): Renamed from STACK_BOUNDARY.
1339         (RISCV_STACK_ALIGN): Use PREFERRED_STACK_BOUNDARY instead of
1340         STACK_BOUNDARY.
1341
1342 2018-03-19  Richard Biener  <rguenther@suse.de>
1343
1344         PR tree-optimization/84933
1345         * tree-vrp.c (set_and_canonicalize_value_range): Treat out-of-bound
1346         values as -INF/INF when canonicalizing an ANTI_RANGE to a RANGE.
1347
1348 2018-03-19  Richard Biener  <rguenther@suse.de>
1349
1350         PR tree-optimization/84859
1351         * tree-ssa-phiopt.c (single_trailing_store_in_bb): New function.
1352         (cond_if_else_store_replacement): Perform sinking operation on
1353         single-store BBs regardless of MAX_STORES_TO_SINK setting.
1354         Generalize what a BB with a single eligible store is.
1355
1356 2018-03-19  Richard Biener  <rguenther@suse.de>
1357
1358         PR tree-optimization/84929
1359         * tree-data-ref.c (analyze_siv_subscript_cst_affine): Guard
1360         chrec_is_positive against non-chrec arg.
1361
1362 2018-03-19  Tamar Christina  <tamar.christina@arm.com>
1363
1364         PR target/84711
1365         * config/arm/arm.c (arm_can_change_mode_class): revert r258554.
1366
1367 2018-03-18  Martin Liska  <mliska@suse.cz>
1368
1369         PR rtl-optimization/84635
1370         * regrename.c (build_def_use): Use matches_mode only when
1371         matches >= 0.
1372
1373 2018-03-18  Richard Sandiford  <richard.sandiford@linaro.org>
1374
1375         PR tree-optimization/84913
1376         * tree-vect-loop.c (vectorizable_reduction): Don't try to
1377         vectorize chains of COND_EXPRs.
1378
1379 2018-03-18  Chung-Ju Wu  <jasonwucj@gmail.com>
1380
1381         * config/nds32/nds32.h (MAX_REGS_PER_ADDRESS): Fix the value.
1382
1383 2018-03-18  Chung-Ju Wu  <jasonwucj@gmail.com>
1384
1385         * config/nds32/nds32.h (LOGICAL_OP_NON_SHORT_CIRCUIT): Define.
1386
1387 2018-03-18  Chung-Ju Wu  <jasonwucj@gmail.com>
1388
1389         * config/nds32/nds32.h (CLZ_DEFINED_VALUE_AT_ZERO): Define.
1390
1391 2018-03-17  Chung-Ju Wu  <jasonwucj@gmail.com>
1392             Kito Cheng  <kito.cheng@gmail.com>
1393
1394         * config/nds32/nds32-protos.h (nds32_adjust_reg_alloc_order): Declare.
1395         * config/nds32/nds32.c (nds32_reg_alloc_order_for_speed): New array.
1396         (nds32_adjust_reg_alloc_order): New function.
1397         * config/nds32/nds32.h (ADJUST_REG_ALLOC_ORDER): Define.
1398
1399 2018-03-17  Kito Cheng  <kito.cheng@gmail.com>
1400
1401         * config/nds32/nds32.c (nds32_asm_output_mi_thunk,
1402         nds32_print_operand, nds32_print_operand_address): Use
1403         HOST_WIDE_INT_PRINT_DEC instead.
1404
1405 2018-03-17  Chung-Ju Wu  <jasonwucj@gmail.com>
1406
1407         * config/nds32/nds32.c (nds32_register_priority): Modify cost.
1408
1409 2018-03-17  Jakub Jelinek  <jakub@redhat.com>
1410
1411         PR target/84902
1412         * config/i386/i386.c (initial_ix86_tune_features,
1413         initial_ix86_arch_features): Use unsigned HOST_WIDE_INT rather than
1414         unsigned long long.
1415         (set_ix86_tune_features): Change ix86_tune_mask from unsigned int
1416         to unsigned HOST_WIDE_INT, initialize to HOST_WIDE_INT_1U << ix86_tune
1417         rather than 1u << ix86_tune.  Formatting fix.
1418         (ix86_option_override_internal): Change ix86_arch_mask from
1419         unsigned int to unsigned HOST_WIDE_INT, initialize to
1420         HOST_WIDE_INT_1U << ix86_arch rather than 1u << ix86_arch.
1421         (ix86_function_specific_restore): Likewise.
1422
1423 2018-03-16  Jakub Jelinek  <jakub@redhat.com>
1424
1425         PR target/84899
1426         * postreload.c (reload_combine_recognize_pattern): Perform
1427         INTVAL addition in unsigned HOST_WIDE_INT type to avoid UB and
1428         truncate_int_for_mode the result for the destination's mode.
1429
1430         PR c/84909
1431         * hsa-gen.c (mem_type_for_type): Fix comment typo.
1432         * tree-vect-loop-manip.c (vect_create_cond_for_niters_checks):
1433         Likewise.
1434         * gimple-ssa-warn-restrict.c (builtin_memref::set_base_and_offset):
1435         Likewise.
1436
1437 2018-03-16  Vladimir Makarov  <vmakarov@redhat.com>
1438
1439         PR target/84876
1440         * lra-assigns.c (lra_split_hard_reg_for): Don't use
1441         regno_allocno_class_array and sorted_pseudos.
1442         * lra-constraints.c (spill_hard_reg_in_range): Ignore hard regs in
1443         insns where regno is used.
1444
1445 2018-03-16  Martin Liska  <mliska@suse.cz>
1446
1447         PR ipa/84833
1448         * multiple_target.c (create_dispatcher_calls): Redirect
1449         reference in the symbol table.
1450
1451 2018-03-16  Martin Liska  <mliska@suse.cz>
1452
1453         PR ipa/84722
1454         * multiple_target.c (create_dispatcher_calls): Redirect also
1455         an alias.
1456
1457 2018-03-16  Jakub Jelinek  <jakub@redhat.com>
1458
1459         PR c++/79937
1460         PR c++/82410
1461         * tree.h (TARGET_EXPR_NO_ELIDE): Define.
1462         * gimplify.c (gimplify_modify_expr_rhs): Don't elide TARGET_EXPRs with
1463         TARGET_EXPR_NO_ELIDE flag set unless *expr_p is INIT_EXPR.
1464
1465 2018-03-16  Julia Koval  <julia.koval@intel.com>
1466
1467         * doc/invoke.texi (Skylake Server): Add CLWB.
1468         Cannonlake): Remove CLWB.
1469
1470 2018-03-16  Jakub Jelinek  <jakub@redhat.com>
1471
1472         PR tree-optimization/84841
1473         * tree-ssa-reassoc.c (INTEGER_CONST_TYPE): Change to 1 << 4 from
1474         1 << 3.
1475         (FLOAT_ONE_CONST_TYPE): Define.
1476         (constant_type): Return FLOAT_ONE_CONST_TYPE for -1.0 and 1.0.
1477         (sort_by_operand_rank): Put entries with higher constant_type last
1478         rather than first to match comments.
1479
1480 2018-03-15  Sandra Loosemore  <sandra@codesourcery.com>
1481
1482         * config/nios2/nios2.md (movsi_internal): Fix thinko in
1483         split predicate.
1484
1485 2018-03-15  Jakub Jelinek  <jakub@redhat.com>
1486
1487         PR c++/79085
1488         * calls.c (expand_call): For TREE_ADDRESSABLE rettype ignore alignment
1489         check and use address of target always.
1490
1491 2018-03-15  H.J. Lu  <hongjiu.lu@intel.com>
1492
1493         PR target/84574
1494         * config/i386/i386.c (indirect_thunk_needed): Update comments.
1495         (indirect_thunk_bnd_needed): Likewise.
1496         (indirect_thunks_used): Likewise.
1497         (indirect_thunks_bnd_used): Likewise.
1498         (indirect_return_needed): New.
1499         (indirect_return_bnd_needed): Likewise.
1500         (output_indirect_thunk_function): Add a bool argument for
1501         function return.
1502         (output_indirect_thunk_function): Don't generate alias for
1503         function return thunk.
1504         (ix86_code_end): Call output_indirect_thunk_function to generate
1505         function return thunks.
1506         (ix86_output_function_return): Set indirect_return_bnd_needed
1507         and indirect_return_needed instead of indirect_thunk_bnd_needed
1508         and indirect_thunk_needed.
1509
1510 2018-03-15  Olga Makhotina  <olga.makhotina@intel.com>
1511
1512         * config/i386/sgxintrin.h (_enclv_u32): New intrinsic.
1513         (__enclv_bc, __enclv_cd, __enclv_generic): New definitions.
1514         (ERDINFO, ETRACKC, ELDBC, ELDUC): New leaves.
1515
1516 2018-03-15  David Malcolm  <dmalcolm@redhat.com>
1517             Paul Hua <paul.hua.gm@gmail.com>
1518
1519         PR c/84852
1520         * gcc.dg/fixits-pr84852-1.c: Fix filename in dg-regexp.
1521
1522 2018-03-15  Segher Boessenkool  <segher@kernel.crashing.org>
1523
1524         * config/rs6000/rs6000.c (abi_v4_pass_in_fpr): Add back the
1525         TARGET_DOUBLE_FLOAT and TARGET_SINGLE_FLOAT conditions on the DFmode
1526         resp. SFmode cases.
1527
1528 2018-03-15  Tamar Christina  <tamar.christina@arm.com>
1529
1530         PR target/84711
1531         * config/arm/arm.c (arm_can_change_mode_class): Use GET_MODE_UNIT_SIZE
1532         instead of GET_MODE_SIZE when comparing Units.
1533
1534 2018-03-15  Vladimir Mezentsev  <vladimir.mezentsev@oracle.com>
1535
1536         PR target/68256
1537         * varasm.c (hash_section): Return an unchangeble hash value
1538         * config/aarch64/aarch64.c (aarch64_use_blocks_for_constant_p):
1539         Return !aarch64_can_use_per_function_literal_pools_p ().
1540
1541 2018-03-15  Jakub Jelinek  <jakub@redhat.com>
1542
1543         PR target/84860
1544         * optabs.c (emit_conditional_move): Pass address of cmode's copy
1545         rather than address of cmode as last argument to prepare_cmp_insn.
1546
1547 2018-03-15  Julia Koval  <julia.koval@intel.com>
1548
1549         * config/i386/i386.c (F_AVX512VBMI2, F_GFNI, F_VPCLMULQDQ,
1550         F_AVX512VNNI, F_AVX512BITALG): New.
1551
1552 2018-03-14  John David Anglin  <danglin@gcc.gnu.org>
1553
1554         PR target/83451
1555         * config/pa/pa.c (pa_emit_move_sequence):  Always emit secondary reload
1556         insn for floating-point loads and stores.
1557
1558 2018-03-14  Carl Love  <cel@us.ibm.com>
1559
1560         * config/rs6000/rs6000-c.c: Add macro definitions for
1561         ALTIVEC_BUILTIN_VEC_PERMXOR.
1562         * config/rs6000/rs6000.h: Add #define for vec_permxor builtin.
1563         * config/rs6000/rs6000-builtin.def: Add macro expansions for VPERMXOR.
1564         * config/rs6000/altivec.md (altivec_vpermxor): New define expand.
1565         * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Add case
1566         UNSPEC_VPERMXOR.
1567         * config/doc/extend.texi: Add prototypes for vec_permxor.
1568
1569 2018-03-14  David Malcolm  <dmalcolm@redhat.com>
1570
1571         PR c/84852
1572         * diagnostic-show-locus.c (class layout_point): Convert m_line
1573         from int to linenum_type.
1574         (line_span::comparator): Use linenum "compare" function when
1575         comparing line numbers.
1576         (test_line_span): New function.
1577         (layout_range::contains_point): Convert param "row" from int to
1578         linenum_type.
1579         (layout_range::intersects_line_p): Likewise.
1580         (layout::will_show_line_p): Likewise.
1581         (layout::print_source_line): Likewise.
1582         (layout::should_print_annotation_line_p): Likewise.
1583         (layout::print_annotation_line): Likewise.
1584         (layout::print_leading_fixits): Likewise.
1585         (layout::annotation_line_showed_range_p): Likewise.
1586         (struct line_corrections): Likewise for field m_row.
1587         (line_corrections::line_corrections): Likewise for param "row".
1588         (layout::print_trailing_fixits): Likewise.
1589         (layout::get_state_at_point): Likewise.
1590         (layout::get_x_bound_for_row): Likewise.
1591         (layout::print_line): Likewise.
1592         (diagnostic_show_locus): Likewise for locals "last_line" and
1593         "row".
1594         (selftest::diagnostic_show_locus_c_tests): Call test_line_span.
1595         * input.c (selftest::test_linenum_comparisons): New function.
1596         (selftest::input_c_tests): Call it.
1597         * selftest.c (selftest::test_assertions): Test ASSERT_GT,
1598         ASSERT_GT_AT, ASSERT_LT, and ASSERT_LT_AT.
1599         * selftest.h (ASSERT_GT): New macro.
1600         (ASSERT_GT_AT): New macro.
1601         (ASSERT_LT): New macro.
1602         (ASSERT_LT_AT): New macro.
1603
1604 2018-03-14  Segher Boessenkool  <segher@kernel.crashing.org>
1605
1606         PR rtl-optimization/84780
1607         * combine.c (distribute_links): Don't make a link based on pc_rtx.
1608
1609 2018-03-14  Martin Liska  <mliska@suse.cz>
1610
1611         * tree.c (record_node_allocation_statistics): Use
1612         get_stats_node_kind.
1613         (get_stats_node_kind): New function extracted from
1614         record_node_allocation_statistics.
1615         (free_node): Use get_stats_node_kind.
1616
1617 2018-03-14  Richard Biener  <rguenther@suse.de>
1618
1619         * tree-ssa-pre.c (compute_antic_aux): Remove code that asserts
1620         that the value-set of ANTIC_IN doesn't grow.
1621
1622         Revert
1623         * tree-ssa-pre.c (struct bb_bitmap_sets): Add visited_with_visited_succs
1624         member.
1625         (BB_VISITED_WITH_VISITED_SUCCS): New define.
1626         (compute_antic): Initialize BB_VISITED_WITH_VISITED_SUCCS.
1627
1628 2018-03-14  Julia Koval  <julia.koval@intel.com>
1629
1630         * config.gcc (icelake-client, icelake-server): New.
1631         (icelake): Remove.
1632         * config/i386/i386.c (initial_ix86_tune_features): Extend to 64 bit.
1633         (initial_ix86_arch_features): Ditto.
1634         (PTA_SKYLAKE): Add SGX.
1635         (PTA_ICELAKE): Remove.
1636         (PTA_ICELAKE_CLIENT): New.
1637         (PTA_ICELAKE_SERVER): New.
1638         (ix86_option_override_internal): Split up icelake on icelake client and
1639         icelake server.
1640         (get_builtin_code_for_version): Ditto.
1641         (fold_builtin_cpu): Ditto.
1642         * config/i386/driver-i386.c (config/i386/driver-i386.c): Ditto.
1643         * config/i386/i386-c.c (ix86_target_macros_internal): Ditto
1644         * config/i386/i386.h (processor_type): Ditto.
1645         * doc/invoke.texi: Ditto.
1646
1647 2018-03-14  Jakub Jelinek  <jakub@redhat.com>
1648
1649         PR sanitizer/83392
1650         * sanopt.c (maybe_optimize_ubsan_ptr_ifn): Handle also
1651         INTEGER_CST offset, add it together with bitpos / 8 and
1652         sign extend based on POINTER_SIZE.
1653
1654         PR target/84844
1655         Revert
1656         2017-04-20  Uros Bizjak  <ubizjak@gmail.com>
1657
1658         PR target/78090
1659         * config/i386/constraints.md (Yc): New register constraint.
1660         * config/i386/i386.md (*float<SWI48:mode><MODEF:mode>2_mixed):
1661         Use Yc constraint for alternative 2 of operand 0.  Remove
1662         preferred_for_speed attribute.
1663
1664 2018-03-14  Richard Biener  <rguenther@suse.de>
1665
1666         PR tree-optimization/84830
1667         * tree-ssa-pre.c (compute_antic_aux): Intersect the new ANTIC_IN
1668         with the old one to avoid oscillations.
1669
1670 2018-03-13  Vladimir Makarov  <vmakarov@redhat.com>
1671
1672         PR target/83712
1673         * lra-assigns.c (find_all_spills_for): Ignore uninteresting
1674         pseudos.
1675         (assign_by_spills): Return a flag of reload assignment failure.
1676         Do not process the reload assignment failures.  Do not spill other
1677         reload pseudos if they has the same reg class.  Update n if
1678         necessary.
1679         (lra_assign): Add a return arg.  Set up from the result of
1680         assign_by_spills call.
1681         (find_reload_regno_insns, lra_split_hard_reg_for): New functions.
1682         * lra-constraints.c (split_reg): Add a new arg.  Use it instead of
1683         usage_insns if it is not NULL.
1684         (spill_hard_reg_in_range): New function.
1685         (split_if_necessary, inherit_in_ebb): Pass a new arg to split_reg.
1686         * lra-int.h (spill_hard_reg_in_range, lra_split_hard_reg_for): New
1687         function prototypes.
1688         (lra_assign): Change prototype.
1689         * lra.c (lra): Add code to deal with fails by splitting hard reg
1690         live ranges.
1691
1692 2018-03-01  Palmer Dabbelt  <palmer@sifive.com>
1693
1694         * config/riscv/riscv.opt (mrelax): New option.
1695         * config/riscv/riscv.c (riscv_file_start): Emit ".option
1696         "norelax" when riscv_mrelax is disabled.
1697         * doc/invoke.texi (RISC-V): Document "-mrelax" and "-mno-relax".
1698
1699 2018-03-13  Aaron Sawdey  <acsawdey@linux.vnet.ibm.com>
1700
1701         PR target/84743
1702         * config/rs6000/rs6000.c (rs6000_reassociation_width): Disable parallel
1703         reassociation for int modes.
1704
1705 2018-03-13  Richard Sandiford  <richard.sandiford@linaro.org>
1706
1707         * tree-vect-loop-manip.c (vect_maybe_permute_loop_masks):
1708         Reverse the choice between VEC_UNPACK_LO_EXPR and VEC_UNPACK_HI_EXPR
1709         for big-endian.
1710         * config/aarch64/iterators.md (hi_lanes_optab): New int attribute.
1711         * config/aarch64/aarch64-sve.md
1712         (*aarch64_sve_<perm_insn><perm_hilo><mode>): Rename to...
1713         (aarch64_sve_<perm_insn><perm_hilo><mode>): ...this.
1714         (*extend<mode><Vwide>2): Rename to...
1715         (aarch64_sve_extend<mode><Vwide>2): ...this.
1716         (vec_unpack<su>_<perm_hilo>_<mode>): Turn into a define_expand,
1717         renaming the old pattern to...
1718         (aarch64_sve_punpk<perm_hilo>_<mode>): ...this.  Only define
1719         unsigned packs.
1720         (vec_unpack<su>_<perm_hilo>_<SVE_BHSI:mode>): Turn into a
1721         define_expand, renaming the old pattern to...
1722         (aarch64_sve_<su>unpk<perm_hilo>_<SVE_BHSI:mode>): ...this.
1723         (*vec_unpacku_<perm_hilo>_<mode>_no_convert): Delete.
1724         (vec_unpacks_<perm_hilo>_<mode>): Take BYTES_BIG_ENDIAN into
1725         account when deciding which SVE instruction the optab should use.
1726         (vec_unpack<su_optab>_float_<perm_hilo>_vnx4si): Likewise.
1727
1728 2018-03-13  Richard Sandiford  <richard.sandiford@linaro.org>
1729
1730         * config/aarch64/aarch64.md (V4_REGNUM, V8_REGNUM, V12_REGNUM)
1731         (V20_REGNUM, V24_REGNUM, V28_REGNUM, P1_REGNUM, P2_REGNUM, P3_REGNUM)
1732         (P4_REGNUM, P5_REGNUM, P6_REGNUM, P8_REGNUM, P9_REGNUM, P10_REGNUM)
1733         (P11_REGNUM, P12_REGNUM, P13_REGNUM, P14_REGNUM): New define_constants.
1734         (tlsdesc_small_<mode>): Turn a define_expand and use
1735         tlsdesc_small_sve_<mode> for SVE.  Rename original define_insn to...
1736         (tlsdesc_small_advsimd_<mode>): ...this.
1737         (tlsdesc_small_sve_<mode>): New pattern.
1738
1739 2018-03-13  Richard Sandiford  <richard.sandiford@linaro.org>
1740
1741         * config/aarch64/iterators.md (UNSPEC_SMUL_HIGHPART)
1742         (UNSPEC_UMUL_HIGHPART): New constants.
1743         (MUL_HIGHPART): New int iteraor.
1744         (su): Handle UNSPEC_SMUL_HIGHPART and UNSPEC_UMUL_HIGHPART.
1745         * config/aarch64/aarch64-sve.md (<su>mul<mode>3_highpart): New
1746         define_expand.
1747         (*<su>mul<mode>3_highpart): New define_insn.
1748
1749 2018-03-13  Eric Botcazou  <ebotcazou@adacore.com>
1750
1751         PR lto/84805
1752         * ipa-devirt.c (odr_subtypes_equivalent_p): Do not get the ODR type of
1753         incomplete types.
1754
1755 2018-03-13  Martin Liska  <mliska@suse.cz>
1756
1757         PR ipa/84658.
1758         * (sem_item_optimizer::sem_item_optimizer): Initialize new
1759         vector.
1760         (sem_item_optimizer::~sem_item_optimizer): Release it.
1761         (sem_item_optimizer::merge_classes): Register variable aliases.
1762         (sem_item_optimizer::fixup_pt_set): New function.
1763         (sem_item_optimizer::fixup_points_to_sets): Likewise.
1764         * ipa-icf.h: Declare new variables and functions.
1765
1766 2018-03-13  Jakub Jelinek  <jakub@redhat.com>
1767
1768         PR middle-end/84834
1769         * match.pd ((A & C) != 0 ? D : 0): Use INTEGER_CST@2 instead of
1770         integer_pow2p@2 and test integer_pow2p in condition.
1771         (A < 0 ? C : 0): Similarly for @1.
1772
1773         PR middle-end/84831
1774         * stmt.c (parse_output_constraint): If the CONSTRAINT_LEN (*p, p)
1775         characters starting at p contain '\0' character, don't look beyond
1776         that.
1777
1778         PR target/84827
1779         * config/i386/i386.md (round<mode>2): For 387 fancy math, disable
1780         pattern if -ftrapping-math -fno-fp-int-builtin-inexact.
1781
1782         PR target/84828
1783         * reg-stack.c (change_stack): Change update_end var from int to
1784         rtx_insn *, if non-NULL don't update just BB_END (current_block), but
1785         also call set_block_for_insn on the newly added insns and rescan.
1786
1787         PR target/84786
1788         * config/i386/sse.md (sse2_loadhpd): Use Yv constraint rather than v
1789         on the last operand.
1790
1791         PR c++/84704
1792         * tree.c (stabilize_reference_1): Return save_expr (e) for
1793         STATEMENT_LIST even if it doesn't have side-effects.
1794
1795 2018-03-12  Jonathan Wakely  <jwakely@redhat.com>
1796
1797         * doc/invoke.texi (-mclflushopt): Fix spelling of option.
1798
1799 2018-03-12  Renlin Li  <renlin.li@arm.com>
1800
1801         * config/aarch64/aarch64.md (movhf_aarch64): Fix mode argument to
1802         aarch64_output_scalar_simd_mov_immediate.
1803
1804 2018-03-12  Martin Sebor  <msebor@redhat.com>
1805
1806         PR tree-optimization/83456
1807         * gimple-fold.c (gimple_fold_builtin_memory_op): Avoid warning
1808         for perfectly overlapping calls to memcpy.
1809         (gimple_fold_builtin_memory_chk): Same.
1810         (gimple_fold_builtin_strcpy): Handle no-warning.
1811         (gimple_fold_builtin_stxcpy_chk): Same.
1812         * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Handle no-warning.
1813
1814 2018-03-12  Segher Boessenkool  <segher@kernel.crashing.org>
1815
1816         * config/rs6000/rs6000.c (abi_v4_pass_in_fpr): Add bool "named"
1817         parameter.  Use it for SFmode.
1818         (rs6000_function_arg_advance_1): Adjust.
1819         (rs6000_function_arg): Adjust.
1820         (rs6000_gimplify_va_arg): Pass false for that new parameter.
1821
1822 2018-03-12  Segher Boessenkool  <segher@kernel.crashing.org>
1823
1824         PR rtl-optimization/84169
1825         PR rtl-optimization/84780
1826         * combine.c (can_combine_p): Check for a 2-insn combination whether
1827         the destination register is used between the two insns, too.
1828
1829 2018-03-12  Richard Biener  <rguenther@suse.de>
1830
1831         PR tree-optimization/84803
1832         * tree-if-conv.c (ifcvt_memrefs_wont_trap): Don't do anything
1833         for refs DR analysis didn't process.
1834
1835 2018-03-12  Richard Biener  <rguenther@suse.de>
1836
1837         PR tree-optimization/84777
1838         * tree-ssa-loop-ch.c (should_duplicate_loop_header_p): For
1839         force-vectorize loops ignore whether we are optimizing for size.
1840
1841 2018-03-12  Chung-Ju Wu  <jasonwucj@gmail.com>
1842
1843         * config/nds32/nds32.c (nds32_md_asm_adjust): New function.
1844         (TARGET_MD_ASM_ADJUST): Define.
1845
1846 2018-03-12  Monk Chiang  <sh.chiang04@gmail.com>
1847             Kito Cheng  <kito.cheng@gmail.com>
1848             Chung-Ju Wu  <jasonwucj@gmail.com>
1849
1850         * config/nds32/nds32.c (nds32_compute_stack_frame,
1851         nds32_emit_stack_push_multiple, nds32_emit_stack_pop_multiple,
1852         nds32_emit_stack_v3push, nds32_emit_stack_v3pop,
1853         nds32_emit_adjust_frame, nds32_expand_prologue, nds32_expand_epilogue,
1854         nds32_expand_prologue_v3push, nds32_expand_epilogue_v3pop): Refine.
1855         * config/nds32/nds32.h (NDS32_FIRST_CALLEE_SAVE_GPR_REGNUM,
1856         NDS32_LAST_CALLEE_SAVE_GPR_REGNUM, NDS32_V3PUSH_AVAILABLE_P): New.
1857         * config/nds32/nds32.md (prologue, epilogue): Use macro
1858         NDS32_V3PUSH_AVAILABLE_P to do checking.
1859
1860 2018-03-11  Jakub Jelinek  <jakub@redhat.com>
1861
1862         PR debug/58150
1863         * dwarf2out.c (gen_enumeration_type_die): Don't guard adding
1864         DW_AT_declaration for ENUM_IS_OPAQUE on -gdwarf-4 or -gno-strict-dwarf,
1865         but on TYPE_SIZE.  Don't do anything for ENUM_IS_OPAQUE if not creating
1866         a new die.  Don't set TREE_ASM_WRITTEN if ENUM_IS_OPAQUE.  Guard
1867         addition of most attributes on !orig_type_die or the attribute not
1868         being present already.  Assert TYPE_VALUES is NULL for ENUM_IS_OPAQUE.
1869
1870 2018-03-11  Kito Cheng  <kito.cheng@gmail.com>
1871             Chung-Ju Wu  <jasonwucj@gmail.com>
1872
1873         * config/nds32/nds32.c (nds32_cpu_cpp_builtins): Modify to define
1874         __NDS32_VH__ macro.
1875         * config/nds32/nds32.opt (mvh): New option.
1876
1877 2018-03-11  Kito Cheng  <kito.cheng@gmail.com>
1878             Chung-Ju Wu  <jasonwucj@gmail.com>
1879
1880         * config/nds32/nds32-protos.h (nds32_cpu_cpp_builtins): Declare
1881         function.
1882         * config/nds32/nds32.c (nds32_cpu_cpp_builtins): New function.
1883         * config/nds32/nds32.h (TARGET_CPU_CPP_BUILTINS): Modify its
1884         definition.
1885
1886 2018-03-11  Kito Cheng  <kito.cheng@gmail.com>
1887             Chung-Ju Wu  <jasonwucj@gmail.com>
1888
1889         * config/nds32/nds32-memory-manipulation.c (nds32_expand_strlen): New
1890         function.
1891         * config/nds32/nds32-multiple.md (strlensi): New pattern.
1892         * config/nds32/nds32-protos.h (nds32_expand_strlen): Declare function.
1893
1894 2018-03-11  Monk Chiang  <sh.chiang04@gmail.com>
1895             Kito Cheng  <kito.cheng@gmail.com>
1896             Chung-Ju Wu  <jasonwucj@gmail.com>
1897
1898         * config/nds32/constants.md (unspec_element): Add UNSPEC_FFB,
1899         UNSPEC_FFMISM and UNSPEC_FLMISM.
1900         * config/nds32/nds32-intrinsic.c (bdesc_2arg): Add builtin description
1901         for ffb, ffmism and flmism.
1902         * config/nds32/nds32-intrinsic.md (unspec_ffb): Define new pattern.
1903         (unspec_ffmism): Ditto.
1904         (unspec_flmism): Ditto.
1905         (nds32_expand_builtin_impl): Check if string extension is available.
1906         * config/nds32/nds32.h (nds32_builtins): Add NDS32_BUILTIN_FFB,
1907         NDS32_BUILTIN_FFMISM and NDS32_BUILTIN_FLMISM.
1908
1909 2018-03-10  Vladimir Makarov  <vmakarov@redhat.com>
1910
1911         Reverting patch:
1912         2018-03-09  Vladimir Makarov  <vmakarov@redhat.com>
1913
1914         PR target/83712
1915         * lra-assigns.c (assign_by_spills): Return a flag of reload
1916         assignment failure.  Do not process the reload assignment
1917         failures.  Do not spill other reload pseudos if they has the same
1918         reg class.
1919         (lra_assign): Add a return arg.  Set up from the result of
1920         assign_by_spills call.
1921         (find_reload_regno_insns, lra_split_hard_reg_for): New functions.
1922         * lra-constraints.c (split_reg): Add a new arg.  Use it instead of
1923         usage_insns if it is not NULL.
1924         (spill_hard_reg_in_range): New function.
1925         (split_if_necessary, inherit_in_ebb): Pass a new arg to split_reg.
1926         * lra-int.h (spill_hard_reg_in_range, lra_split_hard_reg_for): New
1927         function prototypes.
1928         (lra_assign): Change prototype.
1929         * lra.c (lra): Add code to deal with fails by splitting hard reg
1930         live ranges.
1931
1932 2018-03-10  H.J. Lu  <hongjiu.lu@intel.com>
1933
1934         PR target/84807
1935         * config/i386/i386.opt: Replace Enforcment with Enforcement.
1936
1937 2018-03-10  Alexandre Oliva  <aoliva@redhat.com>
1938
1939         PR debug/84620
1940         * dwarf2out.h (dw_val_class): Add dw_val_class_symview.
1941         (dw_val_node): Add val_symbolic_view.
1942         * dwarf2out.c (dw_line_info_table): Add symviews_since_reset.
1943         (symview_upper_bound): New.
1944         (new_line_info_table): Initialize symviews_since_reset.
1945         (dwarf2out_source_line): Count symviews_since_reset and set
1946         symview_upper_bound.
1947         (dw_val_equal_p): Handle symview.
1948         (add_AT_symview): New.
1949         (print_dw_val): Handle symview.
1950         (attr_checksum, attr_checksum_ordered): Likewise.
1951         (same_dw_val_p, size_of_die): Likewise.
1952         (value_format, output_die): Likewise.
1953         (add_high_low_attributes): Use add_AT_symview for entry_view.
1954         (dwarf2out_finish): Reset symview_upper_bound, clear
1955         zero_view_p.
1956
1957 2018-03-09  Peter Bergner  <bergner@vnet.ibm.com>
1958
1959         PR target/83969
1960         * config/rs6000/rs6000.c (rs6000_offsettable_memref_p): New prototype.
1961         Add strict argument and use it.
1962         (rs6000_split_multireg_move): Update for new strict argument.
1963         (mem_operand_gpr): Disallow all non-offsettable addresses.
1964         * config/rs6000/rs6000.md (*movdi_internal64): Use YZ constraint.
1965
1966 2018-03-09  Jakub Jelinek  <jakub@redhat.com>
1967
1968         PR target/84772
1969         * config/rs6000/rs6000.c (rs6000_gimplify_va_arg): Mark va_arg_tmp
1970         temporary TREE_ADDRESSABLE before gimplification of BUILT_IN_MEMCPY.
1971         * config/powerpcspe/powerpcspe.c (rs6000_gimplify_va_arg): Likewise.
1972
1973         PR c++/84767
1974         * tree-inline.c (copy_tree_body_r): For INDIRECT_REF of a remapped
1975         decl, use remap_type if we want to use the type.
1976
1977 2018-03-09  Martin Sebor  <msebor@redhat.com>
1978
1979         PR tree-optimization/84526
1980         * gimple-ssa-warn-restrict.c (builtin_memref::set_base_and_offset):
1981         Remove dead code.
1982         (builtin_access::generic_overlap): Be prepared to handle non-array
1983         base objects.
1984
1985 2018-03-09  Alexandre Oliva  <aoliva@redhat.com>
1986
1987         PR rtl-optimization/84682
1988         * lra-constraints.c (process_address_1): Check is_address flag
1989         for address constraints.
1990         (process_alt_operands): Likewise.
1991         * lra.c (lra_set_insn_recog_data): Pass asm operand locs to
1992         preprocess_constraints.
1993         * recog.h (preprocess_constraints): Add oploc parameter.
1994         Adjust callers.
1995         * recog.c (preprocess_constraints): Test address_operand for
1996         CT_ADDRESS constraints.
1997
1998 2018-03-09  Vladimir Makarov  <vmakarov@redhat.com>
1999
2000         PR target/83712
2001         * lra-assigns.c (assign_by_spills): Return a flag of reload
2002         assignment failure.  Do not process the reload assignment
2003         failures.  Do not spill other reload pseudos if they has the same
2004         reg class.
2005         (lra_assign): Add a return arg.  Set up from the result of
2006         assign_by_spills call.
2007         (find_reload_regno_insns, lra_split_hard_reg_for): New functions.
2008         * lra-constraints.c (split_reg): Add a new arg.  Use it instead of
2009         usage_insns if it is not NULL.
2010         (spill_hard_reg_in_range): New function.
2011         (split_if_necessary, inherit_in_ebb): Pass a new arg to split_reg.
2012         * lra-int.h (spill_hard_reg_in_range, lra_split_hard_reg_for): New
2013         function prototypes.
2014         (lra_assign): Change prototype.
2015         * lra.c (lra): Add code to deal with fails by splitting hard reg
2016         live ranges.
2017
2018 2018-03-09  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
2019
2020         PR target/83193
2021         * common/config/arm/arm-common.c (arm_parse_arch_option_name):
2022         Accept complain bool parameter.  Only emit errors if it is true.
2023         (arm_parse_cpu_option_name): Likewise.
2024         (arm_target_thumb_only): Adjust callers of the above.
2025         * config/arm/arm-protos.h (arm_parse_cpu_option_name): Adjust
2026         prototype to take a default true bool parameter.
2027         (arm_parse_arch_option_name): Likewise.
2028
2029 2018-03-09  David Malcolm  <dmalcolm@redhat.com>
2030             Francois-Xavier Coudert  <fxcoudert@gcc.gnu.org>
2031
2032         PR jit/64089
2033         PR jit/84288
2034         * Makefile.in (LD_VERSION_SCRIPT_OPTION, LD_SONAME_OPTION): New.
2035         * configure: Regenerate.
2036         * configure.ac ("linker --version-script option"): New.
2037         ("linker soname option"): New.
2038
2039 2018-03-09  Richard Biener  <rguenther@suse.de>
2040
2041         PR tree-optimization/84775
2042         * tree-if-conv.c (add_bb_predicate_gimplified_stmts): Delink
2043         immediate uses of predicate stmts and mark them modified.
2044
2045         Revert
2046         PR tree-optimization/84178
2047         * tree-if-conv.c (combine_blocks): Move insert_gimplified_predicates
2048         to caller.
2049         (version_loop_for_if_conversion): Delay update_ssa call.
2050         (tree_if_conversion): Delay update_ssa until after predicate
2051         insertion.
2052
2053 2018-03-09  Eric Botcazou  <ebotcazou@adacore.com>
2054
2055         PR target/84763
2056         * config/i386/winnt.c (i386_pe_seh_cold_init): Use small pre-allocation
2057         when the function accesses prior frames.
2058
2059 2018-03-08  Jakub Jelinek  <jakub@redhat.com>
2060
2061         PR debug/84456
2062         * dwarf2out.c (dw_loc_list): If list && loc_list->first->next, call
2063         gen_llsym, otherwise call maybe_gen_llsym.
2064
2065         PR inline-asm/84742
2066         * recog.c (asm_operand_ok): Return 0 if multi-character constraint
2067         has ',' character inside of it.
2068
2069 2018-03-08  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
2070
2071         PR target/84748
2072         * config/aarch64/aarch64.md (*compare_cstore<mode>_insn): Mark pattern
2073         as clobbering CC_REGNUM.
2074
2075 2018-03-08  Richard Biener  <rguenther@suse.de>
2076
2077         PR middle-end/84552
2078         * tree-scalar-evolution.c: Include tree-into-ssa.h.
2079         (follow_copies_to_constant): Do not follow SSA names registered
2080         for update.
2081
2082 2018-03-08  Richard Biener  <rguenther@suse.de>
2083
2084         PR tree-optimization/84178
2085         * tree-if-conv.c (combine_blocks): Move insert_gimplified_predicates
2086         to caller.
2087         (version_loop_for_if_conversion): Delay update_ssa call.
2088         (tree_if_conversion): Delay update_ssa until after predicate
2089         insertion.
2090
2091 2018-03-08  David Malcolm  <dmalcolm@redhat.com>
2092
2093         PR tree-optimization/84178
2094         * tree-if-conv.c (release_bb_predicate): Remove the
2095         the assertion that the stmts have NULL use_ops.
2096         Discard the statements, asserting that they haven't
2097         yet been added to a BB.
2098
2099 2018-03-08  Richard Biener  <rguenther@suse.de>
2100
2101         PR tree-optimization/84746
2102         * tree-ssa-pre.c (find_leader_in_sets): Deal with SET1 being NULL.
2103         (phi_translate): Pass in destination ANTIC_OUT set.
2104         (phi_translate_1): Likewise.  For a simplified result lookup
2105         a leader in ANTIC_OUT and AVAIL_OUT, not the ANTIC_IN sets.
2106         (phi_translate_set): Adjust.
2107         (do_pre_regular_insertion): Likewise.
2108         (do_pre_partial_partial_insertion): Likewise.
2109
2110 2018-03-08  Martin Liska  <mliska@suse.cz>
2111
2112         PR gcov-profile/84735
2113         * doc/gcov.texi: Document usage of profile files.
2114         * gcov-io.h: Document changes in the format.
2115
2116 2018-03-08  Alexandre Oliva  <aoliva@redhat.com>
2117
2118         PR debug/84404
2119         PR debug/84408
2120         * dwarf2out.c (struct dw_line_info_table): Update comments for
2121         view == -1.
2122         (FORCE_RESET_NEXT_VIEW): New.
2123         (FORCE_RESETTING_VIEW_P): New.
2124         (RESETTING_VIEW_P): Check for -1 too.
2125         (ZERO_VIEW_P): Likewise.
2126         (new_line_info_table): Force-reset next view.
2127         (dwarf2out_begin_function): Likewise.
2128         (dwarf2out_source_line): Simplify zero_view_p initialization.
2129         Test FORCE_RESETTING_VIEW_P and RESETTING_VIEW_P instead of
2130         view directly.  Omit view when omitting .loc at line 0.
2131
2132 2018-03-08  Jakub Jelinek  <jakub@redhat.com>
2133
2134         PR tree-optimization/84740
2135         * tree-switch-conversion.c (process_switch): Call build_constructors
2136         only if info.phi_count is non-zero.
2137
2138         PR tree-optimization/84739
2139         * tree-tailcall.c (find_tail_calls): Check call arguments against
2140         DECL_ARGUMENTS (current_function_decl) rather than
2141         DECL_ARGUMENTS (func) when checking for tail recursion.
2142
2143 2018-03-07  Jakub Jelinek  <jakub@redhat.com>
2144
2145         * doc/contrib.texi: Add entries for Martin Liska, David Malcolm,
2146         Marek Polacek, extend Vladimir Makarov's, Jonathan Wakely's and
2147         Volker Reichelt's entry and add entries for people that perform
2148         GCC fuzzy testing and report numerous bugs.
2149
2150 2018-03-07  Segher Boessenkool  <segher@kernel.crashing.org>
2151
2152         PR target/82411
2153         * config/rs6000/rs6000.c (rs6000_elf_in_small_data_p): Don't put
2154         readonly data in sdata, if that is disabled.
2155         * config/rs6000/sysv4.opt (mreadonly-in-sdata): New option.
2156         * doc/invoke.texi (RS/6000 and PowerPC Options): Document
2157         -mreadonly-in-sdata option.
2158
2159 2018-03-07  Martin Sebor  <msebor@redhat.com>
2160
2161         PR tree-optimization/84468
2162         * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Consider successor
2163         basic block when looking for nul assignment.
2164
2165 2018-03-07  Eric Botcazou  <ebotcazou@adacore.com>
2166
2167         PR target/84277
2168         * except.h (output_function_exception_table): Adjust prototype.
2169         * except.c (output_function_exception_table): Remove FNNAME parameter
2170         and add SECTION parameter.  Ouput one part of the table at a time.
2171         * final.c (final_scan_insn_1) <NOTE_INSN_SWITCH_TEXT_SECTIONS>: Output
2172         the first part of the exception table and emit unwind directives.
2173         * config/i386/i386-protos.h (i386_pe_end_cold_function): Declare.
2174         (i386_pe_seh_cold_init): Likewise.
2175         * config/i386/cygming.h (ASM_DECLARE_COLD_FUNCTION_NAME): New macro.
2176         (ASM_DECLARE_COLD_FUNCTION_SIZE): Likewise.
2177         * config/i386/i386.c (x86_expand_epilogue): Fix wording in comment.
2178         (ix86_output_call_insn): Emit a nop in one more case for SEH.
2179         * config/i386/winnt.c: Include except.h.
2180         (struct seh_frame_state): Add reg_offset, after_prologue and
2181         in_cold_section fields.
2182         (i386_pe_seh_end_prologue): Set seh->after_prologue.
2183         (i386_pe_seh_cold_init): New function.
2184         (i386_pe_seh_fini): Add COLD parameter and bail out if it is not equal
2185         to seh->in_cold_section.
2186         (seh_emit_push): Record the offset of the push.
2187         (seh_emit_save): Record the offet of the save.
2188         (i386_pe_seh_unwind_emit): Deal with NOTE_INSN_SWITCH_TEXT_SECTIONS.
2189         Test seh->after_prologue to disregard the epilogue.
2190         (i386_pe_end_function): Pass FALSE to i386_pe_seh_fini.
2191         (i386_pe_end_cold_function): New function.
2192
2193 2018-03-07  Jakub Jelinek  <jakub@redhat.com>
2194
2195         PR fortran/84565
2196         * config/aarch64/predicates.md (aarch64_simd_reg_or_zero): Use
2197         aarch64_simd_or_scalar_imm_zero rather than aarch64_simd_imm_zero.
2198
2199         PR c++/84704
2200         * gimple-expr.c (create_tmp_var_raw): Set DECL_NAMELESS flag
2201         on tmp_var.
2202         * tree-pretty-print.c (dump_decl_name): For TDF_COMPARE_DEBUG,
2203         don't print names of DECL_NAMELESS DECL_IGNORED_P decls.
2204
2205         PR middle-end/84723
2206         * multiple_target.c: Include tree-inline.h and intl.h.
2207         (expand_target_clones): Diagnose and fail if node->definition and
2208         !tree_versionable_function_p (node->decl).
2209
2210 2018-03-06  John David Anglin  <danglin@gcc.gnu.org>
2211
2212         * config/pa/pa.h (ASM_GENERATE_INTERNAL_LABEL): Revise to use
2213         sprint_ul.
2214         (ASM_OUTPUT_ADDR_VEC_ELT): Revise for above change.
2215         (ASM_OUTPUT_ADDR_DIFF_ELT): Likewise.
2216         * config/pa/pa64-hpux.h (ASM_GENERATE_INTERNAL_LABEL): Revise as above.
2217
2218 2018-03-06  Jakub Jelinek  <jakub@redhat.com>
2219
2220         PR target/84710
2221         * combine.c (try_combine): Use reg_or_subregno instead of handling
2222         just paradoxical SUBREGs and REGs.
2223
2224 2018-03-06  Claudiu Zissulescu  <claziss@synopsys.com>
2225
2226          * config/arc/arc.c (arc_finalize_pic): Remove function.
2227          (arc_must_save_register): We use single base PIC register, remove
2228          checks to save/restore the PIC register.
2229          (arc_expand_prologue): Likewise.
2230          * config/arc/arc-protos.h (arc_set_default_type_attributes):
2231          Remove.
2232          (arc_verify_short): Likewise.
2233          (arc_attr_type): Likewise.
2234          * config/arc/arc.c (arc_set_default_type_attributes): Remove.
2235          (walk_stores): Likewise.
2236          (arc_address_cost): Make it static.
2237          (arc_verify_short): Likewise.
2238          (branch_dest): Likewise.
2239          (arc_attr_type): Likewise.
2240          * config/arc/arc.c (TARGET_ADJUST_INSN_LENGTH): Remove.
2241          (TARGET_INSN_LENGTH_PARAMETERS): Likewise.
2242          (arc_final_prescan_insn): Remove inserting the nops due to
2243          hardware hazards.  It is done in reorg step.
2244          (insn_length_variant_t): Remove.
2245          (insn_length_parameters_t): Likewise.
2246          (arc_insn_length_parameters): Likewise.
2247          (arc_get_insn_variants): Likewise.
2248          * config/arc/arc.h (TARGET_UPSIZE_DBR): Remove.
2249
2250 2018-03-06  Jakub Jelinek  <jakub@redhat.com>
2251
2252         PR inline-asm/84683
2253         * reg-stack.c (move_for_stack_reg): If any_malformed_asm, avoid
2254         assertion failure.
2255
2256         PR tree-optimization/84687
2257         * omp-simd-clone.c (simd_clone_create): Clear DECL_BUILT_IN_CLASS
2258         on new_node->decl.
2259         * match.pd (pow(C,x)*expN(y) -> expN(logN(C)*x+y)): New optimization.
2260
2261 2018-03-05  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
2262
2263         * config/rs6000/rs6000-builtin.def (rs6000_speculation_barrier):
2264         Rename to ppc_speculation_barrier.
2265         * config/rs6000/rs6000.c (rs6000_init_builtins): Rename builtin to
2266         __builtin_ppc_speculation_barrier.
2267
2268 2018-03-05  Jakub Jelinek  <jakub@redhat.com>
2269
2270         PR target/84700
2271         * combine.c (combine_simplify_rtx): Don't try to simplify if
2272         if_then_else_cond returned non-NULL, but either true_rtx or false_rtx
2273         are equal to x.
2274
2275 2018-03-05  Segher Boessenkool  <segher@kernel.crashing.org>
2276
2277         * config/rs6000/rs6000.c (rs6000_loop_align): Don't align tiny loops
2278         to 32 bytes when compiling for POWER9.
2279
2280 2018-03-05  Jakub Jelinek  <jakub@redhat.com>
2281
2282         PR target/84564
2283         * config/i386/i386.c (ix86_function_ok_for_sibcall): Check for
2284         regparm >= 3 with no arg reg available also for calls with
2285         flag_force_indirect_call.  Pass decl to ix86_function_regparm.
2286
2287         PR target/84524
2288         * config/i386/sse.md (*<code><mode>3): Replace <mask_prefix3> with
2289         orig,vex.
2290         (*<plusminus_insn><mode>3): Likewise.  Remove <mask_operand3> uses.
2291
2292 2018-03-05  Peter Bergner  <bergner@vnet.ibm.com>
2293
2294         PR target/84264
2295         * config/rs6000/vector.md (mov<mode>): Disallow altivec memory operands.
2296
2297 2018-03-05  Richard Biener  <rguenther@suse.de>
2298
2299         PR tree-optimization/84486
2300         * tree-ssa-pre.c (create_expression_by_pieces): Remove dead code.
2301         When inserting a __builtin_assume_aligned call set the LHS
2302         SSA name alignment info accordingly.
2303
2304 2018-03-05  Wilco Dijkstra  <wdijkstr@arm.com>
2305
2306         PR tree-optimization/84114
2307         * config/aarch64/aarch64.c (aarch64_reassociation_width)
2308         Avoid reassociation of FLOAT_MODE addition.
2309
2310 2018-03-05  Olga Makhotina  <olga.makhotina@intel.com>
2311
2312         * common/config/i386/i386-common.c (OPTION_MASK_ISA_PCONFIG_SET,
2313         OPTION_MASK_ISA_PCONFIG_UNSET, OPTION_MASK_ISA_WBNOINVD_SET,
2314         OPTION_MASK_ISA_WBNOINVD_UNSET): New definitions.
2315         (ix86_handle_option): Handle -mpconfig and -mwbnoinvd.
2316         * config.gcc (pconfigintrin.h, wbnoinvdintrin.h) : Add headers.
2317         * config/i386/cpuid.h (bit_PCONFIG, bit_WBNOINVD): New.
2318         * config/i386/driver-i386.c (host_detect_local_cpu): Detect -mpconfig
2319         and -mwbnoinvd.
2320         * config/i386/i386-builtin.def (__builtin_ia32_wbnoinvd,
2321         __builtin_ia32_wbinvd): New builtins.
2322         (SPECIAL_ARGS2): New.
2323         * config/i386/i386-c.c (__WBNOINVD__, __PCONFIG__): New.
2324         (SPECIAL_ARGS2): New.
2325         * config/i386/i386.c (ix86_target_string): Add -mpconfig and -mwbnoinvd.
2326         (ix86_valid_target_attribute_inner_p): Ditto.
2327         (ix86_init_mmx_sse_builtins): Add special_args2.
2328         * config/i386/i386.h (TARGET_PCONFIG, TARGET_PCONFIG_P, TARGET_WBNOINVD,
2329         TARGET_WBNOINVD_P): New.
2330         * config/i386/i386.md (UNSPECV_WBINVD, UNSPECV_WBNOINVD): New.
2331         (define_insn "wbinvd", define_insn "wbnoinvd"): New.
2332         * config/i386/i386.opt: Add -mpconfig and -mwbnoinvd.
2333         * config/i386/immintrin.h (_wbinvd): New intrinsic.
2334         * config/i386/pconfigintrin.h: New file.
2335         * config/i386/wbnoinvdintrin.h: Ditto.
2336         * config/i386/x86intrin.h: Add headers pconfigintrin.h and wbnoinvdintrin.h.
2337         * doc/invoke.texi (-mpconfig, -mwbnoinvd): New.
2338
2339 2018-03-05  Richard Biener  <rguenther@suse.de>
2340
2341         PR tree-optimization/84670
2342         * tree-ssa-pre.c (struct bb_bitmap_sets): Add visited_with_visited_succs
2343         member.
2344         (BB_VISITED_WITH_VISITED_SUCCS): New define.
2345         (compute_antic): Initialize BB_VISITED_WITH_VISITED_SUCCS.
2346         (compute_antic_aux): Only assert the number of values in ANTIC_IN
2347         doesn't grow if all successors (recursively) were visited at least
2348         once.
2349
2350 2018-03-05  Richard Biener  <rguenther@suse.de>
2351
2352         PR tree-optimization/84650
2353         * tree-ssa-loop-im.c (pass_lim::execute): Reset the SCEV cache
2354         if executed in the loop pipeline.
2355
2356 2018-03-05  Sandra Loosemore  <sandra@codesourcery.com>
2357
2358         * doc/configfiles.texi (Configuration Files): Move info about
2359         conditionalizing $target-protos.h to...
2360         * doc/sourcebuild.texi (Back End): Here.  Explain how $target.h
2361         differs from $target-protos.h.
2362
2363 2018-03-05  Kito Cheng  <kito.cheng@gmail.com>
2364             Chung-Ju Wu  <jasonwucj@gmail.com>
2365
2366         * config/nds32/nds32-protos.h (nds32_expand_setmem): Declare.
2367         * config/nds32/nds32-multiple.md (setmemsi): Define.
2368         * config/nds32/nds32-memory-manipulation.c
2369         (nds32_gen_dup_4_byte_to_word_value): New.
2370         (emit_setmem_word_loop): New.
2371         (emit_setmem_byte_loop): New.
2372         (nds32_expand_setmem_loop): New.
2373         (nds32_expand_setmem_loop_v3m): New.
2374         (nds32_expand_setmem_unroll): New.
2375         (nds32_expand_setmem): New.
2376
2377 2018-03-04  Kito Cheng  <kito.cheng@gmail.com>
2378             Chung-Ju Wu  <jasonwucj@gmail.com>
2379
2380         * config/nds32/nds32-memory-manipulation.c
2381         (nds32_emit_load_store): New.
2382         (nds32_emit_post_inc_load_store): New.
2383         (nds32_emit_mem_move): New.
2384         (nds32_emit_mem_move_block): New.
2385         (nds32_expand_movmemsi_loop_unknown_size): New.
2386         (nds32_expand_movmemsi_loop_known_size): New.
2387         (nds32_expand_movmemsi_loop): New.
2388         (nds32_expand_movmemsi_unroll): New.
2389         (nds32_expand_movmemqi): Rename ...
2390         (nds32_expand_movmemsi): ... to this.
2391         * config/nds32/nds32-multiple.md (movmemqi): Rename ...
2392         (movmemsi): ... to this.
2393         * config/nds32/nds32-protos.h (nds32_expand_movmemqi): Rename ...
2394         (nds32_expand_movmemsi): ... to this.
2395
2396 2018-03-04  Kito Cheng  <kito.cheng@gmail.com>
2397             Monk Chiang  <sh.chiang04@gmail.com>
2398             Chung-Ju Wu  <jasonwucj@gmail.com>
2399
2400         * config/nds32/nds32-protos.h
2401         (nds32_expand_load_multiple): New arguments.
2402         (nds32_expand_store_multiple): Ditto.
2403         (nds32_valid_multiple_load_store): Rename ...
2404         (nds32_valid_multiple_load_store_p): ... to this.
2405         * config/nds32/nds32-memory-manipulation.c
2406         (nds32_expand_load_multiple): Refine implementation.
2407         (nds32_expand_store_multiple): Ditto.
2408         * config/nds32/nds32-multiple.md
2409         (load_multiple): Update nds32_expand_load_multiple interface.
2410         (store_multiple): Update nds32_expand_store_multiple interface.
2411         * config/nds32/nds32-predicates.c
2412         (nds32_valid_multiple_load_store): Rename ...
2413         (nds32_valid_multiple_load_store_p): ... to this and refine
2414         implementation.
2415         * config/nds32/predicates.md
2416         (nds32_load_multiple_and_update_address_operation): New predicate.
2417         (nds32_store_multiple_and_update_address_operation): New predicate.
2418
2419 2018-03-04  Kito Cheng  <kito.cheng@gmail.com>
2420             Chung-Ju Wu  <jasonwucj@gmail.com>
2421
2422         * config/nds32/nds32.md (type): Add load_multiple and store_multiple.
2423         (combo): New attribute.
2424         * config/nds32/nds32-multiple.md: Refine patterns with new attributes.
2425
2426 2018-03-03  Chung-Ju Wu  <jasonwucj@gmail.com>
2427
2428         * config/nds32/nds32.opt: Change -mcmodel= default value.
2429
2430 2018-03-03  Kito Cheng  <kito.cheng@gmail.com>
2431             Monk Chiang  <sh.chiang04@gmail.com>
2432             Chung-Ju Wu  <jasonwucj@gmail.com>
2433
2434         * config/nds32/constants.md (unspec_element): New enum.
2435         * config/nds32/constraints.md (Umw): New constraint.
2436         * config/nds32/nds32-intrinsic.c: Add more builtin functions.
2437         * config/nds32/nds32-intrinsic.md: Likewise.
2438         * config/nds32/nds32-md-auxiliary.c (nds32_regno_to_enable4): New.
2439         (nds32_valid_smw_lwm_base_p): New.
2440         (nds32_output_smw_single_word): New.
2441         (nds32_output_lmw_single_word): New.
2442         (nds32_expand_unaligned_load): New.
2443         (nds32_expand_unaligned_store): New.
2444         * config/nds32/nds32-protos.h (nds32_valid_smw_lwm_base_p): Declare.
2445         (nds32_output_smw_single_word): Declare.
2446         (nds32_output_lmw_single_word): Declare.
2447         (nds32_expand_unaligned_load): Declare.
2448         (nds32_expand_unaligned_store): Declare.
2449         * config/nds32/nds32.h (nds32_builtins): Add NDS32_BUILTIN_UALOAD_HW,
2450         NDS32_BUILTIN_UALOAD_W, NDS32_BUILTIN_UALOAD_DW,
2451         NDS32_BUILTIN_UASTORE_HW, NDS32_BUILTIN_UASTORE_W,
2452         NDS32_BUILTIN_UASTORE_DW.
2453         * config/nds32/predicates.md (nds32_lmw_smw_base_operand): New
2454         predicate.
2455
2456 2018-03-03  Monk Chiang  <sh.chiang04@gmail.com>
2457             Kito Cheng  <kito.cheng@gmail.com>
2458             Chung-Ju Wu  <jasonwucj@gmail.com>
2459
2460         * config/nds32/nds32-intrinsic.c
2461         (nds32_expand_builtin_null_ftype_reg): Delete.
2462         (nds32_expand_builtin_reg_ftype_imm): Ditto.
2463         (nds32_expand_builtin_null_ftype_reg_imm): Ditto.
2464         (nds32_read_argument): New.
2465         (nds32_legitimize_target): Ditto.
2466         (nds32_legitimize_argument): Ditto.
2467         (nds32_check_constant_argument): Ditto.
2468         (nds32_expand_unop_builtin): Ditto.
2469         (nds32_expand_unopimm_builtin): Ditto.
2470         (nds32_expand_binop_builtin): Ditto.
2471         (nds32_builtin_decl_impl): Ditto.
2472         (builtin_description): Ditto.
2473         (nds32_expand_builtin_impl): Rewrite with new infrastructure.
2474         (nds32_init_builtins_impl): Ditto.
2475         * config/nds32/nds32.c (TARGET_BUILTIN_DECL): Define.
2476         (nds32_builtin_decl): New.
2477         * config/nds32/nds32.h (nds32_builtins): Add NDS32_BUILTIN_COUNT.
2478         * config/nds32/nds32-protos.h (nds32_builtin_decl_impl): Declare.
2479
2480 2018-03-02  Jeff Law  <law@redhat.com>
2481
2482         * reorg.c (stop_search_p): Handle DEBUG_INSN.
2483         (redundant_insn, fill_simple_delay_slots): Likewise.
2484         (fill_slots_from_thread): Likewise.
2485         * resource.c (mark_referenced_resources): Likewise.
2486         (mark_set_resources, find_dead_or_set_registers): Likewise.
2487
2488 2018-03-02  Jakub Jelinek  <jakub@redhat.com>
2489
2490         * substring-locations.h (format_warning_va): Formatting fix for
2491         ATTRIBUTE_GCC_DIAG.
2492         (format_warning_at_substring): Fix up ATTRIBUTE_GCC_DIAG second
2493         argument.
2494         (format_warning_n_va, format_warning_at_substring_n): New prototypes.
2495         * substring-locations.c: Include intl.h.
2496         (format_warning_va): Turned into small wrapper around
2497         format_warning_n_va, renamed to ...
2498         (format_warning_n_va): ... this, add N and PLURAL_GMSGID arguments,
2499         rename GMSGID to SINGULAR_GMSGID, if SINGULAR_GMSGID != PLURAL_GMSGID,
2500         use ngettext.
2501         (format_warning_at_substring_n): New function.
2502         * gimple-ssa-sprintf.c: Remove GCC diagnostic ignored pragma.
2503         (fmtwarn): Add ATTRIBUTE_GCC_DIAG.  Turn into a copy of
2504         format_warning_at_substring with just a shorter name instead of
2505         const function pointer.
2506         (fmtwarn_n): New function.
2507         (maybe_warn, format_directive, parse_directive): Use fmtwarn_n where
2508         appropriate, get rid of all the fmtstr temporaries, move conditionals
2509         with G_() wrapped string literals directly into fmtwarn arguments,
2510         cast dir.len to (int), formatting fixes.
2511
2512 2018-03-02  Thomas Schwinge  <thomas@codesourcery.com>
2513
2514         * doc/invoke.texi: Remove "Cilk Plus" references.
2515
2516 2018-03-02  Jakub Jelinek  <jakub@redhat.com>
2517             Richard Biener  <rguenther@suse.de>
2518
2519         PR ipa/84628
2520         * expr.c (expand_expr_real_1) <case CALL_EXPR>: Don't emit diagnostics
2521         for error or warning attributes if CALL_FROM_THUNK_P is set.
2522         Formatting fixes.
2523
2524 2018-03-02  Jakub Jelinek  <jakub@redhat.com>
2525
2526         PR target/56540
2527         * config/pa/pa.h (TARGET_CPU_CPP_BUILTINS): Predefine
2528         __SIZEOF_128__ macro if HPUX_LONG_DOUBLE_LIBRARY.
2529
2530         PR target/56540
2531         * config/ia64/ia64.h (TARGET_CPU_CPP_BUILTINS): Predefine
2532         __SIZEOF_{FPREG,FLOAT{80,128}}__ macros.
2533
2534         * predict.c (test_prediction_value_range): Use PROB_UNINITIALIZED
2535         instead of -1U in last predictors element's probability member.
2536
2537 2018-03-02  Eric Botcazou  <ebotcazou@adacore.com>
2538
2539         PR ipa/83983
2540         * ipa-devirt.c (odr_subtypes_equivalent_p): Get the ODR type of both
2541         arguments if they are comparable.
2542
2543 2018-03-02  Richard Sandiford  <richard.sandiford@linaro.org>
2544
2545         PR tree-optimization/84634
2546         * tree-vect-stmts.c (vectorizable_store, vectorizable_load): Replace
2547         masks and masked_loop_p with a single loop_masks, making sure it's
2548         null for bb vectorization.
2549
2550 2018-03-02  Richard Sandiford  <richard.sandiford@linaro.org>
2551
2552         * tree-vect-data-refs.c (vect_analyze_data_ref_dependence)
2553         (vect_analyze_data_ref_access): Use loop->safe_len rather than
2554         loop->force_vectorize to check whether there is no alias.
2555
2556 2018-03-02  Jakub Jelinek  <jakub@redhat.com>
2557
2558         PR target/84614
2559         * rtl.h (prev_real_nondebug_insn, next_real_nondebug_insn): New
2560         prototypes.
2561         * emit-rtl.c (next_real_insn, prev_real_insn): Fix up function
2562         comments.
2563         (next_real_nondebug_insn, prev_real_nondebug_insn): New functions.
2564         * cfgcleanup.c (try_head_merge_bb): Use prev_real_nondebug_insn
2565         instead of a loop around prev_real_insn.
2566         * combine.c (move_deaths): Use prev_real_nondebug_insn instead of
2567         prev_real_insn.
2568
2569         PR inline-asm/84625
2570         * config/i386/i386.c (ix86_print_operand): Use conditional
2571         output_operand_lossage instead of gcc_assert if CONST_VECTOR is not
2572         zero vector.
2573
2574 2018-03-02  Richard Biener  <rguenther@suse.de>
2575
2576         PR tree-optimization/84427
2577         * tree-ssa-pre.c (bitmap_remove_expr_from_set): Remove.
2578         (bitmap_set_subtract_values): Rewrite to handle multiple
2579         exprs per value.
2580         (clean): Likewise.
2581         (prune_clobbered_mems): Likewise.
2582         (phi_translate): Take edge instead of pred/phiblock.
2583         (phi_translate_1): Likewise.
2584         (phi_translate_set): Likewise.  Insert all translated
2585         exprs for a value into the set, keeping possibly multiple
2586         expressions per value.
2587         (compute_antic_aux): Adjust for phi_translate changes.
2588         When intersecting union the expressions and prune those
2589         not in the final value set, keeping possibly multiple
2590         expressions per value.  Do not use value-insertion
2591         for unioning ANTIC_OUT U EXP_GEN - TMP_GEN but merge
2592         all expressions.  Add verification that the value-sets
2593         only shrink during iteration.
2594         (compute_partial_antic_aux): Adjust for the phi_translate changes.
2595         (do_pre_regular_insertion): Likewise.
2596         (do_pre_partial_partial_insertion): Likewise.
2597
2598 2018-03-02  Richard Biener  <rguenther@suse.de>
2599
2600         PR target/82005
2601         * config/darwin.c (saved_debug_info_level): New static global.
2602         (darwin_asm_lto_start): Disable debug info generation for LTO out.
2603         (darwin_asm_lto_end): Restore debug info generation settings.
2604
2605 2018-03-01  Martin Liska  <mliska@suse.cz>
2606
2607         PR sanitizer/82484
2608         * sanopt.c (sanitize_rewrite_addressable_params): Do not handle
2609         volatile arguments.
2610
2611 2018-03-01  Richard Biener  <rguenther@suse.de>
2612
2613         PR debug/84645
2614         * dwarf2out.c (gen_variable_die): Properly handle late VLA
2615         type annotation with LTO when debug was disabled at compile-time.
2616
2617 2018-03-01  Matthew Fortune  <mfortune@gmail.com>
2618
2619         * config/mips/mips.c (mips_final_prescan_insn): Fix incorrect
2620         XINT with INTVAL.
2621         (mips_final_postscan_insn): Likewise.
2622
2623 2018-03-01  Richard Sandiford  <richard.sandiford@linaro.org>
2624
2625         PR rtl-optimization/84528
2626         * alias.c (init_alias_target): Add commentary.
2627         (init_alias_analysis): Only give HARD_FRAME_POINTER_REGNUM
2628         a unique base value if the frame pointer is not eliminated
2629         to the stack pointer.
2630
2631 2018-03-01  Tom de Vries  <tom@codesourcery.com>
2632
2633         PR rtl-optimization/83327
2634         * lra-int.h (hard_regs_spilled_into): Declare.
2635         * lra.c (hard_regs_spilled_into): Define.
2636         (init_reg_info): Init hard_regs_spilled_into.
2637         * lra-spills.c (assign_spill_hard_regs): Update hard_regs_spilled_into.
2638         * lra-lives.c (make_hard_regno_born, make_hard_regno_dead)
2639         (process_bb_lives): Handle hard_regs_spilled_into.
2640         (lra_create_live_ranges_1): Before doing liveness propagation, clear
2641         regs in all_hard_regs_bitmap if set in hard_regs_spilled_into.
2642
2643 2018-02-28  David Edelsohn  <dje.gcc@gmail.com>
2644
2645         * config.gcc (powerpc-ibm-aix7.1.*): New stanza.
2646         (powerpc-ibm-aix[789]*): Default to AIX 7.2.
2647         * config/rs6000/aix71.h (TARGET_DEFAULT): Revert to Power4 ISA.
2648         * config/rs6000/aix72.h: New file.
2649
2650 2018-02-28  Jakub Jelinek  <jakub@redhat.com>
2651
2652         * gimple-ssa-warn-restrict.c (maybe_diag_overlap): Use warning_n
2653         instead of warning_at with conditional singular and plural messages
2654         where possible.
2655
2656         PR target/52991
2657         * stor-layout.c (update_alignment_for_field): For
2658         targetm.ms_bitfield_layout_p (rli->t), if !is_bitfield
2659         && !DECL_PACKED (field), do the alignment update, just use
2660         only desired_align instead of MAX (type_align, desired_align)
2661         as the alignment.
2662         (place_field): Don't do known_align < desired_align handling
2663         early if targetm.ms_bitfield_layout_p (rli->t) and rli->prev_field
2664         is non-NULL, instead do it after rli->prev_field handling and
2665         only if not within a bitfield word.  For DECL_PACKED (field)
2666         use type_align of BITS_PER_UNIT.
2667
2668 2018-02-28  Eric Botcazou  <ebotcazou@adacore.com>
2669
2670         * config/aarch64/aarch64.c (aarch64_emit_probe_stack_range): Remove
2671         superfluous parentheses and trailing spaces.
2672
2673 2018-02-28  Richard Biener  <rguenther@suse.de>
2674
2675         PR tree-optimization/84584
2676         * graphite-scop-detection.c (scop_detection::add_scop): Discard
2677         SCoPs with fake exit edge.
2678
2679 2018-02-28  Martin Liska  <mliska@suse.cz>
2680
2681         PR testsuite/84597
2682         * timevar.c (timer::print): Fix format to properly print 100%
2683         values.
2684
2685 2018-02-28  Richard Biener  <rguenther@suse.de>
2686
2687         PR middle-end/84607
2688         * genmatch.c (capture_info::walk_match): Do not mark
2689         captured expressions without operands as expr_p given
2690         they act more like predicates and should be subject to
2691         "lost tail" side-effect preserving.
2692
2693 2018-02-28  Alexandre Oliva  <aoliva@redhat.com>
2694
2695         PR rtl-optimization/81611
2696         * auto-inc-dec.c (attempt_change): Move dead note from
2697         mem_insn if it's the next use of regno
2698         (find_address): Take address use of reg holding
2699         non-incremented value.  Add parm to limit search to the named
2700         reg only.
2701         (merge_in_block): Attempt to use a mem insn that is the next
2702         use of the original regno.
2703
2704 2018-02-27  Martin Sebor  <msebor@redhat.com>
2705
2706         PR c++/83871
2707         * gcc/doc/invoke.texi (-Wmissing-attributes): New option.
2708         * gcc/print-tree.c (print_node): Handle DECL_UNINLINABLE.
2709
2710 2018-02-27  Martin Sebor  <msebor@redhat.com>
2711
2712         PR translation/84207
2713         * diagnostic-core.h (warning_n, error_n, inform_n): Change
2714         n argument to unsigned HOST_WIDE_INT.
2715         * diagnostic.c (warning_n, error_n, inform_n): Ditto.
2716         (diagnostic_n_impl): Ditto.  Handle arguments in excess of LONG_MAX.
2717         * gimple-ssa-sprintf.c (format_directive): Simplify inform_n call.
2718         * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Use warning_n.
2719
2720 2018-02-27  Richard Biener  <rguenther@suse.de>
2721
2722         PR tree-optimization/84512
2723         * tree-vect-loop.c (vect_compute_single_scalar_iteration_cost):
2724         Do not use the estimate returned from record_stmt_cost for
2725         the scalar iteration cost but sum properly using add_stmt_cost.
2726
2727 2018-02-27  Richard Biener  <rguenther@suse.de>
2728
2729         PR tree-optimization/84466
2730         * graphite-scop-detection.c (scop_detection::stmt_simple_for_scop_p):
2731         Adjust last change to less strictly validate use operands.
2732
2733 2018-02-27  Martin Liska  <mliska@suse.cz>
2734
2735         PR gcov-profile/84548
2736         * gcov.c (process_file): Allow partial overlap and consider it
2737         also as group functions.
2738         (output_lines): Properly calculate range of lines for a group.
2739
2740 2018-02-27  Martin Liska  <mliska@suse.cz>
2741
2742         * timevar.c (timer::print_row): Remove 'usr', 'sys', 'wall' and
2743         'ggc' suffixes.  Change first column width.
2744         (timer::print): Fix formatting of the column.
2745
2746 2018-02-27  Alexandre Oliva  <aoliva@redhat.com>
2747
2748         * tree-ssa-live.c (remove_unused_scope_block_p): Do not
2749         preserve inline entry blocks for the sake of debug inline
2750         entry point markers alone.
2751         (remove_unused_locals): Suggest in comments a better place to
2752         force the preservation of inline entry blocks that are
2753         otherwise unused, but do not preserve them.
2754
2755 2018-02-26  H.J. Lu  <hongjiu.lu@intel.com>
2756
2757         * config/i386/i386.c (ix86_output_indirect_jmp): Update comments.
2758
2759 2018-02-26  H.J. Lu  <hongjiu.lu@intel.com>
2760
2761         PR target/84039
2762         * config/i386/constraints.md (Bs): Replace
2763         ix86_indirect_branch_register with
2764         TARGET_INDIRECT_BRANCH_REGISTER.
2765         (Bw): Likewise.
2766         * config/i386/i386.md (indirect_jump): Likewise.
2767         (tablejump): Likewise.
2768         (*sibcall_memory): Likewise.
2769         (*sibcall_value_memory): Likewise.
2770         Peepholes of indirect call and jump via memory: Likewise.
2771         (*sibcall_GOT_32): Disallowed for TARGET_INDIRECT_BRANCH_REGISTER.
2772         (*sibcall_value_GOT_32): Likewise.
2773         * config/i386/predicates.md (indirect_branch_operand): Likewise.
2774         (GOT_memory_operand): Likewise.
2775         (call_insn_operand): Likewise.
2776         (sibcall_insn_operand): Likewise.
2777         (GOT32_symbol_operand): Likewise.
2778         * config/i386/i386.h (TARGET_INDIRECT_BRANCH_REGISTER): New.
2779
2780 2018-02-26  Eric Botcazou  <ebotcazou@adacore.com>
2781
2782         PR rtl-optimization/83496
2783         * reorg.c (steal_delay_list_from_target): Change REDUNDANT array from
2784         booleans to RTXes.  Call fix_reg_dead_note on every non-null element.
2785         (steal_delay_list_from_fallthrough): Call fix_reg_dead_note on a
2786         redundant insn, if any.
2787         (relax_delay_slots): Likewise.
2788         (update_reg_unused_notes): Rename REDUNDANT_INSN to OTHER_INSN.
2789
2790 2018-02-26  Richard Sandiford  <richard.sandiford@linaro.org>
2791
2792         PR tree-optimization/83965
2793         * tree-vect-patterns.c (vect_reassociating_reduction_p): Assume
2794         that grouped statements are part of a reduction chain.  Return
2795         true if the statement is not marked as a reduction itself but
2796         is part of a group.
2797         (vect_recog_dot_prod_pattern): Don't check whether the statement
2798         is part of a group here.
2799         (vect_recog_sad_pattern): Likewise.
2800         (vect_recog_widen_sum_pattern): Likewise.
2801
2802 2018-02-26  Eric Botcazou  <ebotcazou@adacore.com>
2803
2804         PR debug/84545
2805         * final.c (rest_of_clean_state): Also look for calls inside sequences.
2806
2807 2018-02-26  H.J. Lu  <hongjiu.lu@intel.com>
2808
2809         PR target/84530
2810         * config/i386/i386-protos.h (ix86_output_indirect_jmp): Remove
2811         the bool argument.
2812         (ix86_output_indirect_function_return): New prototype.
2813         (ix86_split_simple_return_pop_internal): Likewise.
2814         * config/i386/i386.c (indirect_return_via_cx): New.
2815         (indirect_return_via_cx_bnd): Likewise.
2816         (indirect_thunk_name): Handle return va CX_REG.
2817         (output_indirect_thunk_function): Create alias for
2818         __x86_return_thunk_[re]cx and __x86_return_thunk_[re]cx_bnd.
2819         (ix86_output_indirect_jmp): Remove the bool argument.
2820         (ix86_output_indirect_function_return): New function.
2821         (ix86_split_simple_return_pop_internal): Likewise.
2822         * config/i386/i386.md (*indirect_jump): Don't pass false
2823         to ix86_output_indirect_jmp.
2824         (*tablejump_1): Likewise.
2825         (simple_return_pop_internal): Change it to define_insn_and_split.
2826         Call ix86_split_simple_return_pop_internal to split it for
2827         -mfunction-return=.
2828         (simple_return_indirect_internal): Call
2829         ix86_output_indirect_function_return instead of
2830         ix86_output_indirect_jmp.
2831
2832 2018-02-26  Jakub Jelinek  <jakub@redhat.com>
2833
2834         PR bootstrap/84405
2835         * vec.h (vec_default_construct): For BROKEN_VALUE_INITIALIZATION use
2836         memset and value initialization afterwards.
2837
2838 2018-02-26  Christophe Lyon  <christophe.lyon@linaro.org>
2839
2840         * Makefile.in (lto-wrapper): Use ALL_LINKERFLAGS.
2841
2842 2018-02-26  Ramana Radhakrishnan  <ramana.radhakrishnan@arm.com>
2843
2844         PR target/84521
2845         * common/config/aarch64/aarch64-common.c
2846         (aarch_option_optimization_table[]): Switch
2847         off fomit-frame-pointer
2848
2849 2018-02-26  Kito Cheng  <kito.cheng@gmail.com>
2850             Chung-Ju Wu  <jasonwucj@gmail.com>
2851
2852         * config/nds32/nds32-multiple.md (load_multiple): Disallow
2853         volatile memory.
2854         (store_multiple): Ditto.
2855
2856 2018-02-26  Kito Cheng  <kito.cheng@gmail.com>
2857
2858         * config.gcc: Add --with-cpu support for nds32 target.
2859         * config/nds32/nds32-opts.h (nds32_cpu_type): New.
2860         * config/nds32/nds32.opt: Add -mcpu= option.
2861
2862 2018-02-25  Segher Boessenkool  <segher@kernel.crashing.org>
2863
2864         * config/rs6000/rs6000.opt (mvrsave=no, mvrsave=yes, isel=no,
2865         isel=yes): Warn for these deprecated options.
2866
2867 2018-02-23  David Edelsohn  <dje.gcc@gmail.com>
2868
2869         * config/rs6000/aix71.h (TARGET_DEFAULT): Change to
2870         ISA_2_5_MASKS_EMBEDDED.
2871
2872 2018-02-23  Jakub Jelinek  <jakub@redhat.com>
2873
2874         * ipa-prop.c (ipa_vr_ggc_hash_traits::hash): Hash p->min and
2875         p->max as pointers rather than using iterative_hash_expr.
2876
2877 2018-02-23  Carl Love  <cel@us.ibm.com>
2878
2879         * config/rs6000/rs6000-builtin.def: Change VSIGNED2 and VUNSIGNED2
2880         macro expansions from BU_VSX_2 to BU_P8V_VSX_2 and BU_VSX_OVERLOAD_2 to
2881         BU_P8V_OVERLOAD_2.
2882         * config/rs6000/rs6000-c.c: Change VSX_BUILTIN_VEC_VSIGNED2 to
2883         P8V_BUILTIN_VEC_VSIGNED2.  Change VSX_BUILTIN_VEC_VUNSIGNED2 to
2884         P8V_BUILTIN_VEC_VUNSIGNED2.
2885
2886 2018-02-22  Vladimir Makarov  <vmakarov@redhat.com>
2887
2888         PR target/81572
2889         * lra-int.h (LRA_UNKNOWN_ALT, LRA_NON_CLOBBERED_ALT): New macros.
2890         * lra.c (lra_set_insn_recog_data, lra_update_insn_recog_data): Use
2891         LRA_UNKNOWN_ALT.
2892         * lra-constraints.c (curr_insn_transform): Set up
2893         LRA_NON_CLOBBERED_ALT for moves processed on the fast path.  Use
2894         LRA_UNKNOWN_ALT.
2895         (remove_inheritance_pseudos): Use LRA_UNKNOWN_ALT.
2896         * lra-eliminations.c (spill_pseudos): Ditto.
2897         (process_insn_for_elimination): Ditto.
2898         * lra-lives.c (reg_early_clobber_p): Use the new macros.
2899         * lra-spills.c (spill_pseudos): Use LRA_UNKNOWN_ALT and
2900         LRA_NON_CLOBBERED_ALT.
2901
2902 2018-02-22  Martin Sebor  <msebor@redhat.com>
2903
2904         PR tree-optimization/84480
2905         * gimple-fold.c (gimple_fold_builtin_strcpy): Move warnings
2906         to maybe_diag_stxncpy_trunc.  Call it.
2907         * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Integrate warnings
2908         from gimple_fold_builtin_strcpy.  Print inlining stack.
2909         (handle_builtin_stxncpy): Print inlining stack.
2910         * tree-ssa-strlen.h (maybe_diag_stxncpy_trunc): Declare.
2911
2912 2018-02-22  H.J. Lu  <hongjiu.lu@intel.com>
2913
2914         PR target/84176
2915         * config/i386/i386.c (ix86_set_indirect_branch_type): Issue an
2916         error when -mindirect-branch=thunk-extern, -fcf-protection=branch
2917         and -fcheck-pointer-bounds are used together.
2918         (indirect_thunk_prefix): New enum.
2919         (indirect_thunk_need_prefix): New function.
2920         (indirect_thunk_name): Replace need_bnd_p with need_prefix.  Use
2921         "_nt" instead of "_bnd" for NOTRACK prefix.
2922         (output_indirect_thunk): Replace need_bnd_p with need_prefix.
2923         (output_indirect_thunk_function): Likewise.
2924         (): Likewise.
2925         (ix86_code_end): Update output_indirect_thunk_function calls.
2926         (ix86_output_indirect_branch_via_reg): Replace
2927         ix86_bnd_prefixed_insn_p with indirect_thunk_need_prefix.
2928         (ix86_output_indirect_branch_via_push): Likewise.
2929         (ix86_output_function_return): Likewise.
2930         * doc/invoke.texi: Document -mindirect-branch=thunk-extern is
2931         incompatible with -fcf-protection=branch and
2932         -fcheck-pointer-bounds.
2933
2934 2018-02-22  Steve Ellcey  <sellcey@cavium.com>
2935
2936         PR target/83335
2937         * config/aarch64/aarch64.c (aarch64_print_address_internal):
2938         Change gcc_assert call to output_operand_lossage.
2939
2940 2018-02-22  Steve Ellcey  <sellcey@cavium.com>
2941
2942         * doc/extend.texi (__builtin_extend_pointer): Document builtin.
2943
2944 2018-02-22  DJ Delorie  <dj@redhat.com>
2945             Sebastian Perta  <sebastian.perta@renesas.com>
2946             Oleg Endo  <olegendo@gcc.gnu.org>
2947
2948         * config/rx/rx.c (rx_rtx_costs): New function.
2949         (TARGET_RTX_COSTS): Override to use rx_rtx_costs.
2950
2951 2018-02-22  Thomas Preud'homme  <thomas.preudhomme@arm.com>
2952
2953         * config/arm/t-multilib: Map Armv8-R to Armv7 multilibs.
2954
2955 2018-02-22  Martin Liska  <mliska@suse.cz>
2956
2957         PR driver/83193
2958         * common/config/arm/arm-common.c (arm_print_hint_for_cpu_option):
2959         Add "native" as a possible value.
2960
2961 2018-02-22  Martin Liska  <mliska@suse.cz>
2962
2963         PR driver/83193
2964         * config/i386/i386.c (ix86_option_override_internal):
2965         Add "native" as a possible value for -march and -mtune.
2966
2967 2018-02-22  Jakub Jelinek  <jakub@redhat.com>
2968
2969         PR target/84502
2970         * stor-layout.c (finalize_type_size): Propagate TYPE_EMPTY_P flag
2971         to all type variants.
2972
2973         PR tree-optimization/84503
2974         * gimple-ssa-store-merging.c (merged_store_group::merge_into): Compute
2975         width as info->bitpos + info->bitsize - start.
2976         (merged_store_group::merge_overlapping): Simplify width computation.
2977         (check_no_overlap): New function.
2978         (imm_store_chain_info::try_coalesce_bswap): Compute expected
2979         start + width and last_order of the group, fail if check_no_overlap
2980         fails.
2981         (imm_store_chain_info::coalesce_immediate_stores): Don't merge info
2982         to group if check_no_overlap fails.
2983
2984 2018-02-21  Segher Boessenkool  <segher@kernel.crashing.org>
2985
2986         * config/rs6000/altivec.md: Delete contraint arguments to
2987         define_expand, define_split, and define_peephole2, and in
2988         define_insn_and_split if always unused.
2989         * config/rs6000/darwin.md: Ditto.
2990         * config/rs6000/dfp.md: Ditto.
2991         * config/rs6000/rs6000.md: Ditto.
2992         * config/rs6000/sync.md: Ditto.
2993         * config/rs6000/vector.md: Ditto.
2994         * config/rs6000/vsx.md: Ditto.
2995
2996 2018-02-21  Segher Boessenkool  <segher@kernel.crashing.org>
2997
2998         * config/rs6000/altivec.md: Write output control strings as braced
2999         blocks instead of double-quoted strings.
3000         * config/rs6000/darwin.md: Ditto.
3001         * config/rs6000/rs6000.md: Ditto.
3002         * config/rs6000/vector.md: Ditto.
3003         * config/rs6000/vsx.md: Ditto.
3004
3005 2018-02-21  Jason Merrill  <jason@redhat.com>
3006
3007         PR c++/84314 - ICE with templates and fastcall attribute.
3008         * attribs.c (build_type_attribute_qual_variant): Remove assert.
3009
3010 2018-02-21  Jan Hubicka  <hubicka@ucw.cz>
3011
3012         * ipa-cp.c (determine_versionability): Fix comment typos.
3013
3014 2018-02-21  Jan Hubicka  <hubicka@ucw.cz>
3015
3016         PR c/84229
3017         * ipa-cp.c (determine_versionability): Do not version functions caling
3018         va_arg_pack.
3019
3020 2018-02-21  Martin Liska  <mliska@suse.cz>
3021
3022         PR driver/83193
3023         * config/aarch64/aarch64.c (aarch64_print_hint_for_core_or_arch):
3024         Add "native" as a possible value.
3025         * config/aarch64/aarch64.h (HAVE_LOCAL_CPU_DETECT):  Define
3026         the macro when native cpu detection is available.
3027
3028 2018-02-21  Martin Liska  <mliska@suse.cz>
3029
3030         PR driver/83193
3031         * common/config/arm/arm-common.c (arm_print_hint_for_arch_option):
3032         Add "native" as a possible value.
3033         * config/arm/arm.h (HAVE_LOCAL_CPU_DETECT): Define the macro
3034         when native cpu detection is available.
3035
3036 2018-02-21  Jakub Jelinek  <jakub@redhat.com>
3037             Martin Sebor  <msebor@redhat.com>
3038
3039         PR tree-optimization/84478
3040         * gimple-fold.h (get_range_strlen): Add a bool argument defaulted to
3041         false.
3042         * gimple-fold.c (get_range_strlen): Make minlen const and assume it
3043         can't be NULL.  Change FUZZY from bool to int, for 1 add PHI/COND_EXPR
3044         support which is conservatively correct, for 2 only stay conservative
3045         for maxlen.  Formatting and comment capitalization fixes.  Add STRICT
3046         argument to the 2 argument get_range_strlen, adjust 6 arg
3047         get_range_strlen caller and clear minmaxlen[0] and [1] if it returned
3048         false.
3049         (get_maxval_strlen): Adjust 6 arg get_range_strlen caller.
3050         (gimple_fold_builtin_strlen): Pass true as last argument to
3051         get_range_strlen.
3052
3053 2018-02-20  Martin Sebor  <msebor@redhat.com>
3054
3055         PR middle-end/84095
3056         * gimple-ssa-warn-restrict.c (builtin_memref::extend_offset_range): New.
3057         (builtin_memref::set_base_and_offset): Same.  Handle inner references.
3058         (builtin_memref::builtin_memref): Factor out parts into
3059         set_base_and_offset and call it.
3060
3061 2018-02-20  Richard Sandiford  <richard.sandiford@linaro.org>
3062
3063         PR middle-end/84406
3064         * optabs-query.c (find_widening_optab_handler_and_mode): If from_mode
3065         is a scalar_int_mode, assert that to_mode is a scalar_int_mode with
3066         greater precision.  If to_mode is a MODE_PARTIAL_INT, stop the
3067         search at the associated MODE_INT.
3068
3069 2018-02-20  Jeff Law  <law@redhat.com>
3070
3071         PR middle-end/82123
3072         PR tree-optimization/81592
3073         PR middle-end/79257
3074         * gimple-ssa-sprintf.c (format_integer): Query EVRP range analyzer
3075         for range data rather than using global data.
3076         * gimple-ssa-sprintf.c (get_int_range): Query EVRP range analyzer for
3077         range data rather than using global data.
3078         * gimple-ssa-sprintf.c (get_int_range): Accept vr_values parameter
3079         pass it to children as needed.
3080         (struct directive::fmtresult): Similarly.
3081         (struct directive::set_width): Similarly.
3082         (struct directive::set_precision): Similarly.
3083         (format_integer, format_directive, parse_directive): Similarly.
3084         (format_none): Accept unnamed vr_values parameter.
3085         (format_percent, format_floating, format_character): Similarly.
3086         (format_string, format_plain): Similarly.
3087         * gimple-ssa-sprintf.c (sprintf_dom_walker::handle_gimple_call): Query
3088         the EVRP range analyzer for range data rather than using global data.
3089         * gimple-ssa-sprintf.c: Include alloc-pool.h, vr-values.h and
3090         gimple-ssa-evrp-analyze.h
3091         (class sprintf_dom_walker): Add after_dom_children member function.
3092         Add evrp_range_analyzer member.
3093         (sprintf_dom_walker::before_dom_children): Call into the EVRP
3094         range analyzer as needed.
3095         (sprintf_dom_walker::after_dom_children): New member function.
3096         * gimple-ssa-evrp-analyze.c (evrp_range_analyzer::enter): Do nothing
3097         if not optimizing.
3098         (evrp_range_analyzer::record_ranges_from_stmt): Likewise.
3099         (evrp_range_analyzer::pop_to_marker): Likewise.
3100
3101 2018-02-20  Richard Sandiford  <richard.sandiford@linaro.org>
3102
3103         PR tree-optimization/84419
3104         * internal-fn.c (expand_call_mem_ref): Create a TARGET_MEM_REF
3105         with the required type if its current type is compatible but
3106         different.
3107
3108 2018-02-20  Jakub Jelinek  <jakub@redhat.com>
3109
3110         PR middle-end/82004
3111         * match.pd (pow(C,x) -> exp(log(C)*x)): Delay all folding until
3112         after vectorization.
3113
3114 2018-02-20  Martin Liska  <mliska@suse.cz>
3115
3116         PR driver/83193
3117         * config/aarch64/aarch64.c (aarch64_print_hint_for_core_or_arch): Print
3118         possible values if we don't have a hint.
3119
3120 2018-02-20  Martin Liska  <mliska@suse.cz>
3121
3122         PR c/84310
3123         PR target/79747
3124         * final.c (shorten_branches): Build align_tab array with one
3125         more element.
3126         * opts.c (finish_options): Add alignment option limit check.
3127         (MAX_CODE_ALIGN): Likewise.
3128         (MAX_CODE_ALIGN_VALUE): Likewise.
3129         * doc/invoke.texi: Document maximum allowed option value for
3130         all -falign-* options.
3131
3132 2018-02-19  Jakub Jelinek  <jakub@redhat.com>
3133
3134         PR target/84146
3135         * reg-notes.def (REG_CALL_ARG_LOCATION): New reg note.
3136         * insn-notes.def (NOTE_INSN_CALL_ARG_LOCATION): Remove.
3137         * var-tracking.c (emit_note_insn_var_location): Remove all references
3138         to NOTE_INSN_CALL_ARG_LOCATION.
3139         (emit_notes_in_bb): Emit arguments as REG_CALL_ARG_LOCATION note on
3140         the CALL_INSN rather than separate NOTE_INSN_CALL_ARG_LOCATION note.
3141         Use copy_rtx_if_shared.
3142         * dwarf2out.c (gen_subprogram_die): Use XEXP with 0 instead of
3143         NOTE_VAR_LOCATION on ca_loc->call_arg_loc_note.
3144         (dwarf2out_var_location): Remove handling of
3145         NOTE_INSN_CALL_ARG_LOCATION, instead handle REG_CALL_ARG_LOCATION note
3146         on call_insn.
3147         * final.c (final_scan_insn): Remove all references to
3148         NOTE_INSN_CALL_ARG_LOCATION.
3149         (rest_of_clean_state): Likewise.  Remove REG_CALL_ARG_LOCATION notes
3150         before dumping final insns.
3151         * except.c (emit_note_eh_region_end): Remove all references to
3152         NOTE_INSN_CALL_ARG_LOCATION.
3153         * config/alpha/alpha.c (alpha_pad_function_end): Likewise.
3154         * config/c6x/c6x.c (c6x_gen_bundles): Likewise.
3155         * config/arc/arc.c (hwloop_optimize): Likewise.
3156         * config/arm/arm.c (create_fix_barrier): Likewise.
3157         * config/s390/s390.c (s390_chunkify_start): Likewise.
3158         * config/sh/sh.c (find_barrier): Likewise.
3159         * config/i386/i386.c (rest_of_insert_endbranch,
3160         ix86_seh_fixup_eh_fallthru): Likewise.
3161         * config/xtensa/xtensa.c (hwloop_optimize): Likewise.
3162         * config/iq2000/iq2000.c (final_prescan_insn): Likewise.
3163         * config/frv/frv.c (frv_function_prologue): Likewise.
3164         * emit-rtl.c (try_split): Likewise.  Copy over REG_CALL_ARG_LOCATION
3165         reg note.
3166         (note_outside_basic_block_p): Remove all references to
3167         NOTE_INSN_CALL_ARG_LOCATION.
3168         * gengtype.c (adjust_field_rtx_def): Likewise.
3169         * print-rtl.c (rtx_writer::print_rtx_operand_code_0, print_insn):
3170         Likewise.
3171         * jump.c (cleanup_barriers, delete_related_insns): Likewise.
3172         * cfgrtl.c (force_nonfallthru_and_redirect): Likewise.
3173
3174         PR c++/84444
3175         * builtins.c (builtin_mathfn_code): Don't check if CALL_EXPR_FN (t)
3176         is ADDR_EXPR.
3177
3178         PR tree-optimization/84452
3179         * tree-vect-patterns.c (vect_recog_pow_pattern): Don't call
3180         expand_simd_clones if targetm.simd_clone.compute_vecsize_and_simdlen
3181         is NULL.
3182
3183 2018-02-19  Martin Liska  <mliska@suse.cz>
3184
3185         PR sanitizer/82183
3186         * passes.def: Put pass_sancov_O0 before pass_lower_switch with -O0.
3187
3188 2018-02-19  Martin Liska  <mliska@suse.cz>
3189             Richard Sandiford  <richard.sandiford@linaro.org>
3190
3191         PR tree-optimization/82491
3192         * gimple-fold.c (get_base_constructor): Make earlier bail out
3193         to prevent ubsan.
3194
3195 2018-02-19  Carl Love  <cel@us.ibm.com>
3196
3197         * config/rs6000/rs6000-builtin.def: Change NEG macro expansions from
3198         BU_ALTIVEC_A to BU_P8V_AV_1 and BU_ALTIVEC_OVERLOAD_1 to
3199         BU_P8V_OVERLOAD_1.
3200         * config/rs6000/rs6000-c.c: Change ALTIVEC_BUILTIN_VEC_NEG to
3201         P8V_BUILTIN_VEC_NEG.
3202
3203 2018-02-19  Sebastian Perta  <sebastian.perta@renesas.com>
3204
3205         * config/rl78/rl78.md (movdf): New define expand.
3206
3207 2018-02-19  Martin Liska  <mliska@suse.cz>
3208
3209         PR other/80589
3210         * doc/invoke.texi: Fix typo.
3211         * params.def (PARAM_MAX_LOOP_HEADER_INSNS): Likewise.
3212
3213 2018-02-18  Segher Boessenkool  <segher@kernel.crashing.org>
3214
3215         * config/rs6000/rs6000.c (rs6000_option_override_internal): Don't
3216         handle rs6000_single_float and rs6000_double_float specially for
3217         e500 family CPUs.
3218
3219 2018-02-16  Jeff Law  <law@redhat.com>
3220
3221         * config/rx/rx.c (add_pop_cfi_notes): New function.;
3222         (pop_regs): Use it.
3223
3224 2018-02-16  Jakub Jelinek  <jakub@redhat.com>
3225
3226         PR ipa/84425
3227         * ipa-inline.c (inline_small_functions): Fix a typo.
3228
3229 2018-02-16  Nathan Sidwell  <nathan@acm.org>
3230
3231         * doc/extend.texi (Backwards Compatibility): Americanize 'behaviour'.
3232
3233 2018-02-16  Carl Love  <cel@us.ibm.com>
3234
3235         * config/rs6000/rs6000-builtin.def: Add BU_P8V_VSX_2 macro definition.
3236         Change FLOAT2 expansions from BU_VSX_2 to BU_P8V_VSX_2 and
3237         from BU_VSX_OVERLOAD_2 to BU_P8V_OVERLOAD_2.
3238         * config/rs6000/rs6000-c.c: Changed macro VSX_BUILTIN_VEC_FLOAT2
3239         expansion to P8V_BUILTIN_VEC_FLOAT2.
3240
3241 2018-02-16  Vladimir Makarov  <vmakarov@redhat.com>
3242
3243         PR rtl-optimization/70023
3244         * lra-constraints.c (inherit_in_ebb): Take hard reg mode of
3245         src_regno into account.
3246
3247 2018-02-16  Carl Love  <cel@us.ibm.com>
3248
3249         * config/rs6000/altivec.h: Remove vec_vextract4b and vec_vinsert4b.
3250         * config/rs6000/rs6000-builtin.def: Remove macro expansion for
3251         VEXTRACT4B, VINSERT4B, VINSERT4B_DI and VEXTRACT4B.
3252         * config/rs6000/rs6000.c: Remove case statements for
3253         P9V_BUILTIN_VEXTRACT4B, P9V_BUILTIN_VEC_VEXTRACT4B,
3254         P9V_BUILTIN_VINSERT4B, P9V_BUILTIN_VINSERT4B_DI,
3255         and P9V_BUILTIN_VEC_VINSERT4B.
3256         * config/rs6000/rs6000-c.c (altivec_expand_builtin): Remove entries for
3257         P9V_BUILTIN_VEC_VEXTRACT4B and P9V_BUILTIN_VEC_VINSERT4B.
3258         * config/rs6000/vsx.md:
3259         * doc/extend.texi: Remove vec_vextract4b, non ABI definitions for
3260         vec_insert4b.
3261
3262 2018-02-16  Carl Love  <cel@us.ibm.com>
3263
3264         * config/rs6000/altivec.h: Add builtin names vec_extract4b
3265         vec_insert4b.
3266         * config/rs6000/rs6000-builtin.def: Add INSERT4B and EXTRACT4B
3267         definitions.
3268         * config/rs6000/rs6000-c.c: Add the definitions for
3269         P9V_BUILTIN_VEC_EXTRACT4B and P9V_BUILTIN_VEC_INSERT4B.
3270         * config/rs6000/rs6000.c (altivec_expand_builtin): Add
3271         P9V_BUILTIN_EXTRACT4B and P9V_BUILTIN_INSERT4B case statements.
3272         * config/rs6000/vsx.md: Add define_insn extract4b.  Add define_expand
3273         definition for insert4b and define insn *insert3b_internal.
3274         * doc/extend.texi: Add documentation for vec_extract4b.
3275
3276 2018-02-16  Nathan Sidwell  <nathan@acm.org>
3277
3278         * doc/extend.texi (Backwards Compatibility): Mention friend
3279         injection.  Note for-scope is deprecated.
3280         * doc/invoke.texi (-ffriend-injection): Deprecate.
3281
3282 2018-02-16  Segher Boessenkool  <segher@kernel.crashing.org>
3283
3284         * combine.c (try_combine): When adjusting LOG_LINKS for the destination
3285         that moved to I2, also allow destinations that are a paradoxical
3286         subreg (instead of a normal reg).
3287
3288 2018-02-16  Oleg Endo  <olegendo@gcc.gnu.org>
3289
3290         PR target/83831
3291         * config/rx/rx.c (rx_fuse_in_memory_bitop): Convert shift operand
3292         to QImode.
3293
3294 2018-02-16  Richard Biener  <rguenther@suse.de>
3295
3296         PR tree-optimization/84037
3297         PR tree-optimization/84016
3298         PR target/82862
3299         * config/i386/i386.c (ix86_builtin_vectorization_cost):
3300         Adjust vec_construct for the fact we need additional higher latency
3301         128bit inserts for AVX256 and AVX512 vector builds.
3302         (ix86_add_stmt_cost): Scale vector construction cost for
3303         elementwise loads.
3304
3305 2018-02-16  Richard Biener  <rguenther@suse.de>
3306
3307         PR tree-optimization/84417
3308         * tree-ssa.c (non_rewritable_mem_ref_base): Properly constrain
3309         the MEM_REF offset when conversion to BIT_FIELD_REF is desired.
3310         (non_rewritable_lvalue_p): Likewise, use poly-ints.
3311
3312 2018-02-16  Martin Liska  <mliska@suse.cz>
3313
3314         PR sanitizer/84307
3315         * internal-fn.def (ASAN_CHECK): Set proper flags.
3316         (ASAN_MARK): Likewise.
3317
3318 2018-02-16  Julia Koval  <julia.koval@intel.com>
3319
3320         * config/i386/i386.c (ix86_option_override_internal): Remove PTA_CLWB
3321         from PTA_CANNONLAKE.
3322
3323 2018-02-16  Jakub Jelinek  <jakub@redhat.com>
3324
3325         PR target/84272
3326         * config/aarch64/cortex-a57-fma-steering.c (fma_forest::merge_forest):
3327         Use ++iter rather than iter++ for std::list iterators.
3328         (func_fma_steering::dfs): Likewise.  Don't delete nodes right away,
3329         defer deleting them until all nodes in the forest are processed.  Do
3330         free even leaf nodes.  Change to_process into auto_vec.
3331
3332         PR bootstrap/84405
3333         * system.h (BROKEN_VALUE_INITIALIZATION): Define for GCC < 4.3.
3334         * vec.h (vec_default_construct): Use memset instead of placement new
3335         if BROKEN_VALUE_INITIALIZATION is defined.
3336         * hash-table.h (hash_table<Descriptor, Allocator>::empty_slow): Use
3337         memset instead of value initialization if BROKEN_VALUE_INITIALIZATION
3338         is defined.
3339
3340         PR rtl-optimization/83723
3341         * lra-int.h (lra_substitute_pseudo): Add DEBUG_P argument.
3342         * lra.c (lra_substitute_pseudo): Likewise.  If true, use
3343         gen_rtx_raw_SUBREG instead of gen_rtx_SUBREG.  Pass DEBUG_P to
3344         recursive calls.
3345         (lra_substitute_pseudo_within_insn): Adjust lra_substitute_pseudo
3346         callers.
3347         * lra-constraints.c (inherit_reload_reg, split_reg): Likewise.
3348
3349 2018-02-16  Eric Botcazou  <ebotcazou@adacore.com>
3350
3351         PR rtl-optimization/81443
3352         * rtlanal.c (num_sign_bit_copies1) <SUBREG>: Do not propagate results
3353         from inner REGs to paradoxical SUBREGs.
3354
3355 2018-02-16  Richard Biener  <rguenther@suse.de>
3356
3357         PR tree-optimization/84399
3358         * graphite-scop-detection.c (scop_detection::stmt_simple_for_scop_p):
3359         For operands we can analyze at their definition make sure we can
3360         analyze them at each use as well.
3361
3362 2018-02-16  Richard Biener  <rguenther@suse.de>
3363
3364         PR tree-optimization/84190
3365         * tree-ssa.c (non_rewritable_mem_ref_base): Do not touch
3366         volatile accesses if the decl isn't volatile.
3367
3368 2018-02-15  Jason Merrill  <jason@redhat.com>
3369
3370         PR c++/84314 - ICE with templates and fastcall attribute.
3371         * attribs.c (build_type_attribute_qual_variant): Don't clobber
3372         TYPE_CANONICAL on an existing type.
3373
3374 2018-02-15  Jakub Jelinek  <jakub@redhat.com>
3375
3376         PR tree-optimization/84383
3377         * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Don't look at
3378         dstoff nor call operand_equal_p if dstbase is NULL.
3379
3380         PR tree-optimization/84334
3381         * match.pd ((A +- CST1) +- CST2 -> A + CST3): If A is
3382         also a CONSTANT_CLASS_P, punt.
3383
3384 2018-02-14  Jim Wilson  <jimw@sifive.com>
3385
3386         * config/riscv/riscv.c (riscv_first_stack_step): Move locals after
3387         first SMALL_OPERAND check.  New local min_second_step.  Move assert
3388         to where locals are set.  Add TARGET_RVC support.
3389         * config/riscv/riscv.h (C_SxSP_BITS, SWSP_REACH, SDSP_REACH): New.
3390
3391 2018-02-14  Indu Bhagat  <indu.bhagat@oracle.com>
3392
3393         * doc/invoke.texi: Correct -Wformat-overflow code sample.
3394
3395 2018-02-14  Martin Sebor  <msebor@redhat.com>
3396
3397         PR tree-optimization/83698
3398         * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For
3399         arrays constrain the offset range to their bounds.
3400         (builtin_access::strcat_overlap): Adjust the bounds of overlap offset.
3401         (builtin_access::overlap): Avoid setting the size of overlap if it's
3402         already been set.
3403         (maybe_diag_overlap): Also consider arrays when deciding what values
3404         of offsets to include in diagnostics.
3405
3406 2018-02-14  Martin Sebor  <msebor@redhat.com>
3407
3408         PR c/84108
3409         * attribs.c (diag_attr_exclusions): Consider the exclusion(s)
3410         that correspond to the kind of a declaration.
3411
3412 2018-02-14  John David Anglin  <danglin@gcc.gnu.org>
3413
3414         PR target/83984
3415         * config/pa/pa.md: Load address of PIC label using the linkage table
3416         if the label is nonlocal.
3417
3418 2018-02-14  Kelvin Nilsen  <kelvin@gcc.gnu.org>
3419
3420         * config/rs6000/rs6000.c (rs6000_option_override_internal): Issue
3421         warning message if user requests -maltivec=be.
3422         * doc/invoke.texi: Document deprecation of -maltivec=be.
3423
3424 2018-02-14  Will Schmidt  <will_schmidt@vnet.ibm.com>
3425
3426         PR target/84220
3427         * config/rs6000/rs6000-c.c: Update definitions for
3428         ALTIVEC_BUILTIN_VEC_SLD, ALTIVEC_BUILTIN_VEC_SLDW,
3429         VEC_XXSLDWI and ALTIVEC_BUILTIN_VEC_XXPERMDI builtins.
3430
3431 2018-02-14  Igor Tsimbalist  <igor.v.tsimbalist@intel.com>
3432
3433         PR target/84239
3434         * config/i386/cetintrin.h: Remove _rdssp[d|q] and
3435         add _get_ssp intrinsics. Remove argument from
3436         __builtin_ia32_rdssp[d|q].
3437         * config/i386/i386-builtin-types.def: Add UINT_FTYPE_VOID.
3438         * config/i386/i386-builtin.def: Remove argument from
3439         __builtin_ia32_rdssp[d|q].
3440         * config/i386/i386.c: Use UINT_FTYPE_VOID. Use
3441         ix86_expand_special_args_builtin for _rdssp[d|q].
3442         * config/i386/i386.md: Remove argument from rdssp[si|di] insn.
3443         Clear register before usage.
3444         * doc/extend.texi: Remove argument from __builtin_ia32_rdssp[d|q].
3445         Add documentation for new _get_ssp and _inc_ssp intrinsics.
3446
3447 2018-02-14  Richard Sandiford  <richard.sandiford@linaro.org>
3448
3449         PR tree-optimization/84357
3450         * tree-data-ref.c (object_address_invariant_in_loop_p): Check
3451         operand 1 of an ARRAY_REF too.
3452
3453 2018-02-14  Oleg Endo  <olegendo@gcc.gnu.org>
3454
3455         PR target/83831
3456         * config/rx/rx-protos.h (rx_reg_dead_or_unused_after_insn,
3457         rx_copy_reg_dead_or_unused_notes, rx_fuse_in_memory_bitop): New
3458         declarations.
3459         (set_of_reg): New struct.
3460         (rx_find_set_of_reg, rx_find_use_of_reg): New functions.
3461         * config/rx/rx.c (rx_reg_dead_or_unused_after_insn,
3462         rx_copy_reg_dead_or_unused_notes, rx_fuse_in_memory_bitop): New
3463         functions.
3464         * config/rx/rx.md (andsi3, iorsi3, xorsi3): Convert to insn_and_split.
3465         Split into bitclr, bitset, bitinvert patterns if appropriate.
3466         (*bitset, *bitinvert, *bitclr): Convert to named insn_and_split and
3467         use rx_fuse_in_memory_bitop.
3468         (*bitset_in_memory, *bitinvert_in_memory, *bitclr_in_memory): Convert
3469         to named insn, correct maximum insn length.
3470
3471 2018-02-14  Jozef Lawrynowicz  <jozefl.gcc@gmail.com>
3472
3473         PR target/79242
3474         * machmode.def: Define a complex mode for PARTIAL_INT.
3475         * genmodes.c (complex_class): Return MODE_COMPLEX_INT for
3476         MODE_PARTIAL_INT.
3477         * doc/rtl.texi: Document CSPImode.
3478         * config/msp430/msp430.c (msp430_hard_regno_nregs): Add CPSImode
3479         handling.
3480         (msp430_hard_regno_nregs_with_padding): Likewise.
3481
3482 2018-02-13  Peter Bergner  <bergner@vnet.ibm.com>
3483
3484         PR target/84279
3485         * config/rs6000/rs6000.c (mem_operand_gpr): Disallow altivec addresses.
3486
3487 2018-02-13  Segher Boessenkool  <segher@kernel.crashing.org>
3488
3489         PR rtl-optimization/84169
3490         * combine.c (try_combine): New variable split_i2i3.  Set it to true if
3491         we generated a parallel as new i3 and we split that to new i2 and i3
3492         instructions.  Handle split_i2i3 similar to swap_i2i3: scan the
3493         LOG_LINKs of i3 to see which of those need to link to i2 now.  Link
3494         those to i2, not i1.  Partially rewrite this scan code.
3495
3496 2018-02-13  Jakub Jelinek  <jakub@redhat.com>
3497
3498         PR c/82210
3499         * stor-layout.c (place_field): For variable length fields, adjust
3500         offset_align afterwards not just based on the field's alignment,
3501         but also on the size.
3502
3503         PR middle-end/84309
3504         * match.pd (pow(C,x) -> exp(log(C)*x)): Use exp2s and log2s instead
3505         of exps and logs in the use_exp2 case.
3506
3507 2018-02-13  Jeff Law  <law@redhat.com>
3508
3509         * config/rl/rl78.c (rl78_attribute_table): Fix terminator and
3510         entry for "vector".
3511
3512         * config/rl78/rl78.c (rl78_handle_func_attribute): Mark
3513         ARGS as unused.
3514
3515 2018-02-13  Alexandre Oliva  <aoliva@redhat.com>
3516
3517         PR debug/84342
3518         PR debug/84319
3519         * common.opt (gas-loc-support, gas-locview-support): New.
3520         (ginline-points, ginternal-reset-location-views): New.
3521         * doc/invoke.texi: Document them.  Use @itemx where intended.
3522         (gvariable-location-views): Adjust.
3523         * target.def (reset_location_view): New.
3524         * doc/tm.texi.in (DWARF2_ASM_VIEW_DEBUG_INFO): New.
3525         (TARGET_RESET_LOCATION_VIEW): New.
3526         * doc/tm.texi: Rebuilt.
3527         * dwarf2out.c (dwarf2out_default_as_loc_support): New.
3528         (dwarf2out_default_as_locview_support): New.
3529         (output_asm_line_debug_info): Use option variables.
3530         (dwarf2out_maybe_output_loclist_view_pair): Likewise.
3531         (output_loc_list): Likewise.
3532         (add_high_low_attributes): Check option variables.
3533         Don't output entry view attribute in strict mode.
3534         (gen_inlined_subroutine_die): Check option variables.
3535         (dwarf2out_inline_entry): Likewise.
3536         (init_sections_and_labels): Likewise.
3537         (dwarf2out_early_finish): Likewise.
3538         (maybe_reset_location_view): New, from...
3539         (dwarf2out_var_location): ... here.  Call it.
3540         * debug.h (dwarf2out_default_as_loc_support): Declare.
3541         (dwarf2out_default_as_locview_support): Declare.
3542         * hooks.c (hook_int_rtx_insn_0): New.
3543         * hooks.h (hook_int_rtx_insn_0): Declare.
3544         * toplev.c (process_options): Take -gas-loc-support and
3545         -gas-locview-support from dwarf2out.  Enable
3546         -gvariable-location-views by default only with locview
3547         assembler support.  Enable -ginternal-reset-location-views by
3548         default only if the target defines the corresponding hook.
3549         Enable -ginline-points by default if location views are
3550         enabled; force it disabled if statement frontiers are
3551         disabled.
3552         * tree-inline.c (expand_call_inline): Check option variables.
3553         * tree-ssa-live.c (remove_unused_scope_block_p): Likewise.
3554
3555 2018-02-13  Richard Sandiford  <richard.sandiford@linaro.org>
3556
3557         PR tree-optimization/84321
3558         * tree-vrp.c (intersect_range_with_nonzero_bits): Fix VR_ANTI_RANGE
3559         handling.  Also check whether the anti-range contains any values
3560         that satisfy the mask; switch to a VR_RANGE if not.
3561
3562 2018-02-13  Paolo Bonzini  <bonzini@gnu.org>
3563
3564         PR sanitizer/84340
3565         * internal-fn.def (ASAN_CHECK, ASAN_MARK): Revert changes to fnspec.
3566
3567 2018-02-13  Martin Jambor  <mjambor@suse.cz>
3568
3569         PR c++/83990
3570         * ipa-param-manipulation.c (ipa_modify_call_arguments): Use location
3571         of call statements, also set location of a load to a temporary.
3572
3573 2018-02-13  Sebastian Perta  <sebastian.perta@renesas.com>
3574
3575         * config/rl78/rl78.c (add_vector_labels): New function.
3576         * config/rl78/rl78.c (rl78_handle_vector_attribute): New function.
3577         * config/rl78/rl78.c (rl78_start_function): Call add_vector_labels.
3578         * config/rl78/rl78.c (rl78_handle_func_attribute): Removed the assert
3579         which checks that no arguments are passed.
3580         * config/rl78/rl78.c (rl78_attribute_table): Add "vector" attribute.
3581         * doc/extend.texi: Documentation for the new attribute.
3582
3583 2018-02-13  Andreas Schwab  <schwab@suse.de>
3584
3585         * config/riscv/linux.h (CPP_SPEC): Define.
3586
3587 2018-02-13  Jakub Jelinek  <jakub@redhat.com>
3588
3589         PR target/84335
3590         * config/i386/i386.c (ix86_init_mmx_sse_builtins): Pass
3591         OPTION_MASK_ISA_AES | OPTION_MASK_ISA_SSE2 instead of
3592         OPTION_MASK_ISA_AES as first argument to def_builtin_const
3593         for AES builtins.  Pass OPTION_MASK_ISA_PCLMUL | OPTION_MASK_ISA_SSE2
3594         instead of OPTION_MASK_ISA_PCLMUL as first argument to
3595         def_builtin_const for __builtin_ia32_pclmulqdq128 builtin.
3596         * config/i386/wmmintrin.h: If __SSE2__ is not defined, enable it
3597         temporarily for AES and PCLMUL builtins.
3598
3599         PR tree-optimization/84339
3600         * gimple-fold.c (get_range_strlen): Set *FLEXP to true when handling
3601         ARRAY_REF where first operand is array_at_struct_end_p COMPONENT_REF.
3602         Formatting fixes.
3603
3604         PR middle-end/84309
3605         * match.pd (pow(C,x) -> exp(log(C)*x)): Optimize instead into
3606         exp2(log2(C)*x) if C is a power of 2 and c99 runtime is available.
3607         * generic-match-head.c (canonicalize_math_after_vectorization_p): New
3608         inline function.
3609         * gimple-match-head.c (canonicalize_math_after_vectorization_p): New
3610         inline function.
3611         * omp-simd-clone.h: New file.
3612         * omp-simd-clone.c: Include omp-simd-clone.h.
3613         (expand_simd_clones): No longer static.
3614         * tree-vect-patterns.c: Include fold-const-call.h, attribs.h,
3615         cgraph.h and omp-simd-clone.h.
3616         (vect_recog_pow_pattern): Optimize pow(C,x) to exp(log(C)*x).
3617         (vect_recog_widen_shift_pattern): Formatting fix.
3618         (vect_pattern_recog_1): Don't check optab for calls.
3619
3620         PR target/84336
3621         * config/i386/sse.md (<avx512>_vpermi2var<mode>3_mask): Force
3622         operands[2] into a REG before using gen_lowpart on it.
3623
3624 2018-02-12  Jeff Law  <law@redhat.com>
3625
3626         PR target/83760
3627         * config/sh/sh.c (find_barrier): Consider a sibling call
3628         a barrier as well.
3629
3630         * cse.c (try_back_substitute_reg): Move any REG_ARGS_SIZE note when
3631         successfully back substituting a reg.
3632
3633 2018-02-12  Richard Biener  <rguenther@suse.de>
3634
3635         PR tree-optimization/84037
3636         * tree-vect-slp.c (vect_analyze_slp_cost): Add visited
3637         parameter, move visited init to caller.
3638         (vect_slp_analyze_operations): Separate cost from validity
3639         check, initialize visited once for all instances.
3640         (vect_schedule_slp): Analyze map to CSE vectorized nodes once
3641         for all instances.
3642         * tree-vect-stmts.c (vect_model_simple_cost): Make early
3643         out an assert.
3644         (vect_model_promotion_demotion_cost): Likewise.
3645         (vectorizable_bswap): Guard cost modeling with !slp_node
3646         instead of !PURE_SLP_STMT to avoid double-counting on hybrid
3647         SLP stmts.
3648         (vectorizable_call): Likewise.
3649         (vectorizable_conversion): Likewise.
3650         (vectorizable_assignment): Likewise.
3651         (vectorizable_shift): Likewise.
3652         (vectorizable_operation): Likewise.
3653         (vectorizable_store): Likewise.
3654         (vectorizable_load): Likewise.
3655         (vectorizable_condition): Likewise.
3656         (vectorizable_comparison): Likewise.
3657
3658 2018-02-12  Paolo Bonzini  <bonzini@gnu.org>
3659
3660         PR sanitizer/84307
3661         * internal-fn.def (ASAN_CHECK): Fix fnspec to account for return value.
3662         (ASAN_MARK): Fix fnspec to account for return value, change pointer
3663         argument from 'R' to 'W' so that the pointed-to datum is clobbered.
3664
3665 2018-02-08  Jan Hubicka  <hubicka@ucw.cz>
3666
3667         PR middle-end/83665
3668         * params.def (inline-min-speedup): Increase from 8 to 15.
3669         (max-inline-insns-auto): Decrease from 40 to 30.
3670         * ipa-split.c (consider_split): Add some buffer for function to
3671         be considered inlining candidate.
3672         * invoke.texi (max-inline-insns-auto, inline-min-speedup): UPdate
3673         default values.
3674
3675 2018-02-12  Richard Biener  <rguenther@suse.de>
3676
3677         PR tree-optimization/84037
3678         * tree-vect-slp.c (vect_build_slp_tree_2): Try swapping the
3679         matched stmts if we cannot swap the non-matched ones.
3680
3681 2018-02-12  Olga Makhotina  <olga.makhotina@intel.com>
3682
3683         * config/i386/avx512fintrin.h (_mm_mask_scalef_round_sd,
3684         _mm_maskz_scalef_round_sd, _mm_mask_scalef_round_ss,
3685         _mm_maskz_scalef_round_ss): New intrinsics.
3686         (__builtin_ia32_scalefsd_round, __builtin_ia32_scalefss_round): Fix.
3687         * config/i386/i386-builtin.def (__builtin_ia32_scalefsd_round,
3688         __builtin_ia32_scalefss_round): Remove.
3689         (__builtin_ia32_scalefsd_mask_round,
3690         __builtin_ia32_scalefss_mask_round): New intrinsics.
3691         * config/i386/sse.md (vmscalef<mode><round_name>): Renamed to ...
3692         (vmscalef<mode><mask_scalar_name><round_scalar_name>): ... this.
3693         ((match_operand:VF_128 2 "<round_nimm_predicate>"
3694         "<round_constraint>")): Changed to ...
3695         ((match_operand:VF_128 2 "<round_scalar_nimm_predicate>"
3696         "<round_scalar_constraint>")): ... this.
3697         ("vscalef<ssescalarmodesuffix>\t{<round_op3>%2, %1, %0|
3698         %0, %1, %2<round_op3>}"): Changed to ...
3699         ("vscalef<ssescalarmodesuffix>\t{<round_scalar_mask_op3>%2, %1,
3700         %0<mask_scalar_operand3>|%0<mask_scalar_operand3>, %1,
3701         %2<round_scalar_mask_op3>}"): ... this.
3702         * config/i386/subst.md (round_scalar_nimm_predicate): New.
3703
3704 2018-02-12  Olga Makhotina  <olga.makhotina@intel.com>
3705
3706         * config/i386/avx512fintrin.h (_mm_mask_sqrt_round_sd)
3707         (_mm_maskz_sqrt_round_sd, _mm_mask_sqrt_round_ss)
3708         (_mm_maskz_sqrt_round_ss): New intrinsics.
3709         (__builtin_ia32_sqrtsd_round, __builtin_ia32_sqrtss_round): Remove.
3710         (__builtin_ia32_sqrtsd_mask_round)
3711         (__builtin_ia32_sqrtss_mask_round): New builtins.
3712         * config/i386/i386-builtin.def (__builtin_ia32_sqrtsd_round)
3713         (__builtin_ia32_sqrtss_round): Remove.
3714         (__builtin_ia32_sqrtsd_mask_round)
3715         (__builtin_ia32_sqrtss_mask_round): New builtins.
3716         * config/i386/sse.md (vmsqrt<mode>2<round_name>): Renamed to ...
3717         (vmsqrt<mode>2<mask_scalar_name><round_scalar_name>): ... this.
3718         ((match_operand:VF_128 1 "vector_operand"
3719         "xBm,<round_constraint>")): Changed to ...
3720         ((match_operand:VF_128 1 "vector_operand"
3721         "xBm,<round_scalar_constraint>")): ... this.
3722         (vsqrt<ssescalarmodesuffix>\t{<round_op3>%1, %2, %0|
3723         %0, %2, %<iptr>1<round_op3>}): Changed to ...
3724         (vsqrt<ssescalarmodesuffix>\t{<round_scalar_mask_op3>%1, %2,
3725         %0<mask_scalar_operand3>|%0<mask_scalar_operand3>, %2,
3726         %<iptr>1<round_scalar_mask_op3>}): ... this.
3727         ((set_attr "prefix" "<round_prefix>")): Changed to ...
3728         ((set_attr "prefix" "<round_scalar_prefix>")): ... this.
3729
3730 2018-02-11  Steven Munroe  <munroesj@gcc.gnu.org>
3731
3732         PR target/84266
3733         * config/rs6000/mmintrin.h (_mm_cmpeq_pi32 [_ARCH_PWR9]):
3734         Cast vec_cmpeq result to correct type.
3735         * config/rs6000/mmintrin.h (_mm_cmpgt_pi32 [_ARCH_PWR9]):
3736         Cast vec_cmpgt result to correct type.
3737
3738 2018-02-11  Alexandre Oliva  <aoliva@redhat.com>
3739
3740         * final.c (final_scan_insn_1): Renamed from...
3741         (final_scan_insn): ... this.  New wrapper, to recover
3742         seen from the outermost call in recursive ones.
3743         * config/sparc/sparc.c (output_return): Drop seen from call.
3744         (output_sibcall): Likewise.
3745         * config/visium/visium.c (output_branch): Likewise.
3746
3747 2018-02-10  John David Anglin  <danglin@gcc.gnu.org>
3748
3749         * config/pa/pa.c (hppa_profile_hook): Mark SYMBOL_REF for _mcount as
3750         function label.
3751
3752 2018-02-10  Alan Modra  <amodra@gmail.com>
3753
3754         PR target/84300
3755         * config/rs6000/rs6000.md (split_stack_return): Remove (use ..).
3756         Specify LR as an input.
3757
3758 2018-02-10  Jakub Jelinek  <jakub@redhat.com>
3759
3760         PR sanitizer/83987
3761         * omp-low.c (maybe_remove_omp_member_access_dummy_vars,
3762         remove_member_access_dummy_vars): New functions.
3763         (lower_omp_for, lower_omp_taskreg, lower_omp_target,
3764         lower_omp_1, execute_lower_omp): Use them.
3765
3766         PR rtl-optimization/84308
3767         * shrink-wrap.c (spread_components): Release todo vector.
3768
3769 2018-02-09  Vladimir Makarov  <vmakarov@redhat.com>
3770
3771         PR rtl-optimization/57193
3772         * ira-color.c (struct allocno_color_data): Add member
3773         conflict_allocno_hard_prefs.
3774         (update_conflict_allocno_hard_prefs): New.
3775         (bucket_allocno_compare_func): Add a preference based on
3776         conflict_allocno_hard_prefs.
3777         (push_allocno_to_stack): Update conflict_allocno_hard_prefs.
3778         (color_allocnos): Remove a dead code.  Initiate
3779         conflict_allocno_hard_prefs.  Call update_costs_from_prefs.
3780
3781 2018-02-09  Jakub Jelinek  <jakub@redhat.com>
3782
3783         PR target/84226
3784         * config/rs6000/vsx.md (p9_xxbrq_v16qi): Change input operand
3785         constraint from =wa to wa.  Avoid a subreg on the output operand,
3786         instead use a pseudo and subreg it in a move.
3787         (p9_xxbrd_<mode>): Changed to ...
3788         (p9_xxbrd_v2di): ... this insn, without VSX_D iterator.
3789         (p9_xxbrd_v2df): New expander.
3790         (p9_xxbrw_<mode>): Changed to ...
3791         (p9_xxbrw_v4si): ... this insn, without VSX_W iterator.
3792         (p9_xxbrw_v4sf): New expander.
3793
3794 2018-02-09  Sebastian Perta  <sebastian.perta@renesas.com>
3795
3796         * config/rx/rx.md (movsicc): Update expander to be matched by GCC.
3797
3798 2018-02-09  Peter Bergner  <bergner@vnet.ibm.com>
3799
3800         PR target/83926
3801         * config/rs6000/vsx.md (vsx_mul_v2di): Handle generating a 64-bit
3802         multiply in 32-bit mode.
3803         (vsx_div_v2di): Handle generating a 64-bit signed divide in 32-bit mode.
3804         (vsx_udiv_v2di): Handle generating a 64-bit unsigned divide in 32-bit
3805         mode.
3806
3807 2018-02-09  Sebastian Perta  <sebastian.perta@renesas.com>
3808
3809         * config/rx/constraints.md (CALL_OP_SYMBOL_REF): Added new constraint
3810         to allow or block "symbol_ref" depending on the value of TARGET_JSR.
3811         * config/rx/rx.md (call_internal): Use CALL_OP_SYMBOL_REF.
3812         * config/rx/rx.md (call_value_internal): Use CALL_OP_SYMBOL_REF.
3813
3814 2018-02-09  Pierre-Marie de Rodat  <derodat@adacore.com>
3815
3816         PR lto/84213
3817         * dwarf2out.c (is_trivial_indirect_ref): New function.
3818         (dwarf2out_late_global_decl): Do not generate a location
3819         attribute for variables that have a non-trivial DECL_VALUE_EXPR
3820         and that are not defined in the current unit.
3821
3822 2018-02-09  Eric Botcazou  <ebotcazou@adacore.com>
3823
3824         * optabs.c (prepare_cmp_insn): Try harder to emit a direct comparison
3825         instead of a libcall for UNORDERED.
3826
3827 2018-02-09  Tamar Christina  <tamar.christina@arm.com>
3828
3829         PR target/82641
3830         * config/arm/arm-c.c (arm_cpu_builtins): Un-define __ARM_FEATURE_LDREX,
3831         __ARM_ARCH_PROFILE, __ARM_ARCH_ISA_THUMB, __ARM_FP and __ARM_NEON_FP.
3832
3833 2018-02-09  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>
3834
3835         PR target/PR84295
3836         * config/s390/s390.c (s390_set_current_function): Invoke
3837         s390_indirect_branch_settings also if fndecl didn't change.
3838
3839 2018-02-09  Alexandre Oliva  <aoliva@redhat.com>
3840
3841         * config/rs6000/rs6000.md (blockage): Set length to zero.
3842
3843 2018-02-09  Eric Botcazou  <ebotcazou@adacore.com>
3844
3845         * expr.c (optimize_bitfield_assignment_op): Remove obsolete assertion.
3846
3847 2018-02-09  Jakub Jelinek  <jakub@redhat.com>
3848
3849         PR sanitizer/84285
3850         * gcc.c (STATIC_LIBASAN_LIBS, STATIC_LIBTSAN_LIBS,
3851         STATIC_LIBLSAN_LIBS, STATIC_LIBUBSAN_LIBS): Handle -static like
3852         -static-lib*san.
3853
3854         PR debug/84252
3855         * var-tracking.c (vt_add_function_parameter): Punt for non-onepart
3856         PARALLEL incoming that failed vt_get_decl_and_offset check.
3857
3858         PR middle-end/84237
3859         * output.h (bss_initializer_p): Add NAMED argument, defaulted to false.
3860         * varasm.c (bss_initializer_p): Add NAMED argument, if true, ignore
3861         TREE_READONLY bit.
3862         (get_variable_section): For decls in named .bss* sections pass true as
3863         second argument to bss_initializer_p.
3864
3865 2018-02-09  Marek Polacek  <polacek@redhat.com>
3866             Jakub Jelinek  <jakub@redhat.com>
3867
3868         PR c++/83659
3869         * fold-const.c (fold_indirect_ref_1): Use VECTOR_TYPE_P macro.
3870         Formatting fixes.  Verify first that tree_fits_poly_int64_p (op01).
3871         Sync some changes from cxx_fold_indirect_ref.
3872
3873 2018-02-09  Alexandre Oliva  <aoliva@redhat.com>
3874
3875         * cfgexpand.c (expand_gimple_basic_block): Handle inline entry
3876         markers.
3877         * dwarf2out.c (dwarf2_debug_hooks): Enable inline_entry hook.
3878         (BLOCK_INLINE_ENTRY_LABEL): New.
3879         (dwarf2out_var_location): Disregard inline entry markers.
3880         (inline_entry_data): New struct.
3881         (inline_entry_data_hasher): New hashtable type.
3882         (inline_entry_data_hasher::hash): New.
3883         (inline_entry_data_hasher::equal): New.
3884         (inline_entry_data_table): New variable.
3885         (add_high_low_attributes): Add DW_AT_entry_pc and
3886         DW_AT_GNU_entry_view attributes if a pending entry is found
3887         in inline_entry_data_table.  Add old entry_pc attribute only
3888         if debug nonbinding markers are disabled.
3889         (gen_inlined_subroutine_die): Set BLOCK_DIE if nonbinding
3890         markers are enabled.
3891         (block_within_block_p, dwarf2out_inline_entry): New.
3892         (dwarf2out_finish): Check that no entries remained in
3893         inline_entry_data_table.
3894         * final.c (reemit_insn_block_notes): Handle inline entry notes.
3895         (final_scan_insn, notice_source_line): Likewise.
3896         (rest_of_clean_state): Skip inline entry markers.
3897         * gimple-pretty-print.c (dump_gimple_debug): Handle inline entry
3898         markers.
3899         * gimple.c (gimple_build_debug_inline_entry): New.
3900         * gimple.h (enum gimple_debug_subcode): Add
3901         GIMPLE_DEBUG_INLINE_ENTRY.
3902         (gimple_build_debug_inline_entry): Declare.
3903         (gimple_debug_inline_entry_p): New.
3904         (gimple_debug_nonbind_marker_p): Adjust.
3905         * insn-notes.def (INLINE_ENTRY): New.
3906         * print-rtl.c (rtx_writer::print_rtx_operand_code_0): Handle
3907         inline entry marker notes.
3908         (print_insn): Likewise.
3909         * rtl.h (NOTE_MARKER_P): Add INLINE_ENTRY support.
3910         (INSN_DEBUG_MARKER_KIND): Likewise.
3911         (GEN_RTX_DEBUG_MARKER_INLINE_ENTRY_PAT): New.
3912         * tree-inline.c (expand_call_inline): Build and insert
3913         debug_inline_entry stmt.
3914         * tree-ssa-live.c (remove_unused_scope_block_p): Preserve
3915         inline entry blocks early, if nonbind markers are enabled.
3916         (dump_scope_block): Dump fragment info.
3917         * var-tracking.c (reemit_marker_as_note): Handle inline entry note.
3918         * doc/gimple.texi (gimple_debug_inline_entry_p): New.
3919         (gimple_build_debug_inline_entry): New.
3920         * doc/invoke.texi (gstatement-frontiers, gno-statement-frontiers):
3921         Enable/disable inline entry points too.
3922         * doc/rtl.texi (NOTE_INSN_INLINE_ENTRY): New.
3923         (DEBUG_INSN): Describe inline entry markers.
3924
3925         * common.opt (gvariable-location-views): New.
3926         (gvariable-location-views=incompat5): New.
3927         * config.in: Rebuilt.
3928         * configure: Rebuilt.
3929         * configure.ac: Test assembler for view support.
3930         * dwarf2asm.c (dw2_asm_output_symname_uleb128): New.
3931         * dwarf2asm.h (dw2_asm_output_symname_uleb128): Declare.
3932         * dwarf2out.c (var_loc_view): New typedef.
3933         (struct dw_loc_list_struct): Add vl_symbol, vbegin, vend.
3934         (dwarf2out_locviews_in_attribute): New.
3935         (dwarf2out_locviews_in_loclist): New.
3936         (dw_val_equal_p): Compare val_view_list of dw_val_class_view_lists.
3937         (enum dw_line_info_opcode): Add LI_adv_address.
3938         (struct dw_line_info_table): Add view.
3939         (RESET_NEXT_VIEW, RESETTING_VIEW_P): New macros.
3940         (DWARF2_ASM_VIEW_DEBUG_INFO): Define default.
3941         (zero_view_p): New variable.
3942         (ZERO_VIEW_P): New macro.
3943         (output_asm_line_debug_info): New.
3944         (struct var_loc_node): Add view.
3945         (add_AT_view_list, AT_loc_list): New.
3946         (add_var_loc_to_decl): Add view param.  Test it against last.
3947         (new_loc_list): Add view params.  Record them.
3948         (AT_loc_list_ptr): Handle loc and view lists.
3949         (view_list_to_loc_list_val_node): New.
3950         (print_dw_val): Handle dw_val_class_view_list.
3951         (size_of_die): Likewise.
3952         (value_format): Likewise.
3953         (loc_list_has_views): New.
3954         (gen_llsym): Set vl_symbol too.
3955         (maybe_gen_llsym, skip_loc_list_entry): New.
3956         (dwarf2out_maybe_output_loclist_view_pair): New.
3957         (output_loc_list): Output view list or entries too.
3958         (output_view_list_offset): New.
3959         (output_die): Handle dw_val_class_view_list.
3960         (output_dwarf_version): New.
3961         (output_compilation_unit_header): Use it.
3962         (output_skeleton_debug_sections): Likewise.
3963         (output_rnglists, output_line_info): Likewise.
3964         (output_pubnames, output_aranges): Update version comments.
3965         (output_one_line_info_table): Output view numbers in asm comments.
3966         (dw_loc_list): Determine current endview, pass it to new_loc_list.
3967         Call maybe_gen_llsym.
3968         (loc_list_from_tree_1): Adjust.
3969         (add_AT_location_description): Create view list attribute if
3970         needed, check it's absent otherwise.
3971         (convert_cfa_to_fb_loc_list): Adjust.
3972         (maybe_emit_file): Call output_asm_line_debug_info for test.
3973         (dwarf2out_var_location): Reset views as needed.  Precompute
3974         add_var_loc_to_decl args.  Call get_attr_min_length only if we have the
3975         attribute.  Set view.
3976         (new_line_info_table): Reset next view.
3977         (set_cur_line_info_table): Call output_asm_line_debug_info for test.
3978         (dwarf2out_source_line): Likewise.  Output view resets and labels to
3979         the assembler, or select appropriate line info opcodes.
3980         (prune_unused_types_walk_attribs): Handle dw_val_class_view_list.
3981         (optimize_string_length): Catch it.  Adjust.
3982         (resolve_addr): Copy vl_symbol along with ll_symbol.  Handle
3983         dw_val_class_view_list, and remove it if no longer needed.
3984         (hash_loc_list): Hash view numbers.
3985         (loc_list_hasher::equal): Compare them.
3986         (optimize_location_lists): Check whether a view list symbol is
3987         needed, and whether the locview attribute is present, and
3988         whether they match.  Remove the locview attribute if no longer
3989         needed.
3990         (index_location_lists): Call skip_loc_list_entry for test.
3991         (dwarf2out_finish): Call output_asm_line_debug_info for test.
3992         Use output_dwarf_version.
3993         * dwarf2out.h (enum dw_val_class): Add dw_val_class_view_list.
3994         (struct dw_val_node): Add val_view_list.
3995         * final.c (SEEN_NEXT_VIEW): New.
3996         (set_next_view_needed): New.
3997         (clear_next_view_needed): New.
3998         (maybe_output_next_view): New.
3999         (final_start_function): Rename to...
4000         (final_start_function_1): ... this.  Take pointer to FIRST,
4001         add SEEN parameter.  Emit param bindings in the initial view.
4002         (final_start_function): Reintroduce SEEN-less interface.
4003         (final): Rename to...
4004         (final_1): ... this.  Take SEEN parameter.  Output final pending
4005         next view at the end.
4006         (final): Reintroduce seen-less interface.
4007         (final_scan_insn): Output pending next view before switching
4008         sections or ending a block.  Mark the next view as needed when
4009         outputting variable locations.  Notify debug backend of section
4010         changes, and of location view changes.
4011         (rest_of_handle_final): Adjust.
4012         * toplev.c (process_options): Autodetect value for debug variable
4013         location views option.  Warn on incompat5 without -gdwarf-5.
4014         * doc/invoke.texi (gvariable-location-views): New.
4015         (gvariable-location-views=incompat5): New.
4016         (gno-variable-location-views): New.
4017
4018 2018-02-08  David Malcolm  <dmalcolm@redhat.com>
4019
4020         PR tree-optimization/84136
4021         * tree-cfg.c (find_taken_edge_computed_goto): Remove assertion
4022         that the result of find_edge is non-NULL.
4023
4024 2018-02-08  Sergey Shalnov  <sergey.shalnov@intel.com>
4025
4026         PR target/83008
4027         * config/i386/x86-tune-costs.h (skylake_cost): Fix cost of
4028         storing integer register in SImode.  Fix cost of 256 and 512
4029         byte aligned SSE register store.
4030
4031 2018-02-08  Sergey Shalnov  <sergey.shalnov@intel.com>
4032
4033         * config/i386/i386.c (ix86_multiplication_cost): Fix
4034         multiplication cost for TARGET_AVX512DQ.
4035
4036 2018-02-08  Marek Polacek  <polacek@redhat.com>
4037
4038         PR tree-optimization/84238
4039         * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Verify the result of
4040         get_range_strlen.
4041
4042 2018-02-08  Richard Sandiford  <richard.sandiford@linaro.org>
4043
4044         PR tree-optimization/84265
4045         * tree-vect-stmts.c (vectorizable_store): Don't treat
4046         VMAT_CONTIGUOUS accesses as grouped.
4047         (vectorizable_load): Likewise.
4048
4049 2018-02-08  Richard Sandiford  <richard.sandiford@linaro.org>
4050
4051         PR tree-optimization/81635
4052         * wide-int.h (wi::round_down_for_mask, wi::round_up_for_mask): Declare.
4053         * wide-int.cc (wi::round_down_for_mask, wi::round_up_for_mask)
4054         (test_round_for_mask): New functions.
4055         (wide_int_cc_tests): Call test_round_for_mask.
4056         * tree-vrp.h (intersect_range_with_nonzero_bits): Declare.
4057         * tree-vrp.c (intersect_range_with_nonzero_bits): New function.
4058         * tree-data-ref.c (split_constant_offset_1): Use it to refine the
4059         range returned by get_range_info.
4060
4061 2018-02-08  Jan Hubicka  <hubicka@ucw.cz>
4062
4063         PR ipa/81360
4064         * cgraph.h (symtab_node::output_to_lto_symbol_table_p): Declare
4065         * symtab.c: Include builtins.h
4066         (symtab_node::output_to_lto_symbol_table_p): Move here
4067         from lto-streamer-out.c:output_symbol_p.
4068         * lto-streamer-out.c (write_symbol): Turn early exit to assert.
4069         (output_symbol_p): Move all logic to symtab.c
4070         (produce_symtab): Update.
4071
4072 2018-02-08  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>
4073
4074         * config/s390/s390-opts.h (enum indirect_branch): Define.
4075         * config/s390/s390-protos.h (s390_return_addr_from_memory)
4076         (s390_indirect_branch_via_thunk)
4077         (s390_indirect_branch_via_inline_thunk): Add function prototypes.
4078         (enum s390_indirect_branch_type): Define.
4079         * config/s390/s390.c (struct s390_frame_layout, struct
4080         machine_function): Remove.
4081         (indirect_branch_prez10thunk_mask, indirect_branch_z10thunk_mask)
4082         (indirect_branch_table_label_no, indirect_branch_table_name):
4083         Define variables.
4084         (INDIRECT_BRANCH_NUM_OPTIONS): Define macro.
4085         (enum s390_indirect_branch_option): Define.
4086         (s390_return_addr_from_memory): New function.
4087         (s390_handle_string_attribute): New function.
4088         (s390_attribute_table): Add new attribute handler.
4089         (s390_execute_label): Handle UNSPEC_EXECUTE_JUMP patterns.
4090         (s390_indirect_branch_via_thunk): New function.
4091         (s390_indirect_branch_via_inline_thunk): New function.
4092         (s390_function_ok_for_sibcall): When jumping via thunk disallow
4093         sibling call optimization for non z10 compiles.
4094         (s390_emit_call): Force indirect branch target to be a single
4095         register.  Add r1 clobber for non-z10 compiles.
4096         (s390_emit_epilogue): Emit return jump via return_use expander.
4097         (s390_reorg): Handle JUMP_INSNs as execute targets.
4098         (s390_option_override_internal): Perform validity checks for the
4099         new command line options.
4100         (s390_indirect_branch_attrvalue): New function.
4101         (s390_indirect_branch_settings): New function.
4102         (s390_set_current_function): Invoke s390_indirect_branch_settings.
4103         (s390_output_indirect_thunk_function):  New function.
4104         (s390_code_end): Implement target hook.
4105         (s390_case_values_threshold): Implement target hook.
4106         (TARGET_ASM_CODE_END, TARGET_CASE_VALUES_THRESHOLD): Define target
4107         macros.
4108         * config/s390/s390.h (struct s390_frame_layout)
4109         (struct machine_function): Move here from s390.c.
4110         (TARGET_INDIRECT_BRANCH_NOBP_RET)
4111         (TARGET_INDIRECT_BRANCH_NOBP_JUMP)
4112         (TARGET_INDIRECT_BRANCH_NOBP_JUMP_THUNK)
4113         (TARGET_INDIRECT_BRANCH_NOBP_JUMP_INLINE_THUNK)
4114         (TARGET_INDIRECT_BRANCH_NOBP_CALL)
4115         (TARGET_DEFAULT_INDIRECT_BRANCH_TABLE)
4116         (TARGET_INDIRECT_BRANCH_THUNK_NAME_EXRL)
4117         (TARGET_INDIRECT_BRANCH_THUNK_NAME_EX)
4118         (TARGET_INDIRECT_BRANCH_TABLE): Define macros.
4119         * config/s390/s390.md (UNSPEC_EXECUTE_JUMP)
4120         (INDIRECT_BRANCH_THUNK_REGNUM): Define constants.
4121         (mnemonic attribute): Add values which aren't recognized
4122         automatically.
4123         ("*cjump_long", "*icjump_long", "*basr", "*basr_r"): Disable
4124         pattern for branch conversion.  Fix mnemonic attribute.
4125         ("*c<code>", "*sibcall_br", "*sibcall_value_br", "*return"): Emit
4126         indirect branch via thunk if requested.
4127         ("indirect_jump", "<code>"): Expand patterns for branch conversion.
4128         ("*indirect_jump"): Disable for branch conversion using out of
4129         line thunks.
4130         ("indirect_jump_via_thunk<mode>_z10")
4131         ("indirect_jump_via_thunk<mode>")
4132         ("indirect_jump_via_inlinethunk<mode>_z10")
4133         ("indirect_jump_via_inlinethunk<mode>", "*casesi_jump")
4134         ("casesi_jump_via_thunk<mode>_z10", "casesi_jump_via_thunk<mode>")
4135         ("casesi_jump_via_inlinethunk<mode>_z10")
4136         ("casesi_jump_via_inlinethunk<mode>", "*basr_via_thunk<mode>_z10")
4137         ("*basr_via_thunk<mode>", "*basr_r_via_thunk_z10")
4138         ("*basr_r_via_thunk", "return<mode>_prez10"): New pattern.
4139         ("*indirect2_jump"): Disable for branch conversion.
4140         ("casesi_jump"): Turn into expander and expand patterns for branch
4141         conversion.
4142         ("return_use"): New expander.
4143         ("*return"): Emit return via thunk and rename it to ...
4144         ("*return<mode>"): ... this one.
4145         * config/s390/s390.opt: Add new options and and enum for the
4146         option values.
4147
4148 2018-02-08  Richard Sandiford  <richard.sandiford@linaro.org>
4149
4150         * lra-constraints.c (match_reload): Unconditionally use
4151         gen_lowpart_SUBREG, rather than selecting between that
4152         and equivalent gen_rtx_SUBREG code.
4153
4154 2018-02-08  Richard Biener  <rguenther@suse.de>
4155
4156         PR tree-optimization/84233
4157         * tree-ssa-phiprop.c (propagate_with_phi): Use separate
4158         changed flag instead of boguously re-using phi_inserted.
4159
4160 2018-02-08  Martin Jambor  <mjambor@suse.cz>
4161
4162         * hsa-gen.c (get_symbol_for_decl): Set program allocation for
4163         static local variables.
4164
4165 2018-02-08  Richard Biener  <rguenther@suse.de>
4166
4167         PR tree-optimization/84278
4168         * tree-vect-stmts.c (vectorizable_store): When looking for
4169         smaller vector types to perform grouped strided loads/stores
4170         make sure the mode is supported by the target.
4171         (vectorizable_load): Likewise.
4172
4173 2018-02-08  Wilco Dijkstra  <wdijkstr@arm.com>
4174
4175         * config/aarch64/aarch64.c (aarch64_components_for_bb):
4176         Increase LDP/STP opportunities by adding adjacent callee-saves.
4177
4178 2018-02-08  Wilco Dijkstra  <wdijkstr@arm.com>
4179
4180         PR rtl-optimization/84068
4181         PR rtl-optimization/83459
4182         * haifa-sched.c (rank_for_schedule): Fix SCHED_PRESSURE_MODEL sorting.
4183
4184 2018-02-08  Aldy Hernandez  <aldyh@redhat.com>
4185
4186         PR tree-optimization/84224
4187         * gimple-ssa-warn-alloca.c (pass_walloca::execute): Remove assert.
4188         * calls.c (gimple_alloca_call_p): Only return TRUE when we have
4189         non-zero arguments.
4190
4191 2018-02-07  Iain Sandoe  <iain@codesourcery.com>
4192
4193         PR target/84113
4194         * config/rs6000/altivec.md (*restore_world): Remove LR use.
4195         * config/rs6000/predicates.md (restore_world_operation): Adjust op
4196         count, remove one USE.
4197
4198 2018-02-07  Michael Meissner  <meissner@linux.vnet.ibm.com>
4199
4200         * doc/install.texi (Configuration): Document the
4201         --with-long-double-format={ibm,ieee} PowerPC configuration
4202         options.
4203
4204         PR target/84154
4205         * config/rs6000/rs6000.md (fix_trunc<SFDF:mode><QHI:mode>2):
4206         Convert from define_expand to be define_insn_and_split.  Rework
4207         float/double/_Float128 conversions to QI/HI/SImode to work with
4208         both ISA 2.07 (power8) or ISA 3.0 (power9).  Fix regression where
4209         conversions to QI/HImode types did a store and then a load to
4210         truncate the value.  For conversions to VSX registers, don't split
4211         the insn, instead emit the code directly.  Use the code iterator
4212         any_fix to combine signed and unsigned conversions.
4213         (fix<uns>_trunc<SFDF:mode>si2_p8): Likewise.
4214         (fixuns_trunc<SFDF:mode><QHI:mode>2): Likewise.
4215         (fix_trunc<IEEE128:mode><QHI:mode>2): Likewise.
4216         (fix<uns>_trunc<SFDF:mode><QHI:mode>2): Likewise.
4217         (fix_<mode>di2_hw): Likewise.
4218         (fixuns_<mode>di2_hw): Likewise.
4219         (fix_<mode>si2_hw): Likewise.
4220         (fixuns_<mode>si2_hw): Likewise.
4221         (fix<uns>_<IEEE128:mode><SDI:mode>2_hw): Likewise.
4222         (fix<uns>_trunc<IEEE128:mode><QHI:mode>2): Likewise.
4223         (fctiw<u>z_<mode>_smallint): Rename fctiw<u>z_<mode>_smallint to
4224         fix<uns>_trunc<SFDF:mode>si2_p8.
4225         (fix_trunc<SFDF:mode><QHI:mode>2_internal): Delete, no longer
4226         used.
4227         (fixuns_trunc<SFDF:mode><QHI:mode>2_internal): Likewise.
4228         (fix<uns>_<mode>_mem): Likewise.
4229         (fctiw<u>z_<mode>_mem): Likewise.
4230         (fix<uns>_<mode>_mem): Likewise.
4231         (fix<uns>_trunc<SFDF:mode><QHSI:mode>2_mem): On ISA 3.0, prevent
4232         the register allocator from doing a direct move to the GPRs to do
4233         a store, and instead use the ISA 3.0 store byte/half-word from
4234         vector register instruction.  For IEEE 128-bit floating point,
4235         also optimize stores of 32-bit ints.
4236         (fix<uns>_trunc<IEEE128:mode><QHSI:mode>2_mem): Likewise.
4237
4238 2018-02-07  Alan Hayward  <alan.hayward@arm.com>
4239
4240         * genextract.c (push_pathstr_operand): New function to support
4241         [a-zA-Z].
4242         (walk_rtx): Call push_pathstr_operand.
4243         (print_path): Support [a-zA-Z].
4244
4245 2018-02-07  Richard Biener  <rguenther@suse.de>
4246
4247         PR tree-optimization/84037
4248         * tree-vectorizer.h (struct _loop_vec_info): Add ivexpr_map member.
4249         (cse_and_gimplify_to_preheader): Declare.
4250         (vect_get_place_in_interleaving_chain): Likewise.
4251         * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
4252         ivexpr_map.
4253         (_loop_vec_info::~_loop_vec_info): Delete it.
4254         (cse_and_gimplify_to_preheader): New function.
4255         * tree-vect-slp.c (vect_get_place_in_interleaving_chain): Export.
4256         * tree-vect-stmts.c (vectorizable_store): CSE base and steps.
4257         (vectorizable_load): Likewise.  For grouped stores always base
4258         the IV on the first element.
4259         * tree-vect-loop-manip.c (vect_loop_versioning): Unshare versioning
4260         condition before gimplifying.
4261
4262 2018-02-07  Jakub Jelinek  <jakub@redhat.com>
4263
4264         * tree-eh.c (operation_could_trap_helper_p): Ignore honor_trapv for
4265         *DIV_EXPR and *MOD_EXPR.
4266
4267 2018-02-07  H.J. Lu  <hongjiu.lu@intel.com>
4268
4269         PR target/84248
4270         * config/i386/i386.c (ix86_option_override_internal): Mask out
4271         the CF_SET bit when checking -fcf-protection.
4272
4273 2018-02-07  Tom de Vries  <tom@codesourcery.com>
4274
4275         PR libgomp/84217
4276         * omp-expand.c (expand_oacc_collapse_init): Ensure diff_type is large
4277         enough.
4278
4279 2018-02-07  Richard Biener  <rguenther@suse.de>
4280
4281         PR tree-optimization/84204
4282         * tree-chrec.c (chrec_fold_plus_1): Remove size limiting in
4283         this place.
4284
4285         PR tree-optimization/84205
4286         * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Also
4287         special-case isl_ast_op_zdiv_r.
4288
4289         PR tree-optimization/84223
4290         * graphite-scop-detection.c (gather_bbs::before_dom_children):
4291         Only add conditions from within the region.
4292         (gather_bbs::after_dom_children): Adjust.
4293
4294 2018-02-07  Georg-Johann Lay  <avr@gjlay.de>
4295
4296         PR target/84209
4297         * config/avr/avr.h (GENERAL_REGNO_P, GENERAL_REG_P): New macros.
4298         * config/avr/avr.md: Only post-reload split REG-REG moves if
4299         either register is GENERAL_REG_P.
4300
4301 2018-02-07  Jakub Jelinek  <jakub@redhat.com>
4302
4303         PR tree-optimization/84235
4304         * tree-ssa-scopedtables.c
4305         (avail_exprs_stack::simplify_binary_operation): Fir MINUS_EXPR, punt
4306         if the subtraction is performed in floating point type where NaNs are
4307         honored.  For *DIV_EXPR, punt for ALL_FRACT_MODE_Ps where we can't
4308         build 1.  Formatting fix.
4309
4310 2018-02-06  Jakub Jelinek  <jakub@redhat.com>
4311
4312         PR target/84146
4313         * config/i386/i386.c (rest_of_insert_endbranch): Only skip
4314         NOTE_INSN_CALL_ARG_LOCATION after a call, not anything else,
4315         and skip it regardless of bb boundaries.  Use CALL_P macro,
4316         don't test INSN_P (insn) together with CALL_P or JUMP_P check
4317         unnecessarily, formatting fix.
4318
4319 2018-02-06  Michael Collison  <michael.collison@arm.com>
4320
4321         * config/arm/thumb2.md:
4322         (*thumb2_mov_negscc): Split only if TARGET_THUMB2 && !arm_restrict_it.
4323         (*thumb_mov_notscc): Ditto.
4324
4325 2018-02-06  Michael Meissner  <meissner@linux.vnet.ibm.com>
4326
4327         PR target/84154
4328         * config/rs6000/rs6000.md (su code attribute): Use "u" for
4329         unsigned_fix, not "s".
4330
4331 2018-02-06  Rainer Orth  <ro@CeBiTec.Uni-Bielefeld.DE>
4332
4333         * configure.ac (gcc_fn_eh_frame_ro): New function.
4334         (gcc_cv_as_cfi_directive): Check both 32 and 64-bit assembler for
4335         correct .eh_frame permissions.
4336         * configure: Regenerate.
4337
4338 2018-02-06  Andrew Jenner  <andrew@codeourcery.com>
4339
4340         * doc/invoke.texi: Add section for the PowerPC SPE backend. Remove
4341         irrelevant options.
4342
4343 2018-02-06  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
4344
4345         * config/rs6000/rs6000.c (rs6000_option_override_internal):
4346         Display warning message for -mno-speculate-indirect-jumps.
4347
4348 2018-02-06  Andrew Jenner  <andrew@codesourcery.com>
4349
4350         * config/powerpcspe/powerpcspe.opt: (msimple-fpu, mfpu) Add
4351         Undocumented.
4352         * config/powerpcspe/sysv4.opt (mbit-align): Likewise.
4353
4354 2018-02-06  Aldy Hernandez  <aldyh@redhat.com>
4355
4356         PR tree-optimization/84225
4357         * tree-eh.c (find_trapping_overflow): Only call
4358         operation_no_trapping_overflow when ANY_INTEGRAL_TYPE_P.
4359
4360 2018-02-06  Igor Tsimbalist  <igor.v.tsimbalist@intel.com>
4361
4362         PR target/84145
4363         * config/i386/i386.c: Reimplement the check of possible options
4364         -mibt/-mshstk conbination. Change error messages.
4365         * doc/invoke.texi: Fix a typo: remove extra '='.
4366
4367 2018-02-06  Marek Polacek  <polacek@redhat.com>
4368
4369         PR tree-optimization/84228
4370         * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Skip debug statements.
4371
4372 2018-02-06  Tamar Christina  <tamar.christina@arm.com>
4373
4374         PR target/82641
4375         * config/arm/arm.c (arm_print_asm_arch_directives): Record already
4376         emitted arch directives.
4377         * config/arm/arm-c.c (arm_cpu_builtins): Undefine __ARM_ARCH and
4378         __ARM_FEATURE_COPROC before changing architectures.
4379
4380 2018-02-06  Richard Biener  <rguenther@suse.de>
4381
4382         * config/i386/i386.c (print_reg): Fix typo.
4383         (ix86_loop_unroll_adjust): Do not unroll beyond the original nunroll.
4384
4385 2018-02-06  Eric Botcazou  <ebotcazou@adacore.com>
4386
4387         * configure: Regenerate.
4388
4389 2018-02-05  Martin Sebor  <msebor@redhat.com>
4390
4391         PR tree-optimization/83369
4392         * tree-ssa-ccp.c (pass_post_ipa_warn::execute): Use %G to print
4393         inlining context.
4394
4395 2018-02-05  Martin Liska  <mliska@suse.cz>
4396
4397         * doc/invoke.texi: Cherry-pick upstream r323995.
4398
4399 2018-02-05  Richard Sandiford  <richard.sandiford@linaro.org>
4400
4401         * ira.c (ira_init_register_move_cost): Adjust comment.
4402
4403 2018-02-05  Martin Liska  <mliska@suse.cz>
4404
4405         PR gcov-profile/84137
4406         * doc/gcov.texi: Fix typo in documentation.
4407
4408 2018-02-05  Martin Liska  <mliska@suse.cz>
4409
4410         PR gcov-profile/83879
4411         * doc/gcov.texi: Document necessity of --dynamic-list-data when
4412         using dlopen functionality.
4413
4414 2018-02-05  Olga Makhotina  <olga.makhotina@intel.com>
4415
4416         * config/i386/avx512dqintrin.h (_mm_mask_range_sd, _mm_maskz_range_sd,
4417         _mm_mask_range_round_sd, _mm_maskz_range_round_sd, _mm_mask_range_ss,
4418         _mm_maskz_range_ss, _mm_mask_range_round_ss,
4419         _mm_maskz_range_round_ss): New intrinsics.
4420         (__builtin_ia32_rangesd128_round)
4421         (__builtin_ia32_rangess128_round): Remove.
4422         (__builtin_ia32_rangesd128_mask_round,
4423         __builtin_ia32_rangess128_mask_round): New builtins.
4424         * config/i386/i386-builtin.def (__builtin_ia32_rangesd128_round,
4425         __builtin_ia32_rangess128_round): Remove.
4426         (__builtin_ia32_rangesd128_mask_round,
4427         __builtin_ia32_rangess128_mask_round): New builtins.
4428         * config/i386/sse.md (ranges<mode><round_saeonly_name>): Renamed to ...
4429         (ranges<mode><mask_scalar_name><round_saeonly_scalar_name>): ... this.
4430         ((match_operand:VF_128 2 "<round_saeonly_nimm_predicate>"
4431         "<round_saeonly_constraint>")): Changed to ...
4432         ((match_operand:VF_128 2 "<round_saeonly_scalar_nimm_predicate>"
4433         "<round_saeonly_scalar_constraint>")): ... this.
4434         ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_op4>%2, %1, %0|
4435         %0, %1, %2<round_saeonly_op4>, %3}"): Changed to ...
4436         ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_scalar_mask_op4>%2,
4437         %1, %0<mask_scalar_operand4>|%0<mask_scalar_operand4>, %1,
4438         %2<round_saeonly_scalar_mask_op4>, %3}"): ... this.
4439
4440 2018-02-02  Andrew Jenner  <andrew@codesourcery.com>
4441
4442         * config/powerpcspe/powerpcspe.opt: Add Undocumented to irrelevant
4443         options.
4444         * config/powerpcspe/powerpcspe-tables.opt (rs6000_cpu_opt_value):
4445         Remove all values except native, 8540 and 8548.
4446
4447 2018-02-02  H.J. Lu  <hongjiu.lu@intel.com>
4448
4449         * config/i386/i386.c (ix86_output_function_return): Pass
4450         INVALID_REGNUM, instead of -1, as invalid register number to
4451         indirect_thunk_name and output_indirect_thunk.
4452
4453 2018-02-02  Julia Koval  <julia.koval@intel.com>
4454
4455         * config.gcc: Add -march=icelake.
4456         * config/i386/driver-i386.c (host_detect_local_cpu): Detect icelake.
4457         * config/i386/i386-c.c (ix86_target_macros_internal): Handle icelake.
4458         * config/i386/i386.c (processor_costs): Add m_ICELAKE.
4459         (PTA_ICELAKE, PTA_AVX512VNNI, PTA_GFNI, PTA_VAES, PTA_AVX512VBMI2,
4460         PTA_VPCLMULQDQ, PTA_RDPID, PTA_AVX512BITALG): New.
4461         (processor_target_table): Add icelake.
4462         (ix86_option_override_internal): Handle new PTAs.
4463         (get_builtin_code_for_version): Handle icelake.
4464         (M_INTEL_COREI7_ICELAKE): New.
4465         (fold_builtin_cpu): Handle icelake.
4466         * config/i386/i386.h (TARGET_ICELAKE, PROCESSOR_ICELAKE): New.
4467         * doc/invoke.texi: Add -march=icelake.
4468
4469 2018-02-02  Julia Koval  <julia.koval@intel.com>
4470
4471         * config/i386/i386.c (ix86_option_override_internal): Change flags type
4472         to wide_int_bitmask.
4473         * wide-int-bitmask.h: New.
4474
4475 2018-02-02  Igor Tsimbalist  <igor.v.tsimbalist@intel.com>
4476
4477         PR target/84066
4478         * config/i386/i386.md: Replace Pmode with word_mode in
4479         builtin_setjmp_setup and builtin_longjmp to support x32.
4480
4481 2018-02-01  Peter Bergner  <bergner@vnet.ibm.com>
4482
4483         PR target/56010
4484         PR target/83743
4485         * config/rs6000/driver-rs6000.c: #include "diagnostic.h".
4486         #include "opts.h".
4487         (rs6000_supported_cpu_names): New static variable.
4488         (linux_cpu_translation_table): Likewise.
4489         (elf_platform) <cpu>: Define new static variable and use it.
4490         Translate kernel AT_PLATFORM name to canonical name if needed.
4491         Error if platform name is unknown.
4492
4493 2018-02-01  Aldy Hernandez  <aldyh@redhat.com>
4494
4495         PR target/84089
4496         * config/pa/predicates.md (base14_operand): Handle E_VOIDmode.
4497
4498 2018-02-01  Jeff Law  <law@redhat.com>
4499
4500         PR target/84128
4501         * config/i386/i386.c (release_scratch_register_on_entry): Add new
4502         OFFSET and RELEASE_VIA_POP arguments.  Use SP+OFFSET to restore
4503         the scratch if RELEASE_VIA_POP is false.
4504         (ix86_adjust_stack_and_probe_stack_clash): Un-constify SIZE.
4505         If we have to save a temporary register, decrement SIZE appropriately.
4506         Pass new arguments to release_scratch_register_on_entry.
4507         (ix86_adjust_stack_and_probe): Likewise.
4508         (ix86_emit_probe_stack_range): Pass new arguments to
4509         release_scratch_register_on_entry.
4510
4511 2018-02-01  Uros Bizjak  <ubizjak@gmail.com>
4512
4513         PR rtl-optimization/84157
4514         * combine.c (change_zero_ext): Use REG_P predicate in
4515         front of HARD_REGISTER_P predicate.
4516
4517 2018-02-01  Georg-Johann Lay  <avr@gjlay.de>
4518
4519         * config/avr/avr.c (avr_option_override): Move disabling of
4520         -fdelete-null-pointer-checks to...
4521         * common/config/avr/avr-common.c (avr_option_optimization_table):
4522         ...here.
4523
4524 2018-02-01  Richard Sandiford  <richard.sandiford@linaro.org>
4525
4526         PR tree-optimization/81635
4527         * tree-data-ref.c (split_constant_offset_1): For types that
4528         wrap on overflow, try to use range info to prove that wrapping
4529         cannot occur.
4530
4531 2018-02-01  Renlin Li  <renlin.li@arm.com>
4532
4533         PR target/83370
4534         * config/aarch64/aarch64.c (aarch64_class_max_nregs): Handle
4535         TAILCALL_ADDR_REGS.
4536         (aarch64_register_move_cost): Likewise.
4537         * config/aarch64/aarch64.h (reg_class): Rename CALLER_SAVE_REGS to
4538         TAILCALL_ADDR_REGS.
4539         (REG_CLASS_NAMES): Likewise.
4540         (REG_CLASS_CONTENTS): Rename CALLER_SAVE_REGS to
4541         TAILCALL_ADDR_REGS. Remove IP registers.
4542         * config/aarch64/aarch64.md (Ucs): Update register constraint.
4543
4544 2018-02-01  Richard Biener  <rguenther@suse.de>
4545
4546         * domwalk.h (dom_walker::dom_walker): Add additional constructor
4547         for specifying RPO order and allow NULL for that.
4548         * domwalk.c (dom_walker::dom_walker): Likewise.
4549         (dom_walker::walk): Handle NULL RPO order.
4550         * tree-into-ssa.c (rewrite_dom_walker): Do not walk dom children
4551         in RPO order.
4552         (rewrite_update_dom_walker): Likewise.
4553         (mark_def_dom_walker): Likewise.
4554
4555 2018-02-01  Richard Sandiford  <richard.sandiford@linaro.org>
4556
4557         * config/aarch64/aarch64-protos.h (aarch64_split_sve_subreg_move)
4558         (aarch64_maybe_expand_sve_subreg_move): Declare.
4559         * config/aarch64/aarch64.md (UNSPEC_REV_SUBREG): New unspec.
4560         * config/aarch64/predicates.md (aarch64_any_register_operand): New
4561         predicate.
4562         * config/aarch64/aarch64-sve.md (mov<mode>): Optimize subreg moves
4563         that are semantically a reverse operation.
4564         (*aarch64_sve_mov<mode>_subreg_be): New pattern.
4565         * config/aarch64/aarch64.c (aarch64_maybe_expand_sve_subreg_move):
4566         (aarch64_replace_reg_mode, aarch64_split_sve_subreg_move): New
4567         functions.
4568         (aarch64_can_change_mode_class): For big-endian, forbid changes
4569         between two SVE modes if they have different element sizes.
4570
4571 2018-02-01  Richard Sandiford  <richard.sandiford@linaro.org>
4572
4573         * config/aarch64/aarch64.c (aarch64_expand_sve_const_vector): Prefer
4574         the TImode handling for big-endian targets.
4575
4576 2018-02-01  Richard Sandiford  <richard.sandiford@linaro.org>
4577
4578         * config/aarch64/aarch64-sve.md (sve_ld1rq): Replace with...
4579         (*sve_ld1rq<Vesize>): ... this new pattern.  Handle all element sizes,
4580         not just bytes.
4581         * config/aarch64/aarch64.c (aarch64_expand_sve_widened_duplicate):
4582         Remove BSWAP handing for big-endian targets and use the form of
4583         LD1RQ appropariate for the mode.
4584
4585 2018-02-01  Richard Sandiford  <richard.sandiford@linaro.org>
4586
4587         * config/aarch64/aarch64.c (aarch64_simd_valid_immediate): Handle
4588         all CONST_VECTOR_DUPLICATE_P vectors, not just those with a single
4589         duplicated element.
4590
4591 2018-02-01  Richard Sandiford  <richard.sandiford@linaro.org>
4592
4593         PR tearget/83845
4594         * config/aarch64/aarch64.c (aarch64_secondary_reload): Tighten
4595         check for operands that need to go through aarch64_sve_reload_be.
4596
4597 2018-02-01  Jakub Jelinek  <jakub@redhat.com>
4598
4599         PR tree-optimization/81661
4600         PR tree-optimization/84117
4601         * tree-eh.h (rewrite_to_non_trapping_overflow): Declare.
4602         * tree-eh.c: Include gimplify.h.
4603         (find_trapping_overflow, replace_trapping_overflow,
4604         rewrite_to_non_trapping_overflow): New functions.
4605         * tree-vect-loop.c: Include tree-eh.h.
4606         (vect_get_loop_niters): Use rewrite_to_non_trapping_overflow.
4607         * tree-data-ref.c: Include tree-eh.h.
4608         (get_segment_min_max): Use rewrite_to_non_trapping_overflow.
4609
4610 2018-01-31  Uros Bizjak  <ubizjak@gmail.com>
4611
4612         PR rtl-optimization/84123
4613         * combine.c (change_zero_ext): Check if hard register satisfies
4614         can_change_dest_mode before calling gen_lowpart_SUBREG.
4615
4616 2018-01-31  Vladimir Makarov  <vmakarov@redhat.com>
4617
4618         PR target/82444
4619         * ira.c (ira_init_register_move_cost): Remove assert.
4620
4621 2018-01-31  Eric Botcazou  <ebotcazou@adacore.com>
4622
4623         PR rtl-optimization/84071
4624         * doc/tm.texi.in (WORD_REGISTER_OPERATIONS): Add explicit case.
4625         * doc/tm.texi: Regenerate.
4626
4627 2018-01-31  Richard Biener  <rguenther@suse.de>
4628
4629         PR tree-optimization/84132
4630         * tree-data-ref.c (analyze_miv_subscript): Properly
4631         check whether evolution_function_is_affine_multivariate_p
4632         before calling gcd_of_steps_may_divide_p.
4633
4634 2018-01-31  Julia Koval  <julia.koval@intel.com>
4635
4636         PR target/83618
4637         * config/i386/i386.c (ix86_expand_builtin): Handle IX86_BUILTIN_RDPID.
4638         * config/i386/i386.md (rdpid_rex64) New.
4639         (rdpid): Make 32bit only.
4640
4641 2018-01-29  Aldy Hernandez  <aldyh@redhat.com>
4642
4643         PR lto/84105
4644         * tree-pretty-print.c (dump_generic_node): Handle a TYPE_NAME with
4645         an IDENTIFIER_NODE for FUNCTION_TYPE's.
4646
4647 2018-01-31  Eric Botcazou  <ebotcazou@adacore.com>
4648
4649         Revert
4650         2018-01-12  Eric Botcazou  <ebotcazou@adacore.com>
4651
4652         * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
4653
4654 2018-01-31  Eric Botcazou  <ebotcazou@adacore.com>
4655
4656         PR rtl-optimization/84071
4657         * combine.c (record_dead_and_set_regs_1): Record the source unmodified
4658         for a paradoxical SUBREG on a WORD_REGISTER_OPERATIONS target.
4659
4660 2018-01-31  Claudiu Zissulescu  <claziss@synopsys.com>
4661
4662         * config/arc/arc.c (arc_handle_aux_attribute): New function.
4663         (arc_attribute_table): Add 'aux' attribute.
4664         (arc_in_small_data_p): Consider aux like variables.
4665         (arc_is_aux_reg_p): New function.
4666         (arc_asm_output_aligned_decl_local): Ignore 'aux' like variables.
4667         (arc_get_aux_arg): New function.
4668         (prepare_move_operands): Handle aux-register access.
4669         (arc_handle_aux_attribute): New function.
4670         * doc/extend.texi (ARC Variable attributes): Add subsection.
4671
4672 2018-01-31  Claudiu Zissulescu  <claziss@synopsys.com>
4673
4674         * config/arc/arc-protos.h (arc_is_uncached_mem_p): Function proto.
4675         * config/arc/arc.c (arc_handle_uncached_attribute): New function.
4676         (arc_attribute_table): Add 'uncached' attribute.
4677         (arc_print_operand): Print '.di' flag for uncached memory
4678         accesses.
4679         (arc_in_small_data_p): Do not consider for small data the uncached
4680         types.
4681         (arc_is_uncached_mem_p): New function.
4682         * config/arc/predicates.md (compact_store_memory_operand): Check
4683         for uncached memory accesses.
4684         (nonvol_nonimm_operand): Likewise.
4685         * gcc/doc/extend.texi (ARC Type Attribute): New subsection.
4686
4687 2018-01-31  Jakub Jelinek  <jakub@redhat.com>
4688
4689         PR c/84100
4690         * common.opt (falign-functions=, falign-jumps=, falign-labels=,
4691         falign-loops=): Add Optimization flag.
4692
4693 2018-01-30  Jeff Law  <law@redhat.com>
4694
4695         PR target/84064
4696         * i386.c (ix86_adjust_stack_and_probe_stack_clash): New argument
4697         INT_REGISTERS_SAVED.  Check it prior to calling
4698         get_scratch_register_on_entry.
4699         (ix86_adjust_stack_and_probe): Similarly.
4700         (ix86_emit_probe_stack_range): Similarly.
4701         (ix86_expand_prologue): Corresponding changes.
4702
4703 2018-01-30  Rainer Orth  <ro@CeBiTec.Uni-Bielefeld.DE>
4704
4705         PR target/40411
4706         * config/sol2.h (STARTFILE_ARCH_SPEC): Use -std=c*,
4707         -std=iso9899:199409 instead of -pedantic to select values-Xc.o.
4708
4709 2018-01-30  Vladimir Makarov  <vmakarov@redhat.com>
4710
4711         PR target/84112
4712         * lra-constraints.c (curr_insn_transform): Process AND in the
4713         address.
4714
4715 2018-01-30  Jakub Jelinek  <jakub@redhat.com>
4716
4717         PR rtl-optimization/83986
4718         * sched-deps.c (sched_analyze_insn): For frame related insns, add anti
4719         dependence against last_pending_memory_flush in addition to
4720         pending_jump_insns.
4721
4722 2018-01-30  Alexandre Oliva  <aoliva@redhat.com>
4723
4724         PR tree-optimization/81611
4725         * tree-ssa-dom.c (simple_iv_increment_p): Skip intervening
4726         copies.
4727
4728 2018-01-30  Aaron Sawdey  <acsawdey@linux.vnet.ibm.com>
4729
4730         PR target/83758
4731         * config/rs6000/rs6000.c (rs6000_internal_arg_pointer): Only return
4732         a reg rtx.
4733
4734 2018-01-30  Richard Biener  <rguenther@suse.de>
4735             Jakub Jelinek  <jakub@redhat.com>
4736
4737         PR tree-optimization/84111
4738         * tree-ssa-loop-ivcanon.c (tree_unroll_loops_completely_1): Skip
4739         inner loops added during recursion, as they don't have up-to-date
4740         SSA form.
4741
4742 2018-01-30  Jan Hubicka  <hubicka@ucw.cz>
4743
4744         PR ipa/81360
4745         * ipa-inline.c (can_inline_edge_p): Break out late tests to...
4746         (can_inline_edge_by_limits_p): ... here.
4747         (can_early_inline_edge_p, check_callers,
4748         update_caller_keys, update_callee_keys, recursive_inlining,
4749         add_new_edges_to_heap, speculation_useful_p,
4750         inline_small_functions,
4751         inline_small_functions, flatten_function,
4752         inline_to_all_callers_1): Update.
4753
4754 2018-01-30  Jan Hubicka  <hubicka@ucw.cz>
4755
4756         * profile-count.c (profile_count::combine_with_ipa_count): Handle
4757         zeros correctly.
4758
4759 2018-01-30  Richard Biener  <rguenther@suse.de>
4760
4761         PR tree-optimization/83008
4762         * tree-vect-slp.c (vect_analyze_slp_cost_1): Properly cost
4763         invariant and constant vector uses in stmts when they need
4764         more than one stmt.
4765
4766 2018-01-30  Rainer Orth  <ro@CeBiTec.Uni-Bielefeld.DE>
4767
4768         PR bootstrap/84017
4769         * configure.ac (gcc_cv_as_shf_merge): Disable on Solaris 10/x86.
4770         * configure: Regenerate.
4771
4772 2018-01-30  Richard Sandiford  <richard.sandiford@linaro.org>
4773
4774         * config/aarch64/aarch64-sve.md (*vec_extract<mode><Vel>_0): New
4775         pattern.
4776         (*vec_extract<mode><Vel>_v128): Require a nonzero lane number.
4777         Use gen_rtx_REG rather than gen_lowpart.
4778
4779 2018-01-30  Richard Sandiford  <richard.sandiford@linaro.org>
4780
4781         * lra-constraints.c (match_reload): Use subreg_lowpart_offset
4782         rather than 0 when creating partial subregs.
4783
4784 2018-01-30  Richard Sandiford  <richard.sandiford@linaro.org>
4785
4786         * vec-perm-indices.c (vec_perm_indices::series_p): Give examples
4787         of usage.
4788
4789 2018-01-29  Michael Meissner  <meissner@linux.vnet.ibm.com>
4790
4791         PR target/81550
4792         * config/rs6000/rs6000.c (rs6000_setup_reg_addr_masks): If DFmode
4793         and SFmode can go in Altivec registers (-mcpu=power7 for DFmode,
4794         -mcpu=power8 for SFmode) don't set the PRE_INCDEC or PRE_MODIFY
4795         flags.  This restores the settings used before the 2017-07-24.
4796         Turning off pre increment/decrement/modify allows IVOPTS to
4797         optimize DF/SF loops where the index is an int.
4798
4799 2018-01-29  Richard Biener  <rguenther@suse.de>
4800             Kelvin Nilsen  <kelvin@gcc.gnu.org>
4801
4802         PR bootstrap/80867
4803         * tree-vect-stmts.c (vectorizable_call): Don't call
4804         targetm.vectorize_builtin_md_vectorized_function if callee is
4805         NULL.
4806
4807 2018-01-22  Carl Love  <cel@us.ibm.com>
4808
4809         * doc/extend.tex: Fix typo in second arg in
4810         __builtin_bcdadd_{lt|eq|gt|ov} and __builtin_bcdsub_{lt|eq|gt|ov}.
4811
4812 2018-01-29  Richard Biener  <rguenther@suse.de>
4813
4814         PR tree-optimization/84086
4815         * tree-ssanames.c: Include cfgloop.h and tree-scalar-evolution.h.
4816         (flush_ssaname_freelist): When SSA names were released reset
4817         the SCEV hash table.
4818
4819 2018-01-29  Richard Biener  <rguenther@suse.de>
4820
4821         PR tree-optimization/84057
4822         * tree-ssa-loop-ivcanon.c (unloop_loops): Deal with already
4823         removed paths when removing edges.
4824
4825 2018-01-27  H.J. Lu  <hongjiu.lu@intel.com>
4826
4827         * doc/invoke.texi: Replace -mfunction-return==@var{choice} with
4828         -mfunction-return=@var{choice}.
4829
4830 2018-01-27  Bernd Edlinger  <bernd.edlinger@hotmail.de>
4831
4832         PR diagnostic/84034
4833         * diagnostic-show-locus.c (get_line_width_without_trailing_whitespace):
4834         Handle CR like TAB.
4835         (layout::print_source_line): Likewise.
4836         (test_get_line_width_without_trailing_whitespace): Add test cases.
4837
4838 2018-01-27  Jakub Jelinek  <jakub@redhat.com>
4839
4840         PR middle-end/84040
4841         * sched-deps.c (sched_macro_fuse_insns): Return immediately for
4842         debug insns.
4843
4844 2018-01-26  Jim Wilson  <jimw@sifive.com>
4845
4846         * config/riscv/riscv.h (MAX_FIXED_MODE_SIZE): New.
4847
4848         * config/riscv/elf.h (LIB_SPEC): Don't include -lgloss when nosys.specs
4849         specified.
4850
4851 2018-01-26  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
4852
4853         * config/aarch64/aarch64.md: Add peepholes for CMP + SUB -> SUBS
4854         and CMP + SUB-immediate -> SUBS.
4855
4856 2018-01-26  Martin Sebor  <msebor@redhat.com>
4857
4858         PR tree-optimization/83896
4859         * tree-ssa-strlen.c (get_string_len): Rename...
4860         (get_string_cst_length): ...to this.  Return HOST_WIDE_INT.
4861         Avoid assuming length is constant.
4862         (handle_char_store): Use HOST_WIDE_INT for string length.
4863
4864 2018-01-26  Uros Bizjak  <ubizjak@gmail.com>
4865
4866         PR target/81763
4867         * config/i386/i386.md (*andndi3_doubleword): Add earlyclobber
4868         to (=&r,r,rm) alternative. Add (=r,0,rm) and (=r,r,0) alternatives.
4869
4870 2018-01-26  Richard Biener  <rguenther@suse.de>
4871
4872         PR rtl-optimization/84003
4873         * dse.c (record_store): Only record redundant stores when
4874         the earlier store aliases at least all accesses the later one does.
4875
4876 2018-01-26  Jakub Jelinek  <jakub@redhat.com>
4877
4878         PR rtl-optimization/83985
4879         * dce.c (deletable_insn_p): Return false for separate shrink wrapping
4880         REG_CFA_RESTORE insns.
4881         (delete_unmarked_insns): Don't ignore separate shrink wrapping
4882         REG_CFA_RESTORE insns here.
4883
4884         PR c/83989
4885         * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Don't
4886         use SSA_NAME_VAR as base for SSA_NAMEs with non-NULL SSA_NAME_VAR.
4887
4888 2018-01-26  Claudiu Zissulescu  <claziss@synopsys.com>
4889
4890         * config/arc/arc-arch.h (arc_tune_attr): Add ARC_TUNE_CORE_3.
4891         * config/arc/arc.c (arc_sched_issue_rate): Use ARC_TUNE_... .
4892         (arc_init): Likewise.
4893         (arc_override_options): Likewise.
4894         (arc_file_start): Choose Tag_ARC_CPU_variation based on arc_tune
4895         value.
4896         (hwloop_fail): Use TARGET_DBNZ when we want to check for dbnz insn
4897         support.
4898         * config/arc/arc.h (TARGET_DBNZ): Define.
4899         * config/arc/arc.md (attr tune): Add core_3, use ARC_TUNE_... to
4900         properly set the tune attribute.
4901         (dbnz): Use TARGET_DBNZ guard.
4902         * config/arc/arc.opt (mtune): Add core3 option.
4903
4904 2018-01-26  Claudiu Zissulescu  <claziss@synopsys.com>
4905
4906         * config/arc/arc.c (arc_delegitimize_address_0): Refactored to
4907         recognize new pic like addresses.
4908         (arc_delegitimize_address): Clean up.
4909
4910 2018-01-26  Claudiu Zissulescu  <claziss@synopsys.com>
4911
4912         * config/arc/arc-arches.def: Option mrf16 valid for all
4913         architectures.
4914         * config/arc/arc-c.def (__ARC_RF16__): New predefined macro.
4915         * config/arc/arc-cpus.def (em_mini): New cpu with rf16 on.
4916         * config/arc/arc-options.def (FL_RF16): Add mrf16 option.
4917         * config/arc/arc-tables.opt: Regenerate.
4918         * config/arc/arc.c (arc_conditional_register_usage): Handle
4919         reduced register file case.
4920         (arc_file_start): Set must have build attributes.
4921         * config/arc/arc.h (MAX_ARC_PARM_REGS): Conditional define using
4922         mrf16 option value.
4923         * config/arc/arc.opt (mrf16): Add new option.
4924         * config/arc/elf.h (ATTRIBUTE_PCS): Define.
4925         * config/arc/genmultilib.awk: Handle new mrf16 option.
4926         * config/arc/linux.h (ATTRIBUTE_PCS): Define.
4927         * config/arc/t-multilib: Regenerate.
4928         * doc/invoke.texi (ARC Options): Document mrf16 option.
4929
4930 2018-01-26  Claudiu Zissulescu  <claziss@synopsys.com>
4931
4932         * config/arc/arc-protos.h: Add arc_is_secure_call_p proto.
4933         * config/arc/arc.c (arc_handle_secure_attribute): New function.
4934         (arc_attribute_table): Add 'secure_call' attribute.
4935         (arc_print_operand): Print secure call operand.
4936         (arc_function_ok_for_sibcall): Don't optimize tail calls when
4937         secure.
4938         (arc_is_secure_call_p): New function.  * config/arc/arc.md
4939         (call_i): Add support for sjli instruction.
4940         (call_value_i): Likewise.
4941         * config/arc/constraints.md (Csc): New constraint.
4942
4943 2018-01-26  Claudiu Zissulescu  <claziss@synopsys.com>
4944             John Eric Martin  <John.Martin@emmicro-us.com>
4945
4946         * config/arc/arc-protos.h: Add arc_is_jli_call_p proto.
4947         * config/arc/arc.c (_arc_jli_section): New struct.
4948         (arc_jli_section): New type.
4949         (rc_jli_sections): New static variable.
4950         (arc_handle_jli_attribute): New function.
4951         (arc_attribute_table): Add jli_always and jli_fixed attribute.
4952         (arc_file_end): New function.
4953         (TARGET_ASM_FILE_END): Define.
4954         (arc_print_operand): Reuse 'S' letter for JLI output instruction.
4955         (arc_add_jli_section): New function.
4956         (jli_call_scan): Likewise.
4957         (arc_reorg): Call jli_call_scan.
4958         (arc_output_addsi): Remove 'S' from printing asm operand.
4959         (arc_is_jli_call_p): New function.
4960         * config/arc/arc.md (movqi_insn): Remove 'S' from printing asm
4961         operand.
4962         (movhi_insn): Likewise.
4963         (movsi_insn): Likewise.
4964         (movsi_set_cc_insn): Likewise.
4965         (loadqi_update): Likewise.
4966         (load_zeroextendqisi_update): Likewise.
4967         (load_signextendqisi_update): Likewise.
4968         (loadhi_update): Likewise.
4969         (load_zeroextendhisi_update): Likewise.
4970         (load_signextendhisi_update): Likewise.
4971         (loadsi_update): Likewise.
4972         (loadsf_update): Likewise.
4973         (movsicc_insn): Likewise.
4974         (bset_insn): Likewise.
4975         (bxor_insn): Likewise.
4976         (bclr_insn): Likewise.
4977         (bmsk_insn): Likewise.
4978         (bicsi3_insn): Likewise.
4979         (cmpsi_cc_c_insn): Likewise.
4980         (movsi_ne): Likewise.
4981         (movsi_cond_exec): Likewise.
4982         (clrsbsi2): Likewise.
4983         (norm_f): Likewise.
4984         (normw): Likewise.
4985         (swap): Likewise.
4986         (divaw): Likewise.
4987         (flag): Likewise.
4988         (sr): Likewise.
4989         (kflag): Likewise.
4990         (ffs): Likewise.
4991         (ffs_f): Likewise.
4992         (fls): Likewise.
4993         (call_i): Remove 'S' asm letter, add jli instruction.
4994         (call_value_i): Likewise.
4995         * config/arc/arc.op (mjli-always): New option.
4996         * config/arc/constraints.md (Cji): New constraint.
4997         * config/arc/fpx.md (addsf3_fpx): Remove 'S' from printing asm
4998         operand.
4999         (subsf3_fpx): Likewise.
5000         (mulsf3_fpx): Likewise.
5001         * config/arc/simdext.md (vendrec_insn): Remove 'S' from printing
5002         asm operand.
5003         * doc/extend.texi (ARC): Document 'jli-always' and 'jli-fixed'
5004         function attrbutes.
5005         * doc/invoke.texi (ARC): Document mjli-always option.
5006
5007 2018-01-26  Sebastian Perta  <sebastian.perta@renesas.com>
5008
5009         * config/rl78/rl78.c (rl78_addsi3_internal): If operand 2 is const
5010         avoid addition with 0 and use incw and decw where possible.
5011
5012 2018-01-26  Richard Biener  <rguenther@suse.de>
5013
5014         PR tree-optimization/81082
5015         * fold-const.c (fold_plusminus_mult_expr): Do not perform the
5016         association if it requires casting to unsigned.
5017         * match.pd ((A * C) +- (B * C) -> (A+-B)): New patterns derived
5018         from fold_plusminus_mult_expr to catch important cases late when
5019         range info is available.
5020
5021 2018-01-26  Rainer Orth  <ro@CeBiTec.Uni-Bielefeld.DE>
5022
5023         * config/i386/sol2.h (USE_HIDDEN_LINKONCE): Remove.
5024         * configure.ac (hidden_linkonce): New test.
5025         * configure: Regenerate.
5026         * config.in: Regenerate.
5027
5028 2018-01-26  Julia Koval  <julia.koval@intel.com>
5029
5030         * config/i386/avx512bitalgintrin.h (_mm512_bitshuffle_epi64_mask,
5031         _mm512_mask_bitshuffle_epi64_mask, _mm256_bitshuffle_epi64_mask,
5032         _mm256_mask_bitshuffle_epi64_mask, _mm_bitshuffle_epi64_mask,
5033         _mm_mask_bitshuffle_epi64_mask): Fix type.
5034         * config/i386/i386-builtin-types.def (UHI_FTYPE_V2DI_V2DI_UHI,
5035         USI_FTYPE_V4DI_V4DI_USI): Remove.
5036         * config/i386/i386-builtin.def (__builtin_ia32_vpshufbitqmb512_mask,
5037         __builtin_ia32_vpshufbitqmb256_mask,
5038         __builtin_ia32_vpshufbitqmb128_mask): Fix types.
5039         * config/i386/i386.c (ix86_expand_args_builtin): Remove old types.
5040         * config/i386/sse.md (VI1_AVX512VLBW): Change types.
5041
5042 2018-01-26  Alan Modra  <amodra@gmail.com>
5043
5044         PR target/84033
5045         * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Exclude
5046         UNSPEC_VBPERMQ.  Sort other unspecs.
5047
5048 2018-01-25  David Edelsohn  <dje.gcc@gmail.com>
5049
5050         * doc/invoke.texi (PowerPC Options): Document 'native' cpu type.
5051
5052 2018-01-25  Jan Hubicka  <hubicka@ucw.cz>
5053
5054         PR middle-end/83055
5055         * predict.c (drop_profile): Do not push/pop cfun; update also
5056         node->count.
5057         (handle_missing_profiles): Fix logic looking for zero profiles.
5058
5059 2018-01-25  Jakub Jelinek  <jakub@redhat.com>
5060
5061         PR middle-end/83977
5062         * ipa-fnsummary.c (compute_fn_summary): Clear can_change_signature
5063         on functions with #pragma omp declare simd or functions with simd
5064         attribute.
5065         * omp-simd-clone.c (expand_simd_clones): Revert 2018-01-24 change.
5066         * config/i386/i386.c (ix86_simd_clone_compute_vecsize_and_simdlen):
5067         Remove trailing \n from warning_at calls.
5068
5069 2018-01-25  Tom de Vries  <tom@codesourcery.com>
5070
5071         PR target/84028
5072         * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
5073         for neutered workers.
5074
5075 2018-01-24  Joseph Myers  <joseph@codesourcery.com>
5076
5077         PR target/68467
5078         * config/m68k/m68k.c (m68k_promote_function_mode): New function.
5079         (TARGET_PROMOTE_FUNCTION_MODE): New macro.
5080
5081 2018-01-24  Jeff Law  <law@redhat.com>
5082
5083         PR target/83994
5084         * i386.c (get_probe_interval): Move to earlier point.
5085         (ix86_compute_frame_layout): If -fstack-clash-protection and
5086         the frame is larger than the probe interval, then use pushes
5087         to save registers rather than reg->mem moves.
5088         (ix86_expand_prologue): Remove conditional for int_registers_saved
5089         assertion.
5090
5091 2018-01-24  Vladimir Makarov  <vmakarov@redhat.com>
5092
5093         PR target/84014
5094         * ira-build.c (setup_min_max_allocno_live_range_point): Set up
5095         min/max for never referenced object.
5096
5097 2018-01-24  Jakub Jelinek  <jakub@redhat.com>
5098
5099         PR middle-end/83977
5100         * tree.c (free_lang_data_in_decl): Don't clear DECL_ABSTRACT_ORIGIN
5101         here.
5102         * omp-low.c (create_omp_child_function): Remove "omp declare simd"
5103         attributes from DECL_ATTRIBUTES (decl) without affecting
5104         DECL_ATTRIBUTES (current_function_decl).
5105         * omp-simd-clone.c (expand_simd_clones): Ignore DECL_ARTIFICIAL
5106         functions with non-NULL DECL_ABSTRACT_ORIGIN.
5107
5108 2018-01-24  Richard Sandiford  <richard.sandiford@linaro.org>
5109
5110         PR tree-optimization/83979
5111         * fold-const.c (fold_comparison): Use constant_boolean_node
5112         instead of boolean_{true,false}_node.
5113
5114 2018-01-24  Jan Hubicka  <hubicka@ucw.cz>
5115
5116         * ipa-profile.c (ipa_propagate_frequency_1): Fix logic skipping calls
5117         with zero counts.
5118
5119 2018-01-24  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
5120
5121         * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
5122         Simplify the clause that sets the length attribute.
5123         (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
5124         (*sibcall_nonlocal_sysv<mode>): Clean up code block; simplify the
5125         clause that sets the length attribute.
5126         (*sibcall_value_nonlocal_sysv<mode>): Likewise.
5127
5128 2018-01-24  Tom de Vries  <tom@codesourcery.com>
5129
5130         PR target/83589
5131         * config/nvptx/nvptx.c (WORKAROUND_PTXJIT_BUG_2): Define to 1.
5132         (nvptx_pc_set, nvptx_condjump_label): New function. Copy from jump.c.
5133         Add strict parameter.
5134         (prevent_branch_around_nothing): Insert dummy insn between branch to
5135         label and label with no ptx insn inbetween.
5136         * config/nvptx/nvptx.md (define_insn "fake_nop"): New insn.
5137
5138 2018-01-24  Tom de Vries  <tom@codesourcery.com>
5139
5140         PR target/81352
5141         * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
5142         for neutered threads in warp.
5143         * config/nvptx/nvptx.md (define_insn "exit"): New insn.
5144
5145 2018-01-24  Richard Biener  <rguenther@suse.de>
5146
5147         PR tree-optimization/83176
5148         * tree-chrec.c (chrec_fold_plus_1): Handle (signed T){(T) .. }
5149         operands.
5150
5151 2018-01-24  Richard Biener  <rguenther@suse.de>
5152
5153         PR tree-optimization/82819
5154         * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Avoid
5155         code generating pluses that are no-ops in the target precision.
5156
5157 2018-01-24  Richard Biener  <rguenther@suse.de>
5158
5159         PR middle-end/84000
5160         * tree-cfg.c (replace_loop_annotate): Handle annot_expr_parallel_kind.
5161
5162 2018-01-23  Jan Hubicka  <hubicka@ucw.cz>
5163
5164         * cfgcleanup.c (try_crossjump_to_edge): Use combine_with_count
5165         to merge probabilities.
5166         * predict.c (probably_never_executed): Also mark as cold functions
5167         with global 0 profile and guessed local profile.
5168         * profile-count.c (profile_probability::combine_with_count): New
5169         member function.
5170         * profile-count.h (profile_probability::operator*,
5171         profile_probability::operator*=, profile_probability::operator/,
5172         profile_probability::operator/=): Reduce precision to adjusted
5173         and set value to guessed on contradictory divisions.
5174         (profile_probability::combine_with_freq): Remove.
5175         (profile_probability::combine_wiht_count): Declare.
5176         (profile_count::force_nonzero):: Set to adjusted.
5177         (profile_count::probability_in):: Set quality to adjusted.
5178         * tree-ssa-tail-merge.c (replace_block_by): Use
5179         combine_with_count.
5180
5181 2018-01-23  Andrew Waterman  <andrew@sifive.com>
5182             Jim Wilson  <jimw@sifive.com>
5183
5184         * config/riscv/riscv.c (riscv_stack_boundary): New.
5185         (riscv_option_override): Set riscv_stack_boundary.  Handle
5186         riscv_preferred_stack_boundary_arg.
5187         * config/riscv/riscv.h (MIN_STACK_BOUNDARY, ABI_STACK_BOUNDARY): New.
5188         (BIGGEST_ALIGNMENT): Set to STACK_BOUNDARY.
5189         (STACK_BOUNDARY): Set to riscv_stack_boundary.
5190         (RISCV_STACK_ALIGN): Use STACK_BOUNDARY.
5191         * config/riscv/riscv.opt (mpreferred-stack-boundary): New.
5192         * doc/invoke.tex (RISC-V Options): Add -mpreferred-stack-boundary.
5193
5194 2018-01-23  H.J. Lu  <hongjiu.lu@intel.com>
5195
5196         PR target/83905
5197         * config/i386/i386.c (ix86_expand_prologue): Use cost reference
5198         of struct ix86_frame.
5199         (ix86_expand_epilogue): Likewise.  Add a local variable for
5200         the reg_save_offset field in struct ix86_frame.
5201
5202 2018-01-23  Bin Cheng  <bin.cheng@arm.com>
5203
5204         PR tree-optimization/82604
5205         * tree-loop-distribution.c (enum partition_kind): New enum item
5206         PKIND_PARTIAL_MEMSET.
5207         (partition_builtin_p): Support above new enum item.
5208         (generate_code_for_partition): Ditto.
5209         (compute_access_range): Differentiate cases that equality can be
5210         proven at all loops, the innermost loops or no loops.
5211         (classify_builtin_st, classify_builtin_ldst): Adjust call to above
5212         function.  Set PKIND_PARTIAL_MEMSET for partition appropriately.
5213         (finalize_partitions, distribute_loop): Don't fuse partition of
5214         PKIND_PARTIAL_MEMSET kind when distributing 3-level loop nest.
5215         (prepare_perfect_loop_nest): Distribute 3-level loop nest only if
5216         parloop is enabled.
5217
5218 2018-01-23  Martin Liska  <mliska@suse.cz>
5219
5220         * predict.def (PRED_INDIR_CALL): Set probability to PROB_EVEN in
5221         order to ignore the predictor.
5222         (PRED_POLYMORPHIC_CALL): Likewise.
5223         (PRED_RECURSIVE_CALL): Likewise.
5224
5225 2018-01-23  Martin Liska  <mliska@suse.cz>
5226
5227         * tree-profile.c (tree_profiling): Print function header to
5228         aware reader which function we are working on.
5229         * value-prof.c (gimple_find_values_to_profile): Do not print
5230         not interesting value histograms.
5231
5232 2018-01-23  Martin Liska  <mliska@suse.cz>
5233
5234         * profile-count.h (enum profile_quality): Add
5235         profile_uninitialized as the first value. Do not number values
5236         as they are zero based.
5237         (profile_count::verify): Update sanity check.
5238         (profile_probability::verify): Likewise.
5239
5240 2018-01-23  Nathan Sidwell  <nathan@acm.org>
5241
5242         * doc/invoke.texi (ffor-scope): Deprecate.
5243
5244 2018-01-23  David Malcolm  <dmalcolm@redhat.com>
5245
5246         PR tree-optimization/83510
5247         * domwalk.c (set_all_edges_as_executable): New function.
5248         (dom_walker::dom_walker): Convert bool param
5249         "skip_unreachable_blocks" to enum reachability.  Move setup of
5250         edge flags to set_all_edges_as_executable and only do it when
5251         reachability is REACHABLE_BLOCKS.
5252         * domwalk.h (enum dom_walker::reachability): New enum.
5253         (dom_walker::dom_walker): Convert bool param
5254         "skip_unreachable_blocks" to enum reachability.
5255         (set_all_edges_as_executable): New decl.
5256         * graphite-scop-detection.c  (gather_bbs::gather_bbs): Convert
5257         from false for "skip_unreachable_blocks" to ALL_BLOCKS for
5258         "reachability".
5259         * tree-ssa-dom.c (dom_opt_dom_walker::dom_opt_dom_walker): Likewise,
5260         but converting true to REACHABLE_BLOCKS.
5261         * tree-ssa-sccvn.c (sccvn_dom_walker::sccvn_dom_walker): Likewise.
5262         * tree-vrp.c
5263         (check_array_bounds_dom_walker::check_array_bounds_dom_walker):
5264         Likewise, but converting it to REACHABLE_BLOCKS_PRESERVING_FLAGS.
5265         (vrp_dom_walker::vrp_dom_walker): Likewise, but converting it to
5266         REACHABLE_BLOCKS.
5267         (vrp_prop::vrp_finalize): Call set_all_edges_as_executable
5268         if check_all_array_refs will be called.
5269
5270 2018-01-23  David Malcolm  <dmalcolm@redhat.com>
5271
5272         * tree.c (selftest::test_location_wrappers): Add more test
5273         coverage.
5274
5275 2018-01-23  David Malcolm  <dmalcolm@redhat.com>
5276
5277         * sbitmap.c (selftest::test_set_range): Fix memory leaks.
5278         (selftest::test_bit_in_range): Likewise.
5279
5280 2018-01-23  Richard Sandiford  <richard.sandiford@linaro.org>
5281
5282         PR testsuite/83888
5283         * doc/sourcebuild.texi (vect_float): Say that the selector
5284         only describes the situation when -funsafe-math-optimizations is on.
5285         (vect_float_strict): Document.
5286
5287 2018-01-23  Richard Sandiford  <richard.sandiford@linaro.org>
5288
5289         PR tree-optimization/83965
5290         * tree-vect-patterns.c (vect_reassociating_reduction_p): New function.
5291         (vect_recog_dot_prod_pattern, vect_recog_sad_pattern): Use it
5292         instead of checking only for a reduction.
5293         (vect_recog_widen_sum_pattern): Likewise.
5294
5295 2018-01-23  Jan Hubicka  <hubicka@ucw.cz>
5296
5297         * predict.c (probably_never_executed): Only use precise profile info.
5298         (compute_function_frequency): Skip after inlining hack since we now
5299         have quality checking.
5300
5301 2018-01-23  Jan Hubicka  <hubicka@ucw.cz>
5302
5303         * profile-count.h (profile_probability::very_unlikely,
5304         profile_probability::unlikely, profile_probability::even): Set
5305         precision to guessed.
5306
5307 2018-01-23  Richard Biener  <rguenther@suse.de>
5308
5309         PR tree-optimization/83963
5310         * graphite-scop-detection.c (scop_detection::harmful_loop_in_region):
5311         Properly terminate dominator walk when crossing the exit edge not
5312         when visiting its source block.
5313
5314 2018-01-23  Jakub Jelinek  <jakub@redhat.com>
5315
5316         PR c++/83918
5317         * tree.c (maybe_wrap_with_location): Use NON_LVALUE_EXPR rather than
5318         VIEW_CONVERT_EXPR to wrap CONST_DECLs.
5319
5320 2018-01-22  Jakub Jelinek  <jakub@redhat.com>
5321
5322         PR tree-optimization/83957
5323         * omp-expand.c (expand_omp_for_generic): Ignore virtual PHIs.  Remove
5324         semicolon after for body surrounded by braces.
5325
5326         PR tree-optimization/83081
5327         * profile-count.h (profile_probability::split): New method.
5328         * dojump.c (do_jump_1) <case TRUTH_ANDIF_EXPR, case TRUTH_ORIF_EXPR>:
5329         Use profile_probability::split.
5330         (do_compare_rtx_and_jump): Fix adjustment of probabilities
5331         when splitting a single conditional jump into 2.
5332
5333 2018-01-22  David Malcolm  <dmalcolm@redhat.com>
5334
5335         PR tree-optimization/69452
5336         * tree-ssa-loop-im.c (class move_computations_dom_walker): Remove
5337         decl.
5338
5339 2018-01-22  Sebastian Perta  <sebastian.perta@renesas.com>
5340
5341         * config/rl78/rl78-expand.md (bswaphi2): New define_expand.
5342         * config/rl78/rl78-virt.md (*bswaphi2_virt): New define_insn.
5343         * config/rl78/rl78-real.md (*bswaphi2_real): New define_insn.
5344
5345 2018-01-22  Sebastian Perta  <sebastian.perta@renesas.com>
5346
5347         * config/rl78/rl78-protos.h (rl78_split_movdi): New function declaration.
5348         * config/rl78/rl78.md (movdi): New define_expand.
5349         * config/rl78/rl78.c (rl78_split_movdi): New function.
5350
5351 2018-01-22  Michael Meissner  <meissner@linux.vnet.ibm.com>
5352
5353         PR target/83862
5354         * config/rs6000/rs6000-protos.h (rs6000_split_signbit): Delete,
5355         no longer used.
5356         * config/rs6000/rs6000.c (rs6000_split_signbit): Likewise.
5357         * config/rs6000/rs6000.md (signbit<mode>2): Change code for IEEE
5358         128-bit to produce an UNSPEC move to get the double word with the
5359         signbit and then a shift directly to do signbit.
5360         (signbit<mode>2_dm): Replace old IEEE 128-bit signbit
5361         implementation with a new version that just does either a direct
5362         move or a regular move.  Move memory interface to separate insns.
5363         Move insns so they are next to the expander.
5364         (signbit<mode>2_dm_mem_be): New combiner insns to combine load
5365         with signbit move.  Split big and little endian case.
5366         (signbit<mode>2_dm_mem_le): Likewise.
5367         (signbit<mode>2_dm_<su>ext): Delete, no longer used.
5368         (signbit<mode>2_dm2): Likewise.
5369
5370 2018-01-22  Sebastian Perta  <sebastian.perta@renesas.com>
5371
5372         * config/rl78/rl78.md (anddi3): New define_expand.
5373
5374 2018-01-22  Sebastian Perta  <sebastian.perta@renesas.com>
5375
5376         * config/rl78/rl78.md (umindi3): New define_expand.
5377
5378 2018-01-22  Sebastian Perta  <sebastian.perta@renesas.com>
5379
5380         * config/rl78/rl78.md (smindi3): New define_expand.
5381
5382 2018-01-22  Sebastian Perta  <sebastian.perta@renesas.com>
5383
5384         * config/rl78/rl78.md (smaxdi3): New define_expand.
5385
5386 2018-01-22  Carl Love  <cel@us.ibm.com>
5387
5388         * config/rs6000/rs6000-builtin.def (ST_ELEMREV_V1TI, LD_ELEMREV_V1TI,
5389         LVX_V1TI): Add macro expansion.
5390         * config/rs6000/rs6000-c.c (altivec_builtin_types): Add argument
5391         definitions for VSX_BUILTIN_VEC_XST_BE, VSX_BUILTIN_VEC_ST,
5392         VSX_BUILTIN_VEC_XL, LD_ELEMREV_V1TI builtins.
5393         * config/rs6000/rs6000-p8swap.c (insn_is_swappable_p);
5394         Change check to determine if the instruction is a byte reversing
5395         entry.  Fix typo in comment.
5396         * config/rs6000/rs6000.c (altivec_expand_builtin): Add case entry
5397         for VSX_BUILTIN_ST_ELEMREV_V1TI and VSX_BUILTIN_LD_ELEMREV_V1TI.
5398         Add def_builtin calls for new builtins.
5399         * config/rs6000/vsx.md (vsx_st_elemrev_v1ti, vsx_ld_elemrev_v1ti):
5400         Add define_insn expansion.
5401
5402 2018-01-22  Sebastian Perta  <sebastian.perta@renesas.com>
5403
5404         * config/rl78/rl78.md (umaxdi3): New define_expand.
5405
5406 2018-01-22  Sebastian Perta  <sebastian.perta@renesas.com>
5407
5408         * config/rl78/rl78.c (rl78_note_reg_set): Fixed dead reg check
5409         for non-QImode registers.
5410
5411 2018-01-22  Richard Biener  <rguenther@suse.de>
5412
5413         PR tree-optimization/83963
5414         * graphite-scop-detection.c (scop_detection::get_sese): Delay
5415         including the loop exit block.
5416         (scop_detection::merge_sese): Likewise.
5417         (scop_detection::add_scop): Do it here instead.
5418
5419 2018-01-22  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
5420
5421         * doc/sourcebuild.texi (arm_softfloat): Document.
5422
5423 2018-01-21  John David Anglin  <danglin@gcc.gnu.org>
5424
5425         PR gcc/77734
5426         * config/pa/pa.c (pa_function_ok_for_sibcall): Use
5427         targetm.binds_local_p instead of TREE_PUBLIC to check local binding.
5428         Move TARGET_PORTABLE_RUNTIME check after TARGET_64BIT check.
5429
5430 2018-01-21  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
5431             David Edelsohn  <dje.gcc@gmail.com>
5432
5433         PR target/83946
5434         * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
5435         Change "crset eq" to "crset 2".
5436         (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
5437         (*call_indirect_aix<mode>_nospec): Likewise.
5438         (*call_value_indirect_aix<mode>_nospec): Likewise.
5439         (*call_indirect_elfv2<mode>_nospec): Likewise.
5440         (*call_value_indirect_elfv2<mode>_nospec): Likewise.
5441         (*sibcall_nonlocal_sysv<mode>): Change "crset eq" to "crset 2";
5442         change assembly output from . to $.
5443         (*sibcall_value_nonlocal_sysv<mode>): Likewise.
5444         (indirect_jump<mode>_nospec): Change assembly output from . to $.
5445         (*tablejump<mode>_internal1_nospec): Likewise.
5446
5447 2018-01-21  Oleg Endo  <olegendo@gcc.gnu.org>
5448
5449         PR target/80870
5450         * config/sh/sh_optimize_sett_clrt.cc:
5451         Use INCLUDE_ALGORITHM and INCLUDE_VECTOR instead of direct includes.
5452
5453 2018-01-20  Richard Sandiford  <richard.sandiford@linaro.org>
5454
5455         PR tree-optimization/83940
5456         * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): Set
5457         offset_dt to vect_constant_def rather than vect_unknown_def_type.
5458         (vect_check_load_store_mask): Add a mask_dt_out parameter and
5459         use it to pass back the definition type.
5460         (vect_check_store_rhs): Likewise rhs_dt_out.
5461         (vect_build_gather_load_calls): Add a mask_dt argument and use
5462         it instead of a call to vect_is_simple_use.
5463         (vectorizable_store): Update calls to vect_check_load_store_mask
5464         and vect_check_store_rhs.  Use the dt returned by the latter instead
5465         of scatter_src_dt.  Use the cached mask_dt and gs_info.offset_dt
5466         instead of calls to vect_is_simple_use.  Pass the scalar rather
5467         than the vector operand to vect_is_simple_use when handling
5468         second and subsequent copies of an rhs value.
5469         (vectorizable_load): Update calls to vect_check_load_store_mask
5470         and vect_build_gather_load_calls.  Use the cached mask_dt and
5471         gs_info.offset_dt instead of calls to vect_is_simple_use.
5472
5473 2018-01-20  Jakub Jelinek  <jakub@redhat.com>
5474
5475         PR middle-end/83945
5476         * tree-emutls.c: Include gimplify.h.
5477         (lower_emutls_2): New function.
5478         (lower_emutls_1): If ADDR_EXPR is a gimple invariant and walk_tree
5479         with lower_emutls_2 callback finds some TLS decl in it, unshare_expr
5480         it before further processing.
5481
5482         PR target/83930
5483         * simplify-rtx.c (simplify_binary_operation_1) <case UMOD>: Use
5484         UINTVAL (trueop1) instead of INTVAL (op1).
5485
5486 2018-01-19  Jakub Jelinek  <jakub@redhat.com>
5487
5488         PR debug/81570
5489         PR debug/83728
5490         * dwarf2cfi.c (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define to
5491         INCOMING_FRAME_SP_OFFSET if not defined.
5492         (scan_trace): Add ENTRY argument.  If true and
5493         DEFAULT_INCOMING_FRAME_SP_OFFSET != INCOMING_FRAME_SP_OFFSET,
5494         emit a note to adjust the CFA offset.
5495         (create_cfi_notes): Adjust scan_trace callers.
5496         (create_cie_data): Use DEFAULT_INCOMING_FRAME_SP_OFFSET rather than
5497         INCOMING_FRAME_SP_OFFSET in the CIE.
5498         * config/i386/i386.h (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define.
5499         * config/stormy16/stormy16.h (DEFAULT_INCOMING_FRAME_SP_OFFSET):
5500         Likewise.
5501         * doc/tm.texi.in (DEFAULT_INCOMING_FRAME_SP_OFFSET): Document.
5502         * doc/tm.texi: Regenerated.
5503
5504 2018-01-19  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>
5505
5506         PR rtl-optimization/83147
5507         * lra-constraints.c (remove_inheritance_pseudos): Use
5508         lra_substitute_pseudo_within_insn.
5509
5510 2018-01-19  Tom de Vries  <tom@codesourcery.com>
5511             Cesar Philippidis  <cesar@codesourcery.com>
5512
5513         PR target/83920
5514         * config/nvptx/nvptx.c (nvptx_single): Fix jit workaround.
5515
5516 2018-01-19  Cesar Philippidis  <cesar@codesourcery.com>
5517
5518         PR target/83790
5519         * config/nvptx/nvptx.c (output_init_frag): Don't use generic address
5520         spaces for function labels.
5521
5522 2018-01-19  Martin Liska  <mliska@suse.cz>
5523
5524         * predict.def (PRED_LOOP_EXIT): Change from 85 to 89.
5525         (PRED_LOOP_EXIT_WITH_RECURSION): Change from 72 to 78.
5526         (PRED_LOOP_EXTRA_EXIT): Change from 83 to 67.
5527         (PRED_OPCODE_POSITIVE): Change from 64 to 59.
5528         (PRED_TREE_OPCODE_POSITIVE): Change from 64 to 59.
5529         (PRED_CONST_RETURN): Change from 69 to 65.
5530         (PRED_NULL_RETURN): Change from 91 to 71.
5531         (PRED_LOOP_IV_COMPARE_GUESS): Change from 98 to 64.
5532         (PRED_LOOP_GUARD): Change from 66 to 73.
5533
5534 2018-01-19  Martin Liska  <mliska@suse.cz>
5535
5536         * predict.c (predict_insn_def): Add new assert.
5537         (struct branch_predictor): Change type to signed integer.
5538         (test_prediction_value_range): Amend test to cover
5539         PROB_UNINITIALIZED.
5540         * predict.def (PRED_LOOP_ITERATIONS): Use the new constant.
5541         (PRED_LOOP_ITERATIONS_GUESSED): Likewise.
5542         (PRED_LOOP_ITERATIONS_MAX): Likewise.
5543         (PRED_LOOP_IV_COMPARE): Likewise.
5544         * predict.h (PROB_UNINITIALIZED): Define new constant.
5545
5546 2018-01-19  Martin Liska  <mliska@suse.cz>
5547
5548         * predict.c (dump_prediction): Add new format for
5549         analyze_brprob.py script which is enabled with -details
5550         suboption.
5551         * profile-count.h (precise_p): New function.
5552
5553 2018-01-19  Richard Sandiford  <richard.sandiford@linaro.org>
5554
5555         PR tree-optimization/83922
5556         * tree-vect-loop.c (vect_verify_full_masking): Return false if
5557         there are no statements that need masking.
5558         (vect_active_double_reduction_p): New function.
5559         (vect_analyze_loop_operations): Use it when handling phis that
5560         are not in the loop header.
5561
5562 2018-01-19  Richard Sandiford  <richard.sandiford@linaro.org>
5563
5564         PR tree-optimization/83914
5565         * tree-vect-loop.c (vectorizable_induction): Don't convert
5566         init_expr or apply the peeling adjustment for inductions
5567         that are nested within the vectorized loop.
5568
5569 2018-01-19  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
5570
5571         * config/arm/thumb2.md (*thumb2_negsi2_short): Use RSB mnemonic
5572         instead of NEG.
5573
5574 2018-01-18  Jakub Jelinek  <jakub@redhat.com>
5575
5576         PR sanitizer/81715
5577         PR testsuite/83882
5578         * function.h (gimplify_parameters): Add gimple_seq * argument.
5579         * function.c: Include gimple.h and options.h.
5580         (gimplify_parameters): Add cleanup argument, add CLOBBER stmts
5581         for the added local temporaries if needed.
5582         * gimplify.c (gimplify_body): Adjust gimplify_parameters caller,
5583         if there are any parameter cleanups, wrap whole body into a
5584         try/finally with the cleanups.
5585
5586 2018-01-18  Wilco Dijkstra  <wdijkstr@arm.com>
5587
5588         PR target/82964
5589         * config/aarch64/aarch64.c (aarch64_legitimate_constant_p):
5590         Use GET_MODE_CLASS for scalar floating point.
5591
5592 2018-01-18  Jan Hubicka  <hubicka@ucw.cz>
5593
5594         PR ipa/82256
5595         patch by PaX Team
5596         * cgraphclones.c (cgraph_node::create_version_clone_with_body):
5597         Fix call of call_cgraph_insertion_hooks.
5598
5599 2018-01-18  Martin Sebor  <msebor@redhat.com>
5600
5601         * doc/invoke.texi (-Wclass-memaccess): Tweak text.
5602
5603 2018-01-18  Jan Hubicka  <hubicka@ucw.cz>
5604
5605         PR ipa/83619
5606         * cgraph.c (cgraph_edge::redirect_call_stmt_to_callee): Update edge
5607         frequencies.
5608
5609 2018-01-18  Boris Kolpackov  <boris@codesynthesis.com>
5610
5611         PR other/70268
5612         * common.opt: (-ffile-prefix-map): New option.
5613         * opts.c (common_handle_option): Defer it.
5614         * opts-global.c (handle_common_deferred_options): Handle it.
5615         * debug.h (remap_debug_filename, add_debug_prefix_map): Move to...
5616         * file-prefix-map.h: New file.
5617         (remap_debug_filename, add_debug_prefix_map): ...here.
5618         (add_macro_prefix_map, add_file_prefix_map, remap_macro_filename): New.
5619         * final.c (debug_prefix_map, add_debug_prefix_map
5620         remap_debug_filename): Move to...
5621         * file-prefix-map.c: New file.
5622         (file_prefix_map, add_prefix_map, remap_filename) ...here and rename,
5623         generalize, get rid of alloca(), use strrchr() instead of strchr().
5624         (add_macro_prefix_map, add_debug_prefix_map, add_file_prefix_map):
5625         Implement in terms of add_prefix_map().
5626         (remap_macro_filename, remap_debug_filename): Implement in term of
5627         remap_filename().
5628         * Makefile.in (OBJS, PLUGIN_HEADERS): Add new files.
5629         * builtins.c (fold_builtin_FILE): Call remap_macro_filename().
5630         * dbxout.c: Include file-prefix-map.h.
5631         * varasm.c: Likewise.
5632         * vmsdbgout.c: Likewise.
5633         * xcoffout.c: Likewise.
5634         * dwarf2out.c: Likewise plus omit new options from DW_AT_producer.
5635         * doc/cppopts.texi (-fmacro-prefix-map): Document.
5636         * doc/invoke.texi (-ffile-prefix-map): Document.
5637         (-fdebug-prefix-map): Update description.
5638
5639 2018-01-18  Martin Liska  <mliska@suse.cz>
5640
5641         * config/i386/i386.c (indirect_thunk_name): Document that also
5642         lfence is emitted.
5643         (output_indirect_thunk): Document why both instructions
5644         (pause and lfence) are generated.
5645
5646 2018-01-18  Richard Biener  <rguenther@suse.de>
5647
5648         PR tree-optimization/83887
5649         * graphite-scop-detection.c
5650         (scop_detection::get_nearest_dom_with_single_entry): Remove.
5651         (scop_detection::get_nearest_pdom_with_single_exit): Likewise.
5652         (scop_detection::merge_sese): Re-implement with a flood-fill
5653         algorithm that properly finds a SESE region if it exists.
5654
5655 2018-01-18  Jakub Jelinek  <jakub@redhat.com>
5656
5657         PR c/61240
5658         * match.pd ((P + A) - P, P - (P + A), (P + A) - (P + B)): For
5659         pointer_diff optimizations use view_convert instead of convert.
5660
5661 2018-01-17  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
5662
5663         * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
5664         Generate different code for -mno-speculate-indirect-jumps.
5665         (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
5666         (*call_indirect_aix<mode>): Disable for
5667         -mno-speculate-indirect-jumps.
5668         (*call_indirect_aix<mode>_nospec): New define_insn.
5669         (*call_value_indirect_aix<mode>): Disable for
5670         -mno-speculate-indirect-jumps.
5671         (*call_value_indirect_aix<mode>_nospec): New define_insn.
5672         (*sibcall_nonlocal_sysv<mode>): Generate different code for
5673         -mno-speculate-indirect-jumps.
5674         (*sibcall_value_nonlocal_sysv<mode>): Likewise.
5675
5676 2018-01-17  Michael Meissner  <meissner@linux.vnet.ibm.com>
5677
5678         * config/rs6000/rs6000.c (rs6000_emit_move): If we load or store a
5679         long double type, set the flags for noting the default long double
5680         type, even if we don't pass or return a long double type.
5681
5682 2018-01-17  Jan Hubicka  <hubicka@ucw.cz>
5683
5684         PR ipa/83051
5685         * ipa-inline.c (flatten_function): Do not overwrite final inlining
5686         failure.
5687
5688 2018-01-17  Will Schmidt  <will_schmidt@vnet.ibm.com>
5689
5690         * config/rs6000/rs6000.c (rs6000_gimple_builtin): Add gimple folding
5691         support for merge[hl].
5692         (fold_mergehl_helper): New helper function.
5693         (tree-vector-builder.h): New #include for tree_vector_builder usage.
5694         * config/rs6000/altivec.md (altivec_vmrghw_direct): Add xxmrghw insn.
5695         (altivec_vmrglw_direct): Add xxmrglw insn.
5696
5697 2018-01-17  Andrew Waterman  <andrew@sifive.com>
5698
5699         * config/riscv/riscv.c (riscv_conditional_register_usage): If
5700         UNITS_PER_FP_ARG is 0, set call_used_regs to 1 for all FP regs.
5701
5702 2018-01-17  David Malcolm  <dmalcolm@redhat.com>
5703
5704         PR lto/83121
5705         * ipa-devirt.c (add_type_duplicate): When comparing memory layout,
5706         call the lto_location_cache before reading the
5707         DECL_SOURCE_LOCATION of the types.
5708
5709 2018-01-17  Wilco Dijkstra  <wdijkstr@arm.com>
5710             Richard Sandiford  <richard.sandiford@linaro.org>
5711
5712         * config/aarch64/aarch64.md (movti_aarch64): Use Uti constraint.
5713         * config/aarch64/aarch64.c (aarch64_mov128_immediate): New function.
5714         (aarch64_legitimate_constant_p): Just support CONST_DOUBLE
5715         SF/DF/TF mode to avoid creating illegal CONST_WIDE_INT immediates.
5716         * config/aarch64/aarch64-protos.h (aarch64_mov128_immediate):
5717         Add declaration.
5718         * config/aarch64/constraints.md (aarch64_movti_operand):
5719         Limit immediates.
5720         * config/aarch64/predicates.md (Uti): Add new constraint.
5721
5722 2018-01-17 Carl Love  <cel@us.ibm.com>
5723
5724         * config/rs6000/vsx.md (define_expand xl_len_r,
5725         define_expand stxvl, define_expand *stxvl): Add match_dup argument.
5726         (define_insn): Add, match_dup 1 argument to define_insn stxvll and
5727         lxvll.
5728         (define_expand, define_insn): Move the shift left from  the
5729         define_insn to the define_expand for lxvl and stxvl instructions.
5730         * config/rs6000/rs6000-builtin.def (BU_P9V_64BIT_VSX_2): Change LXVL
5731         and XL_LEN_R definitions to PURE.
5732
5733 2018-01-17  Uros Bizjak  <ubizjak@gmail.com>
5734
5735         * config/i386/i386.c (indirect_thunk_name): Declare regno
5736         as unsigned int.  Compare regno with INVALID_REGNUM.
5737         (output_indirect_thunk): Ditto.
5738         (output_indirect_thunk_function): Ditto.
5739         (ix86_code_end): Declare regno as unsigned int.  Use INVALID_REGNUM
5740         in the call to output_indirect_thunk_function.
5741
5742 2018-01-17  Richard Sandiford  <richard.sandiford@linaro.org>
5743
5744         PR middle-end/83884
5745         * expr.c (expand_expr_real_1): Use the size of GET_MODE (op0)
5746         rather than the size of inner_type to determine the stack slot size
5747         when handling VIEW_CONVERT_EXPRs on strict-alignment targets.
5748
5749 2018-01-16  Sebastian Peryt  <sebastian.peryt@intel.com>
5750
5751         PR target/83546
5752         * config/i386/i386.c (ix86_option_override_internal): Add PTA_RDRND
5753         to PTA_SILVERMONT.
5754
5755 2018-01-16  Michael Meissner  <meissner@linux.vnet.ibm.com>
5756
5757         * config.gcc (powerpc*-linux*-*): Add support for 64-bit little
5758         endian Linux systems to optionally enable multilibs for selecting
5759         the long double type if the user configured an explicit type.
5760         * config/rs6000/rs6000.h (TARGET_IEEEQUAD_MULTILIB): Indicate we
5761         have no long double multilibs if not defined.
5762         * config/rs6000/rs6000.c (rs6000_option_override_internal): Do not
5763         warn if the user used -mabi={ieee,ibm}longdouble and we built
5764         multilibs for long double.
5765         * config/rs6000/linux64.h (MULTILIB_DEFAULTS_IEEE): Define as the
5766         appropriate multilib option.
5767         (MULTILIB_DEFAULTS): Add MULTILIB_DEFAULTS_IEEE to the default
5768         multilib options.
5769         * config/rs6000/t-ldouble-linux64le-ibm: New configuration files
5770         for building long double multilibs.
5771         * config/rs6000/t-ldouble-linux64le-ieee: Likewise.
5772
5773 2018-01-16  John David Anglin  <danglin@gcc.gnu.org>
5774
5775         * config.gcc (hppa*-*-linux*): Change callee copies ABI to caller
5776         copies.
5777
5778         * config/pa.h (MALLOC_ABI_ALIGNMENT): Set 32-bit alignment default to
5779         64 bits.
5780         * config/pa/pa32-linux.h (MALLOC_ABI_ALIGNMENT): Set alignment to
5781         128 bits.
5782
5783         * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Cleanup type and mode
5784         variables.
5785
5786         * config/pa/pa.c (pa_function_arg_size): Apply CEIL to GET_MODE_SIZE
5787         return value.
5788
5789 2018-01-16  Eric Botcazou  <ebotcazou@adacore.com>
5790
5791         * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For an
5792         ADDR_EXPR, do not count the offset of a COMPONENT_REF twice.
5793
5794 2018-01-16  Kelvin Nilsen  <kelvin@gcc.gnu.org>
5795
5796         * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Generate
5797         different rtl trees depending on TARGET_64BIT.
5798         (rs6000_gen_lvx): Likewise.
5799
5800 2018-01-16  Eric Botcazou  <ebotcazou@adacore.com>
5801
5802         * config/visium/visium.md (nop): Tweak comment.
5803         (hazard_nop): Likewise.
5804
5805 2018-01-16  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
5806
5807         * config/rs6000/rs6000.c (rs6000_opt_vars): Add entry for
5808         -mspeculate-indirect-jumps.
5809         * config/rs6000/rs6000.md (*call_indirect_elfv2<mode>): Disable
5810         for -mno-speculate-indirect-jumps.
5811         (*call_indirect_elfv2<mode>_nospec): New define_insn.
5812         (*call_value_indirect_elfv2<mode>): Disable for
5813         -mno-speculate-indirect-jumps.
5814         (*call_value_indirect_elfv2<mode>_nospec): New define_insn.
5815         (indirect_jump): Emit different RTL for
5816         -mno-speculate-indirect-jumps.
5817         (*indirect_jump<mode>): Disable for
5818         -mno-speculate-indirect-jumps.
5819         (*indirect_jump<mode>_nospec): New define_insn.
5820         (tablejump): Emit different RTL for
5821         -mno-speculate-indirect-jumps.
5822         (tablejumpsi): Disable for -mno-speculate-indirect-jumps.
5823         (tablejumpsi_nospec): New define_expand.
5824         (tablejumpdi): Disable for -mno-speculate-indirect-jumps.
5825         (tablejumpdi_nospec): New define_expand.
5826         (*tablejump<mode>_internal1): Disable for
5827         -mno-speculate-indirect-jumps.
5828         (*tablejump<mode>_internal1_nospec): New define_insn.
5829         * config/rs6000/rs6000.opt (mspeculate-indirect-jumps): New
5830         option.
5831
5832 2018-01-16  Artyom Skrobov tyomitch@gmail.com
5833
5834         * caller-save.c (insert_save): Drop unnecessary parameter.  All
5835         callers updated.
5836
5837 2018-01-16  Jakub Jelinek  <jakub@redhat.com>
5838             Richard Biener  <rguenth@suse.de>
5839
5840         PR libgomp/83590
5841         * gimplify.c (gimplify_one_sizepos): For is_gimple_constant (expr)
5842         return early, inline manually is_gimple_sizepos.  Make sure if we
5843         call gimplify_expr we don't end up with a gimple constant.
5844         * tree.c (variably_modified_type_p): Don't return true for
5845         is_gimple_constant (_t).  Inline manually is_gimple_sizepos.
5846         * gimplify.h (is_gimple_sizepos): Remove.
5847
5848 2018-01-16  Richard Sandiford  <richard.sandiford@linaro.org>
5849
5850         PR tree-optimization/83857
5851         * tree-vect-loop.c (vect_analyze_loop_operations): Don't call
5852         vectorizable_live_operation for pure SLP statements.
5853         (vectorizable_live_operation): Handle PHIs.
5854
5855 2018-01-16  Richard Biener  <rguenther@suse.de>
5856
5857         PR tree-optimization/83867
5858         * tree-vect-stmts.c (vect_transform_stmt): Precompute
5859         nested_in_vect_loop_p since the scalar stmt may get invalidated.
5860
5861 2018-01-16  Jakub Jelinek  <jakub@redhat.com>
5862
5863         PR c/83844
5864         * stor-layout.c (handle_warn_if_not_align): Use byte_position and
5865         multiple_of_p instead of unchecked tree_to_uhwi and UHWI check.
5866         If off is not INTEGER_CST, issue a may not be aligned warning
5867         rather than isn't aligned.  Use isn%'t rather than isn't.
5868         * fold-const.c (multiple_of_p) <case BIT_AND_EXPR>: Don't fall through
5869         into MULT_EXPR.
5870         <case MULT_EXPR>: Improve the case when bottom and one of the
5871         MULT_EXPR operands are INTEGER_CSTs and bottom is multiple of that
5872         operand, in that case check if the other operand is multiple of
5873         bottom divided by the INTEGER_CST operand.
5874
5875 2018-01-16  Richard Sandiford  <richard.sandiford@linaro.org>
5876
5877         PR target/83858
5878         * config/pa/pa.h (FUNCTION_ARG_SIZE): Delete.
5879         * config/pa/pa-protos.h (pa_function_arg_size): Declare.
5880         * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Use
5881         pa_function_arg_size instead of FUNCTION_ARG_SIZE.
5882         * config/pa/pa.c (pa_function_arg_advance): Likewise.
5883         (pa_function_arg, pa_arg_partial_bytes): Likewise.
5884         (pa_function_arg_size): New function.
5885
5886 2018-01-16  Richard Sandiford  <richard.sandiford@linaro.org>
5887
5888         * fold-const.c (fold_ternary_loc): Construct the vec_perm_indices
5889         in a separate statement.
5890
5891 2018-01-16  Richard Sandiford  <richard.sandiford@linaro.org>
5892
5893         PR tree-optimization/83847
5894         * tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Don't
5895         group gathers and scatters.
5896
5897 2018-01-16  Jakub Jelinek  <jakub@redhat.com>
5898
5899         PR rtl-optimization/86620
5900         * params.def (max-sched-ready-insns): Bump minimum value to 1.
5901
5902         PR rtl-optimization/83213
5903         * recog.c (peep2_attempt): Copy over CROSSING_JUMP_P from peepinsn
5904         to last if both are JUMP_INSNs.
5905
5906         PR tree-optimization/83843
5907         * gimple-ssa-store-merging.c
5908         (imm_store_chain_info::output_merged_store): Handle bit_not_p on
5909         store_immediate_info for bswap/nop orig_stores.
5910
5911 2018-01-15  Andrew Waterman  <andrew@sifive.com>
5912
5913         * config/riscv/riscv.c (riscv_rtx_costs) <MULT>: Increase cost if
5914         !TARGET_MUL.
5915         <UDIV>: Increase cost if !TARGET_DIV.
5916
5917 2018-01-15  Segher Boessenkool  <segher@kernel.crashing.org>
5918
5919         * config/rs6000/rs6000.md (define_attr "type"): Remove delayed_cr.
5920         (define_attr "cr_logical_3op"): New.
5921         (cceq_ior_compare): Adjust.
5922         (cceq_ior_compare_complement): Adjust.
5923         (*cceq_rev_compare): Adjust.
5924         * config/rs6000/rs6000.c (rs6000_adjust_cost): Adjust.
5925         (is_cracked_insn): Adjust.
5926         (insn_must_be_first_in_group): Adjust.
5927         * config/rs6000/40x.md: Adjust.
5928         * config/rs6000/440.md: Adjust.
5929         * config/rs6000/476.md: Adjust.
5930         * config/rs6000/601.md: Adjust.
5931         * config/rs6000/603.md: Adjust.
5932         * config/rs6000/6xx.md: Adjust.
5933         * config/rs6000/7450.md: Adjust.
5934         * config/rs6000/7xx.md: Adjust.
5935         * config/rs6000/8540.md: Adjust.
5936         * config/rs6000/cell.md: Adjust.
5937         * config/rs6000/e300c2c3.md: Adjust.
5938         * config/rs6000/e500mc.md: Adjust.
5939         * config/rs6000/e500mc64.md: Adjust.
5940         * config/rs6000/e5500.md: Adjust.
5941         * config/rs6000/e6500.md: Adjust.
5942         * config/rs6000/mpc.md: Adjust.
5943         * config/rs6000/power4.md: Adjust.
5944         * config/rs6000/power5.md: Adjust.
5945         * config/rs6000/power6.md: Adjust.
5946         * config/rs6000/power7.md: Adjust.
5947         * config/rs6000/power8.md: Adjust.
5948         * config/rs6000/power9.md: Adjust.
5949         * config/rs6000/rs64.md: Adjust.
5950         * config/rs6000/titan.md: Adjust.
5951
5952 2018-01-15  H.J. Lu  <hongjiu.lu@intel.com>
5953
5954         * config/i386/predicates.md (indirect_branch_operand): Rewrite
5955         ix86_indirect_branch_register logic.
5956
5957 2018-01-15  H.J. Lu  <hongjiu.lu@intel.com>
5958
5959         * config/i386/constraints.md (Bs): Update
5960         ix86_indirect_branch_register check.  Don't check
5961         ix86_indirect_branch_register with GOT_memory_operand.
5962         (Bw): Likewise.
5963         * config/i386/predicates.md (GOT_memory_operand): Don't check
5964         ix86_indirect_branch_register here.
5965         (GOT32_symbol_operand): Likewise.
5966
5967 2018-01-15  H.J. Lu  <hongjiu.lu@intel.com>
5968
5969         * config/i386/predicates.md (constant_call_address_operand):
5970         Rewrite ix86_indirect_branch_register logic.
5971         (sibcall_insn_operand): Likewise.
5972
5973 2018-01-15  H.J. Lu  <hongjiu.lu@intel.com>
5974
5975         * config/i386/constraints.md (Bs): Replace
5976         ix86_indirect_branch_thunk_register with
5977         ix86_indirect_branch_register.
5978         (Bw): Likewise.
5979         * config/i386/i386.md (indirect_jump): Likewise.
5980         (tablejump): Likewise.
5981         (*sibcall_memory): Likewise.
5982         (*sibcall_value_memory): Likewise.
5983         Peepholes of indirect call and jump via memory: Likewise.
5984         * config/i386/i386.opt: Likewise.
5985         * config/i386/predicates.md (indirect_branch_operand): Likewise.
5986         (GOT_memory_operand): Likewise.
5987         (call_insn_operand): Likewise.
5988         (sibcall_insn_operand): Likewise.
5989         (GOT32_symbol_operand): Likewise.
5990
5991 2018-01-15  Jakub Jelinek  <jakub@redhat.com>
5992
5993         PR middle-end/83837
5994         * omp-expand.c (expand_omp_atomic_pipeline): Use loaded_val
5995         type rather than type addr's type points to.
5996         (expand_omp_atomic_mutex): Likewise.
5997         (expand_omp_atomic): Likewise.
5998
5999 2018-01-15  H.J. Lu  <hongjiu.lu@intel.com>
6000
6001         PR target/83839
6002         * config/i386/i386.c (output_indirect_thunk_function): Use
6003         ASM_OUTPUT_LABEL, instead of ASM_OUTPUT_DEF, for TARGET_MACHO
6004         for  __x86_return_thunk.
6005
6006 2018-01-15  Richard Biener  <rguenther@suse.de>
6007
6008         PR middle-end/83850
6009         * expmed.c (extract_bit_field_1): Fix typo.
6010
6011 2018-01-15  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
6012
6013         PR target/83687
6014         * config/arm/iterators.md (VF): New mode iterator.
6015         * config/arm/neon.md (neon_vabd<mode>_2): Use the above.
6016         Remove integer-related logic from pattern.
6017         (neon_vabd<mode>_3): Likewise.
6018
6019 2018-01-15  Jakub Jelinek  <jakub@redhat.com>
6020
6021         PR middle-end/82694
6022         * common.opt (fstrict-overflow): No longer an alias.
6023         (fwrapv-pointer): New option.
6024         * tree.h (TYPE_OVERFLOW_WRAPS, TYPE_OVERFLOW_UNDEFINED): Define
6025         also for pointer types based on flag_wrapv_pointer.
6026         * opts.c (common_handle_option) <case OPT_fstrict_overflow>: Set
6027         opts->x_flag_wrap[pv] to !value, clear opts->x_flag_trapv if
6028         opts->x_flag_wrapv got set.
6029         * fold-const.c (fold_comparison, fold_binary_loc): Revert 2017-08-01
6030         changes, just use TYPE_OVERFLOW_UNDEFINED on pointer type instead of
6031         POINTER_TYPE_OVERFLOW_UNDEFINED.
6032         * match.pd: Likewise in address comparison pattern.
6033         * doc/invoke.texi: Document -fwrapv and -fstrict-overflow.
6034
6035 2018-01-15  Richard Biener  <rguenther@suse.de>
6036
6037         PR lto/83804
6038         * tree.c (free_lang_data_in_type): Always unlink TYPE_DECLs
6039         from TYPE_FIELDS.  Free TYPE_BINFO if not used by devirtualization.
6040         Reset type names to their identifier if their TYPE_DECL doesn't
6041         have linkage (and thus is used for ODR and devirt).
6042         (save_debug_info_for_decl): Remove.
6043         (save_debug_info_for_type): Likewise.
6044         (add_tree_to_fld_list): Adjust.
6045         * tree-pretty-print.c (dump_generic_node): Make dumping of
6046         type names more robust.
6047
6048 2018-01-15  Richard Biener  <rguenther@suse.de>
6049
6050         * BASE-VER: Bump to 8.0.1.
6051
6052 2018-01-14  Martin Sebor  <msebor@redhat.com>
6053
6054         PR other/83508
6055         * builtins.c (check_access): Avoid warning when the no-warning bit
6056         is set.
6057
6058 2018-01-14  Cory Fields  <cory-nospam-@coryfields.com>
6059
6060         * tree-ssa-loop-im.c (sort_bbs_in_loop_postorder_cmp): Stabilize sort.
6061         * ira-color (allocno_hard_regs_compare): Likewise.
6062
6063 2018-01-14  Nathan Rossi  <nathan@nathanrossi.com>
6064
6065         PR target/83013
6066         * config/microblaze/microblaze.c (microblaze_asm_output_ident):
6067         Use .pushsection/.popsection.
6068
6069 2018-01-14  Martin Sebor  <msebor@redhat.com>
6070
6071         PR c++/81327
6072         * doc/invoke.texi (-Wlass-memaccess): Document suppression by casting.
6073
6074 2018-01-14  Jakub Jelinek  <jakub@redhat.com>
6075
6076         * config.gcc (i[34567]86-*-*): Remove one duplicate gfniintrin.h
6077         entry from extra_headers.
6078         (x86_64-*-*): Remove two duplicate gfniintrin.h entries from
6079         extra_headers, make the list bitwise identical to the i?86-*-* one.
6080
6081 2018-01-14  H.J. Lu  <hongjiu.lu@intel.com>
6082
6083         * config/i386/i386.c (ix86_set_indirect_branch_type): Disallow
6084         -mcmodel=large with -mindirect-branch=thunk,
6085         -mindirect-branch=thunk-extern, -mfunction-return=thunk and
6086         -mfunction-return=thunk-extern.
6087         * doc/invoke.texi: Document -mcmodel=large is incompatible with
6088         -mindirect-branch=thunk, -mindirect-branch=thunk-extern,
6089         -mfunction-return=thunk and -mfunction-return=thunk-extern.
6090
6091 2018-01-14  H.J. Lu  <hongjiu.lu@intel.com>
6092
6093         * config/i386/i386.c (print_reg): Print the name of the full
6094         integer register without '%'.
6095         (ix86_print_operand): Handle 'V'.
6096          * doc/extend.texi: Document 'V' modifier.
6097
6098 2018-01-14  H.J. Lu  <hongjiu.lu@intel.com>
6099
6100         * config/i386/constraints.md (Bs): Disallow memory operand for
6101         -mindirect-branch-register.
6102         (Bw): Likewise.
6103         * config/i386/predicates.md (indirect_branch_operand): Likewise.
6104         (GOT_memory_operand): Likewise.
6105         (call_insn_operand): Likewise.
6106         (sibcall_insn_operand): Likewise.
6107         (GOT32_symbol_operand): Likewise.
6108         * config/i386/i386.md (indirect_jump): Call convert_memory_address
6109         for -mindirect-branch-register.
6110         (tablejump): Likewise.
6111         (*sibcall_memory): Likewise.
6112         (*sibcall_value_memory): Likewise.
6113         Disallow peepholes of indirect call and jump via memory for
6114         -mindirect-branch-register.
6115         (*call_pop): Replace m with Bw.
6116         (*call_value_pop): Likewise.
6117         (*sibcall_pop_memory): Replace m with Bs.
6118         * config/i386/i386.opt (mindirect-branch-register): New option.
6119         * doc/invoke.texi: Document -mindirect-branch-register option.
6120
6121 2018-01-14  H.J. Lu  <hongjiu.lu@intel.com>
6122
6123         * config/i386/i386-protos.h (ix86_output_function_return): New.
6124         * config/i386/i386.c (ix86_set_indirect_branch_type): Also
6125         set function_return_type.
6126         (indirect_thunk_name): Add ret_p to indicate thunk for function
6127         return.
6128         (output_indirect_thunk_function): Pass false to
6129         indirect_thunk_name.
6130         (ix86_output_indirect_branch_via_reg): Likewise.
6131         (ix86_output_indirect_branch_via_push): Likewise.
6132         (output_indirect_thunk_function): Create alias for function
6133         return thunk if regno < 0.
6134         (ix86_output_function_return): New function.
6135         (ix86_handle_fndecl_attribute): Handle function_return.
6136         (ix86_attribute_table): Add function_return.
6137         * config/i386/i386.h (machine_function): Add
6138         function_return_type.
6139         * config/i386/i386.md (simple_return_internal): Use
6140         ix86_output_function_return.
6141         (simple_return_internal_long): Likewise.
6142         * config/i386/i386.opt (mfunction-return=): New option.
6143         (indirect_branch): Mention -mfunction-return=.
6144         * doc/extend.texi: Document function_return function attribute.
6145         * doc/invoke.texi: Document -mfunction-return= option.
6146
6147 2018-01-14  H.J. Lu  <hongjiu.lu@intel.com>
6148
6149         * config/i386/i386-opts.h (indirect_branch): New.
6150         * config/i386/i386-protos.h (ix86_output_indirect_jmp): Likewise.
6151         * config/i386/i386.c (ix86_using_red_zone): Disallow red-zone
6152         with local indirect jump when converting indirect call and jump.
6153         (ix86_set_indirect_branch_type): New.
6154         (ix86_set_current_function): Call ix86_set_indirect_branch_type.
6155         (indirectlabelno): New.
6156         (indirect_thunk_needed): Likewise.
6157         (indirect_thunk_bnd_needed): Likewise.
6158         (indirect_thunks_used): Likewise.
6159         (indirect_thunks_bnd_used): Likewise.
6160         (INDIRECT_LABEL): Likewise.
6161         (indirect_thunk_name): Likewise.
6162         (output_indirect_thunk): Likewise.
6163         (output_indirect_thunk_function): Likewise.
6164         (ix86_output_indirect_branch_via_reg): Likewise.
6165         (ix86_output_indirect_branch_via_push): Likewise.
6166         (ix86_output_indirect_branch): Likewise.
6167         (ix86_output_indirect_jmp): Likewise.
6168         (ix86_code_end): Call output_indirect_thunk_function if needed.
6169         (ix86_output_call_insn): Call ix86_output_indirect_branch if
6170         needed.
6171         (ix86_handle_fndecl_attribute): Handle indirect_branch.
6172         (ix86_attribute_table): Add indirect_branch.
6173         * config/i386/i386.h (machine_function): Add indirect_branch_type
6174         and has_local_indirect_jump.
6175         * config/i386/i386.md (indirect_jump): Set has_local_indirect_jump
6176         to true.
6177         (tablejump): Likewise.
6178         (*indirect_jump): Use ix86_output_indirect_jmp.
6179         (*tablejump_1): Likewise.
6180         (simple_return_indirect_internal): Likewise.
6181         * config/i386/i386.opt (mindirect-branch=): New option.
6182         (indirect_branch): New.
6183         (keep): Likewise.
6184         (thunk): Likewise.
6185         (thunk-inline): Likewise.
6186         (thunk-extern): Likewise.
6187         * doc/extend.texi: Document indirect_branch function attribute.
6188         * doc/invoke.texi: Document -mindirect-branch= option.
6189
6190 2018-01-14  Jan Hubicka  <hubicka@ucw.cz>
6191
6192         PR ipa/83051
6193         * ipa-inline.c (edge_badness): Tolerate roundoff errors.
6194
6195 2018-01-14  Richard Sandiford  <richard.sandiford@linaro.org>
6196
6197         * ipa-inline.c (want_inline_small_function_p): Return false if
6198         inlining has already failed with CIF_FINAL_ERROR.
6199         (update_caller_keys): Call want_inline_small_function_p before
6200         can_inline_edge_p.
6201         (update_callee_keys): Likewise.
6202
6203 2018-01-10  Kelvin Nilsen  <kelvin@gcc.gnu.org>
6204
6205         * config/rs6000/rs6000-p8swap.c (rs6000_sum_of_two_registers_p):
6206         New function.
6207         (rs6000_quadword_masked_address_p): Likewise.
6208         (quad_aligned_load_p): Likewise.
6209         (quad_aligned_store_p): Likewise.
6210         (const_load_sequence_p): Add comment to describe the outer-most loop.
6211         (mimic_memory_attributes_and_flags): New function.
6212         (rs6000_gen_stvx): Likewise.
6213         (replace_swapped_aligned_store): Likewise.
6214         (rs6000_gen_lvx): Likewise.
6215         (replace_swapped_aligned_load): Likewise.
6216         (replace_swapped_load_constant): Capitalize argument name in
6217         comment describing this function.
6218         (rs6000_analyze_swaps): Add a third pass to search for vector loads
6219         and stores that access quad-word aligned addresses and replace
6220         with stvx or lvx instructions when appropriate.
6221         * config/rs6000/rs6000-protos.h (rs6000_sum_of_two_registers_p):
6222         New function prototype.
6223         (rs6000_quadword_masked_address_p): Likewise.
6224         (rs6000_gen_lvx): Likewise.
6225         (rs6000_gen_stvx): Likewise.
6226         * config/rs6000/vsx.md (*vsx_le_perm_load_<mode>): For modes
6227         VSX_D (V2DF, V2DI), modify this split to select lvx instruction
6228         when memory address is aligned.
6229         (*vsx_le_perm_load_<mode>): For modes VSX_W (V4SF, V4SI), modify
6230         this split to select lvx instruction when memory address is aligned.
6231         (*vsx_le_perm_load_v8hi): Modify this split to select lvx
6232         instruction when memory address is aligned.
6233         (*vsx_le_perm_load_v16qi): Likewise.
6234         (four unnamed splitters): Modify to select the stvx instruction
6235         when memory is aligned.
6236
6237 2018-01-13  Jan Hubicka  <hubicka@ucw.cz>
6238
6239         * predict.c (determine_unlikely_bbs): Handle correctly BBs
6240         which appears in the queue multiple times.
6241
6242 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
6243             Alan Hayward  <alan.hayward@arm.com>
6244             David Sherwood  <david.sherwood@arm.com>
6245
6246         * tree-vectorizer.h (vec_lower_bound): New structure.
6247         (_loop_vec_info): Add check_nonzero and lower_bounds.
6248         (LOOP_VINFO_CHECK_NONZERO): New macro.
6249         (LOOP_VINFO_LOWER_BOUNDS): Likewise.
6250         (LOOP_REQUIRES_VERSIONING_FOR_ALIAS): Check lower_bounds too.
6251         * tree-data-ref.h (dr_with_seg_len): Add access_size and align
6252         fields.  Make seg_len the distance travelled, not including the
6253         access size.
6254         (dr_direction_indicator): Declare.
6255         (dr_zero_step_indicator): Likewise.
6256         (dr_known_forward_stride_p): Likewise.
6257         * tree-data-ref.c: Include stringpool.h, tree-vrp.h and
6258         tree-ssanames.h.
6259         (runtime_alias_check_p): Allow runtime alias checks with
6260         variable strides.
6261         (operator ==): Compare access_size and align.
6262         (prune_runtime_alias_test_list): Rework for new distinction between
6263         the access_size and seg_len.
6264         (create_intersect_range_checks_index): Likewise.  Cope with polynomial
6265         segment lengths.
6266         (get_segment_min_max): New function.
6267         (create_intersect_range_checks): Use it.
6268         (dr_step_indicator): New function.
6269         (dr_direction_indicator): Likewise.
6270         (dr_zero_step_indicator): Likewise.
6271         (dr_known_forward_stride_p): Likewise.
6272         * tree-loop-distribution.c (data_ref_segment_size): Return
6273         DR_STEP * (niters - 1).
6274         (compute_alias_check_pairs): Update call to the dr_with_seg_len
6275         constructor.
6276         * tree-vect-data-refs.c (vect_check_nonzero_value): New function.
6277         (vect_preserves_scalar_order_p): New function, split out from...
6278         (vect_analyze_data_ref_dependence): ...here.  Check for zero steps.
6279         (vect_vfa_segment_size): Return DR_STEP * (length_factor - 1).
6280         (vect_vfa_access_size): New function.
6281         (vect_vfa_align): Likewise.
6282         (vect_compile_time_alias): Take access_size_a and access_b arguments.
6283         (dump_lower_bound): New function.
6284         (vect_check_lower_bound): Likewise.
6285         (vect_small_gap_p): Likewise.
6286         (vectorizable_with_step_bound_p): Likewise.
6287         (vect_prune_runtime_alias_test_list): Ignore cross-iteration
6288         depencies if the vectorization factor is 1.  Convert the checks
6289         for nonzero steps into checks on the bounds of DR_STEP.  Try using
6290         a bunds check for variable steps if the minimum required step is
6291         relatively small. Update calls to the dr_with_seg_len
6292         constructor and to vect_compile_time_alias.
6293         * tree-vect-loop-manip.c (vect_create_cond_for_lower_bounds): New
6294         function.
6295         (vect_loop_versioning): Call it.
6296         * tree-vect-loop.c (vect_analyze_loop_2): Clear LOOP_VINFO_LOWER_BOUNDS
6297         when retrying.
6298         (vect_estimate_min_profitable_iters): Account for any bounds checks.
6299
6300 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
6301             Alan Hayward  <alan.hayward@arm.com>
6302             David Sherwood  <david.sherwood@arm.com>
6303
6304         * doc/sourcebuild.texi (vect_scatter_store): Document.
6305         * optabs.def (scatter_store_optab, mask_scatter_store_optab): New
6306         optabs.
6307         * doc/md.texi (scatter_store@var{m}, mask_scatter_store@var{m}):
6308         Document.
6309         * genopinit.c (main): Add supports_vec_scatter_store and
6310         supports_vec_scatter_store_cached to target_optabs.
6311         * gimple.h (gimple_expr_type): Handle IFN_SCATTER_STORE and
6312         IFN_MASK_SCATTER_STORE.
6313         * internal-fn.def (SCATTER_STORE, MASK_SCATTER_STORE): New internal
6314         functions.
6315         * internal-fn.h (internal_store_fn_p): Declare.
6316         (internal_fn_stored_value_index): Likewise.
6317         * internal-fn.c (scatter_store_direct): New macro.
6318         (expand_scatter_store_optab_fn): New function.
6319         (direct_scatter_store_optab_supported_p): New macro.
6320         (internal_store_fn_p): New function.
6321         (internal_gather_scatter_fn_p): Handle IFN_SCATTER_STORE and
6322         IFN_MASK_SCATTER_STORE.
6323         (internal_fn_mask_index): Likewise.
6324         (internal_fn_stored_value_index): New function.
6325         (internal_gather_scatter_fn_supported_p): Adjust operand numbers
6326         for scatter stores.
6327         * optabs-query.h (supports_vec_scatter_store_p): Declare.
6328         * optabs-query.c (supports_vec_scatter_store_p): New function.
6329         * tree-vectorizer.h (vect_get_store_rhs): Declare.
6330         * tree-vect-data-refs.c (vect_analyze_data_ref_access): Return
6331         true for scatter stores.
6332         (vect_gather_scatter_fn_p): Handle scatter stores too.
6333         (vect_check_gather_scatter): Consider using scatter stores if
6334         supports_vec_scatter_store_p.
6335         * tree-vect-patterns.c (vect_try_gather_scatter_pattern): Handle
6336         scatter stores too.
6337         * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
6338         internal_fn_stored_value_index.
6339         (check_load_store_masking): Handle scatter stores too.
6340         (vect_get_store_rhs): Make public.
6341         (vectorizable_call): Use internal_store_fn_p.
6342         (vectorizable_store): Handle scatter store internal functions.
6343         (vect_transform_stmt): Compare GROUP_STORE_COUNT with GROUP_SIZE
6344         when deciding whether the end of the group has been reached.
6345         * config/aarch64/aarch64.md (UNSPEC_ST1_SCATTER): New unspec.
6346         * config/aarch64/aarch64-sve.md (scatter_store<mode>): New expander.
6347         (mask_scatter_store<mode>): New insns.
6348
6349 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
6350             Alan Hayward  <alan.hayward@arm.com>
6351             David Sherwood  <david.sherwood@arm.com>
6352
6353         * tree-vectorizer.h (vect_gather_scatter_fn_p): Declare.
6354         * tree-vect-data-refs.c (vect_gather_scatter_fn_p): Make public.
6355         * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): New
6356         function.
6357         (vect_use_strided_gather_scatters_p): Take a masked_p argument.
6358         Use vect_truncate_gather_scatter_offset if we can't treat the
6359         operation as a normal gather load or scatter store.
6360         (get_group_load_store_type): Take the gather_scatter_info
6361         as argument.  Try using a gather load or scatter store for
6362         single-element groups.
6363         (get_load_store_type): Update calls to get_group_load_store_type
6364         and vect_use_strided_gather_scatters_p.
6365
6366 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
6367             Alan Hayward  <alan.hayward@arm.com>
6368             David Sherwood  <david.sherwood@arm.com>
6369
6370         * tree-vectorizer.h (vect_create_data_ref_ptr): Take an extra
6371         optional tree argument.
6372         * tree-vect-data-refs.c (vect_check_gather_scatter): Check for
6373         null target hooks.
6374         (vect_create_data_ref_ptr): Take the iv_step as an optional argument,
6375         but continue to use the current value as a fallback.
6376         (bump_vector_ptr): Use operand_equal_p rather than tree_int_cst_compare
6377         to compare the updates.
6378         * tree-vect-stmts.c (vect_use_strided_gather_scatters_p): New function.
6379         (get_load_store_type): Use it when handling a strided access.
6380         (vect_get_strided_load_store_ops): New function.
6381         (vect_get_data_ptr_increment): Likewise.
6382         (vectorizable_load): Handle strided gather loads.  Always pass
6383         a step to vect_create_data_ref_ptr and bump_vector_ptr.
6384
6385 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
6386             Alan Hayward  <alan.hayward@arm.com>
6387             David Sherwood  <david.sherwood@arm.com>
6388
6389         * doc/md.texi (gather_load@var{m}): Document.
6390         (mask_gather_load@var{m}): Likewise.
6391         * genopinit.c (main): Add supports_vec_gather_load and
6392         supports_vec_gather_load_cached to target_optabs.
6393         * optabs-tree.c (init_tree_optimization_optabs): Use
6394         ggc_cleared_alloc to allocate target_optabs.
6395         * optabs.def (gather_load_optab, mask_gather_laod_optab): New optabs.
6396         * internal-fn.def (GATHER_LOAD, MASK_GATHER_LOAD): New internal
6397         functions.
6398         * internal-fn.h (internal_load_fn_p): Declare.
6399         (internal_gather_scatter_fn_p): Likewise.
6400         (internal_fn_mask_index): Likewise.
6401         (internal_gather_scatter_fn_supported_p): Likewise.
6402         * internal-fn.c (gather_load_direct): New macro.
6403         (expand_gather_load_optab_fn): New function.
6404         (direct_gather_load_optab_supported_p): New macro.
6405         (direct_internal_fn_optab): New function.
6406         (internal_load_fn_p): Likewise.
6407         (internal_gather_scatter_fn_p): Likewise.
6408         (internal_fn_mask_index): Likewise.
6409         (internal_gather_scatter_fn_supported_p): Likewise.
6410         * optabs-query.c (supports_at_least_one_mode_p): New function.
6411         (supports_vec_gather_load_p): Likewise.
6412         * optabs-query.h (supports_vec_gather_load_p): Declare.
6413         * tree-vectorizer.h (gather_scatter_info): Add ifn, element_type
6414         and memory_type field.
6415         (NUM_PATTERNS): Bump to 15.
6416         * tree-vect-data-refs.c: Include internal-fn.h.
6417         (vect_gather_scatter_fn_p): New function.
6418         (vect_describe_gather_scatter_call): Likewise.
6419         (vect_check_gather_scatter): Try using internal functions for
6420         gather loads.  Recognize existing calls to a gather load function.
6421         (vect_analyze_data_refs): Consider using gather loads if
6422         supports_vec_gather_load_p.
6423         * tree-vect-patterns.c (vect_get_load_store_mask): New function.
6424         (vect_get_gather_scatter_offset_type): Likewise.
6425         (vect_convert_mask_for_vectype): Likewise.
6426         (vect_add_conversion_to_patterm): Likewise.
6427         (vect_try_gather_scatter_pattern): Likewise.
6428         (vect_recog_gather_scatter_pattern): New pattern recognizer.
6429         (vect_vect_recog_func_ptrs): Add it.
6430         * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
6431         internal_fn_mask_index and internal_gather_scatter_fn_p.
6432         (check_load_store_masking): Take the gather_scatter_info as an
6433         argument and handle gather loads.
6434         (vect_get_gather_scatter_ops): New function.
6435         (vectorizable_call): Check internal_load_fn_p.
6436         (vectorizable_load): Likewise.  Handle gather load internal
6437         functions.
6438         (vectorizable_store): Update call to check_load_store_masking.
6439         * config/aarch64/aarch64.md (UNSPEC_LD1_GATHER): New unspec.
6440         * config/aarch64/iterators.md (SVE_S, SVE_D): New mode iterators.
6441         * config/aarch64/predicates.md (aarch64_gather_scale_operand_w)
6442         (aarch64_gather_scale_operand_d): New predicates.
6443         * config/aarch64/aarch64-sve.md (gather_load<mode>): New expander.
6444         (mask_gather_load<mode>): New insns.
6445
6446 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
6447             Alan Hayward  <alan.hayward@arm.com>
6448             David Sherwood  <david.sherwood@arm.com>
6449
6450         * optabs.def (fold_left_plus_optab): New optab.
6451         * doc/md.texi (fold_left_plus_@var{m}): Document.
6452         * internal-fn.def (IFN_FOLD_LEFT_PLUS): New internal function.
6453         * internal-fn.c (fold_left_direct): Define.
6454         (expand_fold_left_optab_fn): Likewise.
6455         (direct_fold_left_optab_supported_p): Likewise.
6456         * fold-const-call.c (fold_const_fold_left): New function.
6457         (fold_const_call): Use it to fold CFN_FOLD_LEFT_PLUS.
6458         * tree-parloops.c (valid_reduction_p): New function.
6459         (gather_scalar_reductions): Use it.
6460         * tree-vectorizer.h (FOLD_LEFT_REDUCTION): New vect_reduction_type.
6461         (vect_finish_replace_stmt): Declare.
6462         * tree-vect-loop.c (fold_left_reduction_fn): New function.
6463         (needs_fold_left_reduction_p): New function, split out from...
6464         (vect_is_simple_reduction): ...here.  Accept reductions that
6465         forbid reassociation, but give them type FOLD_LEFT_REDUCTION.
6466         (vect_force_simple_reduction): Also store the reduction type in
6467         the assignment's STMT_VINFO_REDUC_TYPE.
6468         (vect_model_reduction_cost): Handle FOLD_LEFT_REDUCTION.
6469         (merge_with_identity): New function.
6470         (vect_expand_fold_left): Likewise.
6471         (vectorize_fold_left_reduction): Likewise.
6472         (vectorizable_reduction): Handle FOLD_LEFT_REDUCTION.  Leave the
6473         scalar phi in place for it.  Check for target support and reject
6474         cases that would reassociate the operation.  Defer the transform
6475         phase to vectorize_fold_left_reduction.
6476         * config/aarch64/aarch64.md (UNSPEC_FADDA): New unspec.
6477         * config/aarch64/aarch64-sve.md (fold_left_plus_<mode>): New expander.
6478         (*fold_left_plus_<mode>, *pred_fold_left_plus_<mode>): New insns.
6479
6480 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
6481
6482         * tree-if-conv.c (predicate_mem_writes): Remove redundant
6483         call to ifc_temp_var.
6484
6485 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
6486             Alan Hayward  <alan.hayward@arm.com>
6487             David Sherwood  <david.sherwood@arm.com>
6488
6489         * target.def (legitimize_address_displacement): Take the original
6490         offset as a poly_int.
6491         * targhooks.h (default_legitimize_address_displacement): Update
6492         accordingly.
6493         * targhooks.c (default_legitimize_address_displacement): Likewise.
6494         * doc/tm.texi: Regenerate.
6495         * lra-constraints.c (base_plus_disp_to_reg): Take the displacement
6496         as an argument, moving assert of ad->disp == ad->disp_term to...
6497         (process_address_1): ...here.  Update calls to base_plus_disp_to_reg.
6498         Try calling targetm.legitimize_address_displacement before expanding
6499         the address rather than afterwards, and adjust for the new interface.
6500         * config/aarch64/aarch64.c (aarch64_legitimize_address_displacement):
6501         Match the new hook interface.  Handle SVE addresses.
6502         * config/sh/sh.c (sh_legitimize_address_displacement): Make the
6503         new hook interface.
6504
6505 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
6506
6507         * Makefile.in (OBJS): Add early-remat.o.
6508         * target.def (select_early_remat_modes): New hook.
6509         * doc/tm.texi.in (TARGET_SELECT_EARLY_REMAT_MODES): New hook.
6510         * doc/tm.texi: Regenerate.
6511         * targhooks.h (default_select_early_remat_modes): Declare.
6512         * targhooks.c (default_select_early_remat_modes): New function.
6513         * timevar.def (TV_EARLY_REMAT): New timevar.
6514         * passes.def (pass_early_remat): New pass.
6515         * tree-pass.h (make_pass_early_remat): Declare.
6516         * early-remat.c: New file.
6517         * config/aarch64/aarch64.c (aarch64_select_early_remat_modes): New
6518         function.
6519         (TARGET_SELECT_EARLY_REMAT_MODES): Define.
6520
6521 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
6522             Alan Hayward  <alan.hayward@arm.com>
6523             David Sherwood  <david.sherwood@arm.com>
6524
6525         * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Replace
6526         vfm1 with a bound_epilog parameter.
6527         (vect_do_peeling): Update calls accordingly, and move the prologue
6528         call earlier in the function.  Treat the base bound_epilog as 0 for
6529         fully-masked loops and retain vf - 1 for other loops.  Add 1 to
6530         this base when peeling for gaps.
6531         * tree-vect-loop.c (vect_analyze_loop_2): Allow peeling for gaps
6532         with fully-masked loops.
6533         (vect_estimate_min_profitable_iters): Handle the single peeled
6534         iteration in that case.
6535
6536 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
6537             Alan Hayward  <alan.hayward@arm.com>
6538             David Sherwood  <david.sherwood@arm.com>
6539
6540         * tree-vect-data-refs.c (vect_analyze_group_access_1): Allow
6541         single-element interleaving even if the size is not a power of 2.
6542         * tree-vect-stmts.c (get_load_store_type): Disallow elementwise
6543         accesses for single-element interleaving if the group size is
6544         not a power of 2.
6545
6546 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
6547             Alan Hayward  <alan.hayward@arm.com>
6548             David Sherwood  <david.sherwood@arm.com>
6549
6550         * doc/md.texi (fold_extract_last_@var{m}): Document.
6551         * doc/sourcebuild.texi (vect_fold_extract_last): Likewise.
6552         * optabs.def (fold_extract_last_optab): New optab.
6553         * internal-fn.def (FOLD_EXTRACT_LAST): New internal function.
6554         * internal-fn.c (fold_extract_direct): New macro.
6555         (expand_fold_extract_optab_fn): Likewise.
6556         (direct_fold_extract_optab_supported_p): Likewise.
6557         * tree-vectorizer.h (EXTRACT_LAST_REDUCTION): New vect_reduction_type.
6558         * tree-vect-loop.c (vect_model_reduction_cost): Handle
6559         EXTRACT_LAST_REDUCTION.
6560         (get_initial_def_for_reduction): Do not create an initial vector
6561         for EXTRACT_LAST_REDUCTION reductions.
6562         (vectorizable_reduction): Leave the scalar phi in place for
6563         EXTRACT_LAST_REDUCTIONs.  Try using EXTRACT_LAST_REDUCTION
6564         ahead of INTEGER_INDUC_COND_REDUCTION.  Do not check for an
6565         epilogue code for EXTRACT_LAST_REDUCTION and defer the
6566         transform phase to vectorizable_condition.
6567         * tree-vect-stmts.c (vect_finish_stmt_generation_1): New function,
6568         split out from...
6569         (vect_finish_stmt_generation): ...here.
6570         (vect_finish_replace_stmt): New function.
6571         (vectorizable_condition): Handle EXTRACT_LAST_REDUCTION.
6572         * config/aarch64/aarch64-sve.md (fold_extract_last_<mode>): New
6573         pattern.
6574         * config/aarch64/aarch64.md (UNSPEC_CLASTB): New unspec.
6575
6576 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
6577             Alan Hayward  <alan.hayward@arm.com>
6578             David Sherwood  <david.sherwood@arm.com>
6579
6580         * doc/md.texi (extract_last_@var{m}): Document.
6581         * optabs.def (extract_last_optab): New optab.
6582         * internal-fn.def (EXTRACT_LAST): New internal function.
6583         * internal-fn.c (cond_unary_direct): New macro.
6584         (expand_cond_unary_optab_fn): Likewise.
6585         (direct_cond_unary_optab_supported_p): Likewise.
6586         * tree-vect-loop.c (vectorizable_live_operation): Allow fully-masked
6587         loops using EXTRACT_LAST.
6588         * config/aarch64/aarch64-sve.md (aarch64_sve_lastb<mode>): Rename to...
6589         (extract_last_<mode>): ...this optab.
6590         (vec_extract<mode><Vel>): Update accordingly.
6591
6592 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
6593             Alan Hayward  <alan.hayward@arm.com>
6594             David Sherwood  <david.sherwood@arm.com>
6595
6596         * target.def (empty_mask_is_expensive): New hook.
6597         * doc/tm.texi.in (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): New hook.
6598         * doc/tm.texi: Regenerate.
6599         * targhooks.h (default_empty_mask_is_expensive): Declare.
6600         * targhooks.c (default_empty_mask_is_expensive): New function.
6601         * tree-vectorizer.c (vectorize_loops): Only call optimize_mask_stores
6602         if the target says that empty masks are expensive.
6603         * config/aarch64/aarch64.c (aarch64_empty_mask_is_expensive):
6604         New function.
6605         (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): Redefine.
6606
6607 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
6608             Alan Hayward  <alan.hayward@arm.com>
6609             David Sherwood  <david.sherwood@arm.com>
6610
6611         * tree-vectorizer.h (_loop_vec_info::mask_skip_niters): New field.
6612         (LOOP_VINFO_MASK_SKIP_NITERS): New macro.
6613         (vect_use_loop_mask_for_alignment_p): New function.
6614         (vect_prepare_for_masked_peels, vect_gen_while_not): Declare.
6615         * tree-vect-loop-manip.c (vect_set_loop_masks_directly): Add an
6616         niters_skip argument.  Make sure that the first niters_skip elements
6617         of the first iteration are inactive.
6618         (vect_set_loop_condition_masked): Handle LOOP_VINFO_MASK_SKIP_NITERS.
6619         Update call to vect_set_loop_masks_directly.
6620         (get_misalign_in_elems): New function, split out from...
6621         (vect_gen_prolog_loop_niters): ...here.
6622         (vect_update_init_of_dr): Take a code argument that specifies whether
6623         the adjustment should be added or subtracted.
6624         (vect_update_init_of_drs): Likewise.
6625         (vect_prepare_for_masked_peels): New function.
6626         (vect_do_peeling): Skip prologue peeling if we're using a mask
6627         instead.  Update call to vect_update_inits_of_drs.
6628         * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
6629         mask_skip_niters.
6630         (vect_analyze_loop_2): Allow fully-masked loops with peeling for
6631         alignment.  Do not include the number of peeled iterations in
6632         the minimum threshold in that case.
6633         (vectorizable_induction): Adjust the start value down by
6634         LOOP_VINFO_MASK_SKIP_NITERS iterations.
6635         (vect_transform_loop): Call vect_prepare_for_masked_peels.
6636         Take the number of skipped iterations into account when calculating
6637         the loop bounds.
6638         * tree-vect-stmts.c (vect_gen_while_not): New function.
6639
6640 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
6641             Alan Hayward  <alan.hayward@arm.com>
6642             David Sherwood  <david.sherwood@arm.com>
6643
6644         * doc/sourcebuild.texi (vect_fully_masked): Document.
6645         * params.def (PARAM_MIN_VECT_LOOP_BOUND): Change minimum and
6646         default value to 0.
6647         * tree-vect-loop.c (vect_analyze_loop_costing): New function,
6648         split out from...
6649         (vect_analyze_loop_2): ...here. Don't check the vectorization
6650         factor against the number of loop iterations if the loop is
6651         fully-masked.
6652
6653 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
6654             Alan Hayward  <alan.hayward@arm.com>
6655             David Sherwood  <david.sherwood@arm.com>
6656
6657         * tree-ssa-loop-ivopts.c (USE_ADDRESS): Split into...
6658         (USE_REF_ADDRESS, USE_PTR_ADDRESS): ...these new use types.
6659         (dump_groups): Update accordingly.
6660         (iv_use::mem_type): New member variable.
6661         (address_p): New function.
6662         (record_use): Add a mem_type argument and initialize the new
6663         mem_type field.
6664         (record_group_use): Add a mem_type argument.  Use address_p.
6665         Remove obsolete null checks of base_object.  Update call to record_use.
6666         (find_interesting_uses_op): Update call to record_group_use.
6667         (find_interesting_uses_cond): Likewise.
6668         (find_interesting_uses_address): Likewise.
6669         (get_mem_type_for_internal_fn): New function.
6670         (find_address_like_use): Likewise.
6671         (find_interesting_uses_stmt): Try find_address_like_use before
6672         calling find_interesting_uses_op.
6673         (addr_offset_valid_p): Use the iv mem_type field as the type
6674         of the addressed memory.
6675         (add_autoinc_candidates): Likewise.
6676         (get_address_cost): Likewise.
6677         (split_small_address_groups_p): Use address_p.
6678         (split_address_groups): Likewise.
6679         (add_iv_candidate_for_use): Likewise.
6680         (autoinc_possible_for_pair): Likewise.
6681         (rewrite_groups): Likewise.
6682         (get_use_type): Check for USE_REF_ADDRESS instead of USE_ADDRESS.
6683         (determine_group_iv_cost): Update after split of USE_ADDRESS.
6684         (get_alias_ptr_type_for_ptr_address): New function.
6685         (rewrite_use_address): Rewrite address uses in calls that were
6686         identified by find_address_like_use.
6687
6688 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
6689             Alan Hayward  <alan.hayward@arm.com>
6690             David Sherwood  <david.sherwood@arm.com>
6691
6692         * expr.c (expand_expr_addr_expr_1): Handle ADDR_EXPRs of
6693         TARGET_MEM_REFs.
6694         * gimple-expr.h (is_gimple_addressable: Likewise.
6695         * gimple-expr.c (is_gimple_address): Likewise.
6696         * internal-fn.c (expand_call_mem_ref): New function.
6697         (expand_mask_load_optab_fn): Use it.
6698         (expand_mask_store_optab_fn): Likewise.
6699
6700 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
6701             Alan Hayward  <alan.hayward@arm.com>
6702             David Sherwood  <david.sherwood@arm.com>
6703
6704         * doc/md.texi (cond_add@var{mode}, cond_sub@var{mode})
6705         (cond_and@var{mode}, cond_ior@var{mode}, cond_xor@var{mode})
6706         (cond_smin@var{mode}, cond_smax@var{mode}, cond_umin@var{mode})
6707         (cond_umax@var{mode}): Document.
6708         * optabs.def (cond_add_optab, cond_sub_optab, cond_and_optab)
6709         (cond_ior_optab, cond_xor_optab, cond_smin_optab, cond_smax_optab)
6710         (cond_umin_optab, cond_umax_optab): New optabs.
6711         * internal-fn.def (COND_ADD, COND_SUB, COND_MIN, COND_MAX, COND_AND)
6712         (COND_IOR, COND_XOR): New internal functions.
6713         * internal-fn.h (get_conditional_internal_fn): Declare.
6714         * internal-fn.c (cond_binary_direct): New macro.
6715         (expand_cond_binary_optab_fn): Likewise.
6716         (direct_cond_binary_optab_supported_p): Likewise.
6717         (get_conditional_internal_fn): New function.
6718         * tree-vect-loop.c (vectorizable_reduction): Handle fully-masked loops.
6719         Cope with reduction statements that are vectorized as calls rather
6720         than assignments.
6721         * config/aarch64/aarch64-sve.md (cond_<optab><mode>): New insns.
6722         * config/aarch64/iterators.md (UNSPEC_COND_ADD, UNSPEC_COND_SUB)
6723         (UNSPEC_COND_SMAX, UNSPEC_COND_UMAX, UNSPEC_COND_SMIN)
6724         (UNSPEC_COND_UMIN, UNSPEC_COND_AND, UNSPEC_COND_ORR)
6725         (UNSPEC_COND_EOR): New unspecs.
6726         (optab): Add mappings for them.
6727         (SVE_COND_INT_OP, SVE_COND_FP_OP): New int iterators.
6728         (sve_int_op, sve_fp_op): New int attributes.
6729
6730 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
6731             Alan Hayward  <alan.hayward@arm.com>
6732             David Sherwood  <david.sherwood@arm.com>
6733
6734         * optabs.def (while_ult_optab): New optab.
6735         * doc/md.texi (while_ult@var{m}@var{n}): Document.
6736         * internal-fn.def (WHILE_ULT): New internal function.
6737         * internal-fn.h (direct_internal_fn_supported_p): New override
6738         that takes two types as argument.
6739         * internal-fn.c (while_direct): New macro.
6740         (expand_while_optab_fn): New function.
6741         (convert_optab_supported_p): Likewise.
6742         (direct_while_optab_supported_p): New macro.
6743         * wide-int.h (wi::udiv_ceil): New function.
6744         * tree-vectorizer.h (rgroup_masks): New structure.
6745         (vec_loop_masks): New typedef.
6746         (_loop_vec_info): Add masks, mask_compare_type, can_fully_mask_p
6747         and fully_masked_p.
6748         (LOOP_VINFO_CAN_FULLY_MASK_P, LOOP_VINFO_FULLY_MASKED_P)
6749         (LOOP_VINFO_MASKS, LOOP_VINFO_MASK_COMPARE_TYPE): New macros.
6750         (vect_max_vf): New function.
6751         (slpeel_make_loop_iterate_ntimes): Delete.
6752         (vect_set_loop_condition, vect_get_loop_mask_type, vect_gen_while)
6753         (vect_halve_mask_nunits, vect_double_mask_nunits): Declare.
6754         (vect_record_loop_mask, vect_get_loop_mask): Likewise.
6755         * tree-vect-loop-manip.c: Include tree-ssa-loop-niter.h,
6756         internal-fn.h, stor-layout.h and optabs-query.h.
6757         (vect_set_loop_mask): New function.
6758         (add_preheader_seq): Likewise.
6759         (add_header_seq): Likewise.
6760         (interleave_supported_p): Likewise.
6761         (vect_maybe_permute_loop_masks): Likewise.
6762         (vect_set_loop_masks_directly): Likewise.
6763         (vect_set_loop_condition_masked): Likewise.
6764         (vect_set_loop_condition_unmasked): New function, split out from
6765         slpeel_make_loop_iterate_ntimes.
6766         (slpeel_make_loop_iterate_ntimes): Rename to..
6767         (vect_set_loop_condition): ...this.  Use vect_set_loop_condition_masked
6768         for fully-masked loops and vect_set_loop_condition_unmasked otherwise.
6769         (vect_do_peeling): Update call accordingly.
6770         (vect_gen_vector_loop_niters): Use VF as the step for fully-masked
6771         loops.
6772         * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
6773         mask_compare_type, can_fully_mask_p and fully_masked_p.
6774         (release_vec_loop_masks): New function.
6775         (_loop_vec_info): Use it to free the loop masks.
6776         (can_produce_all_loop_masks_p): New function.
6777         (vect_get_max_nscalars_per_iter): Likewise.
6778         (vect_verify_full_masking): Likewise.
6779         (vect_analyze_loop_2): Save LOOP_VINFO_CAN_FULLY_MASK_P around
6780         retries, and free the mask rgroups before retrying.  Check loop-wide
6781         reasons for disallowing fully-masked loops.  Make the final decision
6782         about whether use a fully-masked loop or not.
6783         (vect_estimate_min_profitable_iters): Do not assume that peeling
6784         for the number of iterations will be needed for fully-masked loops.
6785         (vectorizable_reduction): Disable fully-masked loops.
6786         (vectorizable_live_operation): Likewise.
6787         (vect_halve_mask_nunits): New function.
6788         (vect_double_mask_nunits): Likewise.
6789         (vect_record_loop_mask): Likewise.
6790         (vect_get_loop_mask): Likewise.
6791         (vect_transform_loop): Handle the case in which the final loop
6792         iteration might handle a partial vector.  Call vect_set_loop_condition
6793         instead of slpeel_make_loop_iterate_ntimes.
6794         * tree-vect-stmts.c: Include tree-ssa-loop-niter.h and gimple-fold.h.
6795         (check_load_store_masking): New function.
6796         (prepare_load_store_mask): Likewise.
6797         (vectorizable_store): Handle fully-masked loops.
6798         (vectorizable_load): Likewise.
6799         (supportable_widening_operation): Use vect_halve_mask_nunits for
6800         booleans.
6801         (supportable_narrowing_operation): Likewise vect_double_mask_nunits.
6802         (vect_gen_while): New function.
6803         * config/aarch64/aarch64.md (umax<mode>3): New expander.
6804         (aarch64_uqdec<mode>): New insn.
6805
6806 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
6807             Alan Hayward  <alan.hayward@arm.com>
6808             David Sherwood  <david.sherwood@arm.com>
6809
6810         * optabs.def (reduc_and_scal_optab, reduc_ior_scal_optab)
6811         (reduc_xor_scal_optab): New optabs.
6812         * doc/md.texi (reduc_and_scal_@var{m}, reduc_ior_scal_@var{m})
6813         (reduc_xor_scal_@var{m}): Document.
6814         * doc/sourcebuild.texi (vect_logical_reduc): Likewise.
6815         * internal-fn.def (IFN_REDUC_AND, IFN_REDUC_IOR, IFN_REDUC_XOR): New
6816         internal functions.
6817         * fold-const-call.c (fold_const_call): Handle them.
6818         * tree-vect-loop.c (reduction_fn_for_scalar_code): Return the new
6819         internal functions for BIT_AND_EXPR, BIT_IOR_EXPR and BIT_XOR_EXPR.
6820         * config/aarch64/aarch64-sve.md (reduc_<bit_reduc>_scal_<mode>):
6821         (*reduc_<bit_reduc>_scal_<mode>): New patterns.
6822         * config/aarch64/iterators.md (UNSPEC_ANDV, UNSPEC_ORV)
6823         (UNSPEC_XORV): New unspecs.
6824         (optab): Add entries for them.
6825         (BITWISEV): New int iterator.
6826         (bit_reduc_op): New int attributes.
6827
6828 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
6829             Alan Hayward  <alan.hayward@arm.com>
6830             David Sherwood  <david.sherwood@arm.com>
6831
6832         * doc/md.texi (vec_shl_insert_@var{m}): New optab.
6833         * internal-fn.def (VEC_SHL_INSERT): New internal function.
6834         * optabs.def (vec_shl_insert_optab): New optab.
6835         * tree-vectorizer.h (can_duplicate_and_interleave_p): Declare.
6836         (duplicate_and_interleave): Likewise.
6837         * tree-vect-loop.c: Include internal-fn.h.
6838         (neutral_op_for_slp_reduction): New function, split out from
6839         get_initial_defs_for_reduction.
6840         (get_initial_def_for_reduction): Handle option 2 for variable-length
6841         vectors by loading the neutral value into a vector and then shifting
6842         the initial value into element 0.
6843         (get_initial_defs_for_reduction): Replace the code argument with
6844         the neutral value calculated by neutral_op_for_slp_reduction.
6845         Use gimple_build_vector for constant-length vectors.
6846         Use IFN_VEC_SHL_INSERT for variable-length vectors if all
6847         but the first group_size elements have a neutral value.
6848         Use duplicate_and_interleave otherwise.
6849         (vect_create_epilog_for_reduction): Take a neutral_op parameter.
6850         Update call to get_initial_defs_for_reduction.  Handle SLP
6851         reductions for variable-length vectors by creating one vector
6852         result for each scalar result, with the elements associated
6853         with other scalar results stubbed out with the neutral value.
6854         (vectorizable_reduction): Call neutral_op_for_slp_reduction.
6855         Require IFN_VEC_SHL_INSERT for double reductions on
6856         variable-length vectors, or SLP reductions that have
6857         a neutral value.  Require can_duplicate_and_interleave_p
6858         support for variable-length unchained SLP reductions if there
6859         is no neutral value, such as for MIN/MAX reductions.  Also require
6860         the number of vector elements to be a multiple of the number of
6861         SLP statements when doing variable-length unchained SLP reductions.
6862         Update call to vect_create_epilog_for_reduction.
6863         * tree-vect-slp.c (can_duplicate_and_interleave_p): Make public
6864         and remove initial values.
6865         (duplicate_and_interleave): Make public.
6866         * config/aarch64/aarch64.md (UNSPEC_INSR): New unspec.
6867         * config/aarch64/aarch64-sve.md (vec_shl_insert_<mode>): New insn.
6868
6869 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
6870             Alan Hayward  <alan.hayward@arm.com>
6871             David Sherwood  <david.sherwood@arm.com>
6872
6873         * tree-vect-slp.c: Include gimple-fold.h and internal-fn.h
6874         (can_duplicate_and_interleave_p): New function.
6875         (vect_get_and_check_slp_defs): Take the vector of statements
6876         rather than just the current one.  Remove excess parentheses.
6877         Restriction rejectinon of vect_constant_def and vect_external_def
6878         for variable-length vectors to boolean types, or types for which
6879         can_duplicate_and_interleave_p is false.
6880         (vect_build_slp_tree_2): Update call to vect_get_and_check_slp_defs.
6881         (duplicate_and_interleave): New function.
6882         (vect_get_constant_vectors): Use gimple_build_vector for
6883         constant-length vectors and suitable variable-length constant
6884         vectors.  Use duplicate_and_interleave for other variable-length
6885         vectors.  Don't defer the update when inserting new statements.
6886
6887 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
6888             Alan Hayward  <alan.hayward@arm.com>
6889             David Sherwood  <david.sherwood@arm.com>
6890
6891         * tree-vect-loop.c (vect_estimate_min_profitable_iters): Make sure
6892         min_profitable_iters doesn't go negative.
6893
6894 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
6895             Alan Hayward  <alan.hayward@arm.com>
6896             David Sherwood  <david.sherwood@arm.com>
6897
6898         * doc/md.texi (vec_mask_load_lanes@var{m}@var{n}): Document.
6899         (vec_mask_store_lanes@var{m}@var{n}): Likewise.
6900         * optabs.def (vec_mask_load_lanes_optab): New optab.
6901         (vec_mask_store_lanes_optab): Likewise.
6902         * internal-fn.def (MASK_LOAD_LANES): New internal function.
6903         (MASK_STORE_LANES): Likewise.
6904         * internal-fn.c (mask_load_lanes_direct): New macro.
6905         (mask_store_lanes_direct): Likewise.
6906         (expand_mask_load_optab_fn): Handle masked operations.
6907         (expand_mask_load_lanes_optab_fn): New macro.
6908         (expand_mask_store_optab_fn): Handle masked operations.
6909         (expand_mask_store_lanes_optab_fn): New macro.
6910         (direct_mask_load_lanes_optab_supported_p): Likewise.
6911         (direct_mask_store_lanes_optab_supported_p): Likewise.
6912         * tree-vectorizer.h (vect_store_lanes_supported): Take a masked_p
6913         parameter.
6914         (vect_load_lanes_supported): Likewise.
6915         * tree-vect-data-refs.c (strip_conversion): New function.
6916         (can_group_stmts_p): Likewise.
6917         (vect_analyze_data_ref_accesses): Use it instead of checking
6918         for a pair of assignments.
6919         (vect_store_lanes_supported): Take a masked_p parameter.
6920         (vect_load_lanes_supported): Likewise.
6921         * tree-vect-loop.c (vect_analyze_loop_2): Update calls to
6922         vect_store_lanes_supported and vect_load_lanes_supported.
6923         * tree-vect-slp.c (vect_analyze_slp_instance): Likewise.
6924         * tree-vect-stmts.c (get_group_load_store_type): Take a masked_p
6925         parameter.  Don't allow gaps for masked accesses.
6926         Use vect_get_store_rhs.  Update calls to vect_store_lanes_supported
6927         and vect_load_lanes_supported.
6928         (get_load_store_type): Take a masked_p parameter and update
6929         call to get_group_load_store_type.
6930         (vectorizable_store): Update call to get_load_store_type.
6931         Handle IFN_MASK_STORE_LANES.
6932         (vectorizable_load): Update call to get_load_store_type.
6933         Handle IFN_MASK_LOAD_LANES.
6934
6935 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
6936             Alan Hayward  <alan.hayward@arm.com>
6937             David Sherwood  <david.sherwood@arm.com>
6938
6939         * config/aarch64/aarch64-modes.def: Define x2, x3 and x4 vector
6940         modes for SVE.
6941         * config/aarch64/aarch64-protos.h
6942         (aarch64_sve_struct_memory_operand_p): Declare.
6943         * config/aarch64/iterators.md (SVE_STRUCT): New mode iterator.
6944         (vector_count, insn_length, VSINGLE, vsingle): New mode attributes.
6945         (VPRED, vpred): Handle SVE structure modes.
6946         * config/aarch64/constraints.md (Utx): New constraint.
6947         * config/aarch64/predicates.md (aarch64_sve_struct_memory_operand)
6948         (aarch64_sve_struct_nonimmediate_operand): New predicates.
6949         * config/aarch64/aarch64.md (UNSPEC_LDN, UNSPEC_STN): New unspecs.
6950         * config/aarch64/aarch64-sve.md (mov<mode>, *aarch64_sve_mov<mode>_le)
6951         (*aarch64_sve_mov<mode>_be, pred_mov<mode>): New patterns for
6952         structure modes.  Split into pieces after RA.
6953         (vec_load_lanes<mode><vsingle>, vec_mask_load_lanes<mode><vsingle>)
6954         (vec_store_lanes<mode><vsingle>, vec_mask_store_lanes<mode><vsingle>):
6955         New patterns.
6956         * config/aarch64/aarch64.c (aarch64_classify_vector_mode): Handle
6957         SVE structure modes.
6958         (aarch64_classify_address): Likewise.
6959         (sizetochar): Move earlier in file.
6960         (aarch64_print_operand): Handle SVE register lists.
6961         (aarch64_array_mode): New function.
6962         (aarch64_sve_struct_memory_operand_p): Likewise.
6963         (TARGET_ARRAY_MODE): Redefine.
6964
6965 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
6966             Alan Hayward  <alan.hayward@arm.com>
6967             David Sherwood  <david.sherwood@arm.com>
6968
6969         * target.def (array_mode): New target hook.
6970         * doc/tm.texi.in (TARGET_ARRAY_MODE): New hook.
6971         * doc/tm.texi: Regenerate.
6972         * hooks.h (hook_optmode_mode_uhwi_none): Declare.
6973         * hooks.c (hook_optmode_mode_uhwi_none): New function.
6974         * tree-vect-data-refs.c (vect_lanes_optab_supported_p): Use
6975         targetm.array_mode.
6976         * stor-layout.c (mode_for_array): Likewise.  Support polynomial
6977         type sizes.
6978
6979 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
6980             Alan Hayward  <alan.hayward@arm.com>
6981             David Sherwood  <david.sherwood@arm.com>
6982
6983         * fold-const.c (fold_binary_loc): Check the argument types
6984         rather than the result type when testing for a vector operation.
6985
6986 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
6987
6988         * doc/tm.texi.in (DWARF_LAZY_REGISTER_VALUE): Document.
6989         * doc/tm.texi: Regenerate.
6990
6991 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
6992             Alan Hayward  <alan.hayward@arm.com>
6993             David Sherwood  <david.sherwood@arm.com>
6994
6995         * doc/invoke.texi (-msve-vector-bits=): Document new option.
6996         (sve): Document new AArch64 extension.
6997         * doc/md.texi (w): Extend the description of the AArch64
6998         constraint to include SVE vectors.
6999         (Upl, Upa): Document new AArch64 predicate constraints.
7000         * config/aarch64/aarch64-opts.h (aarch64_sve_vector_bits_enum): New
7001         enum.
7002         * config/aarch64/aarch64.opt (sve_vector_bits): New enum.
7003         (msve-vector-bits=): New option.
7004         * config/aarch64/aarch64-option-extensions.def (fp, simd): Disable
7005         SVE when these are disabled.
7006         (sve): New extension.
7007         * config/aarch64/aarch64-modes.def: Define SVE vector and predicate
7008         modes.  Adjust their number of units based on aarch64_sve_vg.
7009         (MAX_BITSIZE_MODE_ANY_MODE): Define.
7010         * config/aarch64/aarch64-protos.h (ADDR_QUERY_ANY): New
7011         aarch64_addr_query_type.
7012         (aarch64_const_vec_all_same_in_range_p, aarch64_sve_pred_mode)
7013         (aarch64_sve_cnt_immediate_p, aarch64_sve_addvl_addpl_immediate_p)
7014         (aarch64_sve_inc_dec_immediate_p, aarch64_add_offset_temporaries)
7015         (aarch64_split_add_offset, aarch64_output_sve_cnt_immediate)
7016         (aarch64_output_sve_addvl_addpl, aarch64_output_sve_inc_dec_immediate)
7017         (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): Declare.
7018         (aarch64_simd_imm_zero_p): Delete.
7019         (aarch64_check_zero_based_sve_index_immediate): Declare.
7020         (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
7021         (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
7022         (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
7023         (aarch64_sve_float_mul_immediate_p): Likewise.
7024         (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
7025         rather than an rtx.
7026         (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): Declare.
7027         (aarch64_expand_mov_immediate): Take a gen_vec_duplicate callback.
7028         (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move): Declare.
7029         (aarch64_expand_sve_vec_cmp_int, aarch64_expand_sve_vec_cmp_float)
7030         (aarch64_expand_sve_vcond, aarch64_expand_sve_vec_perm): Declare.
7031         (aarch64_regmode_natural_size): Likewise.
7032         * config/aarch64/aarch64.h (AARCH64_FL_SVE): New macro.
7033         (AARCH64_FL_V8_3, AARCH64_FL_RCPC, AARCH64_FL_DOTPROD): Shift
7034         left one place.
7035         (AARCH64_ISA_SVE, TARGET_SVE): New macros.
7036         (FIXED_REGISTERS, CALL_USED_REGISTERS, REGISTER_NAMES): Add entries
7037         for VG and the SVE predicate registers.
7038         (V_ALIASES): Add a "z"-prefixed alias.
7039         (FIRST_PSEUDO_REGISTER): Change to P15_REGNUM + 1.
7040         (AARCH64_DWARF_VG, AARCH64_DWARF_P0): New macros.
7041         (PR_REGNUM_P, PR_LO_REGNUM_P): Likewise.
7042         (PR_LO_REGS, PR_HI_REGS, PR_REGS): New reg_classes.
7043         (REG_CLASS_NAMES): Add entries for them.
7044         (REG_CLASS_CONTENTS): Likewise.  Update ALL_REGS to include VG
7045         and the predicate registers.
7046         (aarch64_sve_vg): Declare.
7047         (BITS_PER_SVE_VECTOR, BYTES_PER_SVE_VECTOR, BYTES_PER_SVE_PRED)
7048         (SVE_BYTE_MODE, MAX_COMPILE_TIME_VEC_BYTES): New macros.
7049         (REGMODE_NATURAL_SIZE): Define.
7050         * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins): Handle
7051         SVE macros.
7052         * config/aarch64/aarch64.c: Include cfgrtl.h.
7053         (simd_immediate_info): Add a constructor for series vectors,
7054         and an associated step field.
7055         (aarch64_sve_vg): New variable.
7056         (aarch64_dbx_register_number): Handle VG and the predicate registers.
7057         (aarch64_vect_struct_mode_p, aarch64_vector_mode_p): Delete.
7058         (VEC_ADVSIMD, VEC_SVE_DATA, VEC_SVE_PRED, VEC_STRUCT, VEC_ANY_SVE)
7059         (VEC_ANY_DATA, VEC_STRUCT): New constants.
7060         (aarch64_advsimd_struct_mode_p, aarch64_sve_pred_mode_p)
7061         (aarch64_classify_vector_mode, aarch64_vector_data_mode_p)
7062         (aarch64_sve_data_mode_p, aarch64_sve_pred_mode)
7063         (aarch64_get_mask_mode): New functions.
7064         (aarch64_hard_regno_nregs): Handle SVE data modes for FP_REGS
7065         and FP_LO_REGS.  Handle PR_REGS, PR_LO_REGS and PR_HI_REGS.
7066         (aarch64_hard_regno_mode_ok): Handle VG.  Also handle the SVE
7067         predicate modes and predicate registers.  Explicitly restrict
7068         GPRs to modes of 16 bytes or smaller.  Only allow FP registers
7069         to store a vector mode if it is recognized by
7070         aarch64_classify_vector_mode.
7071         (aarch64_regmode_natural_size): New function.
7072         (aarch64_hard_regno_caller_save_mode): Return the original mode
7073         for predicates.
7074         (aarch64_sve_cnt_immediate_p, aarch64_output_sve_cnt_immediate)
7075         (aarch64_sve_addvl_addpl_immediate_p, aarch64_output_sve_addvl_addpl)
7076         (aarch64_sve_inc_dec_immediate_p, aarch64_output_sve_inc_dec_immediate)
7077         (aarch64_add_offset_1_temporaries, aarch64_offset_temporaries): New
7078         functions.
7079         (aarch64_add_offset): Add a temp2 parameter.  Assert that temp1
7080         does not overlap dest if the function is frame-related.  Handle
7081         SVE constants.
7082         (aarch64_split_add_offset): New function.
7083         (aarch64_add_sp, aarch64_sub_sp): Add temp2 parameters and pass
7084         them aarch64_add_offset.
7085         (aarch64_allocate_and_probe_stack_space): Add a temp2 parameter
7086         and update call to aarch64_sub_sp.
7087         (aarch64_add_cfa_expression): New function.
7088         (aarch64_expand_prologue): Pass extra temporary registers to the
7089         functions above.  Handle the case in which we need to emit new
7090         DW_CFA_expressions for registers that were originally saved
7091         relative to the stack pointer, but now have to be expressed
7092         relative to the frame pointer.
7093         (aarch64_output_mi_thunk): Pass extra temporary registers to the
7094         functions above.
7095         (aarch64_expand_epilogue): Likewise.  Prevent inheritance of
7096         IP0 and IP1 values for SVE frames.
7097         (aarch64_expand_vec_series): New function.
7098         (aarch64_expand_sve_widened_duplicate): Likewise.
7099         (aarch64_expand_sve_const_vector): Likewise.
7100         (aarch64_expand_mov_immediate): Add a gen_vec_duplicate parameter.
7101         Handle SVE constants.  Use emit_move_insn to move a force_const_mem
7102         into the register, rather than emitting a SET directly.
7103         (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move)
7104         (aarch64_get_reg_raw_mode, offset_4bit_signed_scaled_p)
7105         (offset_6bit_unsigned_scaled_p, aarch64_offset_7bit_signed_scaled_p)
7106         (offset_9bit_signed_scaled_p): New functions.
7107         (aarch64_replicate_bitmask_imm): New function.
7108         (aarch64_bitmask_imm): Use it.
7109         (aarch64_cannot_force_const_mem): Reject expressions involving
7110         a CONST_POLY_INT.  Update call to aarch64_classify_symbol.
7111         (aarch64_classify_index): Handle SVE indices, by requiring
7112         a plain register index with a scale that matches the element size.
7113         (aarch64_classify_address): Handle SVE addresses.  Assert that
7114         the mode of the address is VOIDmode or an integer mode.
7115         Update call to aarch64_classify_symbol.
7116         (aarch64_classify_symbolic_expression): Update call to
7117         aarch64_classify_symbol.
7118         (aarch64_const_vec_all_in_range_p): New function.
7119         (aarch64_print_vector_float_operand): Likewise.
7120         (aarch64_print_operand): Handle 'N' and 'C'.  Use "zN" rather than
7121         "vN" for FP registers with SVE modes.  Handle (const ...) vectors
7122         and the FP immediates 1.0 and 0.5.
7123         (aarch64_print_address_internal): Handle SVE addresses.
7124         (aarch64_print_operand_address): Use ADDR_QUERY_ANY.
7125         (aarch64_regno_regclass): Handle predicate registers.
7126         (aarch64_secondary_reload): Handle big-endian reloads of SVE
7127         data modes.
7128         (aarch64_class_max_nregs): Handle SVE modes and predicate registers.
7129         (aarch64_rtx_costs): Check for ADDVL and ADDPL instructions.
7130         (aarch64_convert_sve_vector_bits): New function.
7131         (aarch64_override_options): Use it to handle -msve-vector-bits=.
7132         (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
7133         rather than an rtx.
7134         (aarch64_legitimate_constant_p): Use aarch64_classify_vector_mode.
7135         Handle SVE vector and predicate modes.  Accept VL-based constants
7136         that need only one temporary register, and VL offsets that require
7137         no temporary registers.
7138         (aarch64_conditional_register_usage): Mark the predicate registers
7139         as fixed if SVE isn't available.
7140         (aarch64_vector_mode_supported_p): Use aarch64_classify_vector_mode.
7141         Return true for SVE vector and predicate modes.
7142         (aarch64_simd_container_mode): Take the number of bits as a poly_int64
7143         rather than an unsigned int.  Handle SVE modes.
7144         (aarch64_preferred_simd_mode): Update call accordingly.  Handle
7145         SVE modes.
7146         (aarch64_autovectorize_vector_sizes): Add BYTES_PER_SVE_VECTOR
7147         if SVE is enabled.
7148         (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
7149         (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
7150         (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
7151         (aarch64_sve_float_mul_immediate_p): New functions.
7152         (aarch64_sve_valid_immediate): New function.
7153         (aarch64_simd_valid_immediate): Use it as the fallback for SVE vectors.
7154         Explicitly reject structure modes.  Check for INDEX constants.
7155         Handle PTRUE and PFALSE constants.
7156         (aarch64_check_zero_based_sve_index_immediate): New function.
7157         (aarch64_simd_imm_zero_p): Delete.
7158         (aarch64_mov_operand_p): Use aarch64_simd_valid_immediate for
7159         vector modes.  Accept constants in the range of CNT[BHWD].
7160         (aarch64_simd_scalar_immediate_valid_for_move): Explicitly
7161         ask for an Advanced SIMD mode.
7162         (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): New functions.
7163         (aarch64_simd_vector_alignment): Handle SVE predicates.
7164         (aarch64_vectorize_preferred_vector_alignment): New function.
7165         (aarch64_simd_vector_alignment_reachable): Use it instead of
7166         the vector size.
7167         (aarch64_shift_truncation_mask): Use aarch64_vector_data_mode_p.
7168         (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): New
7169         functions.
7170         (MAX_VECT_LEN): Delete.
7171         (expand_vec_perm_d): Add a vec_flags field.
7172         (emit_unspec2, aarch64_expand_sve_vec_perm): New functions.
7173         (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_zip)
7174         (aarch64_evpc_ext): Don't apply a big-endian lane correction
7175         for SVE modes.
7176         (aarch64_evpc_rev): Rename to...
7177         (aarch64_evpc_rev_local): ...this.  Use a predicated operation for SVE.
7178         (aarch64_evpc_rev_global): New function.
7179         (aarch64_evpc_dup): Enforce a 64-byte range for SVE DUP.
7180         (aarch64_evpc_tbl): Use MAX_COMPILE_TIME_VEC_BYTES instead of
7181         MAX_VECT_LEN.
7182         (aarch64_evpc_sve_tbl): New function.
7183         (aarch64_expand_vec_perm_const_1): Update after rename of
7184         aarch64_evpc_rev.  Handle SVE permutes too, trying
7185         aarch64_evpc_rev_global and using aarch64_evpc_sve_tbl rather
7186         than aarch64_evpc_tbl.
7187         (aarch64_vectorize_vec_perm_const): Initialize vec_flags.
7188         (aarch64_sve_cmp_operand_p, aarch64_unspec_cond_code)
7189         (aarch64_gen_unspec_cond, aarch64_expand_sve_vec_cmp_int)
7190         (aarch64_emit_unspec_cond, aarch64_emit_unspec_cond_or)
7191         (aarch64_emit_inverted_unspec_cond, aarch64_expand_sve_vec_cmp_float)
7192         (aarch64_expand_sve_vcond): New functions.
7193         (aarch64_modes_tieable_p): Use aarch64_vector_data_mode_p instead
7194         of aarch64_vector_mode_p.
7195         (aarch64_dwarf_poly_indeterminate_value): New function.
7196         (aarch64_compute_pressure_classes): Likewise.
7197         (aarch64_can_change_mode_class): Likewise.
7198         (TARGET_GET_RAW_RESULT_MODE, TARGET_GET_RAW_ARG_MODE): Redefine.
7199         (TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT): Likewise.
7200         (TARGET_VECTORIZE_GET_MASK_MODE): Likewise.
7201         (TARGET_DWARF_POLY_INDETERMINATE_VALUE): Likewise.
7202         (TARGET_COMPUTE_PRESSURE_CLASSES): Likewise.
7203         (TARGET_CAN_CHANGE_MODE_CLASS): Likewise.
7204         * config/aarch64/constraints.md (Upa, Upl, Uav, Uat, Usv, Usi, Utr)
7205         (Uty, Dm, vsa, vsc, vsd, vsi, vsn, vsl, vsm, vsA, vsM, vsN): New
7206         constraints.
7207         (Dn, Dl, Dr): Accept const as well as const_vector.
7208         (Dz): Likewise.  Compare against CONST0_RTX.
7209         * config/aarch64/iterators.md: Refer to "Advanced SIMD" instead
7210         of "vector" where appropriate.
7211         (SVE_ALL, SVE_BH, SVE_BHS, SVE_BHSI, SVE_HSDI, SVE_HSF, SVE_SD)
7212         (SVE_SDI, SVE_I, SVE_F, PRED_ALL, PRED_BHS): New mode iterators.
7213         (UNSPEC_SEL, UNSPEC_ANDF, UNSPEC_IORF, UNSPEC_XORF, UNSPEC_COND_LT)
7214         (UNSPEC_COND_LE, UNSPEC_COND_EQ, UNSPEC_COND_NE, UNSPEC_COND_GE)
7215         (UNSPEC_COND_GT, UNSPEC_COND_LO, UNSPEC_COND_LS, UNSPEC_COND_HS)
7216         (UNSPEC_COND_HI, UNSPEC_COND_UO): New unspecs.
7217         (Vetype, VEL, Vel, VWIDE, Vwide, vw, vwcore, V_INT_EQUIV)
7218         (v_int_equiv): Extend to SVE modes.
7219         (Vesize, V128, v128, Vewtype, V_FP_EQUIV, v_fp_equiv, VPRED): New
7220         mode attributes.
7221         (LOGICAL_OR, SVE_INT_UNARY, SVE_FP_UNARY): New code iterators.
7222         (optab): Handle popcount, smin, smax, umin, umax, abs and sqrt.
7223         (logical_nn, lr, sve_int_op, sve_fp_op): New code attributs.
7224         (LOGICALF, OPTAB_PERMUTE, UNPACK, UNPACK_UNSIGNED, SVE_COND_INT_CMP)
7225         (SVE_COND_FP_CMP): New int iterators.
7226         (perm_hilo): Handle the new unpack unspecs.
7227         (optab, logicalf_op, su, perm_optab, cmp_op, imm_con): New int
7228         attributes.
7229         * config/aarch64/predicates.md (aarch64_sve_cnt_immediate)
7230         (aarch64_sve_addvl_addpl_immediate, aarch64_split_add_offset_immediate)
7231         (aarch64_pluslong_or_poly_operand, aarch64_nonmemory_operand)
7232         (aarch64_equality_operator, aarch64_constant_vector_operand)
7233         (aarch64_sve_ld1r_operand, aarch64_sve_ldr_operand): New predicates.
7234         (aarch64_sve_nonimmediate_operand): Likewise.
7235         (aarch64_sve_general_operand): Likewise.
7236         (aarch64_sve_dup_operand, aarch64_sve_arith_immediate): Likewise.
7237         (aarch64_sve_sub_arith_immediate, aarch64_sve_inc_dec_immediate)
7238         (aarch64_sve_logical_immediate, aarch64_sve_mul_immediate): Likewise.
7239         (aarch64_sve_dup_immediate, aarch64_sve_cmp_vsc_immediate): Likewise.
7240         (aarch64_sve_cmp_vsd_immediate, aarch64_sve_index_immediate): Likewise.
7241         (aarch64_sve_float_arith_immediate): Likewise.
7242         (aarch64_sve_float_arith_with_sub_immediate): Likewise.
7243         (aarch64_sve_float_mul_immediate, aarch64_sve_arith_operand): Likewise.
7244         (aarch64_sve_add_operand, aarch64_sve_logical_operand): Likewise.
7245         (aarch64_sve_lshift_operand, aarch64_sve_rshift_operand): Likewise.
7246         (aarch64_sve_mul_operand, aarch64_sve_cmp_vsc_operand): Likewise.
7247         (aarch64_sve_cmp_vsd_operand, aarch64_sve_index_operand): Likewise.
7248         (aarch64_sve_float_arith_operand): Likewise.
7249         (aarch64_sve_float_arith_with_sub_operand): Likewise.
7250         (aarch64_sve_float_mul_operand): Likewise.
7251         (aarch64_sve_vec_perm_operand): Likewise.
7252         (aarch64_pluslong_operand): Include aarch64_sve_addvl_addpl_immediate.
7253         (aarch64_mov_operand): Accept const_poly_int and const_vector.
7254         (aarch64_simd_lshift_imm, aarch64_simd_rshift_imm): Accept const
7255         as well as const_vector.
7256         (aarch64_simd_imm_zero, aarch64_simd_imm_minus_one): Move earlier
7257         in file.  Use CONST0_RTX and CONSTM1_RTX.
7258         (aarch64_simd_or_scalar_imm_zero): Likewise.  Add match_codes.
7259         (aarch64_simd_reg_or_zero): Accept const as well as const_vector.
7260         Use aarch64_simd_imm_zero.
7261         * config/aarch64/aarch64-sve.md: New file.
7262         * config/aarch64/aarch64.md: Include it.
7263         (VG_REGNUM, P0_REGNUM, P7_REGNUM, P15_REGNUM): New register numbers.
7264         (UNSPEC_REV, UNSPEC_LD1_SVE, UNSPEC_ST1_SVE, UNSPEC_MERGE_PTRUE)
7265         (UNSPEC_PTEST_PTRUE, UNSPEC_UNPACKSHI, UNSPEC_UNPACKUHI)
7266         (UNSPEC_UNPACKSLO, UNSPEC_UNPACKULO, UNSPEC_PACK)
7267         (UNSPEC_FLOAT_CONVERT, UNSPEC_WHILE_LO): New unspec constants.
7268         (sve): New attribute.
7269         (enabled): Disable instructions with the sve attribute unless
7270         TARGET_SVE.
7271         (movqi, movhi): Pass CONST_POLY_INT operaneds through
7272         aarch64_expand_mov_immediate.
7273         (*mov<mode>_aarch64, *movsi_aarch64, *movdi_aarch64): Handle
7274         CNT[BHSD] immediates.
7275         (movti): Split CONST_POLY_INT moves into two halves.
7276         (add<mode>3): Accept aarch64_pluslong_or_poly_operand.
7277         Split additions that need a temporary here if the destination
7278         is the stack pointer.
7279         (*add<mode>3_aarch64): Handle ADDVL and ADDPL immediates.
7280         (*add<mode>3_poly_1): New instruction.
7281         (set_clobber_cc): New expander.
7282
7283 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
7284
7285         * simplify-rtx.c (simplify_immed_subreg): Add an inner_bytes
7286         parameter and use it instead of GET_MODE_SIZE (innermode).  Use
7287         inner_bytes * BITS_PER_UNIT instead of GET_MODE_BITSIZE (innermode).
7288         Use CEIL (inner_bytes, GET_MODE_UNIT_SIZE (innermode)) instead of
7289         GET_MODE_NUNITS (innermode).  Also add a first_elem parameter.
7290         Change innermode from fixed_mode_size to machine_mode.
7291         (simplify_subreg): Update call accordingly.  Handle a constant-sized
7292         subreg of a variable-length CONST_VECTOR.
7293
7294 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
7295             Alan Hayward  <alan.hayward@arm.com>
7296             David Sherwood  <david.sherwood@arm.com>
7297
7298         * tree-ssa-address.c (mem_ref_valid_without_offset_p): New function.
7299         (add_offset_to_base): New function, split out from...
7300         (create_mem_ref): ...here.  When handling a scale other than 1,
7301         check first whether the address is valid without the offset.
7302         Add it into the base if so, leaving the index and scale as-is.
7303
7304 2018-01-12  Jakub Jelinek  <jakub@redhat.com>
7305
7306         PR c++/83778
7307         * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Call
7308         fold_for_warn before checking if arg2 is INTEGER_CST.
7309
7310 2018-01-12  Segher Boessenkool  <segher@kernel.crashing.org>
7311
7312         * config/rs6000/predicates.md (load_multiple_operation): Delete.
7313         (store_multiple_operation): Delete.
7314         * config/rs6000/rs6000-cpus.def (601): Remove MASK_STRING.
7315         * config/rs6000/rs6000-protos.h (rs6000_output_load_multiple): Delete.
7316         * config/rs6000/rs6000-string.c (expand_block_move): Delete everything
7317         guarded by TARGET_STRING.
7318         (rs6000_output_load_multiple): Delete.
7319         * config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
7320         OPTION_MASK_STRING / TARGET_STRING handling.
7321         (print_operand) <'N', 'O'>: Add comment that these are unused now.
7322         (const rs6000_opt_masks) <"string">: Change mask to 0.
7323         * config/rs6000/rs6000.h (TARGET_DEFAULT): Remove MASK_STRING.
7324         (MASK_STRING): Delete.
7325         * config/rs6000/rs6000.md (*mov<mode>_string): Delete TARGET_STRING
7326         parts.  Simplify.
7327         (load_multiple): Delete.
7328         (*ldmsi8): Delete.
7329         (*ldmsi7): Delete.
7330         (*ldmsi6): Delete.
7331         (*ldmsi5): Delete.
7332         (*ldmsi4): Delete.
7333         (*ldmsi3): Delete.
7334         (store_multiple): Delete.
7335         (*stmsi8): Delete.
7336         (*stmsi7): Delete.
7337         (*stmsi6): Delete.
7338         (*stmsi5): Delete.
7339         (*stmsi4): Delete.
7340         (*stmsi3): Delete.
7341         (movmemsi_8reg): Delete.
7342         (corresponding unnamed define_insn): Delete.
7343         (movmemsi_6reg): Delete.
7344         (corresponding unnamed define_insn): Delete.
7345         (movmemsi_4reg): Delete.
7346         (corresponding unnamed define_insn): Delete.
7347         (movmemsi_2reg): Delete.
7348         (corresponding unnamed define_insn): Delete.
7349         (movmemsi_1reg): Delete.
7350         (corresponding unnamed define_insn): Delete.
7351         * config/rs6000/rs6000.opt (mno-string): New.
7352         (mstring): Replace by deprecation warning stub.
7353         * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mstring.
7354
7355 2018-01-12  Jakub Jelinek  <jakub@redhat.com>
7356
7357         * regrename.c (regrename_do_replace): If replacing the same
7358         reg multiple times, try to reuse last created gen_raw_REG.
7359
7360         PR debug/81155
7361         * bb-reorder.c (pass_partition_blocks::gate): In lto don't partition
7362         main to workaround a bug in GDB.
7363
7364 2018-01-12  Tom de Vries  <tom@codesourcery.com>
7365
7366         PR target/83737
7367         * config.gcc (nvptx*-*-*): Set use_gcc_stdint=wrap.
7368
7369 2018-01-12  Vladimir Makarov  <vmakarov@redhat.com>
7370
7371         PR rtl-optimization/80481
7372         * ira-color.c (get_cap_member): New function.
7373         (allocnos_conflict_by_live_ranges_p): Use it.
7374         (slot_coalesced_allocno_live_ranges_intersect_p): Add assert.
7375         (setup_slot_coalesced_allocno_live_ranges): Ditto.
7376
7377 2018-01-12  Uros Bizjak  <ubizjak@gmail.com>
7378
7379         PR target/83628
7380         * config/alpha/alpha.md (*saddsi_1): New insn_ans_split pattern.
7381         (*saddl_se_1): Ditto.
7382         (*ssubsi_1): Ditto.
7383         (*ssubl_se_1): Ditto.
7384
7385 2018-01-12  Richard Sandiford  <richard.sandiford@linaro.org>
7386
7387         * tree-predcom.c (aff_combination_dr_offset): Use wi::to_poly_widest
7388         rather than wi::to_widest for DR_INITs.
7389         * tree-vect-data-refs.c (vect_find_same_alignment_drs): Use
7390         wi::to_poly_offset rather than wi::to_offset for DR_INIT.
7391         (vect_analyze_data_ref_accesses): Require both DR_INITs to be
7392         INTEGER_CSTs.
7393         (vect_analyze_group_access_1): Note that here.
7394
7395 2018-01-12  Richard Sandiford  <richard.sandiford@linaro.org>
7396
7397         * tree-vectorizer.c (get_vec_alignment_for_array_type): Handle
7398         polynomial type sizes.
7399
7400 2018-01-12  Richard Sandiford  <richard.sandiford@linaro.org>
7401
7402         * gimplify.c (gimple_add_tmp_var_fn): Allow variables to have a
7403         poly_uint64 size, rather than requiring an unsigned HOST_WIDE_INT size.
7404         (gimple_add_tmp_var): Likewise.
7405
7406 2018-01-12  Martin Liska  <mliska@suse.cz>
7407
7408         * gimple.c (gimple_alloc_counts): Use uint64_t instead of int.
7409         (gimple_alloc_sizes): Likewise.
7410         (dump_gimple_statistics): Use PRIu64 in printf format.
7411         * gimple.h: Change uint64_t to int.
7412
7413 2018-01-12  Martin Liska  <mliska@suse.cz>
7414
7415         * tree-core.h: Use uint64_t instead of int.
7416         * tree.c (tree_node_counts): Likewise.
7417         (tree_node_sizes): Likewise.
7418         (dump_tree_statistics): Use PRIu64 in printf format.
7419
7420 2018-01-12  Martin Liska  <mliska@suse.cz>
7421
7422         * Makefile.in: As qsort_chk is implemented in vec.c, add
7423         vec.o to linkage of gencfn-macros.
7424         * tree.c (build_new_poly_int_cst): Add CXX_MEM_STAT_INFO as it's
7425         passing the info to record_node_allocation_statistics.
7426         (test_vector_cst_patterns): Add CXX_MEM_STAT_INFO to declaration
7427         and pass the info.
7428         * ggc-common.c (struct ggc_usage): Add operator== and use
7429         it in operator< and compare function.
7430         * mem-stats.h (struct mem_usage): Likewise.
7431         * vec.c (struct vec_usage): Remove operator< and compare
7432         function. Can be simply inherited.
7433
7434 2018-01-12  Martin Jambor  <mjambor@suse.cz>
7435
7436         PR target/81616
7437         * params.def: New parameter PARAM_AVOID_FMA_MAX_BITS.
7438         * tree-ssa-math-opts.c: Include domwalk.h.
7439         (convert_mult_to_fma_1): New function.
7440         (fma_transformation_info): New type.
7441         (fma_deferring_state): Likewise.
7442         (cancel_fma_deferring): New function.
7443         (result_of_phi): Likewise.
7444         (last_fma_candidate_feeds_initial_phi): Likewise.
7445         (convert_mult_to_fma): Added deferring logic, split actual
7446         transformation to convert_mult_to_fma_1.
7447         (math_opts_dom_walker): New type.
7448         (math_opts_dom_walker::after_dom_children): New method, body moved
7449         here from pass_optimize_widening_mul::execute, added deferring logic
7450         bits.
7451         (pass_optimize_widening_mul::execute): Moved most of code to
7452         math_opts_dom_walker::after_dom_children.
7453         * config/i386/x86-tune.def (X86_TUNE_AVOID_128FMA_CHAINS): New.
7454         * config/i386/i386.c (ix86_option_override_internal): Added
7455         maybe_setting of PARAM_AVOID_FMA_MAX_BITS.
7456
7457 2018-01-12  Richard Biener  <rguenther@suse.de>
7458
7459         PR debug/83157
7460         * dwarf2out.c (gen_variable_die): Do not reset old_die for
7461         inline instance vars.
7462
7463 2018-01-12  Oleg Endo  <olegendo@gcc.gnu.org>
7464
7465         PR target/81819
7466         * config/rx/rx.c (rx_is_restricted_memory_address):
7467         Handle SUBREG case.
7468
7469 2018-01-12  Richard Biener  <rguenther@suse.de>
7470
7471         PR tree-optimization/80846
7472         * target.def (split_reduction): New target hook.
7473         * targhooks.c (default_split_reduction): New function.
7474         * targhooks.h (default_split_reduction): Declare.
7475         * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
7476         target requests first reduce vectors by combining low and high
7477         parts.
7478         * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
7479         (get_vectype_for_scalar_type_and_size): Export.
7480         * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
7481         * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
7482         * doc/tm.texi: Regenerate.
7483         * config/i386/i386.c (ix86_split_reduction): Implement
7484         TARGET_VECTORIZE_SPLIT_REDUCTION.
7485
7486 2018-01-12  Eric Botcazou  <ebotcazou@adacore.com>
7487
7488         PR target/83368
7489         * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
7490         in PIC mode except for TARGET_VXWORKS_RTP.
7491         * config/sparc/sparc.c: Include cfgrtl.h.
7492         (TARGET_INIT_PIC_REG): Define.
7493         (TARGET_USE_PSEUDO_PIC_REG): Likewise.
7494         (sparc_pic_register_p): New predicate.
7495         (sparc_legitimate_address_p): Use it.
7496         (sparc_legitimize_pic_address): Likewise.
7497         (sparc_delegitimize_address): Likewise.
7498         (sparc_mode_dependent_address_p): Likewise.
7499         (gen_load_pcrel_sym): Remove 4th parameter.
7500         (load_got_register): Adjust call to above.  Remove obsolete stuff.
7501         (sparc_expand_prologue): Do not call load_got_register here.
7502         (sparc_flat_expand_prologue): Likewise.
7503         (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
7504         (sparc_use_pseudo_pic_reg): New function.
7505         (sparc_init_pic_reg): Likewise.
7506         * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
7507         (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
7508
7509 2018-01-12  Christophe Lyon  <christophe.lyon@linaro.org>
7510
7511         * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
7512         Add item for branch_cost.
7513
7514 2018-01-12  Eric Botcazou  <ebotcazou@adacore.com>
7515
7516         PR rtl-optimization/83565
7517         * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
7518         not extend the result to a larger mode for rotate operations.
7519         (num_sign_bit_copies1): Likewise.
7520
7521 2018-01-12  Rainer Orth  <ro@CeBiTec.Uni-Bielefeld.DE>
7522
7523         PR target/40411
7524         * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
7525         -symbolic.
7526         Use values-Xc.o for -pedantic.
7527         Link with values-xpg4.o for C90, values-xpg6.o otherwise.
7528
7529 2018-01-12  Martin Liska  <mliska@suse.cz>
7530
7531         PR ipa/83054
7532         * ipa-devirt.c (final_warning_record::grow_type_warnings):
7533         New function.
7534         (possible_polymorphic_call_targets): Use it.
7535         (ipa_devirt): Likewise.
7536
7537 2018-01-12  Martin Liska  <mliska@suse.cz>
7538
7539         * profile-count.h (enum profile_quality): Use 0 as invalid
7540         enum value of profile_quality.
7541
7542 2018-01-12  Chung-Ju Wu  <jasonwucj@gmail.com>
7543
7544         * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
7545         -mext-string options.
7546
7547 2018-01-12  Richard Biener  <rguenther@suse.de>
7548
7549         * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
7550         DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
7551         * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
7552         Likewise.
7553         * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
7554
7555 2018-01-11  Michael Meissner  <meissner@linux.vnet.ibm.com>
7556
7557         * configure.ac (--with-long-double-format): Add support for the
7558         configuration option to change the default long double format on
7559         PowerPC systems.
7560         * config.gcc (powerpc*-linux*-*): Likewise.
7561         * configure: Regenerate.
7562         * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
7563         double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
7564         used without modification.
7565
7566 2018-01-11  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
7567
7568         * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
7569         (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
7570         * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
7571         MISC_BUILTIN_SPEC_BARRIER.
7572         (rs6000_init_builtins): Likewise.
7573         * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
7574         enum value.
7575         (speculation_barrier): New define_insn.
7576         * doc/extend.texi: Document __builtin_speculation_barrier.
7577
7578 2018-01-11  Jakub Jelinek  <jakub@redhat.com>
7579
7580         PR target/83203
7581         * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
7582         is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
7583         * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
7584         iterators.
7585         (ssescalarmodesuffix): Add 512-bit vectors.  Use "d" or "q" for
7586         integral modes instead of "ss" and "sd".
7587         (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
7588         vectors with 32-bit and 64-bit elements.
7589         (vecdupssescalarmodesuffix): New mode attribute.
7590         (vec_dup<mode>): Use it.
7591
7592 2018-01-11  H.J. Lu  <hongjiu.lu@intel.com>
7593
7594         PR target/83330
7595         * config/i386/i386.c (ix86_compute_frame_layout): Align stack
7596         frame if argument is passed on stack.
7597
7598 2018-01-11  Jakub Jelinek  <jakub@redhat.com>
7599
7600         PR target/82682
7601         * ree.c (combine_reaching_defs): Optimize also
7602         reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
7603         reg2=any_extend(exp); reg1=reg2;, formatting fix.
7604
7605 2018-01-11  Jan Hubicka  <hubicka@ucw.cz>
7606
7607         PR middle-end/83189
7608         * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
7609
7610 2018-01-11  Jan Hubicka  <hubicka@ucw.cz>
7611
7612         PR middle-end/83718
7613         * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
7614         after they are computed.
7615
7616 2018-01-11  Bin Cheng  <bin.cheng@arm.com>
7617
7618         PR tree-optimization/83695
7619         * gimple-loop-linterchange.cc
7620         (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
7621         reset cached scev information after interchange.
7622         (pass_linterchange::execute): Remove call to scev_reset_htab.
7623
7624 2018-01-11  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
7625
7626         * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
7627         vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
7628         vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
7629         vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
7630         vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
7631         vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
7632         * config/arm/arm_neon_builtins.def (vfmal_lane_low,
7633         vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
7634         vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
7635         vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
7636         vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
7637         * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
7638         (V_lane_reg): Likewise.
7639         * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
7640         New define_expand.
7641         (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
7642         (vfmal_lane_low<mode>_intrinsic,
7643         vfmal_lane_low<vfmlsel2><mode>_intrinsic,
7644         vfmal_lane_high<vfmlsel2><mode>_intrinsic,
7645         vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
7646         vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
7647         vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
7648         vfmsl_lane_high<mode>_intrinsic): New define_insns.
7649
7650 2018-01-11  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
7651
7652         * config/arm/arm-cpus.in (fp16fml): New feature.
7653         (ALL_SIMD): Add fp16fml.
7654         (armv8.2-a): Add fp16fml as an option.
7655         (armv8.3-a): Likewise.
7656         (armv8.4-a): Add fp16fml as part of fp16.
7657         * config/arm/arm.h (TARGET_FP16FML): Define.
7658         * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
7659         when appropriate.
7660         * config/arm/arm-modes.def (V2HF): Define.
7661         * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
7662         vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
7663         vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
7664         * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
7665         vfmsl_low, vfmsl_high): New set of builtins.
7666         * config/arm/iterators.md (PLUSMINUS): New code iterator.
7667         (vfml_op): New code attribute.
7668         (VFMLHALVES): New int iterator.
7669         (VFML, VFMLSEL): New mode attributes.
7670         (V_reg): Define mapping for V2HF.
7671         (V_hi, V_lo): New mode attributes.
7672         (VF_constraint): Likewise.
7673         (vfml_half, vfml_half_selector): New int attributes.
7674         * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
7675         define_expand.
7676         (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
7677         vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
7678         New define_insn.
7679         * config/arm/t-arm-elf (v8_fps): Add fp16fml.
7680         * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
7681         * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
7682         * doc/invoke.texi (ARM Options): Document fp16fml.  Update armv8.4-a
7683         documentation.
7684         * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
7685         Document new effective target and option set.
7686
7687 2018-01-11  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
7688
7689         * config/arm/arm-cpus.in (armv8_4): New feature.
7690         (ARMv8_4a): New fgroup.
7691         (armv8.4-a): New arch.
7692         * config/arm/arm-tables.opt: Regenerate.
7693         * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
7694         * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
7695         * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
7696         Add matching rules for -march=armv8.4-a and extensions.
7697         * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
7698
7699 2018-01-11  Oleg Endo  <olegendo@gcc.gnu.org>
7700
7701         PR target/81821
7702         * config/rx/rx.md (BW): New mode attribute.
7703         (sync_lock_test_and_setsi): Add mode suffix to insn output.
7704
7705 2018-01-11  Richard Biener  <rguenther@suse.de>
7706
7707         PR tree-optimization/83435
7708         * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
7709         * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
7710         * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
7711
7712 2018-01-11  Richard Sandiford  <richard.sandiford@linaro.org>
7713             Alan Hayward  <alan.hayward@arm.com>
7714             David Sherwood  <david.sherwood@arm.com>
7715
7716         * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
7717         field.
7718         (aarch64_classify_address): Initialize it.  Track polynomial offsets.
7719         (aarch64_print_address_internal): Use it to check for a zero offset.
7720
7721 2018-01-11  Richard Sandiford  <richard.sandiford@linaro.org>
7722             Alan Hayward  <alan.hayward@arm.com>
7723             David Sherwood  <david.sherwood@arm.com>
7724
7725         * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
7726         * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
7727         Return a poly_int64 rather than a HOST_WIDE_INT.
7728         (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
7729         rather than a HOST_WIDE_INT.
7730         * config/aarch64/aarch64.h (aarch64_frame): Protect with
7731         HAVE_POLY_INT_H rather than HOST_WIDE_INT.  Change locals_offset,
7732         hard_fp_offset, frame_size, initial_adjust, callee_offset and
7733         final_offset from HOST_WIDE_INT to poly_int64.
7734         * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
7735         to_constant when getting the number of units in an Advanced SIMD
7736         mode.
7737         (aarch64_builtin_vectorized_function): Check for a constant number
7738         of units.
7739         * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
7740         GET_MODE_SIZE.
7741         (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
7742         attribute instead of GET_MODE_NUNITS.
7743         * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
7744         (aarch64_class_max_nregs): Use the constant_lowest_bound of the
7745         GET_MODE_SIZE for fixed-size registers.
7746         (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
7747         (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
7748         (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
7749         (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
7750         (aarch64_print_operand, aarch64_print_address_internal)
7751         (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
7752         (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
7753         (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
7754         Handle polynomial GET_MODE_SIZE.
7755         (aarch64_hard_regno_caller_save_mode): Likewise.  Return modes
7756         wider than SImode without modification.
7757         (tls_symbolic_operand_type): Use strip_offset instead of split_const.
7758         (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
7759         (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
7760         passing and returning SVE modes.
7761         (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
7762         rather than GEN_INT.
7763         (aarch64_emit_probe_stack_range): Take the size as a poly_int64
7764         rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
7765         (aarch64_allocate_and_probe_stack_space): Likewise.
7766         (aarch64_layout_frame): Cope with polynomial offsets.
7767         (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
7768         start_offset as a poly_int64 rather than a HOST_WIDE_INT.  Track
7769         polynomial offsets.
7770         (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
7771         (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
7772         poly_int64 rather than a HOST_WIDE_INT.
7773         (aarch64_get_separate_components, aarch64_process_components)
7774         (aarch64_expand_prologue, aarch64_expand_epilogue)
7775         (aarch64_use_return_insn_p): Handle polynomial frame offsets.
7776         (aarch64_anchor_offset): New function, split out from...
7777         (aarch64_legitimize_address): ...here.
7778         (aarch64_builtin_vectorization_cost): Handle polynomial
7779         TYPE_VECTOR_SUBPARTS.
7780         (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
7781         GET_MODE_NUNITS.
7782         (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
7783         number of elements from the PARALLEL rather than the mode.
7784         (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
7785         rather than GET_MODE_BITSIZE.
7786         (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
7787         (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
7788         (aarch64_expand_vec_perm_const_1): Handle polynomial
7789         d->perm.length () and d->perm elements.
7790         (aarch64_evpc_tbl): Likewise.  Use nelt rather than GET_MODE_NUNITS.
7791         Apply to_constant to d->perm elements.
7792         (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
7793         polynomial CONST_VECTOR_NUNITS.
7794         (aarch64_move_pointer): Take amount as a poly_int64 rather
7795         than an int.
7796         (aarch64_progress_pointer): Avoid temporary variable.
7797         * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
7798         the mode attribute instead of GET_MODE.
7799
7800 2018-01-11  Richard Sandiford  <richard.sandiford@linaro.org>
7801             Alan Hayward  <alan.hayward@arm.com>
7802             David Sherwood  <david.sherwood@arm.com>
7803
7804         * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
7805         x exists before using it.
7806         (aarch64_add_constant_internal): Rename to...
7807         (aarch64_add_offset_1): ...this.  Replace regnum with separate
7808         src and dest rtxes.  Handle the case in which they're different,
7809         including when the offset is zero.  Replace scratchreg with an rtx.
7810         Use 2 additions if there is no spare register into which we can
7811         move a 16-bit constant.
7812         (aarch64_add_constant): Delete.
7813         (aarch64_add_offset): Replace reg with separate src and dest
7814         rtxes.  Take a poly_int64 offset instead of a HOST_WIDE_INT.
7815         Use aarch64_add_offset_1.
7816         (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
7817         an rtx rather than an int.  Take the delta as a poly_int64
7818         rather than a HOST_WIDE_INT.  Use aarch64_add_offset.
7819         (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
7820         (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
7821         aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
7822         (aarch64_expand_epilogue): Update calls to aarch64_add_offset
7823         and aarch64_add_sp.
7824         (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
7825         aarch64_add_constant.
7826
7827 2018-01-11  Richard Sandiford  <richard.sandiford@linaro.org>
7828
7829         * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
7830         Use scalar_float_mode.
7831
7832 2018-01-11  Richard Sandiford  <richard.sandiford@linaro.org>
7833
7834         * config/aarch64/aarch64-simd.md
7835         (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
7836         (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
7837         (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
7838         (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
7839         (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
7840         (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
7841         (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
7842         (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
7843         (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
7844         (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
7845
7846 2018-01-11  Prathamesh Kulkarni  <prathamesh.kulkarni@linaro.org>
7847
7848         PR target/83514
7849         * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
7850         targ_options->x_arm_arch_string is non NULL.
7851
7852 2018-01-11  Tamar Christina  <tamar.christina@arm.com>
7853
7854         * config/aarch64/aarch64.h
7855         (AARCH64_FL_FOR_ARCH8_4): Add  AARCH64_FL_DOTPROD.
7856
7857 2018-01-11  Sudakshina Das  <sudi.das@arm.com>
7858
7859         PR target/82096
7860         * expmed.c (emit_store_flag_force): Swap if const op0
7861         and change VOIDmode to mode of op0.
7862
7863 2018-01-11  Richard Sandiford  <richard.sandiford@linaro.org>
7864
7865         PR rtl-optimization/83761
7866         * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
7867         than bytes to mode_for_size.
7868
7869 2018-01-10  Jan Hubicka  <hubicka@ucw.cz>
7870
7871         PR middle-end/83189
7872         * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
7873         * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
7874         profile.
7875
7876 2018-01-10  Jan Hubicka  <hubicka@ucw.cz>
7877
7878         PR middle-end/83575
7879         * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
7880         when in layout mode.
7881         (cfg_layout_finalize): Do not verify cfg before we are out of layout.
7882         * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
7883         partition fixup.
7884
7885 2018-01-10  Michael Collison  <michael.collison@arm.com>
7886
7887         * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
7888         * config/aarch64/aarch64-option-extension.def: Add
7889         AARCH64_OPT_EXTENSION of 'fp16fml'.
7890         * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
7891         (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
7892         * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
7893         * config/aarch64/constraints.md (Ui7): New constraint.
7894         * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
7895         (VFMLA_SEL_W): Ditto.
7896         (f16quad): Ditto.
7897         (f16mac1): Ditto.
7898         (VFMLA16_LOW): New int iterator.
7899         (VFMLA16_HIGH): Ditto.
7900         (UNSPEC_FMLAL): New unspec.
7901         (UNSPEC_FMLSL): Ditto.
7902         (UNSPEC_FMLAL2): Ditto.
7903         (UNSPEC_FMLSL2): Ditto.
7904         (f16mac): New code attribute.
7905         * config/aarch64/aarch64-simd-builtins.def
7906         (aarch64_fmlal_lowv2sf): Ditto.
7907         (aarch64_fmlsl_lowv2sf): Ditto.
7908         (aarch64_fmlalq_lowv4sf): Ditto.
7909         (aarch64_fmlslq_lowv4sf): Ditto.
7910         (aarch64_fmlal_highv2sf): Ditto.
7911         (aarch64_fmlsl_highv2sf): Ditto.
7912         (aarch64_fmlalq_highv4sf): Ditto.
7913         (aarch64_fmlslq_highv4sf): Ditto.
7914         (aarch64_fmlal_lane_lowv2sf): Ditto.
7915         (aarch64_fmlsl_lane_lowv2sf): Ditto.
7916         (aarch64_fmlal_laneq_lowv2sf): Ditto.
7917         (aarch64_fmlsl_laneq_lowv2sf): Ditto.
7918         (aarch64_fmlalq_lane_lowv4sf): Ditto.
7919         (aarch64_fmlsl_lane_lowv4sf): Ditto.
7920         (aarch64_fmlalq_laneq_lowv4sf): Ditto.
7921         (aarch64_fmlsl_laneq_lowv4sf): Ditto.
7922         (aarch64_fmlal_lane_highv2sf): Ditto.
7923         (aarch64_fmlsl_lane_highv2sf): Ditto.
7924         (aarch64_fmlal_laneq_highv2sf): Ditto.
7925         (aarch64_fmlsl_laneq_highv2sf): Ditto.
7926         (aarch64_fmlalq_lane_highv4sf): Ditto.
7927         (aarch64_fmlsl_lane_highv4sf): Ditto.
7928         (aarch64_fmlalq_laneq_highv4sf): Ditto.
7929         (aarch64_fmlsl_laneq_highv4sf): Ditto.
7930         * config/aarch64/aarch64-simd.md:
7931         (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
7932         (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
7933         (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
7934         (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
7935         (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
7936         (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
7937         (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
7938         (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
7939         (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
7940         (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
7941         (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
7942         (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
7943         (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
7944         (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
7945         (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
7946         (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
7947         (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
7948         (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
7949         (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
7950         (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
7951         * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
7952         (vfmlsl_low_u32): Ditto.
7953         (vfmlalq_low_u32): Ditto.
7954         (vfmlslq_low_u32): Ditto.
7955         (vfmlal_high_u32): Ditto.
7956         (vfmlsl_high_u32): Ditto.
7957         (vfmlalq_high_u32): Ditto.
7958         (vfmlslq_high_u32): Ditto.
7959         (vfmlal_lane_low_u32): Ditto.
7960         (vfmlsl_lane_low_u32): Ditto.
7961         (vfmlal_laneq_low_u32): Ditto.
7962         (vfmlsl_laneq_low_u32): Ditto.
7963         (vfmlalq_lane_low_u32): Ditto.
7964         (vfmlslq_lane_low_u32): Ditto.
7965         (vfmlalq_laneq_low_u32): Ditto.
7966         (vfmlslq_laneq_low_u32): Ditto.
7967         (vfmlal_lane_high_u32): Ditto.
7968         (vfmlsl_lane_high_u32): Ditto.
7969         (vfmlal_laneq_high_u32): Ditto.
7970         (vfmlsl_laneq_high_u32): Ditto.
7971         (vfmlalq_lane_high_u32): Ditto.
7972         (vfmlslq_lane_high_u32): Ditto.
7973         (vfmlalq_laneq_high_u32): Ditto.
7974         (vfmlslq_laneq_high_u32): Ditto.
7975         * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
7976         (AARCH64_FL_FOR_ARCH8_4): New.
7977         (AARCH64_ISA_F16FML): New ISA flag.
7978         (TARGET_F16FML): New feature flag for fp16fml.
7979         (doc/invoke.texi): Document new fp16fml option.
7980
7981 2018-01-10  Michael Collison  <michael.collison@arm.com>
7982
7983         * config/aarch64/aarch64-builtins.c:
7984         (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
7985         * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
7986         (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
7987         * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
7988         (AARCH64_ISA_SHA3): New ISA flag.
7989         (TARGET_SHA3): New feature flag for sha3.
7990         * config/aarch64/iterators.md (sha512_op): New int attribute.
7991         (CRYPTO_SHA512): New int iterator.
7992         (UNSPEC_SHA512H): New unspec.
7993         (UNSPEC_SHA512H2): Ditto.
7994         (UNSPEC_SHA512SU0): Ditto.
7995         (UNSPEC_SHA512SU1): Ditto.
7996         * config/aarch64/aarch64-simd-builtins.def
7997         (aarch64_crypto_sha512hqv2di): New builtin.
7998         (aarch64_crypto_sha512h2qv2di): Ditto.
7999         (aarch64_crypto_sha512su0qv2di): Ditto.
8000         (aarch64_crypto_sha512su1qv2di): Ditto.
8001         (aarch64_eor3qv8hi): Ditto.
8002         (aarch64_rax1qv2di): Ditto.
8003         (aarch64_xarqv2di): Ditto.
8004         (aarch64_bcaxqv8hi): Ditto.
8005         * config/aarch64/aarch64-simd.md:
8006         (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
8007         (aarch64_crypto_sha512su0qv2di): Ditto.
8008         (aarch64_crypto_sha512su1qv2di): Ditto.
8009         (aarch64_eor3qv8hi): Ditto.
8010         (aarch64_rax1qv2di): Ditto.
8011         (aarch64_xarqv2di): Ditto.
8012         (aarch64_bcaxqv8hi): Ditto.
8013         * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
8014         (vsha512h2q_u64): Ditto.
8015         (vsha512su0q_u64): Ditto.
8016         (vsha512su1q_u64): Ditto.
8017         (veor3q_u16): Ditto.
8018         (vrax1q_u64): Ditto.
8019         (vxarq_u64): Ditto.
8020         (vbcaxq_u16): Ditto.
8021         * config/arm/types.md (crypto_sha512): New type attribute.
8022         (crypto_sha3): Ditto.
8023         (doc/invoke.texi): Document new sha3 option.
8024
8025 2018-01-10  Michael Collison  <michael.collison@arm.com>
8026
8027         * config/aarch64/aarch64-builtins.c:
8028         (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
8029         * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
8030         (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
8031         (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
8032         * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
8033         (AARCH64_ISA_SM4): New ISA flag.
8034         (TARGET_SM4): New feature flag for sm4.
8035         * config/aarch64/aarch64-simd-builtins.def
8036         (aarch64_sm3ss1qv4si): Ditto.
8037         (aarch64_sm3tt1aq4si): Ditto.
8038         (aarch64_sm3tt1bq4si): Ditto.
8039         (aarch64_sm3tt2aq4si): Ditto.
8040         (aarch64_sm3tt2bq4si): Ditto.
8041         (aarch64_sm3partw1qv4si): Ditto.
8042         (aarch64_sm3partw2qv4si): Ditto.
8043         (aarch64_sm4eqv4si): Ditto.
8044         (aarch64_sm4ekeyqv4si): Ditto.
8045         * config/aarch64/aarch64-simd.md:
8046         (aarch64_sm3ss1qv4si): Ditto.
8047         (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
8048         (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
8049         (aarch64_sm4eqv4si): Ditto.
8050         (aarch64_sm4ekeyqv4si): Ditto.
8051         * config/aarch64/iterators.md (sm3tt_op): New int iterator.
8052         (sm3part_op): Ditto.
8053         (CRYPTO_SM3TT): Ditto.
8054         (CRYPTO_SM3PART): Ditto.
8055         (UNSPEC_SM3SS1): New unspec.
8056         (UNSPEC_SM3TT1A): Ditto.
8057         (UNSPEC_SM3TT1B): Ditto.
8058         (UNSPEC_SM3TT2A): Ditto.
8059         (UNSPEC_SM3TT2B): Ditto.
8060         (UNSPEC_SM3PARTW1): Ditto.
8061         (UNSPEC_SM3PARTW2): Ditto.
8062         (UNSPEC_SM4E): Ditto.
8063         (UNSPEC_SM4EKEY): Ditto.
8064         * config/aarch64/constraints.md (Ui2): New constraint.
8065         * config/aarch64/predicates.md (aarch64_imm2): New predicate.
8066         * config/arm/types.md (crypto_sm3): New type attribute.
8067         (crypto_sm4): Ditto.
8068         * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
8069         (vsm3tt1aq_u32): Ditto.
8070         (vsm3tt1bq_u32): Ditto.
8071         (vsm3tt2aq_u32): Ditto.
8072         (vsm3tt2bq_u32): Ditto.
8073         (vsm3partw1q_u32): Ditto.
8074         (vsm3partw2q_u32): Ditto.
8075         (vsm4eq_u32): Ditto.
8076         (vsm4ekeyq_u32): Ditto.
8077         (doc/invoke.texi): Document new sm4 option.
8078
8079 2018-01-10  Michael Collison  <michael.collison@arm.com>
8080
8081         * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
8082         * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
8083         (AARCH64_FL_FOR_ARCH8_4): New.
8084         (AARCH64_FL_V8_4): New flag.
8085         (doc/invoke.texi): Document new armv8.4-a option.
8086
8087 2018-01-10  Michael Collison  <michael.collison@arm.com>
8088
8089         * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
8090         (__ARM_FEATURE_AES): Define if TARGET_AES is true.
8091         (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
8092         * config/aarch64/aarch64-option-extension.def: Add
8093         AARCH64_OPT_EXTENSION of 'sha2'.
8094         (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
8095         (crypto): Disable sha2 and aes if crypto disabled.
8096         (crypto): Enable aes and sha2 if enabled.
8097         (simd): Disable sha2 and aes if simd disabled.
8098         * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
8099         New flags.
8100         (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
8101         (TARGET_SHA2): New feature flag for sha2.
8102         (TARGET_AES): New feature flag for aes.
8103         * config/aarch64/aarch64-simd.md:
8104         (aarch64_crypto_aes<aes_op>v16qi): Make pattern
8105         conditional on TARGET_AES.
8106         (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
8107         (aarch64_crypto_sha1hsi): Make pattern conditional
8108         on TARGET_SHA2.
8109         (aarch64_crypto_sha1hv4si): Ditto.
8110         (aarch64_be_crypto_sha1hv4si): Ditto.
8111         (aarch64_crypto_sha1su1v4si): Ditto.
8112         (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
8113         (aarch64_crypto_sha1su0v4si): Ditto.
8114         (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
8115         (aarch64_crypto_sha256su0v4si): Ditto.
8116         (aarch64_crypto_sha256su1v4si): Ditto.
8117         (doc/invoke.texi): Document new aes and sha2 options.
8118
8119 2018-01-10  Martin Sebor  <msebor@redhat.com>
8120
8121         PR tree-optimization/83781
8122         * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
8123         as string arrays.
8124
8125 2018-01-11  Martin Sebor  <msebor@gmail.com>
8126             Prathamesh Kulkarni  <prathamesh.kulkarni@linaro.org>
8127
8128         PR tree-optimization/83501
8129         PR tree-optimization/81703
8130
8131         * tree-ssa-strlen.c (get_string_cst): Rename...
8132         (get_string_len): ...to this.  Handle global constants.
8133         (handle_char_store): Adjust.
8134
8135 2018-01-10  Kito Cheng  <kito.cheng@gmail.com>
8136             Jim Wilson  <jimw@sifive.com>
8137
8138         * config/riscv/riscv-protos.h (riscv_output_return): New.
8139         * config/riscv/riscv.c (struct machine_function): New naked_p field.
8140         (riscv_attribute_table, riscv_output_return),
8141         (riscv_handle_fndecl_attribute, riscv_naked_function_p),
8142         (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
8143         (riscv_compute_frame_info): Only compute frame->mask if not a naked
8144         function.
8145         (riscv_expand_prologue): Add early return for naked function.
8146         (riscv_expand_epilogue): Likewise.
8147         (riscv_function_ok_for_sibcall): Return false for naked function.
8148         (riscv_set_current_function): New.
8149         (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
8150         (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
8151         * config/riscv/riscv.md (simple_return): Call riscv_output_return.
8152         * doc/extend.texi (RISC-V Function Attributes): New.
8153
8154 2018-01-10  Michael Meissner  <meissner@linux.vnet.ibm.com>
8155
8156         * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
8157         check for 128-bit long double before checking TCmode.
8158         * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
8159         128-bit long doubles before checking TFmode or TCmode.
8160         (FLOAT128_IBM_P): Likewise.
8161
8162 2018-01-10  Martin Sebor  <msebor@redhat.com>
8163
8164         PR tree-optimization/83671
8165         * builtins.c (c_strlen): Unconditionally return zero for the empty
8166         string.
8167         Use -Warray-bounds for warnings.
8168         * gimple-fold.c (get_range_strlen): Handle non-constant lengths
8169         for non-constant array indices with COMPONENT_REF, arrays of
8170         arrays, and pointers to arrays.
8171         (gimple_fold_builtin_strlen): Determine and set length range for
8172         non-constant character arrays.
8173
8174 2018-01-10  Aldy Hernandez  <aldyh@redhat.com>
8175
8176         PR middle-end/81897
8177         * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
8178         empty blocks.
8179
8180 2018-01-10  Eric Botcazou  <ebotcazou@adacore.com>
8181
8182         * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
8183
8184 2018-01-10  Peter Bergner  <bergner@vnet.ibm.com>
8185
8186         PR target/83399
8187         * config/rs6000/rs6000.c (print_operand) <'y'>: Use
8188         VECTOR_MEM_ALTIVEC_OR_VSX_P.
8189         * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
8190         indexed_or_indirect_operand predicate.
8191         (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
8192         (*vsx_le_perm_load_v8hi): Likewise.
8193         (*vsx_le_perm_load_v16qi): Likewise.
8194         (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
8195         (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
8196         (*vsx_le_perm_store_v8hi): Likewise.
8197         (*vsx_le_perm_store_v16qi): Likewise.
8198         (eight unnamed splitters): Likewise.
8199
8200 2018-01-10  Peter Bergner  <bergner@vnet.ibm.com>
8201
8202         * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
8203         * config/rs6000/emmintrin.h: Likewise.
8204         * config/rs6000/mmintrin.h: Likewise.
8205         * config/rs6000/xmmintrin.h: Likewise.
8206
8207 2018-01-10  David Malcolm  <dmalcolm@redhat.com>
8208
8209         PR c++/43486
8210         * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
8211         "public_flag".
8212         * tree.c (tree_nop_conversion): Return true for location wrapper
8213         nodes.
8214         (maybe_wrap_with_location): New function.
8215         (selftest::check_strip_nops): New function.
8216         (selftest::test_location_wrappers): New function.
8217         (selftest::tree_c_tests): Call it.
8218         * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
8219         (maybe_wrap_with_location): New decl.
8220         (EXPR_LOCATION_WRAPPER_P): New macro.
8221         (location_wrapper_p): New inline function.
8222         (tree_strip_any_location_wrapper): New inline function.
8223
8224 2018-01-10  H.J. Lu  <hongjiu.lu@intel.com>
8225
8226         PR target/83735
8227         * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
8228         stack_realign_offset for the largest alignment of stack slot
8229         actually used.
8230         (ix86_find_max_used_stack_alignment): New function.
8231         (ix86_finalize_stack_frame_flags): Use it.  Set
8232         max_used_stack_alignment if we don't realign stack.
8233         * config/i386/i386.h (machine_function): Add
8234         max_used_stack_alignment.
8235
8236 2018-01-10  Christophe Lyon  <christophe.lyon@linaro.org>
8237
8238         * config/arm/arm.opt (-mbranch-cost): New option.
8239         * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
8240         account.
8241
8242 2018-01-10  Segher Boessenkool  <segher@kernel.crashing.org>
8243
8244         PR target/83629
8245         * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
8246         load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
8247
8248 2018-01-10  Richard Biener  <rguenther@suse.de>
8249
8250         PR debug/83765
8251         * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
8252         early out so it also covers the case where we have a non-NULL
8253         origin.
8254
8255 2018-01-10  Richard Sandiford  <richard.sandiford@linaro.org>
8256
8257         PR tree-optimization/83753
8258         * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
8259         for non-strided grouped accesses if the number of elements is 1.
8260
8261 2018-01-10  Jan Hubicka  <hubicka@ucw.cz>
8262
8263         PR target/81616
8264         * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
8265         * i386.h (TARGET_USE_GATHER): Define.
8266         * x86-tune.def (X86_TUNE_USE_GATHER): New.
8267
8268 2018-01-10  Martin Liska  <mliska@suse.cz>
8269
8270         PR bootstrap/82831
8271         * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
8272         * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
8273         partitioning.
8274         * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
8275         CLEANUP_NO_PARTITIONING is not set.
8276
8277 2018-01-10  Richard Sandiford  <richard.sandiford@linaro.org>
8278
8279         * doc/rtl.texi: Remove documentation of (const ...) wrappers
8280         for vectors, as a partial revert of r254296.
8281         * rtl.h (const_vec_p): Delete.
8282         (const_vec_duplicate_p): Don't test for vector CONSTs.
8283         (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
8284         * expmed.c (make_tree): Likewise.
8285
8286         Revert:
8287         * common.md (E, F): Use CONSTANT_P instead of checking for
8288         CONST_VECTOR.
8289         * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
8290         checking for CONST_VECTOR.
8291
8292 2018-01-09  Jan Hubicka  <hubicka@ucw.cz>
8293
8294         PR middle-end/83575
8295         * predict.c (force_edge_cold): Handle in more sane way edges
8296         with no prediction.
8297
8298 2018-01-09  Carl Love  <cel@us.ibm.com>
8299
8300         * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
8301         V4SI, V4SF types.
8302         (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
8303         * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
8304         VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
8305         VMRGOW_V2DI, VMRGOW_V2DF.  Remove definition for VMRGOW.
8306         * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
8307         P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW):  Add definitions.
8308         * config/rs6000/rs6000-protos.h: Add extern defition for
8309         rs6000_generate_float2_double_code.
8310         * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
8311         function.
8312         * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
8313         (float2_v2df): Add define_expand.
8314
8315 2018-01-09  Uros Bizjak  <ubizjak@gmail.com>
8316
8317         PR target/83628
8318         * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
8319         op_mode in the force_to_mode call.
8320
8321 2018-01-09  Richard Sandiford  <richard.sandiford@linaro.org>
8322
8323         * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
8324         instead of checking each element individually.
8325         (aarch64_evpc_uzp): Likewise.
8326         (aarch64_evpc_zip): Likewise.
8327         (aarch64_evpc_ext): Likewise.
8328         (aarch64_evpc_rev): Likewise.
8329         (aarch64_evpc_dup): Test the encoding for a single duplicated element,
8330         instead of checking each element individually.  Return true without
8331         generating rtl if
8332         (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
8333         whether all selected elements come from the same input, instead of
8334         checking each element individually.  Remove calls to gen_rtx_REG,
8335         start_sequence and end_sequence and instead assert that no rtl is
8336         generated.
8337
8338 2018-01-09  Richard Sandiford  <richard.sandiford@linaro.org>
8339
8340         * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
8341         order of HIGH and CONST checks.
8342
8343 2018-01-09  Richard Sandiford  <richard.sandiford@linaro.org>
8344
8345         * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
8346         if the destination isn't an SSA_NAME.
8347
8348 2018-01-09  Richard Biener  <rguenther@suse.de>
8349
8350         PR tree-optimization/83668
8351         * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
8352         move prologue...
8353         (canonicalize_loop_form): ... here, renamed from ...
8354         (canonicalize_loop_closed_ssa_form): ... this and amended to
8355         swap successor edges for loop exit blocks to make us use
8356         the RPO order we need for initial schedule generation.
8357
8358 2018-01-09  Joseph Myers  <joseph@codesourcery.com>
8359
8360         PR tree-optimization/64811
8361         * match.pd: When optimizing comparisons with Inf, avoid
8362         introducing or losing exceptions from comparisons with NaN.
8363
8364 2018-01-09  Martin Liska  <mliska@suse.cz>
8365
8366         PR sanitizer/82517
8367         * asan.c (shadow_mem_size): Add gcc_assert.
8368
8369 2018-01-09  Georg-Johann Lay  <avr@gjlay.de>
8370
8371         Don't save registers in main().
8372
8373         PR target/83738
8374         * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
8375         * config/avr/avr.opt (-mmain-is-OS_task): New target option.
8376         * config/avr/avr.c (avr_set_current_function): Don't error if
8377         naked, OS_task or OS_main are specified at the same time.
8378         (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
8379         OS_main.
8380         (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
8381         attribute.
8382         * common/config/avr/avr-common.c (avr_option_optimization_table):
8383         Switch on -mmain-is-OS_task for optimizing compilations.
8384
8385 2018-01-09  Richard Biener  <rguenther@suse.de>
8386
8387         PR tree-optimization/83572
8388         * graphite.c: Include cfganal.h.
8389         (graphite_transform_loops): Connect infinite loops to exit
8390         and remove fake edges at the end.
8391
8392 2018-01-09  Jan Hubicka  <hubicka@ucw.cz>
8393
8394         * ipa-inline.c (edge_badness): Revert accidental checkin.
8395
8396 2018-01-09  Jan Hubicka  <hubicka@ucw.cz>
8397
8398         PR ipa/80763
8399         * ipa-comdats.c (set_comdat_group): Only set comdat group of real
8400         symbols; not inline clones.
8401
8402 2018-01-09  Jakub Jelinek  <jakub@redhat.com>
8403
8404         PR target/83507
8405         * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
8406         hard registers.  Formatting fixes.
8407
8408         PR preprocessor/83722
8409         * gcc.c (try_generate_repro): Pass
8410         &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
8411         &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
8412         do_report_bug.
8413
8414 2018-01-08  Monk Chiang  <sh.chiang04@gmail.com>
8415             Kito Cheng  <kito.cheng@gmail.com>
8416
8417         * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
8418         (riscv_leaf_function_p): Delete.
8419         (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
8420
8421 2018-01-08  Aaron Sawdey  <acsawdey@linux.vnet.ibm.com>
8422
8423         * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
8424         function.
8425         (do_ifelse): New function.
8426         (do_isel): New function.
8427         (do_sub3): New function.
8428         (do_add3): New function.
8429         (do_load_mask_compare): New function.
8430         (do_overlap_load_compare): New function.
8431         (expand_compare_loop): New function.
8432         (expand_block_compare): Call expand_compare_loop() when appropriate.
8433         * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
8434         option description.
8435         (-mblock-compare-inline-loop-limit): New option.
8436
8437 2018-01-08  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
8438
8439         PR target/83677
8440         * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
8441         Reverse order of second and third operands in first alternative.
8442         * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
8443         of first and second elements in UNSPEC_VPERMR vector.
8444         (altivec_expand_vec_perm_le): Likewise.
8445
8446 2018-01-08  Jeff Law  <law@redhat.com>
8447
8448         PR rtl-optimizatin/81308
8449         * tree-switch-conversion.c (cfg_altered): New file scoped static.
8450         (process_switch): If group_case_labels makes a change, then set
8451         cfg_altered.
8452         (pass_convert_switch::execute): If a switch is converted, then
8453         set cfg_altered.  Return TODO_cfg_cleanup if cfg_altered is true.
8454
8455         PR rtl-optimization/81308
8456         * recog.c (split_all_insns): Conditionally cleanup the CFG after
8457         splitting insns.
8458
8459 2018-01-08  Vidya Praveen  <vidyapraveen@arm.com>
8460
8461         PR target/83663 - Revert r255946
8462         * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
8463         generation for cases where splatting a value is not useful.
8464         * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
8465         across a vec_duplicate and a paradoxical subreg forming a vector
8466         mode to a vec_concat.
8467
8468 2018-01-08  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
8469
8470         * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
8471         -march=armv8.3-a variants.
8472         * config/arm/t-multilib: Likewise.
8473         * config/arm/t-arm-elf: Likewise.  Handle dotprod extension.
8474
8475 2018-01-08  Aaron Sawdey  <acsawdey@linux.vnet.ibm.com>
8476
8477         * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
8478         to generate rtl.
8479         (cceq_ior_compare_complement): Give it a name so I can use it, and
8480         change boolean_or_operator predicate to boolean_operator so it can
8481         be used to generate a crand.
8482         (eqne): New code iterator.
8483         (bd/bd_neg): New code_attrs.
8484         (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
8485         a single define_insn.
8486         (<bd>tf_<mode>): A new insn pattern for the conditional form branch
8487         decrement (bdnzt/bdnzf/bdzt/bdzf).
8488         * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
8489         with the new names of the branch decrement patterns, and added the
8490         names of the branch decrement conditional patterns.
8491
8492 2018-01-08  Richard Biener  <rguenther@suse.de>
8493
8494         PR tree-optimization/83563
8495         * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
8496         cache.
8497
8498 2018-01-08  Richard Biener  <rguenther@suse.de>
8499
8500         PR middle-end/83713
8501         * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
8502
8503 2018-01-08  Richard Biener  <rguenther@suse.de>
8504
8505         PR tree-optimization/83685
8506         * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
8507         references to abnormals.
8508
8509 2018-01-08  Richard Biener  <rguenther@suse.de>
8510
8511         PR lto/83719
8512         * dwarf2out.c (output_indirect_strings): Handle empty
8513         skeleton_debug_str_hash.
8514         (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
8515
8516 2018-01-08  Claudiu Zissulescu  <claziss@synopsys.com>
8517
8518         * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
8519         (emit_store_direct): Likewise.
8520         (arc_trampoline_adjust_address): Likewise.
8521         (arc_asm_trampoline_template): New function.
8522         (arc_initialize_trampoline): Use asm_trampoline_template.
8523         (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
8524         * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
8525         * config/arc/arc.md (flush_icache): Delete pattern.
8526
8527 2018-01-08  Claudiu Zissulescu  <claziss@synopsys.com>
8528
8529         * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
8530         * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
8531         munaligned-access.
8532
8533 2018-01-08  Sebastian Huber  <sebastian.huber@embedded-brains.de>
8534
8535         PR target/83681
8536         * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
8537         by not USED_FOR_TARGET.
8538         (make_pass_resolve_sw_modes): Likewise.
8539
8540 2018-01-08  Sebastian Huber  <sebastian.huber@embedded-brains.de>
8541
8542         * config/nios2/nios2.h (nios2_section_threshold): Guard by not
8543         USED_FOR_TARGET.
8544
8545 2018-01-08  Richard Biener  <rguenther@suse.de>
8546
8547         PR middle-end/83580
8548         * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
8549
8550 2018-01-08  Richard Biener  <rguenther@suse.de>
8551
8552         PR middle-end/83517
8553         * match.pd ((t * 2) / 2) -> t): Add missing :c.
8554
8555 2018-01-06  Aldy Hernandez  <aldyh@redhat.com>
8556
8557         PR middle-end/81897
8558         * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
8559         basic blocks with a small number of successors.
8560         (convert_control_dep_chain_into_preds): Improve handling of
8561         forwarder blocks.
8562         (dump_predicates): Split apart into...
8563         (dump_pred_chain): ...here...
8564         (dump_pred_info): ...and here.
8565         (can_one_predicate_be_invalidated_p): Add debugging printfs.
8566         (can_chain_union_be_invalidated_p): Improve check for invalidation
8567         of paths.
8568         (uninit_uses_cannot_happen): Avoid unnecessary if
8569         convert_control_dep_chain_into_preds yielded nothing.
8570
8571 2018-01-06  Martin Sebor  <msebor@redhat.com>
8572
8573         PR tree-optimization/83640
8574         * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
8575         subtracting negative offset from size.
8576         (builtin_access::overlap): Adjust offset bounds of the access to fall
8577         within the size of the object if possible.
8578
8579 2018-01-06  Richard Sandiford  <richard.sandiford@linaro.org>
8580
8581         PR rtl-optimization/83699
8582         * expmed.c (extract_bit_field_1): Restrict the vector usage of
8583         extract_bit_field_as_subreg to cases in which the extracted
8584         value is also a vector.
8585
8586         * lra-constraints.c (process_alt_operands): Test for the equivalence
8587         substitutions when detecting a possible reload cycle.
8588
8589 2018-01-06  Jakub Jelinek  <jakub@redhat.com>
8590
8591         PR debug/83480
8592         * toplev.c (process_options): Don't enable debug_nonbind_markers_p
8593         by default if flag_selective_schedling{,2}.  Formatting fixes.
8594
8595         PR rtl-optimization/83682
8596         * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
8597         if it has non-VECTOR_MODE element mode.
8598         (vec_duplicate_p): Likewise.
8599
8600         PR middle-end/83694
8601         * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
8602         and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
8603
8604 2018-01-05  Jakub Jelinek  <jakub@redhat.com>
8605
8606         PR target/83604
8607         * config/i386/i386-builtin.def
8608         (__builtin_ia32_vgf2p8affineinvqb_v64qi,
8609         __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
8610         Require also OPTION_MASK_ISA_AVX512F in addition to
8611         OPTION_MASK_ISA_GFNI.
8612         (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
8613         __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
8614         OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
8615         to OPTION_MASK_ISA_GFNI.
8616         (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
8617         OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
8618         OPTION_MASK_ISA_AVX512BW.
8619         (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
8620         OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
8621         addition to OPTION_MASK_ISA_GFNI.
8622         (__builtin_ia32_vgf2p8affineinvqb_v16qi,
8623         __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
8624         Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
8625         to OPTION_MASK_ISA_GFNI.
8626         * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
8627         a requirement for all ISAs rather than any of them with a few
8628         exceptions.
8629         (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
8630         processing.
8631         (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
8632         bitmasks to be enabled with 3 exceptions, instead of requiring any
8633         enabled ISA with lots of exceptions.
8634         * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
8635         vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
8636         Change avx512bw in isa attribute to avx512f.
8637         * config/i386/sgxintrin.h: Add license boilerplate.
8638         * config/i386/vaesintrin.h: Likewise.  Fix macro spelling __AVX512F
8639         to __AVX512F__ and __AVX512VL to __AVX512VL__.
8640         (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
8641         _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
8642         defined.
8643         * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
8644         _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
8645         temporarily sse2 rather than sse if not enabled already.
8646
8647         PR target/83604
8648         * config/i386/sse.md (VI248_VLBW): Rename to ...
8649         (VI248_AVX512VL): ... this.  Don't guard V32HI with TARGET_AVX512BW.
8650         (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
8651         vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
8652         vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
8653         vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
8654         mode iterator instead of VI248_VLBW.
8655
8656 2018-01-05  Jan Hubicka  <hubicka@ucw.cz>
8657
8658         * ipa-fnsummary.c (record_modified_bb_info): Add OP.
8659         (record_modified): Skip clobbers; add debug output.
8660         (param_change_prob): Use sreal frequencies.
8661
8662 2018-01-05  Richard Sandiford  <richard.sandiford@linaro.org>
8663
8664         * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
8665         punt for user-aligned variables.
8666
8667 2018-01-05  Richard Sandiford  <richard.sandiford@linaro.org>
8668
8669         * tree-chrec.c (chrec_contains_symbols): Return true for
8670         POLY_INT_CST.
8671
8672 2018-01-05  Sudakshina Das  <sudi.das@arm.com>
8673
8674         PR target/82439
8675         * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
8676         of (x|y) == x for BICS pattern.
8677
8678 2018-01-05  Jakub Jelinek  <jakub@redhat.com>
8679
8680         PR tree-optimization/83605
8681         * gimple-ssa-strength-reduction.c: Include tree-eh.h.
8682         (find_candidates_dom_walker::before_dom_children): Ignore stmts that
8683         can throw.
8684
8685 2018-01-05  Sebastian Huber  <sebastian.huber@embedded-brains.de>
8686
8687         * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
8688         * config/epiphany/rtems.h: New file.
8689
8690 2018-01-04  Jakub Jelinek  <jakub@redhat.com>
8691             Uros Bizjak  <ubizjak@gmail.com>
8692
8693         PR target/83554
8694         * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
8695         QIreg_operand instead of register_operand predicate.
8696         * config/i386/i386.c (ix86_rop_should_change_byte_p,
8697         set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
8698         comments instead of -fmitigate[-_]rop.
8699
8700 2018-01-04  Rainer Orth  <ro@CeBiTec.Uni-Bielefeld.DE>
8701
8702         PR bootstrap/81926
8703         * cgraphunit.c (symbol_table::compile): Switch to text_section
8704         before calling assembly_start debug hook.
8705         * run-rtl-passes.c (run_rtl_passes): Likewise.
8706         Include output.h.
8707
8708 2018-01-04  Richard Sandiford  <richard.sandiford@linaro.org>
8709
8710         * tree-vrp.c (extract_range_from_binary_expr_1): Check
8711         range_int_cst_p rather than !symbolic_range_p before calling
8712         extract_range_from_multiplicative_op_1.
8713
8714 2018-01-04  Jeff Law  <law@redhat.com>
8715
8716         * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
8717         redundant test in assertion.
8718
8719 2018-01-04  Richard Sandiford  <richard.sandiford@linaro.org>
8720
8721         * doc/rtl.texi: Document machine_mode wrapper classes.
8722
8723 2018-01-04  Richard Sandiford  <richard.sandiford@linaro.org>
8724
8725         * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
8726         using tree_to_uhwi.
8727
8728 2018-01-04  Richard Sandiford  <richard.sandiford@linaro.org>
8729
8730         * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
8731         the VEC_PERM_EXPR fold to fail.
8732
8733 2018-01-04  Jakub Jelinek  <jakub@redhat.com>
8734
8735         PR debug/83585
8736         * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
8737         to switched_sections.
8738
8739 2018-01-04  Richard Sandiford  <richard.sandiford@linaro.org>
8740
8741         PR target/83680
8742         * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
8743         test for d.testing.
8744
8745 2018-01-04  Peter Bergner  <bergner@vnet.ibm.com>
8746
8747         PR target/83387
8748         * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
8749         allow arguments in FP registers if TARGET_HARD_FLOAT is false.
8750
8751 2018-01-04  Jakub Jelinek  <jakub@redhat.com>
8752
8753         PR debug/83666
8754         * cfgexpand.c (expand_debug_expr) <case BIT_FIELD_REF>: Punt if mode
8755         is BLKmode and bitpos not zero or mode change is needed.
8756
8757 2018-01-04  Richard Sandiford  <richard.sandiford@linaro.org>
8758
8759         PR target/83675
8760         * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
8761         TARGET_VIS2.
8762
8763 2018-01-04  Uros Bizjak  <ubizjak@gmail.com>
8764
8765         PR target/83628
8766         * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
8767         instead of MULT rtx.  Update all corresponding splitters.
8768         (*saddl_se): Ditto.
8769         (*ssub<modesuffix>): Ditto.
8770         (*ssubl_se): Ditto.
8771         (*cmp_sadd_di): Update split patterns.
8772         (*cmp_sadd_si): Ditto.
8773         (*cmp_sadd_sidi): Ditto.
8774         (*cmp_ssub_di): Ditto.
8775         (*cmp_ssub_si): Ditto.
8776         (*cmp_ssub_sidi): Ditto.
8777         * config/alpha/predicates.md (const23_operand): New predicate.
8778         * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
8779         Look for ASHIFT, not MULT inner operand.
8780         (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
8781
8782 2018-01-04  Martin Liska  <mliska@suse.cz>
8783
8784         PR gcov-profile/83669
8785         * gcov.c (output_intermediate_file): Add version to intermediate
8786         gcov file.
8787         * doc/gcov.texi: Document new field 'version' in intermediate
8788         file format. Fix location of '-k' option of gcov command.
8789
8790 2018-01-04  Martin Liska  <mliska@suse.cz>
8791
8792         PR ipa/82352
8793         * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
8794
8795 2018-01-04  Jakub Jelinek  <jakub@redhat.com>
8796
8797         * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
8798
8799 2018-01-03  Martin Sebor  <msebor@redhat.com>
8800
8801         PR tree-optimization/83655
8802         * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
8803         checking calls with invalid arguments.
8804
8805 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
8806
8807         * tree-vect-stmts.c (vect_get_store_rhs): New function.
8808         (vectorizable_mask_load_store): Delete.
8809         (vectorizable_call): Return false for masked loads and stores.
8810         (vectorizable_store): Handle IFN_MASK_STORE.  Use vect_get_store_rhs
8811         instead of gimple_assign_rhs1.
8812         (vectorizable_load): Handle IFN_MASK_LOAD.
8813         (vect_transform_stmt): Don't set is_store for call_vec_info_type.
8814
8815 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
8816
8817         * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
8818         split out from..,
8819         (vectorizable_mask_load_store): ...here.
8820         (vectorizable_load): ...and here.
8821
8822 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
8823
8824         * tree-vect-stmts.c (vect_build_all_ones_mask)
8825         (vect_build_zero_merge_argument): New functions, split out from...
8826         (vectorizable_load): ...here.
8827
8828 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
8829
8830         * tree-vect-stmts.c (vect_check_store_rhs): New function,
8831         split out from...
8832         (vectorizable_mask_load_store): ...here.
8833         (vectorizable_store): ...and here.
8834
8835 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
8836
8837         * tree-vect-stmts.c (vect_check_load_store_mask): New function,
8838         split out from...
8839         (vectorizable_mask_load_store): ...here.
8840
8841 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
8842
8843         * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
8844         (vect_model_store_cost): Take a vec_load_store_type instead of a
8845         vect_def_type.
8846         * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
8847         (vect_model_store_cost): Take a vec_load_store_type instead of a
8848         vect_def_type.
8849         (vectorizable_mask_load_store): Update accordingly.
8850         (vectorizable_store): Likewise.
8851         * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
8852
8853 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
8854
8855         * tree-vect-loop.c (vect_transform_loop): Stub out scalar
8856         IFN_MASK_LOAD calls here rather than...
8857         * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
8858
8859 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
8860             Alan Hayward  <alan.hayward@arm.com>
8861             David Sherwood  <david.sherwood@arm.com>
8862
8863         * expmed.c (extract_bit_field_1): For vector extracts,
8864         fall back to extract_bit_field_as_subreg if vec_extract
8865         isn't available.
8866
8867 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
8868             Alan Hayward  <alan.hayward@arm.com>
8869             David Sherwood  <david.sherwood@arm.com>
8870
8871         * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
8872         they are variable or constant sized.
8873         (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
8874         slots for constant-sized data.
8875
8876 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
8877             Alan Hayward  <alan.hayward@arm.com>
8878             David Sherwood  <david.sherwood@arm.com>
8879
8880         * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
8881         handling COND_EXPRs with boolean comparisons, try to find a better
8882         basis for the mask type than the boolean itself.
8883
8884 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
8885
8886         * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
8887         is calculated and how it can be overridden.
8888         * genmodes.c (max_bitsize_mode_any_mode): New variable.
8889         (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
8890         if defined.
8891         (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
8892         if nonzero.
8893
8894 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
8895             Alan Hayward  <alan.hayward@arm.com>
8896             David Sherwood  <david.sherwood@arm.com>
8897
8898         * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
8899         Remove the mode argument.
8900         (aarch64_simd_valid_immediate): Remove the mode and inverse
8901         arguments.
8902         * config/aarch64/iterators.md (bitsize): New iterator.
8903         * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
8904         (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
8905         * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
8906         aarch64_simd_valid_immediate.
8907         * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
8908         (aarch64_reg_or_bic_imm): Likewise.
8909         * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
8910         with an insn_type enum and msl with a modifier_type enum.
8911         Replace element_width with a scalar_mode.  Change the shift
8912         to unsigned int.  Add constructors for scalar_float_mode and
8913         scalar_int_mode elements.
8914         (aarch64_vect_float_const_representable_p): Delete.
8915         (aarch64_can_const_movi_rtx_p)
8916         (aarch64_simd_scalar_immediate_valid_for_move)
8917         (aarch64_simd_make_constant): Update call to
8918         aarch64_simd_valid_immediate.
8919         (aarch64_advsimd_valid_immediate_hs): New function.
8920         (aarch64_advsimd_valid_immediate): Likewise.
8921         (aarch64_simd_valid_immediate): Remove mode and inverse
8922         arguments.  Rewrite to use the above.  Use const_vec_duplicate_p
8923         to detect duplicated constants and use aarch64_float_const_zero_rtx_p
8924         and aarch64_float_const_representable_p on the result.
8925         (aarch64_output_simd_mov_immediate): Remove mode argument.
8926         Update call to aarch64_simd_valid_immediate and use of
8927         simd_immediate_info.
8928         (aarch64_output_scalar_simd_mov_immediate): Update call
8929         accordingly.
8930
8931 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
8932             Alan Hayward  <alan.hayward@arm.com>
8933             David Sherwood  <david.sherwood@arm.com>
8934
8935         * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
8936         (mode_nunits): Likewise CONST_MODE_NUNITS.
8937         * machmode.def (ADJUST_NUNITS): Document.
8938         * genmodes.c (mode_data::need_nunits_adj): New field.
8939         (blank_mode): Update accordingly.
8940         (adj_nunits): New variable.
8941         (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
8942         parameter.
8943         (emit_mode_size_inline): Set need_bytesize_adj for all modes
8944         listed in adj_nunits.
8945         (emit_mode_nunits_inline): Set need_nunits_adj for all modes
8946         listed in adj_nunits.  Don't emit case statements for such modes.
8947         (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
8948         and CONST_MODE_PRECISION.  Make CONST_MODE_SIZE expand to
8949         nothing if adj_nunits is nonnull.
8950         (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
8951         (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
8952         (emit_mode_fbit): Update use of print_maybe_const_decl.
8953         (emit_move_size): Likewise.  Treat the array as non-const
8954         if adj_nunits.
8955         (emit_mode_adjustments): Handle adj_nunits.
8956
8957 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
8958
8959         * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
8960         * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
8961         (VECTOR_MODES): Use it.
8962         (make_vector_modes): Take the prefix as an argument.
8963
8964 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
8965             Alan Hayward  <alan.hayward@arm.com>
8966             David Sherwood  <david.sherwood@arm.com>
8967
8968         * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
8969         * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
8970         for MODE_VECTOR_BOOL.
8971         * machmode.def (VECTOR_BOOL_MODE): Document.
8972         * genmodes.c (VECTOR_BOOL_MODE): New macro.
8973         (make_vector_bool_mode): New function.
8974         (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
8975         MODE_VECTOR_BOOL.
8976         * lto-streamer-in.c (lto_input_mode_table): Likewise.
8977         * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
8978         Likewise.
8979         * stor-layout.c (int_mode_for_mode): Likewise.
8980         * tree.c (build_vector_type_for_mode): Likewise.
8981         * varasm.c (output_constant_pool_2): Likewise.
8982         * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
8983         CONSTM1_RTX (BImode) are the same thing.  Initialize const_tiny_rtx
8984         for MODE_VECTOR_BOOL.
8985         * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
8986         of mode class checks.
8987         * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
8988         instead of a list of mode class checks.
8989         (expand_vector_scalar_condition): Likewise.
8990         (type_for_widest_vector_mode): Handle BImode as an inner mode.
8991
8992 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
8993             Alan Hayward  <alan.hayward@arm.com>
8994             David Sherwood  <david.sherwood@arm.com>
8995
8996         * machmode.h (mode_size): Change from unsigned short to
8997         poly_uint16_pod.
8998         (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
8999         (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
9000         or if measurement_type is not polynomial.
9001         (fixed_size_mode::includes_p): Check for constant-sized modes.
9002         * genmodes.c (emit_mode_size_inline): Make mode_size_inline
9003         return a poly_uint16 rather than an unsigned short.
9004         (emit_mode_size): Change the type of mode_size from unsigned short
9005         to poly_uint16_pod.  Use ZERO_COEFFS for the initializer.
9006         (emit_mode_adjustments): Cope with polynomial vector sizes.
9007         * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
9008         for GET_MODE_SIZE.
9009         * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
9010         for GET_MODE_SIZE.
9011         * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
9012         * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
9013         * caller-save.c (setup_save_areas): Likewise.
9014         (replace_reg_with_saved_mem): Likewise.
9015         * calls.c (emit_library_call_value_1): Likewise.
9016         * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
9017         * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
9018         (gen_lowpart_for_combine): Likewise.
9019         * convert.c (convert_to_integer_1): Likewise.
9020         * cse.c (equiv_constant, cse_insn): Likewise.
9021         * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
9022         (cselib_subst_to_values): Likewise.
9023         * dce.c (word_dce_process_block): Likewise.
9024         * df-problems.c (df_word_lr_mark_ref): Likewise.
9025         * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
9026         * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
9027         (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
9028         (rtl_for_decl_location): Likewise.
9029         * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
9030         * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
9031         * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
9032         (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
9033         (expand_expr_real_1): Likewise.
9034         * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
9035         (pad_below): Likewise.
9036         * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
9037         * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
9038         * ira.c (get_subreg_tracking_sizes): Likewise.
9039         * ira-build.c (ira_create_allocno_objects): Likewise.
9040         * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
9041         (ira_sort_regnos_for_alter_reg): Likewise.
9042         * ira-costs.c (record_operand_costs): Likewise.
9043         * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
9044         (resolve_simple_move): Likewise.
9045         * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
9046         (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
9047         (lra_constraints): Likewise.
9048         (CONST_POOL_OK_P): Reject variable-sized modes.
9049         * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
9050         (add_pseudo_to_slot, lra_spill): Likewise.
9051         * omp-low.c (omp_clause_aligned_alignment): Likewise.
9052         * optabs-query.c (get_best_extraction_insn): Likewise.
9053         * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
9054         * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
9055         (expand_mult_highpart, valid_multiword_target_p): Likewise.
9056         * recog.c (offsettable_address_addr_space_p): Likewise.
9057         * regcprop.c (maybe_mode_change): Likewise.
9058         * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
9059         * regrename.c (build_def_use): Likewise.
9060         * regstat.c (dump_reg_info): Likewise.
9061         * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
9062         (find_reloads, find_reloads_subreg_address): Likewise.
9063         * reload1.c (eliminate_regs_1): Likewise.
9064         * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
9065         * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
9066         (simplify_binary_operation_1, simplify_subreg): Likewise.
9067         * targhooks.c (default_function_arg_padding): Likewise.
9068         (default_hard_regno_nregs, default_class_max_nregs): Likewise.
9069         * tree-cfg.c (verify_gimple_assign_binary): Likewise.
9070         (verify_gimple_assign_ternary): Likewise.
9071         * tree-inline.c (estimate_move_cost): Likewise.
9072         * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
9073         * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
9074         (get_address_cost_ainc): Likewise.
9075         * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
9076         (vect_supportable_dr_alignment): Likewise.
9077         * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
9078         (vectorizable_reduction): Likewise.
9079         * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
9080         (vectorizable_operation, vectorizable_load): Likewise.
9081         * tree.c (build_same_sized_truth_vector_type): Likewise.
9082         * valtrack.c (cleanup_auto_inc_dec): Likewise.
9083         * var-tracking.c (emit_note_insn_var_location): Likewise.
9084         * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
9085         (ADDR_VEC_ALIGN): Likewise.
9086
9087 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
9088             Alan Hayward  <alan.hayward@arm.com>
9089             David Sherwood  <david.sherwood@arm.com>
9090
9091         * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
9092         unsigned short.
9093         (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
9094         or if measurement_type is polynomial.
9095         * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
9096         * combine.c (make_extraction): Likewise.
9097         * dse.c (find_shift_sequence): Likewise.
9098         * dwarf2out.c (mem_loc_descriptor): Likewise.
9099         * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
9100         (extract_bit_field, extract_low_bits): Likewise.
9101         * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
9102         (optimize_bitfield_assignment_op, expand_assignment): Likewise.
9103         (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
9104         * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
9105         * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
9106         * reload.c (find_reloads): Likewise.
9107         * reload1.c (alter_reg): Likewise.
9108         * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
9109         * targhooks.c (default_secondary_memory_needed_mode): Likewise.
9110         * tree-if-conv.c (predicate_mem_writes): Likewise.
9111         * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
9112         * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
9113         * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
9114         * valtrack.c (dead_debug_insert_temp): Likewise.
9115         * varasm.c (mergeable_constant_section): Likewise.
9116         * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
9117
9118 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
9119             Alan Hayward  <alan.hayward@arm.com>
9120             David Sherwood  <david.sherwood@arm.com>
9121
9122         * expr.c (expand_assignment): Cope with polynomial mode sizes
9123         when assigning to a CONCAT.
9124
9125 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
9126             Alan Hayward  <alan.hayward@arm.com>
9127             David Sherwood  <david.sherwood@arm.com>
9128
9129         * machmode.h (mode_precision): Change from unsigned short to
9130         poly_uint16_pod.
9131         (mode_to_precision): Return a poly_uint16 rather than an unsigned
9132         short.
9133         (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
9134         or if measurement_type is not polynomial.
9135         (HWI_COMPUTABLE_MODE_P): Turn into a function.  Optimize the case
9136         in which the mode is already known to be a scalar_int_mode.
9137         * genmodes.c (emit_mode_precision): Change the type of mode_precision
9138         from unsigned short to poly_uint16_pod.  Use ZERO_COEFFS for the
9139         initializer.
9140         * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
9141         for GET_MODE_PRECISION.
9142         * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
9143         for GET_MODE_PRECISION.
9144         * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
9145         as polynomial.
9146         (try_combine, find_split_point, combine_simplify_rtx): Likewise.
9147         (expand_field_assignment, make_extraction): Likewise.
9148         (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
9149         (get_last_value): Likewise.
9150         * convert.c (convert_to_integer_1): Likewise.
9151         * cse.c (cse_insn): Likewise.
9152         * expr.c (expand_expr_real_1): Likewise.
9153         * lra-constraints.c (simplify_operand_subreg): Likewise.
9154         * optabs-query.c (can_atomic_load_p): Likewise.
9155         * optabs.c (expand_atomic_load): Likewise.
9156         (expand_atomic_store): Likewise.
9157         * ree.c (combine_reaching_defs): Likewise.
9158         * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
9159         * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
9160         * tree.h (type_has_mode_precision_p): Likewise.
9161         * ubsan.c (instrument_si_overflow): Likewise.
9162
9163 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
9164             Alan Hayward  <alan.hayward@arm.com>
9165             David Sherwood  <david.sherwood@arm.com>
9166
9167         * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
9168         polynomial numbers of units.
9169         (SET_TYPE_VECTOR_SUBPARTS): Likewise.
9170         (valid_vector_subparts_p): New function.
9171         (build_vector_type): Remove temporary shim and take the number
9172         of units as a poly_uint64 rather than an int.
9173         (build_opaque_vector_type): Take the number of units as a
9174         poly_uint64 rather than an int.
9175         * tree.c (build_vector_from_ctor): Handle polynomial
9176         TYPE_VECTOR_SUBPARTS.
9177         (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
9178         (uniform_vector_p, vector_type_mode, build_vector): Likewise.
9179         (build_vector_from_val): If the number of units is variable,
9180         use build_vec_duplicate_cst for constant operands and
9181         VEC_DUPLICATE_EXPR otherwise.
9182         (make_vector_type): Remove temporary is_constant ().
9183         (build_vector_type, build_opaque_vector_type): Take the number of
9184         units as a poly_uint64 rather than an int.
9185         (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
9186         VECTOR_CST_NELTS.
9187         * cfgexpand.c (expand_debug_expr): Likewise.
9188         * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
9189         (store_constructor, expand_expr_real_1): Likewise.
9190         (const_scalar_mask_from_tree): Likewise.
9191         * fold-const-call.c (fold_const_reduction): Likewise.
9192         * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
9193         (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
9194         (native_encode_vector, vec_cst_ctor_to_array): Likewise.
9195         (fold_relational_const): Likewise.
9196         (native_interpret_vector): Likewise.  Change the size from an
9197         int to an unsigned int.
9198         * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
9199         TYPE_VECTOR_SUBPARTS.
9200         (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
9201         (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
9202         duplicating a non-constant operand into a variable-length vector.
9203         * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
9204         TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
9205         * ipa-icf.c (sem_variable::equals): Likewise.
9206         * match.pd: Likewise.
9207         * omp-simd-clone.c (simd_clone_subparts): Likewise.
9208         * print-tree.c (print_node): Likewise.
9209         * stor-layout.c (layout_type): Likewise.
9210         * targhooks.c (default_builtin_vectorization_cost): Likewise.
9211         * tree-cfg.c (verify_gimple_comparison): Likewise.
9212         (verify_gimple_assign_binary): Likewise.
9213         (verify_gimple_assign_ternary): Likewise.
9214         (verify_gimple_assign_single): Likewise.
9215         * tree-pretty-print.c (dump_generic_node): Likewise.
9216         * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
9217         (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
9218         * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
9219         (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
9220         (vect_shift_permute_load_chain): Likewise.
9221         * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
9222         (expand_vector_condition, optimize_vector_constructor): Likewise.
9223         (lower_vec_perm, get_compute_type): Likewise.
9224         * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
9225         (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
9226         * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
9227         (vect_recog_mask_conversion_pattern): Likewise.
9228         * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
9229         (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
9230         * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
9231         (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
9232         (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
9233         (vectorizable_shift, vectorizable_operation, vectorizable_store)
9234         (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
9235         (supportable_widening_operation): Likewise.
9236         (supportable_narrowing_operation): Likewise.
9237         * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
9238         Likewise.
9239         * varasm.c (output_constant): Likewise.
9240
9241 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
9242             Alan Hayward  <alan.hayward@arm.com>
9243             David Sherwood  <david.sherwood@arm.com>
9244
9245         * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
9246         so that both the length == 3 and length != 3 cases set up their
9247         own permute vectors.  Add comments explaining why we know the
9248         number of elements is constant.
9249         (vect_permute_load_chain): Likewise.
9250
9251 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
9252             Alan Hayward  <alan.hayward@arm.com>
9253             David Sherwood  <david.sherwood@arm.com>
9254
9255         * machmode.h (mode_nunits): Change from unsigned char to
9256         poly_uint16_pod.
9257         (ONLY_FIXED_SIZE_MODES): New macro.
9258         (pod_mode::measurement_type, scalar_int_mode::measurement_type)
9259         (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
9260         (complex_mode::measurement_type, fixed_size_mode::measurement_type):
9261         New typedefs.
9262         (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
9263         (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
9264         or if measurement_type is not polynomial.
9265         * genmodes.c (ZERO_COEFFS): New macro.
9266         (emit_mode_nunits_inline): Make mode_nunits_inline return a
9267         poly_uint16.
9268         (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
9269         Use ZERO_COEFFS when emitting initializers.
9270         * data-streamer.h (bp_pack_poly_value): New function.
9271         (bp_unpack_poly_value): Likewise.
9272         * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
9273         for GET_MODE_NUNITS.
9274         * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
9275         for GET_MODE_NUNITS.
9276         * tree.c (make_vector_type): Remove temporary shim and make
9277         the real function take the number of units as a poly_uint64
9278         rather than an int.
9279         (build_vector_type_for_mode): Handle polynomial nunits.
9280         * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
9281         * emit-rtl.c (const_vec_series_p_1): Likewise.
9282         (gen_rtx_CONST_VECTOR): Likewise.
9283         * fold-const.c (test_vec_duplicate_folding): Likewise.
9284         * genrecog.c (validate_pattern): Likewise.
9285         * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
9286         * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
9287         * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
9288         (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
9289         (expand_vec_cond_expr, expand_mult_highpart): Likewise.
9290         * rtlanal.c (subreg_get_info): Likewise.
9291         * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
9292         (vect_grouped_load_supported): Likewise.
9293         * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
9294         * tree-vect-loop.c (have_whole_vector_shift): Likewise.
9295         * simplify-rtx.c (simplify_unary_operation_1): Likewise.
9296         (simplify_const_unary_operation, simplify_binary_operation_1)
9297         (simplify_const_binary_operation, simplify_ternary_operation)
9298         (test_vector_ops_duplicate, test_vector_ops): Likewise.
9299         (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
9300         instead of CONST_VECTOR_NUNITS.
9301         * varasm.c (output_constant_pool_2): Likewise.
9302         * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
9303         explicit-encoded elements in the XVEC for variable-length vectors.
9304
9305 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
9306
9307         * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
9308
9309 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
9310             Alan Hayward  <alan.hayward@arm.com>
9311             David Sherwood  <david.sherwood@arm.com>
9312
9313         * coretypes.h (fixed_size_mode): Declare.
9314         (fixed_size_mode_pod): New typedef.
9315         * builtins.h (target_builtins::x_apply_args_mode)
9316         (target_builtins::x_apply_result_mode): Change type to
9317         fixed_size_mode_pod.
9318         * builtins.c (apply_args_size, apply_result_size, result_vector)
9319         (expand_builtin_apply_args_1, expand_builtin_apply)
9320         (expand_builtin_return): Update accordingly.
9321
9322 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
9323
9324         * cse.c (hash_rtx_cb): Hash only the encoded elements.
9325         * cselib.c (cselib_hash_rtx): Likewise.
9326         * expmed.c (make_tree): Build VECTOR_CSTs directly from the
9327         CONST_VECTOR encoding.
9328
9329 2018-01-03  Jakub Jelinek  <jakub@redhat.com>
9330             Jeff Law  <law@redhat.com>
9331
9332         PR target/83641
9333         * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
9334         noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
9335         only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
9336         and add REG_CFA_ADJUST_CFA notes in that case to both insns.
9337
9338         PR target/83641
9339         * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
9340         explicitly probe *sp in a noreturn function if there were any callee
9341         register saves or frame pointer is needed.
9342
9343 2018-01-03  Jakub Jelinek  <jakub@redhat.com>
9344
9345         PR debug/83621
9346         * cfgexpand.c (expand_debug_expr): Return NULL if mode is
9347         BLKmode for ternary, binary or unary expressions.
9348
9349         PR debug/83645
9350         * var-tracking.c (delete_vta_debug_insn): New inline function.
9351         (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
9352         insns from get_insns () to NULL instead of each bb separately.
9353         Use delete_vta_debug_insn.  No longer static.
9354         (vt_debug_insns_local, variable_tracking_main_1): Adjust
9355         delete_vta_debug_insns callers.
9356         * rtl.h (delete_vta_debug_insns): Declare.
9357         * final.c (rest_of_handle_final): Call delete_vta_debug_insns
9358         instead of variable_tracking_main.
9359
9360 2018-01-03  Martin Sebor  <msebor@redhat.com>
9361
9362         PR tree-optimization/83603
9363         * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
9364         arguments past the endof the argument list in functions declared
9365         without a prototype.
9366         * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
9367         Avoid checking when arguments are null.
9368
9369 2018-01-03  Martin Sebor  <msebor@redhat.com>
9370
9371         PR c/83559
9372         * doc/extend.texi (attribute const): Fix a typo.
9373         * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
9374         issuing -Wsuggest-attribute for void functions.
9375
9376 2018-01-03  Martin Sebor  <msebor@redhat.com>
9377
9378         * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
9379         offset_int::from instead of wide_int::to_shwi.
9380         (maybe_diag_overlap): Remove assertion.
9381         Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
9382         * gimple-ssa-sprintf.c (format_directive): Same.
9383         (parse_directive): Same.
9384         (sprintf_dom_walker::compute_format_length): Same.
9385         (try_substitute_return_value): Same.
9386
9387 2018-01-03  Jeff Law  <law@redhat.com>
9388
9389         PR middle-end/83654
9390         * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
9391         non-constant residual for zero at runtime and avoid probing in
9392         that case.  Reorganize code for trailing problem to mirror handling
9393         of the residual.
9394
9395 2018-01-03  Prathamesh Kulkarni  <prathamesh.kulkarni@linaro.org>
9396
9397         PR tree-optimization/83501
9398         * tree-ssa-strlen.c (get_string_cst): New.
9399         (handle_char_store): Call get_string_cst.
9400
9401 2018-01-03  Martin Liska  <mliska@suse.cz>
9402
9403         PR tree-optimization/83593
9404         * tree-ssa-strlen.c: Include tree-cfg.h.
9405         (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
9406         (strlen_dom_walker): Add new member variable m_cleanup_cfg.
9407         (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
9408         to false.
9409         (strlen_dom_walker::before_dom_children): Call
9410         gimple_purge_dead_eh_edges. Dump tranformation with details
9411         dump flags.
9412         (strlen_dom_walker::before_dom_children): Update call by adding
9413         new argument cleanup_eh.
9414         (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
9415
9416 2018-01-03  Martin Liska  <mliska@suse.cz>
9417
9418         PR ipa/83549
9419         * cif-code.def (VARIADIC_THUNK): New enum value.
9420         * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
9421         thunks.
9422
9423 2018-01-03  Jan Beulich  <jbeulich@suse.com>
9424
9425         * sse.md (mov<mode>_internal): Tighten condition for when to use
9426         vmovdqu<ssescalarsize> for TI and OI modes.
9427
9428 2018-01-03  Jakub Jelinek  <jakub@redhat.com>
9429
9430         Update copyright years.
9431
9432 2018-01-03  Martin Liska  <mliska@suse.cz>
9433
9434         PR ipa/83594
9435         * ipa-visibility.c (function_and_variable_visibility): Skip
9436         functions with noipa attribure.
9437
9438 2018-01-03  Jakub Jelinek  <jakub@redhat.com>
9439
9440         * gcc.c (process_command): Update copyright notice dates.
9441         * gcov-dump.c (print_version): Ditto.
9442         * gcov.c (print_version): Ditto.
9443         * gcov-tool.c (print_version): Ditto.
9444         * gengtype.c (create_file): Ditto.
9445         * doc/cpp.texi: Bump @copying's copyright year.
9446         * doc/cppinternals.texi: Ditto.
9447         * doc/gcc.texi: Ditto.
9448         * doc/gccint.texi: Ditto.
9449         * doc/gcov.texi: Ditto.
9450         * doc/install.texi: Ditto.
9451         * doc/invoke.texi: Ditto.
9452
9453 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
9454
9455         * vector-builder.h (vector_builder::m_full_nelts): Change from
9456         unsigned int to poly_uint64.
9457         (vector_builder::full_nelts): Update prototype accordingly.
9458         (vector_builder::new_vector): Likewise.
9459         (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
9460         (vector_builder::operator ==): Likewise.
9461         (vector_builder::finalize): Likewise.
9462         * int-vector-builder.h (int_vector_builder::int_vector_builder):
9463         Take the number of elements as a poly_uint64 rather than an
9464         unsigned int.
9465         * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
9466         from unsigned int to poly_uint64.
9467         (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
9468         (vec_perm_indices::new_vector): Likewise.
9469         (vec_perm_indices::length): Likewise.
9470         (vec_perm_indices::nelts_per_input): Likewise.
9471         (vec_perm_indices::input_nelts): Likewise.
9472         * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
9473         number of elements per input as a poly_uint64 rather than an
9474         unsigned int.  Use the original encoding for variable-length
9475         vectors, rather than clamping each individual element.
9476         For the second and subsequent elements in each pattern,
9477         clamp the step and base before clamping their sum.
9478         (vec_perm_indices::series_p): Handle polynomial element counts.
9479         (vec_perm_indices::all_in_range_p): Likewise.
9480         (vec_perm_indices_to_tree): Likewise.
9481         (vec_perm_indices_to_rtx): Likewise.
9482         * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
9483         * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
9484         (tree_vector_builder::new_binary_operation): Handle polynomial
9485         element counts.  Return false if we need to know the number
9486         of elements at compile time.
9487         * fold-const.c (fold_vec_perm): Punt if the number of elements
9488         isn't known at compile time.
9489
9490 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
9491
9492         * vec-perm-indices.h (vec_perm_builder): Change element type
9493         from HOST_WIDE_INT to poly_int64.
9494         (vec_perm_indices::element_type): Update accordingly.
9495         (vec_perm_indices::clamp): Handle polynomial element_types.
9496         * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
9497         (vec_perm_indices::all_in_range_p): Likewise.
9498         (tree_to_vec_perm_builder): Check for poly_int64 trees rather
9499         than shwi trees.
9500         * vector-builder.h (vector_builder::stepped_sequence_p): Handle
9501         polynomial vec_perm_indices element types.
9502         * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
9503         * fold-const.c (fold_vec_perm): Likewise.
9504         * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
9505         * tree-vect-generic.c (lower_vec_perm): Likewise.
9506         * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
9507         * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
9508         element type to HOST_WIDE_INT.
9509
9510 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
9511             Alan Hayward  <alan.hayward@arm.com>
9512             David Sherwood  <david.sherwood@arm.com>
9513
9514         * alias.c (addr_side_effect_eval): Take the size as a poly_int64
9515         rather than an int.  Use plus_constant.
9516         (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
9517         Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
9518
9519 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
9520             Alan Hayward  <alan.hayward@arm.com>
9521             David Sherwood  <david.sherwood@arm.com>
9522
9523         * calls.c (emit_call_1, expand_call): Change struct_value_size from
9524         a HOST_WIDE_INT to a poly_int64.
9525
9526 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
9527             Alan Hayward  <alan.hayward@arm.com>
9528             David Sherwood  <david.sherwood@arm.com>
9529
9530         * calls.c (load_register_parameters): Cope with polynomial
9531         mode sizes.  Require a constant size for BLKmode parameters
9532         that aren't described by a PARALLEL.  If BLOCK_REG_PADDING
9533         forces a parameter to be padded at the lsb end in order to
9534         fill a complete number of words, require the parameter size
9535         to be ordered wrt UNITS_PER_WORD.
9536
9537 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
9538             Alan Hayward  <alan.hayward@arm.com>
9539             David Sherwood  <david.sherwood@arm.com>
9540
9541         * reload1.c (spill_stack_slot_width): Change element type
9542         from unsigned int to poly_uint64_pod.
9543         (alter_reg): Treat mode sizes as polynomial.
9544
9545 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
9546             Alan Hayward  <alan.hayward@arm.com>
9547             David Sherwood  <david.sherwood@arm.com>
9548
9549         * reload.c (complex_word_subreg_p): New function.
9550         (reload_inner_reg_of_subreg, push_reload): Use it.
9551
9552 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
9553             Alan Hayward  <alan.hayward@arm.com>
9554             David Sherwood  <david.sherwood@arm.com>
9555
9556         * lra-constraints.c (process_alt_operands): Reject matched
9557         operands whose sizes aren't ordered.
9558         (match_reload): Refer to this check here.
9559
9560 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
9561             Alan Hayward  <alan.hayward@arm.com>
9562             David Sherwood  <david.sherwood@arm.com>
9563
9564         * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
9565         that the mode size is in the set {1, 2, 4, 8, 16}.
9566
9567 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
9568             Alan Hayward  <alan.hayward@arm.com>
9569             David Sherwood  <david.sherwood@arm.com>
9570
9571         * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
9572         Use plus_constant instead of gen_rtx_PLUS.
9573
9574 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
9575             Alan Hayward  <alan.hayward@arm.com>
9576             David Sherwood  <david.sherwood@arm.com>
9577
9578         * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
9579         * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
9580         * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
9581         * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
9582         * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
9583         * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
9584         * config/i386/i386-protos.h (ix86_push_rounding): Declare.
9585         * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
9586         * config/i386/i386.c (ix86_push_rounding): ...this new function.
9587         * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
9588         a poly_int64.
9589         * config/m32c/m32c.c (m32c_push_rounding): Likewise.
9590         * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
9591         * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
9592         * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
9593         * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
9594         * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
9595         * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
9596         * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
9597         * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
9598         * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
9599         function.
9600         * expr.c (emit_move_resolve_push): Treat the input and result
9601         of PUSH_ROUNDING as a poly_int64.
9602         (emit_move_complex_push, emit_single_push_insn_1): Likewise.
9603         (emit_push_insn): Likewise.
9604         * lra-eliminations.c (mark_not_eliminable): Likewise.
9605         * recog.c (push_operand): Likewise.
9606         * reload1.c (elimination_effects): Likewise.
9607         * rtlanal.c (nonzero_bits1): Likewise.
9608         * calls.c (store_one_arg): Likewise.  Require the padding to be
9609         known at compile time.
9610
9611 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
9612             Alan Hayward  <alan.hayward@arm.com>
9613             David Sherwood  <david.sherwood@arm.com>
9614
9615         * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
9616         Use plus_constant instead of gen_rtx_PLUS.
9617
9618 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
9619             Alan Hayward  <alan.hayward@arm.com>
9620             David Sherwood  <david.sherwood@arm.com>
9621
9622         * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
9623         rather than an int.
9624
9625 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
9626             Alan Hayward  <alan.hayward@arm.com>
9627             David Sherwood  <david.sherwood@arm.com>
9628
9629         * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
9630         instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
9631         via stack temporaries.  Treat the mode size as polynomial too.
9632
9633 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
9634             Alan Hayward  <alan.hayward@arm.com>
9635             David Sherwood  <david.sherwood@arm.com>
9636
9637         * expr.c (expand_expr_real_2): When handling conversions involving
9638         unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
9639         multiplying int_size_in_bytes by BITS_PER_UNIT.  Treat GET_MODE_BISIZE
9640         as a poly_uint64 too.
9641
9642 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
9643             Alan Hayward  <alan.hayward@arm.com>
9644             David Sherwood  <david.sherwood@arm.com>
9645
9646         * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
9647
9648 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
9649             Alan Hayward  <alan.hayward@arm.com>
9650             David Sherwood  <david.sherwood@arm.com>
9651
9652         * combine.c (can_change_dest_mode): Handle polynomial
9653         REGMODE_NATURAL_SIZE.
9654         * expmed.c (store_bit_field_1): Likewise.
9655         * expr.c (store_constructor): Likewise.
9656         * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
9657         and polynomial REGMODE_NATURAL_SIZE.
9658         (gen_lowpart_common): Likewise.
9659         * reginfo.c (record_subregs_of_mode): Likewise.
9660         * rtlanal.c (read_modify_subreg_p): Likewise.
9661
9662 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
9663             Alan Hayward  <alan.hayward@arm.com>
9664             David Sherwood  <david.sherwood@arm.com>
9665
9666         * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
9667         numbers of elements.
9668
9669 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
9670             Alan Hayward  <alan.hayward@arm.com>
9671             David Sherwood  <david.sherwood@arm.com>
9672
9673         * match.pd: Cope with polynomial numbers of vector elements.
9674
9675 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
9676             Alan Hayward  <alan.hayward@arm.com>
9677             David Sherwood  <david.sherwood@arm.com>
9678
9679         * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
9680         in a POINTER_PLUS_EXPR.
9681
9682 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
9683             Alan Hayward  <alan.hayward@arm.com>
9684             David Sherwood  <david.sherwood@arm.com>
9685
9686         * omp-simd-clone.c (simd_clone_subparts): New function.
9687         (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
9688         (ipa_simd_modify_function_body): Likewise.
9689
9690 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
9691             Alan Hayward  <alan.hayward@arm.com>
9692             David Sherwood  <david.sherwood@arm.com>
9693
9694         * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
9695         (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
9696         (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
9697         (expand_vector_condition, vector_element): Likewise.
9698         (subparts_gt): New function.
9699         (get_compute_type): Use subparts_gt.
9700         (count_type_subparts): Delete.
9701         (expand_vector_operations_1): Use subparts_gt instead of
9702         count_type_subparts.
9703
9704 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
9705             Alan Hayward  <alan.hayward@arm.com>
9706             David Sherwood  <david.sherwood@arm.com>
9707
9708         * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
9709         (vect_compile_time_alias): ...this new function.  Do the calculation
9710         on poly_ints rather than trees.
9711         (vect_prune_runtime_alias_test_list): Update call accordingly.
9712
9713 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
9714             Alan Hayward  <alan.hayward@arm.com>
9715             David Sherwood  <david.sherwood@arm.com>
9716
9717         * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
9718         numbers of units.
9719         (vect_schedule_slp_instance): Likewise.
9720
9721 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
9722             Alan Hayward  <alan.hayward@arm.com>
9723             David Sherwood  <david.sherwood@arm.com>
9724
9725         * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
9726         constant and extern definitions for variable-length vectors.
9727         (vect_get_constant_vectors): Note that the number of units
9728         is known to be constant.
9729
9730 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
9731             Alan Hayward  <alan.hayward@arm.com>
9732             David Sherwood  <david.sherwood@arm.com>
9733
9734         * tree-vect-stmts.c (vectorizable_conversion): Treat the number
9735         of units as polynomial.  Choose between WIDE and NARROW based
9736         on multiple_p.
9737
9738 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
9739             Alan Hayward  <alan.hayward@arm.com>
9740             David Sherwood  <david.sherwood@arm.com>
9741
9742         * tree-vect-stmts.c (simd_clone_subparts): New function.
9743         (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
9744
9745 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
9746             Alan Hayward  <alan.hayward@arm.com>
9747             David Sherwood  <david.sherwood@arm.com>
9748
9749         * tree-vect-stmts.c (vectorizable_call): Treat the number of
9750         vectors as polynomial.  Use build_index_vector for
9751         IFN_GOMP_SIMD_LANE.
9752
9753 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
9754             Alan Hayward  <alan.hayward@arm.com>
9755             David Sherwood  <david.sherwood@arm.com>
9756
9757         * tree-vect-stmts.c (get_load_store_type): Treat the number of
9758         units as polynomial.  Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
9759         for variable-length vectors.
9760         (vectorizable_mask_load_store): Treat the number of units as
9761         polynomial, asserting that it is constant if the condition has
9762         already been enforced.
9763         (vectorizable_store, vectorizable_load): Likewise.
9764
9765 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
9766             Alan Hayward  <alan.hayward@arm.com>
9767             David Sherwood  <david.sherwood@arm.com>
9768
9769         * tree-vect-loop.c (vectorizable_live_operation): Treat the number
9770         of units as polynomial.  Punt if we can't tell at compile time
9771         which vector contains the final result.
9772
9773 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
9774             Alan Hayward  <alan.hayward@arm.com>
9775             David Sherwood  <david.sherwood@arm.com>
9776
9777         * tree-vect-loop.c (vectorizable_induction): Treat the number
9778         of units as polynomial.  Punt on SLP inductions.  Use an integer
9779         VEC_SERIES_EXPR for variable-length integer reductions.  Use a
9780         cast of such a series for variable-length floating-point
9781         reductions.
9782
9783 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
9784             Alan Hayward  <alan.hayward@arm.com>
9785             David Sherwood  <david.sherwood@arm.com>
9786
9787         * tree.h (build_index_vector): Declare.
9788         * tree.c (build_index_vector): New function.
9789         * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
9790         of units as polynomial, forcibly converting it to a constant if
9791         vectorizable_reduction has already enforced the condition.
9792         (vect_create_epilog_for_reduction): Likewise.  Use build_index_vector
9793         to create a {1,2,3,...} vector.
9794         (vectorizable_reduction): Treat the number of units as polynomial.
9795         Choose vectype_in based on the largest scalar element size rather
9796         than the smallest number of units.  Enforce the restrictions
9797         relied on above.
9798
9799 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
9800             Alan Hayward  <alan.hayward@arm.com>
9801             David Sherwood  <david.sherwood@arm.com>
9802
9803         * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
9804         number of units as polynomial.
9805
9806 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
9807             Alan Hayward  <alan.hayward@arm.com>
9808             David Sherwood  <david.sherwood@arm.com>
9809
9810         * target.h (vector_sizes, auto_vector_sizes): New typedefs.
9811         * target.def (autovectorize_vector_sizes): Return the vector sizes
9812         by pointer, using vector_sizes rather than a bitmask.
9813         * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
9814         * targhooks.c (default_autovectorize_vector_sizes): Likewise.
9815         * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
9816         Likewise.
9817         * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
9818         * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
9819         * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
9820         * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
9821         * omp-general.c (omp_max_vf): Likewise.
9822         * omp-low.c (omp_clause_aligned_alignment): Likewise.
9823         * optabs-query.c (can_vec_mask_load_store_p): Likewise.
9824         * tree-vect-loop.c (vect_analyze_loop): Likewise.
9825         * tree-vect-slp.c (vect_slp_bb): Likewise.
9826         * doc/tm.texi: Regenerate.
9827         * tree-vectorizer.h (current_vector_size): Change from an unsigned int
9828         to a poly_uint64.
9829         * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
9830         the vector size as a poly_uint64 rather than an unsigned int.
9831         (current_vector_size): Change from an unsigned int to a poly_uint64.
9832         (get_vectype_for_scalar_type): Update accordingly.
9833         * tree.h (build_truth_vector_type): Take the size and number of
9834         units as a poly_uint64 rather than an unsigned int.
9835         (build_vector_type): Add a temporary overload that takes
9836         the number of units as a poly_uint64 rather than an unsigned int.
9837         * tree.c (make_vector_type): Likewise.
9838         (build_truth_vector_type): Take the number of units as a poly_uint64
9839         rather than an unsigned int.
9840
9841 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
9842             Alan Hayward  <alan.hayward@arm.com>
9843             David Sherwood  <david.sherwood@arm.com>
9844
9845         * target.def (get_mask_mode): Take the number of units and length
9846         as poly_uint64s rather than unsigned ints.
9847         * targhooks.h (default_get_mask_mode): Update accordingly.
9848         * targhooks.c (default_get_mask_mode): Likewise.
9849         * config/i386/i386.c (ix86_get_mask_mode): Likewise.
9850         * doc/tm.texi: Regenerate.
9851
9852 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
9853             Alan Hayward  <alan.hayward@arm.com>
9854             David Sherwood  <david.sherwood@arm.com>
9855
9856         * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
9857         * omp-general.c (omp_max_vf): Likewise.
9858         * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
9859         (expand_omp_simd): Handle polynomial safelen.
9860         * omp-low.c (omplow_simd_context): Add a default constructor.
9861         (omplow_simd_context::max_vf): Change from int to poly_uint64.
9862         (lower_rec_simd_input_clauses): Update accordingly.
9863         (lower_rec_input_clauses): Likewise.
9864
9865 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
9866             Alan Hayward  <alan.hayward@arm.com>
9867             David Sherwood  <david.sherwood@arm.com>
9868
9869         * tree-vectorizer.h (vect_nunits_for_cost): New function.
9870         * tree-vect-loop.c (vect_model_reduction_cost): Use it.
9871         * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
9872         (vect_analyze_slp_cost): Likewise.
9873         * tree-vect-stmts.c (vect_model_store_cost): Likewise.
9874         (vect_model_load_cost): Likewise.
9875
9876 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
9877             Alan Hayward  <alan.hayward@arm.com>
9878             David Sherwood  <david.sherwood@arm.com>
9879
9880         * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
9881         (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
9882         from an unsigned int * to a poly_uint64_pod *.
9883         (calculate_unrolling_factor): New function.
9884         (vect_analyze_slp_instance): Use it.  Track polynomial max_nunits.
9885
9886 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
9887             Alan Hayward  <alan.hayward@arm.com>
9888             David Sherwood  <david.sherwood@arm.com>
9889
9890         * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
9891         from an unsigned int to a poly_uint64.
9892         (_loop_vec_info::slp_unrolling_factor): Likewise.
9893         (_loop_vec_info::vectorization_factor): Change from an int
9894         to a poly_uint64.
9895         (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
9896         (vect_get_num_vectors): New function.
9897         (vect_update_max_nunits, vect_vf_for_cost): Likewise.
9898         (vect_get_num_copies): Use vect_get_num_vectors.
9899         (vect_analyze_data_ref_dependences): Change max_vf from an int *
9900         to an unsigned int *.
9901         (vect_analyze_data_refs): Change min_vf from an int * to a
9902         poly_uint64 *.
9903         (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
9904         than an unsigned HOST_WIDE_INT.
9905         * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
9906         (vect_analyze_data_ref_dependence): Change max_vf from an int *
9907         to an unsigned int *.
9908         (vect_analyze_data_ref_dependences): Likewise.
9909         (vect_compute_data_ref_alignment): Handle polynomial vf.
9910         (vect_enhance_data_refs_alignment): Likewise.
9911         (vect_prune_runtime_alias_test_list): Likewise.
9912         (vect_shift_permute_load_chain): Likewise.
9913         (vect_supportable_dr_alignment): Likewise.
9914         (dependence_distance_ge_vf): Take the vectorization factor as a
9915         poly_uint64 rather than an unsigned HOST_WIDE_INT.
9916         (vect_analyze_data_refs): Change min_vf from an int * to a
9917         poly_uint64 *.
9918         * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
9919         vfm1 as a poly_uint64 rather than an int.  Make the same change
9920         for the returned bound_scalar.
9921         (vect_gen_vector_loop_niters): Handle polynomial vf.
9922         (vect_do_peeling): Likewise.  Update call to
9923         vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
9924         (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
9925         be constant.
9926         * tree-vect-loop.c (vect_determine_vectorization_factor)
9927         (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
9928         (vect_get_known_peeling_cost): Likewise.
9929         (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
9930         (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
9931         (vect_transform_loop): Likewise.  Use the lowest possible VF when
9932         updating the upper bounds of the loop.
9933         (vect_min_worthwhile_factor): Make static.  Return an unsigned int
9934         rather than an int.
9935         * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
9936         polynomial unroll factors.
9937         (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
9938         (vect_make_slp_decision): Likewise.
9939         (vect_supported_load_permutation_p): Likewise, and polynomial
9940         vf too.
9941         (vect_analyze_slp_cost): Handle polynomial vf.
9942         (vect_slp_analyze_node_operations): Likewise.
9943         (vect_slp_analyze_bb_1): Likewise.
9944         (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
9945         than an unsigned HOST_WIDE_INT.
9946         * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
9947         (vectorizable_load): Handle polynomial vf.
9948         * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
9949         a poly_uint64.
9950         (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
9951
9952 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
9953             Alan Hayward  <alan.hayward@arm.com>
9954             David Sherwood  <david.sherwood@arm.com>
9955
9956         * match.pd: Handle bit operations involving three constants
9957         and try to fold one pair.
9958
9959 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
9960
9961         * tree-vect-loop-manip.c: Include gimple-fold.h.
9962         (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
9963         niters_maybe_zero parameters.  Handle other cases besides a step of 1.
9964         (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
9965         Add a path that uses a step of VF instead of 1, but disable it
9966         for now.
9967         (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
9968         and niters_no_overflow parameters.  Update calls to
9969         slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
9970         Create a new SSA name if the latter choses to use a ste other
9971         than zero, and return it via niters_vector_mult_vf_var.
9972         * tree-vect-loop.c (vect_transform_loop): Update calls to
9973         vect_do_peeling, vect_gen_vector_loop_niters and
9974         slpeel_make_loop_iterate_ntimes.
9975         * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
9976         (vect_gen_vector_loop_niters): Update declarations after above changes.
9977
9978 2018-01-02  Michael Meissner  <meissner@linux.vnet.ibm.com>
9979
9980         * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
9981         128-bit round to integer instructions.
9982         (ceil<mode>2): Likewise.
9983         (btrunc<mode>2): Likewise.
9984         (round<mode>2): Likewise.
9985
9986 2018-01-02  Aaron Sawdey  <acsawdey@linux.vnet.ibm.com>
9987
9988         * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
9989         unaligned VSX load/store on P8/P9.
9990         (expand_block_clear): Allow the use of unaligned VSX
9991         load/store on P8/P9.
9992
9993 2018-01-02  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
9994
9995         * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
9996         New function.
9997         (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
9998         swap associated with both a load and a store.
9999
10000 2018-01-02  Andrew Waterman  <andrew@sifive.com>
10001
10002         * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
10003         * config/riscv/riscv.md (clear_cache): Use it.
10004
10005 2018-01-02  Artyom Skrobov  <tyomitch@gmail.com>
10006
10007         * web.c: Remove out-of-date comment.
10008
10009 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
10010
10011         * expr.c (fixup_args_size_notes): Check that any existing
10012         REG_ARGS_SIZE notes are correct, and don't try to re-add them.
10013         (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
10014         (emit_single_push_insn): ...here.
10015
10016 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
10017
10018         * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
10019         (const_vector_encoded_nelts): New function.
10020         (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
10021         (const_vector_int_elt, const_vector_elt): Declare.
10022         * emit-rtl.c (const_vector_int_elt_1): New function.
10023         (const_vector_elt): Likewise.
10024         * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
10025         of CONST_VECTOR_ELT.
10026
10027 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
10028
10029         * expr.c: Include rtx-vector-builder.h.
10030         (const_vector_mask_from_tree): Use rtx_vector_builder and operate
10031         directly on the tree encoding.
10032         (const_vector_from_tree): Likewise.
10033         * optabs.c: Include rtx-vector-builder.h.
10034         (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
10035         sequence of "u" values.
10036         * vec-perm-indices.c: Include rtx-vector-builder.h.
10037         (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
10038         directly on the vec_perm_indices encoding.
10039
10040 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
10041
10042         * doc/rtl.texi (const_vector): Describe new encoding scheme.
10043         * Makefile.in (OBJS): Add rtx-vector-builder.o.
10044         * rtx-vector-builder.h: New file.
10045         * rtx-vector-builder.c: Likewise.
10046         * rtl.h (rtx_def::u2): Add a const_vector field.
10047         (CONST_VECTOR_NPATTERNS): New macro.
10048         (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
10049         (CONST_VECTOR_DUPLICATE_P): Likewise.
10050         (CONST_VECTOR_STEPPED_P): Likewise.
10051         (CONST_VECTOR_ENCODED_ELT): Likewise.
10052         (const_vec_duplicate_p): Check for a duplicated vector encoding.
10053         (unwrap_const_vec_duplicate): Likewise.
10054         (const_vec_series_p): Check for a non-duplicated vector encoding.
10055         Say that the function only returns true for integer vectors.
10056         * emit-rtl.c: Include rtx-vector-builder.h.
10057         (gen_const_vec_duplicate_1): Delete.
10058         (gen_const_vector): Call gen_const_vec_duplicate instead of
10059         gen_const_vec_duplicate_1.
10060         (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
10061         (gen_const_vec_duplicate): Use rtx_vector_builder.
10062         (gen_const_vec_series): Likewise.
10063         (gen_rtx_CONST_VECTOR): Likewise.
10064         * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
10065         (swap_const_vector_halves): Take an rtx pointer rather than rtx.
10066         Build a new vector rather than modifying a CONST_VECTOR in-place.
10067         (handle_special_swappables): Update call accordingly.
10068         * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
10069         (swap_const_vector_halves): Take an rtx pointer rather than rtx.
10070         Build a new vector rather than modifying a CONST_VECTOR in-place.
10071         (handle_special_swappables): Update call accordingly.
10072
10073 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
10074
10075         * simplify-rtx.c (simplify_const_binary_operation): Use
10076         CONST_VECTOR_ELT instead of XVECEXP.
10077
10078 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
10079
10080         * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
10081         the selector elements to be different from the data elements
10082         if the selector is a VECTOR_CST.
10083         * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
10084         ssizetype for the selector.
10085
10086 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
10087
10088         * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
10089         before testing each element individually.
10090         * tree-vect-generic.c (lower_vec_perm): Likewise.
10091
10092 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
10093
10094         * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
10095         * selftest-run-tests.c (selftest::run_tests): Call it.
10096         * vector-builder.h (vector_builder::operator ==): New function.
10097         (vector_builder::operator !=): Likewise.
10098         * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
10099         (vec_perm_indices::all_from_input_p): New function.
10100         * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
10101         (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
10102         * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
10103         instead of reading the VECTOR_CST directly.  Detect whether both
10104         vector inputs are the same before constructing the vec_perm_indices,
10105         and update the number of inputs argument accordingly.  Use the
10106         utility functions added above.  Only construct sel2 if we need to.
10107
10108 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
10109
10110         * optabs.c (expand_vec_perm_var): Use an explicit encoding for
10111         the broadcast of the low byte.
10112         (expand_mult_highpart): Use an explicit encoding for the permutes.
10113         * optabs-query.c (can_mult_highpart_p): Likewise.
10114         * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
10115         * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
10116         (vectorizable_bswap): Likewise.
10117         * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
10118         explicit encoding for the power-of-2 permutes.
10119         (vect_permute_store_chain): Likewise.
10120         (vect_grouped_load_supported): Likewise.
10121         (vect_permute_load_chain): Likewise.
10122
10123 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
10124
10125         * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
10126         * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
10127         * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
10128         * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
10129         * tree-vect-stmts.c (vectorizable_bswap): Likewise.
10130         (vect_gen_perm_mask_any): Likewise.
10131
10132 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
10133
10134         * int-vector-builder.h: New file.
10135         * vec-perm-indices.h: Include int-vector-builder.h.
10136         (vec_perm_indices): Redefine as an int_vector_builder.
10137         (auto_vec_perm_indices): Delete.
10138         (vec_perm_builder): Redefine as a stand-alone class.
10139         (vec_perm_indices::vec_perm_indices): New function.
10140         (vec_perm_indices::clamp): Likewise.
10141         * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
10142         (vec_perm_indices::new_vector): New function.
10143         (vec_perm_indices::new_expanded_vector): Update for new
10144         vec_perm_indices class.
10145         (vec_perm_indices::rotate_inputs): New function.
10146         (vec_perm_indices::all_in_range_p): Operate directly on the
10147         encoded form, without computing elided elements.
10148         (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
10149         encoding.  Update for new vec_perm_indices class.
10150         * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
10151         the given vec_perm_builder.
10152         (expand_vec_perm_var): Update vec_perm_builder constructor.
10153         (expand_mult_highpart): Use vec_perm_builder instead of
10154         auto_vec_perm_indices.
10155         * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
10156         vec_perm_indices instead of auto_vec_perm_indices.  Use a single
10157         or double series encoding as appropriate.
10158         * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
10159         vec_perm_indices instead of auto_vec_perm_indices.
10160         * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
10161         * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
10162         (vect_permute_store_chain): Likewise.
10163         (vect_grouped_load_supported): Likewise.
10164         (vect_permute_load_chain): Likewise.
10165         (vect_shift_permute_load_chain): Likewise.
10166         * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
10167         (vect_transform_slp_perm_load): Likewise.
10168         (vect_schedule_slp_instance): Likewise.
10169         * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
10170         (vectorizable_mask_load_store): Likewise.
10171         (vectorizable_bswap): Likewise.
10172         (vectorizable_store): Likewise.
10173         (vectorizable_load): Likewise.
10174         * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
10175         vec_perm_indices instead of auto_vec_perm_indices.  Use
10176         tree_to_vec_perm_builder to read the vector from a tree.
10177         * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
10178         vec_perm_builder instead of a vec_perm_indices.
10179         (have_whole_vector_shift): Use vec_perm_builder and
10180         vec_perm_indices instead of auto_vec_perm_indices.  Leave the
10181         truncation to calc_vec_perm_mask_for_shift.
10182         (vect_create_epilog_for_reduction): Likewise.
10183         * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
10184         from auto_vec_perm_indices to vec_perm_indices.
10185         (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
10186         instead of changing individual elements.
10187         (aarch64_vectorize_vec_perm_const): Use new_vector to install
10188         the vector in d.perm.
10189         * config/arm/arm.c (expand_vec_perm_d::perm): Change
10190         from auto_vec_perm_indices to vec_perm_indices.
10191         (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
10192         instead of changing individual elements.
10193         (arm_vectorize_vec_perm_const): Use new_vector to install
10194         the vector in d.perm.
10195         * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
10196         Update vec_perm_builder constructor.
10197         (rs6000_expand_interleave): Likewise.
10198         * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
10199         (rs6000_expand_interleave): Likewise.
10200
10201 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
10202
10203         * optabs-query.c (can_vec_perm_var_p): Check whether lowering
10204         to qimode could truncate the indices.
10205         * optabs.c (expand_vec_perm_var): Likewise.
10206
10207 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
10208
10209         * Makefile.in (OBJS): Add vec-perm-indices.o.
10210         * vec-perm-indices.h: New file.
10211         * vec-perm-indices.c: Likewise.
10212         * target.h (vec_perm_indices): Replace with a forward class
10213         declaration.
10214         (auto_vec_perm_indices): Move to vec-perm-indices.h.
10215         * optabs.h: Include vec-perm-indices.h.
10216         (expand_vec_perm): Delete.
10217         (selector_fits_mode_p, expand_vec_perm_var): Declare.
10218         (expand_vec_perm_const): Declare.
10219         * target.def (vec_perm_const_ok): Replace with...
10220         (vec_perm_const): ...this new hook.
10221         * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
10222         (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
10223         * doc/tm.texi: Regenerate.
10224         * optabs.def (vec_perm_const): Delete.
10225         * doc/md.texi (vec_perm_const): Likewise.
10226         (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
10227         * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
10228         expand_vec_perm for constant permutation vectors.  Assert that
10229         the mode of variable permutation vectors is the integer equivalent
10230         of the mode that is being permuted.
10231         * optabs-query.h (selector_fits_mode_p): Declare.
10232         * optabs-query.c: Include vec-perm-indices.h.
10233         (selector_fits_mode_p): New function.
10234         (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
10235         is defined, instead of checking whether the vec_perm_const_optab
10236         exists.  Use targetm.vectorize.vec_perm_const instead of
10237         targetm.vectorize.vec_perm_const_ok.  Check whether the indices
10238         fit in the vector mode before using a variable permute.
10239         * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
10240         vec_perm_indices instead of an rtx.
10241         (expand_vec_perm): Replace with...
10242         (expand_vec_perm_const): ...this new function.  Take the selector
10243         as a vec_perm_indices rather than an rtx.  Also take the mode of
10244         the selector.  Update call to shift_amt_for_vec_perm_mask.
10245         Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
10246         Use vec_perm_indices::new_expanded_vector to expand the original
10247         selector into bytes.  Check whether the indices fit in the vector
10248         mode before using a variable permute.
10249         (expand_vec_perm_var): Make global.
10250         (expand_mult_highpart): Use expand_vec_perm_const.
10251         * fold-const.c: Includes vec-perm-indices.h.
10252         * tree-ssa-forwprop.c: Likewise.
10253         * tree-vect-data-refs.c: Likewise.
10254         * tree-vect-generic.c: Likewise.
10255         * tree-vect-loop.c: Likewise.
10256         * tree-vect-slp.c: Likewise.
10257         * tree-vect-stmts.c: Likewise.
10258         * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
10259         Delete.
10260         * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
10261         * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
10262         (aarch64_vectorize_vec_perm_const_ok): Fuse into...
10263         (aarch64_vectorize_vec_perm_const): ...this new function.
10264         (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
10265         (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
10266         * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
10267         * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
10268         * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
10269         (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
10270         (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
10271         into...
10272         (arm_vectorize_vec_perm_const): ...this new function.  Explicitly
10273         check for NEON modes.
10274         * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
10275         * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
10276         * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
10277         (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
10278         into...
10279         (ix86_vectorize_vec_perm_const): ...this new function.  Incorporate
10280         the old VEC_PERM_CONST conditions.
10281         * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
10282         * config/ia64/vect.md (vec_perm_const<mode>): Delete.
10283         * config/ia64/ia64.c (ia64_expand_vec_perm_const)
10284         (ia64_vectorize_vec_perm_const_ok): Merge into...
10285         (ia64_vectorize_vec_perm_const): ...this new function.
10286         * config/mips/loongson.md (vec_perm_const<mode>): Delete.
10287         * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
10288         * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
10289         * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
10290         * config/mips/mips.c (mips_expand_vec_perm_const)
10291         (mips_vectorize_vec_perm_const_ok): Merge into...
10292         (mips_vectorize_vec_perm_const): ...this new function.
10293         * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
10294         * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
10295         * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
10296         * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
10297         * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
10298         (rs6000_expand_vec_perm_const): Delete.
10299         * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
10300         Delete.
10301         (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
10302         (altivec_expand_vec_perm_const_le): Take each operand individually.
10303         Operate on constant selectors rather than rtxes.
10304         (altivec_expand_vec_perm_const): Likewise.  Update call to
10305         altivec_expand_vec_perm_const_le.
10306         (rs6000_expand_vec_perm_const): Delete.
10307         (rs6000_vectorize_vec_perm_const_ok): Delete.
10308         (rs6000_vectorize_vec_perm_const): New function.
10309         (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
10310         an element count and rtx array.
10311         (rs6000_expand_extract_even): Update call accordingly.
10312         (rs6000_expand_interleave): Likewise.
10313         * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
10314         * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
10315         * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
10316         * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
10317         (rs6000_expand_vec_perm_const): Delete.
10318         * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
10319         (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
10320         (altivec_expand_vec_perm_const_le): Take each operand individually.
10321         Operate on constant selectors rather than rtxes.
10322         (altivec_expand_vec_perm_const): Likewise.  Update call to
10323         altivec_expand_vec_perm_const_le.
10324         (rs6000_expand_vec_perm_const): Delete.
10325         (rs6000_vectorize_vec_perm_const_ok): Delete.
10326         (rs6000_vectorize_vec_perm_const): New function.  Remove stray
10327         reference to the SPE evmerge intructions.
10328         (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
10329         an element count and rtx array.
10330         (rs6000_expand_extract_even): Update call accordingly.
10331         (rs6000_expand_interleave): Likewise.
10332         * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
10333         * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
10334         new function.
10335         (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
10336
10337 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
10338
10339         * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
10340         vector mode and that that mode matches the mode of the data
10341         being permuted.
10342         (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
10343         out into expand_vec_perm_var.  Do all CONST_VECTOR handling here,
10344         directly using expand_vec_perm_1 when forcing selectors into
10345         registers.
10346         (expand_vec_perm_var): New function, split out from expand_vec_perm.
10347
10348 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
10349
10350         * optabs-query.h (can_vec_perm_p): Delete.
10351         (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
10352         * optabs-query.c (can_vec_perm_p): Split into...
10353         (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
10354         (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
10355         particular selector is valid.
10356         * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
10357         * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
10358         (vect_grouped_load_supported): Likewise.
10359         (vect_shift_permute_load_chain): Likewise.
10360         * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
10361         (vect_transform_slp_perm_load): Likewise.
10362         * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
10363         (vectorizable_bswap): Likewise.
10364         (vect_gen_perm_mask_checked): Likewise.
10365         * fold-const.c (fold_ternary_loc): Likewise.  Don't take
10366         implementations of variable permutation vectors into account
10367         when deciding which selector to use.
10368         * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
10369         vec_perm_const_optab is supported; instead use can_vec_perm_const_p
10370         with a false third argument.
10371         * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
10372         to test whether the constant selector is valid and can_vec_perm_var_p
10373         to test whether a variable selector is valid.
10374
10375 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
10376
10377         * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
10378         * optabs-query.c (can_vec_perm_p): Likewise.
10379         * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
10380         instead of vec_perm_indices.
10381         * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
10382         (vect_gen_perm_mask_checked): Likewise,
10383         * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
10384         (vect_gen_perm_mask_checked): Likewise,
10385
10386 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
10387
10388         * optabs-query.h (qimode_for_vec_perm): Declare.
10389         * optabs-query.c (can_vec_perm_p): Split out qimode search to...
10390         (qimode_for_vec_perm): ...this new function.
10391         * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
10392
10393 2018-01-02  Aaron Sawdey  <acsawdey@linux.vnet.ibm.com>
10394
10395         * rtlanal.c (canonicalize_condition): Return 0 if final rtx
10396         does not have a conditional at the top.
10397
10398 2018-01-02  Richard Biener  <rguenther@suse.de>
10399
10400         * ipa-inline.c (big_speedup_p): Fix expression.
10401
10402 2018-01-02  Jan Hubicka  <hubicka@ucw.cz>
10403
10404         PR target/81616
10405         * config/i386/x86-tune-costs.h: Increase cost of integer load costs
10406         for generic 4->6.
10407
10408 2018-01-02  Jan Hubicka  <hubicka@ucw.cz>
10409
10410         PR target/81616
10411         Generic tuning.
10412         * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
10413         cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
10414         and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
10415         cond_taken_branch_cost 3->4.
10416
10417 2018-01-01  Jakub Jelinek  <jakub@redhat.com>
10418
10419         PR tree-optimization/83581
10420         * tree-loop-distribution.c (pass_loop_distribution::execute): Return
10421         TODO_cleanup_cfg if any changes have been made.
10422
10423         PR middle-end/83608
10424         * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
10425         convert_modes if target mode has the right side, but different mode
10426         class.
10427
10428         PR middle-end/83609
10429         * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
10430         last argument when extracting from CONCAT.  If either from_real or
10431         from_imag is NULL, use expansion through memory.  If result is not
10432         a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
10433         the parts directly to inner mode, if even that fails, use expansion
10434         through memory.
10435
10436         PR middle-end/83623
10437         * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
10438         check for bswap in mode rather than HImode and use that in expand_unop
10439         too.
10440 \f
10441 Copyright (C) 2018 Free Software Foundation, Inc.
10442
10443 Copying and distribution of this file, with or without modification,
10444 are permitted in any medium without royalty provided the copyright
10445 notice and this notice are preserved.