gcc/ChangeLog

   1 2018-05-19  Chung-Ju Wu  <jasonwucj@gmail.com>
   2
   3         * config/nds32/nds32-md-auxiliary.c
   4         (nds32_valid_smw_lwm_base_p): Refine.
   5         (nds32_output_smw_single_word): Refine.
   6         (nds32_output_smw_double_word): New.
   7         * config/nds32/nds32-protos.h (nds32_output_smw_double_word): New.
   8
   9 2018-05-19  Chung-Ju Wu  <jasonwucj@gmail.com>
  10
  11         * config/nds32/nds32-md-auxiliary.c (nds32_output_stack_push): Refine.
  12         (nds32_output_stack_pop): Refine.
  13         (nds32_expand_unaligned_load): Refine.
  14         (nds32_expand_unaligned_store): Refine.
  15
  16 2018-05-19  Kuan-Lin Chen <kuanlinchentw@gmail.com>
  17             Chung-Ju Wu  <jasonwucj@gmail.com>
  18
  19         * config/nds32/constants.md: Add TP_REGNUM constant.
  20         (unspec_element): Add UNSPEC_GOTINIT, UNSPEC_GOT, UNSPEC_GOTOFF,
  21         UNSPEC_PLT, UNSPEC_TLSGD, UNSPEC_TLSLD, UNSPEC_TLSIE, UNSPEC_TLSLE and
  22         UNSPEC_ADD32.
  23         * config/nds32/nds32-doubleword.md: Consider flag_pic.
  24         * config/nds32/nds32-dspext.md (mov<mode>): Expand TLS and PIC cases.
  25         * config/nds32/nds32-predicates.c (nds32_const_unspec_p): New.
  26         * config/nds32/nds32-md-auxiliary.c: Implementation that support TLS
  27         and PIC code generation.
  28         * config/nds32/nds32-protos.h: Declarations that support TLS and PIC
  29         code generation.
  30         * config/nds32/nds32-relax-opt.c: Consider TLS and PIC for relax
  31         optimization.
  32         * config/nds32/nds32.md: Support TLS and PIC.
  33         * config/nds32/nds32.c: Support TLS and PIC.
  34         * config/nds32/nds32.h (nds32_relax_insn_type): New enum type.
  35         * config/nds32/predicates.md (nds32_nonunspec_symbolic_operand): New
  36         predicate.
  37
  38 2018-05-19  Chung-Ju Wu  <jasonwucj@gmail.com>
  39
  40         * config/nds32/nds32-predicates.c (const_vector_to_hwint): Use machine
  41         mode with E_ prefix.
  42
  43 2018-05-19  Kuan-Lin Chen <kuanlinchentw@gmail.com>
  44             Chung-Ju Wu  <jasonwucj@gmail.com>
  45
  46         * config/nds32/constants.md (unspec_element): Add UNSPEC_ICT.
  47         * config/nds32/nds32-md-auxiliary.c
  48         (symbolic_reference_mentioned_p): New.
  49         (nds32_legitimize_ict_address): New.
  50         (nds32_expand_ict_move): New.
  51         (nds32_indirect_call_referenced_p): New.
  52         (nds32_symbol_binds_local_p): Delete.
  53         (nds32_long_call_p): Modify.
  54         * config/nds32/nds32-opts.h (nds32_ict_model_type): New enum type.
  55         * config/nds32/nds32-protos.h
  56         (symbolic_reference_mentioned_p): Declare.
  57         (nds32_legitimize_ict_address): Declare.
  58         (nds32_expand_ict_move): Declare.
  59         (nds32_indirect_call_referenced_p): Declare.
  60         * config/nds32/nds32-relax-opt.c (nds32_ict_const_p): New.
  61         (nds32_relax_group): Use nds32_ict_const_p as condition.
  62         * config/nds32/nds32.c (nds32_attribute_table): Add "indirect_call".
  63         (nds32_asm_file_start): Output ict_model directive in asm code.
  64         (nds32_legitimate_address_p): Consider indirect call.
  65         (nds32_print_operand): Consider indirect call.
  66         (nds32_print_operand_address): Consider indirect call.
  67         (nds32_insert_attributes): Handle "indirect_call" attribute.
  68         (TARGET_LEGITIMATE_ADDRESS_P): Define.
  69         (TARGET_LEGITIMATE_CONSTANT_P): Define.
  70         (TARGET_CANNOT_FORCE_CONST_MEM): Define.
  71         (TARGET_DELEGITIMIZE_ADDRESS): Define.
  72         (TARGET_ASM_OUTPUT_ADDR_CONST_EXTRA): Define.
  73         * config/nds32/nds32.h (SYMBOLIC_CONST_P): Define.
  74         (TARGET_ICT_MODEL_SMALL): Define.
  75         (TARGET_ICT_MODEL_LARGE): Define.
  76         * config/nds32/nds32.md (movsi): Consider ict model.
  77         (call, call_value): Consider ict model.
  78         (sibcall, sibcall_value): Consider ict model.
  79         * config/nds32/nds32.opt (mict-model): New option.
  80         * config/nds32/predicates.md (nds32_symbolic_operand): Consider ict
  81         model.
  82
  83 2018-05-18  Kito Cheng  <kito.cheng@gmail.com>
  84             Monk Chiang  <sh.chiang04@gmail.com>
  85             Jim Wilson <jimw@sifive.com>
  86
  87         * common/config/riscv/riscv-common.c (riscv_parse_arch_string):
  88         Add support to parse rv32e*.  Clear MASK_RVE for rv32i and rv64i.
  89         * config.gcc (riscv*-*-*): Add support for rv32e* and ilp32e.
  90         * config/riscv/riscv-c.c (riscv_cpu_cpp_builtins): Define
  91         __riscv_32e when TARGET_RVE.  Handle ABI_ILP32E as soft-float ABI.
  92         * config/riscv/riscv-opts.h (riscv_abi_type): Add ABI_ILP32E.
  93         * config/riscv/riscv.c (riscv_compute_frame_info): When TARGET_RVE,
  94         compute save_libcall_adjustment properly.
  95         (riscv_option_override): Call error if TARGET_RVE and not ABI_ILP32E.
  96         (riscv_conditional_register_usage): Handle TARGET_RVE and ABI_ILP32E.
  97         * config/riscv/riscv.h (UNITS_PER_FP_ARG): Handle ABI_ILP32E.
  98         (STACK_BOUNDARY, ABI_STACK_BOUNDARY): Handle TARGET_RVE.
  99         (GP_REG_LAST, MAX_ARGS_IN_REGISTERS): Likewise.
 100         (ABI_SPEC): Handle mabi=ilp32e.
 101         * config/riscv/riscv.opt (abi_type): Add ABI_ILP32E.
 102         (RVE): Add RVE mask.
 103         * doc/invoke.texi (RISC-V options) <-mabi>: Add ilp32e info.
 104         <-march>: Add rv32e as an example.
 105
 106 2018-05-18  Marc Glisse  <marc.glisse@inria.fr>
 107
 108         PR c++/82899
 109         * tree-ssa-structalias.c (create_variable_info_for_1): Extra argument.
 110         (intra_create_variable_infos): Handle C++ constructors.
 111
 112 2018-05-18  Martin Liska  <mliska@suse.cz>
 113
 114         * passes.def: Remove a redundant pass.
 115
 116 2018-05-18  Eric Botcazou  <ebotcazou@adacore.com>
 117
 118         PR bootstrap/85838
 119         * config/sparc/sparc.c (sparc_expand_builtin): Always initialize op[0].
 120
 121 2018-05-18  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
 122
 123         * config/arm/arm-cpus.in (armv3m, mode32): Delete features.
 124         (ARMv4): Update.
 125         (ARMv2, ARMv3, ARMv3m): Delete fgroups.
 126         (ARMv6m): Update.
 127         (armv2, armv2a, armv3, armv3m): Delete architectures.
 128         (arm2, arm250, arm3, arm6, arm60, arm600, arm610, arm620,
 129         arm7, arm7d, arm7di, arm70, arm700, arm700i, arm710, arm720,
 130         arm710c, arm7100, arm7500, arm7500fe, arm7m, arm7dm, arm7dmi):
 131         Delete cpus.
 132         * config/arm/arm.md (maddsidi4): Remove check for arm_arch3m.
 133         (*mulsidi3adddi): Likewise.
 134         (mulsidi3): Likewise.
 135         (*mulsidi3_nov6): Likewise.
 136         (umulsidi3): Likewise.
 137         (umulsidi3_nov6): Likewise.
 138         (umaddsidi4): Likewise.
 139         (*umulsidi3adddi): Likewise.
 140         (smulsi3_highpart): Likewise.
 141         (*smulsi3_highpart_nov6): Likewise.
 142         (umulsi3_highpart): Likewise.
 143         (*umulsi3_highpart_nov6): Likewise.
 144         * config/arm/arm.h (arm_arch3m): Delete.
 145         * config/arm/arm.c (arm_arch3m): Delete.
 146         (arm_option_override_internal): Update armv3-related comment.
 147         (arm_configure_build_target): Delete use of isa_bit_mode32.
 148         (arm_option_reconfigure_globals): Delete set of arm_ach3m.
 149         (arm_rtx_costs_internal): Delete check of arm_arch3m.
 150         * config/arm/arm-fixed.md (mulsq3): Delete check for arm_arch3m.
 151         (mulsa3): Likewise.
 152         (mulusa3): Likewise.
 153         * config/arm/arm-protos.h (arm_arch3m): Delete.
 154         * config/arm/arm-tables.opt: Regenerate.
 155         * config/arm/arm-tune.md: Likewise.
 156         * config/arm/t-arm-elf (all_early_nofp): Delete mentions of
 157         deleted architectures.
 158
 159 2018-05-18  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
 160
 161         * config/arm/arm-cpus.in (armv5, armv5e): Delete features.
 162         (armv5t, armv5te): New features.
 163         (ARMv5, ARMv5e): Delete fgroups.
 164         (ARMv5t, ARMv5te): Adjust for above changes.
 165         (ARMv6m): Likewise.
 166         (armv5, armv5e): Delete arches.
 167         * config/arm/arm.md (*call_reg_armv5): Use arm_arch5t instead of
 168         arm_arch5.
 169         (*call_reg_arm): Likewise.
 170         (*call_value_reg_armv5): Likewise.
 171         (*call_value_reg_arm): Likewise.
 172         (*call_symbol): Likewise.
 173         (*call_value_symbol): Likewise.
 174         (*sibcall_insn): Likewise.
 175         (*sibcall_value_insn): Likewise.
 176         (clzsi2): Likewise.
 177         (prefetch): Likewise.
 178         (define_split and define_peephole2 dependent on arm_arch5):
 179         Likewise.
 180         * config/arm/arm.h (TARGET_LDRD): Use arm_arch5te instead of
 181         arm_arch5e.
 182         (TARGET_ARM_QBIT): Likewise.
 183         (TARGET_DSP_MULTIPLY): Likewise.
 184         (enum base_architecture): Delete BASE_ARCH_5, BASE_ARCH_5E.
 185         (arm_arch5, arm_arch5e): Delete.
 186         (arm_arch5t, arm_arch5te): Declare.
 187         * config/arm/arm.c (arm_arch5, arm_arch5e): Delete.
 188         (arm_arch5t): Declare.
 189         (arm_option_reconfigure_globals): Update for the above.
 190         (arm_options_perform_arch_sanity_checks): Update comment, replace
 191         use of arm_arch5 with arm_arch5t.
 192         (use_return_insn): Likewise.
 193         (arm_emit_call_insn): Likewise.
 194         (output_return_instruction): Likewise.
 195         (arm_final_prescan_insn): Likewise.
 196         (arm_coproc_builtin_available): Likewise.
 197         * config/arm/arm-c.c (arm_cpu_builtins): Replace arm_arch5 and
 198         arm_arch5e with arm_arch5t and arm_arch5te.
 199         * config/arm/arm-protos.h (arm_arch5, arm_arch5e): Delete.
 200         (arm_arch5t, arm_arch5te): Declare.
 201         * config/arm/arm-tables.opt: Regenerate.
 202         * config/arm/t-arm-elf: Remove references to armv5, armv5e.
 203         * config/arm/t-multilib: Likewise.
 204         * config/arm/thumb1.md (*call_reg_thumb1_v5): Check arm_arch5t
 205         instead of arm_arch5.
 206         (*call_reg_thumb1): Likewise.
 207         (*call_value_reg_thumb1_v5): Likewise.
 208         (*call_value_reg_thumb1): Likewise.
 209         * config/arm/vxworks.h (TARGET_OS_CPP_BUILTINS): Remove now
 210         unreachable path.
 211         * doc/invoke.texi (ARM Options): Remove references to armv5, armv5e.
 212
 213 2018-05-18  Martin Liska  <mliska@suse.cz>
 214
 215         PR gcov-profile/84846
 216         * doc/gcov.texi: Document -t option of gcov tool.
 217
 218 2018-05-18  Martin Liska  <mliska@suse.cz>
 219
 220         PR gcov-profile/84846
 221         * gcov.c (print_usage): Add new -t option.
 222         (process_args): Handle the option.
 223         (generate_results): Use stdout as output when requested by
 224         the option.
 225
 226 2018-05-18  Martin Liska  <mliska@suse.cz>
 227
 228         PR gcov-profile/84846
 229         * coverage.c (coverage_init): Write PWD to .gcno file.
 230         * doc/gcov.texi: Document how working directory is printed.
 231         * gcov-dump.c (dump_gcov_file): Print PWD.
 232         * gcov.c (output_intermediate_file): Likewise.
 233         (read_graph_file): Read PWD string.
 234         (output_lines): Print PWD.
 235
 236 2018-05-18  Prathamesh Kulkarni  <prathamesh.kulkarni@linaro.org>
 237
 238         PR middle-end/85817
 239         * ipa-pure-const.c (malloc_candidate_p): Remove the check integer_zerop
 240         for retval and return false if all args to phi are zero.
 241
 242 2018-05-18  Richard Biener  <rguenther@suse.de>
 243
 244         * gimple-ssa-evrp.c (class evrp_folder): Add simplify_stmt_using_ranges
 245         method.
 246         (evrp_dom_walker::before_dom_children): Call it.
 247
 248 2018-05-18  Richard Biener  <rguenther@suse.de>
 249
 250         * tree-dfa.c (get_ref_base_and_extent): Use range-info to refine
 251         results when processing array refs with variable index.
 252
 253 2018-05-18  Toon Moene  <toon@moene.org>
 254
 255         * doc/invoke.texi: Move -floop-unroll-and-jam documentation
 256         directly after that of -floop-interchange. Indicate that both
 257         options are enabled by default when specifying -O3.
 258
 259 2018-05-18  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
 260
 261         * config/aarch64/aarch64-simd.md (vec_set<mode>): Use VALL_F16 mode
 262         iterator.  Delete separate integer-mode vec_set<mode> expander.
 263         (aarch64_simd_vec_setv2di): Delete.
 264         (vec_setv2di): Delete.
 265         (aarch64_simd_vec_set<mode>): Delete all other patterns with that name.
 266         Use VALL_F16 mode iterator.  Add LD1 alternative and use vwcore for
 267         the "w, r" alternative.
 268
 269 2018-05-18  Martin Liska  <mliska@suse.cz>
 270
 271         * passes.def: Add pass_lower_switch and pass_lower_switch_O0.
 272         * tree-pass.h (make_pass_lower_switch_O0): New function.
 273         * tree-switch-conversion.c (node_has_low_bound): Remove.
 274         (node_has_high_bound): Likewise.
 275         (node_is_bounded): Likewise.
 276         (class pass_lower_switch): Make it a template type and create
 277         two instances.
 278         (pass_lower_switch::execute): Add template argument.
 279         (make_pass_lower_switch): New function.
 280         (make_pass_lower_switch_O0): New function.
 281         (do_jump_if_equal): Remove.
 282         (emit_case_nodes): Simplify to just handle all 3 cases and leave
 283         all the hard work to tree optimization passes.
 284
 285 2018-05-18  Martin Liska  <mliska@suse.cz>
 286
 287         * dbgcnt.c (limit_low): Renamed from limit.
 288         (limit_high): New variable.
 289         (dbg_cnt_is_enabled): Check for upper limit.
 290         (dbg_cnt): Adjust dumping.
 291         (dbg_cnt_set_limit_by_index): Add new argument for high
 292         value.
 293         (dbg_cnt_set_limit_by_name): Likewise.
 294         (dbg_cnt_process_single_pair): Parse new format.
 295         (dbg_cnt_process_opt): Use strtok.
 296         (dbg_cnt_list_all_counters): Remove 'value' and add
 297         'limit_high'.
 298         * doc/invoke.texi: Document changes.
 299
 300 2018-05-18  Richard Sandiford  <richard.sandiford@linaro.org>
 301
 302         * doc/sourcebuild.texi (scalar_all_fma): Document.
 303         * tree.def (FMA_EXPR): Delete.
 304         * internal-fn.def (FMA, FMS, FNMA, FNMS): New internal functions.
 305         * internal-fn.c (ternary_direct): New macro.
 306         (expand_ternary_optab_fn): Likewise.
 307         (direct_ternary_optab_supported_p): Likewise.
 308         * Makefile.in (build/genmatch.o): Depend on case-fn-macros.h.
 309         * builtins.c (fold_builtin_fma): Delete.
 310         (fold_builtin_3): Don't call it.
 311         * cfgexpand.c (expand_debug_expr): Remove FMA_EXPR handling.
 312         * expr.c (expand_expr_real_2): Likewise.
 313         * fold-const.c (operand_equal_p): Likewise.
 314         (fold_ternary_loc): Likewise.
 315         * gimple-pretty-print.c (dump_ternary_rhs): Likewise.
 316         * gimple.c (DEFTREECODE): Likewise.
 317         * gimplify.c (gimplify_expr): Likewise.
 318         * optabs-tree.c (optab_for_tree_code): Likewise.
 319         * tree-cfg.c (verify_gimple_assign_ternary): Likewise.
 320         * tree-eh.c (operation_could_trap_p): Likewise.
 321         (stmt_could_throw_1_p): Likewise.
 322         * tree-inline.c (estimate_operator_cost): Likewise.
 323         * tree-pretty-print.c (dump_generic_node): Likewise.
 324         (op_code_prio): Likewise.
 325         * tree-ssa-loop-im.c (stmt_cost): Likewise.
 326         * tree-ssa-operands.c (get_expr_operands): Likewise.
 327         * tree.c (commutative_ternary_tree_code, add_expr): Likewise.
 328         * fold-const-call.h (fold_fma): Delete.
 329         * fold-const-call.c (fold_const_call_ssss): Handle CFN_FMS,
 330         CFN_FNMA and CFN_FNMS.
 331         (fold_fma): Delete.
 332         * genmatch.c (combined_fn): New enum.
 333         (commutative_ternary_tree_code): Remove FMA_EXPR handling.
 334         (commutative_op): New function.
 335         (commutate): Use it.  Handle more than 2 operands.
 336         (dt_operand::gen_gimple_expr): Use commutative_op.
 337         (parser::parse_expr): Allow :c to be used with non-binary
 338         operators if the commutative operand is known.
 339         * gimple-ssa-backprop.c (backprop::process_builtin_call_use): Handle
 340         CFN_FMS, CFN_FNMA and CFN_FNMS.
 341         (backprop::process_assign_use): Remove FMA_EXPR handling.
 342         * hsa-gen.c (gen_hsa_insns_for_operation_assignment): Likewise.
 343         (gen_hsa_fma): New function.
 344         (gen_hsa_insn_for_internal_fn_call): Use it for IFN_FMA, IFN_FMS,
 345         IFN_FNMA and IFN_FNMS.
 346         * match.pd: Add folds for IFN_FMS, IFN_FNMA and IFN_FNMS.
 347         * gimple-fold.h (follow_all_ssa_edges): Declare.
 348         * gimple-fold.c (follow_all_ssa_edges): New function.
 349         * tree-ssa-math-opts.c (convert_mult_to_fma_1): Use the
 350         gimple_build interface and use follow_all_ssa_edges to fold the result.
 351         (convert_mult_to_fma): Use direct_internal_fn_suppoerted_p
 352         instead of checking for optabs directly.
 353         * config/i386/i386.c (ix86_add_stmt_cost): Recognize FMAs as calls
 354         rather than FMA_EXPRs.
 355         * config/rs6000/rs6000.c (rs6000_gimple_fold_builtin): Create a
 356         call to IFN_FMA instead of an FMA_EXPR.
 357
 358 2018-05-17  Jim Wilson  <jimw@sifive.com>
 359
 360         * expr.c (do_tablejump): When converting index to Pmode, if we have a
 361         sign extended promoted subreg, and the range does not have the sign bit
 362         set, then do a sign extend.
 363
 364         * config/riscv/riscv.c (riscv_extend_comparands): In unsigned QImode
 365         test, check for sign extended subreg and/or constant operands, and
 366         do a sign extend in that case.
 367
 368 2018-05-17  Steve Ellcey  <sellcey@cavium.com>
 369
 370         * config/aarch64/thunderx2t99.md (thunderx2t99_ls_both): Delete.
 371         (thunderx2t99_multiple): Delete psuedo-units from used cpus.
 372         Add untyped.
 373         (thunderx2t99_alu_shift): Remove alu_shift_reg, alus_shift_reg.
 374         Change logics_shift_reg to logics_shift_imm.
 375         (thunderx2t99_fp_loadpair_basic): Delete.
 376         (thunderx2t99_fp_storepair_basic): Delete.
 377         (thunderx2t99_asimd_int): Add neon_sub and neon_sub_q types.
 378         (thunderx2t99_asimd_polynomial): Delete.
 379         (thunderx2t99_asimd_fp_simple): Add neon_fp_mul_s_scalar_q
 380         and neon_fp_mul_d_scalar_q.
 381         (thunderx2t99_asimd_fp_conv): Add *int_to_fp* types.
 382         (thunderx2t99_asimd_misc): Delete neon_dup and neon_dup_q.
 383         (thunderx2t99_asimd_recip_step): Add missing *sqrt* types.
 384         (thunderx2t99_asimd_lut): Add missing tbl types.
 385         (thunderx2t99_asimd_ext): Delete.
 386         (thunderx2t99_asimd_load1_1_mult): Delete.
 387         (thunderx2t99_asimd_load1_2_mult): Delete.
 388         (thunderx2t99_asimd_load1_ldp): New.
 389         (thunderx2t99_asimd_load1): New.
 390         (thunderx2t99_asimd_load2): Add missing *load2* types.
 391         (thunderx2t99_asimd_load3): New.
 392         (thunderx2t99_asimd_load4): New.
 393         (thunderx2t99_asimd_store1_1_mult): Delete.
 394         (thunderx2t99_asimd_store1_2_mult): Delete.
 395         (thunderx2t99_asimd_store2_mult): Delete.
 396         (thunderx2t99_asimd_store2_onelane): Delete.
 397         (thunderx2t99_asimd_store_stp): New.
 398         (thunderx2t99_asimd_store1): New.
 399         (thunderx2t99_asimd_store2): New.
 400         (thunderx2t99_asimd_store3): New.
 401         (thunderx2t99_asimd_store4): New.
 402
 403 2018-05-17  Jerome Lambourg  <lambourg@adacore.com>
 404
 405         * config/arm/arm_cmse.h (cmse_nsfptr_create, cmse_is_nsfptr): Remove
 406         #include <stdint.h>.  Replace intptr_t with __INTPTR_TYPE__.
 407
 408 2018-05-17  Pat Haugen  <pthaugen@us.ibm.com>
 409             Segher Boessenkool  <segher@kernel.crashing.org>
 410
 411         PR target/85698
 412         * config/rs6000/rs6000.c (rs6000_output_move_128bit): Check dest
 413         operand.
 414
 415 2018-05-17  Richard Biener  <rguenther@suse.de>
 416
 417         * tree-ssa-dse.c (dse_classify_store): Fix iterator increment
 418         for pruning loop and prune defs feeding only already visited PHIs.
 419
 420 2018-05-17  Richard Biener  <rguenther@suse.de>
 421
 422         * tree-ssa-sccvn.c (vn_reference_lookup_3): Improve memset handling.
 423
 424 2018-05-17  Bin Cheng  <bin.cheng@arm.com>
 425             Richard Biener  <rguenther@suse.de>
 426
 427         PR tree-optimization/85793
 428         * tree-vect-stmts.c (vectorizable_load): Handle 1 element-wise load
 429         for VMAT_ELEMENTWISE.
 430
 431 2018-05-17  Richard Sandiford  <richard.sandiford@linaro.org>
 432
 433         * internal-fn.h (lookup_internal_fn): Declare
 434         * internal-fn.c (lookup_internal_fn): New function.
 435         * gimple.c (gimple_build_call_from_tree): Handle calls to
 436         internal functions.
 437         * gimple-pretty-print.c (dump_gimple_call): Print "." before
 438         internal function names.
 439         * tree-pretty-print.c (dump_generic_node): Likewise.
 440         * tree-ssa-scopedtables.c (expr_hash_elt::print): Likewise.
 441
 442 2018-05-17  Richard Sandiford  <richard.sandiford@linaro.org>
 443
 444         * gimple-fold.h (gimple_build): Make the function forms take
 445         combined_fn rather than built_in_function.
 446         (gimple_simplify): Likewise.
 447         * gimple-match-head.c (gimple_simplify): Likewise.
 448         * gimple-fold.c (gimple_build): Likewise.
 449         * tree-vect-loop.c (get_initial_def_for_reduction): Use gimple_build
 450         rather than gimple_build_call_internal.
 451         (get_initial_defs_for_reduction): Likewise.
 452         (vect_create_epilog_for_reduction): Likewise.
 453         (vectorizable_live_operation): Likewise.
 454
 455 2018-05-17  Martin Liska  <mliska@suse.cz>
 456
 457         * gimple-ssa-sprintf.c (format_directive): Do not use
 458         space in between 'G_' and '('.
 459
 460 2018-05-17  Jakub Jelinek  <jakub@redhat.com>
 461
 462         PR target/85323
 463         * config/i386/i386.c (ix86_fold_builtin): Handle masked shifts
 464         even if the mask is not all ones.
 465
 466         PR target/85323
 467         * config/i386/i386.c (ix86_fold_builtin): Fold shift builtins by
 468         vector.
 469         (ix86_gimple_fold_builtin): Likewise.
 470
 471         PR target/85323
 472         * config/i386/i386.c: Include tree-vector-builder.h.
 473         (ix86_vector_shift_count): New function.
 474         (ix86_fold_builtin): Fold shift builtins by scalar count.
 475         (ix86_gimple_fold_builtin): Likewise.
 476
 477         * config/i386/avx512fintrin.h (_mm512_set_epi16, _mm512_set_epi8,
 478         _mm512_setzero): New intrinsics.
 479
 480 2018-05-17  James Greenhalgh  <james.greenhalgh@arm.com>
 481             Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
 482
 483         * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify
 484         code generation for cases where splatting a value is not useful.
 485         * simplify-rtx.c (simplify_ternary_operation): Simplify
 486         vec_merge across a vec_duplicate and a paradoxical subreg forming a vector
 487         mode to a vec_concat.
 488
 489 2018-05-17  Olga Makhotina  <olga.makhotina@intel.com>
 490
 491         * config.gcc: Support "goldmont-plus".
 492         * config/i386/driver-i386.c (host_detect_local_cpu): Detect
 493         "goldmont-plus".
 494         * config/i386/i386-c.c (ix86_target_macros_internal): Handle
 495         PROCESSOR_GOLDMONT_PLUS.
 496         * config/i386/i386.c (m_GOLDMONT_PLUS): Define.
 497         (processor_target_table): Add "goldmont-plus".
 498         (PTA_GOLDMONT_PLUS): Define.
 499         (ix86_lea_outperforms): Add TARGET_GOLDMONT_PLUS.
 500         (get_builtin_code_for_version): Handle PROCESSOR_GOLDMONT_PLUS.
 501         (fold_builtin_cpu): Add M_INTEL_GOLDMONT_PLUS.
 502         (fold_builtin_cpu): Add "goldmont-plus".
 503         (ix86_add_stmt_cost): Add TARGET_GOLDMONT_PLUS.
 504         (ix86_option_override_internal): Add "goldmont-plus".
 505         * config/i386/i386.h (processor_costs): Define TARGET_GOLDMONT_PLUS.
 506         (processor_type): Add PROCESSOR_GOLDMONT_PLUS.
 507         * config/i386/x86-tune.def: Add m_GOLDMONT_PLUS.
 508         * doc/invoke.texi: Add goldmont-plus as x86 -march=/-mtune= CPU type.
 509
 510 2018-05-17  Richard Biener  <rguenther@suse.de>
 511
 512         PR tree-optimization/85757
 513         * tree-ssa-dse.c (dse_classify_store): Record a PHI def and
 514         remove defs that only feed that PHI from further processing.
 515
 516 2018-05-16  Jim Wilson  <jimw@sifive.com>
 517
 518         * config/riscv/riscv.md (<optab>si3_mask, <optab>si3_mask_1): Prepend
 519         asterisk to name.
 520         (<optab>di3_mask, <optab>di3_mask_1): Likewise.
 521
 522 2018-05-16  Mark Wielaard  <mark@klomp.org>
 523
 524         * dwarf2out.c (count_index_strings): New function.
 525         (output_indirect_strings): Call count_index_strings and generate
 526         header for dwarf_version >= 5.
 527
 528 2018-05-16  Mark Wielaard  <mark@klomp.org>
 529
 530         * dwarf2out.c (dwarf_FORM): New function.
 531         (set_indirect_string): Use dwarf_FORM.
 532         (reset_indirect_string): Likewise.
 533         (size_of_die): Likewise.
 534         (value_format): Likewise.
 535         (output_die): Likewise.
 536         (add_skeleton_AT_string): Likewise.
 537         (output_macinfo_op): Likewise.
 538         (index_string): Likewise.
 539         (output_index_string_offset): Likewise.
 540         (output_index_string): Likewise.
 541         (count_index_strings): Likewise.
 542
 543 2018-05-16  Carl Love  <cel@us.ibm.com>
 544
 545         * config/rs6000/rs6000.md (prefetch): Generate ISA 2.06 instructions
 546         dcbt and dcbtstt with TH=16 if operands[2] is 0 and Power 8 or newer.
 547
 548 2018-05-16  Martin Jambor  <mjambor@suse.cz>
 549
 550         * ipa-prop.c (ipa_free_all_edge_args): Remove.
 551         * ipa-prop.h (ipa_free_all_edge_args): Likewise.
 552
 553 2018-05-16  Wilco Dijkstra  <wdijkstr@arm.com>
 554
 555         * config/aarch64/aarch64.md (fma<mode>4): Change into expand pattern.
 556         (fnma<mode>4): Likewise.
 557         (fms<mode>4): Likewise.
 558         (fnms<mode>4): Likewise.
 559         (aarch64_fma<mode>4): Rename insn, reorder accumulator operand.
 560         (aarch64_fnma<mode>4): Likewise.
 561         (aarch64_fms<mode>4): Likewise.
 562         (aarch64_fnms<mode>4): Likewise.
 563         (aarch64_fnmadd<mode>4): Likewise.
 564
 565 2018-05-16  Jason Merrill  <jason@redhat.com>
 566
 567         * tree.c (warn_deprecated_use): Return bool.  Simplify logic.
 568
 569 2018-05-16  Richard Biener  <rguenther@suse.de>
 570
 571         * tree-vectorizer.h (struct stmt_info_for_cost): Add where member.
 572         (dump_stmt_cost): Declare.
 573         (add_stmt_cost): Dump cost we add.
 574         (add_stmt_costs): New function.
 575         (vect_model_simple_cost, vect_model_store_cost, vect_model_load_cost):
 576         No longer exported.
 577         (vect_analyze_stmt): Adjust prototype.
 578         (vectorizable_condition): Likewise.
 579         (vectorizable_live_operation): Likewise.
 580         (vectorizable_reduction): Likewise.
 581         (vectorizable_induction): Likewise.
 582         * tree-vect-loop.c (vect_analyze_loop_operations): Create local
 583         cost vector to pass to vectorizable_ and record afterwards.
 584         (vect_model_reduction_cost): Take cost vector argument and adjust.
 585         (vect_model_induction_cost): Likewise.
 586         (vectorizable_reduction): Likewise.
 587         (vectorizable_induction): Likewise.
 588         (vectorizable_live_operation): Likewise.
 589         * tree-vect-slp.c (vect_create_new_slp_node): Initialize
 590         SLP_TREE_NUMBER_OF_VEC_STMTS.
 591         (vect_analyze_slp_cost_1): Remove.
 592         (vect_analyze_slp_cost): Likewise.
 593         (vect_slp_analyze_node_operations): Take visited args and
 594         a target cost vector.  Avoid processing already visited stmt sets.
 595         (vect_slp_analyze_operations): Use a local cost vector to gather
 596         costs and register those of non-discarded instances.
 597         (vect_bb_vectorization_profitable_p): Use add_stmt_costs.
 598         (vect_schedule_slp_instance): Remove copying of
 599         SLP_TREE_NUMBER_OF_VEC_STMTS.  Instead assert that it is not
 600         zero.
 601         * tree-vect-stmts.c (record_stmt_cost): Remove path directly
 602         adding cost.  Record cost entry location.
 603         (vect_prologue_cost_for_slp_op): Function to compute cost of
 604         a constant or invariant generated for SLP vect in the prologue,
 605         split out from vect_analyze_slp_cost_1.
 606         (vect_model_simple_cost): Make static.  Adjust for SLP costing.
 607         (vect_model_promotion_demotion_cost): Likewise.
 608         (vect_model_store_cost): Likewise, make static.
 609         (vect_model_load_cost): Likewise.
 610         (vectorizable_bswap): Add cost vector arg and adjust.
 611         (vectorizable_call): Likewise.
 612         (vectorizable_simd_clone_call): Likewise.
 613         (vectorizable_conversion): Likewise.
 614         (vectorizable_assignment): Likewise.
 615         (vectorizable_shift): Likewise.
 616         (vectorizable_operation): Likewise.
 617         (vectorizable_store): Likewise.
 618         (vectorizable_load): Likewise.
 619         (vectorizable_condition): Likewise.
 620         (vectorizable_comparison): Likewise.
 621         (can_vectorize_live_stmts): Likewise.
 622         (vect_analyze_stmt): Likewise.
 623         (vect_transform_stmt): Adjust calls to vectorizable_*.
 624         * tree-vectorizer.c: Include gimple-pretty-print.h.
 625         (dump_stmt_cost): New function.
 626
 627 2018-05-16  Richard Biener  <rguenther@suse.de>
 628
 629         * params.def (PARAM_DSE_MAX_ALIAS_QUERIES_PER_STORE): New param.
 630         * doc/invoke.texi (dse-max-alias-queries-per-store): Document.
 631         * tree-ssa-dse.c: Include tree-ssa-loop.h.
 632         (check_name): New callback.
 633         (dse_classify_store): Track cycles via a visited bitmap of PHI
 634         defs and simplify handling of in-loop and across loop dead stores
 635         and properly fail for loop-variant refs.  Handle byte-tracking with
 636         multiple defs.  Use PARAM_DSE_MAX_ALIAS_QUERIES_PER_STORE for
 637         limiting the walk.
 638
 639 2018-05-16  Richard Sandiford  <richard.sandiford@linaro.org>
 640
 641         * tree-vectorizer.h (vect_get_vector_types_for_stmt): Declare.
 642         (vect_get_mask_type_for_stmt): Likewise.
 643         * tree-vect-slp.c (vect_two_operations_perm_ok_p): New function,
 644         split out from...
 645         (vect_build_slp_tree_1): ...here.  Use vect_get_vector_types_for_stmt
 646         to determine the statement's vector type and the vector type that
 647         should be used for calculating nunits.  Deal with cases in which
 648         the type has to be deferred.
 649         (vect_slp_analyze_node_operations): Use vect_get_vector_types_for_stmt
 650         and vect_get_mask_type_for_stmt to calculate STMT_VINFO_VECTYPE.
 651         * tree-vect-loop.c (vect_determine_vf_for_stmt_1)
 652         (vect_determine_vf_for_stmt): New functions, split out from...
 653         (vect_determine_vectorization_factor): ...here.
 654         * tree-vect-stmts.c (vect_get_vector_types_for_stmt)
 655         (vect_get_mask_type_for_stmt): New functions, split out from
 656         vect_determine_vectorization_factor.
 657
 658 2018-05-16  Richard Biener  <rguenther@suse.de>
 659
 660         * tree-cfg.c (verify_gimple_assign_ternary): Properly
 661         verify the [VEC_]COND_EXPR embedded comparison.
 662
 663 2018-05-15  Martin Sebor  <msebor@redhat.com>
 664
 665         PR tree-optimization/85753
 666         * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Handle
 667         RECORD_TYPE in addition to ARRAY_TYPE.
 668
 669 2018-05-15  Martin Sebor  <msebor@redhat.com>
 670
 671         PR middle-end/85643
 672         * calls.c (get_attr_nonstring_decl): Handle MEM_REF.
 673
 674 2018-05-15  Richard Biener  <rguenther@suse.de>
 675
 676         * tree-ssa-dse.c (dse_classify_store): Remove use_stmt parameter,
 677         add by_clobber_p one.  Change algorithm to collect all defs
 678         representing uses we need to walk and try reducing them to
 679         a single one before failing.
 680         (dse_dom_walker::dse_optimize_stmt): Adjust.
 681
 682 2018-05-13  Mark Wielaard  <mark@klomp.org>
 683
 684         * dwarf2out.c (dwarf_OP): Handle DW_OP_addrx and DW_OP_constx.
 685         (size_of_loc_descr): Likewise.
 686         (output_loc_operands): Likewise.
 687         (output_loc_operands_raw): Likewise.
 688         (dw_addr_op): Use dwarf_OP () for DW_OP_constx and DW_OP_addrx.
 689         (resolve_addr_in_expr): Handle DW_OP_addrx and DW_OP_constx.
 690         (hash_loc_operands): Likewise.
 691         (compare_loc_operands): Likewise.
 692
 693 2018-05-14  Mark Wielaard  <mark@klomp.org>
 694
 695         * dwarf2out.c (count_index_addrs): New function.
 696         (dwarf2out_finish): Use count_index_addrs to calculate addrs_length.
 697
 698 2018-05-15  Prathamesh Kulkarni  <prathamesh.kulkarni@linaro.org>
 699
 700         PR tree-optimization/83648
 701         * ipa-pure-const.c (malloc_candidate_p): Allow function with NULL
 702         return value as malloc candidate.
 703
 704 2018-05-15  Prathamesh Kulkarni  <prathamesh.kulkarni@linaro.org>
 705
 706         PR ipa/85734
 707         * ipa-pure-const.c (warn_function_malloc): Pass value of known_finite param
 708         as true in call to suggest_attribute.
 709
 710 2018-05-14  Segher Boessenkool  <segher@kernel.crashing.org>
 711
 712         * doc/invoke.texi (RS/6000 and PowerPC Options): Delete @itemx for
 713         -mreadonly-in-sdata.
 714
 715 2018-05-14  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
 716
 717         * config/aarch64/aarch64-simd.md (*aarch64_crypto_aese_fused):
 718         New pattern.
 719         (aarch64_crypto_aesd_fused): Likewise.
 720
 721 2018-05-14  Wilco Dijkstra  <wdijkstr@arm.com>
 722
 723         * config/aarch64/aarch64.md (mov<mode>): Remove '*' in alternatives.
 724         (movsi_aarch64): Likewise.
 725         (load_pairsi): Likewise.
 726         (load_pairdi): Likewise.
 727         (store_pairsi): Likewise.
 728         (store_pairdi): Likewise.
 729         (load_pairsf): Likewise.
 730         (load_pairdf): Likewise.
 731         (store_pairsf): Likewise.
 732         (store_pairdf): Likewise.
 733         (zero_extend): Likewise.
 734         (trunc): Swap alternatives.
 735         (fcvt_target): Add '?' to prefer w over r.
 736
 737 2018-05-14  Jakub Jelinek  <jakub@redhat.com>
 738
 739         PR target/85756
 740         * config/i386/i386.md: Disallow non-commutative arithmetics in
 741         last twpeephole for mem {+,-,&,|,^}= x; mem != 0 after cmpelim
 742         optimization.  Use COMMUTATIVE_ARITH_P test rather than != MINUS
 743         in the peephole2 before it.
 744
 745 2018-05-14  Sebastian Peryt  <sebastian.peryt@intel.com>
 746
 747         * common/config/i386/i386-common.c (OPTION_MASK_ISA_CLDEMOTE_SET,
 748         OPTION_MASK_ISA_CLDEMOTE_UNSET): New defines.
 749         (ix86_handle_option): Handle -mcldemote.
 750         * config.gcc: New header.
 751         * config/i386/cldemoteintrin.h: New file.
 752         * config/i386/cpuid.h (bit_CLDEMOTE): New bit.
 753         * config/i386/driver-i386.c (host_detect_local_cpu): Detect
 754         -mcldemote.
 755         * config/i386/i386-c.c (ix86_target_macros_internal): Handle
 756         OPTION_MASK_ISA_CLDEMOTE.
 757         * config/i386/i386.c (ix86_target_string): Add -mcldemote.
 758         (ix86_valid_target_attribute_inner_p): Ditto.
 759         (enum ix86_builtins): Add IX86_BUILTIN_CLDEMOTE.
 760         (ix86_init_mmx_sse_builtins): Define __builtin_ia32_cldemote.
 761         (ix86_expand_builtin): Expand IX86_BUILTIN_CLDEMOTE.
 762         * config/i386/i386.h (TARGET_CLDEMOTE, TARGET_CLDEMOTE_P): New.
 763         * config/i386/i386.md (UNSPECV_CLDEMOTE): New.
 764         (cldemote): New.
 765         * config/i386/i386.opt: Add -mcldemote.
 766         * config/i386/x86intrin.h: New header.
 767         * doc/invoke.texi: Add -mcldemote.
 768
 769 2018-05-14  Richard Biener  <rguenther@suse.de>
 770
 771         * doc/match-and-simplify.texi: Adjust :s documentation.
 772
 773 2018-05-14  Alexander Monakov  <amonakov@ispras.ru>
 774
 775         * sort.cc (REORDER_23): Pass the type for the temporaries instead of
 776         intended memcpy size.
 777         (REORDER_45): Likewise.
 778
 779 2018-05-13  Alexander Monakov  <amonakov@ispras.ru>
 780
 781         * sort.cc: New file.
 782         * system.h [!CHECKING_P] (qsort): Redirect to gcc_qsort.
 783         * vec.c (qsort_chk): Use gcc_qsort.
 784         * Makefile.in (OBJS-libcommon): Add sort.o.
 785         (build/sort.o): New target.  Use it...
 786         (BUILD_RTL): ... here, and...
 787         (build/gencfn-macros): ... here, and...
 788         (build/genmatch): ... here.
 789
 790 2018-05-13  Kito Cheng  <kito.cheng@gmail.com>
 791             Chung-Ju Wu  <jasonwucj@gmail.com>
 792
 793         * config.gcc (nds32*-*-*): Check that n15 is valid to --with-cpu.
 794         * config/nds32/nds32-graywolf.md: New file.
 795         * config/nds32/nds32-opts.h (nds32_cpu_type): Add CPU_GRAYWOLF.
 796         * config/nds32/nds32-pipelines-auxiliary.c: Implementation for n15
 797         pipeline.
 798         * config/nds32/nds32-protos.h: More declarations for n15 pipeline.
 799         * config/nds32/nds32-utils.c: More implementations for n15 pipeline.
 800         * config/nds32/nds32.md (pipeline_model): Add graywolf.
 801         * config/nds32/nds32.opt (mcpu): Support n15 pipeline cpus.
 802         * config/nds32/pipelines.md: Include n15 settings.
 803
 804 2018-05-13  Kito Cheng  <kito.cheng@gmail.com>
 805             Chung-Ju Wu  <jasonwucj@gmail.com>
 806
 807         * config.gcc (nds32*-*-*): Check that n12/n13 are valid to --with-cpu.
 808         * config/nds32/nds32-n13.md: New file.
 809         * config/nds32/nds32-opts.h (nds32_cpu_type): Add CPU_N12 and CPU_N13.
 810         * config/nds32/nds32-pipelines-auxiliary.c: Implementation for n12/n13
 811         pipeline.
 812         * config/nds32/nds32-protos.h: More declarations for n12/n13 pipeline.
 813         * config/nds32/nds32.md (pipeline_model): Add n13.
 814         * config/nds32/nds32.opt (mcpu): Support n12/n13 pipeline cpus.
 815         * config/nds32/pipelines.md: Include n13 settings.
 816
 817 2018-05-13  Kito Cheng  <kito.cheng@gmail.com>
 818             Chung-Ju Wu  <jasonwucj@gmail.com>
 819
 820         * config.gcc (nds32*-*-*): Check that n10/d10 are valid to --with-cpu.
 821         * config/nds32/nds32-n10.md: New file.
 822         * config/nds32/nds32-opts.h (nds32_cpu_type): Add CPU_N10.
 823         * config/nds32/nds32-pipelines-auxiliary.c: Implementation for n10
 824         pipeline.
 825         * config/nds32/nds32-protos.h: More declarations for n10 pipeline.
 826         * config/nds32/nds32-utils.c: More implementations for n10 pipeline.
 827         * config/nds32/nds32.md (pipeline_model): Add n10.
 828         * config/nds32/nds32.opt (mcpu): Support n10 pipeline cpus.
 829         * config/nds32/pipelines.md: Include n10 settings.
 830
 831 2018-05-13  Monk Chiang  <sh.chiang04@gmail.com>
 832             Kito Cheng  <kito.cheng@gmail.com>
 833             Chung-Ju Wu  <jasonwucj@gmail.com>
 834
 835         * config.gcc (nds32be-*-*): Handle --with-ext-dsp.
 836         * config/nds32/constants.md (unspec_element, unspec_volatile_element):
 837         Add enum values for DSP extension instructions.
 838         * config/nds32/constraints.md (Iu06, IU06, CVp5, CVs5, CVs2, CVhi):
 839         New constraints.
 840         * config/nds32/iterators.md (shifts, shiftrt, sat_plus, all_plus,
 841         sat_minus, all_minus, plus_minus, extend, sumax, sumin, sumin_max):
 842         New code iterators.
 843         (su, zs, uk, opcode, add_rsub, add_sub): New code attributes.
 844         * config/nds32/nds32-dspext.md: New file for DSP implementation.
 845         * config/nds32/nds32-intrinsic.c: Implementation of DSP extension.
 846         * config/nds32/nds32-intrinsic.md: Likewise.
 847         * config/nds32/nds32_intrinsic.h: Likewise.
 848         * config/nds32/nds32-md-auxiliary.c: Likewise.
 849         * config/nds32/nds32-memory-manipulation.c: Consider DSP extension.
 850         * config/nds32/nds32-predicates.c (const_vector_to_hwint): New.
 851         (nds32_valid_CVp5_p, nds32_valid_CVs5_p): New.
 852         (nds32_valid_CVs2_p, nds32_valid_CVhi_p): New.
 853         * config/nds32/nds32-protos.h: New declarations for DSP extension.
 854         * config/nds32/nds32-utils.c (extract_mac_non_acc_rtx): New case
 855         TYPE_DMAC in switch statement.
 856         * config/nds32/nds32.c: New checking and implementation for DSP
 857         extension instructions.
 858         * config/nds32/nds32.h: Likewise.
 859         * config/nds32/nds32.md: Likewise.
 860         * config/nds32/nds32.opt (mhw-abs, mext-dsp): New options.
 861         * config/nds32/predicates.md: Implement new predicates for DSP
 862         extension.
 863
 864 2018-05-11  Michael Meissner  <meissner@linux.vnet.ibm.com>
 865
 866         * config/rs6000/rs6000.md (mov<mode>_softfloat, FMOVE32):
 867         Reformat alternatives and attributes so it is easier to identify
 868         which constraints/attributes go with which instruction.
 869         (mov<mode>_hardfloat32, FMOVE64): Likewise.
 870         (mov<mode>_softfloat32, FMOVE64): Likewise.
 871         (mov<mode>_hardfloat64, FMOVE64): Likewise.
 872         (mov<mode>_softfloat64, FMOVE64): Likewise.
 873
 874 2018-05-11  Kelvin Nilsen  <kelvin@gcc.gnu.org>
 875
 876         * doc/extend.texi (PowerPC Built-in Functions): Rename this
 877         subsection.
 878         (Basic PowerPC Built-in Functions): The new name of the
 879         subsection previously known as "PowerPC Built-in Functions".
 880         (Basic PowerPC Built-in Functions Available on all Configurations):
 881         New subsubsection.
 882         (Basic PowerPC Built-in Functions Available on ISA 2.05): Likewise.
 883         (Basic PowerPC Built-in Functions Available on ISA 2.06): Likewise.
 884         (Basic PowerPC Built-in Functions Available on ISA 2.07): Likewise.
 885         (Basic PowerPC Built-in Functions Available on ISA 3.0): Likewise.
 886
 887 2018-05-11  Martin Jambor  <mjambor@suse.cz>
 888
 889         PR ipa/85655
 890         * ipa-cp.c (intersect_with_plats): Check that the lattice contains
 891         single const.
 892
 893 2018-05-11  Richard Earnshaw  <rearnsha@arm.com>
 894
 895         PR target/85733
 896         * config/arm/arm-cpus.in (fgroup ARMv6m): Add be8 feature.
 897
 898 2018-05-11  Sebastian Peryt  <sebastian.peryt@intel.com>
 899
 900         * common/config/i386/i386-common.c (OPTION_MASK_ISA_WAITPKG_SET,
 901         OPTION_MASK_ISA_WAITPKG_UNSET): New defines.
 902         (ix86_handle_option): Handle -mwaitpkg.
 903         * config.gcc: New header.
 904         * config/i386/cpuid.h (bit_WAITPKG): New bit.
 905         * config/i386/driver-i386.c (host_detect_local_cpu): Detect -mwaitpkg.
 906         * config/i386/i386-builtin-types.def ((UINT8, UNSIGNED, UINT64)): New
 907         function type.
 908         * config/i386/i386-c.c (ix86_target_macros_internal): Handle
 909         OPTION_MASK_ISA_WAITPKG.
 910         * config/i386/i386.c (ix86_target_string): Add -mwaitpkg.
 911         (ix86_option_override_internal): Add PTA_WAITPKG.
 912         (ix86_valid_target_attribute_inner_p): Add -mwaitpkg.
 913         (enum ix86_builtins): Add IX86_BUILTIN_UMONITOR, IX86_BUILTIN_UMWAIT,
 914         IX86_BUILTIN_TPAUSE.
 915         (ix86_init_mmx_sse_builtins): Define __builtin_ia32_umonitor,
 916         __builtin_ia32_umwait and __builtin_ia32_tpause.
 917         (ix86_expand_builtin): Expand IX86_BUILTIN_UMONITOR,
 918         IX86_BUILTIN_UMWAIT, IX86_BUILTIN_TPAUSE.
 919         * config/i386/i386.h (TARGET_WAITPKG, TARGET_WAITPKG_P): New.
 920         * config/i386/i386.md (UNSPECV_UMWAIT, UNSPECV_UMONITOR,
 921         UNSPECV_TPAUSE): New.
 922         (umwait, umwait_rex64, umonitor_<mode>, tpause, tpause_rex64): New.
 923         * config/i386/i386.opt: Add -mwaitpkg.
 924         * config/i386/waitpkgintrin.h: New file.
 925         * config/i386/x86intrin.h: New header.
 926         * doc/invoke.texi: Add -mwaitpkg.
 927
 928 2018-05-11  Richard Earnshaw  <rearnsha@arm.com>
 929
 930         PR target/85606
 931         * config/arm/arm-cpus.in: Add comment that ARMv6-m and ARMv6S-m are now
 932         equivalent.
 933         (cortex-m0): Use armv6s-m isa.
 934         (cortex-m0plus): Likewise.
 935         (cortex-m1): Likewise.
 936         (cortex-m0.small-multiply): Likewise.
 937         (cortex-m0plus.small-multiply): Likewise.
 938         (cortex-m1.small-multiply): Likewise.
 939
 940 2018-05-11  Allan Sandfeld Jensen  <allan.jensen@qt.io>
 941             Jakub Jelinek  <jakub@redhat.com>
 942
 943         PR tree-optimization/85692
 944         * tree-ssa-forwprop.c (simplify_vector_constructor): Try two
 945         source permute as well.
 946
 947 2018-05-11  Martin Liska  <mliska@suse.cz>
 948
 949         PR sanitizer/85556
 950         * doc/extend.texi: Document LLVM style format for no_sanitize
 951         attribute.
 952
 953 2018-05-10  Michael Meissner  <meissner@linux.vnet.ibm.com>
 954
 955         * config/rs6000/rs6000.c (mode_supports_dq_form): Rename
 956         mode_supports_vsx_dform_quad to mode_supports_dq_form.
 957         (mode_supports_vsx_dform_quad): Likewise.
 958         (mode_supports_vmx_dform): Move these functions to be next to the
 959         other mode_supports functions.
 960         (mode_supports_dq_form): Likewise.
 961         (quad_address_p): Change calls of mode_supports_vsx_dform_quad to
 962         mode_supports_dq_form.
 963         (reg_offset_addressing_ok_p): Likewise.
 964         (offsettable_ok_by_alignment): Likewise.
 965         (rs6000_legitimate_offset_address_p): Likewise.
 966         (legitimate_lo_sum_address_p): Likewise.
 967         (rs6000_legitimize_address): Likewise.
 968         (rs6000_legitimize_reload_address): Likewise.
 969         (rs6000_secondary_reload_inner): Likewise.
 970         (rs6000_preferred_reload_class): Likewise.
 971         (rs6000_output_move_128bit): Likewise.
 972
 973 2018-05-10  Uros Bizjak  <ubizjak@gmail.com>
 974
 975         * config/i386/i386.c (ix86_expand_builtin) <case IX86_BUILTIN_RDPID>:
 976         Generate SImode target register for null target.
 977         <case IX86_BUILTIN_XGETBV>: Ditto.
 978         <case IX86_BUILTIN_XSETBV>: Optimize LSHIFTRT generation.
 979         * config/i386/xsaveintrin.h (_xgetbv): Add missing return.
 980
 981 2018-05-10  Carl Love  <cel@us.ibm.com>
 982
 983         * config/rs6000/rs6000.md (prefetch): Generate ISA 2.06 instructions
 984         dcbtt and dcbtstt if operands[2] is 0.
 985
 986 2018-05-10  Uros Bizjak  <ubizjak@gmail.com>
 987
 988         PR target/85693
 989         * config/i386/sse.md (usadv64qi): New expander.
 990
 991 2018-05-10  Segher Boessenkool  <segher@kernel.crashing.org>
 992
 993         * config/rs6000/altivec.md (altivec_vmrghb, altivec_vmrghh,
 994         altivec_vmrghw, altivec_vmrglb, altivec_vmrglh, altivec_vmrglw): Remove
 995         -maltivec=be support.
 996         (vec_widen_umult_even_v16qi, vec_widen_smult_even_v16qi,
 997         vec_widen_umult_even_v8hi, vec_widen_smult_even_v8hi,
 998         vec_widen_umult_even_v4si, vec_widen_smult_even_v4si,
 999         vec_widen_umult_odd_v16qi, vec_widen_smult_odd_v16qi,
1000         vec_widen_umult_odd_v8hi, vec_widen_smult_odd_v8hi,
1001         vec_widen_umult_odd_v4si, vec_widen_smult_odd_v4si, altivec_vpkpx,
1002         altivec_vpks<VI_char>ss, altivec_vpks<VI_char>us,
1003         altivec_vpku<VI_char>us, altivec_vpku<VI_char>um, altivec_vsum2sws,
1004         altivec_vsumsws): Adjust.
1005         (altivec_vspltb *altivec_vspltb_internal, altivec_vsplth,
1006         *altivec_vsplth_internal, altivec_vspltw, *altivec_vspltw_internal,
1007         altivec_vspltsf, *altivec_vspltsf_internal): Remove -maltivec=be
1008         support.
1009         (altivec_vperm_<mode>, altivec_vperm_<mode>_uns,
1010         altivec_vupkhs<VU_char>, altivec_vupkls<VU_char>, altivec_vupkhpx,
1011         altivec_vupklpx, altivec_lvsl, altivec_lvsr): Adjust.
1012         (altivec_lve<VI_char>x): Delete expand.
1013         (*altivec_lve<VI_char>x_internal): Rename to...
1014         (altivec_lve<VI_char>x): ... this.
1015         (altivec_lvxl_<mode>): Delete expand.
1016         (*altivec_lvxl_<mode>_internal): Rename to ...
1017         (altivec_lvxl_<mode>): ... this.
1018         (altivec_stvxl_<mode>): Delete expand.
1019         (*altivec_stvxl_<mode>_internal): Rename to ...
1020         (altivec_stvxl_<mode>): ... this.
1021         (altivec_stve<VI_char>x): Delete expand.
1022         (*altivec_stve<VI_char>x_internal): Rename to ...
1023         (altivec_stve<VI_char>x): ... this.
1024         (doublee<mode>2, unsdoubleev4si2, doubleo<mode>2, unsdoubleov4si2,
1025         doubleh<mode>2, unsdoublehv4si2, doublel<mode>2, unsdoublelv4si2,
1026         reduc_plus_scal_<mode>): Adjust.
1027         * config/rs6000/rs6000-c.c (rs6000_target_modify_macros): Adjust
1028         comment.
1029         (rs6000_cpu_cpp_builtins): Adjust.
1030         (altivec_resolve_overloaded_builtin): Remove -maltivec=be support.
1031         * config/rs6000/rs6000-protos.h (altivec_expand_lvx_be,
1032         altivec_expand_stvx_be, altivec_expand_stvex_be): Delete.
1033         * config/rs6000/rs6000.c (rs6000_option_override_internal): Remove
1034         -maltivec=be support.
1035         (rs6000_split_vec_extract_var): Adjust.
1036         (rs6000_split_v4si_init): Adjust.
1037         (swap_selector_for_mode): Delete.
1038         (altivec_expand_lvx_be, altivec_expand_stvx_be,
1039         altivec_expand_stvex_be): Delete.
1040         (altivec_expand_lv_builtin, altivec_expand_stv_builtin): Remove
1041         -maltivec=be support.
1042         (rs6000_gimple_fold_builtin): Ditto.
1043         (rs6000_generate_float2_double_code, rs6000_generate_float2_code):
1044         Adjust.
1045         * config/rs6000/rs6000.h (VECTOR_ELT_ORDER_BIG): Delete.
1046         (TARGET_DIRECT_MOVE_64BIT): Adjust.
1047         * config/rs6000/rs6000.md (split for extendsidi2 for vectors): Adjust.
1048         * config/rs6000/rs6000.opt (maltivec=le, maltivec=be): Delete.
1049         * config/rs6000/vsx.md (floate<mode>, unsfloatev2di, floato<mode>,
1050         unsfloatov2di, vsignedo_v2df, vsignede_v2df, vunsignedo_v2df,
1051         vunsignede_v2df, vsx_extract_<mode>_p9, *vsx_extract_si,
1052         *vsx_extract_<mode>_p8, *vsx_extract_si_<uns>float_df,
1053         *vsx_extract_si_<uns>float_<mode>, vsx_set_<mode>_p9, vsx_set_v4sf_p9,
1054         *vsx_insert_extract_v4sf_p9, *vsx_insert_extract_v4sf_p9_2, and an
1055         anonymous split): Adjust.
1056         (vsx_mergel_<mode>, vsx_mergeh_<mode>): Remove -maltivec=be support.
1057         (vsx_xxspltd_<mode>, extract4b, insert4b): Adjust.
1058
1059 2018-05-10  Eric Botcazou  <ebotcazou@adacore.com>
1060
1061         * configure.ac (gcc_gxx_include_dir_add_sysroot): Set it to 1 only
1062         when --with-gxx-include-dir is also specified.
1063         * configure: Regenerate.
1064
1065 2018-05-09  Jim Wilson  <jimw@sifive.com>
1066
1067         PR target/84797
1068         * config.gcc (riscv*-*-*): Handle --with-multilib-list.
1069         * config/riscv/t-withmultilib: New.
1070         * config/riscv/withmultilib.h: New.
1071         * doc/install.texi: Document RISC-V --with-multilib-list support.
1072
1073 2018-05-09  Richard Biener  <rguenther@suse.de>
1074
1075         * tree-vect-slp.c (vect_bb_slp_scalar_cost): Fill a cost
1076         vector.
1077         (vect_bb_vectorization_profitable_p): Adjust.  Compute
1078         actual scalar cost using the cost vector and the add_stmt_cost
1079         machinery.
1080
1081 2018-05-09  Segher Boessenkool  <segher@kernel.crashing.org>
1082
1083         PR rtl-optimization/85645
1084         * config/rs6000/rs6000.c (rs6000_emit_prologue_components): Put a SET
1085         in the REG_CFA_REGISTER note for LR, don't leave it empty.
1086
1087 2018-05-09  Segher Boessenkool  <segher@kernel.crashing.org>
1088
1089         PR rtl-optimization/85645
1090         * shrink-wrap.c (spread_components): Return a boolean saying if
1091         anything was changed.
1092         (try_shrink_wrapping_separate): Iterate spread_components until
1093         nothing changes anymore.
1094
1095 2018-05-09  Segher Boessenkool  <segher@kernel.crashing.org>
1096
1097         PR rtl-optimization/85645
1098         * regrename.c (build_def_use): Also kill the chains that include the
1099         destination of a REG_CFA_REGISTER note.
1100
1101 2018-05-09  Segher Boessenkool  <segher@kernel.crashing.org>
1102
1103         PR rtl-optimization/85645
1104         *  regcprop.c (copyprop_hardreg_forward_1): Don't propagate into an
1105         insn that has a REG_CFA_REGISTER note.
1106
1107 2018-05-09  Richard Sandiford  <richard.sandiford@linaro.org>
1108
1109         * cfgexpand.c (expand_clobber): New function.
1110         (expand_gimple_stmt_1): Use it.
1111         * tree-vect-stmts.c (vect_clobber_variable): New function,
1112         split out from...
1113         (vectorizable_simd_clone_call): ...here.
1114         (vectorizable_store): Emit a clobber either side of an
1115         IFN_STORE_LANES sequence.
1116         (vectorizable_load): Emit a clobber after an IFN_LOAD_LANES sequence.
1117
1118 2018-05-09  Tom de Vries  <tom@codesourcery.com>
1119
1120         PR target/85626
1121         * config/nvptx/nvptx.md (define_insn "trap", define_insn "trap_if_true")
1122         (define_insn "trap_if_false"): Add exit after trap.
1123
1124 2018-05-09  Eric Botcazou  <ebotcazou@adacore.com>
1125
1126         PR rtl-optimization/85638
1127         * bb-reorder.c: Include common/common-target.h.
1128         (create_forwarder_block): New function extracted from...
1129         (fix_up_crossing_landing_pad): ...here.  Rename into...
1130         (dw2_fix_up_crossing_landing_pad): ...this.
1131         (sjlj_fix_up_crossing_landing_pad): New function.
1132         (find_rarely_executed_basic_blocks_and_crossing_edges): In SJLJ mode,
1133         call sjlj_fix_up_crossing_landing_pad if there are incoming EH edges
1134         from both partitions and exit the loop after one iteration.
1135
1136 2018-05-08  Kelvin Nilsen  <kelvin@gcc.gnu.org>
1137
1138         Revert:
1139         * doc/extend.texi (PowerPC Built-in Functions): Rename this
1140         subsection.
1141         (Basic PowerPC Built-in Functions): The new name of the
1142         subsection previously known as "PowerPC Built-in Functions".
1143         (Basic PowerPC Built-in Functions Available on all Configurations):
1144         New subsubsection.
1145         (Basic PowerPC Built-in Functions Available on ISA 2.05): New
1146         subsubsection.
1147         (Basic PowerPC Built-in Functions Available on ISA 2.06): New
1148         subsubsection.
1149         (Basic PowerPC Built-in Functions Available on ISA 2.07): New
1150         subsubsection.
1151         (Basic PowerPC Built-in Functions Available on ISA 3.0): New
1152         subsubsection.
1153
1154 2018-05-08  Jim Wilson  <jimw@sifive.com>
1155
1156         * config/riscv/linux.h (MUSL_ABI_SUFFIX): Delete unnecessary backslash.
1157         (LD_EMUL_SUFFIX): New.
1158         (LINK_SPEC): Use it.
1159
1160 2018-05-08  Kelvin Nilsen  <kelvin@gcc.gnu.org>
1161
1162         * doc/extend.texi (PowerPC Built-in Functions): Rename this
1163         subsection.
1164         (Basic PowerPC Built-in Functions): The new name of the
1165         subsection previously known as "PowerPC Built-in Functions".
1166         (Basic PowerPC Built-in Functions Available on all Configurations):
1167         New subsubsection.
1168         (Basic PowerPC Built-in Functions Available on ISA 2.05): New
1169         subsubsection.
1170         (Basic PowerPC Built-in Functions Available on ISA 2.06): New
1171         subsubsection.
1172         (Basic PowerPC Built-in Functions Available on ISA 2.07): New
1173         subsubsection.
1174         (Basic PowerPC Built-in Functions Available on ISA 3.0): New
1175         subsubsection.
1176
1177 2018-05-08  Jakub Jelinek  <jakub@redhat.com>
1178
1179         PR target/85683
1180         * config/i386/i386.md: Add peepholes for mem {+,-,&,|,^}= x; mem != 0
1181         after cmpelim optimization.
1182
1183 2018-05-08  Olga Makhotina  <olga.makhotina@intel.com>
1184
1185         * config.gcc: Support "goldmont".
1186         * config/i386/driver-i386.c (host_detect_local_cpu): Detect "goldmont".
1187         * config/i386/i386-c.c (ix86_target_macros_internal): Handle
1188         PROCESSOR_GOLDMONT.
1189         * config/i386/i386.c (m_GOLDMONT): Define.
1190         (processor_target_table): Add "goldmont".
1191         (PTA_GOLDMONT): Define.
1192         (ix86_lea_outperforms): Add TARGET_GOLDMONT.
1193         (get_builtin_code_for_version): Handle PROCESSOR_GOLDMONT.
1194         (fold_builtin_cpu): Add M_INTEL_GOLDMONT.
1195         (fold_builtin_cpu): Add "goldmont".
1196         (ix86_add_stmt_cost): Add TARGET_GOLDMONT.
1197         (ix86_option_override_internal): Add "goldmont".
1198         * config/i386/i386.h (processor_costs): Define TARGET_GOLDMONT.
1199         (processor_type): Add PROCESSOR_GOLDMONT.
1200         * config/i386/i386.md: Add CPU "glm".
1201         * config/i386/glm.md: New file.
1202         * config/i386/x86-tune.def: Add m_GOLDMONT.
1203         * doc/invoke.texi: Add goldmont as x86 -march=/-mtune= CPU type.
1204
1205 2018-05-08  Jakub Jelinek  <jakub@redhat.com>
1206
1207         PR target/85572
1208         * config/i386/i386.c (ix86_expand_sse2_abs): Handle E_V2DImode and
1209         E_V4DImode.
1210         * config/i386/sse.md (abs<mode>2): Use VI_AVX2 iterator instead of
1211         VI1248_AVX512VL_AVX512BW.  Handle V2DImode and V4DImode if not
1212         TARGET_AVX512VL using ix86_expand_sse2_abs.  Formatting fixes.
1213
1214         PR target/85317
1215         * config/i386/i386.c (ix86_fold_builtin): Handle
1216         IX86_BUILTIN_{,P}MOVMSK{PS,PD,B}{,128,256}.
1217
1218         PR target/85480
1219         * config/i386/sse.md (ssequaterinsnmode): New mode attribute.
1220         (*<extract_type>_vinsert<shuffletype><extract_suf>_0): New pattern.
1221
1222 2018-05-08  Richard Earnshaw  <rearnsha@arm.com>
1223
1224         PR target/85658
1225         * config/arm/parsecpu.awk (check_cpu): Fix operator precedence.
1226         (check_arch): Likewise.
1227         (check_fpu): Return the result rather than printing it.
1228         (end arch): Fix operator precedence.
1229         (end cpu): Likewise.
1230         (END): Print the result from check_fpu.
1231
1232 2018-05-08  Richard Sandiford  <richard.sandiford@linaro.org>
1233             Alan Hayward  <alan.hayward@arm.com>
1234             David Sherwood  <david.sherwood@arm.com>
1235
1236         * config/aarch64/aarch64-sve.md (*pred_cmp<cmp_op><mode>_combine)
1237         (*pred_cmp<cmp_op><mode>, *fcm<cmp_op><mode>_and_combine)
1238         (*fcmuo<mode>_and_combine, *fcm<cmp_op><mode>_and)
1239         (*fcmuo<mode>_and): New patterns.
1240
1241 2018-05-08  Richard Sandiford  <richard.sandiford@linaro.org>
1242
1243         * config/aarch64/iterators.md (UNSPEC_COND_LO, UNSPEC_COND_LS)
1244         (UNSPEC_COND_HI, UNSPEC_COND_HS, UNSPEC_COND_UO): Delete.
1245         (SVE_INT_CMP, SVE_FP_CMP): New code iterators.
1246         (cmp_op, sve_imm_con): New code attributes.
1247         (SVE_COND_INT_CMP, imm_con): Delete.
1248         (cmp_op): Remove above unspecs from int attribute.
1249         * config/aarch64/aarch64-sve.md (*vec_cmp<cmp_op>_<mode>): Rename
1250         to...
1251         (*cmp<cmp_op><mode>): ...this.  Use UNSPEC_MERGE_PTRUE instead of
1252         comparison-specific unspecs.
1253         (*vec_cmp<cmp_op>_<mode>_ptest): Rename to...
1254         (*cmp<cmp_op><mode>_ptest): ...this and adjust likewise.
1255         (*vec_cmp<cmp_op>_<mode>_cc): Rename to...
1256         (*cmp<cmp_op><mode>_cc): ...this and adjust likewise.
1257         (*vec_fcm<cmp_op><mode>): Rename to...
1258         (*fcm<cmp_op><mode>): ...this and adjust likewise.
1259         (*vec_fcmuo<mode>): Rename to...
1260         (*fcmuo<mode>): ...this and adjust likewise.
1261         (*pred_fcm<cmp_op><mode>): New pattern.
1262         * config/aarch64/aarch64.c (aarch64_emit_unop, aarch64_emit_binop)
1263         (aarch64_emit_sve_ptrue_op, aarch64_emit_sve_ptrue_op_cc): New
1264         functions.
1265         (aarch64_unspec_cond_code): Remove handling of LTU, GTU, LEU, GEU
1266         and UNORDERED.
1267         (aarch64_gen_unspec_cond, aarch64_emit_unspec_cond): Delete.
1268         (aarch64_emit_sve_predicated_cond): New function.
1269         (aarch64_expand_sve_vec_cmp_int): Use aarch64_emit_sve_ptrue_op_cc.
1270         (aarch64_emit_unspec_cond_or): Replace with...
1271         (aarch64_emit_sve_or_conds): ...this new function.  Use
1272         aarch64_emit_sve_ptrue_op for the individual comparisons and
1273         aarch64_emit_binop to OR them together.
1274         (aarch64_emit_inverted_unspec_cond): Replace with...
1275         (aarch64_emit_sve_inverted_cond): ...this new function.  Use
1276         aarch64_emit_sve_ptrue_op for the comparison and
1277         aarch64_emit_unop to invert the result.
1278         (aarch64_expand_sve_vec_cmp_float): Update after the above
1279         changes.  Use aarch64_emit_sve_ptrue_op for native comparisons.
1280
1281 2018-05-07  Nathan Sidwell  <nathan@acm.org>
1282
1283         * doc/invoke.texi (C++ Dialect Options): Remove -ffor-scope.
1284         * doc/extend.texi (Deprecated Features): Remove -fno-for-scope
1285         (Backwards Compatibility): Likewise.
1286
1287 2018-05-07  Luis Machado  <luis.machado@linaro.org>
1288
1289         PR bootstrap/85681
1290         Revert:
1291         2018-05-07  Luis Machado  <luis.machado@linaro.org>
1292
1293         * config/aarch64/aarch64-protos.h (cpu_prefetch_tune)
1294         <prefetch_dynamic_strides>: New const bool field.
1295         * config/aarch64/aarch64.c (generic_prefetch_tune): Update to include
1296         prefetch_dynamic_strides.
1297         (exynosm1_prefetch_tune): Likewise.
1298         (thunderxt88_prefetch_tune): Likewise.
1299         (thunderx_prefetch_tune): Likewise.
1300         (thunderx2t99_prefetch_tune): Likewise.
1301         (qdf24xx_prefetch_tune): Likewise. Set prefetch_dynamic_strides to false.
1302         (aarch64_override_options_internal): Update to set
1303         PARAM_PREFETCH_DYNAMIC_STRIDES.
1304         * doc/invoke.texi (prefetch-dynamic-strides): Document new option.
1305         * params.def (PARAM_PREFETCH_DYNAMIC_STRIDES): New.
1306         * params.h (PARAM_PREFETCH_DYNAMIC_STRIDES): Define.
1307         * tree-ssa-loop-prefetch.c (should_issue_prefetch_p): Account for
1308         prefetch-dynamic-strides setting.
1309
1310         2018-05-07  Luis Machado  <luis.machado@linaro.org>
1311
1312         * config/aarch64/aarch64-protos.h (cpu_prefetch_tune)
1313         <minimum_stride>: New const int field.
1314         * config/aarch64/aarch64.c (generic_prefetch_tune): Update to include
1315         minimum_stride field.
1316         (exynosm1_prefetch_tune): Likewise.
1317         (thunderxt88_prefetch_tune): Likewise.
1318         (thunderx_prefetch_tune): Likewise.
1319         (thunderx2t99_prefetch_tune): Likewise.
1320         (qdf24xx_prefetch_tune): Likewise. Set minimum_stride to 2048.
1321         (aarch64_override_options_internal): Update to set
1322         PARAM_PREFETCH_MINIMUM_STRIDE.
1323         * doc/invoke.texi (prefetch-minimum-stride): Document new option.
1324         * params.def (PARAM_PREFETCH_MINIMUM_STRIDE): New.
1325         * params.h (PARAM_PREFETCH_MINIMUM_STRIDE): Define.
1326         * tree-ssa-loop-prefetch.c (should_issue_prefetch_p): Return false if
1327         stride is constant and is below the minimum stride threshold.
1328
1329 2018-05-07  Luis Machado  <luis.machado@linaro.org>
1330
1331         * config/aarch64/aarch64.c (qdf24xx_prefetch_tune) <l2_cache_size>: Set
1332         to 512.
1333
1334 2018-05-07  Luis Machado  <luis.machado@linaro.org>
1335
1336         * config/aarch64/aarch64-protos.h (cpu_prefetch_tune)
1337         <prefetch_dynamic_strides>: New const bool field.
1338         * config/aarch64/aarch64.c (generic_prefetch_tune): Update to include
1339         prefetch_dynamic_strides.
1340         (exynosm1_prefetch_tune): Likewise.
1341         (thunderxt88_prefetch_tune): Likewise.
1342         (thunderx_prefetch_tune): Likewise.
1343         (thunderx2t99_prefetch_tune): Likewise.
1344         (qdf24xx_prefetch_tune): Likewise. Set prefetch_dynamic_strides to false.
1345         (aarch64_override_options_internal): Update to set
1346         PARAM_PREFETCH_DYNAMIC_STRIDES.
1347         * doc/invoke.texi (prefetch-dynamic-strides): Document new option.
1348         * params.def (PARAM_PREFETCH_DYNAMIC_STRIDES): New.
1349         * params.h (PARAM_PREFETCH_DYNAMIC_STRIDES): Define.
1350         * tree-ssa-loop-prefetch.c (should_issue_prefetch_p): Account for
1351         prefetch-dynamic-strides setting.
1352
1353 2018-05-07  Luis Machado  <luis.machado@linaro.org>
1354
1355         * config/aarch64/aarch64-protos.h (cpu_prefetch_tune)
1356         <minimum_stride>: New const int field.
1357         * config/aarch64/aarch64.c (generic_prefetch_tune): Update to include
1358         minimum_stride field.
1359         (exynosm1_prefetch_tune): Likewise.
1360         (thunderxt88_prefetch_tune): Likewise.
1361         (thunderx_prefetch_tune): Likewise.
1362         (thunderx2t99_prefetch_tune): Likewise.
1363         (qdf24xx_prefetch_tune): Likewise. Set minimum_stride to 2048.
1364         (aarch64_override_options_internal): Update to set
1365         PARAM_PREFETCH_MINIMUM_STRIDE.
1366         * doc/invoke.texi (prefetch-minimum-stride): Document new option.
1367         * params.def (PARAM_PREFETCH_MINIMUM_STRIDE): New.
1368         * params.h (PARAM_PREFETCH_MINIMUM_STRIDE): Define.
1369         * tree-ssa-loop-prefetch.c (should_issue_prefetch_p): Return false if
1370         stride is constant and is below the minimum stride threshold.
1371
1372 2018-05-06  Jakub Jelinek  <jakub@redhat.com>
1373
1374         PR c++/85659
1375         * cfgexpand.c (expand_asm_stmt): Don't create a temporary if
1376         the type is addressable.  Don't force op into register if it has
1377         BLKmode.
1378
1379 2018-05-05  Roland McGrath  <mcgrathr@google.com>
1380
1381         PR other/77609
1382         * varasm.c (default_section_type_flags): Set SECTION_NOTYPE for
1383         any section for which we don't know a specific type it should have,
1384         regardless of name.  Previously this was done only for the exact
1385         names ".init_array", ".fini_array", and ".preinit_array".
1386         (default_elf_asm_named_section): Add comment about
1387         relationship with default_section_type_flags and SECTION_NOTYPE.
1388         (get_section): Don't consider it a type conflict if one side has
1389         SECTION_NOTYPE and the other doesn't, as long as neither has the
1390         SECTION_BSS et al used in the default_section_type_flags logic.
1391
1392 2018-05-05  Tom de Vries  <tom@codesourcery.com>
1393
1394         PR target/85653
1395         * config/nvptx/nvptx.c (WORKAROUND_PTXJIT_BUG_3): Define.
1396         (workaround_barsyncs): New function.
1397         (nvptx_reorg): Use workaround_barsyncs.
1398         * config/nvptx/nvptx.md (define_c_enum "unspecv"): Add UNSPECV_MEMBAR.
1399         (define_expand "nvptx_membar_cta"): New define_expand.
1400         (define_insn "*nvptx_membar_cta"): New insn.
1401
1402 2018-05-04  Pekka Jääskeläinen  <pekka.jaaskelainen@parmance.com>
1403
1404         * brig-builtins.def: Add consts to ptrs etc. in BRIG builtin defs.
1405         To improve optimization opportunities.
1406         * builtin-types.def: The new needed builtin types for the above.
1407
1408 2018-05-04  Richard Biener  <rguenther@suse.de>
1409
1410         * bb-reorder.c (sanitize_hot_paths): Release hot_bbs_to_check.
1411         * gimple-ssa-store-merging.c
1412         (imm_store_chain_info::output_merged_store): Remove redundant create,
1413         release split_store vector contents on failure.
1414         * tree-vect-slp.c (vect_schedule_slp_instance): Avoid leaking
1415         scalar stmt vector on cache hit.
1416
1417 2018-05-04  Segher Boessenkool  <segher@kernel.crashing.org>
1418
1419         * common/config/rs6000/rs6000-common.c (rs6000_handle_option): Remove
1420         Xilinx FP support.
1421         * config.gcc (powerpc-xilinx-eabi*): Remove.
1422         * config/rs6000/predicates.md (easy_fp_constant): Remove Xilinx FP
1423         support.
1424         (fusion_addis_mem_combo_load): Ditto.
1425         * config/rs6000/rs6000-c.c (rs6000_target_modify_macros): Remove Xilinx
1426         FP support.
1427         (rs6000_cpu_cpp_builtins): Ditto.
1428         * config/rs6000/rs6000-linux.c
1429         (rs6000_linux_float_exceptions_rounding_supported_p): Ditto.
1430         * config/rs6000/rs6000-opts.h (enum fpu_type_t): Delete.
1431         * config/rs6000/rs6000.c (rs6000_debug_reg_global): Remove Xilinx FP
1432         support.
1433         (rs6000_setup_reg_addr_masks): Ditto.
1434         (rs6000_init_hard_regno_mode_ok): Ditto.
1435         (rs6000_option_override_internal): Ditto.
1436         (legitimate_lo_sum_address_p): Ditto.
1437         (rs6000_legitimize_address): Ditto.
1438         (rs6000_legitimize_reload_address): Ditto.
1439         (rs6000_legitimate_address_p): Ditto.
1440         (abi_v4_pass_in_fpr): Ditto.
1441         (setup_incoming_varargs): Ditto.
1442         (rs6000_gimplify_va_arg): Ditto.
1443         (rs6000_split_multireg_move): Ditto.
1444         (rs6000_savres_strategy): Ditto.
1445         (rs6000_emit_prologue_components): Ditto.
1446         (rs6000_emit_epilogue_components): Ditto.
1447         (rs6000_emit_prologue): Ditto.
1448         (rs6000_emit_epilogue): Ditto.
1449         (rs6000_elf_file_end): Ditto.
1450         (rs6000_function_value): Ditto.
1451         (rs6000_libcall_value): Ditto.
1452         * config/rs6000/rs6000.h: Ditto.
1453         (TARGET_MINMAX_SF, TARGET_MINMAX_DF): Delete, merge to ...
1454         (TARGET_MINMAX): ... this.  New.
1455         (TARGET_SF_FPR, TARGET_DF_FPR, TARGET_SF_INSN, TARGET_DF_INSN): Delete.
1456         * config/rs6000/rs6000.md: Remove Xilinx FP support.
1457         (*movsi_internal1_single): Delete.
1458         * config/rs6000/rs6000.opt (msingle-float, mdouble-float, msimple-fpu,
1459         mfpu=, mxilinx-fpu): Delete.
1460         * config/rs6000/singlefp.h: Delete.
1461         * config/rs6000/sysv4.h: Remove Xilinx FP support.
1462         * config/rs6000/t-rs6000: Ditto.
1463         * config/rs6000/t-xilinx: Delete.
1464         * gcc/config/rs6000/titan.md: Adjust for fp_type removal.
1465         * gcc/config/rs6000/vsx.md: Remove Xilinx FP support.
1466         (VStype_simple): Delete.
1467         (VSfptype_simple, VSfptype_mul, VSfptype_div, VSfptype_sqrt): Delete.
1468         * config/rs6000/xfpu.h: Delete.
1469         * config/rs6000/xfpu.md: Delete.
1470         * config/rs6000/xilinx.h: Delete.
1471         * config/rs6000/xilinx.opt: Delete.
1472         * gcc/doc/invoke.texi (RS/6000 and PowerPC Options): Remove
1473         -msingle-float, -mdouble-float, -msimple-fpu, -mfpu=, and -mxilinx-fpu.
1474
1475 2018-05-04  Tom de Vries  <tom@codesourcery.com>
1476
1477         PR libgomp/85639
1478         * builtins.c (expand_builtin_goacc_parlevel_id_size): Handle null target
1479         if ignore == 0.
1480
1481 2018-05-04  Richard Biener  <rguenther@suse.de>
1482
1483         PR middle-end/85627
1484         * tree-complex.c (update_complex_assignment): We are always in SSA form.
1485         (expand_complex_div_wide): Likewise.
1486         (expand_complex_operations_1): Likewise.
1487         (expand_complex_libcall): Preserve EH info of the original stmt.
1488         (tree_lower_complex): Handle removed blocks.
1489         * tree.c (build_common_builtin_nodes): Do not set ECF_NOTRHOW
1490         on complex multiplication and division libcall builtins.
1491
1492 2018-05-04  Richard Biener  <rguenther@suse.de>
1493
1494         PR middle-end/85574
1495         * fold-const.c (negate_expr_p): Restrict negation of operand
1496         zero of a division to when we know that can happen without
1497         overflow.
1498         (fold_negate_expr_1): Likewise.
1499
1500 2018-05-04  Jakub Jelinek  <jakub@redhat.com>
1501
1502         PR libstdc++/85466
1503         * real.h (real_nextafter): Declare.
1504         * real.c (real_nextafter): New function.
1505         * fold-const-call.c (fold_const_nextafter): New function.
1506         (fold_const_call_sss): Call it for CASE_CFN_NEXTAFTER and
1507         CASE_CFN_NEXTTOWARD.
1508         (fold_const_call_1): For CASE_CFN_NEXTTOWARD call fold_const_call_sss
1509         even when arg1_mode is different from arg0_mode.
1510
1511 2018-05-03  Nathan Sidwell  <nathan@acm.org>
1512
1513         * doc/extend.texi (Deprecated Features): Remove
1514         -ffriend-injection.
1515         (Backwards Compatibility): Likewise.
1516         * doc/invoke.texi (C++ Language Options): Likewise.
1517         (C++ Dialect Options): Likewise.
1518
1519 2018-05-03  Jakub Jelinek  <jakub@redhat.com>
1520
1521         PR target/85530
1522         * config/i386/avx512fintrin.h (_mm512_mullox_epi64,
1523         _mm512_mask_mullox_epi64): New intrinsics.
1524
1525 2018-05-03  Tom de Vries  <tom@codesourcery.com>
1526
1527         PR testsuite/85106
1528         * doc/sourcebuild.texi (Commands for use in dg-final, Scan optimization
1529         dump files): Add offload-tree.
1530
1531 2018-05-03  Richard Biener  <rguenther@suse.de>
1532
1533         PR tree-optimization/85615
1534         * tree-ssa-threadupdate.c (thread_block_1): Only allow exits
1535         to loops not nested in BBs loop father to avoid creating multi-entry
1536         loops.
1537
1538 2018-05-03  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
1539
1540         PR tree-optimization/70291
1541         * tree-complex.c (expand_complex_libcall): Add type, inplace_p
1542         arguments.  Change return type to tree.  Emit libcall as a new
1543         statement rather than replacing existing one when inplace_p is true.
1544         (expand_complex_multiplication_components): New function.
1545         (expand_complex_multiplication): Expand floating-point complex
1546         multiplication using the above.
1547         (expand_complex_division): Rename inner_type parameter to type.
1548         Update expand_complex_libcall call-site.
1549         (expand_complex_operations_1): Update expand_complex_multiplication
1550         and expand_complex_division call-sites.
1551
1552 2018-05-02  Jakub Jelinek  <jakub@redhat.com>
1553
1554         PR target/85582
1555         * config/i386/i386.md (*ashl<dwi>3_doubleword_mask,
1556         *ashl<dwi>3_doubleword_mask_1, *<shift_insn><dwi>3_doubleword_mask,
1557         *<shift_insn><dwi>3_doubleword_mask_1): In condition require that
1558         the highest significant bit of the shift count mask is clear.  In
1559         check whether and[sq]i3 is needed verify that all significant bits
1560         of the shift count other than the highest are set.
1561
1562 2018-05-02  Tom de Vries  <tom@codesourcery.com>
1563
1564         PR libgomp/82428
1565         * builtins.def (DEF_GOACC_BUILTIN_ONLY): Define.
1566         * omp-builtins.def (BUILT_IN_GOACC_PARLEVEL_ID)
1567         (BUILT_IN_GOACC_PARLEVEL_SIZE): New builtin.
1568         * builtins.c (expand_builtin_goacc_parlevel_id_size): New function.
1569         (expand_builtin): Call expand_builtin_goacc_parlevel_id_size.
1570         * doc/extend.texi (Other Builtins): Add __builtin_goacc_parlevel_id and
1571         __builtin_goacc_parlevel_size.
1572
1573 2018-05-02  Richard Biener  <rguenther@suse.de>
1574
1575         PR tree-optimization/85597
1576         * tree-vect-stmts.c (vectorizable_operation): For ternary SLP
1577         do not use split vect_get_vec_defs call but call vect_get_slp_defs
1578         directly.
1579
1580 2018-05-02  Tom de Vries  <tom@codesourcery.com>
1581
1582         PR testsuite/85106
1583         * doc/sourcebuild.texi (Commands for use in dg-final, Scan optimization
1584         dump files): Add ltrans-tree.
1585
1586 2018-05-02  Tom de Vries  <tom@codesourcery.com>
1587
1588         PR testsuite/85106
1589         * doc/sourcebuild.texi (Commands for use in dg-final, Scan optimization
1590         dump files): Add wpa-ipa.
1591
1592 2018-05-02  Segher Boessenkool  <segher@kernel.crashing.org>
1593
1594         * config.gcc (powerpc*-*-*): Remove paired.h.  Unsupport the
1595         powerpc*-*-linux*paired* target.
1596         * config/rs6000/750cl.h: Delete.
1597         * config/rs6000/paired.h: Delete.
1598         * config/rs6000/paired.md: Delete.
1599         * config/rs6000/predicates.md (easy_vector_constant): Remove paired
1600         float support.
1601         * config/rs6000/rs6000-builtin.def: Remove paired float support.
1602         * config/rs6000/rs6000-c.c (rs6000_target_modify_macros): Update
1603         comment.  Remove paired float support.
1604         * config/rs6000/rs6000-modes.def: Remove V2SF and V2SI.
1605         * config/rs6000/rs6000-opts.h (enum rs6000_vector): Delete
1606         VECTOR_PAIRED.
1607         * config/rs6000/rs6000-protos.h (paired_expand_vector_init,
1608         paired_emit_vector_cond_expr, paired_expand_vector_move): Delete
1609         declarations.
1610         * config/rs6000/rs6000.c: Remove paired float support.
1611         (paired_expand_vector_init, paired_expand_vector_move,
1612         paired_emit_vector_compare, paired_emit_vector_cond_expr,
1613         (paired_expand_lv_builtin, paired_expand_stv_builtin,
1614         paired_expand_builtin, paired_expand_predicate_builtin,
1615         paired_init_builtins): Delete.
1616         * config/rs6000/rs6000.h: Remove paired float support.
1617         * config/rs6000/rs6000.md: Remove paired float support.
1618         (move_from_CR_ov_bit): Delete.
1619         * config/rs6000/rs6000.opt (mpaired): Delete.
1620         * config/rs6000/t-rs6000: Remove paired.md from MD_INCLUDES.
1621         * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mpaired.
1622
1623 2018-05-02  Richard Biener  <rguenther@suse.de>
1624
1625         PR middle-end/85567
1626         * gimplify.c (gimplify_save_expr): When in SSA form allow
1627         SAVE_EXPRs to compute to SSA vars.
1628
1629 2018-05-02  Jakub Jelinek  <jakub@redhat.com>
1630
1631         PR target/85582
1632         * config/i386/i386.md (*ashl<dwi>3_doubleword_mask,
1633         *ashl<dwi>3_doubleword_mask_1, *<shift_insn><dwi>3_doubleword_mask,
1634         *<shift_insn><dwi>3_doubleword_mask_1): If and[sq]i3 is needed, don't
1635         clobber operands[2], instead use a new pseudo.  Formatting fixes.
1636
1637 2018-05-02  Richard Sandiford  <richard.sandiford@linaro.org>
1638
1639         PR tree-optimization/85586
1640         * tree-vect-data-refs.c (vect_analyze_data_ref_dependence): Only
1641         exit early for statements in the same group if the accesses are
1642         not strided.
1643
1644 2018-05-02  Tom de Vries  <tom@codesourcery.com>
1645
1646         PR lto/85451
1647         * lto-wrapper.c (compile_offload_image): Add "could not find mkoffload"
1648         error message.
1649
1650 2018-05-01  Marc Glisse  <marc.glisse@inria.fr>
1651
1652         PR tree-optimization/85143
1653         * match.pd (A<B&A<C): Extend to BIT_IOR_EXPR.
1654
1655 2018-05-01  Tom de Vries  <tom@codesourcery.com>
1656
1657         PR lto/85451
1658         * config/nvptx/mkoffload.c (main): Suggest using -B in "offload compiler
1659         not found" error message.
1660
1661 2018-05-01  Tom de Vries  <tom@codesourcery.com>
1662
1663         PR other/83786
1664         * vec.h (VEC_ORDERED_REMOVE_IF, VEC_ORDERED_REMOVE_IF_FROM_TO): Define.
1665         * vec.c (test_ordered_remove_if): New function.
1666         (vec_c_tests): Call test_ordered_remove_if.
1667         * dwarf2cfi.c (connect_traces): Use VEC_ORDERED_REMOVE_IF_FROM_TO.
1668         * lto-streamer-out.c (prune_offload_funcs): Use VEC_ORDERED_REMOVE_IF.
1669         * tree-vect-patterns.c (vect_pattern_recog_1): Use
1670         VEC_ORDERED_REMOVE_IF.
1671
1672 2018-05-01  Prathamesh Kulkarni  <prathamesh.kulkarni@linaro.org>
1673
1674         PR tree-optimization/82665
1675         * vr-values.c (vr_values::extract_range_from_binary_expr): Handle
1676         pointer subtraction where arguments come from a memchr call.
1677
1678 2018-05-01  Jakub Jelinek  <jakub@redhat.com>
1679
1680         * configure.ac (LD_AS_NEEDED_OPTION, LD_NO_AS_NEEDED_OPTION): Use
1681         --push-state --as-needed and --pop-state instead of --as-needed and
1682         --no-as-needed if ld supports it.
1683         * configure: Regenerated.
1684
1685         PR web/85578
1686         * doc/install.texi2html: Replace _002d with - and _002a with * in
1687         generated html files using sed.
1688
1689 2018-04-30  David Malcolm  <dmalcolm@redhat.com>
1690
1691         PR c++/85523
1692         * gcc-rich-location.c (blank_line_before_p): New function.
1693         (use_new_line): New function.
1694         (gcc_rich_location::add_fixit_insert_formatted): New function.
1695         * gcc-rich-location.h
1696         (gcc_rich_location::add_fixit_insert_formatted): New function.
1697
1698 2018-04-30  David Malcolm  <dmalcolm@redhat.com>
1699
1700         * selftest.c (assert_streq): Rename "expected" and "actual" to
1701         "val1" and "val2".  Extend NULL-handling to cover both inputs
1702         symmetrically, while still requiring both to be non-NULL for a pass.
1703         * selftest.h (assert_streq): Rename "expected" and "actual" to
1704         "val1" and "val2".
1705         (ASSERT_EQ): Likewise.
1706         (ASSERT_EQ_AT): Likewise.
1707         (ASSERT_KNOWN_EQ): Likewise.
1708         (ASSERT_KNOWN_EQ_AT): Likewise.
1709         (ASSERT_NE): Likewise.
1710         (ASSERT_MAYBE_NE): Likewise.
1711         (ASSERT_MAYBE_NE_AT): Likewise.
1712         (ASSERT_STREQ): Likewise.  Clarify that both must be non-NULL for
1713         the assertion to pass.
1714         (ASSERT_STREQ_AT): Likewise.
1715
1716 2018-04-30  Jonathan Wakely  <jwakely@redhat.com>
1717
1718         * doc/invoke.texi (-fpie, -fPIE): Fix grammar and clarify
1719         interaction with -pie.
1720
1721 2018-04-30  David Malcolm  <dmalcolm@redhat.com>
1722
1723         * selftest.h: Fix alphabetization of per-source-file selftest
1724         declarations.
1725
1726 2018-04-30  Jason Merrill  <jason@redhat.com>
1727
1728         PR c++/61982 - dead stores to destroyed objects.
1729         * gimplify.c (gimplify_modify_expr): Simplify complex lvalue on LHS
1730         of clobber.
1731
1732 2018-04-30  Jason Merrill  <jason@redhat.com>
1733
1734         * tree.c (build_clobber): New.
1735         * tree.h: Declare it.
1736         * gimplify.c (gimplify_bind_expr, gimplify_target_expr): Use it.
1737
1738 2018-04-30  David Malcolm  <dmalcolm@redhat.com>
1739
1740         * diagnostic-show-locus.c (layout::layout): Update for
1741         location_get_source_line returning a char_span.
1742         (struct char_span): Move to input.h.
1743         (struct correction): Update for fields in char_span becoming
1744         private.
1745         (struct source_line): Update for location_get_source_line
1746         returning a char_span.
1747         (layout::print_line): Likewise.
1748         * edit-context.c (edited_file::print_content): Likewise.
1749         (edited_file::print_diff_hunk): Likewise.
1750         (edited_file::print_run_of_changed_lines): Likewise.
1751         (edited_file::get_num_lines): Likewise.
1752         (edited_line::edited_line): Likewise.
1753         * final.c (asm_show_source): Likewise.
1754         * input.c (location_get_source_line): Convert return type
1755         from const char * to char_span, losing the final "line_len"
1756         param.
1757         (dump_location_info): Update for the above.
1758         (get_substring_ranges_for_loc): Likewise.  Use a char_span
1759         when handling the literal within the line.
1760         (test_reading_source_line): Update for location_get_source_line
1761         returning a char_span.
1762         * input.h (class char_span): Move here from
1763         diagnostic-show-locus.c, converting from a struct to a class.
1764         Make data members private.
1765         (char_span::operator bool): New.
1766         (char_span::length): New.
1767         (char_span::get_buffer): New.
1768         (char_span::operator[]): New.
1769         (char_span::subspan): Make const.
1770         (char_span::xstrdup): New.
1771         (location_get_source_line): Convert return type from const char *
1772         to char_span, losing the final "line_size" param.
1773
1774 2018-04-30  Jan Hubicka  <jh@suse.cz>
1775
1776         * lto-wrapper.c (ltrans_priorities): New static var.
1777         (cmp_priority): New.
1778         (run_gcc): Read priorities and if doing parallel build order
1779         the Makefile by them.
1780
1781 2018-04-30  David Malcolm  <dmalcolm@redhat.com>
1782
1783         * input.h (builtins_location_check): Convert to a STATIC_ASSERT.
1784
1785 2018-04-30  Richard Biener  <rguenther@suse.de>
1786
1787         * tree-cfg.c (verify_address): Remove base argument, add
1788         flag whether to check TREE_ADDRESSABLE and do that.
1789         (verify_expr): Remove.
1790         (verify_types_in_gimple_reference): Add pieces from verify_expr.
1791         (verify_gimple_assign_single): Likewise.
1792         (verify_gimple_switch): Likewise.
1793         (verify_expr_location_1): Dereference tp once.  Add (disabled)
1794         piece from verify_expr.
1795         (verify_gimple_in_cfg): Do not call verify_expr on all ops.
1796
1797 2018-04-30  Claudiu Zissulescu  <claziss@synopsys.com>
1798
1799         * config/arc/linux.h (CLEAR_INSN_CACHE): Define.
1800
1801 2018-04-30  Claudiu Zissulescu  <claziss@synopsys.com>
1802
1803         * config/arc/arc-protos.h (prepare_extend_operands): Remove.
1804         (small_data_pattern): Likewise.
1805         (arc_rewrite_small_data): Likewise.
1806         * config/arc/arc.c (LEGITIMATE_SMALL_DATA_OFFSET_P): Remove.
1807         (LEGITIMATE_SMALL_DATA_ADDRESS_P): Likewise.
1808         (get_symbol_alignment): New function.
1809         (legitimate_small_data_address_p): Likewise.
1810         (legitimate_scaled_address): Update, call
1811         legitimate_small_data_address_p.
1812         (output_sdata): New static variable.
1813         (arc_print_operand): Update how we handle small data operands.
1814         (arc_print_operand_address): Likewise.
1815         (arc_legitimate_address_p): Update, use
1816         legitimate_small_data_address_p.
1817         (arc_rewrite_small_data_p): Remove.
1818         (arc_rewrite_small_data_1): Likewise.
1819         (arc_rewrite_small_data): Likewise.
1820         (small_data_pattern): Likewise.
1821         (compact_sda_memory_operand): Update to use
1822         legitimate_small_data_address_p and get_symbol_alignment.
1823         (prepare_move_operands): Don't rewite sdata pattern.
1824         (prepare_extend_operands): Remove.
1825         * config/arc/arc.md (zero_extendqihi2): Don't rewrite sdata
1826         pattern.
1827         (zero_extendqisi2): Likewise.
1828         (zero_extendhisi2): Likewise.
1829         (extendqihi2): Likewise.
1830         (extendqisi2): Likewise.
1831         (extendhisi2): Likewise.
1832         (addsi3): Likewise.
1833         (subsi3): Likewise.
1834         (andsi3): Likewise.
1835         * config/arc/constraints.md (Usd): Change it to memory constraint.
1836
1837 2018-04-30  Claudiu Zissulescu  <claziss@synopsys.com>
1838
1839         * config/arc/arc.c (arc_split_move): Allow signed 6-bit constants
1840         as source of std instructions.
1841         * config/arc/arc.md (movsi_insn): Update pattern predicate to
1842         allow 6-bit constants as source for store instructions.
1843         (movdi_insn): Update instruction pattern to allow 6-bit constants
1844         as source for store instructions.
1845
1846 2018-04-30  Jonathan Wakely  <jwakely@redhat.com>
1847
1848         * doc/invoke.texi (-fdebug-types-section): Fix grammar.
1849
1850 2018-04-30  Nathan Sidwell  <nathan@acm.org>
1851             Sandra Loosemore <sandra@codesourcery.com>
1852
1853         * dumpfile.c (dump_open): Allow '-' for stdout.
1854         * doc/invoke.texi (Developer Options): Document dump filename
1855         determination early.  Document stdin/stdout selection.
1856
1857 2018-04-30 Andrew Sadek  <andrew.sadek.se@gmail.com>
1858
1859         Microblaze Target: PIC data text relative
1860
1861         * config/microblaze/microblaze.opt: add new option -mpic-data-text-rel.
1862         * config/microblaze/microblaze-protos.h (microblaze_constant_address_p):
1863         Add declaration.
1864         * gcc/config/microblaze/microblaze.h (microblaze_constant_address_p):
1865         CONSTANT_ADDRESS_P definition to microblaze_constant_address_p.
1866         * config/microblaze/microblaze.c (TARGET_PIC_DATA_TEXT_REL):
1867         New addressing mode for data-text relative position indepenedent code.
1868         (microblaze_classify_unspec): add 'UNSPEC_TEXT' case ->
1869         'ADDRESS_SYMBOLIC_TXT_REL'.
1870         (microblaze_classify_address): Add handling for UNSPEC + CONST_INT.
1871         (microblaze_legitimate_pic_operand): Exclude function calls from
1872         pic operands in case of TARGET_PIC_DATA_TEXT_REL option.
1873         (microblaze_legitimize_address): Generate 'UNSPEC_TEXT' for all possible
1874         addresses cases.
1875         (microblaze_address_insns): Add 'ADDRESS_SYMBOLIC_TXT_REL' case.
1876         (print_operand): Add 'ADDRESS_SYMBOLIC_TXT_REL' case.
1877         (print_operand_address): Add 'ADDRESS_SYMBOLIC_TXT_REL' case + handling
1878         for 'address + offset'.
1879         (microblaze_expand_prologue): Add new function prologue call for
1880         'r20' assignation.
1881         (microblaze_asm_generate_pic_addr_dif_vec): Override new target hook
1882         'TARGET_ASM_GENERATE_PIC_ADDR_DIFF_VEC' to disable address diff vector
1883         table in case of TARGET_PIC_DATA_TEXT_REL.
1884         (expand_pic_symbol_ref): Add handling for 'UNSPEC_TEXT'.
1885         * gcc/config/microblaze/microblaze.md (TARGET_PIC_DATA_TEXT_REL):
1886         Add new macros 'UNSPEC_TEXT',
1887         'UNSPEC_SET_TEXT' + add rule for setting r20 in function prologue
1888         + exclude function calls from 'UNSPEC_PLT' in case of data text
1889         relative mode.
1890         * doc/tm.texi.in (TARGET_ASM_GENERATE_PIC_ADDR_DIFF_VEC): Add
1891         new target hook for generating address diff vector tables in case of
1892         flag_pic.
1893         * doc/tm.texi : Regenerate.
1894         * stmt.c (TARGET_ASM_GENERATE_PIC_ADDR_DIFF_VEC): Append new condition
1895         'targetm.asm_out.generate_pic_addr_diff_vec' to flag_pic in case
1896         of addr diff vector generation.
1897         * target.def (TARGET_ASM_GENERATE_PIC_ADDR_DIFF_VEC): Add
1898         target hook definition.
1899         * targhooks.h, gcc/targhooks.c (TARGET_ASM_GENERATE_PIC_ADDR_DIFF_VEC):
1900         Add default function for generate_pic_addr_diff_vec -> flag_pic.
1901         * doc/invoke.texi (Add new pic option): Add new microblaze pic
1902         option for data text relative.
1903
1904 2018-04-30  Richard Biener  <rguenther@suse.de>
1905
1906         * tree-chrec.h (evolution_function_is_constant_p): Remove
1907         redundant check.
1908         * tree-cfg.c (tree_node_can_be_shared): Re-order checks.
1909
1910 2018-04-30  Richard Biener  <rguenther@suse.de>
1911
1912         PR bootstrap/85571
1913         * dwarf2out.c (gen_producer_string): Ignore -fchecking[=].
1914
1915 2018-04-30  Richard Biener  <rguenther@suse.de>
1916
1917         PR tree-optimization/28364
1918         PR tree-optimization/85275
1919         * tree-ssa-loop-ch.c (ch_base::copy_headers): Stop after
1920         copying first exit test.
1921
1922 2018-04-28  Mark Wielaard  <mark@klomp.org>
1923
1924         * dwarf2out.c (dwarf2out_finish): Add .debug_addr table header for
1925         dwarf_version >= 5.
1926         (dwarf_AT): Handle DW_AT_addr_base.
1927         (add_top_level_skeleton_die_attrs): Use dwarf_AT for DW_AT_addr_base.
1928
1929 2018-04-28  Uros Bizjak  <ubizjak@gmail.com>
1930
1931         PR target/84431
1932         * config/i386/i386.md (*ashl<dwi>3_doubleword_mask): New pattern.
1933         (*ashl<dwi>3_doubleword_mask_1): Ditto.
1934         (*<shift_insn><dwi>3_doubleword_mask): Ditto.
1935         (*<shift_insn><dwi>3_doubleword_mask_1): Ditto.
1936
1937 2018-04-28  Richard Biener  <rguenther@suse.de>
1938
1939         * tree-cfg.c (verify_gimple_phi): Take a gphi * argument.
1940         (verify_gimple_in_cfg): Rename visited_stmts to visited_throwing_stmts
1941         to reflect use.  Only add interesting stmts.
1942
1943 2018-04-27  Martin Jambor  <mjambor@suse.cz>
1944
1945         PR ipa/85549
1946         * ipa-cp.c (find_aggregate_values_for_callers_subset): Make sure
1947         the jump function allows for passing through aggregate values.
1948
1949 2018-04-27  David Malcolm  <dmalcolm@redhat.com>
1950
1951         * input.h (in_system_header_at): Convert from macro to inline
1952         function.
1953         (from_macro_expansion_at): Likewise.
1954         (from_macro_definition_at): Likewise.
1955
1956 2018-04-27  Jeff Law  <law@redhat.com>
1957
1958         * config.gcc: Mark tile* targets as deprecated/obsolete.
1959
1960 2018-04-27  Richard Biener  <rguenther@suse.de>
1961
1962         * config/aarch64/aarch64.c: Simplify ap.__stack advance and
1963         fix for ILP32.
1964
1965 2018-04-27  Richard Biener  <rguenther@suse.de>
1966
1967         * tree-cfg.c (verify_expr): Make dead code hit gcc_unreachable.
1968
1969 2018-04-27  Uros Bizjak  <ubizjak@gmail.com>
1970
1971         * config/i386/i386.md (*movti_internal): Substitute Ye constraint
1972         with Yd constraint. Set "preferred_for_speed" attribute from
1973         TARGET_INTER_UNIT_MOVES_{FROM,TO}_VEC for alternatives
1974         with Yd constraint.
1975         (*movdi_internal): Ditto.
1976         (movti_interunit splitters): Remove
1977         TARGET_INTER_UNIT_MOVES_{FROM,TO}_VEC from insn condition.
1978         (movdi_interunit splitters): Ditto.
1979         * config/i386/constraints.md (Ye): Remove.
1980         (Yd): Do not depend on TARGET_INTER_UNIT_MOVES_{FROM,TO}_VEC.
1981
1982 2018-04-27  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
1983
1984         PR target/85512
1985         * config/aarch64/constraints.md (Usg): Limit to 31.
1986         (Usj): Limit to 63.
1987
1988 2018-04-27  Jakub Jelinek  <jakub@redhat.com>
1989
1990         PR tree-optimization/85529
1991         * tree-ssa-reassoc.c (optimize_range_tests_var_bound): Add FIRST_BB
1992         argument.  Don't call get_nonzero_bits if opcode is ERROR_MARK_NODE,
1993         rhs2 def stmt's bb is dominated by first_bb and it isn't an obvious
1994         zero extension or masking of the MSB bit.
1995         (optimize_range_tests): Add FIRST_BB argument, pass it through
1996         to optimize_range_tests_var_bound.
1997         (maybe_optimize_range_tests, reassociate_bb): Adjust
1998         optimize_range_tests callers.
1999
2000 2018-04-26  Richard Biener  <rguenther@suse.de>
2001             Jakub Jelinek  <jakub@redhat.com>
2002
2003         * cgraph.h (symbol_table): Just declare debug method here.
2004         * symtab.c (symbol_table::debug): Define.
2005
2006 2018-04-26  Eric Botcazou  <ebotcazou@adacore.com>
2007
2008         * loop-invariant.c (may_assign_reg_p): Return false for frame pointer.
2009
2010 2018-04-26  Uros Bizjak  <ubizjak@gmail.com>
2011
2012         * config/i386/i386.md ("isa" attribute): Add x64_sse2.
2013         ("enabled" attribute): Handle x64_sse2 "isa" attribute.
2014         (*movdi_internal): Substitute Yi and Yj constraint with x
2015         and Ym and Yn constraint with y constraint.  Update "isa"
2016         attribute and set "preferred_for_speed" attribute from
2017         TARGET_INTER_UNIT_MOVES_{FROM,TO}_VEC for updated alternatives.
2018         (*movsi_internal): Ditto.
2019         (*movdf_internal): Ditto.
2020         (*movsf_internal): Ditto.
2021         (*zero_extendsidi2): Ditto.
2022         * config/i386/sse.md (vec_set<mode>_0): Ditto.
2023         (sse2_loadld): Ditto.
2024         (*vec_extract<ssevecmodelower>_0): Ditto.
2025         (*vec_extractv4si_0_zext_sse4): Ditto.
2026         (vec_concatv2di): Ditto.
2027         (*vec_dup<mode>): Ditto.
2028         * config/i386/mmx.md (*mov<mode>_internal): Ditto.
2029         * config/i386/constraints.md (Yi): Remove.
2030         (Yj): Remove.
2031         (Ym): Remove.
2032         (Yn): Remove.
2033
2034 2018-04-26  Nathan Sidwell  <nathan@acm.org>
2035
2036         * dumpfile.c (dump_open): New.
2037         (dump_open_alternate_stream, dump_start, dump_begin): Call it.
2038         (dump_finish): Detect stdio/stderr by value not name.
2039
2040 2018-04-26  Jonathan Wakely  <jwakely@redhat.com>
2041
2042         * doc/invoke.texi (-Wreturn-type): Document default status for C++.
2043
2044 2018-04-26  Tom de Vries  <tom@codesourcery.com>
2045
2046         PR target/84952
2047         * config/nvptx/nvptx.c (verify_neutering_jumps)
2048         (verify_neutering_labels): New function
2049         (nvptx_single): Use verify_neutering_jumps and verify_neutering_labels.
2050
2051 2018-04-26  Tom de Vries  <tom@codesourcery.com>
2052
2053         PR target/84025
2054         * config/nvptx/nvptx.c (needs_neutering_p): New function.
2055         (nvptx_single): Use needs_neutering_p to skip over insns that do not
2056         need neutering.
2057
2058 2018-04-26  Richard Biener <rguenther@suse.de>
2059             Tom de Vries  <tom@codesourcery.com>
2060
2061         PR lto/85422
2062         * lto-streamer-out.c (output_function): Fixup loops if required to match
2063         discovery done in the reader.
2064
2065 2018-04-26  Richard Biener  <rguenther@suse.de>
2066
2067         PR tree-optimization/85116
2068         * tree-ssa-loop-ch.c (do_while_loop_p): A do-while loop should
2069         have a loop exit from the single latch predecessor.  Remove
2070         case of header with just condition.
2071         (ch_base::copy_headers): Exclude infinite loops from any
2072         processing.
2073         (pass_ch::execute): Record exits.
2074
2075 2018-04-26  Richard Biener  <rguenther@suse.de>
2076
2077         * tree-vect-data-refs.c (vect_get_data_access_cost): Get
2078         prologue cost vector and pass it to vect_get_load_cost.
2079         (vect_get_peeling_costs_all_drs): Likewise.
2080         (vect_peeling_hash_get_lowest_cost): Likewise.
2081         (vect_enhance_data_refs_alignment): Likewise.
2082
2083 2018-04-26  Richard Biener  <rguenther@suse.de>
2084
2085         PR middle-end/85450
2086         * tree-cfg.c (verify_gimple_assign_unary): Restore proper
2087         checking of integer<->pointer conversions.
2088         * omp-expand.c (expand_omp_for_static_nochunk): Avoid
2089         sign-/zero-extending pointer types.
2090         (expand_omp_for_static_chunk): Likewise.
2091
2092 2018-03-22  Hans-Peter Nilsson  <hp@axis.com>
2093             Jean Lee  <xiaoyur347@gmail.com>
2094
2095         * config/mips/mips.c (mips_asan_shadow_offset): New function.
2096         (TARGET_ASAN_SHADOW_OFFSET): Define.
2097         * config/mips/mips.h (FRAME_GROWS_DOWNWARD): Augment to also be
2098         true for -fsanitize=address.
2099
2100 2018-04-25  Mark Wielaard  <mark@klomp.org>
2101
2102         * dwarf2out.c (file_info_cmp): Sort longer dir prefixes before
2103         shorter ones.
2104
2105 2018-04-25  Jakub Jelinek  <jakub@redhat.com>
2106
2107         * config/i386/i386.md (*x86_mov<mode>cc_0_m1): Use type "alu1" rather
2108         than "alu", remove explicit "memory" and "imm_disp" attributes.
2109         (*x86_mov<mode>cc_0_m1_se, *x86_mov<mode>cc_0_m1_neg): Likewise.
2110
2111         PR middle-end/85414
2112         * simplify-rtx.c (simplify_unary_operation_1) <case SIGN_EXTEND,
2113         case ZERO_EXTEND>: Pass SUBREG_REG (op) rather than op to
2114         gen_lowpart_no_emit.
2115
2116 2018-04-25  Sebastian Peryt  <sebastian.peryt@intel.com>
2117
2118         PR target/85473
2119         * config/i386/i386.c (ix86_expand_builtin): Change memory
2120         operand to XI, extend p0 to Pmode.
2121         * config/i386/i386.md: Change unspec volatile and operand
2122         1 mode to XI, change operand 0 mode to P.
2123
2124 2018-04-25  Chung-Ju Wu  <jasonwucj@gmail.com>
2125
2126         * config/nds32/nds32-predicates.c (nds32_can_use_bclr_p): Mask with
2127         GET_MODE_MASK before any checking.
2128         (nds32_can_use_bset_p): Likewise.
2129         (nds32_can_use_btgl_p): Likewise.
2130
2131 2018-04-25  Chung-Ju Wu  <jasonwucj@gmail.com>
2132
2133         * config/nds32/nds32-doubleword.md: New define_split pattern for
2134         illegal register number.
2135
2136 2018-04-25  Chung-Ju Wu  <jasonwucj@gmail.com>
2137
2138         * config/nds32/nds32.c (nds32_print_operand): Set op_value ealier.
2139
2140 2018-04-25  Chung-Ju Wu  <jasonwucj@gmail.com>
2141
2142         * config/nds32/nds32.h (ASM_APP_ON): Add missing newline character.
2143
2144 2018-04-25  Richard Biener  <rguenther@suse.de>
2145
2146         * lto-streamer.h (LTO_major_version): Bump to 8.
2147
2148 2018-04-25  Jakub Jelinek  <jakub@redhat.com>
2149
2150         * BASE-VER: Set to 9.0.0.
2151
2152 2018-04-24  Segher Boessenkool  <segher@kernel.crashing.org>
2153
2154         * config/rs6000/rs6000.c (init_float128_ieee): Fix spelling mistakes
2155         in __abskf2 and __powikf2.
2156
2157 2018-04-24  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
2158
2159         PR target/85512
2160         * config/aarch64/constraints.md (Usg, Usj): New constraints.
2161         * config/aarch64/iterators.md (cmode_simd): New mode attribute.
2162         * config/aarch64/aarch64.md (*aarch64_ashr_sisd_or_int_<mode>3):
2163         Use the above on operand 2.  Reindent.
2164         (*aarch64_lshr_sisd_or_int_<mode>3): Likewise.
2165
2166 2018-04-24  H.J. Lu  <hongjiu.lu@intel.com>
2167
2168         PR target/85485
2169         * common/config/i386/i386-common.c (ix86_handle_option): Don't
2170         handle OPT_mcet.
2171         * config/i386/i386.opt (mcet): Removed.
2172         * doc/install.texi: Remove -mcet documentation.
2173         * doc/invoke.texi: Likewise.
2174
2175 2018-04-24  H.J. Lu  <hongjiu.lu@intel.com>
2176
2177         PR target/85485
2178         * doc/install.texi: Remove -mcet from bootstrap-cet.
2179
2180 2018-04-24  Jakub Jelinek  <jakub@redhat.com>
2181
2182         PR target/85511
2183         * config/i386/i386.c (ix86_init_mmx_sse_builtins): Don't define
2184         __builtin_ia32_readeflags_u32 and __builtin_ia32_writeeflags_u32
2185         if TARGET_64BIT.
2186
2187         PR target/85503
2188         * config/rs6000/rs6000-p8swap.c (const_load_sequence_p): Punt if
2189         const_vector is not CONST_VECTOR or SYMBOL_REF for a constant pool
2190         containing a CONST_VECTOR.
2191
2192 2018-04-24  Cesar Philippidis  <cesar@codesourcery.com>
2193
2194         * doc/install.texi: Update newlib dependency for nvptx.
2195
2196 2018-04-24  Jakub Jelinek  <jakub@redhat.com>
2197
2198         PR target/85508
2199         * config/i386/i386.c (ix86_expand_vector_init_one_var): Use UINTVAL
2200         instead of INTVAL when shifting x left.
2201
2202 2018-04-24  Andreas Krebbel  <krebbel@linux.ibm.com>
2203
2204         PR tree-optimization/85478
2205         * tree-vect-loop.c (vect_analyze_loop_2): Do not call
2206         vect_grouped_store_supported for single element vectors.
2207
2208 2018-04-24  Richard Biener  <rguenther@suse.de>
2209
2210         PR target/85491
2211         * config/i386/i386.c (ix86_add_stmt_cost): Restrict strided
2212         load cost increase to the case of non-constant step.
2213
2214 2018-04-24  Jakub Jelinek  <jakub@redhat.com>
2215
2216         PR target/84828
2217         * reg-stack.c (move_for_stack_reg): Don't fail assertion about dead
2218         destination if any_malformed_asm.
2219
2220 2018-04-23  Eric Botcazou  <ebotcazou@adacore.com>
2221
2222         PR middle-end/85496
2223         * expr.c (store_field): In the bitfield case, if the value comes from
2224         a function call and is returned in registers by means of a PARALLEL,
2225         do not change the mode of the temporary unless BLKmode and VOIDmode.
2226
2227 2018-04-23  Andrey Belevantsev  <abel@ispras.ru>
2228
2229         PR rtl-optimization/85423
2230         * sel-sched-ir.c (has_dependence_note_mem_dep): Only discard
2231         dependencies to debug insns when the previous insn is non-debug.
2232
2233 2018-04-23  Claudiu Zissulescu  <claziss@synopsys.com>
2234
2235         * config/arc/arc.md ("vunspec"): Delete it, unify all the unspec
2236         enums into a single definition.
2237         (fls): Fix predicates and printing.
2238         (seti): Likewise.
2239
2240 2018-04-23  Claudiu Zissulescu  <claziss@synopsys.com>
2241
2242         * config/arc/arc-protos.h (check_if_valid_sleep_operand): Remove.
2243         * config/arc/arc.c (arc_expand_builtin): Sleep accepts registers
2244         and short u6 immediate.
2245         (check_if_valid_sleep_operand): Remove.
2246         * config/arc/arc.md (Sleep): Accepts registers and u6 immediates.
2247
2248 2018-04-22  Chung-Ju Wu  <jasonwucj@gmail.com>
2249
2250         * config/nds32/nds32.c (nds32_compute_stack_frame): Consider
2251         flag_always_save_lp condition.
2252         * config/nds32/nds32.opt (malways-save-lp): New option.
2253
2254 2018-04-22  Shiva Chen  <shiva0217@gmail.com>
2255
2256         * config/nds32/nds32-protos.h (nds32_use_load_post_increment): Declare.
2257         * config/nds32/nds32.c (nds32_use_load_post_increment): New.
2258         * config/nds32/nds32.h
2259         (USE_LOAD_POST_INCREMENT, USE_LOAD_POST_DECREMENT): Define.
2260         (USE_STORE_POST_INCREMENT, USE_STORE_POST_DECREMENT): Define.
2261
2262 2018-04-22  Shiva Chen  <shiva0217@gmail.com>
2263
2264         * config/nds32/nds32-protos.h (nds32_ls_333_p): Remove.
2265         * config/nds32/nds32.c (nds32_ls_333_p): Remove.
2266
2267 2018-04-22  Shiva Chen  <shiva0217@gmail.com>
2268             Chung-Ju Wu  <jasonwucj@gmail.com>
2269
2270         * config/nds32/nds32-protos.h (nds32_case_vector_shorten_mode):
2271         Declare.
2272         * config/nds32/nds32.c (nds32_case_vector_shorten_mode): New function.
2273         * config/nds32/nds32.h (CASE_VECTOR_SHORTEN_MODE): Modify.
2274
2275 2018-04-22  Chung-Ju Wu  <jasonwucj@gmail.com>
2276
2277         * config/nds32/nds32.c (nds32_compute_stack_frame): Fix wrong value.
2278
2279 2018-04-22  Chung-Ju Wu  <jasonwucj@gmail.com>
2280
2281         * config/nds32/nds32-protos.h (nds32_data_alignment,
2282         nds32_local_alignment): Declare.
2283         * config/nds32/nds32.c (nds32_data_alignment, nds32_constant_alignment,
2284         nds32_local_alignment): New functions.
2285         (TARGET_CONSTANT_ALIGNMENT): Define.
2286         * config/nds32/nds32.h (DATA_ALIGNMENT, LOCAL_ALIGNMENT): Define.
2287
2288 2018-04-22  Chung-Ju Wu  <jasonwucj@gmail.com>
2289
2290         * config/nds32/nds32.c
2291         (TARGET_HARD_REGNO_MODE_OK): Move to the bottom of file.
2292         (TARGET_MODES_TIEABLE_P): Likewise.
2293
2294 2018-04-22  Chung-Ju Wu  <jasonwucj@gmail.com>
2295
2296         * config/nds32/nds32.c (nds32_asm_file_start): Display optimization
2297         level Ofast and Og.
2298
2299 2018-04-22  Monk Chiang  <sh.chiang04@gmail.com>
2300             Chung-Ju Wu  <jasonwucj@gmail.com>
2301
2302         * config/nds32/constants.md (unspec_volatile_element): Add enum values
2303         for unaligned access.
2304         * config/nds32/nds32-intrinsic.c: Implementation of expanding
2305         unaligned access.
2306         * config/nds32/nds32-intrinsic.md: Likewise.
2307         * config/nds32/nds32_intrinsic.h: Likewise.
2308         * config/nds32/nds32.h (nds32_builtins): Likewise.
2309         * config/nds32/nds32.opt (munaligned-access): New option.
2310         * config/nds32/nds32.c (nds32_asm_file_start): Display
2311         flag_unaligned_access status.
2312
2313 2018-04-20  Kito Cheng  <kito.cheng@gmail.com>
2314
2315         * config/riscv/elf.h (LINK_SPEC): Pass --no-relax if
2316         -mno-relax is present.
2317         * config/riscv/linux.h (LINK_SPEC): Ditto.
2318
2319 2018-04-20  Martin Sebor  <msebor@redhat.com>
2320
2321         PR c/85365
2322         * gimple-fold.c (gimple_fold_builtin_strcpy): Suppress -Wrestrict
2323         for null pointers.
2324         (gimple_fold_builtin_stxcpy_chk): Same.
2325         * gimple-ssa-warn-restrict.c (check_bounds_or_overlap): Same.
2326
2327 2018-04-20  Michael Meissner  <meissner@linux.ibm.com>
2328
2329         PR target/85456
2330         * config/rs6000/rs6000.c (init_float128_ieee): Add support to call
2331         __powikf2 when long double is IEEE 128-bit.
2332
2333 2018-04-20  Kito Cheng  <kito.cheng@gmail.com>
2334
2335         * config/riscv/riscv.c (riscv_first_stack_step): Round up min
2336         step to make sure stack always aligned.
2337
2338 2018-04-20  Carl Love  <cel@us.ibm.com>
2339
2340         PR target/83402
2341         * config/rs6000/rs6000-c.c (rs6000_gimple_fold_builtin): Add
2342         size check for arg0.
2343
2344 2018-04-20  Nathan Sidwell  <nathan@codesourcery.com>
2345             Tom de Vries  <tom@codesourcery.com>
2346
2347         PR target/85445
2348         * config/nvptx/nvptx.c (nvptx_emit_forking, nvptx_emit_joining):
2349         Emit insns for calls too.
2350         (nvptx_find_par): Always look for worker-level predecessor insn.
2351         (nvptx_propagate): Add is_call parm, return bool.  Copy frame for
2352         calls.
2353         (nvptx_vpropagate, nvptx_wpropagate): Adjust.
2354         (nvptx_process_pars): Propagate frames for calls.
2355
2356 2018-04-20  H.J. Lu  <hongjiu.lu@intel.com>
2357
2358         PR target/85469
2359         * common/config/i386/i386-common.c (OPTION_MASK_ISA_IBT_SET):
2360         Removed.
2361         (OPTION_MASK_ISA_IBT_UNSET): Likewise.
2362         (ix86_handle_option): Don't handle OPT_mibt.
2363         * config/i386/cet.h: Check __CET__ instead of __IBT__ and
2364         __SHSTK__.
2365         * config/i386/driver-i386.c (host_detect_local_cpu): Remove
2366         has_ibt and ibt.
2367         * config/i386/i386-c.c (ix86_target_macros_internal): Don't
2368         check OPTION_MASK_ISA_IBT nor flag_cf_protection.
2369         (ix86_target_macros): Define __CET__ with flag_cf_protection
2370         for -fcf-protection.
2371         * config/i386/i386.c (isa2_opts): Remove -mibt.
2372         * config/i386/i386.h (TARGET_IBT): Removed.
2373         (TARGET_IBT_P): Likewise.
2374         (ix86_valid_target_attribute_inner_p): Don't check OPT_mibt.
2375         * config/i386/i386.md (nop_endbr): Don't check TARGET_IBT.
2376         * config/i386/i386.opt (mcet): Update help message.
2377         (mshstk): Likewise.
2378         (mibt): Removed.
2379         * doc/invoke.texi: Remove -mibt.  Document __CET__.  Document
2380         -mcet as an alias for -mshstk.
2381
2382 2018-04-20  Richard Biener <rguenther@suse.de>
2383
2384         PR middle-end/85475
2385         * match.pd ((X * CST) * Y -> (X * Y) * CST): Avoid exponential
2386         complexity by forcing a single use of the multiply operand.
2387
2388 2018-04-20  Martin Jambor  <mjambor@suse.cz>
2389
2390         ipa/85449
2391         * ipa-cp.c (cgraph_edge_brings_value_p): Move check for self-feeding
2392         recursion dependency to only apply to non-clones.
2393
2394 2018-04-20  Martin Jambor  <mjambor@suse.cz>
2395
2396         ipa/85447
2397         * ipa-cp.c (create_specialized_node): Check that clones of
2398         self-recursive edges exist during IPA-CP.
2399
2400 2018-04-19  Toon Moene  <toon@moene.org>
2401
2402         * doc/invoke.texi: Add -floop-unroll-and-jam to options enabled
2403         by -O3.
2404
2405 2018-04-19  Jakub Jelinek  <jakub@redhat.com>
2406
2407         PR tree-optimization/85467
2408         * fold-const.c (fold_ternary_loc) <case BIT_FIELD_REF>: Use
2409         VECTOR_TYPE_P macro.  If type is vector type, VIEW_CONVERT_EXPR the
2410         VECTOR_CST element to type.
2411
2412 2018-04-19  H.J. Lu  <hongjiu.lu@intel.com>
2413
2414         PR target/85397
2415         * config/i386/i386.h (STACK_SAVEAREA_MODE): New.
2416         * config/i386/i386.md (builtin_setjmp_setup): Removed.
2417         (builtin_longjmp): Likewise.
2418         (save_stack_nonlocal): New pattern.
2419         (restore_stack_nonlocal): Likewise.
2420
2421 2018-04-19  H.J. Lu  <hongjiu.lu@intel.com>
2422
2423         PR target/85404
2424         * config/i386/cet.c (file_end_indicate_exec_stack_and_cet):
2425         Replace ASM_OUTPUT_LABEL with fprintf.
2426
2427 2018-04-19  H.J. Lu  <hongjiu.lu@intel.com>
2428
2429         PR target/85417
2430         * config/i386/cet.c (file_end_indicate_exec_stack_and_cet):
2431         Check flag_cf_protection instead of TARGET_IBT and TARGET_SHSTK.
2432         * config/i386/i386-c.c (ix86_target_macros_internal): Also
2433         define __IBT__ and __SHSTK__ for -fcf-protection.
2434         * config/i386/i386.c (pass_insert_endbranch::gate): Don't check
2435         TARGET_IBT.
2436         (ix86_trampoline_init): Likewise.
2437         (x86_output_mi_thunk): Likewise.
2438         (ix86_notrack_prefixed_insn_p): Likewise.
2439         (ix86_option_override_internal): Don't disallow -fcf-protection.
2440         * config/i386/i386.md (rdssp<mode>): Also enable for
2441         -fcf-protection.
2442         (incssp<mode>): Likewise.
2443         (nop_endbr): Likewise.
2444         * config/i386/i386.opt (mcet): Change help message to built-in
2445         functions only.
2446         (mibt): Likewise.
2447         (mshstk): Likewise.
2448         * doc/invoke.texi: Remove -mcet, -mibt and -mshstk condition
2449         on -fcf-protection.  Change -mcet, -mibt and -mshstk to only
2450         enable CET built-in functions.
2451
2452 2018-04-19  Sebastian Peryt  <sebastian.peryt@intel.com>
2453
2454         * common/config/i386/i386-common.c
2455         (OPTION_MASK_ISA_MOVDIRI_SET, OPTION_MASK_ISA_MOVDIR64B_SET,
2456         OPTION_MASK_ISA_MOVDIRI_UNSET,
2457         OPTION_MASK_ISA_MOVDIR64B_UNSET): New defines.
2458         (ix86_handle_option): Handle -mmovdiri and -mmovdir64b.
2459         * config.gcc (movdirintrin.h): New header.
2460         * config/i386/cpuid.h (bit_MOVDIRI,
2461         bit_MOVDIR64B): New bits.
2462         * config/i386/driver-i386.c (host_detect_local_cpu): Detect -mmovdiri
2463         and -mmvodir64b.
2464         * config/i386/i386-builtin-types.def ((VOID, PUNSIGNED, UNSIGNED),
2465         (VOID, PVOID, PCVOID)): New function types.
2466         * config/i386/i386-builtin.def (__builtin_ia32_directstoreu_u32,
2467         __builtin_ia32_directstoreu_u64,
2468          __builtin_ia32_movdir64b): New builtins.
2469         * config/i386/i386-c.c (__MOVDIRI__, __MOVDIR64B__): New.
2470         * config/i386/i386.c (ix86_target_string): Added -mmovdir64b
2471         and -mmovdiri.
2472         (ix86_valid_target_attribute_inner_p): Ditto.
2473         (ix86_expand_special_args_builtin): Added VOID_FTYPE_PUNSIGNED_UNSIGNED
2474         and VOID_FTYPE_PUNSIGNED_UNSIGNED.
2475         (ix86_expand_builtin): Expand IX86_BUILTIN_MOVDIR64B.
2476         * config/i386/i386.h (TARGET_MOVDIRI, TARGET_MOVDIRI_P,
2477         TARGET_MOVDIR64B, TARGET_MOVDIR64B_P): New.
2478         * config/i386/i386.md (UNSPECV_MOVDIRI, UNSPECV_MOVDIR64B): New.
2479         (movdiri<mode>, movdir64b_<mode>): New.
2480         * config/i386/i386.opt: Add -mmovdiri and -mmovdir64b.
2481         * config/i386/immintrin.h: Include movdirintrin.h.
2482         * config/i386/movdirintrin.h: New file.
2483         * doc/invoke.texi: Added -mmovdiri and -mmovdir64b.
2484
2485 2018-04-19  Richard Biener  <rguenther@suse.de>
2486
2487         PR middle-end/85455
2488         * cfg.c (clear_bb_flags): When loop state says we have
2489         marked irreducible regions also preserve BB_IRREDUCIBLE_LOOP.
2490
2491 2018-04-19  Richard Biener  <rguenther@suse.de>
2492
2493         PR tree-optimization/84737
2494         * tree-vect-data-refs.c (vect_copy_ref_info): New function
2495         copying restrict info.
2496         (vect_setup_realignment): Use it.
2497         * tree-vectorizer.h (vect_copy_ref_info): Declare.
2498         * tree-vect-stmts.c (vectorizable_store): Copy ref info from
2499         the first DR to all generated stores.
2500         (vectorizable_load): Likewise for loads.
2501
2502 2018-04-19  Jakub Jelinek  <jakub@redhat.com>
2503
2504         PR tree-optimization/85446
2505         * match.pd ((intptr_t) x eq/ne CST to x eq/ne (typeof x) cst): Require
2506         the integral and pointer types to have the same precision.
2507
2508         * doc/install.texi: Document --disable-cet being the default and
2509         --enable-cet=auto.
2510
2511 2018-04-18  Martin Liska  <mliska@suse.cz>
2512
2513         * ipa-devirt.c (odr_subtypes_equivalent_p): Fix GNU coding
2514         style.
2515
2516 2018-04-18  Martin Liska  <mliska@suse.cz>
2517
2518         Revert
2519         2018-03-02  Eric Botcazou  <ebotcazou@adacore.com>
2520
2521         PR ipa/83983
2522         * ipa-devirt.c (odr_subtypes_equivalent_p): Get the ODR type of both
2523         arguments if they are comparable.
2524
2525 2018-04-18  Martin Liska  <mliska@suse.cz>
2526
2527         Revert
2528         2018-03-13  Eric Botcazou  <ebotcazou@adacore.com>
2529
2530         PR lto/84805
2531         * ipa-devirt.c (odr_subtypes_equivalent_p): Do not get the ODR type of
2532         incomplete types.
2533
2534 2018-04-18  H.J. Lu  <hongjiu.lu@intel.com>
2535
2536         PR target/85388
2537         * config/i386/i386.c (ix86_expand_split_stack_prologue): Insert
2538         ENDBR after calling __morestack.
2539
2540 2018-04-18  David Malcolm  <dmalcolm@redhat.com>
2541
2542         PR jit/85384
2543         * configure.ac (gcc-driver-name.h): Honor --with-gcc-major-version
2544         by using gcc_base_ver to generate a gcc_driver_version, and use
2545         it when generating GCC_DRIVER_NAME.
2546         * configure: Regenerate.
2547
2548 2018-04-18  Jakub Jelinek  <jakub@redhat.com>
2549
2550         PR target/81084
2551         * config.gcc: Obsolete powerpc*-*-*spe*.
2552
2553 2018-04-17  Jakub Jelinek  <jakub@redhat.com>
2554
2555         PR debug/84637
2556         * dbxout.c (dbxout_int): Perform negation in unsigned int type.
2557         (stabstr_D): Change type of unum from unsigned int to
2558         unsigned HOST_WIDE_INT.  Perform negation in unsigned HOST_WIDE_INT
2559         type.
2560
2561 2018-04-17  Jim Wilson  <jimw@sifive.com>
2562
2563         PR 84856
2564         * config/riscv/riscv.c (riscv_compute_frame_info): Add calls to
2565         RISCV_STACK_ALIGN when using outgoing_args_size and pretend_args_size.
2566         Set arg_pointer_offset after using pretend_args_size.
2567
2568 2018-04-17  Jakub Jelinek  <jakub@redhat.com>
2569
2570         PR rtl-optimization/85431
2571         * dse.c (record_store): Ignore zero width stores.
2572
2573         PR sanitizer/85230
2574         * asan.c (handle_builtin_stack_restore): Adjust comment.  Emit
2575         __asan_allocas_unpoison call and last_alloca_addr = new_sp before
2576         __builtin_stack_restore rather than after it.
2577         * builtins.c (expand_asan_emit_allocas_unpoison): Pass
2578         arg1 + (virtual_dynamic_stack_rtx - stack_pointer_rtx) as second
2579         argument instead of virtual_dynamic_stack_rtx.
2580
2581 2018-04-17  Kelvin Nilsen  <kelvin@gcc.gnu.org>
2582
2583         * config/rs6000/rs6000-protos.h (rs6000_builtin_is_supported_p):
2584         New prototype.
2585         * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin):
2586         Add note to error message to explain internal mapping of overloaded
2587         built-in function name to non-overloaded built-in function name.
2588         * config/rs6000/rs6000.c (rs6000_builtin_is_supported_p): New
2589         function.
2590
2591 2018-04-17  Michael Meissner  <meissner@linux.vnet.ibm.com>
2592
2593         PR target/85424
2594         * config/rs6000/rs6000.md (pack<mode>): Do not try handle a pack
2595         where the inputs overlap with the output.
2596
2597 2018-04-17  Jakub Jelinek  <jakub@redhat.com>
2598
2599         * config/i386/sse.md (vec_extract_lo_<mode><mask_name>): Add
2600         (=v, v) alternative and explicit "memory" attribute.
2601         (vec_extract_lo_<mode><mask_name>): Likewise.  Also add
2602         "type", "prefix", "prefix_extra", "length_immediate" and "mode"
2603         attributes.
2604         (vec_extract_lo_<mode><mask_name>): Add (=v, v) alternative and use
2605         "sselog1" type instead of "sselog".
2606         (vec_extract_hi_<mode><mask_name>): Use "sselog1" type instead of
2607         "sselog".  Remove explicit "memory" attribute.
2608         (vec_extract_lo_v32hi): Add (=v, v) alternative and explicit "memory",
2609         "type", "prefix", "prefix_extra", "length_immediate" and "mode"
2610         attributes.
2611         (vec_extract_hi_v32hi): Merge all alternatives into one, use
2612         "sselog1" type instead of "sselog".  Remove explicit "memory"
2613         attribute.
2614         (vec_extract_hi_v16hi): Merge each pair of alternatives into one,
2615         use "sselog1" type instead of "sselog".  Remove explicit "memory"
2616         attribute.
2617         (vec_extract_lo_v64qi): Add (=v, v) alternative and explicit "memory",
2618         "type", "prefix", "prefix_extra", "length_immediate" and "mode"
2619         attributes.
2620         (vec_extract_hi_v64qi): Merge all alternatives into one, use
2621         "sselog1" type instead of "sselog".  Remove explicit "memory"
2622         attribute.
2623         (vec_extract_hi_v32qi): Merge each pair of alternatives into one,
2624         use "sselog1" type instead of "sselog".  Remove explicit "memory"
2625         attribute.
2626
2627         PR target/85430
2628         * config/i386/i386.md (*ashlqi3_1_slp): Use alu1 type instead of alu.
2629
2630         PR middle-end/85414
2631         * rtlhooks.c (gen_lowpart_if_possible): Don't call gen_lowpart_SUBREG
2632         on a SUBREG.
2633
2634 2018-04-17  Martin Jambor  <mjambor@suse.cz>
2635
2636         PR ipa/85421
2637         * ipa-cp.c (create_specialized_node): Call
2638         expand_all_artificial_thunks if necessary.
2639
2640 2018-04-17  Martin Liska  <mliska@suse.cz>
2641
2642         PR lto/85405
2643         * ipa-devirt.c (odr_types_equivalent_p): Remove trailing
2644         in message, remote space in between '_G' and '('.
2645
2646 2018-04-17  Jakub Jelinek  <jakub@redhat.com>
2647
2648         PR target/85281
2649         * config/i386/sse.md (reduces<mode><mask_scalar_name>,
2650         avx512f_vmcmp<mode>3<round_saeonly_name>,
2651         avx512f_vmcmp<mode>3_mask<round_saeonly_name>,
2652         avx512f_sgetexp<mode><mask_scalar_name><round_saeonly_scalar_name>,
2653         avx512f_rndscale<mode><round_saeonly_name>,
2654         avx512dq_ranges<mode><mask_scalar_name><round_saeonly_scalar_name>,
2655         avx512f_vgetmant<mode><mask_scalar_name><round_saeonly_scalar_name>):
2656         Use %<iptr>2 instead of %2 for -masm=intel.
2657         (avx512f_vcvtss2usi<round_name>, avx512f_vcvtss2usiq<round_name>,
2658         avx512f_vcvttss2usi<round_saeonly_name>,
2659         avx512f_vcvttss2usiq<round_saeonly_name>): Use %k1 instead of %1 for
2660         -masm=intel.
2661         (avx512f_vcvtsd2usi<round_name>, avx512f_vcvtsd2usiq<round_name>,
2662         avx512f_vcvttsd2usi<round_saeonly_name>,
2663         avx512f_vcvttsd2usiq<round_saeonly_name>, ufloatv2siv2df2<mask_name>):
2664         Use %q1 instead of %1 for -masm=intel.
2665         (avx512f_sfixupimm<mode><sd_maskz_name><round_saeonly_name>,
2666         avx512f_sfixupimm<mode>_mask<round_saeonly_name>): Use %<iptr>3 instead
2667         of %3 for -masm=intel.
2668         (sse2_shufpd_v2df_mask): Fix a typo, change %{6%} to %{%6%} for
2669         -masm=intel.
2670         (*avx512vl_<code>v2div2qi2_store): Use %w0 instead of %0 for
2671         -masm=intel.
2672         (*avx512vl_<code><mode>v4qi2_store): Use %k0 instead of %0 for
2673         -masm=intel.
2674         (avx512vl_<code><mode>v4qi2_mask_store): Use a single pattern with
2675         %k0 and %1 for -masm=intel rather than two patterns, one with %0 and
2676         %g1.
2677         (*avx512vl_<code><mode>v8qi2_store): Use %q0 instead of %0 for
2678         -masm=intel.
2679         (avx512vl_<code><mode>v8qi2_mask_store): Use a single pattern with
2680         %q0 and %1 for -masm=intel rather than two patterns, one with %0 and
2681         %g1 and one with %0 and %1.
2682         (avx512er_vmrcp28<mode><round_saeonly_name>,
2683         avx512er_vmrsqrt28<mode><round_saeonly_name>): Use %<iptr>1 instead of
2684         %1 for -masm=intel.
2685         (avx5124fmaddps_4fmaddps_mask, avx5124fmaddps_4fmaddss_mask,
2686         avx5124fmaddps_4fnmaddps_mask, avx5124fmaddps_4fnmaddss_mask,
2687         avx5124vnniw_vp4dpwssd_mask, avx5124vnniw_vp4dpwssds_mask): Swap order
2688         of %0 and %{%4%} for -masm=intel.
2689         (avx5124fmaddps_4fmaddps_maskz, avx5124fmaddps_4fmaddss_maskz,
2690         avx5124fmaddps_4fnmaddps_maskz, avx5124fmaddps_4fnmaddss_maskz,
2691         avx5124vnniw_vp4dpwssd_maskz, avx5124vnniw_vp4dpwssds_maskz): Swap
2692         order of %0 and %{%5%}%{z%} for -masm=intel.
2693
2694 2018-04-17  Jan Hubicka  <jh@suse.cz>
2695
2696         PR lto/85405
2697         * ipa-devirt.c (odr_types_equivalent_p): Handle bit fields.
2698
2699 2018-04-17  Martin Liska  <mliska@suse.cz>
2700
2701         PR ipa/85329
2702         * multiple_target.c (create_dispatcher_calls): Set apostrophes
2703         for target_clone error message.  Make default implementation
2704         clone to be a local declaration.
2705         (separate_attrs): Add new argument and check for an empty
2706         string.
2707         (expand_target_clones): Handle it.
2708         (ipa_target_clone): Make redirection just for target_clones
2709         functions.
2710
2711 2018-04-16  Cesar Philippidis  <cesar@codesourcery.com>
2712             Tom de Vries  <tom@codesourcery.com>
2713
2714         PR middle-end/84955
2715         * omp-expand.c (expand_oacc_for): Add dummy false branch for
2716         tiled basic blocks without omp continue statements.
2717
2718 2018-04-16  Aaron Sawdey  <acsawdey@linux.ibm.com>
2719
2720         PR target/83660
2721         * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Mark
2722         vec_extract expression as having side effects to make sure it gets
2723         a cleanup point.
2724
2725 2018-04-16  H.J. Lu  <hongjiu.lu@intel.com>
2726
2727         PR target/85403
2728         * config/i386/i386.c (get_builtin_code_for_version): Check
2729         error_mark_node.
2730
2731 2018-04-16  Olga Makhotina  <olga.makhotina@intel.com>
2732
2733         PR target/84331
2734         * gcc/config.gcc: Support "skylake".
2735         * gcc/config/i386/i386-c.c (ix86_target_macros_internal): Handle
2736         PROCESSOR_SKYLAKE.
2737         * gcc/config/i386/i386.c (m_SKYLAKE): Define.
2738         (processor_target_table): Add "skylake".
2739         (ix86_option_override_internal): Add "skylake".
2740         (get_builtin_code_for_version): Handle PROCESSOR_SKYLAKE,
2741         PROCESSOR_CANNONLAKE.
2742         (get_builtin_code_for_version): Fix priority for
2743         PROCESSOR_ICELAKE_CLIENT, PROCESSOR_ICELAKE_SERVER,
2744         PROCESSOR_SKYLAKE-AVX512.
2745         * gcc/config/i386/i386.h (processor_costs): Define TARGET_SKYLAKE.
2746         (processor_type): Add PROCESSOR_SKYLAKE.
2747
2748 2018-04-16  Paolo Carlini  <paolo.carlini@oracle.com>
2749             Jason Merrill  <jason@redhat.com>
2750
2751         PR c++/85112
2752         * convert.c (convert_to_integer_1): Use direct recursion for
2753         enumeral types and types with a precision less than the number
2754         of bits in their mode.
2755
2756 2018-04-16  Julia Koval  <julia.koval@intel.com>
2757
2758         PR target/84413
2759         * config/i386/x86-tune.def (X86_TUNE_SSE_UNALIGNED_LOAD_OPTIMAL,
2760         X86_TUNE_SSE_UNALIGNED_STORE_OPTIMAL): Add m_SKYLAKE_AVX512
2761
2762 2018-04-14  Segher Boessenkool  <segher@kernel.crashing.org>
2763
2764         PR target/85293
2765         * config/rs6000/rs6000.opt (mdirect-move): Make deprecated.
2766         * doc/invoke.texi (RS/6000 and PowerPC Options): Remove -mdirect-move
2767         and -mno-direct-move.
2768
2769 2018-04-13  Paul A. Clarke  <pc@us.ibm.com>
2770
2771         PR target/83402
2772         * config/rs6000/emmintrin.h (_mm_slli_epi{16,32,64}):
2773         Ensure that vec_splat_s32 is only called with 0 <= shift < 16.
2774         Ensure negative shifts result in {0}.
2775
2776 2018-04-13  Vladimir Makarov  <vmakarov@redhat.com>
2777
2778         PR rtl-optimization/79916
2779         * config/rs6000/rs6000.c (rs6000_emit_move): Use assigned hard
2780         regs (if any) to define how to gnerate SD moves when LRA is in
2781         progress.
2782
2783 2018-04-13  Jakub Jelinek  <jakub@redhat.com>
2784
2785         PR rtl-optimization/85393
2786         * except.h (expand_dw2_landing_pad_for_region): Remove declaration.
2787         * except.c (expand_dw2_landing_pad_for_region): Make static.
2788         * bb-reorder.c (fix_up_crossing_landing_pad): In new_bb emit just
2789         a label and unconditional jump to old_bb, rather than
2790         expand_dw2_landing_pad_for_region insn(s) and jump to single_succ
2791         basic block.
2792
2793         PR rtl-optimization/85376
2794         * simplify-rtx.c (simplify_const_unary_operation): For CLZ and CTZ and
2795         zero op0, if C?Z_DEFINED_VALUE_AT_ZERO is false, return NULL_RTX
2796         instead of a specific value.
2797
2798 2018-04-13  Jan Hubicka  <hubicka@ucw.cz>
2799             Bin Cheng  <bin.cheng@arm.com>
2800
2801         PR tree-optimization/82965
2802         PR tree-optimization/83991
2803         * cfgloopanal.c (expected_loop_iterations_unbounded): Add
2804         by_profile_only parameter.
2805         * cfgloopmanip.c (scale_loop_profile): Further scale loop's profile
2806         information if the loop was predicted to iterate too many times.
2807         * cfgloop.h (expected_loop_iterations_unbounded): Update prototype
2808
2809 2018-04-13  Jan Hubicka  <hubicka@ucw.cz>
2810
2811         PR lto/71991
2812         * config/i386/i386.c (ix86_can_inline_p): Allow safe transitions for
2813         always inline.
2814
2815 2018-04-13  Martin Liska  <mliska@suse.cz>
2816             Jakub Jelinek  <jakub@redhat.com>
2817
2818         PR middle-end/81657
2819         * expr.h (enum block_op_methods): Add BLOCK_OP_NO_LIBCALL_RET.
2820         * expr.c (emit_block_move_hints): Handle BLOCK_OP_NO_LIBCALL_RET.
2821         * builtins.c (expand_builtin_memory_copy_args): Use
2822         BLOCK_OP_NO_LIBCALL_RET method for mempcpy with non-ignored target,
2823         handle dest_addr == pc_rtx.
2824
2825 2018-04-12  Segher Boessenkool  <segher@kernel.crashing.org>
2826
2827         PR target/85291
2828         * config/rs6000/rs6000.md (fix_trunc<mode>si2): Use legacy code if
2829         asked to not generate direct moves.
2830         (fix_trunc<mode>si2_stfiwx): Similar.
2831         (fix_trunc<mode>si2_internal): Similar.
2832
2833 2018-04-12  Jakub Jelinek  <jakub@redhat.com>
2834
2835         PR debug/83157
2836         * var-tracking.c (add_stores): Handle STRICT_LOW_PART SET_DEST.
2837         * cselib.c (cselib_record_sets): For STRICT_LOW_PART dest,
2838         lookup if dest in some wider mode is known to be const0_rtx and
2839         if so, record permanent equivalence for it to be ZERO_EXTEND of
2840         the narrower mode destination.
2841
2842 2018-04-12  Cesar Philippidis  <cesar@codesourcery.com>
2843
2844         * lto-streamer-out.c (output_function): Revert 259346.
2845         * omp-expand.c (expand_oacc_for): Likewise.
2846
2847 2018-04-12  Alexander Monakov  <amonakov@ispras.ru>
2848
2849         PR rtl-optimization/85354
2850         * sel-sched-ir.c (sel_init_pipelining): Move cfg_cleanup call...
2851         * sel-sched.c (sel_global_init): ... here.
2852
2853 2018-04-12  Eric Botcazou  <ebotcazou@adacore.com>
2854
2855         PR target/85238
2856         * lto-wrapper.c (debug_objcopy): Open the files in binary mode.
2857         * dwarf2out.c (dwarf2out_early_finish): Do not generate assembly in LTO
2858         mode for PE-COFF targets.
2859         * config/i386/i386-protos.h (i386_pe_asm_lto_start): Declare.
2860         (i386_pe_asm_lto_end): Likewise.
2861         * config/i386/cygming.h (TARGET_ASM_LTO_START): Define.
2862         (TARGET_ASM_LTO_END): Likewise.
2863         * config/i386/winnt.c (saved_debug_info_level): New static variable.
2864         (i386_pe_asm_lto_start): New function.
2865         (i386_pe_asm_lto_end): Likewise.
2866
2867 2018-04-12  Cesar Philippidis  <cesar@codesourcery.com>
2868             Richard Biener  <rguenther@suse.de>
2869
2870         PR middle-end/84955
2871         * lto-streamer-out.c (output_function): Fix CFG loop state before
2872         streaming out.
2873         * omp-expand.c (expand_oacc_for): Handle calls to internal
2874         functions like regular functions.
2875
2876 2018-04-12  Richard Biener  <rguenther@suse.de>
2877
2878         PR lto/85371
2879         * dwarf2out.c (init_sections_and_labels): Use debug_line_section[_label]
2880         for the early LTO debug to properly generate references to it
2881         during DIE emission.  Do not re-use that for the skeleton for
2882         split-dwarf.
2883         (dwarf2out_early_finish): Likewise.
2884
2885 2018-04-12  Jakub Jelinek  <jakub@redhat.com>
2886
2887         PR target/85328
2888         * config/i386/sse.md
2889         (<mask_codefor>avx512dq_vextract<shuffletype>64x2_1<mask_name> split,
2890         <mask_codefor>avx512f_vextract<shuffletype>32x4_1<mask_name> split,
2891         vec_extract_lo_<mode><mask_name> split, vec_extract_lo_v32hi,
2892         vec_extract_lo_v64qi): For non-AVX512VL if input is xmm16+ reg
2893         and output is a reg, avoid creating invalid lowpart subreg, but
2894         instead split into a 512-bit move.  Don't split if not AVX512VL,
2895         input is xmm16+ reg and output is a mem.
2896         (vec_extract_lo_<mode><mask_name>, vec_extract_lo_v32hi,
2897         vec_extract_lo_v64qi): Don't require split if not AVX512VL, input is
2898         xmm16+ reg and output is a mem.
2899
2900 2018-04-12  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>
2901
2902         * config/s390/s390.c (s390_output_indirect_thunk_function): Check
2903         also for flag_dwarf2_cfi_asm.
2904
2905 2018-04-12  Jakub Jelinek  <jakub@redhat.com>
2906
2907         PR rtl-optimization/85342
2908         * regcprop.c (copyprop_hardreg_forward_1): Remove replaced array, use
2909         a bool scalar var inside of the loop instead.  Don't try to update
2910         recog_data.operand after failed apply_change_group.
2911
2912 2018-04-12  Tom de Vries  <tom@codesourcery.com>
2913
2914         PR target/85296
2915         * config/nvptx/nvptx.c (flexible_array_member_type_p): New function.
2916         (nvptx_assemble_decl_begin): Add undefined param.  Declare undefined
2917         array with flexible array member as array without given dimension.
2918         (nvptx_assemble_undefined_decl): Set nvptx_assemble_decl_begin call
2919         argument for undefined param to true.
2920
2921 2018-04-11  Aaron Sawdey  <acsawdey@linux.ibm.com>
2922
2923         PR target/85321
2924         * doc/invoke.texi (RS/6000 and PowerPC Options): Document options
2925         -mcall- and -mtraceback=. Remove options -mabi=spe and -mabi=no-spe
2926         from PowerPC section.
2927         * config/rs6000/sysv4.opt (mcall-): Improve help text.
2928         * config/rs6000/rs6000.opt (mblock-compare-inline-limit=): Trim
2929         help text that is too long.
2930         * config/rs6000/rs6000.opt (mblock-compare-inline-loop-limit=): Trim
2931         help text that is too long.
2932         * config/rs6000/rs6000.opt (mstring-compare-inline-limit=): Trim
2933         help text that is too long.
2934
2935 2018-04-11  Uros Bizjak  <ubizjak@gmail.com>
2936
2937         * config/alpha/alpha.md (stack_probe_internal): Rename
2938         from "probe_stack".  Update all callers.
2939
2940 2018-04-11  Alexander Monakov  <amonakov@ispras.ru>
2941
2942         PR rtl-optimization/84566
2943         * sched-deps.c (sched_analyze_insn): Check deps->readonly when invoking
2944         sched_macro_fuse_insns.
2945
2946 2018-04-11  Alexander Monakov  <amonakov@ispras.ru>
2947
2948         PR target/84301
2949         * sched-rgn.c (add_branch_dependences): Move sel_sched_p check here...
2950         (compute_block_dependences): ... from here.
2951
2952 2018-04-11  Jakub Jelinek  <jakub@redhat.com>
2953
2954         PR tree-optimization/85331
2955         * vec-perm-indices.h (vec_perm_indices::clamp): Change input type
2956         from int to HOST_WIDE_INT.
2957
2958 2018-04-11  Martin Jambor  <mjambor@suse.cz>
2959
2960         PR ipa/84149
2961         * ipa-cp.c (propagate_vals_across_pass_through): Expand comment.
2962         (cgraph_edge_brings_value_p): New parameter dest_val, check if it is
2963         not the same as the source val.
2964         (cgraph_edge_brings_value_p): New parameter.
2965         (gather_edges_for_value): Pass destination value to
2966         cgraph_edge_brings_value_p.
2967         (perhaps_add_new_callers): Likewise.
2968         (get_info_about_necessary_edges): Likewise and exclude values brought
2969         only by self-recursive edges.
2970         (create_specialized_node): Redirect only clones of self-calling edges.
2971         (+self_recursive_pass_through_p): New function.
2972         (find_more_scalar_values_for_callers_subset): Use it.
2973         (find_aggregate_values_for_callers_subset): Likewise.
2974         (known_aggs_to_agg_replacement_list): Removed.
2975         (decide_whether_version_node): Re-calculate known constants for all
2976         remaining context clones.
2977
2978 2018-04-11  Richard Biener  <rguenther@suse.de>
2979
2980         PR lto/85339
2981         * dwarf2out.c (dwarf2out_finish): Remove DW_AT_stmt_list attribute
2982         from early DWARF output.
2983         (dwarf2out_early_finish): Output line info unconditionally into
2984         early DWARF and add reference to it.
2985
2986 2018-04-11  Jakub Jelinek  <jakub@redhat.com>
2987
2988         PR target/85281
2989         * config/i386/sse.md (iptr): Add V16SFmode and V8DFmode cases.
2990         (<avx512>_vec_dup<mode><mask_name>): Use a single pattern for modes
2991         other than V2DFmode using iptr mode attribute.
2992         (<avx512>_vec_dup<mode><mask_name>): Use iptr mode attribute.
2993
2994 2018-04-11  Alexander Monakov  <amonakov@ispras.ru>
2995
2996         PR rtl-optimization/84659
2997         * sel-sched-ir.c (sel_init_pipelining): Invoke cleanup_cfg.
2998
2999 2018-04-11  Jakub Jelinek  <jakub@redhat.com>
3000
3001         PR debug/85302
3002         * dwarf2out.c (skip_loc_list_entry): Don't call size_of_locs if
3003         SIZEP is NULL.
3004         (output_loc_list): Pass address of a dummy size variable even in the
3005         locview handling loop.
3006         (index_location_lists): Add comment on why skip_loc_list_entry can't
3007         call size_of_locs.
3008
3009 2018-04-11  Thomas Preud'homme  <thomas.preudhomme@arm.com>
3010
3011         PR target/85261
3012         * config/arm/arm-builtins.c (arm_expand_builtin): Force input operand
3013         into register.
3014
3015 2018-04-10  Aaron Sawdey  <acsawdey@linux.ibm.com>
3016
3017         PR target/85321
3018         * doc/invoke.texi (RS/6000 and PowerPC Options): Document options
3019         -mblock-compare-inline-limit, -mblock-compare-inline-loop-limit,
3020         and -mstring-compare-inline-limit.
3021
3022 2018-04-10  Segher Boessenkool  <segher@kernel.crashing.org>
3023
3024         PR target/85287
3025         * gcc/config/rs6000/rs6000.md (allocate_stack): Put the residual size
3026         for stack clash protection in a register whenever we need it to be in
3027         a register.
3028
3029 2018-04-10  Segher Boessenkool  <segher@kernel.crashing.org>
3030
3031         * common/config/rs6000/rs6000-common.c (rs6000_option_init_struct):
3032         Enable -fasynchronous-unwind-tables by default if OBJECT_FORMAT_ELF.
3033
3034 2018-04-10  Segher Boessenkool  <segher@kernel.crashing.org>
3035
3036         PR target/85321
3037         * config/rs6000/rs6000.opt (mtraceback=): Show the allowed values in
3038         the help text.
3039         (mlong-double-): Ditto.
3040         * config/rs6000/sysv4.opt (msdata=): Ditto.
3041         (mtls-size=): Ditto.
3042
3043 2018-04-10  Kelvin Nilsen  <kelvin@gcc.gnu.org>
3044
3045         * config/rs6000/rs6000-c.c (altivec_overloaded_builtins): Remove
3046         erroneous entries for
3047         "vector int vec_ldl (int, long int *)", and
3048         "vector unsigned int vec_ldl (int, unsigned long int *)".
3049         Add comments and entries for
3050         "vector bool char vec_ldl (int, bool char *)",
3051         "vector bool short vec_ldl (int, bool short *)",
3052         "vector bool int vec_ldl (int, bool int *)",
3053         "vector bool long long vec_ldl (int, bool long long *)",
3054         "vector pixel vec_ldl (int, pixel *)",
3055         "vector long long vec_ldl (int, long long *)",
3056         "vector unsigned long long vec_ldl (int, unsigned long long *)".
3057         * config/rs6000/rs6000.c (rs6000_init_builtins): Initialize new
3058         type tree bool_long_long_type_node and correct definition of
3059         bool_V2DI_type_node to make reference to this new type tree.
3060         (rs6000_mangle_type): Replace erroneous reference to
3061         bool_long_type_node with bool_long_long_type_node.
3062         * config/rs6000/rs6000.h (enum rs6000_builtin_type_index): Add
3063         comments to emphasize sign distinctions for char and int types and
3064         replace RS6000_BTI_bool_long constant with
3065         RS6000_BTI_bool_long_long constant.  Also add comment to restrict
3066         use of RS6000_BTI_pixel.
3067         (bool_long_type_node): Remove this macro definition.
3068         (bool_long_long_type_node): New macro definition
3069
3070 2018-04-10  Jakub Jelinek  <jakub@redhat.com>
3071
3072         PR rtl-optimization/85300
3073         * combine.c (subst): Handle subst of CONST_SCALAR_INT_P new_rtx also
3074         into FLOAT and UNSIGNED_FLOAT like ZERO_EXTEND, return a CLOBBER if
3075         simplify_unary_operation fails.
3076
3077 2018-04-10  Martin Liska  <mliska@suse.cz>
3078
3079         * gdbhooks.py: Add pretty-printers for varpool_node, symtab_node,
3080         cgraph_edge and ipa_ref.
3081
3082 2018-04-10  Jakub Jelinek  <jakub@redhat.com>
3083
3084         PR target/85177
3085         PR target/85255
3086         * config/i386/sse.md
3087         (<extract_type>_vinsert<shuffletype><extract_suf>_mask): Fix
3088         computation of the VEC_MERGE selector from mask.
3089         (<extract_type>_vinsert<shuffletype><extract_suf>_1<mask_name>):
3090         Fix decoding of the VEC_MERGE selector into mask.
3091
3092 2018-04-10  Richard Sandiford  <richard.sandiford@linaro.org>
3093
3094         PR tree-optimization/85286
3095         * tree-vect-data-refs.c (vect_get_smallest_scalar_type):
3096
3097 2018-04-10  Richard Sandiford  <richard.sandiford@linaro.org>
3098
3099         * final.c (final_1): Set insn_last_address as well as
3100         insn_current_address.
3101
3102 2018-04-10  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
3103
3104         PR target/85173
3105         * explow.c (emit_stack_probe): Call validize_mem on memory location
3106         before passing it to gen_probe_stack.  Create address operand and
3107         legitimize it for the probe_stack_address case.
3108
3109 2018-04-09  Jan Hubicka  <jh@suse.cz>
3110
3111         PR lto/85078
3112         * ipa-devirt.c (rebuild_type_inheritance-hash): New.
3113         * ipa-utils.h (rebuild_type_inheritance-hash): Declare.
3114         * tree.c (free_lang_data_in_type): Fix handling of binfos;
3115         walk basetypes.
3116         (free_lang_data): Rebuild type inheritance graph.
3117
3118 2018-04-09  Martin Sebor  <msebor@redhat.com>
3119
3120         * invoke.texi (-finline-small-functions): Mention other optimization
3121         options.
3122         (-findirect-inlining, -fpartial-inlining): Same.
3123         (-finline-functions-called-once): Same.
3124         (-freorder-blocks-and-partition): Same.
3125
3126 2018-04-09  Jan Hubicka  <jh@suse.cz>
3127
3128         PR rtl/84058
3129         * cfgcleanup.c (try_forward_edges): Do not give up on crossing
3130         jumps; choose last target that matches the criteria (i.e.
3131         no partition changes for non-crossing jumps).
3132         * cfgrtl.c (cfg_layout_redirect_edge_and_branch): Add basic
3133         support for redirecting crossing jumps to non-crossing.
3134
3135 2018-04-09  Alexey Brodkin  <abrodkin@synopsys.com>
3136
3137         * config/arc/arc.c (arc_expand_prologue): Set stack usage info
3138         also for naked functions.
3139
3140 2018-04-09  Claudiu Zissulescu  <claziss@synopsys.com>
3141
3142         * config/arc/arc.md (add_shift): New pattern.
3143         (add_shift2): Likewise.
3144         (sub_shift): Likewise.
3145         (sub_shift_cmp0_noout): Likewise.
3146         (compare_si_ashiftsi): Likewise.
3147         (xbfu_cmp0_noout): New combine pattern.
3148         (xbfu_cmp0"): Likewise.
3149         (movsi_set_cc_insn): Place the predicable variant first.
3150         (commutative_binary_cmp0_noout): Remove clobber.
3151         (commutative_binary_cmp0): New pattern.
3152         (noncommutative_binary_cmp0): Likewise.
3153         (noncommutative_binary_cmp0_noout): Likewise.
3154         (noncommutative_binary_comparison_result_used): Removed.
3155         (rsub_cmp0): New pattern.
3156         (rsub_cmp0_noout): Likewise.
3157         (extzvsi): Changed, keep only meaningful variants.
3158         (SQH, SEZ): New iterators.
3159         (SQH_postfix): New mode attribute.
3160         (SEZ_prefix): New code attribute.
3161         (<SEZ_prefix>xt<SQH_postfix>_cmp0_noout): New instruction pattern.
3162         (<SEZ_prefix>xt<SQH_postfix>_cmp0): Likewise.
3163         * config/arc/predicates.md (cc_set_register): Use CC_REG instead
3164         of numerical value.
3165         (noncommutative_operator): Check the availability of barrel
3166         shifter option.
3167
3168 2018-04-09  Richard Biener  <rguenther@suse.de>
3169
3170         PR tree-optimization/85284
3171         * tree-ssa-loop-niter.c (number_of_iterations_exit_assumptions):
3172         Only use the niter constraining form of simple_iv when the exit
3173         is always executed.
3174
3175 2018-04-09  Tom de Vries  <tom@codesourcery.com>
3176
3177         PR target/84041
3178         * config/nvptx/nvptx.md (define_c_enum "unspecv"): Add UNSPECV_MEMBAR.
3179         (define_expand "*memory_barrier"): New define_expand.
3180         (define_insn "memory_barrier"): New insn.
3181
3182 2018-04-09  Andrey Belevantsev  <abel@ispras.ru>
3183
3184         PR rtl-optimization/80463
3185         PR rtl-optimization/83972
3186         PR rtl-optimization/83480
3187
3188         * sel-sched-ir.c (has_dependence_note_mem_dep): Take into account the
3189         correct producer for the insn.
3190         (tidy_control_flow): Fixup seqnos in case of debug insns.
3191
3192 2018-04-09  Andrey Belevantsev  <abel@ispras.ru>
3193
3194         PR rtl-optimization/83913
3195
3196         * sel-sched-ir.c (merge_expr_data): Choose the middle between two
3197         different sched-times when merging exprs.
3198
3199 2018-04-09  Andrey Belevantsev  <abel@ispras.ru>
3200
3201         PR rtl-optimization/83962
3202
3203         * sel-sched-ir.c (tidy_control_flow): Correct the order in which we call
3204         tidy_fallthru_edge and tidy_control_flow.
3205
3206 2018-04-09  Andrey Belevantsev  <abel@ispras.ru>
3207
3208         PR rtl-optimization/83530
3209
3210         * sel-sched.c (force_next_insn): New global variable.
3211         (remove_insn_for_debug): When force_next_insn is true, also leave only
3212         next insn in the ready list.
3213         (sel_sched_region): When the region wasn't scheduled, make another pass
3214         over it with force_next_insn set to 1.
3215
3216 2018-04-08  Monk Chiang  <sh.chiang04@gmail.com>
3217
3218         * config.gcc (nds32le-*-*, nds32be-*-*): Add nds32/nds32_intrinsic.h
3219         into tm_file.
3220         * config/nds32/constants.md (unspec_volatile_element): Add enum values
3221         for interrupt control.
3222         * config/nds32/nds32-intrinsic.c: Implementation of intrinsic
3223         functions for interrupt control.
3224         * config/nds32/nds32-intrinsic.md: Likewise.
3225         * config/nds32/nds32_intrinsic.h: Likewise.
3226         * config/nds32/nds32.h (nds32_builtins): Likewise.
3227
3228 2018-04-08  Chung-Ju Wu  <jasonwucj@gmail.com>
3229
3230         * config/nds32/nds32.c (nds32_init_machine_status,
3231         nds32_legitimate_index_p, nds32_legitimate_address_p): Consider
3232         strict_aligned_p field.
3233         (nds32_expand_to_rtl_hook): New function.
3234         (TARGET_EXPAND_TO_RTL_HOOK): Define.
3235         * config/nds32/nds32.h (machine_function): Add strict_aligned_p field.
3236
3237 2018-04-08  Kito Cheng  <kito.cheng@gmail.com>
3238             Chung-Ju Wu  <jasonwucj@gmail.com>
3239
3240         * config.gcc (nds32*-*-*): Check that n7 is valid to --with-cpu.
3241         * config/nds32/nds32-n7.md: New file.
3242         * config/nds32/nds32-opts.h (nds32_cpu_type): Add CPU_N7.
3243         * config/nds32/nds32-pipelines-auxiliary.c: Implementation for n7
3244         pipeline.
3245         * config/nds32/nds32-protos.h: More declarations for n7 pipeline.
3246         * config/nds32/nds32.md (pipeline_model): Add n7.
3247         * config/nds32/nds32.opt (mcpu): Support n7 pipeline cpus.
3248         * config/nds32/pipelines.md: Include n7 settings.
3249
3250 2018-04-08  Kito Cheng  <kito.cheng@gmail.com>
3251             Chung-Ju Wu  <jasonwucj@gmail.com>
3252
3253         * config.gcc (nds32*-*-*): Check that e8 is valid to --with-cpu.
3254         * config/nds32/nds32-e8.md: New file.
3255         * config/nds32/nds32-opts.h (nds32-cpu_type): Add CPU_E8.
3256         * config/nds32/nds32-pipelines-auxiliary.c: Implementation for e8
3257         pipeline.
3258         * config/nds32/nds32-protos.h: More declarations for e8 pipeline.
3259         * config/nds32/nds32.md (pipeline_model): Add e8.
3260         * config/nds32/nds32.opt (mcpu): Support e8 pipeline cpus.
3261         * config/nds32/pipelines.md: Include e8 settings.
3262
3263 2018-04-08  Kito Cheng  <kito.cheng@gmail.com>
3264             Chung-Ju Wu  <jasonwucj@gmail.com>
3265
3266         * config.gcc (nds32*-*-*): Check that n6/n8/s8 are valid to --with-cpu.
3267         * config/nds32/nds32-n8.md: New file.
3268         * config/nds32/nds32-opts.h (nds32_cpu_type): Add CPU_N6 and CPU_N8.
3269         * config/nds32/nds32-pipelines-auxiliary.c: Implementation for n8
3270         pipeline.
3271         * config/nds32/nds32-protos.h: More declarations for n8 pipeline.
3272         * config/nds32/nds32-utils.c: More implementations for n8 pipeline.
3273         * config/nds32/nds32.md (pipeline_model): Add n8.
3274         * config/nds32/nds32.opt (mcpu): Support n8 pipeline cpus.
3275         * config/nds32/pipelines.md: Include n8 settings.
3276
3277 2018-04-08  Kito Cheng  <kito.cheng@gmail.com>
3278             Chung-Ju Wu  <jasonwucj@gmail.com>
3279
3280         * config.gcc (nds32*): Add nds32-utils.o into extra_objs.
3281         * config/nds32/nds32-n9-2r1w.md: New file.
3282         * config/nds32/nds32-n9-3r2w.md: New file.
3283         * config/nds32/nds32-opts.h (nds32_cpu_type, nds32_mul_type,
3284         nds32_register_ports): New or modify for cpu n9.
3285         * config/nds32/nds32-pipelines-auxiliary.c: Implementation for n9
3286         pipeline.
3287         * config/nds32/nds32-protos.h: More declarations for n9 pipeline.
3288         * config/nds32/nds32-utils.c: New file.
3289         * config/nds32/nds32.h (TARGET_PIPELINE_N9, TARGET_PIPELINE_SIMPLE,
3290         TARGET_MUL_SLOW): Define.
3291         * config/nds32/nds32.md (pipeline_model): New attribute.
3292         * config/nds32/nds32.opt (mcpu, mconfig-mul, mconfig-register-ports):
3293         New options that support cpu n9.
3294         * config/nds32/pipelines.md: Include n9 settings.
3295         * config/nds32/t-nds32 (nds32-utils.o): Add dependency.
3296
3297 2018-04-08  Chung-Ju Wu  <jasonwucj@gmail.com>
3298
3299         * config/nds32/nds32-md-auxiliary.c (output_cond_branch): Output align
3300         information if necessary.
3301         (output_cond_branch_compare_zero): Likewise.
3302         * config/nds32/nds32.c (nds32_adjust_insn_length): Consider align case.
3303         (nds32_target_alignment): Refine for alignment.
3304         * config/nds32/nds32.h (NDS32_ALIGN_P): Define.
3305         (FUNCTION_BOUNDARY): Modify.
3306         * config/nds32/nds32.md (call_internal, call_value_internal): Consider
3307         align case.
3308         * config/nds32/nds32.opt (malways-align, malign-functions): New.
3309
3310 2018-04-08  Monk Chiang  <sh.chiang04@gmail.com>
3311
3312         * config/nds32/constants.md (unspec_volatile_element): Add values for
3313         TLB operation and data prefetch.
3314         * config/nds32/nds32-intrinsic.c: Implementation of intrinsic
3315         functions for TLB operation and data prefetch.
3316         * config/nds32/nds32-intrinsic.md: Likewise.
3317         * config/nds32/nds32_intrinsic.h: Likewise.
3318         * config/nds32/nds32.c (nds32_dpref_names): Likewise.
3319         (nds32_print_operand): Likewise.
3320         * config/nds32/nds32.h (nds32_builtins): Likewise.
3321
3322 2018-04-07  Thomas Koenig  <tkoenig@gcc.gnu.org>
3323         Andrew Pinski <pinsika@gcc.gnu.org>
3324
3325         PR middle-end/82976
3326         * match.pd: Use constant_boolean_node of correct type instead of
3327         boolean_true_node or boolean_false_node for simplifying
3328         pointer comparisons to zero.
3329
3330 2018-04-07  Jakub Jelinek  <jakub@redhat.com>
3331
3332         PR tree-optimization/80021
3333         * tree.c (verify_type_variant): Make error call in verify_variant_match
3334         translatable and remove final full stop.
3335
3336 2018-04-07  Chung-Ju Wu  <jasonwucj@gmail.com>
3337
3338         * config/nds32/constants.md (unspec_volatile_element): Add
3339         UNSPEC_VOLATILE_EH_RETURN.
3340         * config/nds32/nds32-md-auxiliary.c (nds32_output_stack_push,
3341         nds32_output_stack_pop): Support dwarf exception handling process.
3342         * config/nds32/nds32-protos.h (nds32_dynamic_chain_address): Declare.
3343         * config/nds32/nds32.c (nds32_init_machine_status): Support dwarf
3344         exception handling process.
3345         (nds32_compute_stack_frame): Likewise.
3346         (nds32_return_addr_rtx): Likewise.
3347         (nds32_initial_elimination_offset): Likewise.
3348         (nds32_expand_prologue): Likewise.
3349         (nds32_expand_epilogue): Likewise.
3350         (nds32_dynamic_chain_address): New function.
3351         * config/nds32/nds32.h (machine_function): Add fields for dwarf
3352         exception handling.
3353         (DYNAMIC_CHAIN_ADDRESS): Define.
3354         (EH_RETURN_DATA_REGNO): Define.
3355         (EH_RETURN_STACKADJ_RTX): Define.
3356         * config/nds32/nds32.md (eh_return, nds32_eh_return): Implement
3357         patterns for dwarf exception handling.
3358
3359 2018-04-07  Chung-Ju Wu  <jasonwucj@gmail.com>
3360
3361         * config/nds32/nds32.h: Clean up obsolete macros.
3362
3363 2018-04-07  Monk Chiang  <sh.chiang04@gmail.com>
3364
3365         * config/nds32/constants.md (unspec_element, unspec_volatile_element):
3366         Add enum values for particular instructions.
3367         * config/nds32/nds32-intrinsic.c: Implementation of expanding
3368         particular intrinsic functions.
3369         * config/nds32/nds32-intrinsic.md: Likewise.
3370         * config/nds32/nds32_intrinsic.h: Likewise.
3371         * config/nds32/nds32.h (nds32_builtins): Likewise.
3372         * config/nds32/nds32.md (type): Add pbsad and pbsada.
3373         (btst, ave): New patterns for particular instructions.
3374
3375 2018-04-07  Monk Chiang  <sh.chiang04@gmail.com>
3376
3377         * config/nds32/constants.md (unspec_element, unspec_volatile_element):
3378         Add enum values for atomic load/store and memory sync.
3379         * config/nds32/nds32-intrinsic.c: Implementation for atomic load/store
3380         and memory sync.
3381         * config/nds32/nds32-intrinsic.md: Likewise.
3382         * config/nds32/nds32_intrinsic.h: Likewise.
3383         * config/nds32/nds32.h (nds32_builtins): Likewise.
3384
3385 2018-04-07  Jakub Jelinek  <jakub@redhat.com>
3386
3387         PR tree-optimization/85257
3388         * fold-const.c (native_encode_vector): If not all elts could fit
3389         and off is -1, return 0 rather than offset.
3390         * tree-ssa-sccvn.c (vn_reference_lookup_3): Pass
3391         (offseti - offset2) / BITS_PER_UNIT as 4th argument to
3392         native_encode_expr.  Verify len * BITS_PER_UNIT >= maxsizei.  Don't
3393         adjust buffer in native_interpret_expr call.
3394
3395 2018-04-07  Monk Chiang  <sh.chiang04@gmail.com>
3396
3397         * config/nds32/constants.md (unspec_volatile_element): Add cache
3398         control enum values.
3399         * config/nds32/nds32-intrinsic.c: Add cache control expand functions.
3400         * config/nds32/nds32-intrinsic.md: Add cache control patterns.
3401         * config/nds32/nds32.c (nds32_cctl_names): New.
3402         (nds32_print_operand): Handle cache control register names.
3403         * config/nds32/nds32.h (nds32_builtins): New enum values.
3404         * config/nds32/nds32_intrinsic.h: Add cache control enum types and
3405         macros.
3406         * config/nds32/nds32.md (type): Add mmu.
3407         * config/nds32/pipelines.md (simple_insn): Add mmu.
3408
3409 2018-04-07  Chung-Ju Wu  <jasonwucj@gmail.com>
3410
3411         * config/nds32/nds32.md (type): Remove call.
3412         * config/nds32/pipelines.md (simple_insn): Likewise.
3413
3414 2018-04-07  Monk Chiang  <sh.chiang04@gmail.com>
3415
3416         * config/nds32/constants.md (unspec_volatile_element): Add
3417         UNSPEC_VOLATILE_FMFCSR, UNSPEC_VOLATILE_FMTCSR and
3418         UNSPEC_VOLATILE_FMFCFG.
3419         * config/nds32/nds32-intrinsic.c (bdesc_noarg): New builtin
3420         description for fmfcfg and fmfcsr.
3421         (bdesc_1arg): Add fmtcsr.
3422         (bdesc_2arg): Add fcpynss, fcpyss, fcpynsd and fcpysd.
3423         (nds32_expand_builtin_impl): Deal with FPU intrinsic functions.
3424         * config/nds32/nds32-intrinsic.md (unspec_fcpynsd, unspec_fcpysd,
3425         unspec_fcpynss, unspec_fcpysd, unspec_fcpyss, unspec_fmfcsr,
3426         unspec_fmfcfg): New patterns.
3427         * config/nds32/nds32.h (nds32_builtins): Add NDS32_BUILTIN_FMFCFG,
3428         NDS32_BUILTIN_FMFCSR, NDS32_BUILTIN_FMTCSR, NDS32_BUILTIN_FCPYNSS,
3429         NDS32_BUILTIN_FCPYSS,NDS32_BUILTIN_FCPYNSD and NDS32_BUILTIN_FCPYSD.
3430         * config/nds32/nds32_intrinsic.h (__nds32__fcpynsd, __nds32__fcpynss,
3431         __nds32__fcpysd, __nds32__fcpyss, __nds32__fmfcsr, __nds32__fmtcsr,
3432         __nds32__fmfcfg): Define.
3433
3434 2018-04-07  Monk Chiang  <sh.chiang04@gmail.com>
3435
3436         * config/nds32/nds32.c (nds32_intrinsic_register_names): Add more
3437         intrinsic register names.
3438         * config/nds32/nds32_intrinsic.h (nds32_intrinsic_registers): Add more
3439         intrinsic register enum values and macros.
3440
3441 2018-04-07  Chung-Ju Wu  <jasonwucj@gmail.com>
3442
3443         * config/nds32/nds32.c (nds32_legitimate_index_p): Modify condition
3444         for load/store addressing form.
3445         (nds32_print_operand_address): Likewise.
3446
3447 2018-04-06  Eric Botcazou  <ebotcazou@adacore.com>
3448
3449         PR target/85196
3450         * config/sparc/sparc.c (sparc_expand_move): Deal with symbolic operands
3451         based on LABEL_REF.  Remove useless assertion.
3452         (pic_address_needs_scratch): Fix formatting.
3453         (sparc_legitimize_pic_address): Minor tweaks.
3454         (sparc_delegitimize_address): Adjust assertion accordingly.
3455         * config/sparc/sparc.md (movsi_pic_label_ref): Change label_ref_operand
3456         into symbolic_operand.
3457         (movsi_high_pic_label_ref): Likewise.
3458         (movsi_lo_sum_pic_label_ref): Likewise.
3459         (movdi_pic_label_ref): Likewise.
3460         (movdi_high_pic_label_ref): Likewise.
3461         (movdi_lo_sum_pic_label_ref): Likewise.
3462
3463 2018-04-06  Amaan Cheval  <amaan.cheval@gmail.com>
3464
3465         * config.gcc (x86_64-*-rtems*): Add rtems.h to tm_file for
3466         custom LIB_SPEC setup.
3467
3468 2018-04-06  Ruslan Bukin  <br@bsdpad.com>
3469             Kito Cheng  <kito.cheng@gmail.com>
3470
3471         * config.gcc (riscv*-*-freebsd*): Add RISC-V FreeBSD support.
3472         * config/riscv/freebsd.h: New.
3473
3474 2018-04-06  Chung-Ju Wu  <jasonwucj@gmail.com>
3475
3476         * config/nds32/nds32.c (nds32_adjust_insn_length): Refine.
3477         * config/nds32/nds32.h (ADJUST_INSN_LENGTH): Change the location in
3478         file.
3479
3480 2018-04-06  Chung-Ju Wu  <jasonwucj@gmail.com>
3481             Kito Cheng  <kito.cheng@gmail.com>
3482
3483         * config/nds32/nds32-md-auxiliary.c (nds32_output_return,
3484         nds32_output_call, nds32_symbol_binds_local_p): New functions.
3485         * config/nds32/nds32-protos.h (nds32_output_call,
3486         nds32_output_return): Declare.
3487         * config/nds32/nds32.md: Refine all the call and return patterns.
3488
3489 2018-04-06  Jakub Jelinek  <jakub@redhat.com>
3490
3491         PR debug/85252
3492         * dwarf2out.c (rtl_for_decl_init): For STRING_CST initializer only
3493         build CONST_STRING if TYPE_MAX_VALUE is non-NULL and is INTEGER_CST.
3494
3495         PR rtl-optimization/84872
3496         * cfgloopmanip.c (create_preheader): Use make_forwarder_block even if
3497         nentry == 1 when CP_FALLTHRU_PREHEADERS and single_entry is
3498         EDGE_CROSSING edge.
3499
3500 2018-04-06  Tamar Christina  <tamar.christina@arm.com>
3501
3502         * expr.c (copy_blkmode_to_reg): Revert 254862.
3503         * doc/sourcebuild.texi (word_mode_no_slow_unalign): Likewise.
3504
3505 2018-04-06  Richard Biener  <rguenther@suse.de>
3506
3507         PR middle-end/85244
3508         * tree-dfa.c (get_ref_base_and_extent): Reset seen_variable_array_ref
3509         after seeing a component reference with an adjacent field.  Treat
3510         refs to arrays at struct end of external decls similar to
3511         refs to unconstrained commons.
3512
3513 2018-04-06  Jakub Jelinek  <jakub@redhat.com>
3514
3515         PR sanitizer/85213
3516         * fold-const.c (twoval_comparison_p): Remove SAVE_P argument and don't
3517         look through SAVE_EXPRs with non-side-effects argument.  Adjust
3518         recursive calls.
3519         (fold_comparison): Adjust twoval_comparison_p caller, don't handle
3520         save_p here.
3521
3522 2018-04-06  Richard Biener  <rguenther@suse.de>
3523
3524         PR middle-end/85180
3525         * alias.c (find_base_term): New wrapper around find_base_term
3526         unwinding CSELIB_VAL_PTR changes.
3527         (find_base_term): Do not restore CSELIB_VAL_PTR during the
3528         recursion.
3529
3530 2018-04-06  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>
3531
3532         * config/s390/s390.c (s390_z10_optimize_cmp): Expand dedicated NOP
3533         instructions.
3534         * config/s390/s390.md (UNSPECV_NOP_LR_0, UNSPECV_NOP_LR_1): New
3535         constant definitions.
3536         ("nop"): lr 0,0 -> nopr r0
3537         ("nop_lr0", "nop_lr1"): New insn definitions.
3538
3539 2018-04-06  Chung-Ju Wu  <jasonwucj@gmail.com>
3540
3541         * config/nds32/nds32.md (*stack_push, *stack_pop): Use
3542         NDS32_V3PUSH_AVAILABLE_P macro.
3543
3544 2018-04-06  Monk Chiang  <sh.chiang04@gmail.com>
3545             Chung-Ju Wu  <jasonwucj@gmail.com>
3546
3547         * config.gcc (nds32*-*-*): Add v2j v3f v3s checking.
3548         (nds32*-*-*): Add float and fpu_config into supported_defaults.
3549         * common/config/nds32/nds32-common.c (TARGET_DEFAULT_TARGET_FLAGS):
3550         Include TARGET_DEFAULT_FPU_ISA and TARGET_DEFAULT_FPU_FMA.
3551         * config/nds32/constants.md (unspec_element): Add UNSPEC_COPYSIGN,
3552         UNSPEC_FCPYNSD, UNSPEC_FCPYNSS, UNSPEC_FCPYSD and UNSPEC_FCPYSS.
3553         * config/nds32/constraints.md: New constraints and checking for hard
3554         float configuration.
3555         * config/nds32/iterators.md: New mode iterator and attribute for hard
3556         float configuration.
3557         * config/nds32/nds32-doubleword.md: Use hard float alternatives and
3558         patterns.
3559         * config/nds32/nds32-fpu.md: New file.
3560         * config/nds32/nds32-md-auxiliary.c: New functions and checkings to
3561         deal with hard float code generation.
3562         * config/nds32/nds32-opts.h (nds32_arch_type): Add ARCH_V3F and
3563         ARCH_V3S.
3564         (abi_type, float_reg_number): New enum type.
3565         * config/nds32/nds32-predicates.c: New predicates for hard float.
3566         * config/nds32/nds32-protos.h: Declare functions for hard float.
3567         * config/nds32/nds32.c: Implementation for hard float configuration.
3568         * config/nds32/nds32.h: Definitions for hard float configuration.
3569         * config/nds32/nds32.md: Include hard float machine description and
3570         modify patterns for hard float configuration.
3571         * config/nds32/nds32.opt: New options for hard float configuration.
3572         * config/nds32/predicates.md: New predicates for hard float
3573         configuration.
3574
3575 2018-04-06  Kuan-Lin Chen  <kuanlinchentw@gmail.com>
3576
3577         * common/config/nds32/nds32-common.c
3578         (nds32_option_optimization_table): Enable -mreleax-hint by default.
3579
3580 2018-04-05  Jakub Jelinek  <jakub@redhat.com>
3581
3582         PR middle-end/85195
3583         * match.pd (BIT_FIELD_REF CONSTRUCTOR@0 @1 @2): Use view_convert around
3584         CONSTRUCTOR_ELT (ctor, ...)->value.
3585
3586 2018-04-05  Uros Bizjak  <ubizjak@gmail.com>
3587
3588         PR target/85193
3589         * config/i386/i386.md (define_attr "memory"): Handle rotate1 type.
3590
3591 2018-04-05  Tom de Vries  <tom@codesourcery.com>
3592
3593         PR target/85204
3594         * config/nvptx/nvptx.c (nvptx_single): Fix neutering of bb with only
3595         cond jump.
3596
3597 2018-04-05  Shiva Chen  <shiva0217@gmail.com>
3598             Kito Cheng  <kito.cheng@gmail.com>
3599
3600         * config/nds32/constraints.md (U33): Fine-tune checking condition.
3601         * config/nds32/nds32-md-auxiliary.c (nds32_mem_format): Ditto.
3602         * config/nds32/nds32.h (nds32_16bit_address_type): Add
3603         ADDRESS_POST_MODIFY_LO_REG_IMM3U.
3604
3605 2018-04-05  Shiva Chen  <shiva0217@gmail.com>
3606             Kito Cheng  <kito.cheng@gmail.com>
3607
3608         * config/nds32/constraints.md (Ufe): New memory constraint.
3609         * config/nds32/nds32-md-auxiliary.c (nds32_mem_format,
3610         nds32_output_16bit_load): Consider r8 register for lwi45.fe format.
3611         * config/nds32/nds32.c (nds32_print_operand): Output lwi45.fe
3612         operands.
3613         * config/nds32/nds32.h (nds32_16bit_address_type): Add ADDRESS_R8_IMM7U.
3614         * config/nds32/nds32.md (*mov<mode>): Adjust pattern.
3615
3616 2018-04-05  Chung-Ju Wu  <jasonwucj@gmail.com>
3617
3618         * config/nds32/nds32.md: Use optimize_size in the condition for
3619         alu-shift instructions.
3620
3621 2018-04-05  Chung-Ju Wu  <jasonwucj@gmail.com>
3622
3623         * config/nds32/nds32.md (divsi4, udivsi4): New patterns.
3624
3625 2018-04-05  Chung-Ju Wu  <jasonwucj@gmail.com>
3626
3627         * config/nds32/nds32.md (negsi2): Refine pattern.
3628
3629 2018-04-05  Kito Cheng  <kito.cheng@gmail.com>
3630             Chung-Ju Wu  <jasonwucj@gmail.com>
3631
3632         * config/nds32/iterators.md (shift_rotate): New code iterator.
3633         (shift): New code attribute.
3634         * config/nds32/nds32-md-auxiliary.c (nds32_expand_constant): New.
3635         * config/nds32/nds32-protos.h (nds32_expand_constant): Declare.
3636         * config/nds32/nds32.c (nds32_print_operand): Deal with more cases.
3637         * config/nds32/nds32.md (addsi3, *add_srli): Refine implementation for
3638         bit-wise operations.
3639         (andsi3, *andsi3): Ditto.
3640         (iorsi3, *iorsi3, *or_slli, *or_srli): Ditto.
3641         (xorsi3, *xorsi3, *xor_slli, *xor_srli): Ditto.
3642         (<shift>si3, *ashlsi3, *ashrsi3, *lshrsi3, *rotrsi3): Ditto.
3643         * config/nds32/predicates.md (nds32_rimm5u_operand, nds32_and_operand,
3644         nds32_ior_operand, nds32_xor_operand): New predicates.
3645
3646 2018-04-05  Chung-Ju Wu  <jasonwucj@gmail.com>
3647
3648         * config/nds32/nds32.md (add<mode>3, sub<mode>3): Rename to ...
3649         (addsi3, subsi3): ... this.
3650
3651 2018-04-05  Chung-Ju Wu  <jasonwucj@gmail.com>
3652
3653         * config/nds32/nds32.md (*sub_srli, *and_slli): Fine-tune predicator.
3654
3655 2018-04-05  Chung-Ju Wu  <jasonwucj@gmail.com>
3656
3657         * config/nds32/nds32.md: Adjust indention.
3658
3659 2018-04-05  Kito Cheng  <kito.cheng@gmail.com>
3660
3661         * config/nds32/nds32.md (feature): New attribute.
3662
3663 2018-04-05  Chung-Ju Wu  <jasonwucj@gmail.com>
3664
3665         * config/nds32/nds32.md (subtype): New attribute.
3666
3667 2018-04-04  Thomas Preud'homme  <thomas.preudhomme@arm.com>
3668
3669         PR target/85203
3670         * config/arm/arm-builtins.c (arm_expand_builtin): Change
3671         expansion to perform a bitwise AND of the argument followed by a
3672         boolean negation of the result.
3673
3674 2018-04-04  Peter Bergner  <bergner@vnet.ibm.com>
3675
3676         PR rtl-optimization/84878
3677         * ddg.c (add_cross_iteration_register_deps): Use DF_REF_BB to determine
3678         the basic block.  Assert the use reference is not artificial and that
3679         it has an associated insn.
3680
3681 2018-04-04  Michael Matz  <matz@suse.de>
3682
3683         * builtins.c (compute_objsize): Pass correct operand
3684         to array_at_struct_end_p.
3685
3686 2018-04-04  Richard Biener  <rguenther@suse.de>
3687
3688         PR lto/85176
3689         * dwarf2out.c (dwarf2out_register_external_die): Peel namespaces
3690         from contexts for DINFO_LEVEL_TERSE and below.
3691
3692 2018-04-04  Kito Cheng  <kito.cheng@gmail.com>
3693
3694         * config/nds32/nds32-doubleword.md (move_<mode>): Require
3695         resiter_operand condition.
3696         * config/nds32/nds32.md (*move<mode>): Ditto.
3697
3698 2018-04-04  Kito Cheng  <kito.cheng@gmail.com>
3699             Monk Chiang  <sh.chiang04@gmail.com>
3700
3701         * config/nds32/nds32.md (movmisalign<mode>): New pattern.
3702
3703 2018-04-04  Chung-Ju Wu  <jasonwucj@gmail.com>
3704
3705         * config/nds32/nds32.md (movqi, movhi): Merge into mov<mode>.
3706
3707 2018-04-04  Chung-Ju Wu  <jasonwucj@gmail.com>
3708             Kito Cheng  <kito.cheng@gmail.com>
3709
3710         * config/nds32/nds32-md-auxiliary.c (nds32_inverse_cond_code,
3711         nds32_cond_code_str, output_cond_branch,
3712         output_cond_branch_compare_zero, nds32_expand_cbranch,
3713         nds32_expand_cstore, nds32_expand_movcc,
3714         nds32_output_cbranchsi4_equality_zero,
3715         nds32_output_cbranchsi4_equality_reg,
3716         nds32_output_cbranchsi4_equality_reg_or_const_int,
3717         nds32_output_cbranchsi4_greater_less_zero: New functions.
3718         * config/nds32/nds32-protos.h (nds32_expand_cbranch,
3719         nds32_expand_cstore, nds32_expand_movcc,
3720         nds32_output_cbranchsi4_equality_zero,
3721         nds32_output_cbranchsi4_equality_reg,
3722         nds32_output_cbranchsi4_equality_reg_or_const_int,
3723         nds32_output_cbranchsi4_greater_less_zero): Declare.
3724         * config/nds32/predicates.md (nds32_movecc_comparison_operator,
3725         nds32_rimm11s_operand): New predicates.
3726         * config/nds32/nds32.h (nds32_expand_result_type): New enum type.
3727         * config/nds32/nds32.md: Rewrite all the branch and conditional move
3728         patterns.
3729
3730 2018-04-04  Kito Cheng  <kito.cheng@gmail.com>
3731
3732         * config/nds32/nds32-doubleword.md: Refine all the instruction type.
3733         * config/nds32/nds32.md: Ditto.
3734         * config/nds32/pipelines.md: Ditto.
3735
3736 2018-04-04  Richard Biener  <rguenther@suse.de>
3737
3738         PR tree-optimization/85168
3739         * tree-ssa-sccvn.c (vn_reference_maybe_forwprop_address): Avoid
3740         propagating abnormals.
3741
3742 2018-04-04  Chung-Ju Wu  <jasonwucj@gmail.com>
3743
3744         * config/nds32/nds32.md (enabled): Use yes/no for this attribute.
3745
3746 2018-04-04  Chung-Ju Wu  <jasonwucj@gmail.com>
3747             Kito Cheng  <kito.cheng@gmail.com>
3748
3749         * config/nds32/nds32-md-auxiliary.c (nds32_long_call_p): New function.
3750         * config/nds32/nds32-protos.h (nds32_long_call_p): Declare.
3751         * config/nds32/nds32.c (nds32_function_ok_for_sibcall): New function.
3752         (TARGET_FUNCTION_OK_FOR_SIBCALL): Define.
3753         * config/nds32/nds32.md (sibcall_internal): New.
3754         (sibcall_register): Remove.
3755         (sibcall_immediate): Remove.
3756         (sibcall_value_internal): New.
3757         (sibcall_value_register): Remove.
3758         (sibcall_value_immediate): Remove.
3759         * config/nds32/predicates.md (nds32_general_register_operand): New.
3760         (nds32_call_address_operand): New.
3761
3762 2018-04-03  Jakub Jelinek  <jakub@redhat.com>
3763
3764         PR rtl-optimization/85167
3765         * shrink-wrap.c (move_insn_for_shrink_wrap): Don't set bb_uses and
3766         bb_defs if *split_p, instead preinitialize it to NULL.
3767
3768         PR tree-optimization/85156
3769         * builtins.c (fold_builtin_expect): Use save_expr on arg1 to avoid
3770         evaluating the argument multiple times.
3771
3772 2018-04-03  Bill Schmidt  <wschmidt@linux.ibm.com>
3773
3774         * config/rs6000/emmintrin.h (_mm_cvtpd_epi32): Use __vector rather
3775         than vector.
3776         (_mm_cvtpd_ps): Likewise.
3777         (_mm_cvttpd_epi32): Likewise.
3778         * config/rs6000/mmintrin.h (_mm_unpacklo_pi8): Likewise.
3779         * config/rs6000/xmmintrin.h: For strict-ANSI C++ or C11, undefine
3780         vector, pixel, and bool following altivec.h include.
3781
3782 2018-04-03  Martin Sebor  <msebor@redhat.com>
3783
3784         * doc/extend.texi (Common Function Attributes): Clarify.
3785         (const attribute): Likewise.
3786         (pure attribute): Likewise.
3787
3788 2018-04-03  Jakub Jelinek  <jakub@redhat.com>
3789
3790         PR target/85169
3791         * config/i386/i386.c (ix86_expand_vector_set): Use
3792         HOST_WIDE_INT_1U << elt instead of 1 << elt.  Formatting fix.
3793
3794 2018-04-03  Uros Bizjak  <ubizjak@gmail.com>
3795
3796         * config/i386/i386.c (emit_i387_cw_initialization): Always use logic
3797         instructions when changing rounding bits to preserve precision bits
3798         in the x87 control word.
3799
3800 2018-04-03  Martin Liska  <mliska@suse.cz>
3801
3802         PR tree-optimization/82491
3803         * rtl.h (strip_offset_and_add): Replace += suboffset with
3804         poly_uint64 () + suboffset.
3805
3806 2018-03-29  Martin Liska  <mliska@suse.cz>
3807             Martin Jambor  <mjambor@suse.cz>
3808
3809         PR ipa/84947
3810         * ipa-cp.c (propagate_bits_across_jump_function): Bail out if
3811         param_type is not an integral or pointer type.
3812
3813 2018-04-03  Richard Biener  <rguenther@suse.de>
3814
3815         * sese.h (recompute_all_dominators): Remove.
3816
3817 2018-04-02  Martin Sebor  <msebor@redhat.com>
3818
3819         * doc/invoke.texi (-Wrestrict): Fix typos.
3820
3821 2018-04-02  Jim Wilson  <jimw@sifive.com>
3822
3823         * config/riscv/riscv.h (SHIFT_COUNT_TRUNCATED): Set to zero.
3824         * config/riscv/riscv.md (<optab>si3): Use QImode shift count.
3825         (<optab>di3, <optab>si3_extend): Likewise.
3826         (<optab>si3_mask, <optab>si3_mask_1): New.
3827         (<optab>di3_mask, <optab>di3_mask_1): New.
3828         (<optab>si3_extend_mask, <optab>si3_extend_mask_1): New.
3829         (lshrsi3_zero_extend_1): Use VOIDmode shift count.
3830         * config/riscv/sync.md (atomic_test_and_set): Emit QImode shift count.
3831
3832 2018-04-02  Gerald Pfeifer  <gerald@pfeifer.com>
3833
3834         * doc/cpp.texi (Variadic Macros): Fix line continuation in an
3835         example.
3836
3837 2018-04-02  Chung-Ju Wu  <jasonwucj@gmail.com>
3838
3839         * config/nds32/nds32.c (TARGET_CANONICALIZE_COMPARISON): Define.
3840         (nds32_canonicalize_comparison): New function.
3841
3842 2018-04-01  Chung-Ju Wu  <jasonwucj@gmail.com>
3843             Kito Cheng  <kito.cheng@gmail.com>
3844             Kuan-Lin Chen  <kuanlinchentw@gmail.com>
3845
3846         * config.gcc (nds32): Add nds32-relax-opt.o into extra_objs.
3847         * config/nds32/constants.md (unspec_volatile_element): Add
3848         UNSPEC_VOLATILE_RELAX_GROUP.
3849         * config/nds32/nds32-relax-opt.c: New file.
3850         * config/nds32/nds32-predicates.c
3851         (nds32_symbol_load_store_p): New function.
3852         * config/nds32/nds32-protos.h
3853         (nds32_symbol_load_store_p): Declare function.
3854         (make_pass_nds32_relax_opt): Declare new rtl pass function.
3855         * config/nds32/nds32.c
3856         (nds32_register_pass): New function to register pass.
3857         (nds32_register_passes): New function to register passes.
3858         * config/nds32/nds32.md (relax_group): New pattern.
3859         * config/nds32/nds32.opt (mrelax-hint): New option.
3860         * config/nds32/t-nds32 (nds32-relax-opt.o): New dependency.
3861
3862 2018-04-01  Kito Cheng  <kito.cheng@gmail.com>
3863
3864         * config/nds32/t-nds32: Modify files dependency.
3865
3866 2018-04-01  Chung-Ju Wu  <jasonwucj@gmail.com>
3867
3868         * config/nds32/nds32.h (FUNCTION_PROFILER): Output newline character.
3869         (PROFILE_HOOK): Define its implementation.
3870
3871 2018-04-01  Chung-Ju Wu  <jasonwucj@gmail.com>
3872
3873         * config/nds32/nds32.h (WCHAR_TYPE, WCHAR_TYPE_SIZE): Use unsigned int
3874         type and 32-bit size.
3875
3876 2018-04-01  Jakub Jelinek  <jakub@redhat.com>
3877
3878         PR middle-end/85090
3879         * config/i386/sse.md (V): Add V64QI and V32HI for TARGET_AVX512F.
3880         (V_128_256): New mode iterator.
3881         (*avx512dq_vextract<shuffletype>64x2_1 splitter): New define_split.
3882         (*avx512f_vextract<shuffletype>32x4_1 splitter): Likewise.
3883         (xop_pcmov_<mode><avxsizesuffix>): Use V_128_256 mode iterator instead
3884         of V.
3885         * config/i386/i386.c (ix86_expand_vector_set): Improve V32HImode and
3886         V64QImode expansion for !TARGET_AVX512BW && TARGET_AVX512F.
3887
3888 2018-03-31  Segher Boessenkool  <segher@kernel.crashing.org>
3889
3890         PR target/83315
3891         * config/rs6000/xmmintrin.h (_mm_set_ps, _mm_max_ps): Handle (quiet)
3892         NaN inputs correctly.
3893
3894 2018-03-30  Peter Bergner  <bergner@vnet.ibm.com>
3895
3896         PR target/80546
3897         * config/rs6000/vsx.md (??r): New mode attribute.
3898         (*vsx_mov<mode>_64bit): Use it.
3899         (*vsx_mov<mode>_32bit): Likewise.
3900
3901 2018-03-30  Martin Sebor  <msebor@redhat.com>
3902
3903         PR tree-optimization/84818
3904         * builtins.c (check_access): Use warning_n.
3905
3906 2018-03-30  Aaron Sawdey  <acsawdey@linux.vnet.ibm.com>
3907
3908         PR target/83822
3909         * config/rs6000/rs6000-string.c (expand_compare_loop): Fix redundant
3910         condition.
3911         * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): Fix redundant
3912         condition.
3913
3914 2018-03-30  Julia Koval  <julia.koval@intel.com>
3915
3916         PR target/84413
3917         * x86-tune.def (movx, partial_reg_dependency): Enable for
3918         m_SKYLAKE_AVX512.
3919
3920 2018-03-29  Vladimir Makarov  <vmakarov@redhat.com>
3921
3922         PR inline-asm/84985
3923         * lra-constraints.c (process_alt_operands): Move setting
3924         this_alternative_matches below.
3925
3926 2018-03-29  Martin Liska  <mliska@suse.cz>
3927
3928         PR lto/84995.
3929         * doc/invoke.texi: Document how LTO works with debug info.
3930         Describe auto-load support of binutils.  Mention 'x86-64'
3931         as valid option value of -march option.
3932
3933 2018-03-29  Jakub Jelinek  <jakub@redhat.com>
3934
3935         * config/i386/sse.md (<avx512>_blendm<mode>): Use <sseintprefix>.
3936
3937         PR c/85094
3938         * fold-const.c (operand_equal_p): Handle DEBUG_BEGIN_STMT.
3939         For STATEMENT_LIST, pass down OEP_LEXICOGRAPHIC and maybe
3940         OEP_NO_HASH_CHECK for recursive call, to avoid exponential
3941         checking.
3942
3943 2018-03-28  Peter Bergner  <bergner@vnet.ibm.com>
3944
3945         PR target/84912
3946         * config/rs6000/rs6000.h (RS6000_BTM_POWERPC64): New define.
3947         (RS6000_BTM_COMMON): Add RS6000_BTM_POWERPC64.
3948         * config/rs6000/rs6000.c (rs6000_builtin_mask_calculate): Add support
3949         for RS6000_BTM_POWERPC64.
3950         (rs6000_invalid_builtin): Add handling for RS6000_BTM_POWERPC64
3951         (rs6000_builtin_mask_names): Add RS6000_BTM_POWERPC64.
3952         * config/rs6000/rs6000-builtin.def (BU_P7_POWERPC64_MISC_2): New macro
3953         definition.
3954         (DIVDE): Use it.
3955         (DIVDEU): Likewise.
3956
3957 2018-03-28 Carl Love  <cel@us.ibm.com>
3958
3959         Revert
3960         2017-09-27  Carl Love  <cel@us.ibm.com>
3961
3962         * config/rs6000/rs6000-builtin.def (BU_FP_1MISC_1): Add define macro.
3963         (FCTID, FCTIW): Add BU_FP_MISC_1 macro expansion for builtins.
3964         * config/rs6000/rs6000.md (lrintsfsi2): Add define_insn for the
3965         fctiw instruction.
3966
3967 2018-03-28  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
3968
3969         * config/rs6000/xmmintrin.h (_mm_max_pi16): Use __vector __bool
3970         instead of __vector bool.
3971         (_mm_max_pu8): Likewise.
3972         (_mm_min_pi16): Likewise.
3973
3974 2018-03-28  Peter Bergner  <bergner@vnet.ibm.com>
3975
3976         PR target/84912
3977         * config/rs6000/rs6000-builtin.def (DIVWEO): Delete macro expansion.
3978         (DIVWEUO): Likewise.
3979         (DIVDEO): Likewise.
3980         (DIVDEUO): Likewise.
3981         * config/rs6000/rs6000.c (builtin_function_type): Remove support for
3982         DIVWEUO and DIVDEUO.
3983         * config/rs6000/rs6000.md (UNSPEC_DIVEO, UNSPEC_DIVEUO): Delete unspecs.
3984         (UNSPEC_DIV_EXTEND): Remove deleted unspecs.
3985         (div_extend): Likewise.
3986         * doc/extend.texi (__builtin_divweo): Remove documentation for deleted
3987         builtin function.
3988         (__builtin_divweuo): Likewise.
3989         (__builtin_divdeo): Likewise.
3990         (__builtin_divdeuo): Likewise.
3991
3992 2018-03-28  Jakub Jelinek  <jakub@redhat.com>
3993
3994         PR target/85095
3995         * config/i386/i386.md (*add<mode>3_carry_0, *addsi3_carry_zext_0,
3996         *sub<mode>3_carry_0, *subsi3_carry_zext_0): New patterns.
3997
3998         PR tree-optimization/82004
3999         * gimple-match-head.c (optimize_pow_to_exp): New function.
4000         * match.pd (pow(C,x) -> exp(log(C)*x)): Wrap with #if GIMPLE.
4001         Don't fold to exp if optimize_pow_to_exp is false.
4002
4003 2018-03-28  Martin Liska  <mliska@suse.cz>
4004
4005         PR other/84819
4006         * calls.c (initialize_argument_information): Fix trailing space.
4007         * common.opt: Fix typo and provide better explanation for
4008         -fsanitize-coverage option.
4009         * config/i386/i386.opt: Fix typo.
4010
4011 2018-03-28  Jakub Jelinek  <jakub@redhat.com>
4012             Martin Liska  <mliska@suse.cz>
4013
4014         PR sanitizer/85081
4015         * gimplify.c (asan_poison_variable): Don't do the check for
4016         gimplify_omp_ctxp here.
4017         (gimplify_decl_expr): Do it here.
4018         (gimplify_target_expr): Likewise.
4019
4020 2018-03-28  Martin Liska  <mliska@suse.cz>
4021
4022         PR target/84988
4023         * config/i386/i386.c (ix86_function_arg_advance): Do not call
4024         chkp_type_bounds_count if MPX is not enabled.
4025
4026 2018-03-27  Chung-Ju Wu  <jasonwucj@gmail.com>
4027
4028         * config/nds32/nds32.h (BRANCH_COST): Adjust cost.
4029
4030 2018-03-27  Michael Meissner  <meissner@linux.vnet.ibm.com>
4031
4032         PR target/84914
4033         * config/rs6000/rs6000.c (create_complex_muldiv): New helper
4034         function to create the function decl for complex long double
4035         multiply and divide for -mabi=ieeelongdouble.
4036         (init_float128_ieee): Call it.
4037
4038 2018-03-27  H.J. Lu  <hongjiu.lu@intel.com>
4039
4040         PR target/85044
4041         * config/i386/i386.c (ix86_trampoline_init): Insert ENDBR for
4042         -fcf-protection=branch -mibt.
4043         * config/i386/i386.h (TRAMPOLINE_SIZE): Increased by 4 bytes.
4044
4045 2018-03-27  Ramana Radhakrishnan  <ramana.radhakrishnan@arm.com>
4046
4047         PR target/81863
4048         * config/arm/arm.c (arm_valid_symbolic_address): Handle arm_word_relocations
4049
4050 2018-03-27  Cesar Philippidis  <cesar@codesourcery.com>
4051
4052         PR target/85056
4053         * config/nvptx/nvptx.c (nvptx_assemble_decl_begin): Add '[]' to
4054         extern array declarations.
4055
4056 2018-03-27  Richard Biener  <rguenther@suse.de>
4057
4058         PR middle-end/84067
4059         * match.pd ((A * C) +- (B * C) -> (A+-B) * C): Guard with
4060         explicit single_use checks.
4061
4062 2018-03-27  Richard Biener  <rguenther@suse.de>
4063
4064         PR tree-optimization/85082
4065         * tree-ssa-sccvn.c (vn_reference_lookup_or_insert_for_pieces):
4066         Valueize the VUSE.
4067
4068 2018-03-27  Ramana Radhakrishnan  <ramana.radhakrishnan@arm.com>
4069
4070         * config.gcc (aarch64*-*-linux*): New TARGET_DEFAULT_ASYNC_UNWIND_TABLES
4071         * common/config/aarch64/aarch64-common.c (aarch64_optimization_table[]):
4072         Turn on fasynchronous-unwind-tables and funwind-tables.
4073
4074 2018-03-26  Uros Bizjak  <ubizjak@gmail.com>
4075
4076         PR target/85073
4077         * config/i386/i386.md (*bmi_blsr_<mode>_cmp): New insn pattern.
4078         (*bmi_blsr_<mode>_ccz): Ditto.
4079
4080 2018-03-26  Tom de Vries  <tom@codesourcery.com>
4081
4082         PR tree-optimization/85063
4083         * omp-general.c (offloading_function_p): New function.  Factor out
4084         of ...
4085         * omp-offload.c (pass_omp_target_link::gate): ... here.
4086         * omp-general.h (offloading_function_p): Declare.
4087         * tree-switch-conversion.c (build_one_array): Mark CSWTCH.x variable
4088         with attribute omp declare target for offloading functions.
4089
4090 2018-03-24  Richard Sandiford  <richard.sandiford@linaro.org>
4091
4092         PR tree-optimization/84005
4093         * tree-data-ref.h (get_base_for_alignment): Declare.
4094         * tree-data-ref.c (get_base_for_alignment_1): New function.
4095         (get_base_for_alignment): Likewise.
4096         * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Use
4097         get_base_for_alignment to find a suitable base object, instead
4098         of always using drb->base_address.
4099
4100 2018-03-23  Jakub Jelinek  <jakub@redhat.com>
4101
4102         PR inline-asm/85022
4103         * emit-rtl.c (init_emit_regs): Indicate that VOIDmode MEMs don't have
4104         known size by default.
4105
4106 2018-03-23  Vladimir Makarov  <vmakarov@redhat.com>
4107
4108         PR inline-asm/85030
4109         * lra-constraints.c (process_alt_operands): Don't match BLKmode
4110         and non BLKmode operands.
4111
4112 2018-03-23  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
4113
4114         PR target/85026
4115         * config/arm/arm.md (unaligned_loadhis): Remove first alternative.
4116         Clean up attributes.
4117
4118 2018-03-23  Richard Biener  <rguenther@suse.de>
4119
4120         PR debug/85020
4121         * dwarf2out.c (rtl_for_decl_location): Do not generate RTL early when
4122         we are going to emit early debug for LTO.
4123
4124 2018-03-23  Jakub Jelinek  <jakub@redhat.com>
4125
4126         PR inline-asm/85034
4127         * function.c (match_asm_constraints_1): Don't optimize if input
4128         doesn't satisfy general_operand predicate for output's mode.
4129
4130         PR inline-asm/85022
4131         * alias.c (write_dependence_p): Don't require for x_canonicalized
4132         non-VOIDmode if x has VOIDmode.
4133
4134         PR sanitizer/85029
4135         * sanopt.c (maybe_optimize_ubsan_ptr_ifn): If DECL_REGISTER (base),
4136         just don't try to optimize it rather than assert it never happens.
4137
4138 2018-03-22  Kelvin Nilsen  <kelvin@gcc.gnu.org>
4139
4140         * config/rs6000/rs6000-builtin.def: Remove various BU_ALTIVEC_X
4141         macro expansions for definition of ST_INTERNAL_<mode> and
4142         LD_INTERNAL_<mode> builtins.
4143         * config/rs6000/rs6000-protos.h (rs6000_address_for_altivec):
4144         Remove prototype.
4145         * config/rs6000/rs6000.c (altivec_expand_ld_builtin): Delete this
4146         function.
4147         (altivec_expand_st_builtin): Likewise.
4148         (altivec_expand_builtin): Remove calls to deleted functions.
4149         (rs6000_address_for_altivec): Delete this function.
4150         * config/rs6000/vector.md: Remove expands for
4151         vector_altivec_load_<mode> and vector_altivec_store_<mode>.
4152
4153 2018-03-22  Sudakshina Das  <sudi.das@arm.com>
4154
4155         PR target/84826
4156         * config/arm/arm.h (machine_function): Add static_chain_stack_bytes.
4157         * config/arm/arm.c (arm_compute_static_chain_stack_bytes): Avoid
4158         re-computing once computed.
4159         (arm_expand_prologue): Compute machine->static_chain_stack_bytes.
4160         (arm_init_machine_status): Initialize
4161         machine->static_chain_stack_bytes.
4162
4163 2018-03-22  Kelvin Nilsen  <kelvin@gcc.gnu.org>
4164
4165         PR target/84760
4166         * doc/extend.texi: Add four new prototypes for vec_ld.
4167         * config/rs6000/rs6000-builtin.def (LVX_V1TI): Reorder symbol
4168         definitions for more logical presentation.
4169         * config/rs6000/rs6000-c.c: (altivec_overloaded_builtins): Add
4170         entries for V1TI variants of __builtin_altivec_ld builtin.
4171         * config/rs6000/rs6000.c: (altivec_expand_lv_builtin): Add test and
4172         handling of V1TI variant of LVX icode pattern.
4173         (altivec_expand_builtin): Add case for ALTIVEC_BUILTIN_LVX_V1TI.
4174         (rs6000_gimple_fold_builtin): Likewise.
4175         (altivec_init_builtins): Add code to define
4176         __builtin_altivec_lvx_v1ti function.
4177
4178 2018-03-22  Jakub Jelinek  <jakub@redhat.com>
4179
4180         PR inline-asm/84941
4181         * function.c (match_asm_constraints_1): Don't do the optimization
4182         if input isn't a REG, SUBREG, MEM or constant.
4183
4184 2018-03-22  Tom de Vries  <tom@codesourcery.com>
4185
4186         PR tree-optimization/84956
4187         * tree-ssa-tail-merge.c (find_clusters_1): Skip bbs with
4188         bb_has_abnormal_pred.
4189
4190 2018-03-22  Jakub Jelinek  <jakub@redhat.com>
4191
4192         PR sanitizer/85018
4193         * dwarf2asm.c (dw2_output_indirect_constant_1): Set
4194         DECL_INITIAL (decl) to decl at the end.
4195         * varasm.c (use_blocks_for_decl_p): Revert the 2018-03-20 change,
4196         adjust the comment.
4197
4198 2018-03-21  Joseph Myers  <joseph@codesourcery.com>
4199
4200         * doc/extend.texi (__builtin_tgmath): Document when complex
4201         integer types are treated as _Complex _Float64.
4202
4203 2018-03-21  Tom de Vries  <tom@codesourcery.com>
4204
4205         * doc/extend.texi (__builtin_extend_pointer): Remove pasto.
4206
4207 2018-03-21  Jakub Jelinek  <jakub@redhat.com>
4208
4209         PR tree-optimization/84960
4210         * tree-cfg.c (remove_bb): Don't move forced labels into bb->prev_bb
4211         if it is ENTRY block, move them into single succ of ENTRY in that case.
4212
4213 2018-03-21  Richard Sandiford  <richard.sandiford@linaro.org>
4214
4215         PR tree-optimization/84811
4216         * poly-int.h (poly_span_traits): Remove the T3 parameter and
4217         promote HOST_WIDE_INT T2 - T1 results to unsigned HOST_WIDE_INT.
4218         (maybe_in_range_p, known_in_range_p, ranges_known_overlap_p):
4219         (known_subrange_p): Update accordingly.  Cast each value involved
4220         in the size comparison, rather than casting the result of the
4221         subtraction.
4222
4223 2018-03-21  Jakub Jelinek  <jakub@redhat.com>
4224
4225         PR tree-optimization/84982
4226         * gimple-ssa-store-merging.c (invert_op): Handle boolean inversion
4227         by flipping the least significant bit rather than all bits from
4228         bitpos to bitpos + bitsize - 1.
4229
4230 2018-03-21  Nathan Sidwell  <nathan@acm.org>
4231
4232         * doc/extend.texi (Deprecated Features): Remove mention of
4233         long-deleted deprecations.
4234
4235 2018-03-21  Rainer Orth  <ro@CeBiTec.Uni-Bielefeld.DE>
4236
4237         PR jit/84288
4238         * configure.ac (gcc_cv_ld_soname) <*-*-solaris2*>: Set.
4239         * configure: Regenerate.
4240
4241 2018-03-21  Tom de Vries  <tom@codesourcery.com>
4242
4243         PR tree-optimization/83126
4244         * tree-parloops.c (num_phis): New function.
4245         (gen_parallel_loop): Detect and handle canonicalize_loop_ivs failure.
4246
4247 2018-03-21  Nathan Sidwell  <nathan@acm.org>
4248
4249         * doc/extend.texi (Deprecated Features): Update deprecated flags,
4250         mention anon-struct/union members and trailing attributes.
4251
4252 2018-03-21  Bin Cheng  <bin.cheng@arm.com>
4253
4254         PR tree-optimization/84969
4255         * tree-loop-distribution.c (fuse_memset_builtins): Don't reorder
4256         builtin memset partitions if they set different rhs values.
4257
4258 2018-03-21  Jakub Jelinek  <jakub@redhat.com>
4259
4260         PR rtl-optimization/84989
4261         * simplify-rtx.c (simplify_unary_operation_1): Don't try to simplify
4262         VEC_DUPLICATE with scalar result mode.
4263
4264 2018-03-21  Martin Liska  <mliska@suse.cz>
4265
4266         PR ipa/84963
4267         * ipa-icf.c (sem_item_optimizer::fixup_points_to_sets): Remove
4268         not intended return statement.
4269
4270 2018-03-21  Martin Liska  <mliska@suse.cz>
4271
4272         PR target/84988
4273         * tree-chkp.c (CHKP_ARRAY_MAX_CHECK_STEPS): Define a new macro.
4274         (chkp_find_bound_slots_1): Limit number of iterations.
4275
4276 2018-03-20  David H. Gutteridge  <dhgutteridge@sympatico.ca>
4277
4278         PR target/84838
4279         * Minor grammar fixes for x86 options.
4280
4281 2018-03-20  Jakub Jelinek  <jakub@redhat.com>
4282
4283         PR debug/84875
4284         * dce.c (delete_unmarked_insns): Don't remove frame related noop moves
4285         holding REG_CFA_RESTORE notes, instead turn them into a USE.
4286
4287 2018-03-20  Peter Bergner  <bergner@vnet.ibm.com>
4288
4289         PR target/83789
4290         * config/rs6000/altivec.md (altivec_lvx_<mode>_2op): Delete define_insn.
4291         (altivec_lvx_<mode>_1op): Likewise.
4292         (altivec_stvx_<mode>_2op): Likewise.
4293         (altivec_stvx_<mode>_1op): Likewise.
4294         (altivec_lvx_<VM2:mode>): New define_expand.
4295         (altivec_stvx_<VM2:mode>): Likewise.
4296         (altivec_lvx_<VM2:mode>_2op_<P:mptrsize>): New define_insn.
4297         (altivec_lvx_<VM2:mode>_1op_<P:mptrsize>): Likewise.
4298         (altivec_stvx_<VM2:mode>_2op_<P:mptrsize>): Likewise.
4299         (altivec_stvx_<VM2:mode>_1op_<P:mptrsize>): Likewise.
4300         * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Use new expanders.
4301         (rs6000_gen_lvx): Likewise.
4302         * config/rs6000/rs6000.c (altivec_expand_lv_builtin): Likewise.
4303         (altivec_expand_stv_builtin): Likewise.
4304         (altivec_expand_builtin): Likewise.
4305         * config/rs6000/vector.md: Likewise.
4306
4307 2018-03-20  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
4308
4309         PR target/82518
4310         * config/arm/arm.c (arm_array_mode_supported_p): Return false for
4311         BYTES_BIG_ENDIAN.
4312
4313 2018-03-20  Richard Biener  <rguenther@suse.de>
4314
4315         PR target/84986
4316         * config/i386/i386.c (ix86_add_stmt_cost): Only cost
4317         sign-conversions as zero, fall back to standard scalar_stmt
4318         cost for the rest.
4319
4320 2018-03-20  Martin Liska  <mliska@suse.cz>
4321
4322         PR ipa/84825
4323         * predict.c (rebuild_frequencies): Handle case when we have
4324         PROFILE_ABSENT, but flag_guess_branch_prob is false.
4325
4326 2018-03-20  Jakub Jelinek  <jakub@redhat.com>
4327
4328         PR target/84990
4329         * dwarf2asm.c (dw2_output_indirect_constant_1): Temporarily turn off
4330         flag_section_anchors.
4331         * varasm.c (use_blocks_for_decl_p): Remove hack for
4332         dw2_force_const_mem.
4333
4334         PR target/84845
4335         * config/aarch64/aarch64.md (*aarch64_reg_<mode>3_neg_mask2): Rename
4336         to ...
4337         (*aarch64_<optab>_reg_<mode>3_neg_mask2): ... this.  If pseudos can't
4338         be created, use lowpart_subreg of operands[0] rather than operands[0]
4339         itself.
4340         (*aarch64_reg_<mode>3_minus_mask): Rename to ...
4341         (*aarch64_ashl_reg_<mode>3_minus_mask): ... this.
4342         (*aarch64_<optab>_reg_di3_mask2): Use const_int_operand predicate
4343         and n constraint instead of aarch64_shift_imm_di and Usd.
4344         (*aarch64_reg_<optab>_minus<mode>3): Rename to ...
4345         (*aarch64_<optab>_reg_minus<mode>3): ... this.
4346
4347 2018-03-20  Sudakshina Das  <sudi.das@arm.com>
4348
4349         PR target/82989
4350         * config/arm/neon.md (ashldi3_neon): Update ?s for constraints
4351         to favor GPR over NEON registers.
4352         (<shift>di3_neon): Likewise.
4353
4354 2018-03-20  Tom de Vries  <tom@codesourcery.com>
4355
4356         PR target/84952
4357         * config/nvptx/nvptx.c (nvptx_single): Don't neuter bar.sync.
4358         (nvptx_process_pars): Emit bar.sync asap and alap.
4359
4360 2018-03-20  Tom de Vries  <tom@codesourcery.com>
4361
4362         PR target/84954
4363         * config/nvptx/nvptx.c (prevent_branch_around_nothing): Also update
4364         seen_label if seen_label is already set.
4365
4366 2018-03-20  Jakub Jelinek  <jakub@redhat.com>
4367
4368         PR target/84945
4369         * config/i386/i386.c (fold_builtin_cpu): For features above 31
4370         use __cpu_features2 variable instead of __cpu_model.__cpu_features[0].
4371         Use 1U instead of 1.  Formatting fixes.
4372
4373         PR c/84953
4374         * builtins.c (fold_builtin_strpbrk): For strpbrk(x, "") use type
4375         instead of TREE_TYPE (s1) for the return value.
4376
4377 2018-03-19  Jakub Jelinek  <jakub@redhat.com>
4378
4379         PR tree-optimization/84946
4380         * gimple-ssa-store-merging.c (mem_valid_for_store_merging): Compute
4381         bitsize + bitsize in poly_uint64 rather than poly_int64.
4382
4383         PR sanitizer/78651
4384         * dwarf2asm.c: Include fold-const.c.
4385         (dw2_output_indirect_constant_1): Set DECL_INITIAL (decl) to ADDR_EXPR
4386         of decl rather than decl itself.
4387
4388         PR rtl-optimization/84643
4389         * memmodel.h (enum memmodel): Add MEMMODEL_MAX enumerator.
4390
4391 2018-03-19  Maxim Ostapenko  <m.ostapenko@samsung.com>
4392
4393         PR sanitizer/78651
4394         * dwarf2asm.c (dw2_output_indirect_constant_1): Disable ASan before
4395         calling assemble_variable.
4396
4397 2018-03-19  Sudakshina Das  <sudi.das@arm.com>
4398
4399         PR target/81647
4400         * config/aarch64/aarch64-simd.md (vec_cmp<mode><v_int_equiv>): Modify
4401         instructions for UNLT, UNLE, UNGT, UNGE, UNEQ, UNORDERED and ORDERED.
4402
4403 2018-03-19  Jim Wilson  <jimw@sifive.com>
4404
4405         PR bootstrap/84856
4406         * config/riscv/riscv.c (riscv_function_arg_boundary): Use
4407         PREFERRED_STACK_BOUNDARY instead of STACK_BOUNDARY.
4408         (riscv_first_stack_step): Likewise.
4409         (riscv_option_override): Use STACK_BOUNDARY instead of
4410         MIN_STACK_BOUNDARY.
4411         * config/riscv/riscv.h (STACK_BOUNDARY): Renamed from
4412         MIN_STACK_BOUNDARY.
4413         (BIGGEST_ALIGNMENT): Set to 128.
4414         (PREFERRED_STACK_BOUNDARY): Renamed from STACK_BOUNDARY.
4415         (RISCV_STACK_ALIGN): Use PREFERRED_STACK_BOUNDARY instead of
4416         STACK_BOUNDARY.
4417
4418 2018-03-19  Richard Biener  <rguenther@suse.de>
4419
4420         PR tree-optimization/84933
4421         * tree-vrp.c (set_and_canonicalize_value_range): Treat out-of-bound
4422         values as -INF/INF when canonicalizing an ANTI_RANGE to a RANGE.
4423
4424 2018-03-19  Richard Biener  <rguenther@suse.de>
4425
4426         PR tree-optimization/84859
4427         * tree-ssa-phiopt.c (single_trailing_store_in_bb): New function.
4428         (cond_if_else_store_replacement): Perform sinking operation on
4429         single-store BBs regardless of MAX_STORES_TO_SINK setting.
4430         Generalize what a BB with a single eligible store is.
4431
4432 2018-03-19  Richard Biener  <rguenther@suse.de>
4433
4434         PR tree-optimization/84929
4435         * tree-data-ref.c (analyze_siv_subscript_cst_affine): Guard
4436         chrec_is_positive against non-chrec arg.
4437
4438 2018-03-19  Tamar Christina  <tamar.christina@arm.com>
4439
4440         PR target/84711
4441         * config/arm/arm.c (arm_can_change_mode_class): revert r258554.
4442
4443 2018-03-18  Martin Liska  <mliska@suse.cz>
4444
4445         PR rtl-optimization/84635
4446         * regrename.c (build_def_use): Use matches_mode only when
4447         matches >= 0.
4448
4449 2018-03-18  Richard Sandiford  <richard.sandiford@linaro.org>
4450
4451         PR tree-optimization/84913
4452         * tree-vect-loop.c (vectorizable_reduction): Don't try to
4453         vectorize chains of COND_EXPRs.
4454
4455 2018-03-18  Chung-Ju Wu  <jasonwucj@gmail.com>
4456
4457         * config/nds32/nds32.h (MAX_REGS_PER_ADDRESS): Fix the value.
4458
4459 2018-03-18  Chung-Ju Wu  <jasonwucj@gmail.com>
4460
4461         * config/nds32/nds32.h (LOGICAL_OP_NON_SHORT_CIRCUIT): Define.
4462
4463 2018-03-18  Chung-Ju Wu  <jasonwucj@gmail.com>
4464
4465         * config/nds32/nds32.h (CLZ_DEFINED_VALUE_AT_ZERO): Define.
4466
4467 2018-03-17  Chung-Ju Wu  <jasonwucj@gmail.com>
4468             Kito Cheng  <kito.cheng@gmail.com>
4469
4470         * config/nds32/nds32-protos.h (nds32_adjust_reg_alloc_order): Declare.
4471         * config/nds32/nds32.c (nds32_reg_alloc_order_for_speed): New array.
4472         (nds32_adjust_reg_alloc_order): New function.
4473         * config/nds32/nds32.h (ADJUST_REG_ALLOC_ORDER): Define.
4474
4475 2018-03-17  Kito Cheng  <kito.cheng@gmail.com>
4476
4477         * config/nds32/nds32.c (nds32_asm_output_mi_thunk,
4478         nds32_print_operand, nds32_print_operand_address): Use
4479         HOST_WIDE_INT_PRINT_DEC instead.
4480
4481 2018-03-17  Chung-Ju Wu  <jasonwucj@gmail.com>
4482
4483         * config/nds32/nds32.c (nds32_register_priority): Modify cost.
4484
4485 2018-03-17  Jakub Jelinek  <jakub@redhat.com>
4486
4487         PR target/84902
4488         * config/i386/i386.c (initial_ix86_tune_features,
4489         initial_ix86_arch_features): Use unsigned HOST_WIDE_INT rather than
4490         unsigned long long.
4491         (set_ix86_tune_features): Change ix86_tune_mask from unsigned int
4492         to unsigned HOST_WIDE_INT, initialize to HOST_WIDE_INT_1U << ix86_tune
4493         rather than 1u << ix86_tune.  Formatting fix.
4494         (ix86_option_override_internal): Change ix86_arch_mask from
4495         unsigned int to unsigned HOST_WIDE_INT, initialize to
4496         HOST_WIDE_INT_1U << ix86_arch rather than 1u << ix86_arch.
4497         (ix86_function_specific_restore): Likewise.
4498
4499 2018-03-16  Jakub Jelinek  <jakub@redhat.com>
4500
4501         PR target/84899
4502         * postreload.c (reload_combine_recognize_pattern): Perform
4503         INTVAL addition in unsigned HOST_WIDE_INT type to avoid UB and
4504         truncate_int_for_mode the result for the destination's mode.
4505
4506         PR c/84909
4507         * hsa-gen.c (mem_type_for_type): Fix comment typo.
4508         * tree-vect-loop-manip.c (vect_create_cond_for_niters_checks):
4509         Likewise.
4510         * gimple-ssa-warn-restrict.c (builtin_memref::set_base_and_offset):
4511         Likewise.
4512
4513 2018-03-16  Vladimir Makarov  <vmakarov@redhat.com>
4514
4515         PR target/84876
4516         * lra-assigns.c (lra_split_hard_reg_for): Don't use
4517         regno_allocno_class_array and sorted_pseudos.
4518         * lra-constraints.c (spill_hard_reg_in_range): Ignore hard regs in
4519         insns where regno is used.
4520
4521 2018-03-16  Martin Liska  <mliska@suse.cz>
4522
4523         PR ipa/84833
4524         * multiple_target.c (create_dispatcher_calls): Redirect
4525         reference in the symbol table.
4526
4527 2018-03-16  Martin Liska  <mliska@suse.cz>
4528
4529         PR ipa/84722
4530         * multiple_target.c (create_dispatcher_calls): Redirect also
4531         an alias.
4532
4533 2018-03-16  Jakub Jelinek  <jakub@redhat.com>
4534
4535         PR c++/79937
4536         PR c++/82410
4537         * tree.h (TARGET_EXPR_NO_ELIDE): Define.
4538         * gimplify.c (gimplify_modify_expr_rhs): Don't elide TARGET_EXPRs with
4539         TARGET_EXPR_NO_ELIDE flag set unless *expr_p is INIT_EXPR.
4540
4541 2018-03-16  Julia Koval  <julia.koval@intel.com>
4542
4543         * doc/invoke.texi (Skylake Server): Add CLWB.
4544         Cannonlake): Remove CLWB.
4545
4546 2018-03-16  Jakub Jelinek  <jakub@redhat.com>
4547
4548         PR tree-optimization/84841
4549         * tree-ssa-reassoc.c (INTEGER_CONST_TYPE): Change to 1 << 4 from
4550         1 << 3.
4551         (FLOAT_ONE_CONST_TYPE): Define.
4552         (constant_type): Return FLOAT_ONE_CONST_TYPE for -1.0 and 1.0.
4553         (sort_by_operand_rank): Put entries with higher constant_type last
4554         rather than first to match comments.
4555
4556 2018-03-15  Sandra Loosemore  <sandra@codesourcery.com>
4557
4558         * config/nios2/nios2.md (movsi_internal): Fix thinko in
4559         split predicate.
4560
4561 2018-03-15  Jakub Jelinek  <jakub@redhat.com>
4562
4563         PR c++/79085
4564         * calls.c (expand_call): For TREE_ADDRESSABLE rettype ignore alignment
4565         check and use address of target always.
4566
4567 2018-03-15  H.J. Lu  <hongjiu.lu@intel.com>
4568
4569         PR target/84574
4570         * config/i386/i386.c (indirect_thunk_needed): Update comments.
4571         (indirect_thunk_bnd_needed): Likewise.
4572         (indirect_thunks_used): Likewise.
4573         (indirect_thunks_bnd_used): Likewise.
4574         (indirect_return_needed): New.
4575         (indirect_return_bnd_needed): Likewise.
4576         (output_indirect_thunk_function): Add a bool argument for
4577         function return.
4578         (output_indirect_thunk_function): Don't generate alias for
4579         function return thunk.
4580         (ix86_code_end): Call output_indirect_thunk_function to generate
4581         function return thunks.
4582         (ix86_output_function_return): Set indirect_return_bnd_needed
4583         and indirect_return_needed instead of indirect_thunk_bnd_needed
4584         and indirect_thunk_needed.
4585
4586 2018-03-15  Olga Makhotina  <olga.makhotina@intel.com>
4587
4588         * config/i386/sgxintrin.h (_enclv_u32): New intrinsic.
4589         (__enclv_bc, __enclv_cd, __enclv_generic): New definitions.
4590         (ERDINFO, ETRACKC, ELDBC, ELDUC): New leaves.
4591
4592 2018-03-15  David Malcolm  <dmalcolm@redhat.com>
4593             Paul Hua <paul.hua.gm@gmail.com>
4594
4595         PR c/84852
4596         * gcc.dg/fixits-pr84852-1.c: Fix filename in dg-regexp.
4597
4598 2018-03-15  Segher Boessenkool  <segher@kernel.crashing.org>
4599
4600         * config/rs6000/rs6000.c (abi_v4_pass_in_fpr): Add back the
4601         TARGET_DOUBLE_FLOAT and TARGET_SINGLE_FLOAT conditions on the DFmode
4602         resp. SFmode cases.
4603
4604 2018-03-15  Tamar Christina  <tamar.christina@arm.com>
4605
4606         PR target/84711
4607         * config/arm/arm.c (arm_can_change_mode_class): Use GET_MODE_UNIT_SIZE
4608         instead of GET_MODE_SIZE when comparing Units.
4609
4610 2018-03-15  Vladimir Mezentsev  <vladimir.mezentsev@oracle.com>
4611
4612         PR target/68256
4613         * varasm.c (hash_section): Return an unchangeble hash value
4614         * config/aarch64/aarch64.c (aarch64_use_blocks_for_constant_p):
4615         Return !aarch64_can_use_per_function_literal_pools_p ().
4616
4617 2018-03-15  Jakub Jelinek  <jakub@redhat.com>
4618
4619         PR target/84860
4620         * optabs.c (emit_conditional_move): Pass address of cmode's copy
4621         rather than address of cmode as last argument to prepare_cmp_insn.
4622
4623 2018-03-15  Julia Koval  <julia.koval@intel.com>
4624
4625         * config/i386/i386.c (F_AVX512VBMI2, F_GFNI, F_VPCLMULQDQ,
4626         F_AVX512VNNI, F_AVX512BITALG): New.
4627
4628 2018-03-14  John David Anglin  <danglin@gcc.gnu.org>
4629
4630         PR target/83451
4631         * config/pa/pa.c (pa_emit_move_sequence):  Always emit secondary reload
4632         insn for floating-point loads and stores.
4633
4634 2018-03-14  Carl Love  <cel@us.ibm.com>
4635
4636         * config/rs6000/rs6000-c.c: Add macro definitions for
4637         ALTIVEC_BUILTIN_VEC_PERMXOR.
4638         * config/rs6000/rs6000.h: Add #define for vec_permxor builtin.
4639         * config/rs6000/rs6000-builtin.def: Add macro expansions for VPERMXOR.
4640         * config/rs6000/altivec.md (altivec_vpermxor): New define expand.
4641         * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Add case
4642         UNSPEC_VPERMXOR.
4643         * config/doc/extend.texi: Add prototypes for vec_permxor.
4644
4645 2018-03-14  David Malcolm  <dmalcolm@redhat.com>
4646
4647         PR c/84852
4648         * diagnostic-show-locus.c (class layout_point): Convert m_line
4649         from int to linenum_type.
4650         (line_span::comparator): Use linenum "compare" function when
4651         comparing line numbers.
4652         (test_line_span): New function.
4653         (layout_range::contains_point): Convert param "row" from int to
4654         linenum_type.
4655         (layout_range::intersects_line_p): Likewise.
4656         (layout::will_show_line_p): Likewise.
4657         (layout::print_source_line): Likewise.
4658         (layout::should_print_annotation_line_p): Likewise.
4659         (layout::print_annotation_line): Likewise.
4660         (layout::print_leading_fixits): Likewise.
4661         (layout::annotation_line_showed_range_p): Likewise.
4662         (struct line_corrections): Likewise for field m_row.
4663         (line_corrections::line_corrections): Likewise for param "row".
4664         (layout::print_trailing_fixits): Likewise.
4665         (layout::get_state_at_point): Likewise.
4666         (layout::get_x_bound_for_row): Likewise.
4667         (layout::print_line): Likewise.
4668         (diagnostic_show_locus): Likewise for locals "last_line" and
4669         "row".
4670         (selftest::diagnostic_show_locus_c_tests): Call test_line_span.
4671         * input.c (selftest::test_linenum_comparisons): New function.
4672         (selftest::input_c_tests): Call it.
4673         * selftest.c (selftest::test_assertions): Test ASSERT_GT,
4674         ASSERT_GT_AT, ASSERT_LT, and ASSERT_LT_AT.
4675         * selftest.h (ASSERT_GT): New macro.
4676         (ASSERT_GT_AT): New macro.
4677         (ASSERT_LT): New macro.
4678         (ASSERT_LT_AT): New macro.
4679
4680 2018-03-14  Segher Boessenkool  <segher@kernel.crashing.org>
4681
4682         PR rtl-optimization/84780
4683         * combine.c (distribute_links): Don't make a link based on pc_rtx.
4684
4685 2018-03-14  Martin Liska  <mliska@suse.cz>
4686
4687         * tree.c (record_node_allocation_statistics): Use
4688         get_stats_node_kind.
4689         (get_stats_node_kind): New function extracted from
4690         record_node_allocation_statistics.
4691         (free_node): Use get_stats_node_kind.
4692
4693 2018-03-14  Richard Biener  <rguenther@suse.de>
4694
4695         * tree-ssa-pre.c (compute_antic_aux): Remove code that asserts
4696         that the value-set of ANTIC_IN doesn't grow.
4697
4698         Revert
4699         * tree-ssa-pre.c (struct bb_bitmap_sets): Add visited_with_visited_succs
4700         member.
4701         (BB_VISITED_WITH_VISITED_SUCCS): New define.
4702         (compute_antic): Initialize BB_VISITED_WITH_VISITED_SUCCS.
4703
4704 2018-03-14  Julia Koval  <julia.koval@intel.com>
4705
4706         * config.gcc (icelake-client, icelake-server): New.
4707         (icelake): Remove.
4708         * config/i386/i386.c (initial_ix86_tune_features): Extend to 64 bit.
4709         (initial_ix86_arch_features): Ditto.
4710         (PTA_SKYLAKE): Add SGX.
4711         (PTA_ICELAKE): Remove.
4712         (PTA_ICELAKE_CLIENT): New.
4713         (PTA_ICELAKE_SERVER): New.
4714         (ix86_option_override_internal): Split up icelake on icelake client and
4715         icelake server.
4716         (get_builtin_code_for_version): Ditto.
4717         (fold_builtin_cpu): Ditto.
4718         * config/i386/driver-i386.c (config/i386/driver-i386.c): Ditto.
4719         * config/i386/i386-c.c (ix86_target_macros_internal): Ditto
4720         * config/i386/i386.h (processor_type): Ditto.
4721         * doc/invoke.texi: Ditto.
4722
4723 2018-03-14  Jakub Jelinek  <jakub@redhat.com>
4724
4725         PR sanitizer/83392
4726         * sanopt.c (maybe_optimize_ubsan_ptr_ifn): Handle also
4727         INTEGER_CST offset, add it together with bitpos / 8 and
4728         sign extend based on POINTER_SIZE.
4729
4730         PR target/84844
4731         Revert
4732         2017-04-20  Uros Bizjak  <ubizjak@gmail.com>
4733
4734         PR target/78090
4735         * config/i386/constraints.md (Yc): New register constraint.
4736         * config/i386/i386.md (*float<SWI48:mode><MODEF:mode>2_mixed):
4737         Use Yc constraint for alternative 2 of operand 0.  Remove
4738         preferred_for_speed attribute.
4739
4740 2018-03-14  Richard Biener  <rguenther@suse.de>
4741
4742         PR tree-optimization/84830
4743         * tree-ssa-pre.c (compute_antic_aux): Intersect the new ANTIC_IN
4744         with the old one to avoid oscillations.
4745
4746 2018-03-13  Vladimir Makarov  <vmakarov@redhat.com>
4747
4748         PR target/83712
4749         * lra-assigns.c (find_all_spills_for): Ignore uninteresting
4750         pseudos.
4751         (assign_by_spills): Return a flag of reload assignment failure.
4752         Do not process the reload assignment failures.  Do not spill other
4753         reload pseudos if they has the same reg class.  Update n if
4754         necessary.
4755         (lra_assign): Add a return arg.  Set up from the result of
4756         assign_by_spills call.
4757         (find_reload_regno_insns, lra_split_hard_reg_for): New functions.
4758         * lra-constraints.c (split_reg): Add a new arg.  Use it instead of
4759         usage_insns if it is not NULL.
4760         (spill_hard_reg_in_range): New function.
4761         (split_if_necessary, inherit_in_ebb): Pass a new arg to split_reg.
4762         * lra-int.h (spill_hard_reg_in_range, lra_split_hard_reg_for): New
4763         function prototypes.
4764         (lra_assign): Change prototype.
4765         * lra.c (lra): Add code to deal with fails by splitting hard reg
4766         live ranges.
4767
4768 2018-03-01  Palmer Dabbelt  <palmer@sifive.com>
4769
4770         * config/riscv/riscv.opt (mrelax): New option.
4771         * config/riscv/riscv.c (riscv_file_start): Emit ".option
4772         "norelax" when riscv_mrelax is disabled.
4773         * doc/invoke.texi (RISC-V): Document "-mrelax" and "-mno-relax".
4774
4775 2018-03-13  Aaron Sawdey  <acsawdey@linux.vnet.ibm.com>
4776
4777         PR target/84743
4778         * config/rs6000/rs6000.c (rs6000_reassociation_width): Disable parallel
4779         reassociation for int modes.
4780
4781 2018-03-13  Richard Sandiford  <richard.sandiford@linaro.org>
4782
4783         * tree-vect-loop-manip.c (vect_maybe_permute_loop_masks):
4784         Reverse the choice between VEC_UNPACK_LO_EXPR and VEC_UNPACK_HI_EXPR
4785         for big-endian.
4786         * config/aarch64/iterators.md (hi_lanes_optab): New int attribute.
4787         * config/aarch64/aarch64-sve.md
4788         (*aarch64_sve_<perm_insn><perm_hilo><mode>): Rename to...
4789         (aarch64_sve_<perm_insn><perm_hilo><mode>): ...this.
4790         (*extend<mode><Vwide>2): Rename to...
4791         (aarch64_sve_extend<mode><Vwide>2): ...this.
4792         (vec_unpack<su>_<perm_hilo>_<mode>): Turn into a define_expand,
4793         renaming the old pattern to...
4794         (aarch64_sve_punpk<perm_hilo>_<mode>): ...this.  Only define
4795         unsigned packs.
4796         (vec_unpack<su>_<perm_hilo>_<SVE_BHSI:mode>): Turn into a
4797         define_expand, renaming the old pattern to...
4798         (aarch64_sve_<su>unpk<perm_hilo>_<SVE_BHSI:mode>): ...this.
4799         (*vec_unpacku_<perm_hilo>_<mode>_no_convert): Delete.
4800         (vec_unpacks_<perm_hilo>_<mode>): Take BYTES_BIG_ENDIAN into
4801         account when deciding which SVE instruction the optab should use.
4802         (vec_unpack<su_optab>_float_<perm_hilo>_vnx4si): Likewise.
4803
4804 2018-03-13  Richard Sandiford  <richard.sandiford@linaro.org>
4805
4806         * config/aarch64/aarch64.md (V4_REGNUM, V8_REGNUM, V12_REGNUM)
4807         (V20_REGNUM, V24_REGNUM, V28_REGNUM, P1_REGNUM, P2_REGNUM, P3_REGNUM)
4808         (P4_REGNUM, P5_REGNUM, P6_REGNUM, P8_REGNUM, P9_REGNUM, P10_REGNUM)
4809         (P11_REGNUM, P12_REGNUM, P13_REGNUM, P14_REGNUM): New define_constants.
4810         (tlsdesc_small_<mode>): Turn a define_expand and use
4811         tlsdesc_small_sve_<mode> for SVE.  Rename original define_insn to...
4812         (tlsdesc_small_advsimd_<mode>): ...this.
4813         (tlsdesc_small_sve_<mode>): New pattern.
4814
4815 2018-03-13  Richard Sandiford  <richard.sandiford@linaro.org>
4816
4817         * config/aarch64/iterators.md (UNSPEC_SMUL_HIGHPART)
4818         (UNSPEC_UMUL_HIGHPART): New constants.
4819         (MUL_HIGHPART): New int iteraor.
4820         (su): Handle UNSPEC_SMUL_HIGHPART and UNSPEC_UMUL_HIGHPART.
4821         * config/aarch64/aarch64-sve.md (<su>mul<mode>3_highpart): New
4822         define_expand.
4823         (*<su>mul<mode>3_highpart): New define_insn.
4824
4825 2018-03-13  Eric Botcazou  <ebotcazou@adacore.com>
4826
4827         PR lto/84805
4828         * ipa-devirt.c (odr_subtypes_equivalent_p): Do not get the ODR type of
4829         incomplete types.
4830
4831 2018-03-13  Martin Liska  <mliska@suse.cz>
4832
4833         PR ipa/84658.
4834         * (sem_item_optimizer::sem_item_optimizer): Initialize new
4835         vector.
4836         (sem_item_optimizer::~sem_item_optimizer): Release it.
4837         (sem_item_optimizer::merge_classes): Register variable aliases.
4838         (sem_item_optimizer::fixup_pt_set): New function.
4839         (sem_item_optimizer::fixup_points_to_sets): Likewise.
4840         * ipa-icf.h: Declare new variables and functions.
4841
4842 2018-03-13  Jakub Jelinek  <jakub@redhat.com>
4843
4844         PR middle-end/84834
4845         * match.pd ((A & C) != 0 ? D : 0): Use INTEGER_CST@2 instead of
4846         integer_pow2p@2 and test integer_pow2p in condition.
4847         (A < 0 ? C : 0): Similarly for @1.
4848
4849         PR middle-end/84831
4850         * stmt.c (parse_output_constraint): If the CONSTRAINT_LEN (*p, p)
4851         characters starting at p contain '\0' character, don't look beyond
4852         that.
4853
4854         PR target/84827
4855         * config/i386/i386.md (round<mode>2): For 387 fancy math, disable
4856         pattern if -ftrapping-math -fno-fp-int-builtin-inexact.
4857
4858         PR target/84828
4859         * reg-stack.c (change_stack): Change update_end var from int to
4860         rtx_insn *, if non-NULL don't update just BB_END (current_block), but
4861         also call set_block_for_insn on the newly added insns and rescan.
4862
4863         PR target/84786
4864         * config/i386/sse.md (sse2_loadhpd): Use Yv constraint rather than v
4865         on the last operand.
4866
4867         PR c++/84704
4868         * tree.c (stabilize_reference_1): Return save_expr (e) for
4869         STATEMENT_LIST even if it doesn't have side-effects.
4870
4871 2018-03-12  Jonathan Wakely  <jwakely@redhat.com>
4872
4873         * doc/invoke.texi (-mclflushopt): Fix spelling of option.
4874
4875 2018-03-12  Renlin Li  <renlin.li@arm.com>
4876
4877         * config/aarch64/aarch64.md (movhf_aarch64): Fix mode argument to
4878         aarch64_output_scalar_simd_mov_immediate.
4879
4880 2018-03-12  Martin Sebor  <msebor@redhat.com>
4881
4882         PR tree-optimization/83456
4883         * gimple-fold.c (gimple_fold_builtin_memory_op): Avoid warning
4884         for perfectly overlapping calls to memcpy.
4885         (gimple_fold_builtin_memory_chk): Same.
4886         (gimple_fold_builtin_strcpy): Handle no-warning.
4887         (gimple_fold_builtin_stxcpy_chk): Same.
4888         * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Handle no-warning.
4889
4890 2018-03-12  Segher Boessenkool  <segher@kernel.crashing.org>
4891
4892         * config/rs6000/rs6000.c (abi_v4_pass_in_fpr): Add bool "named"
4893         parameter.  Use it for SFmode.
4894         (rs6000_function_arg_advance_1): Adjust.
4895         (rs6000_function_arg): Adjust.
4896         (rs6000_gimplify_va_arg): Pass false for that new parameter.
4897
4898 2018-03-12  Segher Boessenkool  <segher@kernel.crashing.org>
4899
4900         PR rtl-optimization/84169
4901         PR rtl-optimization/84780
4902         * combine.c (can_combine_p): Check for a 2-insn combination whether
4903         the destination register is used between the two insns, too.
4904
4905 2018-03-12  Richard Biener  <rguenther@suse.de>
4906
4907         PR tree-optimization/84803
4908         * tree-if-conv.c (ifcvt_memrefs_wont_trap): Don't do anything
4909         for refs DR analysis didn't process.
4910
4911 2018-03-12  Richard Biener  <rguenther@suse.de>
4912
4913         PR tree-optimization/84777
4914         * tree-ssa-loop-ch.c (should_duplicate_loop_header_p): For
4915         force-vectorize loops ignore whether we are optimizing for size.
4916
4917 2018-03-12  Chung-Ju Wu  <jasonwucj@gmail.com>
4918
4919         * config/nds32/nds32.c (nds32_md_asm_adjust): New function.
4920         (TARGET_MD_ASM_ADJUST): Define.
4921
4922 2018-03-12  Monk Chiang  <sh.chiang04@gmail.com>
4923             Kito Cheng  <kito.cheng@gmail.com>
4924             Chung-Ju Wu  <jasonwucj@gmail.com>
4925
4926         * config/nds32/nds32.c (nds32_compute_stack_frame,
4927         nds32_emit_stack_push_multiple, nds32_emit_stack_pop_multiple,
4928         nds32_emit_stack_v3push, nds32_emit_stack_v3pop,
4929         nds32_emit_adjust_frame, nds32_expand_prologue, nds32_expand_epilogue,
4930         nds32_expand_prologue_v3push, nds32_expand_epilogue_v3pop): Refine.
4931         * config/nds32/nds32.h (NDS32_FIRST_CALLEE_SAVE_GPR_REGNUM,
4932         NDS32_LAST_CALLEE_SAVE_GPR_REGNUM, NDS32_V3PUSH_AVAILABLE_P): New.
4933         * config/nds32/nds32.md (prologue, epilogue): Use macro
4934         NDS32_V3PUSH_AVAILABLE_P to do checking.
4935
4936 2018-03-11  Jakub Jelinek  <jakub@redhat.com>
4937
4938         PR debug/58150
4939         * dwarf2out.c (gen_enumeration_type_die): Don't guard adding
4940         DW_AT_declaration for ENUM_IS_OPAQUE on -gdwarf-4 or -gno-strict-dwarf,
4941         but on TYPE_SIZE.  Don't do anything for ENUM_IS_OPAQUE if not creating
4942         a new die.  Don't set TREE_ASM_WRITTEN if ENUM_IS_OPAQUE.  Guard
4943         addition of most attributes on !orig_type_die or the attribute not
4944         being present already.  Assert TYPE_VALUES is NULL for ENUM_IS_OPAQUE.
4945
4946 2018-03-11  Kito Cheng  <kito.cheng@gmail.com>
4947             Chung-Ju Wu  <jasonwucj@gmail.com>
4948
4949         * config/nds32/nds32.c (nds32_cpu_cpp_builtins): Modify to define
4950         __NDS32_VH__ macro.
4951         * config/nds32/nds32.opt (mvh): New option.
4952
4953 2018-03-11  Kito Cheng  <kito.cheng@gmail.com>
4954             Chung-Ju Wu  <jasonwucj@gmail.com>
4955
4956         * config/nds32/nds32-protos.h (nds32_cpu_cpp_builtins): Declare
4957         function.
4958         * config/nds32/nds32.c (nds32_cpu_cpp_builtins): New function.
4959         * config/nds32/nds32.h (TARGET_CPU_CPP_BUILTINS): Modify its
4960         definition.
4961
4962 2018-03-11  Kito Cheng  <kito.cheng@gmail.com>
4963             Chung-Ju Wu  <jasonwucj@gmail.com>
4964
4965         * config/nds32/nds32-memory-manipulation.c (nds32_expand_strlen): New
4966         function.
4967         * config/nds32/nds32-multiple.md (strlensi): New pattern.
4968         * config/nds32/nds32-protos.h (nds32_expand_strlen): Declare function.
4969
4970 2018-03-11  Monk Chiang  <sh.chiang04@gmail.com>
4971             Kito Cheng  <kito.cheng@gmail.com>
4972             Chung-Ju Wu  <jasonwucj@gmail.com>
4973
4974         * config/nds32/constants.md (unspec_element): Add UNSPEC_FFB,
4975         UNSPEC_FFMISM and UNSPEC_FLMISM.
4976         * config/nds32/nds32-intrinsic.c (bdesc_2arg): Add builtin description
4977         for ffb, ffmism and flmism.
4978         * config/nds32/nds32-intrinsic.md (unspec_ffb): Define new pattern.
4979         (unspec_ffmism): Ditto.
4980         (unspec_flmism): Ditto.
4981         (nds32_expand_builtin_impl): Check if string extension is available.
4982         * config/nds32/nds32.h (nds32_builtins): Add NDS32_BUILTIN_FFB,
4983         NDS32_BUILTIN_FFMISM and NDS32_BUILTIN_FLMISM.
4984
4985 2018-03-10  Vladimir Makarov  <vmakarov@redhat.com>
4986
4987         Reverting patch:
4988         2018-03-09  Vladimir Makarov  <vmakarov@redhat.com>
4989
4990         PR target/83712
4991         * lra-assigns.c (assign_by_spills): Return a flag of reload
4992         assignment failure.  Do not process the reload assignment
4993         failures.  Do not spill other reload pseudos if they has the same
4994         reg class.
4995         (lra_assign): Add a return arg.  Set up from the result of
4996         assign_by_spills call.
4997         (find_reload_regno_insns, lra_split_hard_reg_for): New functions.
4998         * lra-constraints.c (split_reg): Add a new arg.  Use it instead of
4999         usage_insns if it is not NULL.
5000         (spill_hard_reg_in_range): New function.
5001         (split_if_necessary, inherit_in_ebb): Pass a new arg to split_reg.
5002         * lra-int.h (spill_hard_reg_in_range, lra_split_hard_reg_for): New
5003         function prototypes.
5004         (lra_assign): Change prototype.
5005         * lra.c (lra): Add code to deal with fails by splitting hard reg
5006         live ranges.
5007
5008 2018-03-10  H.J. Lu  <hongjiu.lu@intel.com>
5009
5010         PR target/84807
5011         * config/i386/i386.opt: Replace Enforcment with Enforcement.
5012
5013 2018-03-10  Alexandre Oliva  <aoliva@redhat.com>
5014
5015         PR debug/84620
5016         * dwarf2out.h (dw_val_class): Add dw_val_class_symview.
5017         (dw_val_node): Add val_symbolic_view.
5018         * dwarf2out.c (dw_line_info_table): Add symviews_since_reset.
5019         (symview_upper_bound): New.
5020         (new_line_info_table): Initialize symviews_since_reset.
5021         (dwarf2out_source_line): Count symviews_since_reset and set
5022         symview_upper_bound.
5023         (dw_val_equal_p): Handle symview.
5024         (add_AT_symview): New.
5025         (print_dw_val): Handle symview.
5026         (attr_checksum, attr_checksum_ordered): Likewise.
5027         (same_dw_val_p, size_of_die): Likewise.
5028         (value_format, output_die): Likewise.
5029         (add_high_low_attributes): Use add_AT_symview for entry_view.
5030         (dwarf2out_finish): Reset symview_upper_bound, clear
5031         zero_view_p.
5032
5033 2018-03-09  Peter Bergner  <bergner@vnet.ibm.com>
5034
5035         PR target/83969
5036         * config/rs6000/rs6000.c (rs6000_offsettable_memref_p): New prototype.
5037         Add strict argument and use it.
5038         (rs6000_split_multireg_move): Update for new strict argument.
5039         (mem_operand_gpr): Disallow all non-offsettable addresses.
5040         * config/rs6000/rs6000.md (*movdi_internal64): Use YZ constraint.
5041
5042 2018-03-09  Jakub Jelinek  <jakub@redhat.com>
5043
5044         PR target/84772
5045         * config/rs6000/rs6000.c (rs6000_gimplify_va_arg): Mark va_arg_tmp
5046         temporary TREE_ADDRESSABLE before gimplification of BUILT_IN_MEMCPY.
5047         * config/powerpcspe/powerpcspe.c (rs6000_gimplify_va_arg): Likewise.
5048
5049         PR c++/84767
5050         * tree-inline.c (copy_tree_body_r): For INDIRECT_REF of a remapped
5051         decl, use remap_type if we want to use the type.
5052
5053 2018-03-09  Martin Sebor  <msebor@redhat.com>
5054
5055         PR tree-optimization/84526
5056         * gimple-ssa-warn-restrict.c (builtin_memref::set_base_and_offset):
5057         Remove dead code.
5058         (builtin_access::generic_overlap): Be prepared to handle non-array
5059         base objects.
5060
5061 2018-03-09  Alexandre Oliva  <aoliva@redhat.com>
5062
5063         PR rtl-optimization/84682
5064         * lra-constraints.c (process_address_1): Check is_address flag
5065         for address constraints.
5066         (process_alt_operands): Likewise.
5067         * lra.c (lra_set_insn_recog_data): Pass asm operand locs to
5068         preprocess_constraints.
5069         * recog.h (preprocess_constraints): Add oploc parameter.
5070         Adjust callers.
5071         * recog.c (preprocess_constraints): Test address_operand for
5072         CT_ADDRESS constraints.
5073
5074 2018-03-09  Vladimir Makarov  <vmakarov@redhat.com>
5075
5076         PR target/83712
5077         * lra-assigns.c (assign_by_spills): Return a flag of reload
5078         assignment failure.  Do not process the reload assignment
5079         failures.  Do not spill other reload pseudos if they has the same
5080         reg class.
5081         (lra_assign): Add a return arg.  Set up from the result of
5082         assign_by_spills call.
5083         (find_reload_regno_insns, lra_split_hard_reg_for): New functions.
5084         * lra-constraints.c (split_reg): Add a new arg.  Use it instead of
5085         usage_insns if it is not NULL.
5086         (spill_hard_reg_in_range): New function.
5087         (split_if_necessary, inherit_in_ebb): Pass a new arg to split_reg.
5088         * lra-int.h (spill_hard_reg_in_range, lra_split_hard_reg_for): New
5089         function prototypes.
5090         (lra_assign): Change prototype.
5091         * lra.c (lra): Add code to deal with fails by splitting hard reg
5092         live ranges.
5093
5094 2018-03-09  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
5095
5096         PR target/83193
5097         * common/config/arm/arm-common.c (arm_parse_arch_option_name):
5098         Accept complain bool parameter.  Only emit errors if it is true.
5099         (arm_parse_cpu_option_name): Likewise.
5100         (arm_target_thumb_only): Adjust callers of the above.
5101         * config/arm/arm-protos.h (arm_parse_cpu_option_name): Adjust
5102         prototype to take a default true bool parameter.
5103         (arm_parse_arch_option_name): Likewise.
5104
5105 2018-03-09  David Malcolm  <dmalcolm@redhat.com>
5106             Francois-Xavier Coudert  <fxcoudert@gcc.gnu.org>
5107
5108         PR jit/64089
5109         PR jit/84288
5110         * Makefile.in (LD_VERSION_SCRIPT_OPTION, LD_SONAME_OPTION): New.
5111         * configure: Regenerate.
5112         * configure.ac ("linker --version-script option"): New.
5113         ("linker soname option"): New.
5114
5115 2018-03-09  Richard Biener  <rguenther@suse.de>
5116
5117         PR tree-optimization/84775
5118         * tree-if-conv.c (add_bb_predicate_gimplified_stmts): Delink
5119         immediate uses of predicate stmts and mark them modified.
5120
5121         Revert
5122         PR tree-optimization/84178
5123         * tree-if-conv.c (combine_blocks): Move insert_gimplified_predicates
5124         to caller.
5125         (version_loop_for_if_conversion): Delay update_ssa call.
5126         (tree_if_conversion): Delay update_ssa until after predicate
5127         insertion.
5128
5129 2018-03-09  Eric Botcazou  <ebotcazou@adacore.com>
5130
5131         PR target/84763
5132         * config/i386/winnt.c (i386_pe_seh_cold_init): Use small pre-allocation
5133         when the function accesses prior frames.
5134
5135 2018-03-08  Jakub Jelinek  <jakub@redhat.com>
5136
5137         PR debug/84456
5138         * dwarf2out.c (dw_loc_list): If list && loc_list->first->next, call
5139         gen_llsym, otherwise call maybe_gen_llsym.
5140
5141         PR inline-asm/84742
5142         * recog.c (asm_operand_ok): Return 0 if multi-character constraint
5143         has ',' character inside of it.
5144
5145 2018-03-08  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
5146
5147         PR target/84748
5148         * config/aarch64/aarch64.md (*compare_cstore<mode>_insn): Mark pattern
5149         as clobbering CC_REGNUM.
5150
5151 2018-03-08  Richard Biener  <rguenther@suse.de>
5152
5153         PR middle-end/84552
5154         * tree-scalar-evolution.c: Include tree-into-ssa.h.
5155         (follow_copies_to_constant): Do not follow SSA names registered
5156         for update.
5157
5158 2018-03-08  Richard Biener  <rguenther@suse.de>
5159
5160         PR tree-optimization/84178
5161         * tree-if-conv.c (combine_blocks): Move insert_gimplified_predicates
5162         to caller.
5163         (version_loop_for_if_conversion): Delay update_ssa call.
5164         (tree_if_conversion): Delay update_ssa until after predicate
5165         insertion.
5166
5167 2018-03-08  David Malcolm  <dmalcolm@redhat.com>
5168
5169         PR tree-optimization/84178
5170         * tree-if-conv.c (release_bb_predicate): Remove the
5171         the assertion that the stmts have NULL use_ops.
5172         Discard the statements, asserting that they haven't
5173         yet been added to a BB.
5174
5175 2018-03-08  Richard Biener  <rguenther@suse.de>
5176
5177         PR tree-optimization/84746
5178         * tree-ssa-pre.c (find_leader_in_sets): Deal with SET1 being NULL.
5179         (phi_translate): Pass in destination ANTIC_OUT set.
5180         (phi_translate_1): Likewise.  For a simplified result lookup
5181         a leader in ANTIC_OUT and AVAIL_OUT, not the ANTIC_IN sets.
5182         (phi_translate_set): Adjust.
5183         (do_pre_regular_insertion): Likewise.
5184         (do_pre_partial_partial_insertion): Likewise.
5185
5186 2018-03-08  Martin Liska  <mliska@suse.cz>
5187
5188         PR gcov-profile/84735
5189         * doc/gcov.texi: Document usage of profile files.
5190         * gcov-io.h: Document changes in the format.
5191
5192 2018-03-08  Alexandre Oliva  <aoliva@redhat.com>
5193
5194         PR debug/84404
5195         PR debug/84408
5196         * dwarf2out.c (struct dw_line_info_table): Update comments for
5197         view == -1.
5198         (FORCE_RESET_NEXT_VIEW): New.
5199         (FORCE_RESETTING_VIEW_P): New.
5200         (RESETTING_VIEW_P): Check for -1 too.
5201         (ZERO_VIEW_P): Likewise.
5202         (new_line_info_table): Force-reset next view.
5203         (dwarf2out_begin_function): Likewise.
5204         (dwarf2out_source_line): Simplify zero_view_p initialization.
5205         Test FORCE_RESETTING_VIEW_P and RESETTING_VIEW_P instead of
5206         view directly.  Omit view when omitting .loc at line 0.
5207
5208 2018-03-08  Jakub Jelinek  <jakub@redhat.com>
5209
5210         PR tree-optimization/84740
5211         * tree-switch-conversion.c (process_switch): Call build_constructors
5212         only if info.phi_count is non-zero.
5213
5214         PR tree-optimization/84739
5215         * tree-tailcall.c (find_tail_calls): Check call arguments against
5216         DECL_ARGUMENTS (current_function_decl) rather than
5217         DECL_ARGUMENTS (func) when checking for tail recursion.
5218
5219 2018-03-07  Jakub Jelinek  <jakub@redhat.com>
5220
5221         * doc/contrib.texi: Add entries for Martin Liska, David Malcolm,
5222         Marek Polacek, extend Vladimir Makarov's, Jonathan Wakely's and
5223         Volker Reichelt's entry and add entries for people that perform
5224         GCC fuzzy testing and report numerous bugs.
5225
5226 2018-03-07  Segher Boessenkool  <segher@kernel.crashing.org>
5227
5228         PR target/82411
5229         * config/rs6000/rs6000.c (rs6000_elf_in_small_data_p): Don't put
5230         readonly data in sdata, if that is disabled.
5231         * config/rs6000/sysv4.opt (mreadonly-in-sdata): New option.
5232         * doc/invoke.texi (RS/6000 and PowerPC Options): Document
5233         -mreadonly-in-sdata option.
5234
5235 2018-03-07  Martin Sebor  <msebor@redhat.com>
5236
5237         PR tree-optimization/84468
5238         * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Consider successor
5239         basic block when looking for nul assignment.
5240
5241 2018-03-07  Eric Botcazou  <ebotcazou@adacore.com>
5242
5243         PR target/84277
5244         * except.h (output_function_exception_table): Adjust prototype.
5245         * except.c (output_function_exception_table): Remove FNNAME parameter
5246         and add SECTION parameter.  Ouput one part of the table at a time.
5247         * final.c (final_scan_insn_1) <NOTE_INSN_SWITCH_TEXT_SECTIONS>: Output
5248         the first part of the exception table and emit unwind directives.
5249         * config/i386/i386-protos.h (i386_pe_end_cold_function): Declare.
5250         (i386_pe_seh_cold_init): Likewise.
5251         * config/i386/cygming.h (ASM_DECLARE_COLD_FUNCTION_NAME): New macro.
5252         (ASM_DECLARE_COLD_FUNCTION_SIZE): Likewise.
5253         * config/i386/i386.c (x86_expand_epilogue): Fix wording in comment.
5254         (ix86_output_call_insn): Emit a nop in one more case for SEH.
5255         * config/i386/winnt.c: Include except.h.
5256         (struct seh_frame_state): Add reg_offset, after_prologue and
5257         in_cold_section fields.
5258         (i386_pe_seh_end_prologue): Set seh->after_prologue.
5259         (i386_pe_seh_cold_init): New function.
5260         (i386_pe_seh_fini): Add COLD parameter and bail out if it is not equal
5261         to seh->in_cold_section.
5262         (seh_emit_push): Record the offset of the push.
5263         (seh_emit_save): Record the offet of the save.
5264         (i386_pe_seh_unwind_emit): Deal with NOTE_INSN_SWITCH_TEXT_SECTIONS.
5265         Test seh->after_prologue to disregard the epilogue.
5266         (i386_pe_end_function): Pass FALSE to i386_pe_seh_fini.
5267         (i386_pe_end_cold_function): New function.
5268
5269 2018-03-07  Jakub Jelinek  <jakub@redhat.com>
5270
5271         PR fortran/84565
5272         * config/aarch64/predicates.md (aarch64_simd_reg_or_zero): Use
5273         aarch64_simd_or_scalar_imm_zero rather than aarch64_simd_imm_zero.
5274
5275         PR c++/84704
5276         * gimple-expr.c (create_tmp_var_raw): Set DECL_NAMELESS flag
5277         on tmp_var.
5278         * tree-pretty-print.c (dump_decl_name): For TDF_COMPARE_DEBUG,
5279         don't print names of DECL_NAMELESS DECL_IGNORED_P decls.
5280
5281         PR middle-end/84723
5282         * multiple_target.c: Include tree-inline.h and intl.h.
5283         (expand_target_clones): Diagnose and fail if node->definition and
5284         !tree_versionable_function_p (node->decl).
5285
5286 2018-03-06  John David Anglin  <danglin@gcc.gnu.org>
5287
5288         * config/pa/pa.h (ASM_GENERATE_INTERNAL_LABEL): Revise to use
5289         sprint_ul.
5290         (ASM_OUTPUT_ADDR_VEC_ELT): Revise for above change.
5291         (ASM_OUTPUT_ADDR_DIFF_ELT): Likewise.
5292         * config/pa/pa64-hpux.h (ASM_GENERATE_INTERNAL_LABEL): Revise as above.
5293
5294 2018-03-06  Jakub Jelinek  <jakub@redhat.com>
5295
5296         PR target/84710
5297         * combine.c (try_combine): Use reg_or_subregno instead of handling
5298         just paradoxical SUBREGs and REGs.
5299
5300 2018-03-06  Claudiu Zissulescu  <claziss@synopsys.com>
5301
5302          * config/arc/arc.c (arc_finalize_pic): Remove function.
5303          (arc_must_save_register): We use single base PIC register, remove
5304          checks to save/restore the PIC register.
5305          (arc_expand_prologue): Likewise.
5306          * config/arc/arc-protos.h (arc_set_default_type_attributes):
5307          Remove.
5308          (arc_verify_short): Likewise.
5309          (arc_attr_type): Likewise.
5310          * config/arc/arc.c (arc_set_default_type_attributes): Remove.
5311          (walk_stores): Likewise.
5312          (arc_address_cost): Make it static.
5313          (arc_verify_short): Likewise.
5314          (branch_dest): Likewise.
5315          (arc_attr_type): Likewise.
5316          * config/arc/arc.c (TARGET_ADJUST_INSN_LENGTH): Remove.
5317          (TARGET_INSN_LENGTH_PARAMETERS): Likewise.
5318          (arc_final_prescan_insn): Remove inserting the nops due to
5319          hardware hazards.  It is done in reorg step.
5320          (insn_length_variant_t): Remove.
5321          (insn_length_parameters_t): Likewise.
5322          (arc_insn_length_parameters): Likewise.
5323          (arc_get_insn_variants): Likewise.
5324          * config/arc/arc.h (TARGET_UPSIZE_DBR): Remove.
5325
5326 2018-03-06  Jakub Jelinek  <jakub@redhat.com>
5327
5328         PR inline-asm/84683
5329         * reg-stack.c (move_for_stack_reg): If any_malformed_asm, avoid
5330         assertion failure.
5331
5332         PR tree-optimization/84687
5333         * omp-simd-clone.c (simd_clone_create): Clear DECL_BUILT_IN_CLASS
5334         on new_node->decl.
5335         * match.pd (pow(C,x)*expN(y) -> expN(logN(C)*x+y)): New optimization.
5336
5337 2018-03-05  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
5338
5339         * config/rs6000/rs6000-builtin.def (rs6000_speculation_barrier):
5340         Rename to ppc_speculation_barrier.
5341         * config/rs6000/rs6000.c (rs6000_init_builtins): Rename builtin to
5342         __builtin_ppc_speculation_barrier.
5343
5344 2018-03-05  Jakub Jelinek  <jakub@redhat.com>
5345
5346         PR target/84700
5347         * combine.c (combine_simplify_rtx): Don't try to simplify if
5348         if_then_else_cond returned non-NULL, but either true_rtx or false_rtx
5349         are equal to x.
5350
5351 2018-03-05  Segher Boessenkool  <segher@kernel.crashing.org>
5352
5353         * config/rs6000/rs6000.c (rs6000_loop_align): Don't align tiny loops
5354         to 32 bytes when compiling for POWER9.
5355
5356 2018-03-05  Jakub Jelinek  <jakub@redhat.com>
5357
5358         PR target/84564
5359         * config/i386/i386.c (ix86_function_ok_for_sibcall): Check for
5360         regparm >= 3 with no arg reg available also for calls with
5361         flag_force_indirect_call.  Pass decl to ix86_function_regparm.
5362
5363         PR target/84524
5364         * config/i386/sse.md (*<code><mode>3): Replace <mask_prefix3> with
5365         orig,vex.
5366         (*<plusminus_insn><mode>3): Likewise.  Remove <mask_operand3> uses.
5367
5368 2018-03-05  Peter Bergner  <bergner@vnet.ibm.com>
5369
5370         PR target/84264
5371         * config/rs6000/vector.md (mov<mode>): Disallow altivec memory operands.
5372
5373 2018-03-05  Richard Biener  <rguenther@suse.de>
5374
5375         PR tree-optimization/84486
5376         * tree-ssa-pre.c (create_expression_by_pieces): Remove dead code.
5377         When inserting a __builtin_assume_aligned call set the LHS
5378         SSA name alignment info accordingly.
5379
5380 2018-03-05  Wilco Dijkstra  <wdijkstr@arm.com>
5381
5382         PR tree-optimization/84114
5383         * config/aarch64/aarch64.c (aarch64_reassociation_width)
5384         Avoid reassociation of FLOAT_MODE addition.
5385
5386 2018-03-05  Olga Makhotina  <olga.makhotina@intel.com>
5387
5388         * common/config/i386/i386-common.c (OPTION_MASK_ISA_PCONFIG_SET,
5389         OPTION_MASK_ISA_PCONFIG_UNSET, OPTION_MASK_ISA_WBNOINVD_SET,
5390         OPTION_MASK_ISA_WBNOINVD_UNSET): New definitions.
5391         (ix86_handle_option): Handle -mpconfig and -mwbnoinvd.
5392         * config.gcc (pconfigintrin.h, wbnoinvdintrin.h) : Add headers.
5393         * config/i386/cpuid.h (bit_PCONFIG, bit_WBNOINVD): New.
5394         * config/i386/driver-i386.c (host_detect_local_cpu): Detect -mpconfig
5395         and -mwbnoinvd.
5396         * config/i386/i386-builtin.def (__builtin_ia32_wbnoinvd,
5397         __builtin_ia32_wbinvd): New builtins.
5398         (SPECIAL_ARGS2): New.
5399         * config/i386/i386-c.c (__WBNOINVD__, __PCONFIG__): New.
5400         (SPECIAL_ARGS2): New.
5401         * config/i386/i386.c (ix86_target_string): Add -mpconfig and -mwbnoinvd.
5402         (ix86_valid_target_attribute_inner_p): Ditto.
5403         (ix86_init_mmx_sse_builtins): Add special_args2.
5404         * config/i386/i386.h (TARGET_PCONFIG, TARGET_PCONFIG_P, TARGET_WBNOINVD,
5405         TARGET_WBNOINVD_P): New.
5406         * config/i386/i386.md (UNSPECV_WBINVD, UNSPECV_WBNOINVD): New.
5407         (define_insn "wbinvd", define_insn "wbnoinvd"): New.
5408         * config/i386/i386.opt: Add -mpconfig and -mwbnoinvd.
5409         * config/i386/immintrin.h (_wbinvd): New intrinsic.
5410         * config/i386/pconfigintrin.h: New file.
5411         * config/i386/wbnoinvdintrin.h: Ditto.
5412         * config/i386/x86intrin.h: Add headers pconfigintrin.h and wbnoinvdintrin.h.
5413         * doc/invoke.texi (-mpconfig, -mwbnoinvd): New.
5414
5415 2018-03-05  Richard Biener  <rguenther@suse.de>
5416
5417         PR tree-optimization/84670
5418         * tree-ssa-pre.c (struct bb_bitmap_sets): Add visited_with_visited_succs
5419         member.
5420         (BB_VISITED_WITH_VISITED_SUCCS): New define.
5421         (compute_antic): Initialize BB_VISITED_WITH_VISITED_SUCCS.
5422         (compute_antic_aux): Only assert the number of values in ANTIC_IN
5423         doesn't grow if all successors (recursively) were visited at least
5424         once.
5425
5426 2018-03-05  Richard Biener  <rguenther@suse.de>
5427
5428         PR tree-optimization/84650
5429         * tree-ssa-loop-im.c (pass_lim::execute): Reset the SCEV cache
5430         if executed in the loop pipeline.
5431
5432 2018-03-05  Sandra Loosemore  <sandra@codesourcery.com>
5433
5434         * doc/configfiles.texi (Configuration Files): Move info about
5435         conditionalizing $target-protos.h to...
5436         * doc/sourcebuild.texi (Back End): Here.  Explain how $target.h
5437         differs from $target-protos.h.
5438
5439 2018-03-05  Kito Cheng  <kito.cheng@gmail.com>
5440             Chung-Ju Wu  <jasonwucj@gmail.com>
5441
5442         * config/nds32/nds32-protos.h (nds32_expand_setmem): Declare.
5443         * config/nds32/nds32-multiple.md (setmemsi): Define.
5444         * config/nds32/nds32-memory-manipulation.c
5445         (nds32_gen_dup_4_byte_to_word_value): New.
5446         (emit_setmem_word_loop): New.
5447         (emit_setmem_byte_loop): New.
5448         (nds32_expand_setmem_loop): New.
5449         (nds32_expand_setmem_loop_v3m): New.
5450         (nds32_expand_setmem_unroll): New.
5451         (nds32_expand_setmem): New.
5452
5453 2018-03-04  Kito Cheng  <kito.cheng@gmail.com>
5454             Chung-Ju Wu  <jasonwucj@gmail.com>
5455
5456         * config/nds32/nds32-memory-manipulation.c
5457         (nds32_emit_load_store): New.
5458         (nds32_emit_post_inc_load_store): New.
5459         (nds32_emit_mem_move): New.
5460         (nds32_emit_mem_move_block): New.
5461         (nds32_expand_movmemsi_loop_unknown_size): New.
5462         (nds32_expand_movmemsi_loop_known_size): New.
5463         (nds32_expand_movmemsi_loop): New.
5464         (nds32_expand_movmemsi_unroll): New.
5465         (nds32_expand_movmemqi): Rename ...
5466         (nds32_expand_movmemsi): ... to this.
5467         * config/nds32/nds32-multiple.md (movmemqi): Rename ...
5468         (movmemsi): ... to this.
5469         * config/nds32/nds32-protos.h (nds32_expand_movmemqi): Rename ...
5470         (nds32_expand_movmemsi): ... to this.
5471
5472 2018-03-04  Kito Cheng  <kito.cheng@gmail.com>
5473             Monk Chiang  <sh.chiang04@gmail.com>
5474             Chung-Ju Wu  <jasonwucj@gmail.com>
5475
5476         * config/nds32/nds32-protos.h
5477         (nds32_expand_load_multiple): New arguments.
5478         (nds32_expand_store_multiple): Ditto.
5479         (nds32_valid_multiple_load_store): Rename ...
5480         (nds32_valid_multiple_load_store_p): ... to this.
5481         * config/nds32/nds32-memory-manipulation.c
5482         (nds32_expand_load_multiple): Refine implementation.
5483         (nds32_expand_store_multiple): Ditto.
5484         * config/nds32/nds32-multiple.md
5485         (load_multiple): Update nds32_expand_load_multiple interface.
5486         (store_multiple): Update nds32_expand_store_multiple interface.
5487         * config/nds32/nds32-predicates.c
5488         (nds32_valid_multiple_load_store): Rename ...
5489         (nds32_valid_multiple_load_store_p): ... to this and refine
5490         implementation.
5491         * config/nds32/predicates.md
5492         (nds32_load_multiple_and_update_address_operation): New predicate.
5493         (nds32_store_multiple_and_update_address_operation): New predicate.
5494
5495 2018-03-04  Kito Cheng  <kito.cheng@gmail.com>
5496             Chung-Ju Wu  <jasonwucj@gmail.com>
5497
5498         * config/nds32/nds32.md (type): Add load_multiple and store_multiple.
5499         (combo): New attribute.
5500         * config/nds32/nds32-multiple.md: Refine patterns with new attributes.
5501
5502 2018-03-03  Chung-Ju Wu  <jasonwucj@gmail.com>
5503
5504         * config/nds32/nds32.opt: Change -mcmodel= default value.
5505
5506 2018-03-03  Kito Cheng  <kito.cheng@gmail.com>
5507             Monk Chiang  <sh.chiang04@gmail.com>
5508             Chung-Ju Wu  <jasonwucj@gmail.com>
5509
5510         * config/nds32/constants.md (unspec_element): New enum.
5511         * config/nds32/constraints.md (Umw): New constraint.
5512         * config/nds32/nds32-intrinsic.c: Add more builtin functions.
5513         * config/nds32/nds32-intrinsic.md: Likewise.
5514         * config/nds32/nds32-md-auxiliary.c (nds32_regno_to_enable4): New.
5515         (nds32_valid_smw_lwm_base_p): New.
5516         (nds32_output_smw_single_word): New.
5517         (nds32_output_lmw_single_word): New.
5518         (nds32_expand_unaligned_load): New.
5519         (nds32_expand_unaligned_store): New.
5520         * config/nds32/nds32-protos.h (nds32_valid_smw_lwm_base_p): Declare.
5521         (nds32_output_smw_single_word): Declare.
5522         (nds32_output_lmw_single_word): Declare.
5523         (nds32_expand_unaligned_load): Declare.
5524         (nds32_expand_unaligned_store): Declare.
5525         * config/nds32/nds32.h (nds32_builtins): Add NDS32_BUILTIN_UALOAD_HW,
5526         NDS32_BUILTIN_UALOAD_W, NDS32_BUILTIN_UALOAD_DW,
5527         NDS32_BUILTIN_UASTORE_HW, NDS32_BUILTIN_UASTORE_W,
5528         NDS32_BUILTIN_UASTORE_DW.
5529         * config/nds32/predicates.md (nds32_lmw_smw_base_operand): New
5530         predicate.
5531
5532 2018-03-03  Monk Chiang  <sh.chiang04@gmail.com>
5533             Kito Cheng  <kito.cheng@gmail.com>
5534             Chung-Ju Wu  <jasonwucj@gmail.com>
5535
5536         * config/nds32/nds32-intrinsic.c
5537         (nds32_expand_builtin_null_ftype_reg): Delete.
5538         (nds32_expand_builtin_reg_ftype_imm): Ditto.
5539         (nds32_expand_builtin_null_ftype_reg_imm): Ditto.
5540         (nds32_read_argument): New.
5541         (nds32_legitimize_target): Ditto.
5542         (nds32_legitimize_argument): Ditto.
5543         (nds32_check_constant_argument): Ditto.
5544         (nds32_expand_unop_builtin): Ditto.
5545         (nds32_expand_unopimm_builtin): Ditto.
5546         (nds32_expand_binop_builtin): Ditto.
5547         (nds32_builtin_decl_impl): Ditto.
5548         (builtin_description): Ditto.
5549         (nds32_expand_builtin_impl): Rewrite with new infrastructure.
5550         (nds32_init_builtins_impl): Ditto.
5551         * config/nds32/nds32.c (TARGET_BUILTIN_DECL): Define.
5552         (nds32_builtin_decl): New.
5553         * config/nds32/nds32.h (nds32_builtins): Add NDS32_BUILTIN_COUNT.
5554         * config/nds32/nds32-protos.h (nds32_builtin_decl_impl): Declare.
5555
5556 2018-03-02  Jeff Law  <law@redhat.com>
5557
5558         * reorg.c (stop_search_p): Handle DEBUG_INSN.
5559         (redundant_insn, fill_simple_delay_slots): Likewise.
5560         (fill_slots_from_thread): Likewise.
5561         * resource.c (mark_referenced_resources): Likewise.
5562         (mark_set_resources, find_dead_or_set_registers): Likewise.
5563
5564 2018-03-02  Jakub Jelinek  <jakub@redhat.com>
5565
5566         * substring-locations.h (format_warning_va): Formatting fix for
5567         ATTRIBUTE_GCC_DIAG.
5568         (format_warning_at_substring): Fix up ATTRIBUTE_GCC_DIAG second
5569         argument.
5570         (format_warning_n_va, format_warning_at_substring_n): New prototypes.
5571         * substring-locations.c: Include intl.h.
5572         (format_warning_va): Turned into small wrapper around
5573         format_warning_n_va, renamed to ...
5574         (format_warning_n_va): ... this, add N and PLURAL_GMSGID arguments,
5575         rename GMSGID to SINGULAR_GMSGID, if SINGULAR_GMSGID != PLURAL_GMSGID,
5576         use ngettext.
5577         (format_warning_at_substring_n): New function.
5578         * gimple-ssa-sprintf.c: Remove GCC diagnostic ignored pragma.
5579         (fmtwarn): Add ATTRIBUTE_GCC_DIAG.  Turn into a copy of
5580         format_warning_at_substring with just a shorter name instead of
5581         const function pointer.
5582         (fmtwarn_n): New function.
5583         (maybe_warn, format_directive, parse_directive): Use fmtwarn_n where
5584         appropriate, get rid of all the fmtstr temporaries, move conditionals
5585         with G_() wrapped string literals directly into fmtwarn arguments,
5586         cast dir.len to (int), formatting fixes.
5587
5588 2018-03-02  Thomas Schwinge  <thomas@codesourcery.com>
5589
5590         * doc/invoke.texi: Remove "Cilk Plus" references.
5591
5592 2018-03-02  Jakub Jelinek  <jakub@redhat.com>
5593             Richard Biener  <rguenther@suse.de>
5594
5595         PR ipa/84628
5596         * expr.c (expand_expr_real_1) <case CALL_EXPR>: Don't emit diagnostics
5597         for error or warning attributes if CALL_FROM_THUNK_P is set.
5598         Formatting fixes.
5599
5600 2018-03-02  Jakub Jelinek  <jakub@redhat.com>
5601
5602         PR target/56540
5603         * config/pa/pa.h (TARGET_CPU_CPP_BUILTINS): Predefine
5604         __SIZEOF_128__ macro if HPUX_LONG_DOUBLE_LIBRARY.
5605
5606         PR target/56540
5607         * config/ia64/ia64.h (TARGET_CPU_CPP_BUILTINS): Predefine
5608         __SIZEOF_{FPREG,FLOAT{80,128}}__ macros.
5609
5610         * predict.c (test_prediction_value_range): Use PROB_UNINITIALIZED
5611         instead of -1U in last predictors element's probability member.
5612
5613 2018-03-02  Eric Botcazou  <ebotcazou@adacore.com>
5614
5615         PR ipa/83983
5616         * ipa-devirt.c (odr_subtypes_equivalent_p): Get the ODR type of both
5617         arguments if they are comparable.
5618
5619 2018-03-02  Richard Sandiford  <richard.sandiford@linaro.org>
5620
5621         PR tree-optimization/84634
5622         * tree-vect-stmts.c (vectorizable_store, vectorizable_load): Replace
5623         masks and masked_loop_p with a single loop_masks, making sure it's
5624         null for bb vectorization.
5625
5626 2018-03-02  Richard Sandiford  <richard.sandiford@linaro.org>
5627
5628         * tree-vect-data-refs.c (vect_analyze_data_ref_dependence)
5629         (vect_analyze_data_ref_access): Use loop->safe_len rather than
5630         loop->force_vectorize to check whether there is no alias.
5631
5632 2018-03-02  Jakub Jelinek  <jakub@redhat.com>
5633
5634         PR target/84614
5635         * rtl.h (prev_real_nondebug_insn, next_real_nondebug_insn): New
5636         prototypes.
5637         * emit-rtl.c (next_real_insn, prev_real_insn): Fix up function
5638         comments.
5639         (next_real_nondebug_insn, prev_real_nondebug_insn): New functions.
5640         * cfgcleanup.c (try_head_merge_bb): Use prev_real_nondebug_insn
5641         instead of a loop around prev_real_insn.
5642         * combine.c (move_deaths): Use prev_real_nondebug_insn instead of
5643         prev_real_insn.
5644
5645         PR inline-asm/84625
5646         * config/i386/i386.c (ix86_print_operand): Use conditional
5647         output_operand_lossage instead of gcc_assert if CONST_VECTOR is not
5648         zero vector.
5649
5650 2018-03-02  Richard Biener  <rguenther@suse.de>
5651
5652         PR tree-optimization/84427
5653         * tree-ssa-pre.c (bitmap_remove_expr_from_set): Remove.
5654         (bitmap_set_subtract_values): Rewrite to handle multiple
5655         exprs per value.
5656         (clean): Likewise.
5657         (prune_clobbered_mems): Likewise.
5658         (phi_translate): Take edge instead of pred/phiblock.
5659         (phi_translate_1): Likewise.
5660         (phi_translate_set): Likewise.  Insert all translated
5661         exprs for a value into the set, keeping possibly multiple
5662         expressions per value.
5663         (compute_antic_aux): Adjust for phi_translate changes.
5664         When intersecting union the expressions and prune those
5665         not in the final value set, keeping possibly multiple
5666         expressions per value.  Do not use value-insertion
5667         for unioning ANTIC_OUT U EXP_GEN - TMP_GEN but merge
5668         all expressions.  Add verification that the value-sets
5669         only shrink during iteration.
5670         (compute_partial_antic_aux): Adjust for the phi_translate changes.
5671         (do_pre_regular_insertion): Likewise.
5672         (do_pre_partial_partial_insertion): Likewise.
5673
5674 2018-03-02  Richard Biener  <rguenther@suse.de>
5675
5676         PR target/82005
5677         * config/darwin.c (saved_debug_info_level): New static global.
5678         (darwin_asm_lto_start): Disable debug info generation for LTO out.
5679         (darwin_asm_lto_end): Restore debug info generation settings.
5680
5681 2018-03-01  Martin Liska  <mliska@suse.cz>
5682
5683         PR sanitizer/82484
5684         * sanopt.c (sanitize_rewrite_addressable_params): Do not handle
5685         volatile arguments.
5686
5687 2018-03-01  Richard Biener  <rguenther@suse.de>
5688
5689         PR debug/84645
5690         * dwarf2out.c (gen_variable_die): Properly handle late VLA
5691         type annotation with LTO when debug was disabled at compile-time.
5692
5693 2018-03-01  Matthew Fortune  <mfortune@gmail.com>
5694
5695         * config/mips/mips.c (mips_final_prescan_insn): Fix incorrect
5696         XINT with INTVAL.
5697         (mips_final_postscan_insn): Likewise.
5698
5699 2018-03-01  Richard Sandiford  <richard.sandiford@linaro.org>
5700
5701         PR rtl-optimization/84528
5702         * alias.c (init_alias_target): Add commentary.
5703         (init_alias_analysis): Only give HARD_FRAME_POINTER_REGNUM
5704         a unique base value if the frame pointer is not eliminated
5705         to the stack pointer.
5706
5707 2018-03-01  Tom de Vries  <tom@codesourcery.com>
5708
5709         PR rtl-optimization/83327
5710         * lra-int.h (hard_regs_spilled_into): Declare.
5711         * lra.c (hard_regs_spilled_into): Define.
5712         (init_reg_info): Init hard_regs_spilled_into.
5713         * lra-spills.c (assign_spill_hard_regs): Update hard_regs_spilled_into.
5714         * lra-lives.c (make_hard_regno_born, make_hard_regno_dead)
5715         (process_bb_lives): Handle hard_regs_spilled_into.
5716         (lra_create_live_ranges_1): Before doing liveness propagation, clear
5717         regs in all_hard_regs_bitmap if set in hard_regs_spilled_into.
5718
5719 2018-02-28  David Edelsohn  <dje.gcc@gmail.com>
5720
5721         * config.gcc (powerpc-ibm-aix7.1.*): New stanza.
5722         (powerpc-ibm-aix[789]*): Default to AIX 7.2.
5723         * config/rs6000/aix71.h (TARGET_DEFAULT): Revert to Power4 ISA.
5724         * config/rs6000/aix72.h: New file.
5725
5726 2018-02-28  Jakub Jelinek  <jakub@redhat.com>
5727
5728         * gimple-ssa-warn-restrict.c (maybe_diag_overlap): Use warning_n
5729         instead of warning_at with conditional singular and plural messages
5730         where possible.
5731
5732         PR target/52991
5733         * stor-layout.c (update_alignment_for_field): For
5734         targetm.ms_bitfield_layout_p (rli->t), if !is_bitfield
5735         && !DECL_PACKED (field), do the alignment update, just use
5736         only desired_align instead of MAX (type_align, desired_align)
5737         as the alignment.
5738         (place_field): Don't do known_align < desired_align handling
5739         early if targetm.ms_bitfield_layout_p (rli->t) and rli->prev_field
5740         is non-NULL, instead do it after rli->prev_field handling and
5741         only if not within a bitfield word.  For DECL_PACKED (field)
5742         use type_align of BITS_PER_UNIT.
5743
5744 2018-02-28  Eric Botcazou  <ebotcazou@adacore.com>
5745
5746         * config/aarch64/aarch64.c (aarch64_emit_probe_stack_range): Remove
5747         superfluous parentheses and trailing spaces.
5748
5749 2018-02-28  Richard Biener  <rguenther@suse.de>
5750
5751         PR tree-optimization/84584
5752         * graphite-scop-detection.c (scop_detection::add_scop): Discard
5753         SCoPs with fake exit edge.
5754
5755 2018-02-28  Martin Liska  <mliska@suse.cz>
5756
5757         PR testsuite/84597
5758         * timevar.c (timer::print): Fix format to properly print 100%
5759         values.
5760
5761 2018-02-28  Richard Biener  <rguenther@suse.de>
5762
5763         PR middle-end/84607
5764         * genmatch.c (capture_info::walk_match): Do not mark
5765         captured expressions without operands as expr_p given
5766         they act more like predicates and should be subject to
5767         "lost tail" side-effect preserving.
5768
5769 2018-02-28  Alexandre Oliva  <aoliva@redhat.com>
5770
5771         PR rtl-optimization/81611
5772         * auto-inc-dec.c (attempt_change): Move dead note from
5773         mem_insn if it's the next use of regno
5774         (find_address): Take address use of reg holding
5775         non-incremented value.  Add parm to limit search to the named
5776         reg only.
5777         (merge_in_block): Attempt to use a mem insn that is the next
5778         use of the original regno.
5779
5780 2018-02-27  Martin Sebor  <msebor@redhat.com>
5781
5782         PR c++/83871
5783         * gcc/doc/invoke.texi (-Wmissing-attributes): New option.
5784         * gcc/print-tree.c (print_node): Handle DECL_UNINLINABLE.
5785
5786 2018-02-27  Martin Sebor  <msebor@redhat.com>
5787
5788         PR translation/84207
5789         * diagnostic-core.h (warning_n, error_n, inform_n): Change
5790         n argument to unsigned HOST_WIDE_INT.
5791         * diagnostic.c (warning_n, error_n, inform_n): Ditto.
5792         (diagnostic_n_impl): Ditto.  Handle arguments in excess of LONG_MAX.
5793         * gimple-ssa-sprintf.c (format_directive): Simplify inform_n call.
5794         * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Use warning_n.
5795
5796 2018-02-27  Richard Biener  <rguenther@suse.de>
5797
5798         PR tree-optimization/84512
5799         * tree-vect-loop.c (vect_compute_single_scalar_iteration_cost):
5800         Do not use the estimate returned from record_stmt_cost for
5801         the scalar iteration cost but sum properly using add_stmt_cost.
5802
5803 2018-02-27  Richard Biener  <rguenther@suse.de>
5804
5805         PR tree-optimization/84466
5806         * graphite-scop-detection.c (scop_detection::stmt_simple_for_scop_p):
5807         Adjust last change to less strictly validate use operands.
5808
5809 2018-02-27  Martin Liska  <mliska@suse.cz>
5810
5811         PR gcov-profile/84548
5812         * gcov.c (process_file): Allow partial overlap and consider it
5813         also as group functions.
5814         (output_lines): Properly calculate range of lines for a group.
5815
5816 2018-02-27  Martin Liska  <mliska@suse.cz>
5817
5818         * timevar.c (timer::print_row): Remove 'usr', 'sys', 'wall' and
5819         'ggc' suffixes.  Change first column width.
5820         (timer::print): Fix formatting of the column.
5821
5822 2018-02-27  Alexandre Oliva  <aoliva@redhat.com>
5823
5824         * tree-ssa-live.c (remove_unused_scope_block_p): Do not
5825         preserve inline entry blocks for the sake of debug inline
5826         entry point markers alone.
5827         (remove_unused_locals): Suggest in comments a better place to
5828         force the preservation of inline entry blocks that are
5829         otherwise unused, but do not preserve them.
5830
5831 2018-02-26  H.J. Lu  <hongjiu.lu@intel.com>
5832
5833         * config/i386/i386.c (ix86_output_indirect_jmp): Update comments.
5834
5835 2018-02-26  H.J. Lu  <hongjiu.lu@intel.com>
5836
5837         PR target/84039
5838         * config/i386/constraints.md (Bs): Replace
5839         ix86_indirect_branch_register with
5840         TARGET_INDIRECT_BRANCH_REGISTER.
5841         (Bw): Likewise.
5842         * config/i386/i386.md (indirect_jump): Likewise.
5843         (tablejump): Likewise.
5844         (*sibcall_memory): Likewise.
5845         (*sibcall_value_memory): Likewise.
5846         Peepholes of indirect call and jump via memory: Likewise.
5847         (*sibcall_GOT_32): Disallowed for TARGET_INDIRECT_BRANCH_REGISTER.
5848         (*sibcall_value_GOT_32): Likewise.
5849         * config/i386/predicates.md (indirect_branch_operand): Likewise.
5850         (GOT_memory_operand): Likewise.
5851         (call_insn_operand): Likewise.
5852         (sibcall_insn_operand): Likewise.
5853         (GOT32_symbol_operand): Likewise.
5854         * config/i386/i386.h (TARGET_INDIRECT_BRANCH_REGISTER): New.
5855
5856 2018-02-26  Eric Botcazou  <ebotcazou@adacore.com>
5857
5858         PR rtl-optimization/83496
5859         * reorg.c (steal_delay_list_from_target): Change REDUNDANT array from
5860         booleans to RTXes.  Call fix_reg_dead_note on every non-null element.
5861         (steal_delay_list_from_fallthrough): Call fix_reg_dead_note on a
5862         redundant insn, if any.
5863         (relax_delay_slots): Likewise.
5864         (update_reg_unused_notes): Rename REDUNDANT_INSN to OTHER_INSN.
5865
5866 2018-02-26  Richard Sandiford  <richard.sandiford@linaro.org>
5867
5868         PR tree-optimization/83965
5869         * tree-vect-patterns.c (vect_reassociating_reduction_p): Assume
5870         that grouped statements are part of a reduction chain.  Return
5871         true if the statement is not marked as a reduction itself but
5872         is part of a group.
5873         (vect_recog_dot_prod_pattern): Don't check whether the statement
5874         is part of a group here.
5875         (vect_recog_sad_pattern): Likewise.
5876         (vect_recog_widen_sum_pattern): Likewise.
5877
5878 2018-02-26  Eric Botcazou  <ebotcazou@adacore.com>
5879
5880         PR debug/84545
5881         * final.c (rest_of_clean_state): Also look for calls inside sequences.
5882
5883 2018-02-26  H.J. Lu  <hongjiu.lu@intel.com>
5884
5885         PR target/84530
5886         * config/i386/i386-protos.h (ix86_output_indirect_jmp): Remove
5887         the bool argument.
5888         (ix86_output_indirect_function_return): New prototype.
5889         (ix86_split_simple_return_pop_internal): Likewise.
5890         * config/i386/i386.c (indirect_return_via_cx): New.
5891         (indirect_return_via_cx_bnd): Likewise.
5892         (indirect_thunk_name): Handle return va CX_REG.
5893         (output_indirect_thunk_function): Create alias for
5894         __x86_return_thunk_[re]cx and __x86_return_thunk_[re]cx_bnd.
5895         (ix86_output_indirect_jmp): Remove the bool argument.
5896         (ix86_output_indirect_function_return): New function.
5897         (ix86_split_simple_return_pop_internal): Likewise.
5898         * config/i386/i386.md (*indirect_jump): Don't pass false
5899         to ix86_output_indirect_jmp.
5900         (*tablejump_1): Likewise.
5901         (simple_return_pop_internal): Change it to define_insn_and_split.
5902         Call ix86_split_simple_return_pop_internal to split it for
5903         -mfunction-return=.
5904         (simple_return_indirect_internal): Call
5905         ix86_output_indirect_function_return instead of
5906         ix86_output_indirect_jmp.
5907
5908 2018-02-26  Jakub Jelinek  <jakub@redhat.com>
5909
5910         PR bootstrap/84405
5911         * vec.h (vec_default_construct): For BROKEN_VALUE_INITIALIZATION use
5912         memset and value initialization afterwards.
5913
5914 2018-02-26  Christophe Lyon  <christophe.lyon@linaro.org>
5915
5916         * Makefile.in (lto-wrapper): Use ALL_LINKERFLAGS.
5917
5918 2018-02-26  Ramana Radhakrishnan  <ramana.radhakrishnan@arm.com>
5919
5920         PR target/84521
5921         * common/config/aarch64/aarch64-common.c
5922         (aarch_option_optimization_table[]): Switch
5923         off fomit-frame-pointer
5924
5925 2018-02-26  Kito Cheng  <kito.cheng@gmail.com>
5926             Chung-Ju Wu  <jasonwucj@gmail.com>
5927
5928         * config/nds32/nds32-multiple.md (load_multiple): Disallow
5929         volatile memory.
5930         (store_multiple): Ditto.
5931
5932 2018-02-26  Kito Cheng  <kito.cheng@gmail.com>
5933
5934         * config.gcc: Add --with-cpu support for nds32 target.
5935         * config/nds32/nds32-opts.h (nds32_cpu_type): New.
5936         * config/nds32/nds32.opt: Add -mcpu= option.
5937
5938 2018-02-25  Segher Boessenkool  <segher@kernel.crashing.org>
5939
5940         * config/rs6000/rs6000.opt (mvrsave=no, mvrsave=yes, isel=no,
5941         isel=yes): Warn for these deprecated options.
5942
5943 2018-02-23  David Edelsohn  <dje.gcc@gmail.com>
5944
5945         * config/rs6000/aix71.h (TARGET_DEFAULT): Change to
5946         ISA_2_5_MASKS_EMBEDDED.
5947
5948 2018-02-23  Jakub Jelinek  <jakub@redhat.com>
5949
5950         * ipa-prop.c (ipa_vr_ggc_hash_traits::hash): Hash p->min and
5951         p->max as pointers rather than using iterative_hash_expr.
5952
5953 2018-02-23  Carl Love  <cel@us.ibm.com>
5954
5955         * config/rs6000/rs6000-builtin.def: Change VSIGNED2 and VUNSIGNED2
5956         macro expansions from BU_VSX_2 to BU_P8V_VSX_2 and BU_VSX_OVERLOAD_2 to
5957         BU_P8V_OVERLOAD_2.
5958         * config/rs6000/rs6000-c.c: Change VSX_BUILTIN_VEC_VSIGNED2 to
5959         P8V_BUILTIN_VEC_VSIGNED2.  Change VSX_BUILTIN_VEC_VUNSIGNED2 to
5960         P8V_BUILTIN_VEC_VUNSIGNED2.
5961
5962 2018-02-22  Vladimir Makarov  <vmakarov@redhat.com>
5963
5964         PR target/81572
5965         * lra-int.h (LRA_UNKNOWN_ALT, LRA_NON_CLOBBERED_ALT): New macros.
5966         * lra.c (lra_set_insn_recog_data, lra_update_insn_recog_data): Use
5967         LRA_UNKNOWN_ALT.
5968         * lra-constraints.c (curr_insn_transform): Set up
5969         LRA_NON_CLOBBERED_ALT for moves processed on the fast path.  Use
5970         LRA_UNKNOWN_ALT.
5971         (remove_inheritance_pseudos): Use LRA_UNKNOWN_ALT.
5972         * lra-eliminations.c (spill_pseudos): Ditto.
5973         (process_insn_for_elimination): Ditto.
5974         * lra-lives.c (reg_early_clobber_p): Use the new macros.
5975         * lra-spills.c (spill_pseudos): Use LRA_UNKNOWN_ALT and
5976         LRA_NON_CLOBBERED_ALT.
5977
5978 2018-02-22  Martin Sebor  <msebor@redhat.com>
5979
5980         PR tree-optimization/84480
5981         * gimple-fold.c (gimple_fold_builtin_strcpy): Move warnings
5982         to maybe_diag_stxncpy_trunc.  Call it.
5983         * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Integrate warnings
5984         from gimple_fold_builtin_strcpy.  Print inlining stack.
5985         (handle_builtin_stxncpy): Print inlining stack.
5986         * tree-ssa-strlen.h (maybe_diag_stxncpy_trunc): Declare.
5987
5988 2018-02-22  H.J. Lu  <hongjiu.lu@intel.com>
5989
5990         PR target/84176
5991         * config/i386/i386.c (ix86_set_indirect_branch_type): Issue an
5992         error when -mindirect-branch=thunk-extern, -fcf-protection=branch
5993         and -fcheck-pointer-bounds are used together.
5994         (indirect_thunk_prefix): New enum.
5995         (indirect_thunk_need_prefix): New function.
5996         (indirect_thunk_name): Replace need_bnd_p with need_prefix.  Use
5997         "_nt" instead of "_bnd" for NOTRACK prefix.
5998         (output_indirect_thunk): Replace need_bnd_p with need_prefix.
5999         (output_indirect_thunk_function): Likewise.
6000         (): Likewise.
6001         (ix86_code_end): Update output_indirect_thunk_function calls.
6002         (ix86_output_indirect_branch_via_reg): Replace
6003         ix86_bnd_prefixed_insn_p with indirect_thunk_need_prefix.
6004         (ix86_output_indirect_branch_via_push): Likewise.
6005         (ix86_output_function_return): Likewise.
6006         * doc/invoke.texi: Document -mindirect-branch=thunk-extern is
6007         incompatible with -fcf-protection=branch and
6008         -fcheck-pointer-bounds.
6009
6010 2018-02-22  Steve Ellcey  <sellcey@cavium.com>
6011
6012         PR target/83335
6013         * config/aarch64/aarch64.c (aarch64_print_address_internal):
6014         Change gcc_assert call to output_operand_lossage.
6015
6016 2018-02-22  Steve Ellcey  <sellcey@cavium.com>
6017
6018         * doc/extend.texi (__builtin_extend_pointer): Document builtin.
6019
6020 2018-02-22  DJ Delorie  <dj@redhat.com>
6021             Sebastian Perta  <sebastian.perta@renesas.com>
6022             Oleg Endo  <olegendo@gcc.gnu.org>
6023
6024         * config/rx/rx.c (rx_rtx_costs): New function.
6025         (TARGET_RTX_COSTS): Override to use rx_rtx_costs.
6026
6027 2018-02-22  Thomas Preud'homme  <thomas.preudhomme@arm.com>
6028
6029         * config/arm/t-multilib: Map Armv8-R to Armv7 multilibs.
6030
6031 2018-02-22  Martin Liska  <mliska@suse.cz>
6032
6033         PR driver/83193
6034         * common/config/arm/arm-common.c (arm_print_hint_for_cpu_option):
6035         Add "native" as a possible value.
6036
6037 2018-02-22  Martin Liska  <mliska@suse.cz>
6038
6039         PR driver/83193
6040         * config/i386/i386.c (ix86_option_override_internal):
6041         Add "native" as a possible value for -march and -mtune.
6042
6043 2018-02-22  Jakub Jelinek  <jakub@redhat.com>
6044
6045         PR target/84502
6046         * stor-layout.c (finalize_type_size): Propagate TYPE_EMPTY_P flag
6047         to all type variants.
6048
6049         PR tree-optimization/84503
6050         * gimple-ssa-store-merging.c (merged_store_group::merge_into): Compute
6051         width as info->bitpos + info->bitsize - start.
6052         (merged_store_group::merge_overlapping): Simplify width computation.
6053         (check_no_overlap): New function.
6054         (imm_store_chain_info::try_coalesce_bswap): Compute expected
6055         start + width and last_order of the group, fail if check_no_overlap
6056         fails.
6057         (imm_store_chain_info::coalesce_immediate_stores): Don't merge info
6058         to group if check_no_overlap fails.
6059
6060 2018-02-21  Segher Boessenkool  <segher@kernel.crashing.org>
6061
6062         * config/rs6000/altivec.md: Delete contraint arguments to
6063         define_expand, define_split, and define_peephole2, and in
6064         define_insn_and_split if always unused.
6065         * config/rs6000/darwin.md: Ditto.
6066         * config/rs6000/dfp.md: Ditto.
6067         * config/rs6000/rs6000.md: Ditto.
6068         * config/rs6000/sync.md: Ditto.
6069         * config/rs6000/vector.md: Ditto.
6070         * config/rs6000/vsx.md: Ditto.
6071
6072 2018-02-21  Segher Boessenkool  <segher@kernel.crashing.org>
6073
6074         * config/rs6000/altivec.md: Write output control strings as braced
6075         blocks instead of double-quoted strings.
6076         * config/rs6000/darwin.md: Ditto.
6077         * config/rs6000/rs6000.md: Ditto.
6078         * config/rs6000/vector.md: Ditto.
6079         * config/rs6000/vsx.md: Ditto.
6080
6081 2018-02-21  Jason Merrill  <jason@redhat.com>
6082
6083         PR c++/84314 - ICE with templates and fastcall attribute.
6084         * attribs.c (build_type_attribute_qual_variant): Remove assert.
6085
6086 2018-02-21  Jan Hubicka  <hubicka@ucw.cz>
6087
6088         * ipa-cp.c (determine_versionability): Fix comment typos.
6089
6090 2018-02-21  Jan Hubicka  <hubicka@ucw.cz>
6091
6092         PR c/84229
6093         * ipa-cp.c (determine_versionability): Do not version functions caling
6094         va_arg_pack.
6095
6096 2018-02-21  Martin Liska  <mliska@suse.cz>
6097
6098         PR driver/83193
6099         * config/aarch64/aarch64.c (aarch64_print_hint_for_core_or_arch):
6100         Add "native" as a possible value.
6101         * config/aarch64/aarch64.h (HAVE_LOCAL_CPU_DETECT):  Define
6102         the macro when native cpu detection is available.
6103
6104 2018-02-21  Martin Liska  <mliska@suse.cz>
6105
6106         PR driver/83193
6107         * common/config/arm/arm-common.c (arm_print_hint_for_arch_option):
6108         Add "native" as a possible value.
6109         * config/arm/arm.h (HAVE_LOCAL_CPU_DETECT): Define the macro
6110         when native cpu detection is available.
6111
6112 2018-02-21  Jakub Jelinek  <jakub@redhat.com>
6113             Martin Sebor  <msebor@redhat.com>
6114
6115         PR tree-optimization/84478
6116         * gimple-fold.h (get_range_strlen): Add a bool argument defaulted to
6117         false.
6118         * gimple-fold.c (get_range_strlen): Make minlen const and assume it
6119         can't be NULL.  Change FUZZY from bool to int, for 1 add PHI/COND_EXPR
6120         support which is conservatively correct, for 2 only stay conservative
6121         for maxlen.  Formatting and comment capitalization fixes.  Add STRICT
6122         argument to the 2 argument get_range_strlen, adjust 6 arg
6123         get_range_strlen caller and clear minmaxlen[0] and [1] if it returned
6124         false.
6125         (get_maxval_strlen): Adjust 6 arg get_range_strlen caller.
6126         (gimple_fold_builtin_strlen): Pass true as last argument to
6127         get_range_strlen.
6128
6129 2018-02-20  Martin Sebor  <msebor@redhat.com>
6130
6131         PR middle-end/84095
6132         * gimple-ssa-warn-restrict.c (builtin_memref::extend_offset_range): New.
6133         (builtin_memref::set_base_and_offset): Same.  Handle inner references.
6134         (builtin_memref::builtin_memref): Factor out parts into
6135         set_base_and_offset and call it.
6136
6137 2018-02-20  Richard Sandiford  <richard.sandiford@linaro.org>
6138
6139         PR middle-end/84406
6140         * optabs-query.c (find_widening_optab_handler_and_mode): If from_mode
6141         is a scalar_int_mode, assert that to_mode is a scalar_int_mode with
6142         greater precision.  If to_mode is a MODE_PARTIAL_INT, stop the
6143         search at the associated MODE_INT.
6144
6145 2018-02-20  Jeff Law  <law@redhat.com>
6146
6147         PR middle-end/82123
6148         PR tree-optimization/81592
6149         PR middle-end/79257
6150         * gimple-ssa-sprintf.c (format_integer): Query EVRP range analyzer
6151         for range data rather than using global data.
6152         * gimple-ssa-sprintf.c (get_int_range): Query EVRP range analyzer for
6153         range data rather than using global data.
6154         * gimple-ssa-sprintf.c (get_int_range): Accept vr_values parameter
6155         pass it to children as needed.
6156         (struct directive::fmtresult): Similarly.
6157         (struct directive::set_width): Similarly.
6158         (struct directive::set_precision): Similarly.
6159         (format_integer, format_directive, parse_directive): Similarly.
6160         (format_none): Accept unnamed vr_values parameter.
6161         (format_percent, format_floating, format_character): Similarly.
6162         (format_string, format_plain): Similarly.
6163         * gimple-ssa-sprintf.c (sprintf_dom_walker::handle_gimple_call): Query
6164         the EVRP range analyzer for range data rather than using global data.
6165         * gimple-ssa-sprintf.c: Include alloc-pool.h, vr-values.h and
6166         gimple-ssa-evrp-analyze.h
6167         (class sprintf_dom_walker): Add after_dom_children member function.
6168         Add evrp_range_analyzer member.
6169         (sprintf_dom_walker::before_dom_children): Call into the EVRP
6170         range analyzer as needed.
6171         (sprintf_dom_walker::after_dom_children): New member function.
6172         * gimple-ssa-evrp-analyze.c (evrp_range_analyzer::enter): Do nothing
6173         if not optimizing.
6174         (evrp_range_analyzer::record_ranges_from_stmt): Likewise.
6175         (evrp_range_analyzer::pop_to_marker): Likewise.
6176
6177 2018-02-20  Richard Sandiford  <richard.sandiford@linaro.org>
6178
6179         PR tree-optimization/84419
6180         * internal-fn.c (expand_call_mem_ref): Create a TARGET_MEM_REF
6181         with the required type if its current type is compatible but
6182         different.
6183
6184 2018-02-20  Jakub Jelinek  <jakub@redhat.com>
6185
6186         PR middle-end/82004
6187         * match.pd (pow(C,x) -> exp(log(C)*x)): Delay all folding until
6188         after vectorization.
6189
6190 2018-02-20  Martin Liska  <mliska@suse.cz>
6191
6192         PR driver/83193
6193         * config/aarch64/aarch64.c (aarch64_print_hint_for_core_or_arch): Print
6194         possible values if we don't have a hint.
6195
6196 2018-02-20  Martin Liska  <mliska@suse.cz>
6197
6198         PR c/84310
6199         PR target/79747
6200         * final.c (shorten_branches): Build align_tab array with one
6201         more element.
6202         * opts.c (finish_options): Add alignment option limit check.
6203         (MAX_CODE_ALIGN): Likewise.
6204         (MAX_CODE_ALIGN_VALUE): Likewise.
6205         * doc/invoke.texi: Document maximum allowed option value for
6206         all -falign-* options.
6207
6208 2018-02-19  Jakub Jelinek  <jakub@redhat.com>
6209
6210         PR target/84146
6211         * reg-notes.def (REG_CALL_ARG_LOCATION): New reg note.
6212         * insn-notes.def (NOTE_INSN_CALL_ARG_LOCATION): Remove.
6213         * var-tracking.c (emit_note_insn_var_location): Remove all references
6214         to NOTE_INSN_CALL_ARG_LOCATION.
6215         (emit_notes_in_bb): Emit arguments as REG_CALL_ARG_LOCATION note on
6216         the CALL_INSN rather than separate NOTE_INSN_CALL_ARG_LOCATION note.
6217         Use copy_rtx_if_shared.
6218         * dwarf2out.c (gen_subprogram_die): Use XEXP with 0 instead of
6219         NOTE_VAR_LOCATION on ca_loc->call_arg_loc_note.
6220         (dwarf2out_var_location): Remove handling of
6221         NOTE_INSN_CALL_ARG_LOCATION, instead handle REG_CALL_ARG_LOCATION note
6222         on call_insn.
6223         * final.c (final_scan_insn): Remove all references to
6224         NOTE_INSN_CALL_ARG_LOCATION.
6225         (rest_of_clean_state): Likewise.  Remove REG_CALL_ARG_LOCATION notes
6226         before dumping final insns.
6227         * except.c (emit_note_eh_region_end): Remove all references to
6228         NOTE_INSN_CALL_ARG_LOCATION.
6229         * config/alpha/alpha.c (alpha_pad_function_end): Likewise.
6230         * config/c6x/c6x.c (c6x_gen_bundles): Likewise.
6231         * config/arc/arc.c (hwloop_optimize): Likewise.
6232         * config/arm/arm.c (create_fix_barrier): Likewise.
6233         * config/s390/s390.c (s390_chunkify_start): Likewise.
6234         * config/sh/sh.c (find_barrier): Likewise.
6235         * config/i386/i386.c (rest_of_insert_endbranch,
6236         ix86_seh_fixup_eh_fallthru): Likewise.
6237         * config/xtensa/xtensa.c (hwloop_optimize): Likewise.
6238         * config/iq2000/iq2000.c (final_prescan_insn): Likewise.
6239         * config/frv/frv.c (frv_function_prologue): Likewise.
6240         * emit-rtl.c (try_split): Likewise.  Copy over REG_CALL_ARG_LOCATION
6241         reg note.
6242         (note_outside_basic_block_p): Remove all references to
6243         NOTE_INSN_CALL_ARG_LOCATION.
6244         * gengtype.c (adjust_field_rtx_def): Likewise.
6245         * print-rtl.c (rtx_writer::print_rtx_operand_code_0, print_insn):
6246         Likewise.
6247         * jump.c (cleanup_barriers, delete_related_insns): Likewise.
6248         * cfgrtl.c (force_nonfallthru_and_redirect): Likewise.
6249
6250         PR c++/84444
6251         * builtins.c (builtin_mathfn_code): Don't check if CALL_EXPR_FN (t)
6252         is ADDR_EXPR.
6253
6254         PR tree-optimization/84452
6255         * tree-vect-patterns.c (vect_recog_pow_pattern): Don't call
6256         expand_simd_clones if targetm.simd_clone.compute_vecsize_and_simdlen
6257         is NULL.
6258
6259 2018-02-19  Martin Liska  <mliska@suse.cz>
6260
6261         PR sanitizer/82183
6262         * passes.def: Put pass_sancov_O0 before pass_lower_switch with -O0.
6263
6264 2018-02-19  Martin Liska  <mliska@suse.cz>
6265             Richard Sandiford  <richard.sandiford@linaro.org>
6266
6267         PR tree-optimization/82491
6268         * gimple-fold.c (get_base_constructor): Make earlier bail out
6269         to prevent ubsan.
6270
6271 2018-02-19  Carl Love  <cel@us.ibm.com>
6272
6273         * config/rs6000/rs6000-builtin.def: Change NEG macro expansions from
6274         BU_ALTIVEC_A to BU_P8V_AV_1 and BU_ALTIVEC_OVERLOAD_1 to
6275         BU_P8V_OVERLOAD_1.
6276         * config/rs6000/rs6000-c.c: Change ALTIVEC_BUILTIN_VEC_NEG to
6277         P8V_BUILTIN_VEC_NEG.
6278
6279 2018-02-19  Sebastian Perta  <sebastian.perta@renesas.com>
6280
6281         * config/rl78/rl78.md (movdf): New define expand.
6282
6283 2018-02-19  Martin Liska  <mliska@suse.cz>
6284
6285         PR other/80589
6286         * doc/invoke.texi: Fix typo.
6287         * params.def (PARAM_MAX_LOOP_HEADER_INSNS): Likewise.
6288
6289 2018-02-18  Segher Boessenkool  <segher@kernel.crashing.org>
6290
6291         * config/rs6000/rs6000.c (rs6000_option_override_internal): Don't
6292         handle rs6000_single_float and rs6000_double_float specially for
6293         e500 family CPUs.
6294
6295 2018-02-16  Jeff Law  <law@redhat.com>
6296
6297         * config/rx/rx.c (add_pop_cfi_notes): New function.;
6298         (pop_regs): Use it.
6299
6300 2018-02-16  Jakub Jelinek  <jakub@redhat.com>
6301
6302         PR ipa/84425
6303         * ipa-inline.c (inline_small_functions): Fix a typo.
6304
6305 2018-02-16  Nathan Sidwell  <nathan@acm.org>
6306
6307         * doc/extend.texi (Backwards Compatibility): Americanize 'behaviour'.
6308
6309 2018-02-16  Carl Love  <cel@us.ibm.com>
6310
6311         * config/rs6000/rs6000-builtin.def: Add BU_P8V_VSX_2 macro definition.
6312         Change FLOAT2 expansions from BU_VSX_2 to BU_P8V_VSX_2 and
6313         from BU_VSX_OVERLOAD_2 to BU_P8V_OVERLOAD_2.
6314         * config/rs6000/rs6000-c.c: Changed macro VSX_BUILTIN_VEC_FLOAT2
6315         expansion to P8V_BUILTIN_VEC_FLOAT2.
6316
6317 2018-02-16  Vladimir Makarov  <vmakarov@redhat.com>
6318
6319         PR rtl-optimization/70023
6320         * lra-constraints.c (inherit_in_ebb): Take hard reg mode of
6321         src_regno into account.
6322
6323 2018-02-16  Carl Love  <cel@us.ibm.com>
6324
6325         * config/rs6000/altivec.h: Remove vec_vextract4b and vec_vinsert4b.
6326         * config/rs6000/rs6000-builtin.def: Remove macro expansion for
6327         VEXTRACT4B, VINSERT4B, VINSERT4B_DI and VEXTRACT4B.
6328         * config/rs6000/rs6000.c: Remove case statements for
6329         P9V_BUILTIN_VEXTRACT4B, P9V_BUILTIN_VEC_VEXTRACT4B,
6330         P9V_BUILTIN_VINSERT4B, P9V_BUILTIN_VINSERT4B_DI,
6331         and P9V_BUILTIN_VEC_VINSERT4B.
6332         * config/rs6000/rs6000-c.c (altivec_expand_builtin): Remove entries for
6333         P9V_BUILTIN_VEC_VEXTRACT4B and P9V_BUILTIN_VEC_VINSERT4B.
6334         * config/rs6000/vsx.md:
6335         * doc/extend.texi: Remove vec_vextract4b, non ABI definitions for
6336         vec_insert4b.
6337
6338 2018-02-16  Carl Love  <cel@us.ibm.com>
6339
6340         * config/rs6000/altivec.h: Add builtin names vec_extract4b
6341         vec_insert4b.
6342         * config/rs6000/rs6000-builtin.def: Add INSERT4B and EXTRACT4B
6343         definitions.
6344         * config/rs6000/rs6000-c.c: Add the definitions for
6345         P9V_BUILTIN_VEC_EXTRACT4B and P9V_BUILTIN_VEC_INSERT4B.
6346         * config/rs6000/rs6000.c (altivec_expand_builtin): Add
6347         P9V_BUILTIN_EXTRACT4B and P9V_BUILTIN_INSERT4B case statements.
6348         * config/rs6000/vsx.md: Add define_insn extract4b.  Add define_expand
6349         definition for insert4b and define insn *insert3b_internal.
6350         * doc/extend.texi: Add documentation for vec_extract4b.
6351
6352 2018-02-16  Nathan Sidwell  <nathan@acm.org>
6353
6354         * doc/extend.texi (Backwards Compatibility): Mention friend
6355         injection.  Note for-scope is deprecated.
6356         * doc/invoke.texi (-ffriend-injection): Deprecate.
6357
6358 2018-02-16  Segher Boessenkool  <segher@kernel.crashing.org>
6359
6360         * combine.c (try_combine): When adjusting LOG_LINKS for the destination
6361         that moved to I2, also allow destinations that are a paradoxical
6362         subreg (instead of a normal reg).
6363
6364 2018-02-16  Oleg Endo  <olegendo@gcc.gnu.org>
6365
6366         PR target/83831
6367         * config/rx/rx.c (rx_fuse_in_memory_bitop): Convert shift operand
6368         to QImode.
6369
6370 2018-02-16  Richard Biener  <rguenther@suse.de>
6371
6372         PR tree-optimization/84037
6373         PR tree-optimization/84016
6374         PR target/82862
6375         * config/i386/i386.c (ix86_builtin_vectorization_cost):
6376         Adjust vec_construct for the fact we need additional higher latency
6377         128bit inserts for AVX256 and AVX512 vector builds.
6378         (ix86_add_stmt_cost): Scale vector construction cost for
6379         elementwise loads.
6380
6381 2018-02-16  Richard Biener  <rguenther@suse.de>
6382
6383         PR tree-optimization/84417
6384         * tree-ssa.c (non_rewritable_mem_ref_base): Properly constrain
6385         the MEM_REF offset when conversion to BIT_FIELD_REF is desired.
6386         (non_rewritable_lvalue_p): Likewise, use poly-ints.
6387
6388 2018-02-16  Martin Liska  <mliska@suse.cz>
6389
6390         PR sanitizer/84307
6391         * internal-fn.def (ASAN_CHECK): Set proper flags.
6392         (ASAN_MARK): Likewise.
6393
6394 2018-02-16  Julia Koval  <julia.koval@intel.com>
6395
6396         * config/i386/i386.c (ix86_option_override_internal): Remove PTA_CLWB
6397         from PTA_CANNONLAKE.
6398
6399 2018-02-16  Jakub Jelinek  <jakub@redhat.com>
6400
6401         PR target/84272
6402         * config/aarch64/cortex-a57-fma-steering.c (fma_forest::merge_forest):
6403         Use ++iter rather than iter++ for std::list iterators.
6404         (func_fma_steering::dfs): Likewise.  Don't delete nodes right away,
6405         defer deleting them until all nodes in the forest are processed.  Do
6406         free even leaf nodes.  Change to_process into auto_vec.
6407
6408         PR bootstrap/84405
6409         * system.h (BROKEN_VALUE_INITIALIZATION): Define for GCC < 4.3.
6410         * vec.h (vec_default_construct): Use memset instead of placement new
6411         if BROKEN_VALUE_INITIALIZATION is defined.
6412         * hash-table.h (hash_table<Descriptor, Allocator>::empty_slow): Use
6413         memset instead of value initialization if BROKEN_VALUE_INITIALIZATION
6414         is defined.
6415
6416         PR rtl-optimization/83723
6417         * lra-int.h (lra_substitute_pseudo): Add DEBUG_P argument.
6418         * lra.c (lra_substitute_pseudo): Likewise.  If true, use
6419         gen_rtx_raw_SUBREG instead of gen_rtx_SUBREG.  Pass DEBUG_P to
6420         recursive calls.
6421         (lra_substitute_pseudo_within_insn): Adjust lra_substitute_pseudo
6422         callers.
6423         * lra-constraints.c (inherit_reload_reg, split_reg): Likewise.
6424
6425 2018-02-16  Eric Botcazou  <ebotcazou@adacore.com>
6426
6427         PR rtl-optimization/81443
6428         * rtlanal.c (num_sign_bit_copies1) <SUBREG>: Do not propagate results
6429         from inner REGs to paradoxical SUBREGs.
6430
6431 2018-02-16  Richard Biener  <rguenther@suse.de>
6432
6433         PR tree-optimization/84399
6434         * graphite-scop-detection.c (scop_detection::stmt_simple_for_scop_p):
6435         For operands we can analyze at their definition make sure we can
6436         analyze them at each use as well.
6437
6438 2018-02-16  Richard Biener  <rguenther@suse.de>
6439
6440         PR tree-optimization/84190
6441         * tree-ssa.c (non_rewritable_mem_ref_base): Do not touch
6442         volatile accesses if the decl isn't volatile.
6443
6444 2018-02-15  Jason Merrill  <jason@redhat.com>
6445
6446         PR c++/84314 - ICE with templates and fastcall attribute.
6447         * attribs.c (build_type_attribute_qual_variant): Don't clobber
6448         TYPE_CANONICAL on an existing type.
6449
6450 2018-02-15  Jakub Jelinek  <jakub@redhat.com>
6451
6452         PR tree-optimization/84383
6453         * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Don't look at
6454         dstoff nor call operand_equal_p if dstbase is NULL.
6455
6456         PR tree-optimization/84334
6457         * match.pd ((A +- CST1) +- CST2 -> A + CST3): If A is
6458         also a CONSTANT_CLASS_P, punt.
6459
6460 2018-02-14  Jim Wilson  <jimw@sifive.com>
6461
6462         * config/riscv/riscv.c (riscv_first_stack_step): Move locals after
6463         first SMALL_OPERAND check.  New local min_second_step.  Move assert
6464         to where locals are set.  Add TARGET_RVC support.
6465         * config/riscv/riscv.h (C_SxSP_BITS, SWSP_REACH, SDSP_REACH): New.
6466
6467 2018-02-14  Indu Bhagat  <indu.bhagat@oracle.com>
6468
6469         * doc/invoke.texi: Correct -Wformat-overflow code sample.
6470
6471 2018-02-14  Martin Sebor  <msebor@redhat.com>
6472
6473         PR tree-optimization/83698
6474         * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For
6475         arrays constrain the offset range to their bounds.
6476         (builtin_access::strcat_overlap): Adjust the bounds of overlap offset.
6477         (builtin_access::overlap): Avoid setting the size of overlap if it's
6478         already been set.
6479         (maybe_diag_overlap): Also consider arrays when deciding what values
6480         of offsets to include in diagnostics.
6481
6482 2018-02-14  Martin Sebor  <msebor@redhat.com>
6483
6484         PR c/84108
6485         * attribs.c (diag_attr_exclusions): Consider the exclusion(s)
6486         that correspond to the kind of a declaration.
6487
6488 2018-02-14  John David Anglin  <danglin@gcc.gnu.org>
6489
6490         PR target/83984
6491         * config/pa/pa.md: Load address of PIC label using the linkage table
6492         if the label is nonlocal.
6493
6494 2018-02-14  Kelvin Nilsen  <kelvin@gcc.gnu.org>
6495
6496         * config/rs6000/rs6000.c (rs6000_option_override_internal): Issue
6497         warning message if user requests -maltivec=be.
6498         * doc/invoke.texi: Document deprecation of -maltivec=be.
6499
6500 2018-02-14  Will Schmidt  <will_schmidt@vnet.ibm.com>
6501
6502         PR target/84220
6503         * config/rs6000/rs6000-c.c: Update definitions for
6504         ALTIVEC_BUILTIN_VEC_SLD, ALTIVEC_BUILTIN_VEC_SLDW,
6505         VEC_XXSLDWI and ALTIVEC_BUILTIN_VEC_XXPERMDI builtins.
6506
6507 2018-02-14  Igor Tsimbalist  <igor.v.tsimbalist@intel.com>
6508
6509         PR target/84239
6510         * config/i386/cetintrin.h: Remove _rdssp[d|q] and
6511         add _get_ssp intrinsics. Remove argument from
6512         __builtin_ia32_rdssp[d|q].
6513         * config/i386/i386-builtin-types.def: Add UINT_FTYPE_VOID.
6514         * config/i386/i386-builtin.def: Remove argument from
6515         __builtin_ia32_rdssp[d|q].
6516         * config/i386/i386.c: Use UINT_FTYPE_VOID. Use
6517         ix86_expand_special_args_builtin for _rdssp[d|q].
6518         * config/i386/i386.md: Remove argument from rdssp[si|di] insn.
6519         Clear register before usage.
6520         * doc/extend.texi: Remove argument from __builtin_ia32_rdssp[d|q].
6521         Add documentation for new _get_ssp and _inc_ssp intrinsics.
6522
6523 2018-02-14  Richard Sandiford  <richard.sandiford@linaro.org>
6524
6525         PR tree-optimization/84357
6526         * tree-data-ref.c (object_address_invariant_in_loop_p): Check
6527         operand 1 of an ARRAY_REF too.
6528
6529 2018-02-14  Oleg Endo  <olegendo@gcc.gnu.org>
6530
6531         PR target/83831
6532         * config/rx/rx-protos.h (rx_reg_dead_or_unused_after_insn,
6533         rx_copy_reg_dead_or_unused_notes, rx_fuse_in_memory_bitop): New
6534         declarations.
6535         (set_of_reg): New struct.
6536         (rx_find_set_of_reg, rx_find_use_of_reg): New functions.
6537         * config/rx/rx.c (rx_reg_dead_or_unused_after_insn,
6538         rx_copy_reg_dead_or_unused_notes, rx_fuse_in_memory_bitop): New
6539         functions.
6540         * config/rx/rx.md (andsi3, iorsi3, xorsi3): Convert to insn_and_split.
6541         Split into bitclr, bitset, bitinvert patterns if appropriate.
6542         (*bitset, *bitinvert, *bitclr): Convert to named insn_and_split and
6543         use rx_fuse_in_memory_bitop.
6544         (*bitset_in_memory, *bitinvert_in_memory, *bitclr_in_memory): Convert
6545         to named insn, correct maximum insn length.
6546
6547 2018-02-14  Jozef Lawrynowicz  <jozefl.gcc@gmail.com>
6548
6549         PR target/79242
6550         * machmode.def: Define a complex mode for PARTIAL_INT.
6551         * genmodes.c (complex_class): Return MODE_COMPLEX_INT for
6552         MODE_PARTIAL_INT.
6553         * doc/rtl.texi: Document CSPImode.
6554         * config/msp430/msp430.c (msp430_hard_regno_nregs): Add CPSImode
6555         handling.
6556         (msp430_hard_regno_nregs_with_padding): Likewise.
6557
6558 2018-02-13  Peter Bergner  <bergner@vnet.ibm.com>
6559
6560         PR target/84279
6561         * config/rs6000/rs6000.c (mem_operand_gpr): Disallow altivec addresses.
6562
6563 2018-02-13  Segher Boessenkool  <segher@kernel.crashing.org>
6564
6565         PR rtl-optimization/84169
6566         * combine.c (try_combine): New variable split_i2i3.  Set it to true if
6567         we generated a parallel as new i3 and we split that to new i2 and i3
6568         instructions.  Handle split_i2i3 similar to swap_i2i3: scan the
6569         LOG_LINKs of i3 to see which of those need to link to i2 now.  Link
6570         those to i2, not i1.  Partially rewrite this scan code.
6571
6572 2018-02-13  Jakub Jelinek  <jakub@redhat.com>
6573
6574         PR c/82210
6575         * stor-layout.c (place_field): For variable length fields, adjust
6576         offset_align afterwards not just based on the field's alignment,
6577         but also on the size.
6578
6579         PR middle-end/84309
6580         * match.pd (pow(C,x) -> exp(log(C)*x)): Use exp2s and log2s instead
6581         of exps and logs in the use_exp2 case.
6582
6583 2018-02-13  Jeff Law  <law@redhat.com>
6584
6585         * config/rl/rl78.c (rl78_attribute_table): Fix terminator and
6586         entry for "vector".
6587
6588         * config/rl78/rl78.c (rl78_handle_func_attribute): Mark
6589         ARGS as unused.
6590
6591 2018-02-13  Alexandre Oliva  <aoliva@redhat.com>
6592
6593         PR debug/84342
6594         PR debug/84319
6595         * common.opt (gas-loc-support, gas-locview-support): New.
6596         (ginline-points, ginternal-reset-location-views): New.
6597         * doc/invoke.texi: Document them.  Use @itemx where intended.
6598         (gvariable-location-views): Adjust.
6599         * target.def (reset_location_view): New.
6600         * doc/tm.texi.in (DWARF2_ASM_VIEW_DEBUG_INFO): New.
6601         (TARGET_RESET_LOCATION_VIEW): New.
6602         * doc/tm.texi: Rebuilt.
6603         * dwarf2out.c (dwarf2out_default_as_loc_support): New.
6604         (dwarf2out_default_as_locview_support): New.
6605         (output_asm_line_debug_info): Use option variables.
6606         (dwarf2out_maybe_output_loclist_view_pair): Likewise.
6607         (output_loc_list): Likewise.
6608         (add_high_low_attributes): Check option variables.
6609         Don't output entry view attribute in strict mode.
6610         (gen_inlined_subroutine_die): Check option variables.
6611         (dwarf2out_inline_entry): Likewise.
6612         (init_sections_and_labels): Likewise.
6613         (dwarf2out_early_finish): Likewise.
6614         (maybe_reset_location_view): New, from...
6615         (dwarf2out_var_location): ... here.  Call it.
6616         * debug.h (dwarf2out_default_as_loc_support): Declare.
6617         (dwarf2out_default_as_locview_support): Declare.
6618         * hooks.c (hook_int_rtx_insn_0): New.
6619         * hooks.h (hook_int_rtx_insn_0): Declare.
6620         * toplev.c (process_options): Take -gas-loc-support and
6621         -gas-locview-support from dwarf2out.  Enable
6622         -gvariable-location-views by default only with locview
6623         assembler support.  Enable -ginternal-reset-location-views by
6624         default only if the target defines the corresponding hook.
6625         Enable -ginline-points by default if location views are
6626         enabled; force it disabled if statement frontiers are
6627         disabled.
6628         * tree-inline.c (expand_call_inline): Check option variables.
6629         * tree-ssa-live.c (remove_unused_scope_block_p): Likewise.
6630
6631 2018-02-13  Richard Sandiford  <richard.sandiford@linaro.org>
6632
6633         PR tree-optimization/84321
6634         * tree-vrp.c (intersect_range_with_nonzero_bits): Fix VR_ANTI_RANGE
6635         handling.  Also check whether the anti-range contains any values
6636         that satisfy the mask; switch to a VR_RANGE if not.
6637
6638 2018-02-13  Paolo Bonzini  <bonzini@gnu.org>
6639
6640         PR sanitizer/84340
6641         * internal-fn.def (ASAN_CHECK, ASAN_MARK): Revert changes to fnspec.
6642
6643 2018-02-13  Martin Jambor  <mjambor@suse.cz>
6644
6645         PR c++/83990
6646         * ipa-param-manipulation.c (ipa_modify_call_arguments): Use location
6647         of call statements, also set location of a load to a temporary.
6648
6649 2018-02-13  Sebastian Perta  <sebastian.perta@renesas.com>
6650
6651         * config/rl78/rl78.c (add_vector_labels): New function.
6652         * config/rl78/rl78.c (rl78_handle_vector_attribute): New function.
6653         * config/rl78/rl78.c (rl78_start_function): Call add_vector_labels.
6654         * config/rl78/rl78.c (rl78_handle_func_attribute): Removed the assert
6655         which checks that no arguments are passed.
6656         * config/rl78/rl78.c (rl78_attribute_table): Add "vector" attribute.
6657         * doc/extend.texi: Documentation for the new attribute.
6658
6659 2018-02-13  Andreas Schwab  <schwab@suse.de>
6660
6661         * config/riscv/linux.h (CPP_SPEC): Define.
6662
6663 2018-02-13  Jakub Jelinek  <jakub@redhat.com>
6664
6665         PR target/84335
6666         * config/i386/i386.c (ix86_init_mmx_sse_builtins): Pass
6667         OPTION_MASK_ISA_AES | OPTION_MASK_ISA_SSE2 instead of
6668         OPTION_MASK_ISA_AES as first argument to def_builtin_const
6669         for AES builtins.  Pass OPTION_MASK_ISA_PCLMUL | OPTION_MASK_ISA_SSE2
6670         instead of OPTION_MASK_ISA_PCLMUL as first argument to
6671         def_builtin_const for __builtin_ia32_pclmulqdq128 builtin.
6672         * config/i386/wmmintrin.h: If __SSE2__ is not defined, enable it
6673         temporarily for AES and PCLMUL builtins.
6674
6675         PR tree-optimization/84339
6676         * gimple-fold.c (get_range_strlen): Set *FLEXP to true when handling
6677         ARRAY_REF where first operand is array_at_struct_end_p COMPONENT_REF.
6678         Formatting fixes.
6679
6680         PR middle-end/84309
6681         * match.pd (pow(C,x) -> exp(log(C)*x)): Optimize instead into
6682         exp2(log2(C)*x) if C is a power of 2 and c99 runtime is available.
6683         * generic-match-head.c (canonicalize_math_after_vectorization_p): New
6684         inline function.
6685         * gimple-match-head.c (canonicalize_math_after_vectorization_p): New
6686         inline function.
6687         * omp-simd-clone.h: New file.
6688         * omp-simd-clone.c: Include omp-simd-clone.h.
6689         (expand_simd_clones): No longer static.
6690         * tree-vect-patterns.c: Include fold-const-call.h, attribs.h,
6691         cgraph.h and omp-simd-clone.h.
6692         (vect_recog_pow_pattern): Optimize pow(C,x) to exp(log(C)*x).
6693         (vect_recog_widen_shift_pattern): Formatting fix.
6694         (vect_pattern_recog_1): Don't check optab for calls.
6695
6696         PR target/84336
6697         * config/i386/sse.md (<avx512>_vpermi2var<mode>3_mask): Force
6698         operands[2] into a REG before using gen_lowpart on it.
6699
6700 2018-02-12  Jeff Law  <law@redhat.com>
6701
6702         PR target/83760
6703         * config/sh/sh.c (find_barrier): Consider a sibling call
6704         a barrier as well.
6705
6706         * cse.c (try_back_substitute_reg): Move any REG_ARGS_SIZE note when
6707         successfully back substituting a reg.
6708
6709 2018-02-12  Richard Biener  <rguenther@suse.de>
6710
6711         PR tree-optimization/84037
6712         * tree-vect-slp.c (vect_analyze_slp_cost): Add visited
6713         parameter, move visited init to caller.
6714         (vect_slp_analyze_operations): Separate cost from validity
6715         check, initialize visited once for all instances.
6716         (vect_schedule_slp): Analyze map to CSE vectorized nodes once
6717         for all instances.
6718         * tree-vect-stmts.c (vect_model_simple_cost): Make early
6719         out an assert.
6720         (vect_model_promotion_demotion_cost): Likewise.
6721         (vectorizable_bswap): Guard cost modeling with !slp_node
6722         instead of !PURE_SLP_STMT to avoid double-counting on hybrid
6723         SLP stmts.
6724         (vectorizable_call): Likewise.
6725         (vectorizable_conversion): Likewise.
6726         (vectorizable_assignment): Likewise.
6727         (vectorizable_shift): Likewise.
6728         (vectorizable_operation): Likewise.
6729         (vectorizable_store): Likewise.
6730         (vectorizable_load): Likewise.
6731         (vectorizable_condition): Likewise.
6732         (vectorizable_comparison): Likewise.
6733
6734 2018-02-12  Paolo Bonzini  <bonzini@gnu.org>
6735
6736         PR sanitizer/84307
6737         * internal-fn.def (ASAN_CHECK): Fix fnspec to account for return value.
6738         (ASAN_MARK): Fix fnspec to account for return value, change pointer
6739         argument from 'R' to 'W' so that the pointed-to datum is clobbered.
6740
6741 2018-02-08  Jan Hubicka  <hubicka@ucw.cz>
6742
6743         PR middle-end/83665
6744         * params.def (inline-min-speedup): Increase from 8 to 15.
6745         (max-inline-insns-auto): Decrease from 40 to 30.
6746         * ipa-split.c (consider_split): Add some buffer for function to
6747         be considered inlining candidate.
6748         * invoke.texi (max-inline-insns-auto, inline-min-speedup): UPdate
6749         default values.
6750
6751 2018-02-12  Richard Biener  <rguenther@suse.de>
6752
6753         PR tree-optimization/84037
6754         * tree-vect-slp.c (vect_build_slp_tree_2): Try swapping the
6755         matched stmts if we cannot swap the non-matched ones.
6756
6757 2018-02-12  Olga Makhotina  <olga.makhotina@intel.com>
6758
6759         * config/i386/avx512fintrin.h (_mm_mask_scalef_round_sd,
6760         _mm_maskz_scalef_round_sd, _mm_mask_scalef_round_ss,
6761         _mm_maskz_scalef_round_ss): New intrinsics.
6762         (__builtin_ia32_scalefsd_round, __builtin_ia32_scalefss_round): Fix.
6763         * config/i386/i386-builtin.def (__builtin_ia32_scalefsd_round,
6764         __builtin_ia32_scalefss_round): Remove.
6765         (__builtin_ia32_scalefsd_mask_round,
6766         __builtin_ia32_scalefss_mask_round): New intrinsics.
6767         * config/i386/sse.md (vmscalef<mode><round_name>): Renamed to ...
6768         (vmscalef<mode><mask_scalar_name><round_scalar_name>): ... this.
6769         ((match_operand:VF_128 2 "<round_nimm_predicate>"
6770         "<round_constraint>")): Changed to ...
6771         ((match_operand:VF_128 2 "<round_scalar_nimm_predicate>"
6772         "<round_scalar_constraint>")): ... this.
6773         ("vscalef<ssescalarmodesuffix>\t{<round_op3>%2, %1, %0|
6774         %0, %1, %2<round_op3>}"): Changed to ...
6775         ("vscalef<ssescalarmodesuffix>\t{<round_scalar_mask_op3>%2, %1,
6776         %0<mask_scalar_operand3>|%0<mask_scalar_operand3>, %1,
6777         %2<round_scalar_mask_op3>}"): ... this.
6778         * config/i386/subst.md (round_scalar_nimm_predicate): New.
6779
6780 2018-02-12  Olga Makhotina  <olga.makhotina@intel.com>
6781
6782         * config/i386/avx512fintrin.h (_mm_mask_sqrt_round_sd)
6783         (_mm_maskz_sqrt_round_sd, _mm_mask_sqrt_round_ss)
6784         (_mm_maskz_sqrt_round_ss): New intrinsics.
6785         (__builtin_ia32_sqrtsd_round, __builtin_ia32_sqrtss_round): Remove.
6786         (__builtin_ia32_sqrtsd_mask_round)
6787         (__builtin_ia32_sqrtss_mask_round): New builtins.
6788         * config/i386/i386-builtin.def (__builtin_ia32_sqrtsd_round)
6789         (__builtin_ia32_sqrtss_round): Remove.
6790         (__builtin_ia32_sqrtsd_mask_round)
6791         (__builtin_ia32_sqrtss_mask_round): New builtins.
6792         * config/i386/sse.md (vmsqrt<mode>2<round_name>): Renamed to ...
6793         (vmsqrt<mode>2<mask_scalar_name><round_scalar_name>): ... this.
6794         ((match_operand:VF_128 1 "vector_operand"
6795         "xBm,<round_constraint>")): Changed to ...
6796         ((match_operand:VF_128 1 "vector_operand"
6797         "xBm,<round_scalar_constraint>")): ... this.
6798         (vsqrt<ssescalarmodesuffix>\t{<round_op3>%1, %2, %0|
6799         %0, %2, %<iptr>1<round_op3>}): Changed to ...
6800         (vsqrt<ssescalarmodesuffix>\t{<round_scalar_mask_op3>%1, %2,
6801         %0<mask_scalar_operand3>|%0<mask_scalar_operand3>, %2,
6802         %<iptr>1<round_scalar_mask_op3>}): ... this.
6803         ((set_attr "prefix" "<round_prefix>")): Changed to ...
6804         ((set_attr "prefix" "<round_scalar_prefix>")): ... this.
6805
6806 2018-02-11  Steven Munroe  <munroesj@gcc.gnu.org>
6807
6808         PR target/84266
6809         * config/rs6000/mmintrin.h (_mm_cmpeq_pi32 [_ARCH_PWR9]):
6810         Cast vec_cmpeq result to correct type.
6811         * config/rs6000/mmintrin.h (_mm_cmpgt_pi32 [_ARCH_PWR9]):
6812         Cast vec_cmpgt result to correct type.
6813
6814 2018-02-11  Alexandre Oliva  <aoliva@redhat.com>
6815
6816         * final.c (final_scan_insn_1): Renamed from...
6817         (final_scan_insn): ... this.  New wrapper, to recover
6818         seen from the outermost call in recursive ones.
6819         * config/sparc/sparc.c (output_return): Drop seen from call.
6820         (output_sibcall): Likewise.
6821         * config/visium/visium.c (output_branch): Likewise.
6822
6823 2018-02-10  John David Anglin  <danglin@gcc.gnu.org>
6824
6825         * config/pa/pa.c (hppa_profile_hook): Mark SYMBOL_REF for _mcount as
6826         function label.
6827
6828 2018-02-10  Alan Modra  <amodra@gmail.com>
6829
6830         PR target/84300
6831         * config/rs6000/rs6000.md (split_stack_return): Remove (use ..).
6832         Specify LR as an input.
6833
6834 2018-02-10  Jakub Jelinek  <jakub@redhat.com>
6835
6836         PR sanitizer/83987
6837         * omp-low.c (maybe_remove_omp_member_access_dummy_vars,
6838         remove_member_access_dummy_vars): New functions.
6839         (lower_omp_for, lower_omp_taskreg, lower_omp_target,
6840         lower_omp_1, execute_lower_omp): Use them.
6841
6842         PR rtl-optimization/84308
6843         * shrink-wrap.c (spread_components): Release todo vector.
6844
6845 2018-02-09  Vladimir Makarov  <vmakarov@redhat.com>
6846
6847         PR rtl-optimization/57193
6848         * ira-color.c (struct allocno_color_data): Add member
6849         conflict_allocno_hard_prefs.
6850         (update_conflict_allocno_hard_prefs): New.
6851         (bucket_allocno_compare_func): Add a preference based on
6852         conflict_allocno_hard_prefs.
6853         (push_allocno_to_stack): Update conflict_allocno_hard_prefs.
6854         (color_allocnos): Remove a dead code.  Initiate
6855         conflict_allocno_hard_prefs.  Call update_costs_from_prefs.
6856
6857 2018-02-09  Jakub Jelinek  <jakub@redhat.com>
6858
6859         PR target/84226
6860         * config/rs6000/vsx.md (p9_xxbrq_v16qi): Change input operand
6861         constraint from =wa to wa.  Avoid a subreg on the output operand,
6862         instead use a pseudo and subreg it in a move.
6863         (p9_xxbrd_<mode>): Changed to ...
6864         (p9_xxbrd_v2di): ... this insn, without VSX_D iterator.
6865         (p9_xxbrd_v2df): New expander.
6866         (p9_xxbrw_<mode>): Changed to ...
6867         (p9_xxbrw_v4si): ... this insn, without VSX_W iterator.
6868         (p9_xxbrw_v4sf): New expander.
6869
6870 2018-02-09  Sebastian Perta  <sebastian.perta@renesas.com>
6871
6872         * config/rx/rx.md (movsicc): Update expander to be matched by GCC.
6873
6874 2018-02-09  Peter Bergner  <bergner@vnet.ibm.com>
6875
6876         PR target/83926
6877         * config/rs6000/vsx.md (vsx_mul_v2di): Handle generating a 64-bit
6878         multiply in 32-bit mode.
6879         (vsx_div_v2di): Handle generating a 64-bit signed divide in 32-bit mode.
6880         (vsx_udiv_v2di): Handle generating a 64-bit unsigned divide in 32-bit
6881         mode.
6882
6883 2018-02-09  Sebastian Perta  <sebastian.perta@renesas.com>
6884
6885         * config/rx/constraints.md (CALL_OP_SYMBOL_REF): Added new constraint
6886         to allow or block "symbol_ref" depending on the value of TARGET_JSR.
6887         * config/rx/rx.md (call_internal): Use CALL_OP_SYMBOL_REF.
6888         * config/rx/rx.md (call_value_internal): Use CALL_OP_SYMBOL_REF.
6889
6890 2018-02-09  Pierre-Marie de Rodat  <derodat@adacore.com>
6891
6892         PR lto/84213
6893         * dwarf2out.c (is_trivial_indirect_ref): New function.
6894         (dwarf2out_late_global_decl): Do not generate a location
6895         attribute for variables that have a non-trivial DECL_VALUE_EXPR
6896         and that are not defined in the current unit.
6897
6898 2018-02-09  Eric Botcazou  <ebotcazou@adacore.com>
6899
6900         * optabs.c (prepare_cmp_insn): Try harder to emit a direct comparison
6901         instead of a libcall for UNORDERED.
6902
6903 2018-02-09  Tamar Christina  <tamar.christina@arm.com>
6904
6905         PR target/82641
6906         * config/arm/arm-c.c (arm_cpu_builtins): Un-define __ARM_FEATURE_LDREX,
6907         __ARM_ARCH_PROFILE, __ARM_ARCH_ISA_THUMB, __ARM_FP and __ARM_NEON_FP.
6908
6909 2018-02-09  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>
6910
6911         PR target/PR84295
6912         * config/s390/s390.c (s390_set_current_function): Invoke
6913         s390_indirect_branch_settings also if fndecl didn't change.
6914
6915 2018-02-09  Alexandre Oliva  <aoliva@redhat.com>
6916
6917         * config/rs6000/rs6000.md (blockage): Set length to zero.
6918
6919 2018-02-09  Eric Botcazou  <ebotcazou@adacore.com>
6920
6921         * expr.c (optimize_bitfield_assignment_op): Remove obsolete assertion.
6922
6923 2018-02-09  Jakub Jelinek  <jakub@redhat.com>
6924
6925         PR sanitizer/84285
6926         * gcc.c (STATIC_LIBASAN_LIBS, STATIC_LIBTSAN_LIBS,
6927         STATIC_LIBLSAN_LIBS, STATIC_LIBUBSAN_LIBS): Handle -static like
6928         -static-lib*san.
6929
6930         PR debug/84252
6931         * var-tracking.c (vt_add_function_parameter): Punt for non-onepart
6932         PARALLEL incoming that failed vt_get_decl_and_offset check.
6933
6934         PR middle-end/84237
6935         * output.h (bss_initializer_p): Add NAMED argument, defaulted to false.
6936         * varasm.c (bss_initializer_p): Add NAMED argument, if true, ignore
6937         TREE_READONLY bit.
6938         (get_variable_section): For decls in named .bss* sections pass true as
6939         second argument to bss_initializer_p.
6940
6941 2018-02-09  Marek Polacek  <polacek@redhat.com>
6942             Jakub Jelinek  <jakub@redhat.com>
6943
6944         PR c++/83659
6945         * fold-const.c (fold_indirect_ref_1): Use VECTOR_TYPE_P macro.
6946         Formatting fixes.  Verify first that tree_fits_poly_int64_p (op01).
6947         Sync some changes from cxx_fold_indirect_ref.
6948
6949 2018-02-09  Alexandre Oliva  <aoliva@redhat.com>
6950
6951         * cfgexpand.c (expand_gimple_basic_block): Handle inline entry
6952         markers.
6953         * dwarf2out.c (dwarf2_debug_hooks): Enable inline_entry hook.
6954         (BLOCK_INLINE_ENTRY_LABEL): New.
6955         (dwarf2out_var_location): Disregard inline entry markers.
6956         (inline_entry_data): New struct.
6957         (inline_entry_data_hasher): New hashtable type.
6958         (inline_entry_data_hasher::hash): New.
6959         (inline_entry_data_hasher::equal): New.
6960         (inline_entry_data_table): New variable.
6961         (add_high_low_attributes): Add DW_AT_entry_pc and
6962         DW_AT_GNU_entry_view attributes if a pending entry is found
6963         in inline_entry_data_table.  Add old entry_pc attribute only
6964         if debug nonbinding markers are disabled.
6965         (gen_inlined_subroutine_die): Set BLOCK_DIE if nonbinding
6966         markers are enabled.
6967         (block_within_block_p, dwarf2out_inline_entry): New.
6968         (dwarf2out_finish): Check that no entries remained in
6969         inline_entry_data_table.
6970         * final.c (reemit_insn_block_notes): Handle inline entry notes.
6971         (final_scan_insn, notice_source_line): Likewise.
6972         (rest_of_clean_state): Skip inline entry markers.
6973         * gimple-pretty-print.c (dump_gimple_debug): Handle inline entry
6974         markers.
6975         * gimple.c (gimple_build_debug_inline_entry): New.
6976         * gimple.h (enum gimple_debug_subcode): Add
6977         GIMPLE_DEBUG_INLINE_ENTRY.
6978         (gimple_build_debug_inline_entry): Declare.
6979         (gimple_debug_inline_entry_p): New.
6980         (gimple_debug_nonbind_marker_p): Adjust.
6981         * insn-notes.def (INLINE_ENTRY): New.
6982         * print-rtl.c (rtx_writer::print_rtx_operand_code_0): Handle
6983         inline entry marker notes.
6984         (print_insn): Likewise.
6985         * rtl.h (NOTE_MARKER_P): Add INLINE_ENTRY support.
6986         (INSN_DEBUG_MARKER_KIND): Likewise.
6987         (GEN_RTX_DEBUG_MARKER_INLINE_ENTRY_PAT): New.
6988         * tree-inline.c (expand_call_inline): Build and insert
6989         debug_inline_entry stmt.
6990         * tree-ssa-live.c (remove_unused_scope_block_p): Preserve
6991         inline entry blocks early, if nonbind markers are enabled.
6992         (dump_scope_block): Dump fragment info.
6993         * var-tracking.c (reemit_marker_as_note): Handle inline entry note.
6994         * doc/gimple.texi (gimple_debug_inline_entry_p): New.
6995         (gimple_build_debug_inline_entry): New.
6996         * doc/invoke.texi (gstatement-frontiers, gno-statement-frontiers):
6997         Enable/disable inline entry points too.
6998         * doc/rtl.texi (NOTE_INSN_INLINE_ENTRY): New.
6999         (DEBUG_INSN): Describe inline entry markers.
7000
7001         * common.opt (gvariable-location-views): New.
7002         (gvariable-location-views=incompat5): New.
7003         * config.in: Rebuilt.
7004         * configure: Rebuilt.
7005         * configure.ac: Test assembler for view support.
7006         * dwarf2asm.c (dw2_asm_output_symname_uleb128): New.
7007         * dwarf2asm.h (dw2_asm_output_symname_uleb128): Declare.
7008         * dwarf2out.c (var_loc_view): New typedef.
7009         (struct dw_loc_list_struct): Add vl_symbol, vbegin, vend.
7010         (dwarf2out_locviews_in_attribute): New.
7011         (dwarf2out_locviews_in_loclist): New.
7012         (dw_val_equal_p): Compare val_view_list of dw_val_class_view_lists.
7013         (enum dw_line_info_opcode): Add LI_adv_address.
7014         (struct dw_line_info_table): Add view.
7015         (RESET_NEXT_VIEW, RESETTING_VIEW_P): New macros.
7016         (DWARF2_ASM_VIEW_DEBUG_INFO): Define default.
7017         (zero_view_p): New variable.
7018         (ZERO_VIEW_P): New macro.
7019         (output_asm_line_debug_info): New.
7020         (struct var_loc_node): Add view.
7021         (add_AT_view_list, AT_loc_list): New.
7022         (add_var_loc_to_decl): Add view param.  Test it against last.
7023         (new_loc_list): Add view params.  Record them.
7024         (AT_loc_list_ptr): Handle loc and view lists.
7025         (view_list_to_loc_list_val_node): New.
7026         (print_dw_val): Handle dw_val_class_view_list.
7027         (size_of_die): Likewise.
7028         (value_format): Likewise.
7029         (loc_list_has_views): New.
7030         (gen_llsym): Set vl_symbol too.
7031         (maybe_gen_llsym, skip_loc_list_entry): New.
7032         (dwarf2out_maybe_output_loclist_view_pair): New.
7033         (output_loc_list): Output view list or entries too.
7034         (output_view_list_offset): New.
7035         (output_die): Handle dw_val_class_view_list.
7036         (output_dwarf_version): New.
7037         (output_compilation_unit_header): Use it.
7038         (output_skeleton_debug_sections): Likewise.
7039         (output_rnglists, output_line_info): Likewise.
7040         (output_pubnames, output_aranges): Update version comments.
7041         (output_one_line_info_table): Output view numbers in asm comments.
7042         (dw_loc_list): Determine current endview, pass it to new_loc_list.
7043         Call maybe_gen_llsym.
7044         (loc_list_from_tree_1): Adjust.
7045         (add_AT_location_description): Create view list attribute if
7046         needed, check it's absent otherwise.
7047         (convert_cfa_to_fb_loc_list): Adjust.
7048         (maybe_emit_file): Call output_asm_line_debug_info for test.
7049         (dwarf2out_var_location): Reset views as needed.  Precompute
7050         add_var_loc_to_decl args.  Call get_attr_min_length only if we have the
7051         attribute.  Set view.
7052         (new_line_info_table): Reset next view.
7053         (set_cur_line_info_table): Call output_asm_line_debug_info for test.
7054         (dwarf2out_source_line): Likewise.  Output view resets and labels to
7055         the assembler, or select appropriate line info opcodes.
7056         (prune_unused_types_walk_attribs): Handle dw_val_class_view_list.
7057         (optimize_string_length): Catch it.  Adjust.
7058         (resolve_addr): Copy vl_symbol along with ll_symbol.  Handle
7059         dw_val_class_view_list, and remove it if no longer needed.
7060         (hash_loc_list): Hash view numbers.
7061         (loc_list_hasher::equal): Compare them.
7062         (optimize_location_lists): Check whether a view list symbol is
7063         needed, and whether the locview attribute is present, and
7064         whether they match.  Remove the locview attribute if no longer
7065         needed.
7066         (index_location_lists): Call skip_loc_list_entry for test.
7067         (dwarf2out_finish): Call output_asm_line_debug_info for test.
7068         Use output_dwarf_version.
7069         * dwarf2out.h (enum dw_val_class): Add dw_val_class_view_list.
7070         (struct dw_val_node): Add val_view_list.
7071         * final.c (SEEN_NEXT_VIEW): New.
7072         (set_next_view_needed): New.
7073         (clear_next_view_needed): New.
7074         (maybe_output_next_view): New.
7075         (final_start_function): Rename to...
7076         (final_start_function_1): ... this.  Take pointer to FIRST,
7077         add SEEN parameter.  Emit param bindings in the initial view.
7078         (final_start_function): Reintroduce SEEN-less interface.
7079         (final): Rename to...
7080         (final_1): ... this.  Take SEEN parameter.  Output final pending
7081         next view at the end.
7082         (final): Reintroduce seen-less interface.
7083         (final_scan_insn): Output pending next view before switching
7084         sections or ending a block.  Mark the next view as needed when
7085         outputting variable locations.  Notify debug backend of section
7086         changes, and of location view changes.
7087         (rest_of_handle_final): Adjust.
7088         * toplev.c (process_options): Autodetect value for debug variable
7089         location views option.  Warn on incompat5 without -gdwarf-5.
7090         * doc/invoke.texi (gvariable-location-views): New.
7091         (gvariable-location-views=incompat5): New.
7092         (gno-variable-location-views): New.
7093
7094 2018-02-08  David Malcolm  <dmalcolm@redhat.com>
7095
7096         PR tree-optimization/84136
7097         * tree-cfg.c (find_taken_edge_computed_goto): Remove assertion
7098         that the result of find_edge is non-NULL.
7099
7100 2018-02-08  Sergey Shalnov  <sergey.shalnov@intel.com>
7101
7102         PR target/83008
7103         * config/i386/x86-tune-costs.h (skylake_cost): Fix cost of
7104         storing integer register in SImode.  Fix cost of 256 and 512
7105         byte aligned SSE register store.
7106
7107 2018-02-08  Sergey Shalnov  <sergey.shalnov@intel.com>
7108
7109         * config/i386/i386.c (ix86_multiplication_cost): Fix
7110         multiplication cost for TARGET_AVX512DQ.
7111
7112 2018-02-08  Marek Polacek  <polacek@redhat.com>
7113
7114         PR tree-optimization/84238
7115         * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Verify the result of
7116         get_range_strlen.
7117
7118 2018-02-08  Richard Sandiford  <richard.sandiford@linaro.org>
7119
7120         PR tree-optimization/84265
7121         * tree-vect-stmts.c (vectorizable_store): Don't treat
7122         VMAT_CONTIGUOUS accesses as grouped.
7123         (vectorizable_load): Likewise.
7124
7125 2018-02-08  Richard Sandiford  <richard.sandiford@linaro.org>
7126
7127         PR tree-optimization/81635
7128         * wide-int.h (wi::round_down_for_mask, wi::round_up_for_mask): Declare.
7129         * wide-int.cc (wi::round_down_for_mask, wi::round_up_for_mask)
7130         (test_round_for_mask): New functions.
7131         (wide_int_cc_tests): Call test_round_for_mask.
7132         * tree-vrp.h (intersect_range_with_nonzero_bits): Declare.
7133         * tree-vrp.c (intersect_range_with_nonzero_bits): New function.
7134         * tree-data-ref.c (split_constant_offset_1): Use it to refine the
7135         range returned by get_range_info.
7136
7137 2018-02-08  Jan Hubicka  <hubicka@ucw.cz>
7138
7139         PR ipa/81360
7140         * cgraph.h (symtab_node::output_to_lto_symbol_table_p): Declare
7141         * symtab.c: Include builtins.h
7142         (symtab_node::output_to_lto_symbol_table_p): Move here
7143         from lto-streamer-out.c:output_symbol_p.
7144         * lto-streamer-out.c (write_symbol): Turn early exit to assert.
7145         (output_symbol_p): Move all logic to symtab.c
7146         (produce_symtab): Update.
7147
7148 2018-02-08  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>
7149
7150         * config/s390/s390-opts.h (enum indirect_branch): Define.
7151         * config/s390/s390-protos.h (s390_return_addr_from_memory)
7152         (s390_indirect_branch_via_thunk)
7153         (s390_indirect_branch_via_inline_thunk): Add function prototypes.
7154         (enum s390_indirect_branch_type): Define.
7155         * config/s390/s390.c (struct s390_frame_layout, struct
7156         machine_function): Remove.
7157         (indirect_branch_prez10thunk_mask, indirect_branch_z10thunk_mask)
7158         (indirect_branch_table_label_no, indirect_branch_table_name):
7159         Define variables.
7160         (INDIRECT_BRANCH_NUM_OPTIONS): Define macro.
7161         (enum s390_indirect_branch_option): Define.
7162         (s390_return_addr_from_memory): New function.
7163         (s390_handle_string_attribute): New function.
7164         (s390_attribute_table): Add new attribute handler.
7165         (s390_execute_label): Handle UNSPEC_EXECUTE_JUMP patterns.
7166         (s390_indirect_branch_via_thunk): New function.
7167         (s390_indirect_branch_via_inline_thunk): New function.
7168         (s390_function_ok_for_sibcall): When jumping via thunk disallow
7169         sibling call optimization for non z10 compiles.
7170         (s390_emit_call): Force indirect branch target to be a single
7171         register.  Add r1 clobber for non-z10 compiles.
7172         (s390_emit_epilogue): Emit return jump via return_use expander.
7173         (s390_reorg): Handle JUMP_INSNs as execute targets.
7174         (s390_option_override_internal): Perform validity checks for the
7175         new command line options.
7176         (s390_indirect_branch_attrvalue): New function.
7177         (s390_indirect_branch_settings): New function.
7178         (s390_set_current_function): Invoke s390_indirect_branch_settings.
7179         (s390_output_indirect_thunk_function):  New function.
7180         (s390_code_end): Implement target hook.
7181         (s390_case_values_threshold): Implement target hook.
7182         (TARGET_ASM_CODE_END, TARGET_CASE_VALUES_THRESHOLD): Define target
7183         macros.
7184         * config/s390/s390.h (struct s390_frame_layout)
7185         (struct machine_function): Move here from s390.c.
7186         (TARGET_INDIRECT_BRANCH_NOBP_RET)
7187         (TARGET_INDIRECT_BRANCH_NOBP_JUMP)
7188         (TARGET_INDIRECT_BRANCH_NOBP_JUMP_THUNK)
7189         (TARGET_INDIRECT_BRANCH_NOBP_JUMP_INLINE_THUNK)
7190         (TARGET_INDIRECT_BRANCH_NOBP_CALL)
7191         (TARGET_DEFAULT_INDIRECT_BRANCH_TABLE)
7192         (TARGET_INDIRECT_BRANCH_THUNK_NAME_EXRL)
7193         (TARGET_INDIRECT_BRANCH_THUNK_NAME_EX)
7194         (TARGET_INDIRECT_BRANCH_TABLE): Define macros.
7195         * config/s390/s390.md (UNSPEC_EXECUTE_JUMP)
7196         (INDIRECT_BRANCH_THUNK_REGNUM): Define constants.
7197         (mnemonic attribute): Add values which aren't recognized
7198         automatically.
7199         ("*cjump_long", "*icjump_long", "*basr", "*basr_r"): Disable
7200         pattern for branch conversion.  Fix mnemonic attribute.
7201         ("*c<code>", "*sibcall_br", "*sibcall_value_br", "*return"): Emit
7202         indirect branch via thunk if requested.
7203         ("indirect_jump", "<code>"): Expand patterns for branch conversion.
7204         ("*indirect_jump"): Disable for branch conversion using out of
7205         line thunks.
7206         ("indirect_jump_via_thunk<mode>_z10")
7207         ("indirect_jump_via_thunk<mode>")
7208         ("indirect_jump_via_inlinethunk<mode>_z10")
7209         ("indirect_jump_via_inlinethunk<mode>", "*casesi_jump")
7210         ("casesi_jump_via_thunk<mode>_z10", "casesi_jump_via_thunk<mode>")
7211         ("casesi_jump_via_inlinethunk<mode>_z10")
7212         ("casesi_jump_via_inlinethunk<mode>", "*basr_via_thunk<mode>_z10")
7213         ("*basr_via_thunk<mode>", "*basr_r_via_thunk_z10")
7214         ("*basr_r_via_thunk", "return<mode>_prez10"): New pattern.
7215         ("*indirect2_jump"): Disable for branch conversion.
7216         ("casesi_jump"): Turn into expander and expand patterns for branch
7217         conversion.
7218         ("return_use"): New expander.
7219         ("*return"): Emit return via thunk and rename it to ...
7220         ("*return<mode>"): ... this one.
7221         * config/s390/s390.opt: Add new options and and enum for the
7222         option values.
7223
7224 2018-02-08  Richard Sandiford  <richard.sandiford@linaro.org>
7225
7226         * lra-constraints.c (match_reload): Unconditionally use
7227         gen_lowpart_SUBREG, rather than selecting between that
7228         and equivalent gen_rtx_SUBREG code.
7229
7230 2018-02-08  Richard Biener  <rguenther@suse.de>
7231
7232         PR tree-optimization/84233
7233         * tree-ssa-phiprop.c (propagate_with_phi): Use separate
7234         changed flag instead of boguously re-using phi_inserted.
7235
7236 2018-02-08  Martin Jambor  <mjambor@suse.cz>
7237
7238         * hsa-gen.c (get_symbol_for_decl): Set program allocation for
7239         static local variables.
7240
7241 2018-02-08  Richard Biener  <rguenther@suse.de>
7242
7243         PR tree-optimization/84278
7244         * tree-vect-stmts.c (vectorizable_store): When looking for
7245         smaller vector types to perform grouped strided loads/stores
7246         make sure the mode is supported by the target.
7247         (vectorizable_load): Likewise.
7248
7249 2018-02-08  Wilco Dijkstra  <wdijkstr@arm.com>
7250
7251         * config/aarch64/aarch64.c (aarch64_components_for_bb):
7252         Increase LDP/STP opportunities by adding adjacent callee-saves.
7253
7254 2018-02-08  Wilco Dijkstra  <wdijkstr@arm.com>
7255
7256         PR rtl-optimization/84068
7257         PR rtl-optimization/83459
7258         * haifa-sched.c (rank_for_schedule): Fix SCHED_PRESSURE_MODEL sorting.
7259
7260 2018-02-08  Aldy Hernandez  <aldyh@redhat.com>
7261
7262         PR tree-optimization/84224
7263         * gimple-ssa-warn-alloca.c (pass_walloca::execute): Remove assert.
7264         * calls.c (gimple_alloca_call_p): Only return TRUE when we have
7265         non-zero arguments.
7266
7267 2018-02-07  Iain Sandoe  <iain@codesourcery.com>
7268
7269         PR target/84113
7270         * config/rs6000/altivec.md (*restore_world): Remove LR use.
7271         * config/rs6000/predicates.md (restore_world_operation): Adjust op
7272         count, remove one USE.
7273
7274 2018-02-07  Michael Meissner  <meissner@linux.vnet.ibm.com>
7275
7276         * doc/install.texi (Configuration): Document the
7277         --with-long-double-format={ibm,ieee} PowerPC configuration
7278         options.
7279
7280         PR target/84154
7281         * config/rs6000/rs6000.md (fix_trunc<SFDF:mode><QHI:mode>2):
7282         Convert from define_expand to be define_insn_and_split.  Rework
7283         float/double/_Float128 conversions to QI/HI/SImode to work with
7284         both ISA 2.07 (power8) or ISA 3.0 (power9).  Fix regression where
7285         conversions to QI/HImode types did a store and then a load to
7286         truncate the value.  For conversions to VSX registers, don't split
7287         the insn, instead emit the code directly.  Use the code iterator
7288         any_fix to combine signed and unsigned conversions.
7289         (fix<uns>_trunc<SFDF:mode>si2_p8): Likewise.
7290         (fixuns_trunc<SFDF:mode><QHI:mode>2): Likewise.
7291         (fix_trunc<IEEE128:mode><QHI:mode>2): Likewise.
7292         (fix<uns>_trunc<SFDF:mode><QHI:mode>2): Likewise.
7293         (fix_<mode>di2_hw): Likewise.
7294         (fixuns_<mode>di2_hw): Likewise.
7295         (fix_<mode>si2_hw): Likewise.
7296         (fixuns_<mode>si2_hw): Likewise.
7297         (fix<uns>_<IEEE128:mode><SDI:mode>2_hw): Likewise.
7298         (fix<uns>_trunc<IEEE128:mode><QHI:mode>2): Likewise.
7299         (fctiw<u>z_<mode>_smallint): Rename fctiw<u>z_<mode>_smallint to
7300         fix<uns>_trunc<SFDF:mode>si2_p8.
7301         (fix_trunc<SFDF:mode><QHI:mode>2_internal): Delete, no longer
7302         used.
7303         (fixuns_trunc<SFDF:mode><QHI:mode>2_internal): Likewise.
7304         (fix<uns>_<mode>_mem): Likewise.
7305         (fctiw<u>z_<mode>_mem): Likewise.
7306         (fix<uns>_<mode>_mem): Likewise.
7307         (fix<uns>_trunc<SFDF:mode><QHSI:mode>2_mem): On ISA 3.0, prevent
7308         the register allocator from doing a direct move to the GPRs to do
7309         a store, and instead use the ISA 3.0 store byte/half-word from
7310         vector register instruction.  For IEEE 128-bit floating point,
7311         also optimize stores of 32-bit ints.
7312         (fix<uns>_trunc<IEEE128:mode><QHSI:mode>2_mem): Likewise.
7313
7314 2018-02-07  Alan Hayward  <alan.hayward@arm.com>
7315
7316         * genextract.c (push_pathstr_operand): New function to support
7317         [a-zA-Z].
7318         (walk_rtx): Call push_pathstr_operand.
7319         (print_path): Support [a-zA-Z].
7320
7321 2018-02-07  Richard Biener  <rguenther@suse.de>
7322
7323         PR tree-optimization/84037
7324         * tree-vectorizer.h (struct _loop_vec_info): Add ivexpr_map member.
7325         (cse_and_gimplify_to_preheader): Declare.
7326         (vect_get_place_in_interleaving_chain): Likewise.
7327         * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
7328         ivexpr_map.
7329         (_loop_vec_info::~_loop_vec_info): Delete it.
7330         (cse_and_gimplify_to_preheader): New function.
7331         * tree-vect-slp.c (vect_get_place_in_interleaving_chain): Export.
7332         * tree-vect-stmts.c (vectorizable_store): CSE base and steps.
7333         (vectorizable_load): Likewise.  For grouped stores always base
7334         the IV on the first element.
7335         * tree-vect-loop-manip.c (vect_loop_versioning): Unshare versioning
7336         condition before gimplifying.
7337
7338 2018-02-07  Jakub Jelinek  <jakub@redhat.com>
7339
7340         * tree-eh.c (operation_could_trap_helper_p): Ignore honor_trapv for
7341         *DIV_EXPR and *MOD_EXPR.
7342
7343 2018-02-07  H.J. Lu  <hongjiu.lu@intel.com>
7344
7345         PR target/84248
7346         * config/i386/i386.c (ix86_option_override_internal): Mask out
7347         the CF_SET bit when checking -fcf-protection.
7348
7349 2018-02-07  Tom de Vries  <tom@codesourcery.com>
7350
7351         PR libgomp/84217
7352         * omp-expand.c (expand_oacc_collapse_init): Ensure diff_type is large
7353         enough.
7354
7355 2018-02-07  Richard Biener  <rguenther@suse.de>
7356
7357         PR tree-optimization/84204
7358         * tree-chrec.c (chrec_fold_plus_1): Remove size limiting in
7359         this place.
7360
7361         PR tree-optimization/84205
7362         * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Also
7363         special-case isl_ast_op_zdiv_r.
7364
7365         PR tree-optimization/84223
7366         * graphite-scop-detection.c (gather_bbs::before_dom_children):
7367         Only add conditions from within the region.
7368         (gather_bbs::after_dom_children): Adjust.
7369
7370 2018-02-07  Georg-Johann Lay  <avr@gjlay.de>
7371
7372         PR target/84209
7373         * config/avr/avr.h (GENERAL_REGNO_P, GENERAL_REG_P): New macros.
7374         * config/avr/avr.md: Only post-reload split REG-REG moves if
7375         either register is GENERAL_REG_P.
7376
7377 2018-02-07  Jakub Jelinek  <jakub@redhat.com>
7378
7379         PR tree-optimization/84235
7380         * tree-ssa-scopedtables.c
7381         (avail_exprs_stack::simplify_binary_operation): Fir MINUS_EXPR, punt
7382         if the subtraction is performed in floating point type where NaNs are
7383         honored.  For *DIV_EXPR, punt for ALL_FRACT_MODE_Ps where we can't
7384         build 1.  Formatting fix.
7385
7386 2018-02-06  Jakub Jelinek  <jakub@redhat.com>
7387
7388         PR target/84146
7389         * config/i386/i386.c (rest_of_insert_endbranch): Only skip
7390         NOTE_INSN_CALL_ARG_LOCATION after a call, not anything else,
7391         and skip it regardless of bb boundaries.  Use CALL_P macro,
7392         don't test INSN_P (insn) together with CALL_P or JUMP_P check
7393         unnecessarily, formatting fix.
7394
7395 2018-02-06  Michael Collison  <michael.collison@arm.com>
7396
7397         * config/arm/thumb2.md:
7398         (*thumb2_mov_negscc): Split only if TARGET_THUMB2 && !arm_restrict_it.
7399         (*thumb_mov_notscc): Ditto.
7400
7401 2018-02-06  Michael Meissner  <meissner@linux.vnet.ibm.com>
7402
7403         PR target/84154
7404         * config/rs6000/rs6000.md (su code attribute): Use "u" for
7405         unsigned_fix, not "s".
7406
7407 2018-02-06  Rainer Orth  <ro@CeBiTec.Uni-Bielefeld.DE>
7408
7409         * configure.ac (gcc_fn_eh_frame_ro): New function.
7410         (gcc_cv_as_cfi_directive): Check both 32 and 64-bit assembler for
7411         correct .eh_frame permissions.
7412         * configure: Regenerate.
7413
7414 2018-02-06  Andrew Jenner  <andrew@codeourcery.com>
7415
7416         * doc/invoke.texi: Add section for the PowerPC SPE backend. Remove
7417         irrelevant options.
7418
7419 2018-02-06  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
7420
7421         * config/rs6000/rs6000.c (rs6000_option_override_internal):
7422         Display warning message for -mno-speculate-indirect-jumps.
7423
7424 2018-02-06  Andrew Jenner  <andrew@codesourcery.com>
7425
7426         * config/powerpcspe/powerpcspe.opt: (msimple-fpu, mfpu) Add
7427         Undocumented.
7428         * config/powerpcspe/sysv4.opt (mbit-align): Likewise.
7429
7430 2018-02-06  Aldy Hernandez  <aldyh@redhat.com>
7431
7432         PR tree-optimization/84225
7433         * tree-eh.c (find_trapping_overflow): Only call
7434         operation_no_trapping_overflow when ANY_INTEGRAL_TYPE_P.
7435
7436 2018-02-06  Igor Tsimbalist  <igor.v.tsimbalist@intel.com>
7437
7438         PR target/84145
7439         * config/i386/i386.c: Reimplement the check of possible options
7440         -mibt/-mshstk conbination. Change error messages.
7441         * doc/invoke.texi: Fix a typo: remove extra '='.
7442
7443 2018-02-06  Marek Polacek  <polacek@redhat.com>
7444
7445         PR tree-optimization/84228
7446         * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Skip debug statements.
7447
7448 2018-02-06  Tamar Christina  <tamar.christina@arm.com>
7449
7450         PR target/82641
7451         * config/arm/arm.c (arm_print_asm_arch_directives): Record already
7452         emitted arch directives.
7453         * config/arm/arm-c.c (arm_cpu_builtins): Undefine __ARM_ARCH and
7454         __ARM_FEATURE_COPROC before changing architectures.
7455
7456 2018-02-06  Richard Biener  <rguenther@suse.de>
7457
7458         * config/i386/i386.c (print_reg): Fix typo.
7459         (ix86_loop_unroll_adjust): Do not unroll beyond the original nunroll.
7460
7461 2018-02-06  Eric Botcazou  <ebotcazou@adacore.com>
7462
7463         * configure: Regenerate.
7464
7465 2018-02-05  Martin Sebor  <msebor@redhat.com>
7466
7467         PR tree-optimization/83369
7468         * tree-ssa-ccp.c (pass_post_ipa_warn::execute): Use %G to print
7469         inlining context.
7470
7471 2018-02-05  Martin Liska  <mliska@suse.cz>
7472
7473         * doc/invoke.texi: Cherry-pick upstream r323995.
7474
7475 2018-02-05  Richard Sandiford  <richard.sandiford@linaro.org>
7476
7477         * ira.c (ira_init_register_move_cost): Adjust comment.
7478
7479 2018-02-05  Martin Liska  <mliska@suse.cz>
7480
7481         PR gcov-profile/84137
7482         * doc/gcov.texi: Fix typo in documentation.
7483
7484 2018-02-05  Martin Liska  <mliska@suse.cz>
7485
7486         PR gcov-profile/83879
7487         * doc/gcov.texi: Document necessity of --dynamic-list-data when
7488         using dlopen functionality.
7489
7490 2018-02-05  Olga Makhotina  <olga.makhotina@intel.com>
7491
7492         * config/i386/avx512dqintrin.h (_mm_mask_range_sd, _mm_maskz_range_sd,
7493         _mm_mask_range_round_sd, _mm_maskz_range_round_sd, _mm_mask_range_ss,
7494         _mm_maskz_range_ss, _mm_mask_range_round_ss,
7495         _mm_maskz_range_round_ss): New intrinsics.
7496         (__builtin_ia32_rangesd128_round)
7497         (__builtin_ia32_rangess128_round): Remove.
7498         (__builtin_ia32_rangesd128_mask_round,
7499         __builtin_ia32_rangess128_mask_round): New builtins.
7500         * config/i386/i386-builtin.def (__builtin_ia32_rangesd128_round,
7501         __builtin_ia32_rangess128_round): Remove.
7502         (__builtin_ia32_rangesd128_mask_round,
7503         __builtin_ia32_rangess128_mask_round): New builtins.
7504         * config/i386/sse.md (ranges<mode><round_saeonly_name>): Renamed to ...
7505         (ranges<mode><mask_scalar_name><round_saeonly_scalar_name>): ... this.
7506         ((match_operand:VF_128 2 "<round_saeonly_nimm_predicate>"
7507         "<round_saeonly_constraint>")): Changed to ...
7508         ((match_operand:VF_128 2 "<round_saeonly_scalar_nimm_predicate>"
7509         "<round_saeonly_scalar_constraint>")): ... this.
7510         ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_op4>%2, %1, %0|
7511         %0, %1, %2<round_saeonly_op4>, %3}"): Changed to ...
7512         ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_scalar_mask_op4>%2,
7513         %1, %0<mask_scalar_operand4>|%0<mask_scalar_operand4>, %1,
7514         %2<round_saeonly_scalar_mask_op4>, %3}"): ... this.
7515
7516 2018-02-02  Andrew Jenner  <andrew@codesourcery.com>
7517
7518         * config/powerpcspe/powerpcspe.opt: Add Undocumented to irrelevant
7519         options.
7520         * config/powerpcspe/powerpcspe-tables.opt (rs6000_cpu_opt_value):
7521         Remove all values except native, 8540 and 8548.
7522
7523 2018-02-02  H.J. Lu  <hongjiu.lu@intel.com>
7524
7525         * config/i386/i386.c (ix86_output_function_return): Pass
7526         INVALID_REGNUM, instead of -1, as invalid register number to
7527         indirect_thunk_name and output_indirect_thunk.
7528
7529 2018-02-02  Julia Koval  <julia.koval@intel.com>
7530
7531         * config.gcc: Add -march=icelake.
7532         * config/i386/driver-i386.c (host_detect_local_cpu): Detect icelake.
7533         * config/i386/i386-c.c (ix86_target_macros_internal): Handle icelake.
7534         * config/i386/i386.c (processor_costs): Add m_ICELAKE.
7535         (PTA_ICELAKE, PTA_AVX512VNNI, PTA_GFNI, PTA_VAES, PTA_AVX512VBMI2,
7536         PTA_VPCLMULQDQ, PTA_RDPID, PTA_AVX512BITALG): New.
7537         (processor_target_table): Add icelake.
7538         (ix86_option_override_internal): Handle new PTAs.
7539         (get_builtin_code_for_version): Handle icelake.
7540         (M_INTEL_COREI7_ICELAKE): New.
7541         (fold_builtin_cpu): Handle icelake.
7542         * config/i386/i386.h (TARGET_ICELAKE, PROCESSOR_ICELAKE): New.
7543         * doc/invoke.texi: Add -march=icelake.
7544
7545 2018-02-02  Julia Koval  <julia.koval@intel.com>
7546
7547         * config/i386/i386.c (ix86_option_override_internal): Change flags type
7548         to wide_int_bitmask.
7549         * wide-int-bitmask.h: New.
7550
7551 2018-02-02  Igor Tsimbalist  <igor.v.tsimbalist@intel.com>
7552
7553         PR target/84066
7554         * config/i386/i386.md: Replace Pmode with word_mode in
7555         builtin_setjmp_setup and builtin_longjmp to support x32.
7556
7557 2018-02-01  Peter Bergner  <bergner@vnet.ibm.com>
7558
7559         PR target/56010
7560         PR target/83743
7561         * config/rs6000/driver-rs6000.c: #include "diagnostic.h".
7562         #include "opts.h".
7563         (rs6000_supported_cpu_names): New static variable.
7564         (linux_cpu_translation_table): Likewise.
7565         (elf_platform) <cpu>: Define new static variable and use it.
7566         Translate kernel AT_PLATFORM name to canonical name if needed.
7567         Error if platform name is unknown.
7568
7569 2018-02-01  Aldy Hernandez  <aldyh@redhat.com>
7570
7571         PR target/84089
7572         * config/pa/predicates.md (base14_operand): Handle E_VOIDmode.
7573
7574 2018-02-01  Jeff Law  <law@redhat.com>
7575
7576         PR target/84128
7577         * config/i386/i386.c (release_scratch_register_on_entry): Add new
7578         OFFSET and RELEASE_VIA_POP arguments.  Use SP+OFFSET to restore
7579         the scratch if RELEASE_VIA_POP is false.
7580         (ix86_adjust_stack_and_probe_stack_clash): Un-constify SIZE.
7581         If we have to save a temporary register, decrement SIZE appropriately.
7582         Pass new arguments to release_scratch_register_on_entry.
7583         (ix86_adjust_stack_and_probe): Likewise.
7584         (ix86_emit_probe_stack_range): Pass new arguments to
7585         release_scratch_register_on_entry.
7586
7587 2018-02-01  Uros Bizjak  <ubizjak@gmail.com>
7588
7589         PR rtl-optimization/84157
7590         * combine.c (change_zero_ext): Use REG_P predicate in
7591         front of HARD_REGISTER_P predicate.
7592
7593 2018-02-01  Georg-Johann Lay  <avr@gjlay.de>
7594
7595         * config/avr/avr.c (avr_option_override): Move disabling of
7596         -fdelete-null-pointer-checks to...
7597         * common/config/avr/avr-common.c (avr_option_optimization_table):
7598         ...here.
7599
7600 2018-02-01  Richard Sandiford  <richard.sandiford@linaro.org>
7601
7602         PR tree-optimization/81635
7603         * tree-data-ref.c (split_constant_offset_1): For types that
7604         wrap on overflow, try to use range info to prove that wrapping
7605         cannot occur.
7606
7607 2018-02-01  Renlin Li  <renlin.li@arm.com>
7608
7609         PR target/83370
7610         * config/aarch64/aarch64.c (aarch64_class_max_nregs): Handle
7611         TAILCALL_ADDR_REGS.
7612         (aarch64_register_move_cost): Likewise.
7613         * config/aarch64/aarch64.h (reg_class): Rename CALLER_SAVE_REGS to
7614         TAILCALL_ADDR_REGS.
7615         (REG_CLASS_NAMES): Likewise.
7616         (REG_CLASS_CONTENTS): Rename CALLER_SAVE_REGS to
7617         TAILCALL_ADDR_REGS. Remove IP registers.
7618         * config/aarch64/aarch64.md (Ucs): Update register constraint.
7619
7620 2018-02-01  Richard Biener  <rguenther@suse.de>
7621
7622         * domwalk.h (dom_walker::dom_walker): Add additional constructor
7623         for specifying RPO order and allow NULL for that.
7624         * domwalk.c (dom_walker::dom_walker): Likewise.
7625         (dom_walker::walk): Handle NULL RPO order.
7626         * tree-into-ssa.c (rewrite_dom_walker): Do not walk dom children
7627         in RPO order.
7628         (rewrite_update_dom_walker): Likewise.
7629         (mark_def_dom_walker): Likewise.
7630
7631 2018-02-01  Richard Sandiford  <richard.sandiford@linaro.org>
7632
7633         * config/aarch64/aarch64-protos.h (aarch64_split_sve_subreg_move)
7634         (aarch64_maybe_expand_sve_subreg_move): Declare.
7635         * config/aarch64/aarch64.md (UNSPEC_REV_SUBREG): New unspec.
7636         * config/aarch64/predicates.md (aarch64_any_register_operand): New
7637         predicate.
7638         * config/aarch64/aarch64-sve.md (mov<mode>): Optimize subreg moves
7639         that are semantically a reverse operation.
7640         (*aarch64_sve_mov<mode>_subreg_be): New pattern.
7641         * config/aarch64/aarch64.c (aarch64_maybe_expand_sve_subreg_move):
7642         (aarch64_replace_reg_mode, aarch64_split_sve_subreg_move): New
7643         functions.
7644         (aarch64_can_change_mode_class): For big-endian, forbid changes
7645         between two SVE modes if they have different element sizes.
7646
7647 2018-02-01  Richard Sandiford  <richard.sandiford@linaro.org>
7648
7649         * config/aarch64/aarch64.c (aarch64_expand_sve_const_vector): Prefer
7650         the TImode handling for big-endian targets.
7651
7652 2018-02-01  Richard Sandiford  <richard.sandiford@linaro.org>
7653
7654         * config/aarch64/aarch64-sve.md (sve_ld1rq): Replace with...
7655         (*sve_ld1rq<Vesize>): ... this new pattern.  Handle all element sizes,
7656         not just bytes.
7657         * config/aarch64/aarch64.c (aarch64_expand_sve_widened_duplicate):
7658         Remove BSWAP handing for big-endian targets and use the form of
7659         LD1RQ appropariate for the mode.
7660
7661 2018-02-01  Richard Sandiford  <richard.sandiford@linaro.org>
7662
7663         * config/aarch64/aarch64.c (aarch64_simd_valid_immediate): Handle
7664         all CONST_VECTOR_DUPLICATE_P vectors, not just those with a single
7665         duplicated element.
7666
7667 2018-02-01  Richard Sandiford  <richard.sandiford@linaro.org>
7668
7669         PR tearget/83845
7670         * config/aarch64/aarch64.c (aarch64_secondary_reload): Tighten
7671         check for operands that need to go through aarch64_sve_reload_be.
7672
7673 2018-02-01  Jakub Jelinek  <jakub@redhat.com>
7674
7675         PR tree-optimization/81661
7676         PR tree-optimization/84117
7677         * tree-eh.h (rewrite_to_non_trapping_overflow): Declare.
7678         * tree-eh.c: Include gimplify.h.
7679         (find_trapping_overflow, replace_trapping_overflow,
7680         rewrite_to_non_trapping_overflow): New functions.
7681         * tree-vect-loop.c: Include tree-eh.h.
7682         (vect_get_loop_niters): Use rewrite_to_non_trapping_overflow.
7683         * tree-data-ref.c: Include tree-eh.h.
7684         (get_segment_min_max): Use rewrite_to_non_trapping_overflow.
7685
7686 2018-01-31  Uros Bizjak  <ubizjak@gmail.com>
7687
7688         PR rtl-optimization/84123
7689         * combine.c (change_zero_ext): Check if hard register satisfies
7690         can_change_dest_mode before calling gen_lowpart_SUBREG.
7691
7692 2018-01-31  Vladimir Makarov  <vmakarov@redhat.com>
7693
7694         PR target/82444
7695         * ira.c (ira_init_register_move_cost): Remove assert.
7696
7697 2018-01-31  Eric Botcazou  <ebotcazou@adacore.com>
7698
7699         PR rtl-optimization/84071
7700         * doc/tm.texi.in (WORD_REGISTER_OPERATIONS): Add explicit case.
7701         * doc/tm.texi: Regenerate.
7702
7703 2018-01-31  Richard Biener  <rguenther@suse.de>
7704
7705         PR tree-optimization/84132
7706         * tree-data-ref.c (analyze_miv_subscript): Properly
7707         check whether evolution_function_is_affine_multivariate_p
7708         before calling gcd_of_steps_may_divide_p.
7709
7710 2018-01-31  Julia Koval  <julia.koval@intel.com>
7711
7712         PR target/83618
7713         * config/i386/i386.c (ix86_expand_builtin): Handle IX86_BUILTIN_RDPID.
7714         * config/i386/i386.md (rdpid_rex64) New.
7715         (rdpid): Make 32bit only.
7716
7717 2018-01-29  Aldy Hernandez  <aldyh@redhat.com>
7718
7719         PR lto/84105
7720         * tree-pretty-print.c (dump_generic_node): Handle a TYPE_NAME with
7721         an IDENTIFIER_NODE for FUNCTION_TYPE's.
7722
7723 2018-01-31  Eric Botcazou  <ebotcazou@adacore.com>
7724
7725         Revert
7726         2018-01-12  Eric Botcazou  <ebotcazou@adacore.com>
7727
7728         * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
7729
7730 2018-01-31  Eric Botcazou  <ebotcazou@adacore.com>
7731
7732         PR rtl-optimization/84071
7733         * combine.c (record_dead_and_set_regs_1): Record the source unmodified
7734         for a paradoxical SUBREG on a WORD_REGISTER_OPERATIONS target.
7735
7736 2018-01-31  Claudiu Zissulescu  <claziss@synopsys.com>
7737
7738         * config/arc/arc.c (arc_handle_aux_attribute): New function.
7739         (arc_attribute_table): Add 'aux' attribute.
7740         (arc_in_small_data_p): Consider aux like variables.
7741         (arc_is_aux_reg_p): New function.
7742         (arc_asm_output_aligned_decl_local): Ignore 'aux' like variables.
7743         (arc_get_aux_arg): New function.
7744         (prepare_move_operands): Handle aux-register access.
7745         (arc_handle_aux_attribute): New function.
7746         * doc/extend.texi (ARC Variable attributes): Add subsection.
7747
7748 2018-01-31  Claudiu Zissulescu  <claziss@synopsys.com>
7749
7750         * config/arc/arc-protos.h (arc_is_uncached_mem_p): Function proto.
7751         * config/arc/arc.c (arc_handle_uncached_attribute): New function.
7752         (arc_attribute_table): Add 'uncached' attribute.
7753         (arc_print_operand): Print '.di' flag for uncached memory
7754         accesses.
7755         (arc_in_small_data_p): Do not consider for small data the uncached
7756         types.
7757         (arc_is_uncached_mem_p): New function.
7758         * config/arc/predicates.md (compact_store_memory_operand): Check
7759         for uncached memory accesses.
7760         (nonvol_nonimm_operand): Likewise.
7761         * gcc/doc/extend.texi (ARC Type Attribute): New subsection.
7762
7763 2018-01-31  Jakub Jelinek  <jakub@redhat.com>
7764
7765         PR c/84100
7766         * common.opt (falign-functions=, falign-jumps=, falign-labels=,
7767         falign-loops=): Add Optimization flag.
7768
7769 2018-01-30  Jeff Law  <law@redhat.com>
7770
7771         PR target/84064
7772         * i386.c (ix86_adjust_stack_and_probe_stack_clash): New argument
7773         INT_REGISTERS_SAVED.  Check it prior to calling
7774         get_scratch_register_on_entry.
7775         (ix86_adjust_stack_and_probe): Similarly.
7776         (ix86_emit_probe_stack_range): Similarly.
7777         (ix86_expand_prologue): Corresponding changes.
7778
7779 2018-01-30  Rainer Orth  <ro@CeBiTec.Uni-Bielefeld.DE>
7780
7781         PR target/40411
7782         * config/sol2.h (STARTFILE_ARCH_SPEC): Use -std=c*,
7783         -std=iso9899:199409 instead of -pedantic to select values-Xc.o.
7784
7785 2018-01-30  Vladimir Makarov  <vmakarov@redhat.com>
7786
7787         PR target/84112
7788         * lra-constraints.c (curr_insn_transform): Process AND in the
7789         address.
7790
7791 2018-01-30  Jakub Jelinek  <jakub@redhat.com>
7792
7793         PR rtl-optimization/83986
7794         * sched-deps.c (sched_analyze_insn): For frame related insns, add anti
7795         dependence against last_pending_memory_flush in addition to
7796         pending_jump_insns.
7797
7798 2018-01-30  Alexandre Oliva  <aoliva@redhat.com>
7799
7800         PR tree-optimization/81611
7801         * tree-ssa-dom.c (simple_iv_increment_p): Skip intervening
7802         copies.
7803
7804 2018-01-30  Aaron Sawdey  <acsawdey@linux.vnet.ibm.com>
7805
7806         PR target/83758
7807         * config/rs6000/rs6000.c (rs6000_internal_arg_pointer): Only return
7808         a reg rtx.
7809
7810 2018-01-30  Richard Biener  <rguenther@suse.de>
7811             Jakub Jelinek  <jakub@redhat.com>
7812
7813         PR tree-optimization/84111
7814         * tree-ssa-loop-ivcanon.c (tree_unroll_loops_completely_1): Skip
7815         inner loops added during recursion, as they don't have up-to-date
7816         SSA form.
7817
7818 2018-01-30  Jan Hubicka  <hubicka@ucw.cz>
7819
7820         PR ipa/81360
7821         * ipa-inline.c (can_inline_edge_p): Break out late tests to...
7822         (can_inline_edge_by_limits_p): ... here.
7823         (can_early_inline_edge_p, check_callers,
7824         update_caller_keys, update_callee_keys, recursive_inlining,
7825         add_new_edges_to_heap, speculation_useful_p,
7826         inline_small_functions,
7827         inline_small_functions, flatten_function,
7828         inline_to_all_callers_1): Update.
7829
7830 2018-01-30  Jan Hubicka  <hubicka@ucw.cz>
7831
7832         * profile-count.c (profile_count::combine_with_ipa_count): Handle
7833         zeros correctly.
7834
7835 2018-01-30  Richard Biener  <rguenther@suse.de>
7836
7837         PR tree-optimization/83008
7838         * tree-vect-slp.c (vect_analyze_slp_cost_1): Properly cost
7839         invariant and constant vector uses in stmts when they need
7840         more than one stmt.
7841
7842 2018-01-30  Rainer Orth  <ro@CeBiTec.Uni-Bielefeld.DE>
7843
7844         PR bootstrap/84017
7845         * configure.ac (gcc_cv_as_shf_merge): Disable on Solaris 10/x86.
7846         * configure: Regenerate.
7847
7848 2018-01-30  Richard Sandiford  <richard.sandiford@linaro.org>
7849
7850         * config/aarch64/aarch64-sve.md (*vec_extract<mode><Vel>_0): New
7851         pattern.
7852         (*vec_extract<mode><Vel>_v128): Require a nonzero lane number.
7853         Use gen_rtx_REG rather than gen_lowpart.
7854
7855 2018-01-30  Richard Sandiford  <richard.sandiford@linaro.org>
7856
7857         * lra-constraints.c (match_reload): Use subreg_lowpart_offset
7858         rather than 0 when creating partial subregs.
7859
7860 2018-01-30  Richard Sandiford  <richard.sandiford@linaro.org>
7861
7862         * vec-perm-indices.c (vec_perm_indices::series_p): Give examples
7863         of usage.
7864
7865 2018-01-29  Michael Meissner  <meissner@linux.vnet.ibm.com>
7866
7867         PR target/81550
7868         * config/rs6000/rs6000.c (rs6000_setup_reg_addr_masks): If DFmode
7869         and SFmode can go in Altivec registers (-mcpu=power7 for DFmode,
7870         -mcpu=power8 for SFmode) don't set the PRE_INCDEC or PRE_MODIFY
7871         flags.  This restores the settings used before the 2017-07-24.
7872         Turning off pre increment/decrement/modify allows IVOPTS to
7873         optimize DF/SF loops where the index is an int.
7874
7875 2018-01-29  Richard Biener  <rguenther@suse.de>
7876             Kelvin Nilsen  <kelvin@gcc.gnu.org>
7877
7878         PR bootstrap/80867
7879         * tree-vect-stmts.c (vectorizable_call): Don't call
7880         targetm.vectorize_builtin_md_vectorized_function if callee is
7881         NULL.
7882
7883 2018-01-22  Carl Love  <cel@us.ibm.com>
7884
7885         * doc/extend.tex: Fix typo in second arg in
7886         __builtin_bcdadd_{lt|eq|gt|ov} and __builtin_bcdsub_{lt|eq|gt|ov}.
7887
7888 2018-01-29  Richard Biener  <rguenther@suse.de>
7889
7890         PR tree-optimization/84086
7891         * tree-ssanames.c: Include cfgloop.h and tree-scalar-evolution.h.
7892         (flush_ssaname_freelist): When SSA names were released reset
7893         the SCEV hash table.
7894
7895 2018-01-29  Richard Biener  <rguenther@suse.de>
7896
7897         PR tree-optimization/84057
7898         * tree-ssa-loop-ivcanon.c (unloop_loops): Deal with already
7899         removed paths when removing edges.
7900
7901 2018-01-27  H.J. Lu  <hongjiu.lu@intel.com>
7902
7903         * doc/invoke.texi: Replace -mfunction-return==@var{choice} with
7904         -mfunction-return=@var{choice}.
7905
7906 2018-01-27  Bernd Edlinger  <bernd.edlinger@hotmail.de>
7907
7908         PR diagnostic/84034
7909         * diagnostic-show-locus.c (get_line_width_without_trailing_whitespace):
7910         Handle CR like TAB.
7911         (layout::print_source_line): Likewise.
7912         (test_get_line_width_without_trailing_whitespace): Add test cases.
7913
7914 2018-01-27  Jakub Jelinek  <jakub@redhat.com>
7915
7916         PR middle-end/84040
7917         * sched-deps.c (sched_macro_fuse_insns): Return immediately for
7918         debug insns.
7919
7920 2018-01-26  Jim Wilson  <jimw@sifive.com>
7921
7922         * config/riscv/riscv.h (MAX_FIXED_MODE_SIZE): New.
7923
7924         * config/riscv/elf.h (LIB_SPEC): Don't include -lgloss when nosys.specs
7925         specified.
7926
7927 2018-01-26  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
7928
7929         * config/aarch64/aarch64.md: Add peepholes for CMP + SUB -> SUBS
7930         and CMP + SUB-immediate -> SUBS.
7931
7932 2018-01-26  Martin Sebor  <msebor@redhat.com>
7933
7934         PR tree-optimization/83896
7935         * tree-ssa-strlen.c (get_string_len): Rename...
7936         (get_string_cst_length): ...to this.  Return HOST_WIDE_INT.
7937         Avoid assuming length is constant.
7938         (handle_char_store): Use HOST_WIDE_INT for string length.
7939
7940 2018-01-26  Uros Bizjak  <ubizjak@gmail.com>
7941
7942         PR target/81763
7943         * config/i386/i386.md (*andndi3_doubleword): Add earlyclobber
7944         to (=&r,r,rm) alternative. Add (=r,0,rm) and (=r,r,0) alternatives.
7945
7946 2018-01-26  Richard Biener  <rguenther@suse.de>
7947
7948         PR rtl-optimization/84003
7949         * dse.c (record_store): Only record redundant stores when
7950         the earlier store aliases at least all accesses the later one does.
7951
7952 2018-01-26  Jakub Jelinek  <jakub@redhat.com>
7953
7954         PR rtl-optimization/83985
7955         * dce.c (deletable_insn_p): Return false for separate shrink wrapping
7956         REG_CFA_RESTORE insns.
7957         (delete_unmarked_insns): Don't ignore separate shrink wrapping
7958         REG_CFA_RESTORE insns here.
7959
7960         PR c/83989
7961         * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Don't
7962         use SSA_NAME_VAR as base for SSA_NAMEs with non-NULL SSA_NAME_VAR.
7963
7964 2018-01-26  Claudiu Zissulescu  <claziss@synopsys.com>
7965
7966         * config/arc/arc-arch.h (arc_tune_attr): Add ARC_TUNE_CORE_3.
7967         * config/arc/arc.c (arc_sched_issue_rate): Use ARC_TUNE_... .
7968         (arc_init): Likewise.
7969         (arc_override_options): Likewise.
7970         (arc_file_start): Choose Tag_ARC_CPU_variation based on arc_tune
7971         value.
7972         (hwloop_fail): Use TARGET_DBNZ when we want to check for dbnz insn
7973         support.
7974         * config/arc/arc.h (TARGET_DBNZ): Define.
7975         * config/arc/arc.md (attr tune): Add core_3, use ARC_TUNE_... to
7976         properly set the tune attribute.
7977         (dbnz): Use TARGET_DBNZ guard.
7978         * config/arc/arc.opt (mtune): Add core3 option.
7979
7980 2018-01-26  Claudiu Zissulescu  <claziss@synopsys.com>
7981
7982         * config/arc/arc.c (arc_delegitimize_address_0): Refactored to
7983         recognize new pic like addresses.
7984         (arc_delegitimize_address): Clean up.
7985
7986 2018-01-26  Claudiu Zissulescu  <claziss@synopsys.com>
7987
7988         * config/arc/arc-arches.def: Option mrf16 valid for all
7989         architectures.
7990         * config/arc/arc-c.def (__ARC_RF16__): New predefined macro.
7991         * config/arc/arc-cpus.def (em_mini): New cpu with rf16 on.
7992         * config/arc/arc-options.def (FL_RF16): Add mrf16 option.
7993         * config/arc/arc-tables.opt: Regenerate.
7994         * config/arc/arc.c (arc_conditional_register_usage): Handle
7995         reduced register file case.
7996         (arc_file_start): Set must have build attributes.
7997         * config/arc/arc.h (MAX_ARC_PARM_REGS): Conditional define using
7998         mrf16 option value.
7999         * config/arc/arc.opt (mrf16): Add new option.
8000         * config/arc/elf.h (ATTRIBUTE_PCS): Define.
8001         * config/arc/genmultilib.awk: Handle new mrf16 option.
8002         * config/arc/linux.h (ATTRIBUTE_PCS): Define.
8003         * config/arc/t-multilib: Regenerate.
8004         * doc/invoke.texi (ARC Options): Document mrf16 option.
8005
8006 2018-01-26  Claudiu Zissulescu  <claziss@synopsys.com>
8007
8008         * config/arc/arc-protos.h: Add arc_is_secure_call_p proto.
8009         * config/arc/arc.c (arc_handle_secure_attribute): New function.
8010         (arc_attribute_table): Add 'secure_call' attribute.
8011         (arc_print_operand): Print secure call operand.
8012         (arc_function_ok_for_sibcall): Don't optimize tail calls when
8013         secure.
8014         (arc_is_secure_call_p): New function.  * config/arc/arc.md
8015         (call_i): Add support for sjli instruction.
8016         (call_value_i): Likewise.
8017         * config/arc/constraints.md (Csc): New constraint.
8018
8019 2018-01-26  Claudiu Zissulescu  <claziss@synopsys.com>
8020             John Eric Martin  <John.Martin@emmicro-us.com>
8021
8022         * config/arc/arc-protos.h: Add arc_is_jli_call_p proto.
8023         * config/arc/arc.c (_arc_jli_section): New struct.
8024         (arc_jli_section): New type.
8025         (rc_jli_sections): New static variable.
8026         (arc_handle_jli_attribute): New function.
8027         (arc_attribute_table): Add jli_always and jli_fixed attribute.
8028         (arc_file_end): New function.
8029         (TARGET_ASM_FILE_END): Define.
8030         (arc_print_operand): Reuse 'S' letter for JLI output instruction.
8031         (arc_add_jli_section): New function.
8032         (jli_call_scan): Likewise.
8033         (arc_reorg): Call jli_call_scan.
8034         (arc_output_addsi): Remove 'S' from printing asm operand.
8035         (arc_is_jli_call_p): New function.
8036         * config/arc/arc.md (movqi_insn): Remove 'S' from printing asm
8037         operand.
8038         (movhi_insn): Likewise.
8039         (movsi_insn): Likewise.
8040         (movsi_set_cc_insn): Likewise.
8041         (loadqi_update): Likewise.
8042         (load_zeroextendqisi_update): Likewise.
8043         (load_signextendqisi_update): Likewise.
8044         (loadhi_update): Likewise.
8045         (load_zeroextendhisi_update): Likewise.
8046         (load_signextendhisi_update): Likewise.
8047         (loadsi_update): Likewise.
8048         (loadsf_update): Likewise.
8049         (movsicc_insn): Likewise.
8050         (bset_insn): Likewise.
8051         (bxor_insn): Likewise.
8052         (bclr_insn): Likewise.
8053         (bmsk_insn): Likewise.
8054         (bicsi3_insn): Likewise.
8055         (cmpsi_cc_c_insn): Likewise.
8056         (movsi_ne): Likewise.
8057         (movsi_cond_exec): Likewise.
8058         (clrsbsi2): Likewise.
8059         (norm_f): Likewise.
8060         (normw): Likewise.
8061         (swap): Likewise.
8062         (divaw): Likewise.
8063         (flag): Likewise.
8064         (sr): Likewise.
8065         (kflag): Likewise.
8066         (ffs): Likewise.
8067         (ffs_f): Likewise.
8068         (fls): Likewise.
8069         (call_i): Remove 'S' asm letter, add jli instruction.
8070         (call_value_i): Likewise.
8071         * config/arc/arc.op (mjli-always): New option.
8072         * config/arc/constraints.md (Cji): New constraint.
8073         * config/arc/fpx.md (addsf3_fpx): Remove 'S' from printing asm
8074         operand.
8075         (subsf3_fpx): Likewise.
8076         (mulsf3_fpx): Likewise.
8077         * config/arc/simdext.md (vendrec_insn): Remove 'S' from printing
8078         asm operand.
8079         * doc/extend.texi (ARC): Document 'jli-always' and 'jli-fixed'
8080         function attrbutes.
8081         * doc/invoke.texi (ARC): Document mjli-always option.
8082
8083 2018-01-26  Sebastian Perta  <sebastian.perta@renesas.com>
8084
8085         * config/rl78/rl78.c (rl78_addsi3_internal): If operand 2 is const
8086         avoid addition with 0 and use incw and decw where possible.
8087
8088 2018-01-26  Richard Biener  <rguenther@suse.de>
8089
8090         PR tree-optimization/81082
8091         * fold-const.c (fold_plusminus_mult_expr): Do not perform the
8092         association if it requires casting to unsigned.
8093         * match.pd ((A * C) +- (B * C) -> (A+-B)): New patterns derived
8094         from fold_plusminus_mult_expr to catch important cases late when
8095         range info is available.
8096
8097 2018-01-26  Rainer Orth  <ro@CeBiTec.Uni-Bielefeld.DE>
8098
8099         * config/i386/sol2.h (USE_HIDDEN_LINKONCE): Remove.
8100         * configure.ac (hidden_linkonce): New test.
8101         * configure: Regenerate.
8102         * config.in: Regenerate.
8103
8104 2018-01-26  Julia Koval  <julia.koval@intel.com>
8105
8106         * config/i386/avx512bitalgintrin.h (_mm512_bitshuffle_epi64_mask,
8107         _mm512_mask_bitshuffle_epi64_mask, _mm256_bitshuffle_epi64_mask,
8108         _mm256_mask_bitshuffle_epi64_mask, _mm_bitshuffle_epi64_mask,
8109         _mm_mask_bitshuffle_epi64_mask): Fix type.
8110         * config/i386/i386-builtin-types.def (UHI_FTYPE_V2DI_V2DI_UHI,
8111         USI_FTYPE_V4DI_V4DI_USI): Remove.
8112         * config/i386/i386-builtin.def (__builtin_ia32_vpshufbitqmb512_mask,
8113         __builtin_ia32_vpshufbitqmb256_mask,
8114         __builtin_ia32_vpshufbitqmb128_mask): Fix types.
8115         * config/i386/i386.c (ix86_expand_args_builtin): Remove old types.
8116         * config/i386/sse.md (VI1_AVX512VLBW): Change types.
8117
8118 2018-01-26  Alan Modra  <amodra@gmail.com>
8119
8120         PR target/84033
8121         * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Exclude
8122         UNSPEC_VBPERMQ.  Sort other unspecs.
8123
8124 2018-01-25  David Edelsohn  <dje.gcc@gmail.com>
8125
8126         * doc/invoke.texi (PowerPC Options): Document 'native' cpu type.
8127
8128 2018-01-25  Jan Hubicka  <hubicka@ucw.cz>
8129
8130         PR middle-end/83055
8131         * predict.c (drop_profile): Do not push/pop cfun; update also
8132         node->count.
8133         (handle_missing_profiles): Fix logic looking for zero profiles.
8134
8135 2018-01-25  Jakub Jelinek  <jakub@redhat.com>
8136
8137         PR middle-end/83977
8138         * ipa-fnsummary.c (compute_fn_summary): Clear can_change_signature
8139         on functions with #pragma omp declare simd or functions with simd
8140         attribute.
8141         * omp-simd-clone.c (expand_simd_clones): Revert 2018-01-24 change.
8142         * config/i386/i386.c (ix86_simd_clone_compute_vecsize_and_simdlen):
8143         Remove trailing \n from warning_at calls.
8144
8145 2018-01-25  Tom de Vries  <tom@codesourcery.com>
8146
8147         PR target/84028
8148         * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
8149         for neutered workers.
8150
8151 2018-01-24  Joseph Myers  <joseph@codesourcery.com>
8152
8153         PR target/68467
8154         * config/m68k/m68k.c (m68k_promote_function_mode): New function.
8155         (TARGET_PROMOTE_FUNCTION_MODE): New macro.
8156
8157 2018-01-24  Jeff Law  <law@redhat.com>
8158
8159         PR target/83994
8160         * i386.c (get_probe_interval): Move to earlier point.
8161         (ix86_compute_frame_layout): If -fstack-clash-protection and
8162         the frame is larger than the probe interval, then use pushes
8163         to save registers rather than reg->mem moves.
8164         (ix86_expand_prologue): Remove conditional for int_registers_saved
8165         assertion.
8166
8167 2018-01-24  Vladimir Makarov  <vmakarov@redhat.com>
8168
8169         PR target/84014
8170         * ira-build.c (setup_min_max_allocno_live_range_point): Set up
8171         min/max for never referenced object.
8172
8173 2018-01-24  Jakub Jelinek  <jakub@redhat.com>
8174
8175         PR middle-end/83977
8176         * tree.c (free_lang_data_in_decl): Don't clear DECL_ABSTRACT_ORIGIN
8177         here.
8178         * omp-low.c (create_omp_child_function): Remove "omp declare simd"
8179         attributes from DECL_ATTRIBUTES (decl) without affecting
8180         DECL_ATTRIBUTES (current_function_decl).
8181         * omp-simd-clone.c (expand_simd_clones): Ignore DECL_ARTIFICIAL
8182         functions with non-NULL DECL_ABSTRACT_ORIGIN.
8183
8184 2018-01-24  Richard Sandiford  <richard.sandiford@linaro.org>
8185
8186         PR tree-optimization/83979
8187         * fold-const.c (fold_comparison): Use constant_boolean_node
8188         instead of boolean_{true,false}_node.
8189
8190 2018-01-24  Jan Hubicka  <hubicka@ucw.cz>
8191
8192         * ipa-profile.c (ipa_propagate_frequency_1): Fix logic skipping calls
8193         with zero counts.
8194
8195 2018-01-24  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
8196
8197         * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
8198         Simplify the clause that sets the length attribute.
8199         (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
8200         (*sibcall_nonlocal_sysv<mode>): Clean up code block; simplify the
8201         clause that sets the length attribute.
8202         (*sibcall_value_nonlocal_sysv<mode>): Likewise.
8203
8204 2018-01-24  Tom de Vries  <tom@codesourcery.com>
8205
8206         PR target/83589
8207         * config/nvptx/nvptx.c (WORKAROUND_PTXJIT_BUG_2): Define to 1.
8208         (nvptx_pc_set, nvptx_condjump_label): New function. Copy from jump.c.
8209         Add strict parameter.
8210         (prevent_branch_around_nothing): Insert dummy insn between branch to
8211         label and label with no ptx insn inbetween.
8212         * config/nvptx/nvptx.md (define_insn "fake_nop"): New insn.
8213
8214 2018-01-24  Tom de Vries  <tom@codesourcery.com>
8215
8216         PR target/81352
8217         * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
8218         for neutered threads in warp.
8219         * config/nvptx/nvptx.md (define_insn "exit"): New insn.
8220
8221 2018-01-24  Richard Biener  <rguenther@suse.de>
8222
8223         PR tree-optimization/83176
8224         * tree-chrec.c (chrec_fold_plus_1): Handle (signed T){(T) .. }
8225         operands.
8226
8227 2018-01-24  Richard Biener  <rguenther@suse.de>
8228
8229         PR tree-optimization/82819
8230         * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Avoid
8231         code generating pluses that are no-ops in the target precision.
8232
8233 2018-01-24  Richard Biener  <rguenther@suse.de>
8234
8235         PR middle-end/84000
8236         * tree-cfg.c (replace_loop_annotate): Handle annot_expr_parallel_kind.
8237
8238 2018-01-23  Jan Hubicka  <hubicka@ucw.cz>
8239
8240         * cfgcleanup.c (try_crossjump_to_edge): Use combine_with_count
8241         to merge probabilities.
8242         * predict.c (probably_never_executed): Also mark as cold functions
8243         with global 0 profile and guessed local profile.
8244         * profile-count.c (profile_probability::combine_with_count): New
8245         member function.
8246         * profile-count.h (profile_probability::operator*,
8247         profile_probability::operator*=, profile_probability::operator/,
8248         profile_probability::operator/=): Reduce precision to adjusted
8249         and set value to guessed on contradictory divisions.
8250         (profile_probability::combine_with_freq): Remove.
8251         (profile_probability::combine_wiht_count): Declare.
8252         (profile_count::force_nonzero):: Set to adjusted.
8253         (profile_count::probability_in):: Set quality to adjusted.
8254         * tree-ssa-tail-merge.c (replace_block_by): Use
8255         combine_with_count.
8256
8257 2018-01-23  Andrew Waterman  <andrew@sifive.com>
8258             Jim Wilson  <jimw@sifive.com>
8259
8260         * config/riscv/riscv.c (riscv_stack_boundary): New.
8261         (riscv_option_override): Set riscv_stack_boundary.  Handle
8262         riscv_preferred_stack_boundary_arg.
8263         * config/riscv/riscv.h (MIN_STACK_BOUNDARY, ABI_STACK_BOUNDARY): New.
8264         (BIGGEST_ALIGNMENT): Set to STACK_BOUNDARY.
8265         (STACK_BOUNDARY): Set to riscv_stack_boundary.
8266         (RISCV_STACK_ALIGN): Use STACK_BOUNDARY.
8267         * config/riscv/riscv.opt (mpreferred-stack-boundary): New.
8268         * doc/invoke.tex (RISC-V Options): Add -mpreferred-stack-boundary.
8269
8270 2018-01-23  H.J. Lu  <hongjiu.lu@intel.com>
8271
8272         PR target/83905
8273         * config/i386/i386.c (ix86_expand_prologue): Use cost reference
8274         of struct ix86_frame.
8275         (ix86_expand_epilogue): Likewise.  Add a local variable for
8276         the reg_save_offset field in struct ix86_frame.
8277
8278 2018-01-23  Bin Cheng  <bin.cheng@arm.com>
8279
8280         PR tree-optimization/82604
8281         * tree-loop-distribution.c (enum partition_kind): New enum item
8282         PKIND_PARTIAL_MEMSET.
8283         (partition_builtin_p): Support above new enum item.
8284         (generate_code_for_partition): Ditto.
8285         (compute_access_range): Differentiate cases that equality can be
8286         proven at all loops, the innermost loops or no loops.
8287         (classify_builtin_st, classify_builtin_ldst): Adjust call to above
8288         function.  Set PKIND_PARTIAL_MEMSET for partition appropriately.
8289         (finalize_partitions, distribute_loop): Don't fuse partition of
8290         PKIND_PARTIAL_MEMSET kind when distributing 3-level loop nest.
8291         (prepare_perfect_loop_nest): Distribute 3-level loop nest only if
8292         parloop is enabled.
8293
8294 2018-01-23  Martin Liska  <mliska@suse.cz>
8295
8296         * predict.def (PRED_INDIR_CALL): Set probability to PROB_EVEN in
8297         order to ignore the predictor.
8298         (PRED_POLYMORPHIC_CALL): Likewise.
8299         (PRED_RECURSIVE_CALL): Likewise.
8300
8301 2018-01-23  Martin Liska  <mliska@suse.cz>
8302
8303         * tree-profile.c (tree_profiling): Print function header to
8304         aware reader which function we are working on.
8305         * value-prof.c (gimple_find_values_to_profile): Do not print
8306         not interesting value histograms.
8307
8308 2018-01-23  Martin Liska  <mliska@suse.cz>
8309
8310         * profile-count.h (enum profile_quality): Add
8311         profile_uninitialized as the first value. Do not number values
8312         as they are zero based.
8313         (profile_count::verify): Update sanity check.
8314         (profile_probability::verify): Likewise.
8315
8316 2018-01-23  Nathan Sidwell  <nathan@acm.org>
8317
8318         * doc/invoke.texi (ffor-scope): Deprecate.
8319
8320 2018-01-23  David Malcolm  <dmalcolm@redhat.com>
8321
8322         PR tree-optimization/83510
8323         * domwalk.c (set_all_edges_as_executable): New function.
8324         (dom_walker::dom_walker): Convert bool param
8325         "skip_unreachable_blocks" to enum reachability.  Move setup of
8326         edge flags to set_all_edges_as_executable and only do it when
8327         reachability is REACHABLE_BLOCKS.
8328         * domwalk.h (enum dom_walker::reachability): New enum.
8329         (dom_walker::dom_walker): Convert bool param
8330         "skip_unreachable_blocks" to enum reachability.
8331         (set_all_edges_as_executable): New decl.
8332         * graphite-scop-detection.c  (gather_bbs::gather_bbs): Convert
8333         from false for "skip_unreachable_blocks" to ALL_BLOCKS for
8334         "reachability".
8335         * tree-ssa-dom.c (dom_opt_dom_walker::dom_opt_dom_walker): Likewise,
8336         but converting true to REACHABLE_BLOCKS.
8337         * tree-ssa-sccvn.c (sccvn_dom_walker::sccvn_dom_walker): Likewise.
8338         * tree-vrp.c
8339         (check_array_bounds_dom_walker::check_array_bounds_dom_walker):
8340         Likewise, but converting it to REACHABLE_BLOCKS_PRESERVING_FLAGS.
8341         (vrp_dom_walker::vrp_dom_walker): Likewise, but converting it to
8342         REACHABLE_BLOCKS.
8343         (vrp_prop::vrp_finalize): Call set_all_edges_as_executable
8344         if check_all_array_refs will be called.
8345
8346 2018-01-23  David Malcolm  <dmalcolm@redhat.com>
8347
8348         * tree.c (selftest::test_location_wrappers): Add more test
8349         coverage.
8350
8351 2018-01-23  David Malcolm  <dmalcolm@redhat.com>
8352
8353         * sbitmap.c (selftest::test_set_range): Fix memory leaks.
8354         (selftest::test_bit_in_range): Likewise.
8355
8356 2018-01-23  Richard Sandiford  <richard.sandiford@linaro.org>
8357
8358         PR testsuite/83888
8359         * doc/sourcebuild.texi (vect_float): Say that the selector
8360         only describes the situation when -funsafe-math-optimizations is on.
8361         (vect_float_strict): Document.
8362
8363 2018-01-23  Richard Sandiford  <richard.sandiford@linaro.org>
8364
8365         PR tree-optimization/83965
8366         * tree-vect-patterns.c (vect_reassociating_reduction_p): New function.
8367         (vect_recog_dot_prod_pattern, vect_recog_sad_pattern): Use it
8368         instead of checking only for a reduction.
8369         (vect_recog_widen_sum_pattern): Likewise.
8370
8371 2018-01-23  Jan Hubicka  <hubicka@ucw.cz>
8372
8373         * predict.c (probably_never_executed): Only use precise profile info.
8374         (compute_function_frequency): Skip after inlining hack since we now
8375         have quality checking.
8376
8377 2018-01-23  Jan Hubicka  <hubicka@ucw.cz>
8378
8379         * profile-count.h (profile_probability::very_unlikely,
8380         profile_probability::unlikely, profile_probability::even): Set
8381         precision to guessed.
8382
8383 2018-01-23  Richard Biener  <rguenther@suse.de>
8384
8385         PR tree-optimization/83963
8386         * graphite-scop-detection.c (scop_detection::harmful_loop_in_region):
8387         Properly terminate dominator walk when crossing the exit edge not
8388         when visiting its source block.
8389
8390 2018-01-23  Jakub Jelinek  <jakub@redhat.com>
8391
8392         PR c++/83918
8393         * tree.c (maybe_wrap_with_location): Use NON_LVALUE_EXPR rather than
8394         VIEW_CONVERT_EXPR to wrap CONST_DECLs.
8395
8396 2018-01-22  Jakub Jelinek  <jakub@redhat.com>
8397
8398         PR tree-optimization/83957
8399         * omp-expand.c (expand_omp_for_generic): Ignore virtual PHIs.  Remove
8400         semicolon after for body surrounded by braces.
8401
8402         PR tree-optimization/83081
8403         * profile-count.h (profile_probability::split): New method.
8404         * dojump.c (do_jump_1) <case TRUTH_ANDIF_EXPR, case TRUTH_ORIF_EXPR>:
8405         Use profile_probability::split.
8406         (do_compare_rtx_and_jump): Fix adjustment of probabilities
8407         when splitting a single conditional jump into 2.
8408
8409 2018-01-22  David Malcolm  <dmalcolm@redhat.com>
8410
8411         PR tree-optimization/69452
8412         * tree-ssa-loop-im.c (class move_computations_dom_walker): Remove
8413         decl.
8414
8415 2018-01-22  Sebastian Perta  <sebastian.perta@renesas.com>
8416
8417         * config/rl78/rl78-expand.md (bswaphi2): New define_expand.
8418         * config/rl78/rl78-virt.md (*bswaphi2_virt): New define_insn.
8419         * config/rl78/rl78-real.md (*bswaphi2_real): New define_insn.
8420
8421 2018-01-22  Sebastian Perta  <sebastian.perta@renesas.com>
8422
8423         * config/rl78/rl78-protos.h (rl78_split_movdi): New function declaration.
8424         * config/rl78/rl78.md (movdi): New define_expand.
8425         * config/rl78/rl78.c (rl78_split_movdi): New function.
8426
8427 2018-01-22  Michael Meissner  <meissner@linux.vnet.ibm.com>
8428
8429         PR target/83862
8430         * config/rs6000/rs6000-protos.h (rs6000_split_signbit): Delete,
8431         no longer used.
8432         * config/rs6000/rs6000.c (rs6000_split_signbit): Likewise.
8433         * config/rs6000/rs6000.md (signbit<mode>2): Change code for IEEE
8434         128-bit to produce an UNSPEC move to get the double word with the
8435         signbit and then a shift directly to do signbit.
8436         (signbit<mode>2_dm): Replace old IEEE 128-bit signbit
8437         implementation with a new version that just does either a direct
8438         move or a regular move.  Move memory interface to separate insns.
8439         Move insns so they are next to the expander.
8440         (signbit<mode>2_dm_mem_be): New combiner insns to combine load
8441         with signbit move.  Split big and little endian case.
8442         (signbit<mode>2_dm_mem_le): Likewise.
8443         (signbit<mode>2_dm_<su>ext): Delete, no longer used.
8444         (signbit<mode>2_dm2): Likewise.
8445
8446 2018-01-22  Sebastian Perta  <sebastian.perta@renesas.com>
8447
8448         * config/rl78/rl78.md (anddi3): New define_expand.
8449
8450 2018-01-22  Sebastian Perta  <sebastian.perta@renesas.com>
8451
8452         * config/rl78/rl78.md (umindi3): New define_expand.
8453
8454 2018-01-22  Sebastian Perta  <sebastian.perta@renesas.com>
8455
8456         * config/rl78/rl78.md (smindi3): New define_expand.
8457
8458 2018-01-22  Sebastian Perta  <sebastian.perta@renesas.com>
8459
8460         * config/rl78/rl78.md (smaxdi3): New define_expand.
8461
8462 2018-01-22  Carl Love  <cel@us.ibm.com>
8463
8464         * config/rs6000/rs6000-builtin.def (ST_ELEMREV_V1TI, LD_ELEMREV_V1TI,
8465         LVX_V1TI): Add macro expansion.
8466         * config/rs6000/rs6000-c.c (altivec_builtin_types): Add argument
8467         definitions for VSX_BUILTIN_VEC_XST_BE, VSX_BUILTIN_VEC_ST,
8468         VSX_BUILTIN_VEC_XL, LD_ELEMREV_V1TI builtins.
8469         * config/rs6000/rs6000-p8swap.c (insn_is_swappable_p);
8470         Change check to determine if the instruction is a byte reversing
8471         entry.  Fix typo in comment.
8472         * config/rs6000/rs6000.c (altivec_expand_builtin): Add case entry
8473         for VSX_BUILTIN_ST_ELEMREV_V1TI and VSX_BUILTIN_LD_ELEMREV_V1TI.
8474         Add def_builtin calls for new builtins.
8475         * config/rs6000/vsx.md (vsx_st_elemrev_v1ti, vsx_ld_elemrev_v1ti):
8476         Add define_insn expansion.
8477
8478 2018-01-22  Sebastian Perta  <sebastian.perta@renesas.com>
8479
8480         * config/rl78/rl78.md (umaxdi3): New define_expand.
8481
8482 2018-01-22  Sebastian Perta  <sebastian.perta@renesas.com>
8483
8484         * config/rl78/rl78.c (rl78_note_reg_set): Fixed dead reg check
8485         for non-QImode registers.
8486
8487 2018-01-22  Richard Biener  <rguenther@suse.de>
8488
8489         PR tree-optimization/83963
8490         * graphite-scop-detection.c (scop_detection::get_sese): Delay
8491         including the loop exit block.
8492         (scop_detection::merge_sese): Likewise.
8493         (scop_detection::add_scop): Do it here instead.
8494
8495 2018-01-22  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
8496
8497         * doc/sourcebuild.texi (arm_softfloat): Document.
8498
8499 2018-01-21  John David Anglin  <danglin@gcc.gnu.org>
8500
8501         PR gcc/77734
8502         * config/pa/pa.c (pa_function_ok_for_sibcall): Use
8503         targetm.binds_local_p instead of TREE_PUBLIC to check local binding.
8504         Move TARGET_PORTABLE_RUNTIME check after TARGET_64BIT check.
8505
8506 2018-01-21  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
8507             David Edelsohn  <dje.gcc@gmail.com>
8508
8509         PR target/83946
8510         * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
8511         Change "crset eq" to "crset 2".
8512         (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
8513         (*call_indirect_aix<mode>_nospec): Likewise.
8514         (*call_value_indirect_aix<mode>_nospec): Likewise.
8515         (*call_indirect_elfv2<mode>_nospec): Likewise.
8516         (*call_value_indirect_elfv2<mode>_nospec): Likewise.
8517         (*sibcall_nonlocal_sysv<mode>): Change "crset eq" to "crset 2";
8518         change assembly output from . to $.
8519         (*sibcall_value_nonlocal_sysv<mode>): Likewise.
8520         (indirect_jump<mode>_nospec): Change assembly output from . to $.
8521         (*tablejump<mode>_internal1_nospec): Likewise.
8522
8523 2018-01-21  Oleg Endo  <olegendo@gcc.gnu.org>
8524
8525         PR target/80870
8526         * config/sh/sh_optimize_sett_clrt.cc:
8527         Use INCLUDE_ALGORITHM and INCLUDE_VECTOR instead of direct includes.
8528
8529 2018-01-20  Richard Sandiford  <richard.sandiford@linaro.org>
8530
8531         PR tree-optimization/83940
8532         * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): Set
8533         offset_dt to vect_constant_def rather than vect_unknown_def_type.
8534         (vect_check_load_store_mask): Add a mask_dt_out parameter and
8535         use it to pass back the definition type.
8536         (vect_check_store_rhs): Likewise rhs_dt_out.
8537         (vect_build_gather_load_calls): Add a mask_dt argument and use
8538         it instead of a call to vect_is_simple_use.
8539         (vectorizable_store): Update calls to vect_check_load_store_mask
8540         and vect_check_store_rhs.  Use the dt returned by the latter instead
8541         of scatter_src_dt.  Use the cached mask_dt and gs_info.offset_dt
8542         instead of calls to vect_is_simple_use.  Pass the scalar rather
8543         than the vector operand to vect_is_simple_use when handling
8544         second and subsequent copies of an rhs value.
8545         (vectorizable_load): Update calls to vect_check_load_store_mask
8546         and vect_build_gather_load_calls.  Use the cached mask_dt and
8547         gs_info.offset_dt instead of calls to vect_is_simple_use.
8548
8549 2018-01-20  Jakub Jelinek  <jakub@redhat.com>
8550
8551         PR middle-end/83945
8552         * tree-emutls.c: Include gimplify.h.
8553         (lower_emutls_2): New function.
8554         (lower_emutls_1): If ADDR_EXPR is a gimple invariant and walk_tree
8555         with lower_emutls_2 callback finds some TLS decl in it, unshare_expr
8556         it before further processing.
8557
8558         PR target/83930
8559         * simplify-rtx.c (simplify_binary_operation_1) <case UMOD>: Use
8560         UINTVAL (trueop1) instead of INTVAL (op1).
8561
8562 2018-01-19  Jakub Jelinek  <jakub@redhat.com>
8563
8564         PR debug/81570
8565         PR debug/83728
8566         * dwarf2cfi.c (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define to
8567         INCOMING_FRAME_SP_OFFSET if not defined.
8568         (scan_trace): Add ENTRY argument.  If true and
8569         DEFAULT_INCOMING_FRAME_SP_OFFSET != INCOMING_FRAME_SP_OFFSET,
8570         emit a note to adjust the CFA offset.
8571         (create_cfi_notes): Adjust scan_trace callers.
8572         (create_cie_data): Use DEFAULT_INCOMING_FRAME_SP_OFFSET rather than
8573         INCOMING_FRAME_SP_OFFSET in the CIE.
8574         * config/i386/i386.h (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define.
8575         * config/stormy16/stormy16.h (DEFAULT_INCOMING_FRAME_SP_OFFSET):
8576         Likewise.
8577         * doc/tm.texi.in (DEFAULT_INCOMING_FRAME_SP_OFFSET): Document.
8578         * doc/tm.texi: Regenerated.
8579
8580 2018-01-19  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>
8581
8582         PR rtl-optimization/83147
8583         * lra-constraints.c (remove_inheritance_pseudos): Use
8584         lra_substitute_pseudo_within_insn.
8585
8586 2018-01-19  Tom de Vries  <tom@codesourcery.com>
8587             Cesar Philippidis  <cesar@codesourcery.com>
8588
8589         PR target/83920
8590         * config/nvptx/nvptx.c (nvptx_single): Fix jit workaround.
8591
8592 2018-01-19  Cesar Philippidis  <cesar@codesourcery.com>
8593
8594         PR target/83790
8595         * config/nvptx/nvptx.c (output_init_frag): Don't use generic address
8596         spaces for function labels.
8597
8598 2018-01-19  Martin Liska  <mliska@suse.cz>
8599
8600         * predict.def (PRED_LOOP_EXIT): Change from 85 to 89.
8601         (PRED_LOOP_EXIT_WITH_RECURSION): Change from 72 to 78.
8602         (PRED_LOOP_EXTRA_EXIT): Change from 83 to 67.
8603         (PRED_OPCODE_POSITIVE): Change from 64 to 59.
8604         (PRED_TREE_OPCODE_POSITIVE): Change from 64 to 59.
8605         (PRED_CONST_RETURN): Change from 69 to 65.
8606         (PRED_NULL_RETURN): Change from 91 to 71.
8607         (PRED_LOOP_IV_COMPARE_GUESS): Change from 98 to 64.
8608         (PRED_LOOP_GUARD): Change from 66 to 73.
8609
8610 2018-01-19  Martin Liska  <mliska@suse.cz>
8611
8612         * predict.c (predict_insn_def): Add new assert.
8613         (struct branch_predictor): Change type to signed integer.
8614         (test_prediction_value_range): Amend test to cover
8615         PROB_UNINITIALIZED.
8616         * predict.def (PRED_LOOP_ITERATIONS): Use the new constant.
8617         (PRED_LOOP_ITERATIONS_GUESSED): Likewise.
8618         (PRED_LOOP_ITERATIONS_MAX): Likewise.
8619         (PRED_LOOP_IV_COMPARE): Likewise.
8620         * predict.h (PROB_UNINITIALIZED): Define new constant.
8621
8622 2018-01-19  Martin Liska  <mliska@suse.cz>
8623
8624         * predict.c (dump_prediction): Add new format for
8625         analyze_brprob.py script which is enabled with -details
8626         suboption.
8627         * profile-count.h (precise_p): New function.
8628
8629 2018-01-19  Richard Sandiford  <richard.sandiford@linaro.org>
8630
8631         PR tree-optimization/83922
8632         * tree-vect-loop.c (vect_verify_full_masking): Return false if
8633         there are no statements that need masking.
8634         (vect_active_double_reduction_p): New function.
8635         (vect_analyze_loop_operations): Use it when handling phis that
8636         are not in the loop header.
8637
8638 2018-01-19  Richard Sandiford  <richard.sandiford@linaro.org>
8639
8640         PR tree-optimization/83914
8641         * tree-vect-loop.c (vectorizable_induction): Don't convert
8642         init_expr or apply the peeling adjustment for inductions
8643         that are nested within the vectorized loop.
8644
8645 2018-01-19  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
8646
8647         * config/arm/thumb2.md (*thumb2_negsi2_short): Use RSB mnemonic
8648         instead of NEG.
8649
8650 2018-01-18  Jakub Jelinek  <jakub@redhat.com>
8651
8652         PR sanitizer/81715
8653         PR testsuite/83882
8654         * function.h (gimplify_parameters): Add gimple_seq * argument.
8655         * function.c: Include gimple.h and options.h.
8656         (gimplify_parameters): Add cleanup argument, add CLOBBER stmts
8657         for the added local temporaries if needed.
8658         * gimplify.c (gimplify_body): Adjust gimplify_parameters caller,
8659         if there are any parameter cleanups, wrap whole body into a
8660         try/finally with the cleanups.
8661
8662 2018-01-18  Wilco Dijkstra  <wdijkstr@arm.com>
8663
8664         PR target/82964
8665         * config/aarch64/aarch64.c (aarch64_legitimate_constant_p):
8666         Use GET_MODE_CLASS for scalar floating point.
8667
8668 2018-01-18  Jan Hubicka  <hubicka@ucw.cz>
8669
8670         PR ipa/82256
8671         patch by PaX Team
8672         * cgraphclones.c (cgraph_node::create_version_clone_with_body):
8673         Fix call of call_cgraph_insertion_hooks.
8674
8675 2018-01-18  Martin Sebor  <msebor@redhat.com>
8676
8677         * doc/invoke.texi (-Wclass-memaccess): Tweak text.
8678
8679 2018-01-18  Jan Hubicka  <hubicka@ucw.cz>
8680
8681         PR ipa/83619
8682         * cgraph.c (cgraph_edge::redirect_call_stmt_to_callee): Update edge
8683         frequencies.
8684
8685 2018-01-18  Boris Kolpackov  <boris@codesynthesis.com>
8686
8687         PR other/70268
8688         * common.opt: (-ffile-prefix-map): New option.
8689         * opts.c (common_handle_option): Defer it.
8690         * opts-global.c (handle_common_deferred_options): Handle it.
8691         * debug.h (remap_debug_filename, add_debug_prefix_map): Move to...
8692         * file-prefix-map.h: New file.
8693         (remap_debug_filename, add_debug_prefix_map): ...here.
8694         (add_macro_prefix_map, add_file_prefix_map, remap_macro_filename): New.
8695         * final.c (debug_prefix_map, add_debug_prefix_map
8696         remap_debug_filename): Move to...
8697         * file-prefix-map.c: New file.
8698         (file_prefix_map, add_prefix_map, remap_filename) ...here and rename,
8699         generalize, get rid of alloca(), use strrchr() instead of strchr().
8700         (add_macro_prefix_map, add_debug_prefix_map, add_file_prefix_map):
8701         Implement in terms of add_prefix_map().
8702         (remap_macro_filename, remap_debug_filename): Implement in term of
8703         remap_filename().
8704         * Makefile.in (OBJS, PLUGIN_HEADERS): Add new files.
8705         * builtins.c (fold_builtin_FILE): Call remap_macro_filename().
8706         * dbxout.c: Include file-prefix-map.h.
8707         * varasm.c: Likewise.
8708         * vmsdbgout.c: Likewise.
8709         * xcoffout.c: Likewise.
8710         * dwarf2out.c: Likewise plus omit new options from DW_AT_producer.
8711         * doc/cppopts.texi (-fmacro-prefix-map): Document.
8712         * doc/invoke.texi (-ffile-prefix-map): Document.
8713         (-fdebug-prefix-map): Update description.
8714
8715 2018-01-18  Martin Liska  <mliska@suse.cz>
8716
8717         * config/i386/i386.c (indirect_thunk_name): Document that also
8718         lfence is emitted.
8719         (output_indirect_thunk): Document why both instructions
8720         (pause and lfence) are generated.
8721
8722 2018-01-18  Richard Biener  <rguenther@suse.de>
8723
8724         PR tree-optimization/83887
8725         * graphite-scop-detection.c
8726         (scop_detection::get_nearest_dom_with_single_entry): Remove.
8727         (scop_detection::get_nearest_pdom_with_single_exit): Likewise.
8728         (scop_detection::merge_sese): Re-implement with a flood-fill
8729         algorithm that properly finds a SESE region if it exists.
8730
8731 2018-01-18  Jakub Jelinek  <jakub@redhat.com>
8732
8733         PR c/61240
8734         * match.pd ((P + A) - P, P - (P + A), (P + A) - (P + B)): For
8735         pointer_diff optimizations use view_convert instead of convert.
8736
8737 2018-01-17  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
8738
8739         * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
8740         Generate different code for -mno-speculate-indirect-jumps.
8741         (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
8742         (*call_indirect_aix<mode>): Disable for
8743         -mno-speculate-indirect-jumps.
8744         (*call_indirect_aix<mode>_nospec): New define_insn.
8745         (*call_value_indirect_aix<mode>): Disable for
8746         -mno-speculate-indirect-jumps.
8747         (*call_value_indirect_aix<mode>_nospec): New define_insn.
8748         (*sibcall_nonlocal_sysv<mode>): Generate different code for
8749         -mno-speculate-indirect-jumps.
8750         (*sibcall_value_nonlocal_sysv<mode>): Likewise.
8751
8752 2018-01-17  Michael Meissner  <meissner@linux.vnet.ibm.com>
8753
8754         * config/rs6000/rs6000.c (rs6000_emit_move): If we load or store a
8755         long double type, set the flags for noting the default long double
8756         type, even if we don't pass or return a long double type.
8757
8758 2018-01-17  Jan Hubicka  <hubicka@ucw.cz>
8759
8760         PR ipa/83051
8761         * ipa-inline.c (flatten_function): Do not overwrite final inlining
8762         failure.
8763
8764 2018-01-17  Will Schmidt  <will_schmidt@vnet.ibm.com>
8765
8766         * config/rs6000/rs6000.c (rs6000_gimple_builtin): Add gimple folding
8767         support for merge[hl].
8768         (fold_mergehl_helper): New helper function.
8769         (tree-vector-builder.h): New #include for tree_vector_builder usage.
8770         * config/rs6000/altivec.md (altivec_vmrghw_direct): Add xxmrghw insn.
8771         (altivec_vmrglw_direct): Add xxmrglw insn.
8772
8773 2018-01-17  Andrew Waterman  <andrew@sifive.com>
8774
8775         * config/riscv/riscv.c (riscv_conditional_register_usage): If
8776         UNITS_PER_FP_ARG is 0, set call_used_regs to 1 for all FP regs.
8777
8778 2018-01-17  David Malcolm  <dmalcolm@redhat.com>
8779
8780         PR lto/83121
8781         * ipa-devirt.c (add_type_duplicate): When comparing memory layout,
8782         call the lto_location_cache before reading the
8783         DECL_SOURCE_LOCATION of the types.
8784
8785 2018-01-17  Wilco Dijkstra  <wdijkstr@arm.com>
8786             Richard Sandiford  <richard.sandiford@linaro.org>
8787
8788         * config/aarch64/aarch64.md (movti_aarch64): Use Uti constraint.
8789         * config/aarch64/aarch64.c (aarch64_mov128_immediate): New function.
8790         (aarch64_legitimate_constant_p): Just support CONST_DOUBLE
8791         SF/DF/TF mode to avoid creating illegal CONST_WIDE_INT immediates.
8792         * config/aarch64/aarch64-protos.h (aarch64_mov128_immediate):
8793         Add declaration.
8794         * config/aarch64/constraints.md (aarch64_movti_operand):
8795         Limit immediates.
8796         * config/aarch64/predicates.md (Uti): Add new constraint.
8797
8798 2018-01-17 Carl Love  <cel@us.ibm.com>
8799
8800         * config/rs6000/vsx.md (define_expand xl_len_r,
8801         define_expand stxvl, define_expand *stxvl): Add match_dup argument.
8802         (define_insn): Add, match_dup 1 argument to define_insn stxvll and
8803         lxvll.
8804         (define_expand, define_insn): Move the shift left from  the
8805         define_insn to the define_expand for lxvl and stxvl instructions.
8806         * config/rs6000/rs6000-builtin.def (BU_P9V_64BIT_VSX_2): Change LXVL
8807         and XL_LEN_R definitions to PURE.
8808
8809 2018-01-17  Uros Bizjak  <ubizjak@gmail.com>
8810
8811         * config/i386/i386.c (indirect_thunk_name): Declare regno
8812         as unsigned int.  Compare regno with INVALID_REGNUM.
8813         (output_indirect_thunk): Ditto.
8814         (output_indirect_thunk_function): Ditto.
8815         (ix86_code_end): Declare regno as unsigned int.  Use INVALID_REGNUM
8816         in the call to output_indirect_thunk_function.
8817
8818 2018-01-17  Richard Sandiford  <richard.sandiford@linaro.org>
8819
8820         PR middle-end/83884
8821         * expr.c (expand_expr_real_1): Use the size of GET_MODE (op0)
8822         rather than the size of inner_type to determine the stack slot size
8823         when handling VIEW_CONVERT_EXPRs on strict-alignment targets.
8824
8825 2018-01-16  Sebastian Peryt  <sebastian.peryt@intel.com>
8826
8827         PR target/83546
8828         * config/i386/i386.c (ix86_option_override_internal): Add PTA_RDRND
8829         to PTA_SILVERMONT.
8830
8831 2018-01-16  Michael Meissner  <meissner@linux.vnet.ibm.com>
8832
8833         * config.gcc (powerpc*-linux*-*): Add support for 64-bit little
8834         endian Linux systems to optionally enable multilibs for selecting
8835         the long double type if the user configured an explicit type.
8836         * config/rs6000/rs6000.h (TARGET_IEEEQUAD_MULTILIB): Indicate we
8837         have no long double multilibs if not defined.
8838         * config/rs6000/rs6000.c (rs6000_option_override_internal): Do not
8839         warn if the user used -mabi={ieee,ibm}longdouble and we built
8840         multilibs for long double.
8841         * config/rs6000/linux64.h (MULTILIB_DEFAULTS_IEEE): Define as the
8842         appropriate multilib option.
8843         (MULTILIB_DEFAULTS): Add MULTILIB_DEFAULTS_IEEE to the default
8844         multilib options.
8845         * config/rs6000/t-ldouble-linux64le-ibm: New configuration files
8846         for building long double multilibs.
8847         * config/rs6000/t-ldouble-linux64le-ieee: Likewise.
8848
8849 2018-01-16  John David Anglin  <danglin@gcc.gnu.org>
8850
8851         * config.gcc (hppa*-*-linux*): Change callee copies ABI to caller
8852         copies.
8853
8854         * config/pa.h (MALLOC_ABI_ALIGNMENT): Set 32-bit alignment default to
8855         64 bits.
8856         * config/pa/pa32-linux.h (MALLOC_ABI_ALIGNMENT): Set alignment to
8857         128 bits.
8858
8859         * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Cleanup type and mode
8860         variables.
8861
8862         * config/pa/pa.c (pa_function_arg_size): Apply CEIL to GET_MODE_SIZE
8863         return value.
8864
8865 2018-01-16  Eric Botcazou  <ebotcazou@adacore.com>
8866
8867         * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For an
8868         ADDR_EXPR, do not count the offset of a COMPONENT_REF twice.
8869
8870 2018-01-16  Kelvin Nilsen  <kelvin@gcc.gnu.org>
8871
8872         * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Generate
8873         different rtl trees depending on TARGET_64BIT.
8874         (rs6000_gen_lvx): Likewise.
8875
8876 2018-01-16  Eric Botcazou  <ebotcazou@adacore.com>
8877
8878         * config/visium/visium.md (nop): Tweak comment.
8879         (hazard_nop): Likewise.
8880
8881 2018-01-16  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
8882
8883         * config/rs6000/rs6000.c (rs6000_opt_vars): Add entry for
8884         -mspeculate-indirect-jumps.
8885         * config/rs6000/rs6000.md (*call_indirect_elfv2<mode>): Disable
8886         for -mno-speculate-indirect-jumps.
8887         (*call_indirect_elfv2<mode>_nospec): New define_insn.
8888         (*call_value_indirect_elfv2<mode>): Disable for
8889         -mno-speculate-indirect-jumps.
8890         (*call_value_indirect_elfv2<mode>_nospec): New define_insn.
8891         (indirect_jump): Emit different RTL for
8892         -mno-speculate-indirect-jumps.
8893         (*indirect_jump<mode>): Disable for
8894         -mno-speculate-indirect-jumps.
8895         (*indirect_jump<mode>_nospec): New define_insn.
8896         (tablejump): Emit different RTL for
8897         -mno-speculate-indirect-jumps.
8898         (tablejumpsi): Disable for -mno-speculate-indirect-jumps.
8899         (tablejumpsi_nospec): New define_expand.
8900         (tablejumpdi): Disable for -mno-speculate-indirect-jumps.
8901         (tablejumpdi_nospec): New define_expand.
8902         (*tablejump<mode>_internal1): Disable for
8903         -mno-speculate-indirect-jumps.
8904         (*tablejump<mode>_internal1_nospec): New define_insn.
8905         * config/rs6000/rs6000.opt (mspeculate-indirect-jumps): New
8906         option.
8907
8908 2018-01-16  Artyom Skrobov tyomitch@gmail.com
8909
8910         * caller-save.c (insert_save): Drop unnecessary parameter.  All
8911         callers updated.
8912
8913 2018-01-16  Jakub Jelinek  <jakub@redhat.com>
8914             Richard Biener  <rguenth@suse.de>
8915
8916         PR libgomp/83590
8917         * gimplify.c (gimplify_one_sizepos): For is_gimple_constant (expr)
8918         return early, inline manually is_gimple_sizepos.  Make sure if we
8919         call gimplify_expr we don't end up with a gimple constant.
8920         * tree.c (variably_modified_type_p): Don't return true for
8921         is_gimple_constant (_t).  Inline manually is_gimple_sizepos.
8922         * gimplify.h (is_gimple_sizepos): Remove.
8923
8924 2018-01-16  Richard Sandiford  <richard.sandiford@linaro.org>
8925
8926         PR tree-optimization/83857
8927         * tree-vect-loop.c (vect_analyze_loop_operations): Don't call
8928         vectorizable_live_operation for pure SLP statements.
8929         (vectorizable_live_operation): Handle PHIs.
8930
8931 2018-01-16  Richard Biener  <rguenther@suse.de>
8932
8933         PR tree-optimization/83867
8934         * tree-vect-stmts.c (vect_transform_stmt): Precompute
8935         nested_in_vect_loop_p since the scalar stmt may get invalidated.
8936
8937 2018-01-16  Jakub Jelinek  <jakub@redhat.com>
8938
8939         PR c/83844
8940         * stor-layout.c (handle_warn_if_not_align): Use byte_position and
8941         multiple_of_p instead of unchecked tree_to_uhwi and UHWI check.
8942         If off is not INTEGER_CST, issue a may not be aligned warning
8943         rather than isn't aligned.  Use isn%'t rather than isn't.
8944         * fold-const.c (multiple_of_p) <case BIT_AND_EXPR>: Don't fall through
8945         into MULT_EXPR.
8946         <case MULT_EXPR>: Improve the case when bottom and one of the
8947         MULT_EXPR operands are INTEGER_CSTs and bottom is multiple of that
8948         operand, in that case check if the other operand is multiple of
8949         bottom divided by the INTEGER_CST operand.
8950
8951 2018-01-16  Richard Sandiford  <richard.sandiford@linaro.org>
8952
8953         PR target/83858
8954         * config/pa/pa.h (FUNCTION_ARG_SIZE): Delete.
8955         * config/pa/pa-protos.h (pa_function_arg_size): Declare.
8956         * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Use
8957         pa_function_arg_size instead of FUNCTION_ARG_SIZE.
8958         * config/pa/pa.c (pa_function_arg_advance): Likewise.
8959         (pa_function_arg, pa_arg_partial_bytes): Likewise.
8960         (pa_function_arg_size): New function.
8961
8962 2018-01-16  Richard Sandiford  <richard.sandiford@linaro.org>
8963
8964         * fold-const.c (fold_ternary_loc): Construct the vec_perm_indices
8965         in a separate statement.
8966
8967 2018-01-16  Richard Sandiford  <richard.sandiford@linaro.org>
8968
8969         PR tree-optimization/83847
8970         * tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Don't
8971         group gathers and scatters.
8972
8973 2018-01-16  Jakub Jelinek  <jakub@redhat.com>
8974
8975         PR rtl-optimization/86620
8976         * params.def (max-sched-ready-insns): Bump minimum value to 1.
8977
8978         PR rtl-optimization/83213
8979         * recog.c (peep2_attempt): Copy over CROSSING_JUMP_P from peepinsn
8980         to last if both are JUMP_INSNs.
8981
8982         PR tree-optimization/83843
8983         * gimple-ssa-store-merging.c
8984         (imm_store_chain_info::output_merged_store): Handle bit_not_p on
8985         store_immediate_info for bswap/nop orig_stores.
8986
8987 2018-01-15  Andrew Waterman  <andrew@sifive.com>
8988
8989         * config/riscv/riscv.c (riscv_rtx_costs) <MULT>: Increase cost if
8990         !TARGET_MUL.
8991         <UDIV>: Increase cost if !TARGET_DIV.
8992
8993 2018-01-15  Segher Boessenkool  <segher@kernel.crashing.org>
8994
8995         * config/rs6000/rs6000.md (define_attr "type"): Remove delayed_cr.
8996         (define_attr "cr_logical_3op"): New.
8997         (cceq_ior_compare): Adjust.
8998         (cceq_ior_compare_complement): Adjust.
8999         (*cceq_rev_compare): Adjust.
9000         * config/rs6000/rs6000.c (rs6000_adjust_cost): Adjust.
9001         (is_cracked_insn): Adjust.
9002         (insn_must_be_first_in_group): Adjust.
9003         * config/rs6000/40x.md: Adjust.
9004         * config/rs6000/440.md: Adjust.
9005         * config/rs6000/476.md: Adjust.
9006         * config/rs6000/601.md: Adjust.
9007         * config/rs6000/603.md: Adjust.
9008         * config/rs6000/6xx.md: Adjust.
9009         * config/rs6000/7450.md: Adjust.
9010         * config/rs6000/7xx.md: Adjust.
9011         * config/rs6000/8540.md: Adjust.
9012         * config/rs6000/cell.md: Adjust.
9013         * config/rs6000/e300c2c3.md: Adjust.
9014         * config/rs6000/e500mc.md: Adjust.
9015         * config/rs6000/e500mc64.md: Adjust.
9016         * config/rs6000/e5500.md: Adjust.
9017         * config/rs6000/e6500.md: Adjust.
9018         * config/rs6000/mpc.md: Adjust.
9019         * config/rs6000/power4.md: Adjust.
9020         * config/rs6000/power5.md: Adjust.
9021         * config/rs6000/power6.md: Adjust.
9022         * config/rs6000/power7.md: Adjust.
9023         * config/rs6000/power8.md: Adjust.
9024         * config/rs6000/power9.md: Adjust.
9025         * config/rs6000/rs64.md: Adjust.
9026         * config/rs6000/titan.md: Adjust.
9027
9028 2018-01-15  H.J. Lu  <hongjiu.lu@intel.com>
9029
9030         * config/i386/predicates.md (indirect_branch_operand): Rewrite
9031         ix86_indirect_branch_register logic.
9032
9033 2018-01-15  H.J. Lu  <hongjiu.lu@intel.com>
9034
9035         * config/i386/constraints.md (Bs): Update
9036         ix86_indirect_branch_register check.  Don't check
9037         ix86_indirect_branch_register with GOT_memory_operand.
9038         (Bw): Likewise.
9039         * config/i386/predicates.md (GOT_memory_operand): Don't check
9040         ix86_indirect_branch_register here.
9041         (GOT32_symbol_operand): Likewise.
9042
9043 2018-01-15  H.J. Lu  <hongjiu.lu@intel.com>
9044
9045         * config/i386/predicates.md (constant_call_address_operand):
9046         Rewrite ix86_indirect_branch_register logic.
9047         (sibcall_insn_operand): Likewise.
9048
9049 2018-01-15  H.J. Lu  <hongjiu.lu@intel.com>
9050
9051         * config/i386/constraints.md (Bs): Replace
9052         ix86_indirect_branch_thunk_register with
9053         ix86_indirect_branch_register.
9054         (Bw): Likewise.
9055         * config/i386/i386.md (indirect_jump): Likewise.
9056         (tablejump): Likewise.
9057         (*sibcall_memory): Likewise.
9058         (*sibcall_value_memory): Likewise.
9059         Peepholes of indirect call and jump via memory: Likewise.
9060         * config/i386/i386.opt: Likewise.
9061         * config/i386/predicates.md (indirect_branch_operand): Likewise.
9062         (GOT_memory_operand): Likewise.
9063         (call_insn_operand): Likewise.
9064         (sibcall_insn_operand): Likewise.
9065         (GOT32_symbol_operand): Likewise.
9066
9067 2018-01-15  Jakub Jelinek  <jakub@redhat.com>
9068
9069         PR middle-end/83837
9070         * omp-expand.c (expand_omp_atomic_pipeline): Use loaded_val
9071         type rather than type addr's type points to.
9072         (expand_omp_atomic_mutex): Likewise.
9073         (expand_omp_atomic): Likewise.
9074
9075 2018-01-15  H.J. Lu  <hongjiu.lu@intel.com>
9076
9077         PR target/83839
9078         * config/i386/i386.c (output_indirect_thunk_function): Use
9079         ASM_OUTPUT_LABEL, instead of ASM_OUTPUT_DEF, for TARGET_MACHO
9080         for  __x86_return_thunk.
9081
9082 2018-01-15  Richard Biener  <rguenther@suse.de>
9083
9084         PR middle-end/83850
9085         * expmed.c (extract_bit_field_1): Fix typo.
9086
9087 2018-01-15  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
9088
9089         PR target/83687
9090         * config/arm/iterators.md (VF): New mode iterator.
9091         * config/arm/neon.md (neon_vabd<mode>_2): Use the above.
9092         Remove integer-related logic from pattern.
9093         (neon_vabd<mode>_3): Likewise.
9094
9095 2018-01-15  Jakub Jelinek  <jakub@redhat.com>
9096
9097         PR middle-end/82694
9098         * common.opt (fstrict-overflow): No longer an alias.
9099         (fwrapv-pointer): New option.
9100         * tree.h (TYPE_OVERFLOW_WRAPS, TYPE_OVERFLOW_UNDEFINED): Define
9101         also for pointer types based on flag_wrapv_pointer.
9102         * opts.c (common_handle_option) <case OPT_fstrict_overflow>: Set
9103         opts->x_flag_wrap[pv] to !value, clear opts->x_flag_trapv if
9104         opts->x_flag_wrapv got set.
9105         * fold-const.c (fold_comparison, fold_binary_loc): Revert 2017-08-01
9106         changes, just use TYPE_OVERFLOW_UNDEFINED on pointer type instead of
9107         POINTER_TYPE_OVERFLOW_UNDEFINED.
9108         * match.pd: Likewise in address comparison pattern.
9109         * doc/invoke.texi: Document -fwrapv and -fstrict-overflow.
9110
9111 2018-01-15  Richard Biener  <rguenther@suse.de>
9112
9113         PR lto/83804
9114         * tree.c (free_lang_data_in_type): Always unlink TYPE_DECLs
9115         from TYPE_FIELDS.  Free TYPE_BINFO if not used by devirtualization.
9116         Reset type names to their identifier if their TYPE_DECL doesn't
9117         have linkage (and thus is used for ODR and devirt).
9118         (save_debug_info_for_decl): Remove.
9119         (save_debug_info_for_type): Likewise.
9120         (add_tree_to_fld_list): Adjust.
9121         * tree-pretty-print.c (dump_generic_node): Make dumping of
9122         type names more robust.
9123
9124 2018-01-15  Richard Biener  <rguenther@suse.de>
9125
9126         * BASE-VER: Bump to 8.0.1.
9127
9128 2018-01-14  Martin Sebor  <msebor@redhat.com>
9129
9130         PR other/83508
9131         * builtins.c (check_access): Avoid warning when the no-warning bit
9132         is set.
9133
9134 2018-01-14  Cory Fields  <cory-nospam-@coryfields.com>
9135
9136         * tree-ssa-loop-im.c (sort_bbs_in_loop_postorder_cmp): Stabilize sort.
9137         * ira-color (allocno_hard_regs_compare): Likewise.
9138
9139 2018-01-14  Nathan Rossi  <nathan@nathanrossi.com>
9140
9141         PR target/83013
9142         * config/microblaze/microblaze.c (microblaze_asm_output_ident):
9143         Use .pushsection/.popsection.
9144
9145 2018-01-14  Martin Sebor  <msebor@redhat.com>
9146
9147         PR c++/81327
9148         * doc/invoke.texi (-Wlass-memaccess): Document suppression by casting.
9149
9150 2018-01-14  Jakub Jelinek  <jakub@redhat.com>
9151
9152         * config.gcc (i[34567]86-*-*): Remove one duplicate gfniintrin.h
9153         entry from extra_headers.
9154         (x86_64-*-*): Remove two duplicate gfniintrin.h entries from
9155         extra_headers, make the list bitwise identical to the i?86-*-* one.
9156
9157 2018-01-14  H.J. Lu  <hongjiu.lu@intel.com>
9158
9159         * config/i386/i386.c (ix86_set_indirect_branch_type): Disallow
9160         -mcmodel=large with -mindirect-branch=thunk,
9161         -mindirect-branch=thunk-extern, -mfunction-return=thunk and
9162         -mfunction-return=thunk-extern.
9163         * doc/invoke.texi: Document -mcmodel=large is incompatible with
9164         -mindirect-branch=thunk, -mindirect-branch=thunk-extern,
9165         -mfunction-return=thunk and -mfunction-return=thunk-extern.
9166
9167 2018-01-14  H.J. Lu  <hongjiu.lu@intel.com>
9168
9169         * config/i386/i386.c (print_reg): Print the name of the full
9170         integer register without '%'.
9171         (ix86_print_operand): Handle 'V'.
9172          * doc/extend.texi: Document 'V' modifier.
9173
9174 2018-01-14  H.J. Lu  <hongjiu.lu@intel.com>
9175
9176         * config/i386/constraints.md (Bs): Disallow memory operand for
9177         -mindirect-branch-register.
9178         (Bw): Likewise.
9179         * config/i386/predicates.md (indirect_branch_operand): Likewise.
9180         (GOT_memory_operand): Likewise.
9181         (call_insn_operand): Likewise.
9182         (sibcall_insn_operand): Likewise.
9183         (GOT32_symbol_operand): Likewise.
9184         * config/i386/i386.md (indirect_jump): Call convert_memory_address
9185         for -mindirect-branch-register.
9186         (tablejump): Likewise.
9187         (*sibcall_memory): Likewise.
9188         (*sibcall_value_memory): Likewise.
9189         Disallow peepholes of indirect call and jump via memory for
9190         -mindirect-branch-register.
9191         (*call_pop): Replace m with Bw.
9192         (*call_value_pop): Likewise.
9193         (*sibcall_pop_memory): Replace m with Bs.
9194         * config/i386/i386.opt (mindirect-branch-register): New option.
9195         * doc/invoke.texi: Document -mindirect-branch-register option.
9196
9197 2018-01-14  H.J. Lu  <hongjiu.lu@intel.com>
9198
9199         * config/i386/i386-protos.h (ix86_output_function_return): New.
9200         * config/i386/i386.c (ix86_set_indirect_branch_type): Also
9201         set function_return_type.
9202         (indirect_thunk_name): Add ret_p to indicate thunk for function
9203         return.
9204         (output_indirect_thunk_function): Pass false to
9205         indirect_thunk_name.
9206         (ix86_output_indirect_branch_via_reg): Likewise.
9207         (ix86_output_indirect_branch_via_push): Likewise.
9208         (output_indirect_thunk_function): Create alias for function
9209         return thunk if regno < 0.
9210         (ix86_output_function_return): New function.
9211         (ix86_handle_fndecl_attribute): Handle function_return.
9212         (ix86_attribute_table): Add function_return.
9213         * config/i386/i386.h (machine_function): Add
9214         function_return_type.
9215         * config/i386/i386.md (simple_return_internal): Use
9216         ix86_output_function_return.
9217         (simple_return_internal_long): Likewise.
9218         * config/i386/i386.opt (mfunction-return=): New option.
9219         (indirect_branch): Mention -mfunction-return=.
9220         * doc/extend.texi: Document function_return function attribute.
9221         * doc/invoke.texi: Document -mfunction-return= option.
9222
9223 2018-01-14  H.J. Lu  <hongjiu.lu@intel.com>
9224
9225         * config/i386/i386-opts.h (indirect_branch): New.
9226         * config/i386/i386-protos.h (ix86_output_indirect_jmp): Likewise.
9227         * config/i386/i386.c (ix86_using_red_zone): Disallow red-zone
9228         with local indirect jump when converting indirect call and jump.
9229         (ix86_set_indirect_branch_type): New.
9230         (ix86_set_current_function): Call ix86_set_indirect_branch_type.
9231         (indirectlabelno): New.
9232         (indirect_thunk_needed): Likewise.
9233         (indirect_thunk_bnd_needed): Likewise.
9234         (indirect_thunks_used): Likewise.
9235         (indirect_thunks_bnd_used): Likewise.
9236         (INDIRECT_LABEL): Likewise.
9237         (indirect_thunk_name): Likewise.
9238         (output_indirect_thunk): Likewise.
9239         (output_indirect_thunk_function): Likewise.
9240         (ix86_output_indirect_branch_via_reg): Likewise.
9241         (ix86_output_indirect_branch_via_push): Likewise.
9242         (ix86_output_indirect_branch): Likewise.
9243         (ix86_output_indirect_jmp): Likewise.
9244         (ix86_code_end): Call output_indirect_thunk_function if needed.
9245         (ix86_output_call_insn): Call ix86_output_indirect_branch if
9246         needed.
9247         (ix86_handle_fndecl_attribute): Handle indirect_branch.
9248         (ix86_attribute_table): Add indirect_branch.
9249         * config/i386/i386.h (machine_function): Add indirect_branch_type
9250         and has_local_indirect_jump.
9251         * config/i386/i386.md (indirect_jump): Set has_local_indirect_jump
9252         to true.
9253         (tablejump): Likewise.
9254         (*indirect_jump): Use ix86_output_indirect_jmp.
9255         (*tablejump_1): Likewise.
9256         (simple_return_indirect_internal): Likewise.
9257         * config/i386/i386.opt (mindirect-branch=): New option.
9258         (indirect_branch): New.
9259         (keep): Likewise.
9260         (thunk): Likewise.
9261         (thunk-inline): Likewise.
9262         (thunk-extern): Likewise.
9263         * doc/extend.texi: Document indirect_branch function attribute.
9264         * doc/invoke.texi: Document -mindirect-branch= option.
9265
9266 2018-01-14  Jan Hubicka  <hubicka@ucw.cz>
9267
9268         PR ipa/83051
9269         * ipa-inline.c (edge_badness): Tolerate roundoff errors.
9270
9271 2018-01-14  Richard Sandiford  <richard.sandiford@linaro.org>
9272
9273         * ipa-inline.c (want_inline_small_function_p): Return false if
9274         inlining has already failed with CIF_FINAL_ERROR.
9275         (update_caller_keys): Call want_inline_small_function_p before
9276         can_inline_edge_p.
9277         (update_callee_keys): Likewise.
9278
9279 2018-01-10  Kelvin Nilsen  <kelvin@gcc.gnu.org>
9280
9281         * config/rs6000/rs6000-p8swap.c (rs6000_sum_of_two_registers_p):
9282         New function.
9283         (rs6000_quadword_masked_address_p): Likewise.
9284         (quad_aligned_load_p): Likewise.
9285         (quad_aligned_store_p): Likewise.
9286         (const_load_sequence_p): Add comment to describe the outer-most loop.
9287         (mimic_memory_attributes_and_flags): New function.
9288         (rs6000_gen_stvx): Likewise.
9289         (replace_swapped_aligned_store): Likewise.
9290         (rs6000_gen_lvx): Likewise.
9291         (replace_swapped_aligned_load): Likewise.
9292         (replace_swapped_load_constant): Capitalize argument name in
9293         comment describing this function.
9294         (rs6000_analyze_swaps): Add a third pass to search for vector loads
9295         and stores that access quad-word aligned addresses and replace
9296         with stvx or lvx instructions when appropriate.
9297         * config/rs6000/rs6000-protos.h (rs6000_sum_of_two_registers_p):
9298         New function prototype.
9299         (rs6000_quadword_masked_address_p): Likewise.
9300         (rs6000_gen_lvx): Likewise.
9301         (rs6000_gen_stvx): Likewise.
9302         * config/rs6000/vsx.md (*vsx_le_perm_load_<mode>): For modes
9303         VSX_D (V2DF, V2DI), modify this split to select lvx instruction
9304         when memory address is aligned.
9305         (*vsx_le_perm_load_<mode>): For modes VSX_W (V4SF, V4SI), modify
9306         this split to select lvx instruction when memory address is aligned.
9307         (*vsx_le_perm_load_v8hi): Modify this split to select lvx
9308         instruction when memory address is aligned.
9309         (*vsx_le_perm_load_v16qi): Likewise.
9310         (four unnamed splitters): Modify to select the stvx instruction
9311         when memory is aligned.
9312
9313 2018-01-13  Jan Hubicka  <hubicka@ucw.cz>
9314
9315         * predict.c (determine_unlikely_bbs): Handle correctly BBs
9316         which appears in the queue multiple times.
9317
9318 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
9319             Alan Hayward  <alan.hayward@arm.com>
9320             David Sherwood  <david.sherwood@arm.com>
9321
9322         * tree-vectorizer.h (vec_lower_bound): New structure.
9323         (_loop_vec_info): Add check_nonzero and lower_bounds.
9324         (LOOP_VINFO_CHECK_NONZERO): New macro.
9325         (LOOP_VINFO_LOWER_BOUNDS): Likewise.
9326         (LOOP_REQUIRES_VERSIONING_FOR_ALIAS): Check lower_bounds too.
9327         * tree-data-ref.h (dr_with_seg_len): Add access_size and align
9328         fields.  Make seg_len the distance travelled, not including the
9329         access size.
9330         (dr_direction_indicator): Declare.
9331         (dr_zero_step_indicator): Likewise.
9332         (dr_known_forward_stride_p): Likewise.
9333         * tree-data-ref.c: Include stringpool.h, tree-vrp.h and
9334         tree-ssanames.h.
9335         (runtime_alias_check_p): Allow runtime alias checks with
9336         variable strides.
9337         (operator ==): Compare access_size and align.
9338         (prune_runtime_alias_test_list): Rework for new distinction between
9339         the access_size and seg_len.
9340         (create_intersect_range_checks_index): Likewise.  Cope with polynomial
9341         segment lengths.
9342         (get_segment_min_max): New function.
9343         (create_intersect_range_checks): Use it.
9344         (dr_step_indicator): New function.
9345         (dr_direction_indicator): Likewise.
9346         (dr_zero_step_indicator): Likewise.
9347         (dr_known_forward_stride_p): Likewise.
9348         * tree-loop-distribution.c (data_ref_segment_size): Return
9349         DR_STEP * (niters - 1).
9350         (compute_alias_check_pairs): Update call to the dr_with_seg_len
9351         constructor.
9352         * tree-vect-data-refs.c (vect_check_nonzero_value): New function.
9353         (vect_preserves_scalar_order_p): New function, split out from...
9354         (vect_analyze_data_ref_dependence): ...here.  Check for zero steps.
9355         (vect_vfa_segment_size): Return DR_STEP * (length_factor - 1).
9356         (vect_vfa_access_size): New function.
9357         (vect_vfa_align): Likewise.
9358         (vect_compile_time_alias): Take access_size_a and access_b arguments.
9359         (dump_lower_bound): New function.
9360         (vect_check_lower_bound): Likewise.
9361         (vect_small_gap_p): Likewise.
9362         (vectorizable_with_step_bound_p): Likewise.
9363         (vect_prune_runtime_alias_test_list): Ignore cross-iteration
9364         depencies if the vectorization factor is 1.  Convert the checks
9365         for nonzero steps into checks on the bounds of DR_STEP.  Try using
9366         a bunds check for variable steps if the minimum required step is
9367         relatively small. Update calls to the dr_with_seg_len
9368         constructor and to vect_compile_time_alias.
9369         * tree-vect-loop-manip.c (vect_create_cond_for_lower_bounds): New
9370         function.
9371         (vect_loop_versioning): Call it.
9372         * tree-vect-loop.c (vect_analyze_loop_2): Clear LOOP_VINFO_LOWER_BOUNDS
9373         when retrying.
9374         (vect_estimate_min_profitable_iters): Account for any bounds checks.
9375
9376 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
9377             Alan Hayward  <alan.hayward@arm.com>
9378             David Sherwood  <david.sherwood@arm.com>
9379
9380         * doc/sourcebuild.texi (vect_scatter_store): Document.
9381         * optabs.def (scatter_store_optab, mask_scatter_store_optab): New
9382         optabs.
9383         * doc/md.texi (scatter_store@var{m}, mask_scatter_store@var{m}):
9384         Document.
9385         * genopinit.c (main): Add supports_vec_scatter_store and
9386         supports_vec_scatter_store_cached to target_optabs.
9387         * gimple.h (gimple_expr_type): Handle IFN_SCATTER_STORE and
9388         IFN_MASK_SCATTER_STORE.
9389         * internal-fn.def (SCATTER_STORE, MASK_SCATTER_STORE): New internal
9390         functions.
9391         * internal-fn.h (internal_store_fn_p): Declare.
9392         (internal_fn_stored_value_index): Likewise.
9393         * internal-fn.c (scatter_store_direct): New macro.
9394         (expand_scatter_store_optab_fn): New function.
9395         (direct_scatter_store_optab_supported_p): New macro.
9396         (internal_store_fn_p): New function.
9397         (internal_gather_scatter_fn_p): Handle IFN_SCATTER_STORE and
9398         IFN_MASK_SCATTER_STORE.
9399         (internal_fn_mask_index): Likewise.
9400         (internal_fn_stored_value_index): New function.
9401         (internal_gather_scatter_fn_supported_p): Adjust operand numbers
9402         for scatter stores.
9403         * optabs-query.h (supports_vec_scatter_store_p): Declare.
9404         * optabs-query.c (supports_vec_scatter_store_p): New function.
9405         * tree-vectorizer.h (vect_get_store_rhs): Declare.
9406         * tree-vect-data-refs.c (vect_analyze_data_ref_access): Return
9407         true for scatter stores.
9408         (vect_gather_scatter_fn_p): Handle scatter stores too.
9409         (vect_check_gather_scatter): Consider using scatter stores if
9410         supports_vec_scatter_store_p.
9411         * tree-vect-patterns.c (vect_try_gather_scatter_pattern): Handle
9412         scatter stores too.
9413         * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
9414         internal_fn_stored_value_index.
9415         (check_load_store_masking): Handle scatter stores too.
9416         (vect_get_store_rhs): Make public.
9417         (vectorizable_call): Use internal_store_fn_p.
9418         (vectorizable_store): Handle scatter store internal functions.
9419         (vect_transform_stmt): Compare GROUP_STORE_COUNT with GROUP_SIZE
9420         when deciding whether the end of the group has been reached.
9421         * config/aarch64/aarch64.md (UNSPEC_ST1_SCATTER): New unspec.
9422         * config/aarch64/aarch64-sve.md (scatter_store<mode>): New expander.
9423         (mask_scatter_store<mode>): New insns.
9424
9425 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
9426             Alan Hayward  <alan.hayward@arm.com>
9427             David Sherwood  <david.sherwood@arm.com>
9428
9429         * tree-vectorizer.h (vect_gather_scatter_fn_p): Declare.
9430         * tree-vect-data-refs.c (vect_gather_scatter_fn_p): Make public.
9431         * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): New
9432         function.
9433         (vect_use_strided_gather_scatters_p): Take a masked_p argument.
9434         Use vect_truncate_gather_scatter_offset if we can't treat the
9435         operation as a normal gather load or scatter store.
9436         (get_group_load_store_type): Take the gather_scatter_info
9437         as argument.  Try using a gather load or scatter store for
9438         single-element groups.
9439         (get_load_store_type): Update calls to get_group_load_store_type
9440         and vect_use_strided_gather_scatters_p.
9441
9442 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
9443             Alan Hayward  <alan.hayward@arm.com>
9444             David Sherwood  <david.sherwood@arm.com>
9445
9446         * tree-vectorizer.h (vect_create_data_ref_ptr): Take an extra
9447         optional tree argument.
9448         * tree-vect-data-refs.c (vect_check_gather_scatter): Check for
9449         null target hooks.
9450         (vect_create_data_ref_ptr): Take the iv_step as an optional argument,
9451         but continue to use the current value as a fallback.
9452         (bump_vector_ptr): Use operand_equal_p rather than tree_int_cst_compare
9453         to compare the updates.
9454         * tree-vect-stmts.c (vect_use_strided_gather_scatters_p): New function.
9455         (get_load_store_type): Use it when handling a strided access.
9456         (vect_get_strided_load_store_ops): New function.
9457         (vect_get_data_ptr_increment): Likewise.
9458         (vectorizable_load): Handle strided gather loads.  Always pass
9459         a step to vect_create_data_ref_ptr and bump_vector_ptr.
9460
9461 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
9462             Alan Hayward  <alan.hayward@arm.com>
9463             David Sherwood  <david.sherwood@arm.com>
9464
9465         * doc/md.texi (gather_load@var{m}): Document.
9466         (mask_gather_load@var{m}): Likewise.
9467         * genopinit.c (main): Add supports_vec_gather_load and
9468         supports_vec_gather_load_cached to target_optabs.
9469         * optabs-tree.c (init_tree_optimization_optabs): Use
9470         ggc_cleared_alloc to allocate target_optabs.
9471         * optabs.def (gather_load_optab, mask_gather_laod_optab): New optabs.
9472         * internal-fn.def (GATHER_LOAD, MASK_GATHER_LOAD): New internal
9473         functions.
9474         * internal-fn.h (internal_load_fn_p): Declare.
9475         (internal_gather_scatter_fn_p): Likewise.
9476         (internal_fn_mask_index): Likewise.
9477         (internal_gather_scatter_fn_supported_p): Likewise.
9478         * internal-fn.c (gather_load_direct): New macro.
9479         (expand_gather_load_optab_fn): New function.
9480         (direct_gather_load_optab_supported_p): New macro.
9481         (direct_internal_fn_optab): New function.
9482         (internal_load_fn_p): Likewise.
9483         (internal_gather_scatter_fn_p): Likewise.
9484         (internal_fn_mask_index): Likewise.
9485         (internal_gather_scatter_fn_supported_p): Likewise.
9486         * optabs-query.c (supports_at_least_one_mode_p): New function.
9487         (supports_vec_gather_load_p): Likewise.
9488         * optabs-query.h (supports_vec_gather_load_p): Declare.
9489         * tree-vectorizer.h (gather_scatter_info): Add ifn, element_type
9490         and memory_type field.
9491         (NUM_PATTERNS): Bump to 15.
9492         * tree-vect-data-refs.c: Include internal-fn.h.
9493         (vect_gather_scatter_fn_p): New function.
9494         (vect_describe_gather_scatter_call): Likewise.
9495         (vect_check_gather_scatter): Try using internal functions for
9496         gather loads.  Recognize existing calls to a gather load function.
9497         (vect_analyze_data_refs): Consider using gather loads if
9498         supports_vec_gather_load_p.
9499         * tree-vect-patterns.c (vect_get_load_store_mask): New function.
9500         (vect_get_gather_scatter_offset_type): Likewise.
9501         (vect_convert_mask_for_vectype): Likewise.
9502         (vect_add_conversion_to_patterm): Likewise.
9503         (vect_try_gather_scatter_pattern): Likewise.
9504         (vect_recog_gather_scatter_pattern): New pattern recognizer.
9505         (vect_vect_recog_func_ptrs): Add it.
9506         * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
9507         internal_fn_mask_index and internal_gather_scatter_fn_p.
9508         (check_load_store_masking): Take the gather_scatter_info as an
9509         argument and handle gather loads.
9510         (vect_get_gather_scatter_ops): New function.
9511         (vectorizable_call): Check internal_load_fn_p.
9512         (vectorizable_load): Likewise.  Handle gather load internal
9513         functions.
9514         (vectorizable_store): Update call to check_load_store_masking.
9515         * config/aarch64/aarch64.md (UNSPEC_LD1_GATHER): New unspec.
9516         * config/aarch64/iterators.md (SVE_S, SVE_D): New mode iterators.
9517         * config/aarch64/predicates.md (aarch64_gather_scale_operand_w)
9518         (aarch64_gather_scale_operand_d): New predicates.
9519         * config/aarch64/aarch64-sve.md (gather_load<mode>): New expander.
9520         (mask_gather_load<mode>): New insns.
9521
9522 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
9523             Alan Hayward  <alan.hayward@arm.com>
9524             David Sherwood  <david.sherwood@arm.com>
9525
9526         * optabs.def (fold_left_plus_optab): New optab.
9527         * doc/md.texi (fold_left_plus_@var{m}): Document.
9528         * internal-fn.def (IFN_FOLD_LEFT_PLUS): New internal function.
9529         * internal-fn.c (fold_left_direct): Define.
9530         (expand_fold_left_optab_fn): Likewise.
9531         (direct_fold_left_optab_supported_p): Likewise.
9532         * fold-const-call.c (fold_const_fold_left): New function.
9533         (fold_const_call): Use it to fold CFN_FOLD_LEFT_PLUS.
9534         * tree-parloops.c (valid_reduction_p): New function.
9535         (gather_scalar_reductions): Use it.
9536         * tree-vectorizer.h (FOLD_LEFT_REDUCTION): New vect_reduction_type.
9537         (vect_finish_replace_stmt): Declare.
9538         * tree-vect-loop.c (fold_left_reduction_fn): New function.
9539         (needs_fold_left_reduction_p): New function, split out from...
9540         (vect_is_simple_reduction): ...here.  Accept reductions that
9541         forbid reassociation, but give them type FOLD_LEFT_REDUCTION.
9542         (vect_force_simple_reduction): Also store the reduction type in
9543         the assignment's STMT_VINFO_REDUC_TYPE.
9544         (vect_model_reduction_cost): Handle FOLD_LEFT_REDUCTION.
9545         (merge_with_identity): New function.
9546         (vect_expand_fold_left): Likewise.
9547         (vectorize_fold_left_reduction): Likewise.
9548         (vectorizable_reduction): Handle FOLD_LEFT_REDUCTION.  Leave the
9549         scalar phi in place for it.  Check for target support and reject
9550         cases that would reassociate the operation.  Defer the transform
9551         phase to vectorize_fold_left_reduction.
9552         * config/aarch64/aarch64.md (UNSPEC_FADDA): New unspec.
9553         * config/aarch64/aarch64-sve.md (fold_left_plus_<mode>): New expander.
9554         (*fold_left_plus_<mode>, *pred_fold_left_plus_<mode>): New insns.
9555
9556 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
9557
9558         * tree-if-conv.c (predicate_mem_writes): Remove redundant
9559         call to ifc_temp_var.
9560
9561 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
9562             Alan Hayward  <alan.hayward@arm.com>
9563             David Sherwood  <david.sherwood@arm.com>
9564
9565         * target.def (legitimize_address_displacement): Take the original
9566         offset as a poly_int.
9567         * targhooks.h (default_legitimize_address_displacement): Update
9568         accordingly.
9569         * targhooks.c (default_legitimize_address_displacement): Likewise.
9570         * doc/tm.texi: Regenerate.
9571         * lra-constraints.c (base_plus_disp_to_reg): Take the displacement
9572         as an argument, moving assert of ad->disp == ad->disp_term to...
9573         (process_address_1): ...here.  Update calls to base_plus_disp_to_reg.
9574         Try calling targetm.legitimize_address_displacement before expanding
9575         the address rather than afterwards, and adjust for the new interface.
9576         * config/aarch64/aarch64.c (aarch64_legitimize_address_displacement):
9577         Match the new hook interface.  Handle SVE addresses.
9578         * config/sh/sh.c (sh_legitimize_address_displacement): Make the
9579         new hook interface.
9580
9581 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
9582
9583         * Makefile.in (OBJS): Add early-remat.o.
9584         * target.def (select_early_remat_modes): New hook.
9585         * doc/tm.texi.in (TARGET_SELECT_EARLY_REMAT_MODES): New hook.
9586         * doc/tm.texi: Regenerate.
9587         * targhooks.h (default_select_early_remat_modes): Declare.
9588         * targhooks.c (default_select_early_remat_modes): New function.
9589         * timevar.def (TV_EARLY_REMAT): New timevar.
9590         * passes.def (pass_early_remat): New pass.
9591         * tree-pass.h (make_pass_early_remat): Declare.
9592         * early-remat.c: New file.
9593         * config/aarch64/aarch64.c (aarch64_select_early_remat_modes): New
9594         function.
9595         (TARGET_SELECT_EARLY_REMAT_MODES): Define.
9596
9597 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
9598             Alan Hayward  <alan.hayward@arm.com>
9599             David Sherwood  <david.sherwood@arm.com>
9600
9601         * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Replace
9602         vfm1 with a bound_epilog parameter.
9603         (vect_do_peeling): Update calls accordingly, and move the prologue
9604         call earlier in the function.  Treat the base bound_epilog as 0 for
9605         fully-masked loops and retain vf - 1 for other loops.  Add 1 to
9606         this base when peeling for gaps.
9607         * tree-vect-loop.c (vect_analyze_loop_2): Allow peeling for gaps
9608         with fully-masked loops.
9609         (vect_estimate_min_profitable_iters): Handle the single peeled
9610         iteration in that case.
9611
9612 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
9613             Alan Hayward  <alan.hayward@arm.com>
9614             David Sherwood  <david.sherwood@arm.com>
9615
9616         * tree-vect-data-refs.c (vect_analyze_group_access_1): Allow
9617         single-element interleaving even if the size is not a power of 2.
9618         * tree-vect-stmts.c (get_load_store_type): Disallow elementwise
9619         accesses for single-element interleaving if the group size is
9620         not a power of 2.
9621
9622 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
9623             Alan Hayward  <alan.hayward@arm.com>
9624             David Sherwood  <david.sherwood@arm.com>
9625
9626         * doc/md.texi (fold_extract_last_@var{m}): Document.
9627         * doc/sourcebuild.texi (vect_fold_extract_last): Likewise.
9628         * optabs.def (fold_extract_last_optab): New optab.
9629         * internal-fn.def (FOLD_EXTRACT_LAST): New internal function.
9630         * internal-fn.c (fold_extract_direct): New macro.
9631         (expand_fold_extract_optab_fn): Likewise.
9632         (direct_fold_extract_optab_supported_p): Likewise.
9633         * tree-vectorizer.h (EXTRACT_LAST_REDUCTION): New vect_reduction_type.
9634         * tree-vect-loop.c (vect_model_reduction_cost): Handle
9635         EXTRACT_LAST_REDUCTION.
9636         (get_initial_def_for_reduction): Do not create an initial vector
9637         for EXTRACT_LAST_REDUCTION reductions.
9638         (vectorizable_reduction): Leave the scalar phi in place for
9639         EXTRACT_LAST_REDUCTIONs.  Try using EXTRACT_LAST_REDUCTION
9640         ahead of INTEGER_INDUC_COND_REDUCTION.  Do not check for an
9641         epilogue code for EXTRACT_LAST_REDUCTION and defer the
9642         transform phase to vectorizable_condition.
9643         * tree-vect-stmts.c (vect_finish_stmt_generation_1): New function,
9644         split out from...
9645         (vect_finish_stmt_generation): ...here.
9646         (vect_finish_replace_stmt): New function.
9647         (vectorizable_condition): Handle EXTRACT_LAST_REDUCTION.
9648         * config/aarch64/aarch64-sve.md (fold_extract_last_<mode>): New
9649         pattern.
9650         * config/aarch64/aarch64.md (UNSPEC_CLASTB): New unspec.
9651
9652 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
9653             Alan Hayward  <alan.hayward@arm.com>
9654             David Sherwood  <david.sherwood@arm.com>
9655
9656         * doc/md.texi (extract_last_@var{m}): Document.
9657         * optabs.def (extract_last_optab): New optab.
9658         * internal-fn.def (EXTRACT_LAST): New internal function.
9659         * internal-fn.c (cond_unary_direct): New macro.
9660         (expand_cond_unary_optab_fn): Likewise.
9661         (direct_cond_unary_optab_supported_p): Likewise.
9662         * tree-vect-loop.c (vectorizable_live_operation): Allow fully-masked
9663         loops using EXTRACT_LAST.
9664         * config/aarch64/aarch64-sve.md (aarch64_sve_lastb<mode>): Rename to...
9665         (extract_last_<mode>): ...this optab.
9666         (vec_extract<mode><Vel>): Update accordingly.
9667
9668 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
9669             Alan Hayward  <alan.hayward@arm.com>
9670             David Sherwood  <david.sherwood@arm.com>
9671
9672         * target.def (empty_mask_is_expensive): New hook.
9673         * doc/tm.texi.in (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): New hook.
9674         * doc/tm.texi: Regenerate.
9675         * targhooks.h (default_empty_mask_is_expensive): Declare.
9676         * targhooks.c (default_empty_mask_is_expensive): New function.
9677         * tree-vectorizer.c (vectorize_loops): Only call optimize_mask_stores
9678         if the target says that empty masks are expensive.
9679         * config/aarch64/aarch64.c (aarch64_empty_mask_is_expensive):
9680         New function.
9681         (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): Redefine.
9682
9683 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
9684             Alan Hayward  <alan.hayward@arm.com>
9685             David Sherwood  <david.sherwood@arm.com>
9686
9687         * tree-vectorizer.h (_loop_vec_info::mask_skip_niters): New field.
9688         (LOOP_VINFO_MASK_SKIP_NITERS): New macro.
9689         (vect_use_loop_mask_for_alignment_p): New function.
9690         (vect_prepare_for_masked_peels, vect_gen_while_not): Declare.
9691         * tree-vect-loop-manip.c (vect_set_loop_masks_directly): Add an
9692         niters_skip argument.  Make sure that the first niters_skip elements
9693         of the first iteration are inactive.
9694         (vect_set_loop_condition_masked): Handle LOOP_VINFO_MASK_SKIP_NITERS.
9695         Update call to vect_set_loop_masks_directly.
9696         (get_misalign_in_elems): New function, split out from...
9697         (vect_gen_prolog_loop_niters): ...here.
9698         (vect_update_init_of_dr): Take a code argument that specifies whether
9699         the adjustment should be added or subtracted.
9700         (vect_update_init_of_drs): Likewise.
9701         (vect_prepare_for_masked_peels): New function.
9702         (vect_do_peeling): Skip prologue peeling if we're using a mask
9703         instead.  Update call to vect_update_inits_of_drs.
9704         * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
9705         mask_skip_niters.
9706         (vect_analyze_loop_2): Allow fully-masked loops with peeling for
9707         alignment.  Do not include the number of peeled iterations in
9708         the minimum threshold in that case.
9709         (vectorizable_induction): Adjust the start value down by
9710         LOOP_VINFO_MASK_SKIP_NITERS iterations.
9711         (vect_transform_loop): Call vect_prepare_for_masked_peels.
9712         Take the number of skipped iterations into account when calculating
9713         the loop bounds.
9714         * tree-vect-stmts.c (vect_gen_while_not): New function.
9715
9716 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
9717             Alan Hayward  <alan.hayward@arm.com>
9718             David Sherwood  <david.sherwood@arm.com>
9719
9720         * doc/sourcebuild.texi (vect_fully_masked): Document.
9721         * params.def (PARAM_MIN_VECT_LOOP_BOUND): Change minimum and
9722         default value to 0.
9723         * tree-vect-loop.c (vect_analyze_loop_costing): New function,
9724         split out from...
9725         (vect_analyze_loop_2): ...here. Don't check the vectorization
9726         factor against the number of loop iterations if the loop is
9727         fully-masked.
9728
9729 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
9730             Alan Hayward  <alan.hayward@arm.com>
9731             David Sherwood  <david.sherwood@arm.com>
9732
9733         * tree-ssa-loop-ivopts.c (USE_ADDRESS): Split into...
9734         (USE_REF_ADDRESS, USE_PTR_ADDRESS): ...these new use types.
9735         (dump_groups): Update accordingly.
9736         (iv_use::mem_type): New member variable.
9737         (address_p): New function.
9738         (record_use): Add a mem_type argument and initialize the new
9739         mem_type field.
9740         (record_group_use): Add a mem_type argument.  Use address_p.
9741         Remove obsolete null checks of base_object.  Update call to record_use.
9742         (find_interesting_uses_op): Update call to record_group_use.
9743         (find_interesting_uses_cond): Likewise.
9744         (find_interesting_uses_address): Likewise.
9745         (get_mem_type_for_internal_fn): New function.
9746         (find_address_like_use): Likewise.
9747         (find_interesting_uses_stmt): Try find_address_like_use before
9748         calling find_interesting_uses_op.
9749         (addr_offset_valid_p): Use the iv mem_type field as the type
9750         of the addressed memory.
9751         (add_autoinc_candidates): Likewise.
9752         (get_address_cost): Likewise.
9753         (split_small_address_groups_p): Use address_p.
9754         (split_address_groups): Likewise.
9755         (add_iv_candidate_for_use): Likewise.
9756         (autoinc_possible_for_pair): Likewise.
9757         (rewrite_groups): Likewise.
9758         (get_use_type): Check for USE_REF_ADDRESS instead of USE_ADDRESS.
9759         (determine_group_iv_cost): Update after split of USE_ADDRESS.
9760         (get_alias_ptr_type_for_ptr_address): New function.
9761         (rewrite_use_address): Rewrite address uses in calls that were
9762         identified by find_address_like_use.
9763
9764 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
9765             Alan Hayward  <alan.hayward@arm.com>
9766             David Sherwood  <david.sherwood@arm.com>
9767
9768         * expr.c (expand_expr_addr_expr_1): Handle ADDR_EXPRs of
9769         TARGET_MEM_REFs.
9770         * gimple-expr.h (is_gimple_addressable: Likewise.
9771         * gimple-expr.c (is_gimple_address): Likewise.
9772         * internal-fn.c (expand_call_mem_ref): New function.
9773         (expand_mask_load_optab_fn): Use it.
9774         (expand_mask_store_optab_fn): Likewise.
9775
9776 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
9777             Alan Hayward  <alan.hayward@arm.com>
9778             David Sherwood  <david.sherwood@arm.com>
9779
9780         * doc/md.texi (cond_add@var{mode}, cond_sub@var{mode})
9781         (cond_and@var{mode}, cond_ior@var{mode}, cond_xor@var{mode})
9782         (cond_smin@var{mode}, cond_smax@var{mode}, cond_umin@var{mode})
9783         (cond_umax@var{mode}): Document.
9784         * optabs.def (cond_add_optab, cond_sub_optab, cond_and_optab)
9785         (cond_ior_optab, cond_xor_optab, cond_smin_optab, cond_smax_optab)
9786         (cond_umin_optab, cond_umax_optab): New optabs.
9787         * internal-fn.def (COND_ADD, COND_SUB, COND_MIN, COND_MAX, COND_AND)
9788         (COND_IOR, COND_XOR): New internal functions.
9789         * internal-fn.h (get_conditional_internal_fn): Declare.
9790         * internal-fn.c (cond_binary_direct): New macro.
9791         (expand_cond_binary_optab_fn): Likewise.
9792         (direct_cond_binary_optab_supported_p): Likewise.
9793         (get_conditional_internal_fn): New function.
9794         * tree-vect-loop.c (vectorizable_reduction): Handle fully-masked loops.
9795         Cope with reduction statements that are vectorized as calls rather
9796         than assignments.
9797         * config/aarch64/aarch64-sve.md (cond_<optab><mode>): New insns.
9798         * config/aarch64/iterators.md (UNSPEC_COND_ADD, UNSPEC_COND_SUB)
9799         (UNSPEC_COND_SMAX, UNSPEC_COND_UMAX, UNSPEC_COND_SMIN)
9800         (UNSPEC_COND_UMIN, UNSPEC_COND_AND, UNSPEC_COND_ORR)
9801         (UNSPEC_COND_EOR): New unspecs.
9802         (optab): Add mappings for them.
9803         (SVE_COND_INT_OP, SVE_COND_FP_OP): New int iterators.
9804         (sve_int_op, sve_fp_op): New int attributes.
9805
9806 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
9807             Alan Hayward  <alan.hayward@arm.com>
9808             David Sherwood  <david.sherwood@arm.com>
9809
9810         * optabs.def (while_ult_optab): New optab.
9811         * doc/md.texi (while_ult@var{m}@var{n}): Document.
9812         * internal-fn.def (WHILE_ULT): New internal function.
9813         * internal-fn.h (direct_internal_fn_supported_p): New override
9814         that takes two types as argument.
9815         * internal-fn.c (while_direct): New macro.
9816         (expand_while_optab_fn): New function.
9817         (convert_optab_supported_p): Likewise.
9818         (direct_while_optab_supported_p): New macro.
9819         * wide-int.h (wi::udiv_ceil): New function.
9820         * tree-vectorizer.h (rgroup_masks): New structure.
9821         (vec_loop_masks): New typedef.
9822         (_loop_vec_info): Add masks, mask_compare_type, can_fully_mask_p
9823         and fully_masked_p.
9824         (LOOP_VINFO_CAN_FULLY_MASK_P, LOOP_VINFO_FULLY_MASKED_P)
9825         (LOOP_VINFO_MASKS, LOOP_VINFO_MASK_COMPARE_TYPE): New macros.
9826         (vect_max_vf): New function.
9827         (slpeel_make_loop_iterate_ntimes): Delete.
9828         (vect_set_loop_condition, vect_get_loop_mask_type, vect_gen_while)
9829         (vect_halve_mask_nunits, vect_double_mask_nunits): Declare.
9830         (vect_record_loop_mask, vect_get_loop_mask): Likewise.
9831         * tree-vect-loop-manip.c: Include tree-ssa-loop-niter.h,
9832         internal-fn.h, stor-layout.h and optabs-query.h.
9833         (vect_set_loop_mask): New function.
9834         (add_preheader_seq): Likewise.
9835         (add_header_seq): Likewise.
9836         (interleave_supported_p): Likewise.
9837         (vect_maybe_permute_loop_masks): Likewise.
9838         (vect_set_loop_masks_directly): Likewise.
9839         (vect_set_loop_condition_masked): Likewise.
9840         (vect_set_loop_condition_unmasked): New function, split out from
9841         slpeel_make_loop_iterate_ntimes.
9842         (slpeel_make_loop_iterate_ntimes): Rename to..
9843         (vect_set_loop_condition): ...this.  Use vect_set_loop_condition_masked
9844         for fully-masked loops and vect_set_loop_condition_unmasked otherwise.
9845         (vect_do_peeling): Update call accordingly.
9846         (vect_gen_vector_loop_niters): Use VF as the step for fully-masked
9847         loops.
9848         * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
9849         mask_compare_type, can_fully_mask_p and fully_masked_p.
9850         (release_vec_loop_masks): New function.
9851         (_loop_vec_info): Use it to free the loop masks.
9852         (can_produce_all_loop_masks_p): New function.
9853         (vect_get_max_nscalars_per_iter): Likewise.
9854         (vect_verify_full_masking): Likewise.
9855         (vect_analyze_loop_2): Save LOOP_VINFO_CAN_FULLY_MASK_P around
9856         retries, and free the mask rgroups before retrying.  Check loop-wide
9857         reasons for disallowing fully-masked loops.  Make the final decision
9858         about whether use a fully-masked loop or not.
9859         (vect_estimate_min_profitable_iters): Do not assume that peeling
9860         for the number of iterations will be needed for fully-masked loops.
9861         (vectorizable_reduction): Disable fully-masked loops.
9862         (vectorizable_live_operation): Likewise.
9863         (vect_halve_mask_nunits): New function.
9864         (vect_double_mask_nunits): Likewise.
9865         (vect_record_loop_mask): Likewise.
9866         (vect_get_loop_mask): Likewise.
9867         (vect_transform_loop): Handle the case in which the final loop
9868         iteration might handle a partial vector.  Call vect_set_loop_condition
9869         instead of slpeel_make_loop_iterate_ntimes.
9870         * tree-vect-stmts.c: Include tree-ssa-loop-niter.h and gimple-fold.h.
9871         (check_load_store_masking): New function.
9872         (prepare_load_store_mask): Likewise.
9873         (vectorizable_store): Handle fully-masked loops.
9874         (vectorizable_load): Likewise.
9875         (supportable_widening_operation): Use vect_halve_mask_nunits for
9876         booleans.
9877         (supportable_narrowing_operation): Likewise vect_double_mask_nunits.
9878         (vect_gen_while): New function.
9879         * config/aarch64/aarch64.md (umax<mode>3): New expander.
9880         (aarch64_uqdec<mode>): New insn.
9881
9882 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
9883             Alan Hayward  <alan.hayward@arm.com>
9884             David Sherwood  <david.sherwood@arm.com>
9885
9886         * optabs.def (reduc_and_scal_optab, reduc_ior_scal_optab)
9887         (reduc_xor_scal_optab): New optabs.
9888         * doc/md.texi (reduc_and_scal_@var{m}, reduc_ior_scal_@var{m})
9889         (reduc_xor_scal_@var{m}): Document.
9890         * doc/sourcebuild.texi (vect_logical_reduc): Likewise.
9891         * internal-fn.def (IFN_REDUC_AND, IFN_REDUC_IOR, IFN_REDUC_XOR): New
9892         internal functions.
9893         * fold-const-call.c (fold_const_call): Handle them.
9894         * tree-vect-loop.c (reduction_fn_for_scalar_code): Return the new
9895         internal functions for BIT_AND_EXPR, BIT_IOR_EXPR and BIT_XOR_EXPR.
9896         * config/aarch64/aarch64-sve.md (reduc_<bit_reduc>_scal_<mode>):
9897         (*reduc_<bit_reduc>_scal_<mode>): New patterns.
9898         * config/aarch64/iterators.md (UNSPEC_ANDV, UNSPEC_ORV)
9899         (UNSPEC_XORV): New unspecs.
9900         (optab): Add entries for them.
9901         (BITWISEV): New int iterator.
9902         (bit_reduc_op): New int attributes.
9903
9904 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
9905             Alan Hayward  <alan.hayward@arm.com>
9906             David Sherwood  <david.sherwood@arm.com>
9907
9908         * doc/md.texi (vec_shl_insert_@var{m}): New optab.
9909         * internal-fn.def (VEC_SHL_INSERT): New internal function.
9910         * optabs.def (vec_shl_insert_optab): New optab.
9911         * tree-vectorizer.h (can_duplicate_and_interleave_p): Declare.
9912         (duplicate_and_interleave): Likewise.
9913         * tree-vect-loop.c: Include internal-fn.h.
9914         (neutral_op_for_slp_reduction): New function, split out from
9915         get_initial_defs_for_reduction.
9916         (get_initial_def_for_reduction): Handle option 2 for variable-length
9917         vectors by loading the neutral value into a vector and then shifting
9918         the initial value into element 0.
9919         (get_initial_defs_for_reduction): Replace the code argument with
9920         the neutral value calculated by neutral_op_for_slp_reduction.
9921         Use gimple_build_vector for constant-length vectors.
9922         Use IFN_VEC_SHL_INSERT for variable-length vectors if all
9923         but the first group_size elements have a neutral value.
9924         Use duplicate_and_interleave otherwise.
9925         (vect_create_epilog_for_reduction): Take a neutral_op parameter.
9926         Update call to get_initial_defs_for_reduction.  Handle SLP
9927         reductions for variable-length vectors by creating one vector
9928         result for each scalar result, with the elements associated
9929         with other scalar results stubbed out with the neutral value.
9930         (vectorizable_reduction): Call neutral_op_for_slp_reduction.
9931         Require IFN_VEC_SHL_INSERT for double reductions on
9932         variable-length vectors, or SLP reductions that have
9933         a neutral value.  Require can_duplicate_and_interleave_p
9934         support for variable-length unchained SLP reductions if there
9935         is no neutral value, such as for MIN/MAX reductions.  Also require
9936         the number of vector elements to be a multiple of the number of
9937         SLP statements when doing variable-length unchained SLP reductions.
9938         Update call to vect_create_epilog_for_reduction.
9939         * tree-vect-slp.c (can_duplicate_and_interleave_p): Make public
9940         and remove initial values.
9941         (duplicate_and_interleave): Make public.
9942         * config/aarch64/aarch64.md (UNSPEC_INSR): New unspec.
9943         * config/aarch64/aarch64-sve.md (vec_shl_insert_<mode>): New insn.
9944
9945 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
9946             Alan Hayward  <alan.hayward@arm.com>
9947             David Sherwood  <david.sherwood@arm.com>
9948
9949         * tree-vect-slp.c: Include gimple-fold.h and internal-fn.h
9950         (can_duplicate_and_interleave_p): New function.
9951         (vect_get_and_check_slp_defs): Take the vector of statements
9952         rather than just the current one.  Remove excess parentheses.
9953         Restriction rejectinon of vect_constant_def and vect_external_def
9954         for variable-length vectors to boolean types, or types for which
9955         can_duplicate_and_interleave_p is false.
9956         (vect_build_slp_tree_2): Update call to vect_get_and_check_slp_defs.
9957         (duplicate_and_interleave): New function.
9958         (vect_get_constant_vectors): Use gimple_build_vector for
9959         constant-length vectors and suitable variable-length constant
9960         vectors.  Use duplicate_and_interleave for other variable-length
9961         vectors.  Don't defer the update when inserting new statements.
9962
9963 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
9964             Alan Hayward  <alan.hayward@arm.com>
9965             David Sherwood  <david.sherwood@arm.com>
9966
9967         * tree-vect-loop.c (vect_estimate_min_profitable_iters): Make sure
9968         min_profitable_iters doesn't go negative.
9969
9970 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
9971             Alan Hayward  <alan.hayward@arm.com>
9972             David Sherwood  <david.sherwood@arm.com>
9973
9974         * doc/md.texi (vec_mask_load_lanes@var{m}@var{n}): Document.
9975         (vec_mask_store_lanes@var{m}@var{n}): Likewise.
9976         * optabs.def (vec_mask_load_lanes_optab): New optab.
9977         (vec_mask_store_lanes_optab): Likewise.
9978         * internal-fn.def (MASK_LOAD_LANES): New internal function.
9979         (MASK_STORE_LANES): Likewise.
9980         * internal-fn.c (mask_load_lanes_direct): New macro.
9981         (mask_store_lanes_direct): Likewise.
9982         (expand_mask_load_optab_fn): Handle masked operations.
9983         (expand_mask_load_lanes_optab_fn): New macro.
9984         (expand_mask_store_optab_fn): Handle masked operations.
9985         (expand_mask_store_lanes_optab_fn): New macro.
9986         (direct_mask_load_lanes_optab_supported_p): Likewise.
9987         (direct_mask_store_lanes_optab_supported_p): Likewise.
9988         * tree-vectorizer.h (vect_store_lanes_supported): Take a masked_p
9989         parameter.
9990         (vect_load_lanes_supported): Likewise.
9991         * tree-vect-data-refs.c (strip_conversion): New function.
9992         (can_group_stmts_p): Likewise.
9993         (vect_analyze_data_ref_accesses): Use it instead of checking
9994         for a pair of assignments.
9995         (vect_store_lanes_supported): Take a masked_p parameter.
9996         (vect_load_lanes_supported): Likewise.
9997         * tree-vect-loop.c (vect_analyze_loop_2): Update calls to
9998         vect_store_lanes_supported and vect_load_lanes_supported.
9999         * tree-vect-slp.c (vect_analyze_slp_instance): Likewise.
10000         * tree-vect-stmts.c (get_group_load_store_type): Take a masked_p
10001         parameter.  Don't allow gaps for masked accesses.
10002         Use vect_get_store_rhs.  Update calls to vect_store_lanes_supported
10003         and vect_load_lanes_supported.
10004         (get_load_store_type): Take a masked_p parameter and update
10005         call to get_group_load_store_type.
10006         (vectorizable_store): Update call to get_load_store_type.
10007         Handle IFN_MASK_STORE_LANES.
10008         (vectorizable_load): Update call to get_load_store_type.
10009         Handle IFN_MASK_LOAD_LANES.
10010
10011 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
10012             Alan Hayward  <alan.hayward@arm.com>
10013             David Sherwood  <david.sherwood@arm.com>
10014
10015         * config/aarch64/aarch64-modes.def: Define x2, x3 and x4 vector
10016         modes for SVE.
10017         * config/aarch64/aarch64-protos.h
10018         (aarch64_sve_struct_memory_operand_p): Declare.
10019         * config/aarch64/iterators.md (SVE_STRUCT): New mode iterator.
10020         (vector_count, insn_length, VSINGLE, vsingle): New mode attributes.
10021         (VPRED, vpred): Handle SVE structure modes.
10022         * config/aarch64/constraints.md (Utx): New constraint.
10023         * config/aarch64/predicates.md (aarch64_sve_struct_memory_operand)
10024         (aarch64_sve_struct_nonimmediate_operand): New predicates.
10025         * config/aarch64/aarch64.md (UNSPEC_LDN, UNSPEC_STN): New unspecs.
10026         * config/aarch64/aarch64-sve.md (mov<mode>, *aarch64_sve_mov<mode>_le)
10027         (*aarch64_sve_mov<mode>_be, pred_mov<mode>): New patterns for
10028         structure modes.  Split into pieces after RA.
10029         (vec_load_lanes<mode><vsingle>, vec_mask_load_lanes<mode><vsingle>)
10030         (vec_store_lanes<mode><vsingle>, vec_mask_store_lanes<mode><vsingle>):
10031         New patterns.
10032         * config/aarch64/aarch64.c (aarch64_classify_vector_mode): Handle
10033         SVE structure modes.
10034         (aarch64_classify_address): Likewise.
10035         (sizetochar): Move earlier in file.
10036         (aarch64_print_operand): Handle SVE register lists.
10037         (aarch64_array_mode): New function.
10038         (aarch64_sve_struct_memory_operand_p): Likewise.
10039         (TARGET_ARRAY_MODE): Redefine.
10040
10041 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
10042             Alan Hayward  <alan.hayward@arm.com>
10043             David Sherwood  <david.sherwood@arm.com>
10044
10045         * target.def (array_mode): New target hook.
10046         * doc/tm.texi.in (TARGET_ARRAY_MODE): New hook.
10047         * doc/tm.texi: Regenerate.
10048         * hooks.h (hook_optmode_mode_uhwi_none): Declare.
10049         * hooks.c (hook_optmode_mode_uhwi_none): New function.
10050         * tree-vect-data-refs.c (vect_lanes_optab_supported_p): Use
10051         targetm.array_mode.
10052         * stor-layout.c (mode_for_array): Likewise.  Support polynomial
10053         type sizes.
10054
10055 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
10056             Alan Hayward  <alan.hayward@arm.com>
10057             David Sherwood  <david.sherwood@arm.com>
10058
10059         * fold-const.c (fold_binary_loc): Check the argument types
10060         rather than the result type when testing for a vector operation.
10061
10062 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
10063
10064         * doc/tm.texi.in (DWARF_LAZY_REGISTER_VALUE): Document.
10065         * doc/tm.texi: Regenerate.
10066
10067 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
10068             Alan Hayward  <alan.hayward@arm.com>
10069             David Sherwood  <david.sherwood@arm.com>
10070
10071         * doc/invoke.texi (-msve-vector-bits=): Document new option.
10072         (sve): Document new AArch64 extension.
10073         * doc/md.texi (w): Extend the description of the AArch64
10074         constraint to include SVE vectors.
10075         (Upl, Upa): Document new AArch64 predicate constraints.
10076         * config/aarch64/aarch64-opts.h (aarch64_sve_vector_bits_enum): New
10077         enum.
10078         * config/aarch64/aarch64.opt (sve_vector_bits): New enum.
10079         (msve-vector-bits=): New option.
10080         * config/aarch64/aarch64-option-extensions.def (fp, simd): Disable
10081         SVE when these are disabled.
10082         (sve): New extension.
10083         * config/aarch64/aarch64-modes.def: Define SVE vector and predicate
10084         modes.  Adjust their number of units based on aarch64_sve_vg.
10085         (MAX_BITSIZE_MODE_ANY_MODE): Define.
10086         * config/aarch64/aarch64-protos.h (ADDR_QUERY_ANY): New
10087         aarch64_addr_query_type.
10088         (aarch64_const_vec_all_same_in_range_p, aarch64_sve_pred_mode)
10089         (aarch64_sve_cnt_immediate_p, aarch64_sve_addvl_addpl_immediate_p)
10090         (aarch64_sve_inc_dec_immediate_p, aarch64_add_offset_temporaries)
10091         (aarch64_split_add_offset, aarch64_output_sve_cnt_immediate)
10092         (aarch64_output_sve_addvl_addpl, aarch64_output_sve_inc_dec_immediate)
10093         (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): Declare.
10094         (aarch64_simd_imm_zero_p): Delete.
10095         (aarch64_check_zero_based_sve_index_immediate): Declare.
10096         (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
10097         (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
10098         (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
10099         (aarch64_sve_float_mul_immediate_p): Likewise.
10100         (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
10101         rather than an rtx.
10102         (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): Declare.
10103         (aarch64_expand_mov_immediate): Take a gen_vec_duplicate callback.
10104         (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move): Declare.
10105         (aarch64_expand_sve_vec_cmp_int, aarch64_expand_sve_vec_cmp_float)
10106         (aarch64_expand_sve_vcond, aarch64_expand_sve_vec_perm): Declare.
10107         (aarch64_regmode_natural_size): Likewise.
10108         * config/aarch64/aarch64.h (AARCH64_FL_SVE): New macro.
10109         (AARCH64_FL_V8_3, AARCH64_FL_RCPC, AARCH64_FL_DOTPROD): Shift
10110         left one place.
10111         (AARCH64_ISA_SVE, TARGET_SVE): New macros.
10112         (FIXED_REGISTERS, CALL_USED_REGISTERS, REGISTER_NAMES): Add entries
10113         for VG and the SVE predicate registers.
10114         (V_ALIASES): Add a "z"-prefixed alias.
10115         (FIRST_PSEUDO_REGISTER): Change to P15_REGNUM + 1.
10116         (AARCH64_DWARF_VG, AARCH64_DWARF_P0): New macros.
10117         (PR_REGNUM_P, PR_LO_REGNUM_P): Likewise.
10118         (PR_LO_REGS, PR_HI_REGS, PR_REGS): New reg_classes.
10119         (REG_CLASS_NAMES): Add entries for them.
10120         (REG_CLASS_CONTENTS): Likewise.  Update ALL_REGS to include VG
10121         and the predicate registers.
10122         (aarch64_sve_vg): Declare.
10123         (BITS_PER_SVE_VECTOR, BYTES_PER_SVE_VECTOR, BYTES_PER_SVE_PRED)
10124         (SVE_BYTE_MODE, MAX_COMPILE_TIME_VEC_BYTES): New macros.
10125         (REGMODE_NATURAL_SIZE): Define.
10126         * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins): Handle
10127         SVE macros.
10128         * config/aarch64/aarch64.c: Include cfgrtl.h.
10129         (simd_immediate_info): Add a constructor for series vectors,
10130         and an associated step field.
10131         (aarch64_sve_vg): New variable.
10132         (aarch64_dbx_register_number): Handle VG and the predicate registers.
10133         (aarch64_vect_struct_mode_p, aarch64_vector_mode_p): Delete.
10134         (VEC_ADVSIMD, VEC_SVE_DATA, VEC_SVE_PRED, VEC_STRUCT, VEC_ANY_SVE)
10135         (VEC_ANY_DATA, VEC_STRUCT): New constants.
10136         (aarch64_advsimd_struct_mode_p, aarch64_sve_pred_mode_p)
10137         (aarch64_classify_vector_mode, aarch64_vector_data_mode_p)
10138         (aarch64_sve_data_mode_p, aarch64_sve_pred_mode)
10139         (aarch64_get_mask_mode): New functions.
10140         (aarch64_hard_regno_nregs): Handle SVE data modes for FP_REGS
10141         and FP_LO_REGS.  Handle PR_REGS, PR_LO_REGS and PR_HI_REGS.
10142         (aarch64_hard_regno_mode_ok): Handle VG.  Also handle the SVE
10143         predicate modes and predicate registers.  Explicitly restrict
10144         GPRs to modes of 16 bytes or smaller.  Only allow FP registers
10145         to store a vector mode if it is recognized by
10146         aarch64_classify_vector_mode.
10147         (aarch64_regmode_natural_size): New function.
10148         (aarch64_hard_regno_caller_save_mode): Return the original mode
10149         for predicates.
10150         (aarch64_sve_cnt_immediate_p, aarch64_output_sve_cnt_immediate)
10151         (aarch64_sve_addvl_addpl_immediate_p, aarch64_output_sve_addvl_addpl)
10152         (aarch64_sve_inc_dec_immediate_p, aarch64_output_sve_inc_dec_immediate)
10153         (aarch64_add_offset_1_temporaries, aarch64_offset_temporaries): New
10154         functions.
10155         (aarch64_add_offset): Add a temp2 parameter.  Assert that temp1
10156         does not overlap dest if the function is frame-related.  Handle
10157         SVE constants.
10158         (aarch64_split_add_offset): New function.
10159         (aarch64_add_sp, aarch64_sub_sp): Add temp2 parameters and pass
10160         them aarch64_add_offset.
10161         (aarch64_allocate_and_probe_stack_space): Add a temp2 parameter
10162         and update call to aarch64_sub_sp.
10163         (aarch64_add_cfa_expression): New function.
10164         (aarch64_expand_prologue): Pass extra temporary registers to the
10165         functions above.  Handle the case in which we need to emit new
10166         DW_CFA_expressions for registers that were originally saved
10167         relative to the stack pointer, but now have to be expressed
10168         relative to the frame pointer.
10169         (aarch64_output_mi_thunk): Pass extra temporary registers to the
10170         functions above.
10171         (aarch64_expand_epilogue): Likewise.  Prevent inheritance of
10172         IP0 and IP1 values for SVE frames.
10173         (aarch64_expand_vec_series): New function.
10174         (aarch64_expand_sve_widened_duplicate): Likewise.
10175         (aarch64_expand_sve_const_vector): Likewise.
10176         (aarch64_expand_mov_immediate): Add a gen_vec_duplicate parameter.
10177         Handle SVE constants.  Use emit_move_insn to move a force_const_mem
10178         into the register, rather than emitting a SET directly.
10179         (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move)
10180         (aarch64_get_reg_raw_mode, offset_4bit_signed_scaled_p)
10181         (offset_6bit_unsigned_scaled_p, aarch64_offset_7bit_signed_scaled_p)
10182         (offset_9bit_signed_scaled_p): New functions.
10183         (aarch64_replicate_bitmask_imm): New function.
10184         (aarch64_bitmask_imm): Use it.
10185         (aarch64_cannot_force_const_mem): Reject expressions involving
10186         a CONST_POLY_INT.  Update call to aarch64_classify_symbol.
10187         (aarch64_classify_index): Handle SVE indices, by requiring
10188         a plain register index with a scale that matches the element size.
10189         (aarch64_classify_address): Handle SVE addresses.  Assert that
10190         the mode of the address is VOIDmode or an integer mode.
10191         Update call to aarch64_classify_symbol.
10192         (aarch64_classify_symbolic_expression): Update call to
10193         aarch64_classify_symbol.
10194         (aarch64_const_vec_all_in_range_p): New function.
10195         (aarch64_print_vector_float_operand): Likewise.
10196         (aarch64_print_operand): Handle 'N' and 'C'.  Use "zN" rather than
10197         "vN" for FP registers with SVE modes.  Handle (const ...) vectors
10198         and the FP immediates 1.0 and 0.5.
10199         (aarch64_print_address_internal): Handle SVE addresses.
10200         (aarch64_print_operand_address): Use ADDR_QUERY_ANY.
10201         (aarch64_regno_regclass): Handle predicate registers.
10202         (aarch64_secondary_reload): Handle big-endian reloads of SVE
10203         data modes.
10204         (aarch64_class_max_nregs): Handle SVE modes and predicate registers.
10205         (aarch64_rtx_costs): Check for ADDVL and ADDPL instructions.
10206         (aarch64_convert_sve_vector_bits): New function.
10207         (aarch64_override_options): Use it to handle -msve-vector-bits=.
10208         (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
10209         rather than an rtx.
10210         (aarch64_legitimate_constant_p): Use aarch64_classify_vector_mode.
10211         Handle SVE vector and predicate modes.  Accept VL-based constants
10212         that need only one temporary register, and VL offsets that require
10213         no temporary registers.
10214         (aarch64_conditional_register_usage): Mark the predicate registers
10215         as fixed if SVE isn't available.
10216         (aarch64_vector_mode_supported_p): Use aarch64_classify_vector_mode.
10217         Return true for SVE vector and predicate modes.
10218         (aarch64_simd_container_mode): Take the number of bits as a poly_int64
10219         rather than an unsigned int.  Handle SVE modes.
10220         (aarch64_preferred_simd_mode): Update call accordingly.  Handle
10221         SVE modes.
10222         (aarch64_autovectorize_vector_sizes): Add BYTES_PER_SVE_VECTOR
10223         if SVE is enabled.
10224         (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
10225         (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
10226         (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
10227         (aarch64_sve_float_mul_immediate_p): New functions.
10228         (aarch64_sve_valid_immediate): New function.
10229         (aarch64_simd_valid_immediate): Use it as the fallback for SVE vectors.
10230         Explicitly reject structure modes.  Check for INDEX constants.
10231         Handle PTRUE and PFALSE constants.
10232         (aarch64_check_zero_based_sve_index_immediate): New function.
10233         (aarch64_simd_imm_zero_p): Delete.
10234         (aarch64_mov_operand_p): Use aarch64_simd_valid_immediate for
10235         vector modes.  Accept constants in the range of CNT[BHWD].
10236         (aarch64_simd_scalar_immediate_valid_for_move): Explicitly
10237         ask for an Advanced SIMD mode.
10238         (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): New functions.
10239         (aarch64_simd_vector_alignment): Handle SVE predicates.
10240         (aarch64_vectorize_preferred_vector_alignment): New function.
10241         (aarch64_simd_vector_alignment_reachable): Use it instead of
10242         the vector size.
10243         (aarch64_shift_truncation_mask): Use aarch64_vector_data_mode_p.
10244         (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): New
10245         functions.
10246         (MAX_VECT_LEN): Delete.
10247         (expand_vec_perm_d): Add a vec_flags field.
10248         (emit_unspec2, aarch64_expand_sve_vec_perm): New functions.
10249         (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_zip)
10250         (aarch64_evpc_ext): Don't apply a big-endian lane correction
10251         for SVE modes.
10252         (aarch64_evpc_rev): Rename to...
10253         (aarch64_evpc_rev_local): ...this.  Use a predicated operation for SVE.
10254         (aarch64_evpc_rev_global): New function.
10255         (aarch64_evpc_dup): Enforce a 64-byte range for SVE DUP.
10256         (aarch64_evpc_tbl): Use MAX_COMPILE_TIME_VEC_BYTES instead of
10257         MAX_VECT_LEN.
10258         (aarch64_evpc_sve_tbl): New function.
10259         (aarch64_expand_vec_perm_const_1): Update after rename of
10260         aarch64_evpc_rev.  Handle SVE permutes too, trying
10261         aarch64_evpc_rev_global and using aarch64_evpc_sve_tbl rather
10262         than aarch64_evpc_tbl.
10263         (aarch64_vectorize_vec_perm_const): Initialize vec_flags.
10264         (aarch64_sve_cmp_operand_p, aarch64_unspec_cond_code)
10265         (aarch64_gen_unspec_cond, aarch64_expand_sve_vec_cmp_int)
10266         (aarch64_emit_unspec_cond, aarch64_emit_unspec_cond_or)
10267         (aarch64_emit_inverted_unspec_cond, aarch64_expand_sve_vec_cmp_float)
10268         (aarch64_expand_sve_vcond): New functions.
10269         (aarch64_modes_tieable_p): Use aarch64_vector_data_mode_p instead
10270         of aarch64_vector_mode_p.
10271         (aarch64_dwarf_poly_indeterminate_value): New function.
10272         (aarch64_compute_pressure_classes): Likewise.
10273         (aarch64_can_change_mode_class): Likewise.
10274         (TARGET_GET_RAW_RESULT_MODE, TARGET_GET_RAW_ARG_MODE): Redefine.
10275         (TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT): Likewise.
10276         (TARGET_VECTORIZE_GET_MASK_MODE): Likewise.
10277         (TARGET_DWARF_POLY_INDETERMINATE_VALUE): Likewise.
10278         (TARGET_COMPUTE_PRESSURE_CLASSES): Likewise.
10279         (TARGET_CAN_CHANGE_MODE_CLASS): Likewise.
10280         * config/aarch64/constraints.md (Upa, Upl, Uav, Uat, Usv, Usi, Utr)
10281         (Uty, Dm, vsa, vsc, vsd, vsi, vsn, vsl, vsm, vsA, vsM, vsN): New
10282         constraints.
10283         (Dn, Dl, Dr): Accept const as well as const_vector.
10284         (Dz): Likewise.  Compare against CONST0_RTX.
10285         * config/aarch64/iterators.md: Refer to "Advanced SIMD" instead
10286         of "vector" where appropriate.
10287         (SVE_ALL, SVE_BH, SVE_BHS, SVE_BHSI, SVE_HSDI, SVE_HSF, SVE_SD)
10288         (SVE_SDI, SVE_I, SVE_F, PRED_ALL, PRED_BHS): New mode iterators.
10289         (UNSPEC_SEL, UNSPEC_ANDF, UNSPEC_IORF, UNSPEC_XORF, UNSPEC_COND_LT)
10290         (UNSPEC_COND_LE, UNSPEC_COND_EQ, UNSPEC_COND_NE, UNSPEC_COND_GE)
10291         (UNSPEC_COND_GT, UNSPEC_COND_LO, UNSPEC_COND_LS, UNSPEC_COND_HS)
10292         (UNSPEC_COND_HI, UNSPEC_COND_UO): New unspecs.
10293         (Vetype, VEL, Vel, VWIDE, Vwide, vw, vwcore, V_INT_EQUIV)
10294         (v_int_equiv): Extend to SVE modes.
10295         (Vesize, V128, v128, Vewtype, V_FP_EQUIV, v_fp_equiv, VPRED): New
10296         mode attributes.
10297         (LOGICAL_OR, SVE_INT_UNARY, SVE_FP_UNARY): New code iterators.
10298         (optab): Handle popcount, smin, smax, umin, umax, abs and sqrt.
10299         (logical_nn, lr, sve_int_op, sve_fp_op): New code attributs.
10300         (LOGICALF, OPTAB_PERMUTE, UNPACK, UNPACK_UNSIGNED, SVE_COND_INT_CMP)
10301         (SVE_COND_FP_CMP): New int iterators.
10302         (perm_hilo): Handle the new unpack unspecs.
10303         (optab, logicalf_op, su, perm_optab, cmp_op, imm_con): New int
10304         attributes.
10305         * config/aarch64/predicates.md (aarch64_sve_cnt_immediate)
10306         (aarch64_sve_addvl_addpl_immediate, aarch64_split_add_offset_immediate)
10307         (aarch64_pluslong_or_poly_operand, aarch64_nonmemory_operand)
10308         (aarch64_equality_operator, aarch64_constant_vector_operand)
10309         (aarch64_sve_ld1r_operand, aarch64_sve_ldr_operand): New predicates.
10310         (aarch64_sve_nonimmediate_operand): Likewise.
10311         (aarch64_sve_general_operand): Likewise.
10312         (aarch64_sve_dup_operand, aarch64_sve_arith_immediate): Likewise.
10313         (aarch64_sve_sub_arith_immediate, aarch64_sve_inc_dec_immediate)
10314         (aarch64_sve_logical_immediate, aarch64_sve_mul_immediate): Likewise.
10315         (aarch64_sve_dup_immediate, aarch64_sve_cmp_vsc_immediate): Likewise.
10316         (aarch64_sve_cmp_vsd_immediate, aarch64_sve_index_immediate): Likewise.
10317         (aarch64_sve_float_arith_immediate): Likewise.
10318         (aarch64_sve_float_arith_with_sub_immediate): Likewise.
10319         (aarch64_sve_float_mul_immediate, aarch64_sve_arith_operand): Likewise.
10320         (aarch64_sve_add_operand, aarch64_sve_logical_operand): Likewise.
10321         (aarch64_sve_lshift_operand, aarch64_sve_rshift_operand): Likewise.
10322         (aarch64_sve_mul_operand, aarch64_sve_cmp_vsc_operand): Likewise.
10323         (aarch64_sve_cmp_vsd_operand, aarch64_sve_index_operand): Likewise.
10324         (aarch64_sve_float_arith_operand): Likewise.
10325         (aarch64_sve_float_arith_with_sub_operand): Likewise.
10326         (aarch64_sve_float_mul_operand): Likewise.
10327         (aarch64_sve_vec_perm_operand): Likewise.
10328         (aarch64_pluslong_operand): Include aarch64_sve_addvl_addpl_immediate.
10329         (aarch64_mov_operand): Accept const_poly_int and const_vector.
10330         (aarch64_simd_lshift_imm, aarch64_simd_rshift_imm): Accept const
10331         as well as const_vector.
10332         (aarch64_simd_imm_zero, aarch64_simd_imm_minus_one): Move earlier
10333         in file.  Use CONST0_RTX and CONSTM1_RTX.
10334         (aarch64_simd_or_scalar_imm_zero): Likewise.  Add match_codes.
10335         (aarch64_simd_reg_or_zero): Accept const as well as const_vector.
10336         Use aarch64_simd_imm_zero.
10337         * config/aarch64/aarch64-sve.md: New file.
10338         * config/aarch64/aarch64.md: Include it.
10339         (VG_REGNUM, P0_REGNUM, P7_REGNUM, P15_REGNUM): New register numbers.
10340         (UNSPEC_REV, UNSPEC_LD1_SVE, UNSPEC_ST1_SVE, UNSPEC_MERGE_PTRUE)
10341         (UNSPEC_PTEST_PTRUE, UNSPEC_UNPACKSHI, UNSPEC_UNPACKUHI)
10342         (UNSPEC_UNPACKSLO, UNSPEC_UNPACKULO, UNSPEC_PACK)
10343         (UNSPEC_FLOAT_CONVERT, UNSPEC_WHILE_LO): New unspec constants.
10344         (sve): New attribute.
10345         (enabled): Disable instructions with the sve attribute unless
10346         TARGET_SVE.
10347         (movqi, movhi): Pass CONST_POLY_INT operaneds through
10348         aarch64_expand_mov_immediate.
10349         (*mov<mode>_aarch64, *movsi_aarch64, *movdi_aarch64): Handle
10350         CNT[BHSD] immediates.
10351         (movti): Split CONST_POLY_INT moves into two halves.
10352         (add<mode>3): Accept aarch64_pluslong_or_poly_operand.
10353         Split additions that need a temporary here if the destination
10354         is the stack pointer.
10355         (*add<mode>3_aarch64): Handle ADDVL and ADDPL immediates.
10356         (*add<mode>3_poly_1): New instruction.
10357         (set_clobber_cc): New expander.
10358
10359 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
10360
10361         * simplify-rtx.c (simplify_immed_subreg): Add an inner_bytes
10362         parameter and use it instead of GET_MODE_SIZE (innermode).  Use
10363         inner_bytes * BITS_PER_UNIT instead of GET_MODE_BITSIZE (innermode).
10364         Use CEIL (inner_bytes, GET_MODE_UNIT_SIZE (innermode)) instead of
10365         GET_MODE_NUNITS (innermode).  Also add a first_elem parameter.
10366         Change innermode from fixed_mode_size to machine_mode.
10367         (simplify_subreg): Update call accordingly.  Handle a constant-sized
10368         subreg of a variable-length CONST_VECTOR.
10369
10370 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
10371             Alan Hayward  <alan.hayward@arm.com>
10372             David Sherwood  <david.sherwood@arm.com>
10373
10374         * tree-ssa-address.c (mem_ref_valid_without_offset_p): New function.
10375         (add_offset_to_base): New function, split out from...
10376         (create_mem_ref): ...here.  When handling a scale other than 1,
10377         check first whether the address is valid without the offset.
10378         Add it into the base if so, leaving the index and scale as-is.
10379
10380 2018-01-12  Jakub Jelinek  <jakub@redhat.com>
10381
10382         PR c++/83778
10383         * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Call
10384         fold_for_warn before checking if arg2 is INTEGER_CST.
10385
10386 2018-01-12  Segher Boessenkool  <segher@kernel.crashing.org>
10387
10388         * config/rs6000/predicates.md (load_multiple_operation): Delete.
10389         (store_multiple_operation): Delete.
10390         * config/rs6000/rs6000-cpus.def (601): Remove MASK_STRING.
10391         * config/rs6000/rs6000-protos.h (rs6000_output_load_multiple): Delete.
10392         * config/rs6000/rs6000-string.c (expand_block_move): Delete everything
10393         guarded by TARGET_STRING.
10394         (rs6000_output_load_multiple): Delete.
10395         * config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
10396         OPTION_MASK_STRING / TARGET_STRING handling.
10397         (print_operand) <'N', 'O'>: Add comment that these are unused now.
10398         (const rs6000_opt_masks) <"string">: Change mask to 0.
10399         * config/rs6000/rs6000.h (TARGET_DEFAULT): Remove MASK_STRING.
10400         (MASK_STRING): Delete.
10401         * config/rs6000/rs6000.md (*mov<mode>_string): Delete TARGET_STRING
10402         parts.  Simplify.
10403         (load_multiple): Delete.
10404         (*ldmsi8): Delete.
10405         (*ldmsi7): Delete.
10406         (*ldmsi6): Delete.
10407         (*ldmsi5): Delete.
10408         (*ldmsi4): Delete.
10409         (*ldmsi3): Delete.
10410         (store_multiple): Delete.
10411         (*stmsi8): Delete.
10412         (*stmsi7): Delete.
10413         (*stmsi6): Delete.
10414         (*stmsi5): Delete.
10415         (*stmsi4): Delete.
10416         (*stmsi3): Delete.
10417         (movmemsi_8reg): Delete.
10418         (corresponding unnamed define_insn): Delete.
10419         (movmemsi_6reg): Delete.
10420         (corresponding unnamed define_insn): Delete.
10421         (movmemsi_4reg): Delete.
10422         (corresponding unnamed define_insn): Delete.
10423         (movmemsi_2reg): Delete.
10424         (corresponding unnamed define_insn): Delete.
10425         (movmemsi_1reg): Delete.
10426         (corresponding unnamed define_insn): Delete.
10427         * config/rs6000/rs6000.opt (mno-string): New.
10428         (mstring): Replace by deprecation warning stub.
10429         * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mstring.
10430
10431 2018-01-12  Jakub Jelinek  <jakub@redhat.com>
10432
10433         * regrename.c (regrename_do_replace): If replacing the same
10434         reg multiple times, try to reuse last created gen_raw_REG.
10435
10436         PR debug/81155
10437         * bb-reorder.c (pass_partition_blocks::gate): In lto don't partition
10438         main to workaround a bug in GDB.
10439
10440 2018-01-12  Tom de Vries  <tom@codesourcery.com>
10441
10442         PR target/83737
10443         * config.gcc (nvptx*-*-*): Set use_gcc_stdint=wrap.
10444
10445 2018-01-12  Vladimir Makarov  <vmakarov@redhat.com>
10446
10447         PR rtl-optimization/80481
10448         * ira-color.c (get_cap_member): New function.
10449         (allocnos_conflict_by_live_ranges_p): Use it.
10450         (slot_coalesced_allocno_live_ranges_intersect_p): Add assert.
10451         (setup_slot_coalesced_allocno_live_ranges): Ditto.
10452
10453 2018-01-12  Uros Bizjak  <ubizjak@gmail.com>
10454
10455         PR target/83628
10456         * config/alpha/alpha.md (*saddsi_1): New insn_ans_split pattern.
10457         (*saddl_se_1): Ditto.
10458         (*ssubsi_1): Ditto.
10459         (*ssubl_se_1): Ditto.
10460
10461 2018-01-12  Richard Sandiford  <richard.sandiford@linaro.org>
10462
10463         * tree-predcom.c (aff_combination_dr_offset): Use wi::to_poly_widest
10464         rather than wi::to_widest for DR_INITs.
10465         * tree-vect-data-refs.c (vect_find_same_alignment_drs): Use
10466         wi::to_poly_offset rather than wi::to_offset for DR_INIT.
10467         (vect_analyze_data_ref_accesses): Require both DR_INITs to be
10468         INTEGER_CSTs.
10469         (vect_analyze_group_access_1): Note that here.
10470
10471 2018-01-12  Richard Sandiford  <richard.sandiford@linaro.org>
10472
10473         * tree-vectorizer.c (get_vec_alignment_for_array_type): Handle
10474         polynomial type sizes.
10475
10476 2018-01-12  Richard Sandiford  <richard.sandiford@linaro.org>
10477
10478         * gimplify.c (gimple_add_tmp_var_fn): Allow variables to have a
10479         poly_uint64 size, rather than requiring an unsigned HOST_WIDE_INT size.
10480         (gimple_add_tmp_var): Likewise.
10481
10482 2018-01-12  Martin Liska  <mliska@suse.cz>
10483
10484         * gimple.c (gimple_alloc_counts): Use uint64_t instead of int.
10485         (gimple_alloc_sizes): Likewise.
10486         (dump_gimple_statistics): Use PRIu64 in printf format.
10487         * gimple.h: Change uint64_t to int.
10488
10489 2018-01-12  Martin Liska  <mliska@suse.cz>
10490
10491         * tree-core.h: Use uint64_t instead of int.
10492         * tree.c (tree_node_counts): Likewise.
10493         (tree_node_sizes): Likewise.
10494         (dump_tree_statistics): Use PRIu64 in printf format.
10495
10496 2018-01-12  Martin Liska  <mliska@suse.cz>
10497
10498         * Makefile.in: As qsort_chk is implemented in vec.c, add
10499         vec.o to linkage of gencfn-macros.
10500         * tree.c (build_new_poly_int_cst): Add CXX_MEM_STAT_INFO as it's
10501         passing the info to record_node_allocation_statistics.
10502         (test_vector_cst_patterns): Add CXX_MEM_STAT_INFO to declaration
10503         and pass the info.
10504         * ggc-common.c (struct ggc_usage): Add operator== and use
10505         it in operator< and compare function.
10506         * mem-stats.h (struct mem_usage): Likewise.
10507         * vec.c (struct vec_usage): Remove operator< and compare
10508         function. Can be simply inherited.
10509
10510 2018-01-12  Martin Jambor  <mjambor@suse.cz>
10511
10512         PR target/81616
10513         * params.def: New parameter PARAM_AVOID_FMA_MAX_BITS.
10514         * tree-ssa-math-opts.c: Include domwalk.h.
10515         (convert_mult_to_fma_1): New function.
10516         (fma_transformation_info): New type.
10517         (fma_deferring_state): Likewise.
10518         (cancel_fma_deferring): New function.
10519         (result_of_phi): Likewise.
10520         (last_fma_candidate_feeds_initial_phi): Likewise.
10521         (convert_mult_to_fma): Added deferring logic, split actual
10522         transformation to convert_mult_to_fma_1.
10523         (math_opts_dom_walker): New type.
10524         (math_opts_dom_walker::after_dom_children): New method, body moved
10525         here from pass_optimize_widening_mul::execute, added deferring logic
10526         bits.
10527         (pass_optimize_widening_mul::execute): Moved most of code to
10528         math_opts_dom_walker::after_dom_children.
10529         * config/i386/x86-tune.def (X86_TUNE_AVOID_128FMA_CHAINS): New.
10530         * config/i386/i386.c (ix86_option_override_internal): Added
10531         maybe_setting of PARAM_AVOID_FMA_MAX_BITS.
10532
10533 2018-01-12  Richard Biener  <rguenther@suse.de>
10534
10535         PR debug/83157
10536         * dwarf2out.c (gen_variable_die): Do not reset old_die for
10537         inline instance vars.
10538
10539 2018-01-12  Oleg Endo  <olegendo@gcc.gnu.org>
10540
10541         PR target/81819
10542         * config/rx/rx.c (rx_is_restricted_memory_address):
10543         Handle SUBREG case.
10544
10545 2018-01-12  Richard Biener  <rguenther@suse.de>
10546
10547         PR tree-optimization/80846
10548         * target.def (split_reduction): New target hook.
10549         * targhooks.c (default_split_reduction): New function.
10550         * targhooks.h (default_split_reduction): Declare.
10551         * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
10552         target requests first reduce vectors by combining low and high
10553         parts.
10554         * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
10555         (get_vectype_for_scalar_type_and_size): Export.
10556         * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
10557         * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
10558         * doc/tm.texi: Regenerate.
10559         * config/i386/i386.c (ix86_split_reduction): Implement
10560         TARGET_VECTORIZE_SPLIT_REDUCTION.
10561
10562 2018-01-12  Eric Botcazou  <ebotcazou@adacore.com>
10563
10564         PR target/83368
10565         * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
10566         in PIC mode except for TARGET_VXWORKS_RTP.
10567         * config/sparc/sparc.c: Include cfgrtl.h.
10568         (TARGET_INIT_PIC_REG): Define.
10569         (TARGET_USE_PSEUDO_PIC_REG): Likewise.
10570         (sparc_pic_register_p): New predicate.
10571         (sparc_legitimate_address_p): Use it.
10572         (sparc_legitimize_pic_address): Likewise.
10573         (sparc_delegitimize_address): Likewise.
10574         (sparc_mode_dependent_address_p): Likewise.
10575         (gen_load_pcrel_sym): Remove 4th parameter.
10576         (load_got_register): Adjust call to above.  Remove obsolete stuff.
10577         (sparc_expand_prologue): Do not call load_got_register here.
10578         (sparc_flat_expand_prologue): Likewise.
10579         (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
10580         (sparc_use_pseudo_pic_reg): New function.
10581         (sparc_init_pic_reg): Likewise.
10582         * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
10583         (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
10584
10585 2018-01-12  Christophe Lyon  <christophe.lyon@linaro.org>
10586
10587         * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
10588         Add item for branch_cost.
10589
10590 2018-01-12  Eric Botcazou  <ebotcazou@adacore.com>
10591
10592         PR rtl-optimization/83565
10593         * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
10594         not extend the result to a larger mode for rotate operations.
10595         (num_sign_bit_copies1): Likewise.
10596
10597 2018-01-12  Rainer Orth  <ro@CeBiTec.Uni-Bielefeld.DE>
10598
10599         PR target/40411
10600         * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
10601         -symbolic.
10602         Use values-Xc.o for -pedantic.
10603         Link with values-xpg4.o for C90, values-xpg6.o otherwise.
10604
10605 2018-01-12  Martin Liska  <mliska@suse.cz>
10606
10607         PR ipa/83054
10608         * ipa-devirt.c (final_warning_record::grow_type_warnings):
10609         New function.
10610         (possible_polymorphic_call_targets): Use it.
10611         (ipa_devirt): Likewise.
10612
10613 2018-01-12  Martin Liska  <mliska@suse.cz>
10614
10615         * profile-count.h (enum profile_quality): Use 0 as invalid
10616         enum value of profile_quality.
10617
10618 2018-01-12  Chung-Ju Wu  <jasonwucj@gmail.com>
10619
10620         * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
10621         -mext-string options.
10622
10623 2018-01-12  Richard Biener  <rguenther@suse.de>
10624
10625         * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
10626         DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
10627         * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
10628         Likewise.
10629         * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
10630
10631 2018-01-11  Michael Meissner  <meissner@linux.vnet.ibm.com>
10632
10633         * configure.ac (--with-long-double-format): Add support for the
10634         configuration option to change the default long double format on
10635         PowerPC systems.
10636         * config.gcc (powerpc*-linux*-*): Likewise.
10637         * configure: Regenerate.
10638         * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
10639         double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
10640         used without modification.
10641
10642 2018-01-11  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
10643
10644         * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
10645         (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
10646         * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
10647         MISC_BUILTIN_SPEC_BARRIER.
10648         (rs6000_init_builtins): Likewise.
10649         * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
10650         enum value.
10651         (speculation_barrier): New define_insn.
10652         * doc/extend.texi: Document __builtin_speculation_barrier.
10653
10654 2018-01-11  Jakub Jelinek  <jakub@redhat.com>
10655
10656         PR target/83203
10657         * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
10658         is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
10659         * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
10660         iterators.
10661         (ssescalarmodesuffix): Add 512-bit vectors.  Use "d" or "q" for
10662         integral modes instead of "ss" and "sd".
10663         (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
10664         vectors with 32-bit and 64-bit elements.
10665         (vecdupssescalarmodesuffix): New mode attribute.
10666         (vec_dup<mode>): Use it.
10667
10668 2018-01-11  H.J. Lu  <hongjiu.lu@intel.com>
10669
10670         PR target/83330
10671         * config/i386/i386.c (ix86_compute_frame_layout): Align stack
10672         frame if argument is passed on stack.
10673
10674 2018-01-11  Jakub Jelinek  <jakub@redhat.com>
10675
10676         PR target/82682
10677         * ree.c (combine_reaching_defs): Optimize also
10678         reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
10679         reg2=any_extend(exp); reg1=reg2;, formatting fix.
10680
10681 2018-01-11  Jan Hubicka  <hubicka@ucw.cz>
10682
10683         PR middle-end/83189
10684         * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
10685
10686 2018-01-11  Jan Hubicka  <hubicka@ucw.cz>
10687
10688         PR middle-end/83718
10689         * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
10690         after they are computed.
10691
10692 2018-01-11  Bin Cheng  <bin.cheng@arm.com>
10693
10694         PR tree-optimization/83695
10695         * gimple-loop-linterchange.cc
10696         (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
10697         reset cached scev information after interchange.
10698         (pass_linterchange::execute): Remove call to scev_reset_htab.
10699
10700 2018-01-11  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
10701
10702         * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
10703         vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
10704         vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
10705         vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
10706         vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
10707         vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
10708         * config/arm/arm_neon_builtins.def (vfmal_lane_low,
10709         vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
10710         vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
10711         vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
10712         vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
10713         * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
10714         (V_lane_reg): Likewise.
10715         * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
10716         New define_expand.
10717         (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
10718         (vfmal_lane_low<mode>_intrinsic,
10719         vfmal_lane_low<vfmlsel2><mode>_intrinsic,
10720         vfmal_lane_high<vfmlsel2><mode>_intrinsic,
10721         vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
10722         vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
10723         vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
10724         vfmsl_lane_high<mode>_intrinsic): New define_insns.
10725
10726 2018-01-11  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
10727
10728         * config/arm/arm-cpus.in (fp16fml): New feature.
10729         (ALL_SIMD): Add fp16fml.
10730         (armv8.2-a): Add fp16fml as an option.
10731         (armv8.3-a): Likewise.
10732         (armv8.4-a): Add fp16fml as part of fp16.
10733         * config/arm/arm.h (TARGET_FP16FML): Define.
10734         * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
10735         when appropriate.
10736         * config/arm/arm-modes.def (V2HF): Define.
10737         * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
10738         vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
10739         vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
10740         * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
10741         vfmsl_low, vfmsl_high): New set of builtins.
10742         * config/arm/iterators.md (PLUSMINUS): New code iterator.
10743         (vfml_op): New code attribute.
10744         (VFMLHALVES): New int iterator.
10745         (VFML, VFMLSEL): New mode attributes.
10746         (V_reg): Define mapping for V2HF.
10747         (V_hi, V_lo): New mode attributes.
10748         (VF_constraint): Likewise.
10749         (vfml_half, vfml_half_selector): New int attributes.
10750         * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
10751         define_expand.
10752         (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
10753         vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
10754         New define_insn.
10755         * config/arm/t-arm-elf (v8_fps): Add fp16fml.
10756         * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
10757         * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
10758         * doc/invoke.texi (ARM Options): Document fp16fml.  Update armv8.4-a
10759         documentation.
10760         * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
10761         Document new effective target and option set.
10762
10763 2018-01-11  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
10764
10765         * config/arm/arm-cpus.in (armv8_4): New feature.
10766         (ARMv8_4a): New fgroup.
10767         (armv8.4-a): New arch.
10768         * config/arm/arm-tables.opt: Regenerate.
10769         * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
10770         * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
10771         * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
10772         Add matching rules for -march=armv8.4-a and extensions.
10773         * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
10774
10775 2018-01-11  Oleg Endo  <olegendo@gcc.gnu.org>
10776
10777         PR target/81821
10778         * config/rx/rx.md (BW): New mode attribute.
10779         (sync_lock_test_and_setsi): Add mode suffix to insn output.
10780
10781 2018-01-11  Richard Biener  <rguenther@suse.de>
10782
10783         PR tree-optimization/83435
10784         * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
10785         * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
10786         * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
10787
10788 2018-01-11  Richard Sandiford  <richard.sandiford@linaro.org>
10789             Alan Hayward  <alan.hayward@arm.com>
10790             David Sherwood  <david.sherwood@arm.com>
10791
10792         * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
10793         field.
10794         (aarch64_classify_address): Initialize it.  Track polynomial offsets.
10795         (aarch64_print_address_internal): Use it to check for a zero offset.
10796
10797 2018-01-11  Richard Sandiford  <richard.sandiford@linaro.org>
10798             Alan Hayward  <alan.hayward@arm.com>
10799             David Sherwood  <david.sherwood@arm.com>
10800
10801         * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
10802         * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
10803         Return a poly_int64 rather than a HOST_WIDE_INT.
10804         (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
10805         rather than a HOST_WIDE_INT.
10806         * config/aarch64/aarch64.h (aarch64_frame): Protect with
10807         HAVE_POLY_INT_H rather than HOST_WIDE_INT.  Change locals_offset,
10808         hard_fp_offset, frame_size, initial_adjust, callee_offset and
10809         final_offset from HOST_WIDE_INT to poly_int64.
10810         * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
10811         to_constant when getting the number of units in an Advanced SIMD
10812         mode.
10813         (aarch64_builtin_vectorized_function): Check for a constant number
10814         of units.
10815         * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
10816         GET_MODE_SIZE.
10817         (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
10818         attribute instead of GET_MODE_NUNITS.
10819         * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
10820         (aarch64_class_max_nregs): Use the constant_lowest_bound of the
10821         GET_MODE_SIZE for fixed-size registers.
10822         (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
10823         (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
10824         (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
10825         (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
10826         (aarch64_print_operand, aarch64_print_address_internal)
10827         (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
10828         (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
10829         (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
10830         Handle polynomial GET_MODE_SIZE.
10831         (aarch64_hard_regno_caller_save_mode): Likewise.  Return modes
10832         wider than SImode without modification.
10833         (tls_symbolic_operand_type): Use strip_offset instead of split_const.
10834         (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
10835         (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
10836         passing and returning SVE modes.
10837         (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
10838         rather than GEN_INT.
10839         (aarch64_emit_probe_stack_range): Take the size as a poly_int64
10840         rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
10841         (aarch64_allocate_and_probe_stack_space): Likewise.
10842         (aarch64_layout_frame): Cope with polynomial offsets.
10843         (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
10844         start_offset as a poly_int64 rather than a HOST_WIDE_INT.  Track
10845         polynomial offsets.
10846         (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
10847         (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
10848         poly_int64 rather than a HOST_WIDE_INT.
10849         (aarch64_get_separate_components, aarch64_process_components)
10850         (aarch64_expand_prologue, aarch64_expand_epilogue)
10851         (aarch64_use_return_insn_p): Handle polynomial frame offsets.
10852         (aarch64_anchor_offset): New function, split out from...
10853         (aarch64_legitimize_address): ...here.
10854         (aarch64_builtin_vectorization_cost): Handle polynomial
10855         TYPE_VECTOR_SUBPARTS.
10856         (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
10857         GET_MODE_NUNITS.
10858         (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
10859         number of elements from the PARALLEL rather than the mode.
10860         (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
10861         rather than GET_MODE_BITSIZE.
10862         (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
10863         (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
10864         (aarch64_expand_vec_perm_const_1): Handle polynomial
10865         d->perm.length () and d->perm elements.
10866         (aarch64_evpc_tbl): Likewise.  Use nelt rather than GET_MODE_NUNITS.
10867         Apply to_constant to d->perm elements.
10868         (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
10869         polynomial CONST_VECTOR_NUNITS.
10870         (aarch64_move_pointer): Take amount as a poly_int64 rather
10871         than an int.
10872         (aarch64_progress_pointer): Avoid temporary variable.
10873         * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
10874         the mode attribute instead of GET_MODE.
10875
10876 2018-01-11  Richard Sandiford  <richard.sandiford@linaro.org>
10877             Alan Hayward  <alan.hayward@arm.com>
10878             David Sherwood  <david.sherwood@arm.com>
10879
10880         * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
10881         x exists before using it.
10882         (aarch64_add_constant_internal): Rename to...
10883         (aarch64_add_offset_1): ...this.  Replace regnum with separate
10884         src and dest rtxes.  Handle the case in which they're different,
10885         including when the offset is zero.  Replace scratchreg with an rtx.
10886         Use 2 additions if there is no spare register into which we can
10887         move a 16-bit constant.
10888         (aarch64_add_constant): Delete.
10889         (aarch64_add_offset): Replace reg with separate src and dest
10890         rtxes.  Take a poly_int64 offset instead of a HOST_WIDE_INT.
10891         Use aarch64_add_offset_1.
10892         (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
10893         an rtx rather than an int.  Take the delta as a poly_int64
10894         rather than a HOST_WIDE_INT.  Use aarch64_add_offset.
10895         (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
10896         (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
10897         aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
10898         (aarch64_expand_epilogue): Update calls to aarch64_add_offset
10899         and aarch64_add_sp.
10900         (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
10901         aarch64_add_constant.
10902
10903 2018-01-11  Richard Sandiford  <richard.sandiford@linaro.org>
10904
10905         * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
10906         Use scalar_float_mode.
10907
10908 2018-01-11  Richard Sandiford  <richard.sandiford@linaro.org>
10909
10910         * config/aarch64/aarch64-simd.md
10911         (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
10912         (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
10913         (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
10914         (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
10915         (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
10916         (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
10917         (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
10918         (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
10919         (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
10920         (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
10921
10922 2018-01-11  Prathamesh Kulkarni  <prathamesh.kulkarni@linaro.org>
10923
10924         PR target/83514
10925         * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
10926         targ_options->x_arm_arch_string is non NULL.
10927
10928 2018-01-11  Tamar Christina  <tamar.christina@arm.com>
10929
10930         * config/aarch64/aarch64.h
10931         (AARCH64_FL_FOR_ARCH8_4): Add  AARCH64_FL_DOTPROD.
10932
10933 2018-01-11  Sudakshina Das  <sudi.das@arm.com>
10934
10935         PR target/82096
10936         * expmed.c (emit_store_flag_force): Swap if const op0
10937         and change VOIDmode to mode of op0.
10938
10939 2018-01-11  Richard Sandiford  <richard.sandiford@linaro.org>
10940
10941         PR rtl-optimization/83761
10942         * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
10943         than bytes to mode_for_size.
10944
10945 2018-01-10  Jan Hubicka  <hubicka@ucw.cz>
10946
10947         PR middle-end/83189
10948         * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
10949         * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
10950         profile.
10951
10952 2018-01-10  Jan Hubicka  <hubicka@ucw.cz>
10953
10954         PR middle-end/83575
10955         * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
10956         when in layout mode.
10957         (cfg_layout_finalize): Do not verify cfg before we are out of layout.
10958         * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
10959         partition fixup.
10960
10961 2018-01-10  Michael Collison  <michael.collison@arm.com>
10962
10963         * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
10964         * config/aarch64/aarch64-option-extension.def: Add
10965         AARCH64_OPT_EXTENSION of 'fp16fml'.
10966         * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
10967         (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
10968         * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
10969         * config/aarch64/constraints.md (Ui7): New constraint.
10970         * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
10971         (VFMLA_SEL_W): Ditto.
10972         (f16quad): Ditto.
10973         (f16mac1): Ditto.
10974         (VFMLA16_LOW): New int iterator.
10975         (VFMLA16_HIGH): Ditto.
10976         (UNSPEC_FMLAL): New unspec.
10977         (UNSPEC_FMLSL): Ditto.
10978         (UNSPEC_FMLAL2): Ditto.
10979         (UNSPEC_FMLSL2): Ditto.
10980         (f16mac): New code attribute.
10981         * config/aarch64/aarch64-simd-builtins.def
10982         (aarch64_fmlal_lowv2sf): Ditto.
10983         (aarch64_fmlsl_lowv2sf): Ditto.
10984         (aarch64_fmlalq_lowv4sf): Ditto.
10985         (aarch64_fmlslq_lowv4sf): Ditto.
10986         (aarch64_fmlal_highv2sf): Ditto.
10987         (aarch64_fmlsl_highv2sf): Ditto.
10988         (aarch64_fmlalq_highv4sf): Ditto.
10989         (aarch64_fmlslq_highv4sf): Ditto.
10990         (aarch64_fmlal_lane_lowv2sf): Ditto.
10991         (aarch64_fmlsl_lane_lowv2sf): Ditto.
10992         (aarch64_fmlal_laneq_lowv2sf): Ditto.
10993         (aarch64_fmlsl_laneq_lowv2sf): Ditto.
10994         (aarch64_fmlalq_lane_lowv4sf): Ditto.
10995         (aarch64_fmlsl_lane_lowv4sf): Ditto.
10996         (aarch64_fmlalq_laneq_lowv4sf): Ditto.
10997         (aarch64_fmlsl_laneq_lowv4sf): Ditto.
10998         (aarch64_fmlal_lane_highv2sf): Ditto.
10999         (aarch64_fmlsl_lane_highv2sf): Ditto.
11000         (aarch64_fmlal_laneq_highv2sf): Ditto.
11001         (aarch64_fmlsl_laneq_highv2sf): Ditto.
11002         (aarch64_fmlalq_lane_highv4sf): Ditto.
11003         (aarch64_fmlsl_lane_highv4sf): Ditto.
11004         (aarch64_fmlalq_laneq_highv4sf): Ditto.
11005         (aarch64_fmlsl_laneq_highv4sf): Ditto.
11006         * config/aarch64/aarch64-simd.md:
11007         (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
11008         (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
11009         (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
11010         (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
11011         (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
11012         (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
11013         (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
11014         (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
11015         (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
11016         (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
11017         (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
11018         (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
11019         (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
11020         (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
11021         (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
11022         (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
11023         (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
11024         (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
11025         (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
11026         (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
11027         * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
11028         (vfmlsl_low_u32): Ditto.
11029         (vfmlalq_low_u32): Ditto.
11030         (vfmlslq_low_u32): Ditto.
11031         (vfmlal_high_u32): Ditto.
11032         (vfmlsl_high_u32): Ditto.
11033         (vfmlalq_high_u32): Ditto.
11034         (vfmlslq_high_u32): Ditto.
11035         (vfmlal_lane_low_u32): Ditto.
11036         (vfmlsl_lane_low_u32): Ditto.
11037         (vfmlal_laneq_low_u32): Ditto.
11038         (vfmlsl_laneq_low_u32): Ditto.
11039         (vfmlalq_lane_low_u32): Ditto.
11040         (vfmlslq_lane_low_u32): Ditto.
11041         (vfmlalq_laneq_low_u32): Ditto.
11042         (vfmlslq_laneq_low_u32): Ditto.
11043         (vfmlal_lane_high_u32): Ditto.
11044         (vfmlsl_lane_high_u32): Ditto.
11045         (vfmlal_laneq_high_u32): Ditto.
11046         (vfmlsl_laneq_high_u32): Ditto.
11047         (vfmlalq_lane_high_u32): Ditto.
11048         (vfmlslq_lane_high_u32): Ditto.
11049         (vfmlalq_laneq_high_u32): Ditto.
11050         (vfmlslq_laneq_high_u32): Ditto.
11051         * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
11052         (AARCH64_FL_FOR_ARCH8_4): New.
11053         (AARCH64_ISA_F16FML): New ISA flag.
11054         (TARGET_F16FML): New feature flag for fp16fml.
11055         (doc/invoke.texi): Document new fp16fml option.
11056
11057 2018-01-10  Michael Collison  <michael.collison@arm.com>
11058
11059         * config/aarch64/aarch64-builtins.c:
11060         (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
11061         * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
11062         (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
11063         * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
11064         (AARCH64_ISA_SHA3): New ISA flag.
11065         (TARGET_SHA3): New feature flag for sha3.
11066         * config/aarch64/iterators.md (sha512_op): New int attribute.
11067         (CRYPTO_SHA512): New int iterator.
11068         (UNSPEC_SHA512H): New unspec.
11069         (UNSPEC_SHA512H2): Ditto.
11070         (UNSPEC_SHA512SU0): Ditto.
11071         (UNSPEC_SHA512SU1): Ditto.
11072         * config/aarch64/aarch64-simd-builtins.def
11073         (aarch64_crypto_sha512hqv2di): New builtin.
11074         (aarch64_crypto_sha512h2qv2di): Ditto.
11075         (aarch64_crypto_sha512su0qv2di): Ditto.
11076         (aarch64_crypto_sha512su1qv2di): Ditto.
11077         (aarch64_eor3qv8hi): Ditto.
11078         (aarch64_rax1qv2di): Ditto.
11079         (aarch64_xarqv2di): Ditto.
11080         (aarch64_bcaxqv8hi): Ditto.
11081         * config/aarch64/aarch64-simd.md:
11082         (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
11083         (aarch64_crypto_sha512su0qv2di): Ditto.
11084         (aarch64_crypto_sha512su1qv2di): Ditto.
11085         (aarch64_eor3qv8hi): Ditto.
11086         (aarch64_rax1qv2di): Ditto.
11087         (aarch64_xarqv2di): Ditto.
11088         (aarch64_bcaxqv8hi): Ditto.
11089         * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
11090         (vsha512h2q_u64): Ditto.
11091         (vsha512su0q_u64): Ditto.
11092         (vsha512su1q_u64): Ditto.
11093         (veor3q_u16): Ditto.
11094         (vrax1q_u64): Ditto.
11095         (vxarq_u64): Ditto.
11096         (vbcaxq_u16): Ditto.
11097         * config/arm/types.md (crypto_sha512): New type attribute.
11098         (crypto_sha3): Ditto.
11099         (doc/invoke.texi): Document new sha3 option.
11100
11101 2018-01-10  Michael Collison  <michael.collison@arm.com>
11102
11103         * config/aarch64/aarch64-builtins.c:
11104         (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
11105         * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
11106         (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
11107         (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
11108         * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
11109         (AARCH64_ISA_SM4): New ISA flag.
11110         (TARGET_SM4): New feature flag for sm4.
11111         * config/aarch64/aarch64-simd-builtins.def
11112         (aarch64_sm3ss1qv4si): Ditto.
11113         (aarch64_sm3tt1aq4si): Ditto.
11114         (aarch64_sm3tt1bq4si): Ditto.
11115         (aarch64_sm3tt2aq4si): Ditto.
11116         (aarch64_sm3tt2bq4si): Ditto.
11117         (aarch64_sm3partw1qv4si): Ditto.
11118         (aarch64_sm3partw2qv4si): Ditto.
11119         (aarch64_sm4eqv4si): Ditto.
11120         (aarch64_sm4ekeyqv4si): Ditto.
11121         * config/aarch64/aarch64-simd.md:
11122         (aarch64_sm3ss1qv4si): Ditto.
11123         (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
11124         (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
11125         (aarch64_sm4eqv4si): Ditto.
11126         (aarch64_sm4ekeyqv4si): Ditto.
11127         * config/aarch64/iterators.md (sm3tt_op): New int iterator.
11128         (sm3part_op): Ditto.
11129         (CRYPTO_SM3TT): Ditto.
11130         (CRYPTO_SM3PART): Ditto.
11131         (UNSPEC_SM3SS1): New unspec.
11132         (UNSPEC_SM3TT1A): Ditto.
11133         (UNSPEC_SM3TT1B): Ditto.
11134         (UNSPEC_SM3TT2A): Ditto.
11135         (UNSPEC_SM3TT2B): Ditto.
11136         (UNSPEC_SM3PARTW1): Ditto.
11137         (UNSPEC_SM3PARTW2): Ditto.
11138         (UNSPEC_SM4E): Ditto.
11139         (UNSPEC_SM4EKEY): Ditto.
11140         * config/aarch64/constraints.md (Ui2): New constraint.
11141         * config/aarch64/predicates.md (aarch64_imm2): New predicate.
11142         * config/arm/types.md (crypto_sm3): New type attribute.
11143         (crypto_sm4): Ditto.
11144         * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
11145         (vsm3tt1aq_u32): Ditto.
11146         (vsm3tt1bq_u32): Ditto.
11147         (vsm3tt2aq_u32): Ditto.
11148         (vsm3tt2bq_u32): Ditto.
11149         (vsm3partw1q_u32): Ditto.
11150         (vsm3partw2q_u32): Ditto.
11151         (vsm4eq_u32): Ditto.
11152         (vsm4ekeyq_u32): Ditto.
11153         (doc/invoke.texi): Document new sm4 option.
11154
11155 2018-01-10  Michael Collison  <michael.collison@arm.com>
11156
11157         * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
11158         * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
11159         (AARCH64_FL_FOR_ARCH8_4): New.
11160         (AARCH64_FL_V8_4): New flag.
11161         (doc/invoke.texi): Document new armv8.4-a option.
11162
11163 2018-01-10  Michael Collison  <michael.collison@arm.com>
11164
11165         * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
11166         (__ARM_FEATURE_AES): Define if TARGET_AES is true.
11167         (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
11168         * config/aarch64/aarch64-option-extension.def: Add
11169         AARCH64_OPT_EXTENSION of 'sha2'.
11170         (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
11171         (crypto): Disable sha2 and aes if crypto disabled.
11172         (crypto): Enable aes and sha2 if enabled.
11173         (simd): Disable sha2 and aes if simd disabled.
11174         * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
11175         New flags.
11176         (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
11177         (TARGET_SHA2): New feature flag for sha2.
11178         (TARGET_AES): New feature flag for aes.
11179         * config/aarch64/aarch64-simd.md:
11180         (aarch64_crypto_aes<aes_op>v16qi): Make pattern
11181         conditional on TARGET_AES.
11182         (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
11183         (aarch64_crypto_sha1hsi): Make pattern conditional
11184         on TARGET_SHA2.
11185         (aarch64_crypto_sha1hv4si): Ditto.
11186         (aarch64_be_crypto_sha1hv4si): Ditto.
11187         (aarch64_crypto_sha1su1v4si): Ditto.
11188         (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
11189         (aarch64_crypto_sha1su0v4si): Ditto.
11190         (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
11191         (aarch64_crypto_sha256su0v4si): Ditto.
11192         (aarch64_crypto_sha256su1v4si): Ditto.
11193         (doc/invoke.texi): Document new aes and sha2 options.
11194
11195 2018-01-10  Martin Sebor  <msebor@redhat.com>
11196
11197         PR tree-optimization/83781
11198         * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
11199         as string arrays.
11200
11201 2018-01-11  Martin Sebor  <msebor@gmail.com>
11202             Prathamesh Kulkarni  <prathamesh.kulkarni@linaro.org>
11203
11204         PR tree-optimization/83501
11205         PR tree-optimization/81703
11206
11207         * tree-ssa-strlen.c (get_string_cst): Rename...
11208         (get_string_len): ...to this.  Handle global constants.
11209         (handle_char_store): Adjust.
11210
11211 2018-01-10  Kito Cheng  <kito.cheng@gmail.com>
11212             Jim Wilson  <jimw@sifive.com>
11213
11214         * config/riscv/riscv-protos.h (riscv_output_return): New.
11215         * config/riscv/riscv.c (struct machine_function): New naked_p field.
11216         (riscv_attribute_table, riscv_output_return),
11217         (riscv_handle_fndecl_attribute, riscv_naked_function_p),
11218         (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
11219         (riscv_compute_frame_info): Only compute frame->mask if not a naked
11220         function.
11221         (riscv_expand_prologue): Add early return for naked function.
11222         (riscv_expand_epilogue): Likewise.
11223         (riscv_function_ok_for_sibcall): Return false for naked function.
11224         (riscv_set_current_function): New.
11225         (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
11226         (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
11227         * config/riscv/riscv.md (simple_return): Call riscv_output_return.
11228         * doc/extend.texi (RISC-V Function Attributes): New.
11229
11230 2018-01-10  Michael Meissner  <meissner@linux.vnet.ibm.com>
11231
11232         * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
11233         check for 128-bit long double before checking TCmode.
11234         * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
11235         128-bit long doubles before checking TFmode or TCmode.
11236         (FLOAT128_IBM_P): Likewise.
11237
11238 2018-01-10  Martin Sebor  <msebor@redhat.com>
11239
11240         PR tree-optimization/83671
11241         * builtins.c (c_strlen): Unconditionally return zero for the empty
11242         string.
11243         Use -Warray-bounds for warnings.
11244         * gimple-fold.c (get_range_strlen): Handle non-constant lengths
11245         for non-constant array indices with COMPONENT_REF, arrays of
11246         arrays, and pointers to arrays.
11247         (gimple_fold_builtin_strlen): Determine and set length range for
11248         non-constant character arrays.
11249
11250 2018-01-10  Aldy Hernandez  <aldyh@redhat.com>
11251
11252         PR middle-end/81897
11253         * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
11254         empty blocks.
11255
11256 2018-01-10  Eric Botcazou  <ebotcazou@adacore.com>
11257
11258         * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
11259
11260 2018-01-10  Peter Bergner  <bergner@vnet.ibm.com>
11261
11262         PR target/83399
11263         * config/rs6000/rs6000.c (print_operand) <'y'>: Use
11264         VECTOR_MEM_ALTIVEC_OR_VSX_P.
11265         * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
11266         indexed_or_indirect_operand predicate.
11267         (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
11268         (*vsx_le_perm_load_v8hi): Likewise.
11269         (*vsx_le_perm_load_v16qi): Likewise.
11270         (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
11271         (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
11272         (*vsx_le_perm_store_v8hi): Likewise.
11273         (*vsx_le_perm_store_v16qi): Likewise.
11274         (eight unnamed splitters): Likewise.
11275
11276 2018-01-10  Peter Bergner  <bergner@vnet.ibm.com>
11277
11278         * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
11279         * config/rs6000/emmintrin.h: Likewise.
11280         * config/rs6000/mmintrin.h: Likewise.
11281         * config/rs6000/xmmintrin.h: Likewise.
11282
11283 2018-01-10  David Malcolm  <dmalcolm@redhat.com>
11284
11285         PR c++/43486
11286         * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
11287         "public_flag".
11288         * tree.c (tree_nop_conversion): Return true for location wrapper
11289         nodes.
11290         (maybe_wrap_with_location): New function.
11291         (selftest::check_strip_nops): New function.
11292         (selftest::test_location_wrappers): New function.
11293         (selftest::tree_c_tests): Call it.
11294         * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
11295         (maybe_wrap_with_location): New decl.
11296         (EXPR_LOCATION_WRAPPER_P): New macro.
11297         (location_wrapper_p): New inline function.
11298         (tree_strip_any_location_wrapper): New inline function.
11299
11300 2018-01-10  H.J. Lu  <hongjiu.lu@intel.com>
11301
11302         PR target/83735
11303         * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
11304         stack_realign_offset for the largest alignment of stack slot
11305         actually used.
11306         (ix86_find_max_used_stack_alignment): New function.
11307         (ix86_finalize_stack_frame_flags): Use it.  Set
11308         max_used_stack_alignment if we don't realign stack.
11309         * config/i386/i386.h (machine_function): Add
11310         max_used_stack_alignment.
11311
11312 2018-01-10  Christophe Lyon  <christophe.lyon@linaro.org>
11313
11314         * config/arm/arm.opt (-mbranch-cost): New option.
11315         * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
11316         account.
11317
11318 2018-01-10  Segher Boessenkool  <segher@kernel.crashing.org>
11319
11320         PR target/83629
11321         * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
11322         load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
11323
11324 2018-01-10  Richard Biener  <rguenther@suse.de>
11325
11326         PR debug/83765
11327         * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
11328         early out so it also covers the case where we have a non-NULL
11329         origin.
11330
11331 2018-01-10  Richard Sandiford  <richard.sandiford@linaro.org>
11332
11333         PR tree-optimization/83753
11334         * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
11335         for non-strided grouped accesses if the number of elements is 1.
11336
11337 2018-01-10  Jan Hubicka  <hubicka@ucw.cz>
11338
11339         PR target/81616
11340         * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
11341         * i386.h (TARGET_USE_GATHER): Define.
11342         * x86-tune.def (X86_TUNE_USE_GATHER): New.
11343
11344 2018-01-10  Martin Liska  <mliska@suse.cz>
11345
11346         PR bootstrap/82831
11347         * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
11348         * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
11349         partitioning.
11350         * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
11351         CLEANUP_NO_PARTITIONING is not set.
11352
11353 2018-01-10  Richard Sandiford  <richard.sandiford@linaro.org>
11354
11355         * doc/rtl.texi: Remove documentation of (const ...) wrappers
11356         for vectors, as a partial revert of r254296.
11357         * rtl.h (const_vec_p): Delete.
11358         (const_vec_duplicate_p): Don't test for vector CONSTs.
11359         (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
11360         * expmed.c (make_tree): Likewise.
11361
11362         Revert:
11363         * common.md (E, F): Use CONSTANT_P instead of checking for
11364         CONST_VECTOR.
11365         * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
11366         checking for CONST_VECTOR.
11367
11368 2018-01-09  Jan Hubicka  <hubicka@ucw.cz>
11369
11370         PR middle-end/83575
11371         * predict.c (force_edge_cold): Handle in more sane way edges
11372         with no prediction.
11373
11374 2018-01-09  Carl Love  <cel@us.ibm.com>
11375
11376         * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
11377         V4SI, V4SF types.
11378         (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
11379         * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
11380         VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
11381         VMRGOW_V2DI, VMRGOW_V2DF.  Remove definition for VMRGOW.
11382         * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
11383         P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW):  Add definitions.
11384         * config/rs6000/rs6000-protos.h: Add extern defition for
11385         rs6000_generate_float2_double_code.
11386         * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
11387         function.
11388         * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
11389         (float2_v2df): Add define_expand.
11390
11391 2018-01-09  Uros Bizjak  <ubizjak@gmail.com>
11392
11393         PR target/83628
11394         * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
11395         op_mode in the force_to_mode call.
11396
11397 2018-01-09  Richard Sandiford  <richard.sandiford@linaro.org>
11398
11399         * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
11400         instead of checking each element individually.
11401         (aarch64_evpc_uzp): Likewise.
11402         (aarch64_evpc_zip): Likewise.
11403         (aarch64_evpc_ext): Likewise.
11404         (aarch64_evpc_rev): Likewise.
11405         (aarch64_evpc_dup): Test the encoding for a single duplicated element,
11406         instead of checking each element individually.  Return true without
11407         generating rtl if
11408         (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
11409         whether all selected elements come from the same input, instead of
11410         checking each element individually.  Remove calls to gen_rtx_REG,
11411         start_sequence and end_sequence and instead assert that no rtl is
11412         generated.
11413
11414 2018-01-09  Richard Sandiford  <richard.sandiford@linaro.org>
11415
11416         * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
11417         order of HIGH and CONST checks.
11418
11419 2018-01-09  Richard Sandiford  <richard.sandiford@linaro.org>
11420
11421         * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
11422         if the destination isn't an SSA_NAME.
11423
11424 2018-01-09  Richard Biener  <rguenther@suse.de>
11425
11426         PR tree-optimization/83668
11427         * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
11428         move prologue...
11429         (canonicalize_loop_form): ... here, renamed from ...
11430         (canonicalize_loop_closed_ssa_form): ... this and amended to
11431         swap successor edges for loop exit blocks to make us use
11432         the RPO order we need for initial schedule generation.
11433
11434 2018-01-09  Joseph Myers  <joseph@codesourcery.com>
11435
11436         PR tree-optimization/64811
11437         * match.pd: When optimizing comparisons with Inf, avoid
11438         introducing or losing exceptions from comparisons with NaN.
11439
11440 2018-01-09  Martin Liska  <mliska@suse.cz>
11441
11442         PR sanitizer/82517
11443         * asan.c (shadow_mem_size): Add gcc_assert.
11444
11445 2018-01-09  Georg-Johann Lay  <avr@gjlay.de>
11446
11447         Don't save registers in main().
11448
11449         PR target/83738
11450         * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
11451         * config/avr/avr.opt (-mmain-is-OS_task): New target option.
11452         * config/avr/avr.c (avr_set_current_function): Don't error if
11453         naked, OS_task or OS_main are specified at the same time.
11454         (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
11455         OS_main.
11456         (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
11457         attribute.
11458         * common/config/avr/avr-common.c (avr_option_optimization_table):
11459         Switch on -mmain-is-OS_task for optimizing compilations.
11460
11461 2018-01-09  Richard Biener  <rguenther@suse.de>
11462
11463         PR tree-optimization/83572
11464         * graphite.c: Include cfganal.h.
11465         (graphite_transform_loops): Connect infinite loops to exit
11466         and remove fake edges at the end.
11467
11468 2018-01-09  Jan Hubicka  <hubicka@ucw.cz>
11469
11470         * ipa-inline.c (edge_badness): Revert accidental checkin.
11471
11472 2018-01-09  Jan Hubicka  <hubicka@ucw.cz>
11473
11474         PR ipa/80763
11475         * ipa-comdats.c (set_comdat_group): Only set comdat group of real
11476         symbols; not inline clones.
11477
11478 2018-01-09  Jakub Jelinek  <jakub@redhat.com>
11479
11480         PR target/83507
11481         * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
11482         hard registers.  Formatting fixes.
11483
11484         PR preprocessor/83722
11485         * gcc.c (try_generate_repro): Pass
11486         &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
11487         &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
11488         do_report_bug.
11489
11490 2018-01-08  Monk Chiang  <sh.chiang04@gmail.com>
11491             Kito Cheng  <kito.cheng@gmail.com>
11492
11493         * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
11494         (riscv_leaf_function_p): Delete.
11495         (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
11496
11497 2018-01-08  Aaron Sawdey  <acsawdey@linux.vnet.ibm.com>
11498
11499         * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
11500         function.
11501         (do_ifelse): New function.
11502         (do_isel): New function.
11503         (do_sub3): New function.
11504         (do_add3): New function.
11505         (do_load_mask_compare): New function.
11506         (do_overlap_load_compare): New function.
11507         (expand_compare_loop): New function.
11508         (expand_block_compare): Call expand_compare_loop() when appropriate.
11509         * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
11510         option description.
11511         (-mblock-compare-inline-loop-limit): New option.
11512
11513 2018-01-08  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
11514
11515         PR target/83677
11516         * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
11517         Reverse order of second and third operands in first alternative.
11518         * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
11519         of first and second elements in UNSPEC_VPERMR vector.
11520         (altivec_expand_vec_perm_le): Likewise.
11521
11522 2018-01-08  Jeff Law  <law@redhat.com>
11523
11524         PR rtl-optimizatin/81308
11525         * tree-switch-conversion.c (cfg_altered): New file scoped static.
11526         (process_switch): If group_case_labels makes a change, then set
11527         cfg_altered.
11528         (pass_convert_switch::execute): If a switch is converted, then
11529         set cfg_altered.  Return TODO_cfg_cleanup if cfg_altered is true.
11530
11531         PR rtl-optimization/81308
11532         * recog.c (split_all_insns): Conditionally cleanup the CFG after
11533         splitting insns.
11534
11535 2018-01-08  Vidya Praveen  <vidyapraveen@arm.com>
11536
11537         PR target/83663 - Revert r255946
11538         * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
11539         generation for cases where splatting a value is not useful.
11540         * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
11541         across a vec_duplicate and a paradoxical subreg forming a vector
11542         mode to a vec_concat.
11543
11544 2018-01-08  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
11545
11546         * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
11547         -march=armv8.3-a variants.
11548         * config/arm/t-multilib: Likewise.
11549         * config/arm/t-arm-elf: Likewise.  Handle dotprod extension.
11550
11551 2018-01-08  Aaron Sawdey  <acsawdey@linux.vnet.ibm.com>
11552
11553         * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
11554         to generate rtl.
11555         (cceq_ior_compare_complement): Give it a name so I can use it, and
11556         change boolean_or_operator predicate to boolean_operator so it can
11557         be used to generate a crand.
11558         (eqne): New code iterator.
11559         (bd/bd_neg): New code_attrs.
11560         (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
11561         a single define_insn.
11562         (<bd>tf_<mode>): A new insn pattern for the conditional form branch
11563         decrement (bdnzt/bdnzf/bdzt/bdzf).
11564         * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
11565         with the new names of the branch decrement patterns, and added the
11566         names of the branch decrement conditional patterns.
11567
11568 2018-01-08  Richard Biener  <rguenther@suse.de>
11569
11570         PR tree-optimization/83563
11571         * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
11572         cache.
11573
11574 2018-01-08  Richard Biener  <rguenther@suse.de>
11575
11576         PR middle-end/83713
11577         * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
11578
11579 2018-01-08  Richard Biener  <rguenther@suse.de>
11580
11581         PR tree-optimization/83685
11582         * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
11583         references to abnormals.
11584
11585 2018-01-08  Richard Biener  <rguenther@suse.de>
11586
11587         PR lto/83719
11588         * dwarf2out.c (output_indirect_strings): Handle empty
11589         skeleton_debug_str_hash.
11590         (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
11591
11592 2018-01-08  Claudiu Zissulescu  <claziss@synopsys.com>
11593
11594         * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
11595         (emit_store_direct): Likewise.
11596         (arc_trampoline_adjust_address): Likewise.
11597         (arc_asm_trampoline_template): New function.
11598         (arc_initialize_trampoline): Use asm_trampoline_template.
11599         (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
11600         * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
11601         * config/arc/arc.md (flush_icache): Delete pattern.
11602
11603 2018-01-08  Claudiu Zissulescu  <claziss@synopsys.com>
11604
11605         * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
11606         * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
11607         munaligned-access.
11608
11609 2018-01-08  Sebastian Huber  <sebastian.huber@embedded-brains.de>
11610
11611         PR target/83681
11612         * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
11613         by not USED_FOR_TARGET.
11614         (make_pass_resolve_sw_modes): Likewise.
11615
11616 2018-01-08  Sebastian Huber  <sebastian.huber@embedded-brains.de>
11617
11618         * config/nios2/nios2.h (nios2_section_threshold): Guard by not
11619         USED_FOR_TARGET.
11620
11621 2018-01-08  Richard Biener  <rguenther@suse.de>
11622
11623         PR middle-end/83580
11624         * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
11625
11626 2018-01-08  Richard Biener  <rguenther@suse.de>
11627
11628         PR middle-end/83517
11629         * match.pd ((t * 2) / 2) -> t): Add missing :c.
11630
11631 2018-01-06  Aldy Hernandez  <aldyh@redhat.com>
11632
11633         PR middle-end/81897
11634         * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
11635         basic blocks with a small number of successors.
11636         (convert_control_dep_chain_into_preds): Improve handling of
11637         forwarder blocks.
11638         (dump_predicates): Split apart into...
11639         (dump_pred_chain): ...here...
11640         (dump_pred_info): ...and here.
11641         (can_one_predicate_be_invalidated_p): Add debugging printfs.
11642         (can_chain_union_be_invalidated_p): Improve check for invalidation
11643         of paths.
11644         (uninit_uses_cannot_happen): Avoid unnecessary if
11645         convert_control_dep_chain_into_preds yielded nothing.
11646
11647 2018-01-06  Martin Sebor  <msebor@redhat.com>
11648
11649         PR tree-optimization/83640
11650         * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
11651         subtracting negative offset from size.
11652         (builtin_access::overlap): Adjust offset bounds of the access to fall
11653         within the size of the object if possible.
11654
11655 2018-01-06  Richard Sandiford  <richard.sandiford@linaro.org>
11656
11657         PR rtl-optimization/83699
11658         * expmed.c (extract_bit_field_1): Restrict the vector usage of
11659         extract_bit_field_as_subreg to cases in which the extracted
11660         value is also a vector.
11661
11662         * lra-constraints.c (process_alt_operands): Test for the equivalence
11663         substitutions when detecting a possible reload cycle.
11664
11665 2018-01-06  Jakub Jelinek  <jakub@redhat.com>
11666
11667         PR debug/83480
11668         * toplev.c (process_options): Don't enable debug_nonbind_markers_p
11669         by default if flag_selective_schedling{,2}.  Formatting fixes.
11670
11671         PR rtl-optimization/83682
11672         * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
11673         if it has non-VECTOR_MODE element mode.
11674         (vec_duplicate_p): Likewise.
11675
11676         PR middle-end/83694
11677         * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
11678         and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
11679
11680 2018-01-05  Jakub Jelinek  <jakub@redhat.com>
11681
11682         PR target/83604
11683         * config/i386/i386-builtin.def
11684         (__builtin_ia32_vgf2p8affineinvqb_v64qi,
11685         __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
11686         Require also OPTION_MASK_ISA_AVX512F in addition to
11687         OPTION_MASK_ISA_GFNI.
11688         (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
11689         __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
11690         OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
11691         to OPTION_MASK_ISA_GFNI.
11692         (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
11693         OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
11694         OPTION_MASK_ISA_AVX512BW.
11695         (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
11696         OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
11697         addition to OPTION_MASK_ISA_GFNI.
11698         (__builtin_ia32_vgf2p8affineinvqb_v16qi,
11699         __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
11700         Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
11701         to OPTION_MASK_ISA_GFNI.
11702         * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
11703         a requirement for all ISAs rather than any of them with a few
11704         exceptions.
11705         (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
11706         processing.
11707         (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
11708         bitmasks to be enabled with 3 exceptions, instead of requiring any
11709         enabled ISA with lots of exceptions.
11710         * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
11711         vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
11712         Change avx512bw in isa attribute to avx512f.
11713         * config/i386/sgxintrin.h: Add license boilerplate.
11714         * config/i386/vaesintrin.h: Likewise.  Fix macro spelling __AVX512F
11715         to __AVX512F__ and __AVX512VL to __AVX512VL__.
11716         (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
11717         _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
11718         defined.
11719         * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
11720         _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
11721         temporarily sse2 rather than sse if not enabled already.
11722
11723         PR target/83604
11724         * config/i386/sse.md (VI248_VLBW): Rename to ...
11725         (VI248_AVX512VL): ... this.  Don't guard V32HI with TARGET_AVX512BW.
11726         (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
11727         vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
11728         vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
11729         vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
11730         mode iterator instead of VI248_VLBW.
11731
11732 2018-01-05  Jan Hubicka  <hubicka@ucw.cz>
11733
11734         * ipa-fnsummary.c (record_modified_bb_info): Add OP.
11735         (record_modified): Skip clobbers; add debug output.
11736         (param_change_prob): Use sreal frequencies.
11737
11738 2018-01-05  Richard Sandiford  <richard.sandiford@linaro.org>
11739
11740         * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
11741         punt for user-aligned variables.
11742
11743 2018-01-05  Richard Sandiford  <richard.sandiford@linaro.org>
11744
11745         * tree-chrec.c (chrec_contains_symbols): Return true for
11746         POLY_INT_CST.
11747
11748 2018-01-05  Sudakshina Das  <sudi.das@arm.com>
11749
11750         PR target/82439
11751         * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
11752         of (x|y) == x for BICS pattern.
11753
11754 2018-01-05  Jakub Jelinek  <jakub@redhat.com>
11755
11756         PR tree-optimization/83605
11757         * gimple-ssa-strength-reduction.c: Include tree-eh.h.
11758         (find_candidates_dom_walker::before_dom_children): Ignore stmts that
11759         can throw.
11760
11761 2018-01-05  Sebastian Huber  <sebastian.huber@embedded-brains.de>
11762
11763         * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
11764         * config/epiphany/rtems.h: New file.
11765
11766 2018-01-04  Jakub Jelinek  <jakub@redhat.com>
11767             Uros Bizjak  <ubizjak@gmail.com>
11768
11769         PR target/83554
11770         * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
11771         QIreg_operand instead of register_operand predicate.
11772         * config/i386/i386.c (ix86_rop_should_change_byte_p,
11773         set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
11774         comments instead of -fmitigate[-_]rop.
11775
11776 2018-01-04  Rainer Orth  <ro@CeBiTec.Uni-Bielefeld.DE>
11777
11778         PR bootstrap/81926
11779         * cgraphunit.c (symbol_table::compile): Switch to text_section
11780         before calling assembly_start debug hook.
11781         * run-rtl-passes.c (run_rtl_passes): Likewise.
11782         Include output.h.
11783
11784 2018-01-04  Richard Sandiford  <richard.sandiford@linaro.org>
11785
11786         * tree-vrp.c (extract_range_from_binary_expr_1): Check
11787         range_int_cst_p rather than !symbolic_range_p before calling
11788         extract_range_from_multiplicative_op_1.
11789
11790 2018-01-04  Jeff Law  <law@redhat.com>
11791
11792         * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
11793         redundant test in assertion.
11794
11795 2018-01-04  Richard Sandiford  <richard.sandiford@linaro.org>
11796
11797         * doc/rtl.texi: Document machine_mode wrapper classes.
11798
11799 2018-01-04  Richard Sandiford  <richard.sandiford@linaro.org>
11800
11801         * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
11802         using tree_to_uhwi.
11803
11804 2018-01-04  Richard Sandiford  <richard.sandiford@linaro.org>
11805
11806         * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
11807         the VEC_PERM_EXPR fold to fail.
11808
11809 2018-01-04  Jakub Jelinek  <jakub@redhat.com>
11810
11811         PR debug/83585
11812         * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
11813         to switched_sections.
11814
11815 2018-01-04  Richard Sandiford  <richard.sandiford@linaro.org>
11816
11817         PR target/83680
11818         * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
11819         test for d.testing.
11820
11821 2018-01-04  Peter Bergner  <bergner@vnet.ibm.com>
11822
11823         PR target/83387
11824         * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
11825         allow arguments in FP registers if TARGET_HARD_FLOAT is false.
11826
11827 2018-01-04  Jakub Jelinek  <jakub@redhat.com>
11828
11829         PR debug/83666
11830         * cfgexpand.c (expand_debug_expr) <case BIT_FIELD_REF>: Punt if mode
11831         is BLKmode and bitpos not zero or mode change is needed.
11832
11833 2018-01-04  Richard Sandiford  <richard.sandiford@linaro.org>
11834
11835         PR target/83675
11836         * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
11837         TARGET_VIS2.
11838
11839 2018-01-04  Uros Bizjak  <ubizjak@gmail.com>
11840
11841         PR target/83628
11842         * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
11843         instead of MULT rtx.  Update all corresponding splitters.
11844         (*saddl_se): Ditto.
11845         (*ssub<modesuffix>): Ditto.
11846         (*ssubl_se): Ditto.
11847         (*cmp_sadd_di): Update split patterns.
11848         (*cmp_sadd_si): Ditto.
11849         (*cmp_sadd_sidi): Ditto.
11850         (*cmp_ssub_di): Ditto.
11851         (*cmp_ssub_si): Ditto.
11852         (*cmp_ssub_sidi): Ditto.
11853         * config/alpha/predicates.md (const23_operand): New predicate.
11854         * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
11855         Look for ASHIFT, not MULT inner operand.
11856         (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
11857
11858 2018-01-04  Martin Liska  <mliska@suse.cz>
11859
11860         PR gcov-profile/83669
11861         * gcov.c (output_intermediate_file): Add version to intermediate
11862         gcov file.
11863         * doc/gcov.texi: Document new field 'version' in intermediate
11864         file format. Fix location of '-k' option of gcov command.
11865
11866 2018-01-04  Martin Liska  <mliska@suse.cz>
11867
11868         PR ipa/82352
11869         * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
11870
11871 2018-01-04  Jakub Jelinek  <jakub@redhat.com>
11872
11873         * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
11874
11875 2018-01-03  Martin Sebor  <msebor@redhat.com>
11876
11877         PR tree-optimization/83655
11878         * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
11879         checking calls with invalid arguments.
11880
11881 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11882
11883         * tree-vect-stmts.c (vect_get_store_rhs): New function.
11884         (vectorizable_mask_load_store): Delete.
11885         (vectorizable_call): Return false for masked loads and stores.
11886         (vectorizable_store): Handle IFN_MASK_STORE.  Use vect_get_store_rhs
11887         instead of gimple_assign_rhs1.
11888         (vectorizable_load): Handle IFN_MASK_LOAD.
11889         (vect_transform_stmt): Don't set is_store for call_vec_info_type.
11890
11891 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11892
11893         * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
11894         split out from..,
11895         (vectorizable_mask_load_store): ...here.
11896         (vectorizable_load): ...and here.
11897
11898 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11899
11900         * tree-vect-stmts.c (vect_build_all_ones_mask)
11901         (vect_build_zero_merge_argument): New functions, split out from...
11902         (vectorizable_load): ...here.
11903
11904 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11905
11906         * tree-vect-stmts.c (vect_check_store_rhs): New function,
11907         split out from...
11908         (vectorizable_mask_load_store): ...here.
11909         (vectorizable_store): ...and here.
11910
11911 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11912
11913         * tree-vect-stmts.c (vect_check_load_store_mask): New function,
11914         split out from...
11915         (vectorizable_mask_load_store): ...here.
11916
11917 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11918
11919         * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
11920         (vect_model_store_cost): Take a vec_load_store_type instead of a
11921         vect_def_type.
11922         * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
11923         (vect_model_store_cost): Take a vec_load_store_type instead of a
11924         vect_def_type.
11925         (vectorizable_mask_load_store): Update accordingly.
11926         (vectorizable_store): Likewise.
11927         * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
11928
11929 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11930
11931         * tree-vect-loop.c (vect_transform_loop): Stub out scalar
11932         IFN_MASK_LOAD calls here rather than...
11933         * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
11934
11935 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11936             Alan Hayward  <alan.hayward@arm.com>
11937             David Sherwood  <david.sherwood@arm.com>
11938
11939         * expmed.c (extract_bit_field_1): For vector extracts,
11940         fall back to extract_bit_field_as_subreg if vec_extract
11941         isn't available.
11942
11943 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11944             Alan Hayward  <alan.hayward@arm.com>
11945             David Sherwood  <david.sherwood@arm.com>
11946
11947         * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
11948         they are variable or constant sized.
11949         (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
11950         slots for constant-sized data.
11951
11952 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11953             Alan Hayward  <alan.hayward@arm.com>
11954             David Sherwood  <david.sherwood@arm.com>
11955
11956         * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
11957         handling COND_EXPRs with boolean comparisons, try to find a better
11958         basis for the mask type than the boolean itself.
11959
11960 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11961
11962         * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
11963         is calculated and how it can be overridden.
11964         * genmodes.c (max_bitsize_mode_any_mode): New variable.
11965         (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
11966         if defined.
11967         (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
11968         if nonzero.
11969
11970 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11971             Alan Hayward  <alan.hayward@arm.com>
11972             David Sherwood  <david.sherwood@arm.com>
11973
11974         * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
11975         Remove the mode argument.
11976         (aarch64_simd_valid_immediate): Remove the mode and inverse
11977         arguments.
11978         * config/aarch64/iterators.md (bitsize): New iterator.
11979         * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
11980         (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
11981         * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
11982         aarch64_simd_valid_immediate.
11983         * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
11984         (aarch64_reg_or_bic_imm): Likewise.
11985         * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
11986         with an insn_type enum and msl with a modifier_type enum.
11987         Replace element_width with a scalar_mode.  Change the shift
11988         to unsigned int.  Add constructors for scalar_float_mode and
11989         scalar_int_mode elements.
11990         (aarch64_vect_float_const_representable_p): Delete.
11991         (aarch64_can_const_movi_rtx_p)
11992         (aarch64_simd_scalar_immediate_valid_for_move)
11993         (aarch64_simd_make_constant): Update call to
11994         aarch64_simd_valid_immediate.
11995         (aarch64_advsimd_valid_immediate_hs): New function.
11996         (aarch64_advsimd_valid_immediate): Likewise.
11997         (aarch64_simd_valid_immediate): Remove mode and inverse
11998         arguments.  Rewrite to use the above.  Use const_vec_duplicate_p
11999         to detect duplicated constants and use aarch64_float_const_zero_rtx_p
12000         and aarch64_float_const_representable_p on the result.
12001         (aarch64_output_simd_mov_immediate): Remove mode argument.
12002         Update call to aarch64_simd_valid_immediate and use of
12003         simd_immediate_info.
12004         (aarch64_output_scalar_simd_mov_immediate): Update call
12005         accordingly.
12006
12007 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12008             Alan Hayward  <alan.hayward@arm.com>
12009             David Sherwood  <david.sherwood@arm.com>
12010
12011         * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
12012         (mode_nunits): Likewise CONST_MODE_NUNITS.
12013         * machmode.def (ADJUST_NUNITS): Document.
12014         * genmodes.c (mode_data::need_nunits_adj): New field.
12015         (blank_mode): Update accordingly.
12016         (adj_nunits): New variable.
12017         (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
12018         parameter.
12019         (emit_mode_size_inline): Set need_bytesize_adj for all modes
12020         listed in adj_nunits.
12021         (emit_mode_nunits_inline): Set need_nunits_adj for all modes
12022         listed in adj_nunits.  Don't emit case statements for such modes.
12023         (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
12024         and CONST_MODE_PRECISION.  Make CONST_MODE_SIZE expand to
12025         nothing if adj_nunits is nonnull.
12026         (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
12027         (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
12028         (emit_mode_fbit): Update use of print_maybe_const_decl.
12029         (emit_move_size): Likewise.  Treat the array as non-const
12030         if adj_nunits.
12031         (emit_mode_adjustments): Handle adj_nunits.
12032
12033 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12034
12035         * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
12036         * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
12037         (VECTOR_MODES): Use it.
12038         (make_vector_modes): Take the prefix as an argument.
12039
12040 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12041             Alan Hayward  <alan.hayward@arm.com>
12042             David Sherwood  <david.sherwood@arm.com>
12043
12044         * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
12045         * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
12046         for MODE_VECTOR_BOOL.
12047         * machmode.def (VECTOR_BOOL_MODE): Document.
12048         * genmodes.c (VECTOR_BOOL_MODE): New macro.
12049         (make_vector_bool_mode): New function.
12050         (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
12051         MODE_VECTOR_BOOL.
12052         * lto-streamer-in.c (lto_input_mode_table): Likewise.
12053         * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
12054         Likewise.
12055         * stor-layout.c (int_mode_for_mode): Likewise.
12056         * tree.c (build_vector_type_for_mode): Likewise.
12057         * varasm.c (output_constant_pool_2): Likewise.
12058         * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
12059         CONSTM1_RTX (BImode) are the same thing.  Initialize const_tiny_rtx
12060         for MODE_VECTOR_BOOL.
12061         * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
12062         of mode class checks.
12063         * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
12064         instead of a list of mode class checks.
12065         (expand_vector_scalar_condition): Likewise.
12066         (type_for_widest_vector_mode): Handle BImode as an inner mode.
12067
12068 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12069             Alan Hayward  <alan.hayward@arm.com>
12070             David Sherwood  <david.sherwood@arm.com>
12071
12072         * machmode.h (mode_size): Change from unsigned short to
12073         poly_uint16_pod.
12074         (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
12075         (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
12076         or if measurement_type is not polynomial.
12077         (fixed_size_mode::includes_p): Check for constant-sized modes.
12078         * genmodes.c (emit_mode_size_inline): Make mode_size_inline
12079         return a poly_uint16 rather than an unsigned short.
12080         (emit_mode_size): Change the type of mode_size from unsigned short
12081         to poly_uint16_pod.  Use ZERO_COEFFS for the initializer.
12082         (emit_mode_adjustments): Cope with polynomial vector sizes.
12083         * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
12084         for GET_MODE_SIZE.
12085         * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
12086         for GET_MODE_SIZE.
12087         * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
12088         * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
12089         * caller-save.c (setup_save_areas): Likewise.
12090         (replace_reg_with_saved_mem): Likewise.
12091         * calls.c (emit_library_call_value_1): Likewise.
12092         * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
12093         * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
12094         (gen_lowpart_for_combine): Likewise.
12095         * convert.c (convert_to_integer_1): Likewise.
12096         * cse.c (equiv_constant, cse_insn): Likewise.
12097         * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
12098         (cselib_subst_to_values): Likewise.
12099         * dce.c (word_dce_process_block): Likewise.
12100         * df-problems.c (df_word_lr_mark_ref): Likewise.
12101         * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
12102         * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
12103         (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
12104         (rtl_for_decl_location): Likewise.
12105         * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
12106         * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
12107         * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
12108         (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
12109         (expand_expr_real_1): Likewise.
12110         * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
12111         (pad_below): Likewise.
12112         * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
12113         * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
12114         * ira.c (get_subreg_tracking_sizes): Likewise.
12115         * ira-build.c (ira_create_allocno_objects): Likewise.
12116         * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
12117         (ira_sort_regnos_for_alter_reg): Likewise.
12118         * ira-costs.c (record_operand_costs): Likewise.
12119         * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
12120         (resolve_simple_move): Likewise.
12121         * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
12122         (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
12123         (lra_constraints): Likewise.
12124         (CONST_POOL_OK_P): Reject variable-sized modes.
12125         * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
12126         (add_pseudo_to_slot, lra_spill): Likewise.
12127         * omp-low.c (omp_clause_aligned_alignment): Likewise.
12128         * optabs-query.c (get_best_extraction_insn): Likewise.
12129         * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
12130         * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
12131         (expand_mult_highpart, valid_multiword_target_p): Likewise.
12132         * recog.c (offsettable_address_addr_space_p): Likewise.
12133         * regcprop.c (maybe_mode_change): Likewise.
12134         * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
12135         * regrename.c (build_def_use): Likewise.
12136         * regstat.c (dump_reg_info): Likewise.
12137         * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
12138         (find_reloads, find_reloads_subreg_address): Likewise.
12139         * reload1.c (eliminate_regs_1): Likewise.
12140         * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
12141         * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
12142         (simplify_binary_operation_1, simplify_subreg): Likewise.
12143         * targhooks.c (default_function_arg_padding): Likewise.
12144         (default_hard_regno_nregs, default_class_max_nregs): Likewise.
12145         * tree-cfg.c (verify_gimple_assign_binary): Likewise.
12146         (verify_gimple_assign_ternary): Likewise.
12147         * tree-inline.c (estimate_move_cost): Likewise.
12148         * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
12149         * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
12150         (get_address_cost_ainc): Likewise.
12151         * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
12152         (vect_supportable_dr_alignment): Likewise.
12153         * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
12154         (vectorizable_reduction): Likewise.
12155         * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
12156         (vectorizable_operation, vectorizable_load): Likewise.
12157         * tree.c (build_same_sized_truth_vector_type): Likewise.
12158         * valtrack.c (cleanup_auto_inc_dec): Likewise.
12159         * var-tracking.c (emit_note_insn_var_location): Likewise.
12160         * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
12161         (ADDR_VEC_ALIGN): Likewise.
12162
12163 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12164             Alan Hayward  <alan.hayward@arm.com>
12165             David Sherwood  <david.sherwood@arm.com>
12166
12167         * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
12168         unsigned short.
12169         (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
12170         or if measurement_type is polynomial.
12171         * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
12172         * combine.c (make_extraction): Likewise.
12173         * dse.c (find_shift_sequence): Likewise.
12174         * dwarf2out.c (mem_loc_descriptor): Likewise.
12175         * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
12176         (extract_bit_field, extract_low_bits): Likewise.
12177         * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
12178         (optimize_bitfield_assignment_op, expand_assignment): Likewise.
12179         (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
12180         * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
12181         * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
12182         * reload.c (find_reloads): Likewise.
12183         * reload1.c (alter_reg): Likewise.
12184         * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
12185         * targhooks.c (default_secondary_memory_needed_mode): Likewise.
12186         * tree-if-conv.c (predicate_mem_writes): Likewise.
12187         * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
12188         * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
12189         * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
12190         * valtrack.c (dead_debug_insert_temp): Likewise.
12191         * varasm.c (mergeable_constant_section): Likewise.
12192         * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
12193
12194 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12195             Alan Hayward  <alan.hayward@arm.com>
12196             David Sherwood  <david.sherwood@arm.com>
12197
12198         * expr.c (expand_assignment): Cope with polynomial mode sizes
12199         when assigning to a CONCAT.
12200
12201 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12202             Alan Hayward  <alan.hayward@arm.com>
12203             David Sherwood  <david.sherwood@arm.com>
12204
12205         * machmode.h (mode_precision): Change from unsigned short to
12206         poly_uint16_pod.
12207         (mode_to_precision): Return a poly_uint16 rather than an unsigned
12208         short.
12209         (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
12210         or if measurement_type is not polynomial.
12211         (HWI_COMPUTABLE_MODE_P): Turn into a function.  Optimize the case
12212         in which the mode is already known to be a scalar_int_mode.
12213         * genmodes.c (emit_mode_precision): Change the type of mode_precision
12214         from unsigned short to poly_uint16_pod.  Use ZERO_COEFFS for the
12215         initializer.
12216         * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
12217         for GET_MODE_PRECISION.
12218         * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
12219         for GET_MODE_PRECISION.
12220         * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
12221         as polynomial.
12222         (try_combine, find_split_point, combine_simplify_rtx): Likewise.
12223         (expand_field_assignment, make_extraction): Likewise.
12224         (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
12225         (get_last_value): Likewise.
12226         * convert.c (convert_to_integer_1): Likewise.
12227         * cse.c (cse_insn): Likewise.
12228         * expr.c (expand_expr_real_1): Likewise.
12229         * lra-constraints.c (simplify_operand_subreg): Likewise.
12230         * optabs-query.c (can_atomic_load_p): Likewise.
12231         * optabs.c (expand_atomic_load): Likewise.
12232         (expand_atomic_store): Likewise.
12233         * ree.c (combine_reaching_defs): Likewise.
12234         * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
12235         * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
12236         * tree.h (type_has_mode_precision_p): Likewise.
12237         * ubsan.c (instrument_si_overflow): Likewise.
12238
12239 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12240             Alan Hayward  <alan.hayward@arm.com>
12241             David Sherwood  <david.sherwood@arm.com>
12242
12243         * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
12244         polynomial numbers of units.
12245         (SET_TYPE_VECTOR_SUBPARTS): Likewise.
12246         (valid_vector_subparts_p): New function.
12247         (build_vector_type): Remove temporary shim and take the number
12248         of units as a poly_uint64 rather than an int.
12249         (build_opaque_vector_type): Take the number of units as a
12250         poly_uint64 rather than an int.
12251         * tree.c (build_vector_from_ctor): Handle polynomial
12252         TYPE_VECTOR_SUBPARTS.
12253         (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
12254         (uniform_vector_p, vector_type_mode, build_vector): Likewise.
12255         (build_vector_from_val): If the number of units is variable,
12256         use build_vec_duplicate_cst for constant operands and
12257         VEC_DUPLICATE_EXPR otherwise.
12258         (make_vector_type): Remove temporary is_constant ().
12259         (build_vector_type, build_opaque_vector_type): Take the number of
12260         units as a poly_uint64 rather than an int.
12261         (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
12262         VECTOR_CST_NELTS.
12263         * cfgexpand.c (expand_debug_expr): Likewise.
12264         * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
12265         (store_constructor, expand_expr_real_1): Likewise.
12266         (const_scalar_mask_from_tree): Likewise.
12267         * fold-const-call.c (fold_const_reduction): Likewise.
12268         * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
12269         (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
12270         (native_encode_vector, vec_cst_ctor_to_array): Likewise.
12271         (fold_relational_const): Likewise.
12272         (native_interpret_vector): Likewise.  Change the size from an
12273         int to an unsigned int.
12274         * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
12275         TYPE_VECTOR_SUBPARTS.
12276         (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
12277         (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
12278         duplicating a non-constant operand into a variable-length vector.
12279         * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
12280         TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
12281         * ipa-icf.c (sem_variable::equals): Likewise.
12282         * match.pd: Likewise.
12283         * omp-simd-clone.c (simd_clone_subparts): Likewise.
12284         * print-tree.c (print_node): Likewise.
12285         * stor-layout.c (layout_type): Likewise.
12286         * targhooks.c (default_builtin_vectorization_cost): Likewise.
12287         * tree-cfg.c (verify_gimple_comparison): Likewise.
12288         (verify_gimple_assign_binary): Likewise.
12289         (verify_gimple_assign_ternary): Likewise.
12290         (verify_gimple_assign_single): Likewise.
12291         * tree-pretty-print.c (dump_generic_node): Likewise.
12292         * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
12293         (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
12294         * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
12295         (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
12296         (vect_shift_permute_load_chain): Likewise.
12297         * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
12298         (expand_vector_condition, optimize_vector_constructor): Likewise.
12299         (lower_vec_perm, get_compute_type): Likewise.
12300         * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
12301         (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
12302         * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
12303         (vect_recog_mask_conversion_pattern): Likewise.
12304         * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
12305         (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
12306         * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
12307         (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
12308         (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
12309         (vectorizable_shift, vectorizable_operation, vectorizable_store)
12310         (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
12311         (supportable_widening_operation): Likewise.
12312         (supportable_narrowing_operation): Likewise.
12313         * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
12314         Likewise.
12315         * varasm.c (output_constant): Likewise.
12316
12317 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12318             Alan Hayward  <alan.hayward@arm.com>
12319             David Sherwood  <david.sherwood@arm.com>
12320
12321         * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
12322         so that both the length == 3 and length != 3 cases set up their
12323         own permute vectors.  Add comments explaining why we know the
12324         number of elements is constant.
12325         (vect_permute_load_chain): Likewise.
12326
12327 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12328             Alan Hayward  <alan.hayward@arm.com>
12329             David Sherwood  <david.sherwood@arm.com>
12330
12331         * machmode.h (mode_nunits): Change from unsigned char to
12332         poly_uint16_pod.
12333         (ONLY_FIXED_SIZE_MODES): New macro.
12334         (pod_mode::measurement_type, scalar_int_mode::measurement_type)
12335         (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
12336         (complex_mode::measurement_type, fixed_size_mode::measurement_type):
12337         New typedefs.
12338         (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
12339         (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
12340         or if measurement_type is not polynomial.
12341         * genmodes.c (ZERO_COEFFS): New macro.
12342         (emit_mode_nunits_inline): Make mode_nunits_inline return a
12343         poly_uint16.
12344         (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
12345         Use ZERO_COEFFS when emitting initializers.
12346         * data-streamer.h (bp_pack_poly_value): New function.
12347         (bp_unpack_poly_value): Likewise.
12348         * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
12349         for GET_MODE_NUNITS.
12350         * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
12351         for GET_MODE_NUNITS.
12352         * tree.c (make_vector_type): Remove temporary shim and make
12353         the real function take the number of units as a poly_uint64
12354         rather than an int.
12355         (build_vector_type_for_mode): Handle polynomial nunits.
12356         * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
12357         * emit-rtl.c (const_vec_series_p_1): Likewise.
12358         (gen_rtx_CONST_VECTOR): Likewise.
12359         * fold-const.c (test_vec_duplicate_folding): Likewise.
12360         * genrecog.c (validate_pattern): Likewise.
12361         * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
12362         * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
12363         * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
12364         (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
12365         (expand_vec_cond_expr, expand_mult_highpart): Likewise.
12366         * rtlanal.c (subreg_get_info): Likewise.
12367         * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
12368         (vect_grouped_load_supported): Likewise.
12369         * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
12370         * tree-vect-loop.c (have_whole_vector_shift): Likewise.
12371         * simplify-rtx.c (simplify_unary_operation_1): Likewise.
12372         (simplify_const_unary_operation, simplify_binary_operation_1)
12373         (simplify_const_binary_operation, simplify_ternary_operation)
12374         (test_vector_ops_duplicate, test_vector_ops): Likewise.
12375         (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
12376         instead of CONST_VECTOR_NUNITS.
12377         * varasm.c (output_constant_pool_2): Likewise.
12378         * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
12379         explicit-encoded elements in the XVEC for variable-length vectors.
12380
12381 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12382
12383         * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
12384
12385 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12386             Alan Hayward  <alan.hayward@arm.com>
12387             David Sherwood  <david.sherwood@arm.com>
12388
12389         * coretypes.h (fixed_size_mode): Declare.
12390         (fixed_size_mode_pod): New typedef.
12391         * builtins.h (target_builtins::x_apply_args_mode)
12392         (target_builtins::x_apply_result_mode): Change type to
12393         fixed_size_mode_pod.
12394         * builtins.c (apply_args_size, apply_result_size, result_vector)
12395         (expand_builtin_apply_args_1, expand_builtin_apply)
12396         (expand_builtin_return): Update accordingly.
12397
12398 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12399
12400         * cse.c (hash_rtx_cb): Hash only the encoded elements.
12401         * cselib.c (cselib_hash_rtx): Likewise.
12402         * expmed.c (make_tree): Build VECTOR_CSTs directly from the
12403         CONST_VECTOR encoding.
12404
12405 2018-01-03  Jakub Jelinek  <jakub@redhat.com>
12406             Jeff Law  <law@redhat.com>
12407
12408         PR target/83641
12409         * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
12410         noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
12411         only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
12412         and add REG_CFA_ADJUST_CFA notes in that case to both insns.
12413
12414         PR target/83641
12415         * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
12416         explicitly probe *sp in a noreturn function if there were any callee
12417         register saves or frame pointer is needed.
12418
12419 2018-01-03  Jakub Jelinek  <jakub@redhat.com>
12420
12421         PR debug/83621
12422         * cfgexpand.c (expand_debug_expr): Return NULL if mode is
12423         BLKmode for ternary, binary or unary expressions.
12424
12425         PR debug/83645
12426         * var-tracking.c (delete_vta_debug_insn): New inline function.
12427         (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
12428         insns from get_insns () to NULL instead of each bb separately.
12429         Use delete_vta_debug_insn.  No longer static.
12430         (vt_debug_insns_local, variable_tracking_main_1): Adjust
12431         delete_vta_debug_insns callers.
12432         * rtl.h (delete_vta_debug_insns): Declare.
12433         * final.c (rest_of_handle_final): Call delete_vta_debug_insns
12434         instead of variable_tracking_main.
12435
12436 2018-01-03  Martin Sebor  <msebor@redhat.com>
12437
12438         PR tree-optimization/83603
12439         * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
12440         arguments past the endof the argument list in functions declared
12441         without a prototype.
12442         * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
12443         Avoid checking when arguments are null.
12444
12445 2018-01-03  Martin Sebor  <msebor@redhat.com>
12446
12447         PR c/83559
12448         * doc/extend.texi (attribute const): Fix a typo.
12449         * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
12450         issuing -Wsuggest-attribute for void functions.
12451
12452 2018-01-03  Martin Sebor  <msebor@redhat.com>
12453
12454         * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
12455         offset_int::from instead of wide_int::to_shwi.
12456         (maybe_diag_overlap): Remove assertion.
12457         Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
12458         * gimple-ssa-sprintf.c (format_directive): Same.
12459         (parse_directive): Same.
12460         (sprintf_dom_walker::compute_format_length): Same.
12461         (try_substitute_return_value): Same.
12462
12463 2018-01-03  Jeff Law  <law@redhat.com>
12464
12465         PR middle-end/83654
12466         * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
12467         non-constant residual for zero at runtime and avoid probing in
12468         that case.  Reorganize code for trailing problem to mirror handling
12469         of the residual.
12470
12471 2018-01-03  Prathamesh Kulkarni  <prathamesh.kulkarni@linaro.org>
12472
12473         PR tree-optimization/83501
12474         * tree-ssa-strlen.c (get_string_cst): New.
12475         (handle_char_store): Call get_string_cst.
12476
12477 2018-01-03  Martin Liska  <mliska@suse.cz>
12478
12479         PR tree-optimization/83593
12480         * tree-ssa-strlen.c: Include tree-cfg.h.
12481         (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
12482         (strlen_dom_walker): Add new member variable m_cleanup_cfg.
12483         (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
12484         to false.
12485         (strlen_dom_walker::before_dom_children): Call
12486         gimple_purge_dead_eh_edges. Dump tranformation with details
12487         dump flags.
12488         (strlen_dom_walker::before_dom_children): Update call by adding
12489         new argument cleanup_eh.
12490         (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
12491
12492 2018-01-03  Martin Liska  <mliska@suse.cz>
12493
12494         PR ipa/83549
12495         * cif-code.def (VARIADIC_THUNK): New enum value.
12496         * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
12497         thunks.
12498
12499 2018-01-03  Jan Beulich  <jbeulich@suse.com>
12500
12501         * sse.md (mov<mode>_internal): Tighten condition for when to use
12502         vmovdqu<ssescalarsize> for TI and OI modes.
12503
12504 2018-01-03  Jakub Jelinek  <jakub@redhat.com>
12505
12506         Update copyright years.
12507
12508 2018-01-03  Martin Liska  <mliska@suse.cz>
12509
12510         PR ipa/83594
12511         * ipa-visibility.c (function_and_variable_visibility): Skip
12512         functions with noipa attribure.
12513
12514 2018-01-03  Jakub Jelinek  <jakub@redhat.com>
12515
12516         * gcc.c (process_command): Update copyright notice dates.
12517         * gcov-dump.c (print_version): Ditto.
12518         * gcov.c (print_version): Ditto.
12519         * gcov-tool.c (print_version): Ditto.
12520         * gengtype.c (create_file): Ditto.
12521         * doc/cpp.texi: Bump @copying's copyright year.
12522         * doc/cppinternals.texi: Ditto.
12523         * doc/gcc.texi: Ditto.
12524         * doc/gccint.texi: Ditto.
12525         * doc/gcov.texi: Ditto.
12526         * doc/install.texi: Ditto.
12527         * doc/invoke.texi: Ditto.
12528
12529 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12530
12531         * vector-builder.h (vector_builder::m_full_nelts): Change from
12532         unsigned int to poly_uint64.
12533         (vector_builder::full_nelts): Update prototype accordingly.
12534         (vector_builder::new_vector): Likewise.
12535         (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
12536         (vector_builder::operator ==): Likewise.
12537         (vector_builder::finalize): Likewise.
12538         * int-vector-builder.h (int_vector_builder::int_vector_builder):
12539         Take the number of elements as a poly_uint64 rather than an
12540         unsigned int.
12541         * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
12542         from unsigned int to poly_uint64.
12543         (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
12544         (vec_perm_indices::new_vector): Likewise.
12545         (vec_perm_indices::length): Likewise.
12546         (vec_perm_indices::nelts_per_input): Likewise.
12547         (vec_perm_indices::input_nelts): Likewise.
12548         * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
12549         number of elements per input as a poly_uint64 rather than an
12550         unsigned int.  Use the original encoding for variable-length
12551         vectors, rather than clamping each individual element.
12552         For the second and subsequent elements in each pattern,
12553         clamp the step and base before clamping their sum.
12554         (vec_perm_indices::series_p): Handle polynomial element counts.
12555         (vec_perm_indices::all_in_range_p): Likewise.
12556         (vec_perm_indices_to_tree): Likewise.
12557         (vec_perm_indices_to_rtx): Likewise.
12558         * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
12559         * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
12560         (tree_vector_builder::new_binary_operation): Handle polynomial
12561         element counts.  Return false if we need to know the number
12562         of elements at compile time.
12563         * fold-const.c (fold_vec_perm): Punt if the number of elements
12564         isn't known at compile time.
12565
12566 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12567
12568         * vec-perm-indices.h (vec_perm_builder): Change element type
12569         from HOST_WIDE_INT to poly_int64.
12570         (vec_perm_indices::element_type): Update accordingly.
12571         (vec_perm_indices::clamp): Handle polynomial element_types.
12572         * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
12573         (vec_perm_indices::all_in_range_p): Likewise.
12574         (tree_to_vec_perm_builder): Check for poly_int64 trees rather
12575         than shwi trees.
12576         * vector-builder.h (vector_builder::stepped_sequence_p): Handle
12577         polynomial vec_perm_indices element types.
12578         * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
12579         * fold-const.c (fold_vec_perm): Likewise.
12580         * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
12581         * tree-vect-generic.c (lower_vec_perm): Likewise.
12582         * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
12583         * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
12584         element type to HOST_WIDE_INT.
12585
12586 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12587             Alan Hayward  <alan.hayward@arm.com>
12588             David Sherwood  <david.sherwood@arm.com>
12589
12590         * alias.c (addr_side_effect_eval): Take the size as a poly_int64
12591         rather than an int.  Use plus_constant.
12592         (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
12593         Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
12594
12595 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12596             Alan Hayward  <alan.hayward@arm.com>
12597             David Sherwood  <david.sherwood@arm.com>
12598
12599         * calls.c (emit_call_1, expand_call): Change struct_value_size from
12600         a HOST_WIDE_INT to a poly_int64.
12601
12602 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12603             Alan Hayward  <alan.hayward@arm.com>
12604             David Sherwood  <david.sherwood@arm.com>
12605
12606         * calls.c (load_register_parameters): Cope with polynomial
12607         mode sizes.  Require a constant size for BLKmode parameters
12608         that aren't described by a PARALLEL.  If BLOCK_REG_PADDING
12609         forces a parameter to be padded at the lsb end in order to
12610         fill a complete number of words, require the parameter size
12611         to be ordered wrt UNITS_PER_WORD.
12612
12613 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12614             Alan Hayward  <alan.hayward@arm.com>
12615             David Sherwood  <david.sherwood@arm.com>
12616
12617         * reload1.c (spill_stack_slot_width): Change element type
12618         from unsigned int to poly_uint64_pod.
12619         (alter_reg): Treat mode sizes as polynomial.
12620
12621 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12622             Alan Hayward  <alan.hayward@arm.com>
12623             David Sherwood  <david.sherwood@arm.com>
12624
12625         * reload.c (complex_word_subreg_p): New function.
12626         (reload_inner_reg_of_subreg, push_reload): Use it.
12627
12628 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12629             Alan Hayward  <alan.hayward@arm.com>
12630             David Sherwood  <david.sherwood@arm.com>
12631
12632         * lra-constraints.c (process_alt_operands): Reject matched
12633         operands whose sizes aren't ordered.
12634         (match_reload): Refer to this check here.
12635
12636 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12637             Alan Hayward  <alan.hayward@arm.com>
12638             David Sherwood  <david.sherwood@arm.com>
12639
12640         * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
12641         that the mode size is in the set {1, 2, 4, 8, 16}.
12642
12643 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12644             Alan Hayward  <alan.hayward@arm.com>
12645             David Sherwood  <david.sherwood@arm.com>
12646
12647         * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
12648         Use plus_constant instead of gen_rtx_PLUS.
12649
12650 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12651             Alan Hayward  <alan.hayward@arm.com>
12652             David Sherwood  <david.sherwood@arm.com>
12653
12654         * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
12655         * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
12656         * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
12657         * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
12658         * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
12659         * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
12660         * config/i386/i386-protos.h (ix86_push_rounding): Declare.
12661         * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
12662         * config/i386/i386.c (ix86_push_rounding): ...this new function.
12663         * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
12664         a poly_int64.
12665         * config/m32c/m32c.c (m32c_push_rounding): Likewise.
12666         * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
12667         * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
12668         * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
12669         * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
12670         * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
12671         * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
12672         * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
12673         * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
12674         * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
12675         function.
12676         * expr.c (emit_move_resolve_push): Treat the input and result
12677         of PUSH_ROUNDING as a poly_int64.
12678         (emit_move_complex_push, emit_single_push_insn_1): Likewise.
12679         (emit_push_insn): Likewise.
12680         * lra-eliminations.c (mark_not_eliminable): Likewise.
12681         * recog.c (push_operand): Likewise.
12682         * reload1.c (elimination_effects): Likewise.
12683         * rtlanal.c (nonzero_bits1): Likewise.
12684         * calls.c (store_one_arg): Likewise.  Require the padding to be
12685         known at compile time.
12686
12687 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12688             Alan Hayward  <alan.hayward@arm.com>
12689             David Sherwood  <david.sherwood@arm.com>
12690
12691         * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
12692         Use plus_constant instead of gen_rtx_PLUS.
12693
12694 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12695             Alan Hayward  <alan.hayward@arm.com>
12696             David Sherwood  <david.sherwood@arm.com>
12697
12698         * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
12699         rather than an int.
12700
12701 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12702             Alan Hayward  <alan.hayward@arm.com>
12703             David Sherwood  <david.sherwood@arm.com>
12704
12705         * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
12706         instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
12707         via stack temporaries.  Treat the mode size as polynomial too.
12708
12709 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12710             Alan Hayward  <alan.hayward@arm.com>
12711             David Sherwood  <david.sherwood@arm.com>
12712
12713         * expr.c (expand_expr_real_2): When handling conversions involving
12714         unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
12715         multiplying int_size_in_bytes by BITS_PER_UNIT.  Treat GET_MODE_BISIZE
12716         as a poly_uint64 too.
12717
12718 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12719             Alan Hayward  <alan.hayward@arm.com>
12720             David Sherwood  <david.sherwood@arm.com>
12721
12722         * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
12723
12724 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12725             Alan Hayward  <alan.hayward@arm.com>
12726             David Sherwood  <david.sherwood@arm.com>
12727
12728         * combine.c (can_change_dest_mode): Handle polynomial
12729         REGMODE_NATURAL_SIZE.
12730         * expmed.c (store_bit_field_1): Likewise.
12731         * expr.c (store_constructor): Likewise.
12732         * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
12733         and polynomial REGMODE_NATURAL_SIZE.
12734         (gen_lowpart_common): Likewise.
12735         * reginfo.c (record_subregs_of_mode): Likewise.
12736         * rtlanal.c (read_modify_subreg_p): Likewise.
12737
12738 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12739             Alan Hayward  <alan.hayward@arm.com>
12740             David Sherwood  <david.sherwood@arm.com>
12741
12742         * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
12743         numbers of elements.
12744
12745 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12746             Alan Hayward  <alan.hayward@arm.com>
12747             David Sherwood  <david.sherwood@arm.com>
12748
12749         * match.pd: Cope with polynomial numbers of vector elements.
12750
12751 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12752             Alan Hayward  <alan.hayward@arm.com>
12753             David Sherwood  <david.sherwood@arm.com>
12754
12755         * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
12756         in a POINTER_PLUS_EXPR.
12757
12758 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12759             Alan Hayward  <alan.hayward@arm.com>
12760             David Sherwood  <david.sherwood@arm.com>
12761
12762         * omp-simd-clone.c (simd_clone_subparts): New function.
12763         (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
12764         (ipa_simd_modify_function_body): Likewise.
12765
12766 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12767             Alan Hayward  <alan.hayward@arm.com>
12768             David Sherwood  <david.sherwood@arm.com>
12769
12770         * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
12771         (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
12772         (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
12773         (expand_vector_condition, vector_element): Likewise.
12774         (subparts_gt): New function.
12775         (get_compute_type): Use subparts_gt.
12776         (count_type_subparts): Delete.
12777         (expand_vector_operations_1): Use subparts_gt instead of
12778         count_type_subparts.
12779
12780 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12781             Alan Hayward  <alan.hayward@arm.com>
12782             David Sherwood  <david.sherwood@arm.com>
12783
12784         * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
12785         (vect_compile_time_alias): ...this new function.  Do the calculation
12786         on poly_ints rather than trees.
12787         (vect_prune_runtime_alias_test_list): Update call accordingly.
12788
12789 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12790             Alan Hayward  <alan.hayward@arm.com>
12791             David Sherwood  <david.sherwood@arm.com>
12792
12793         * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
12794         numbers of units.
12795         (vect_schedule_slp_instance): Likewise.
12796
12797 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12798             Alan Hayward  <alan.hayward@arm.com>
12799             David Sherwood  <david.sherwood@arm.com>
12800
12801         * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
12802         constant and extern definitions for variable-length vectors.
12803         (vect_get_constant_vectors): Note that the number of units
12804         is known to be constant.
12805
12806 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12807             Alan Hayward  <alan.hayward@arm.com>
12808             David Sherwood  <david.sherwood@arm.com>
12809
12810         * tree-vect-stmts.c (vectorizable_conversion): Treat the number
12811         of units as polynomial.  Choose between WIDE and NARROW based
12812         on multiple_p.
12813
12814 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12815             Alan Hayward  <alan.hayward@arm.com>
12816             David Sherwood  <david.sherwood@arm.com>
12817
12818         * tree-vect-stmts.c (simd_clone_subparts): New function.
12819         (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
12820
12821 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12822             Alan Hayward  <alan.hayward@arm.com>
12823             David Sherwood  <david.sherwood@arm.com>
12824
12825         * tree-vect-stmts.c (vectorizable_call): Treat the number of
12826         vectors as polynomial.  Use build_index_vector for
12827         IFN_GOMP_SIMD_LANE.
12828
12829 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12830             Alan Hayward  <alan.hayward@arm.com>
12831             David Sherwood  <david.sherwood@arm.com>
12832
12833         * tree-vect-stmts.c (get_load_store_type): Treat the number of
12834         units as polynomial.  Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
12835         for variable-length vectors.
12836         (vectorizable_mask_load_store): Treat the number of units as
12837         polynomial, asserting that it is constant if the condition has
12838         already been enforced.
12839         (vectorizable_store, vectorizable_load): Likewise.
12840
12841 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12842             Alan Hayward  <alan.hayward@arm.com>
12843             David Sherwood  <david.sherwood@arm.com>
12844
12845         * tree-vect-loop.c (vectorizable_live_operation): Treat the number
12846         of units as polynomial.  Punt if we can't tell at compile time
12847         which vector contains the final result.
12848
12849 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12850             Alan Hayward  <alan.hayward@arm.com>
12851             David Sherwood  <david.sherwood@arm.com>
12852
12853         * tree-vect-loop.c (vectorizable_induction): Treat the number
12854         of units as polynomial.  Punt on SLP inductions.  Use an integer
12855         VEC_SERIES_EXPR for variable-length integer reductions.  Use a
12856         cast of such a series for variable-length floating-point
12857         reductions.
12858
12859 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12860             Alan Hayward  <alan.hayward@arm.com>
12861             David Sherwood  <david.sherwood@arm.com>
12862
12863         * tree.h (build_index_vector): Declare.
12864         * tree.c (build_index_vector): New function.
12865         * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
12866         of units as polynomial, forcibly converting it to a constant if
12867         vectorizable_reduction has already enforced the condition.
12868         (vect_create_epilog_for_reduction): Likewise.  Use build_index_vector
12869         to create a {1,2,3,...} vector.
12870         (vectorizable_reduction): Treat the number of units as polynomial.
12871         Choose vectype_in based on the largest scalar element size rather
12872         than the smallest number of units.  Enforce the restrictions
12873         relied on above.
12874
12875 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12876             Alan Hayward  <alan.hayward@arm.com>
12877             David Sherwood  <david.sherwood@arm.com>
12878
12879         * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
12880         number of units as polynomial.
12881
12882 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12883             Alan Hayward  <alan.hayward@arm.com>
12884             David Sherwood  <david.sherwood@arm.com>
12885
12886         * target.h (vector_sizes, auto_vector_sizes): New typedefs.
12887         * target.def (autovectorize_vector_sizes): Return the vector sizes
12888         by pointer, using vector_sizes rather than a bitmask.
12889         * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
12890         * targhooks.c (default_autovectorize_vector_sizes): Likewise.
12891         * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
12892         Likewise.
12893         * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
12894         * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
12895         * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
12896         * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
12897         * omp-general.c (omp_max_vf): Likewise.
12898         * omp-low.c (omp_clause_aligned_alignment): Likewise.
12899         * optabs-query.c (can_vec_mask_load_store_p): Likewise.
12900         * tree-vect-loop.c (vect_analyze_loop): Likewise.
12901         * tree-vect-slp.c (vect_slp_bb): Likewise.
12902         * doc/tm.texi: Regenerate.
12903         * tree-vectorizer.h (current_vector_size): Change from an unsigned int
12904         to a poly_uint64.
12905         * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
12906         the vector size as a poly_uint64 rather than an unsigned int.
12907         (current_vector_size): Change from an unsigned int to a poly_uint64.
12908         (get_vectype_for_scalar_type): Update accordingly.
12909         * tree.h (build_truth_vector_type): Take the size and number of
12910         units as a poly_uint64 rather than an unsigned int.
12911         (build_vector_type): Add a temporary overload that takes
12912         the number of units as a poly_uint64 rather than an unsigned int.
12913         * tree.c (make_vector_type): Likewise.
12914         (build_truth_vector_type): Take the number of units as a poly_uint64
12915         rather than an unsigned int.
12916
12917 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12918             Alan Hayward  <alan.hayward@arm.com>
12919             David Sherwood  <david.sherwood@arm.com>
12920
12921         * target.def (get_mask_mode): Take the number of units and length
12922         as poly_uint64s rather than unsigned ints.
12923         * targhooks.h (default_get_mask_mode): Update accordingly.
12924         * targhooks.c (default_get_mask_mode): Likewise.
12925         * config/i386/i386.c (ix86_get_mask_mode): Likewise.
12926         * doc/tm.texi: Regenerate.
12927
12928 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12929             Alan Hayward  <alan.hayward@arm.com>
12930             David Sherwood  <david.sherwood@arm.com>
12931
12932         * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
12933         * omp-general.c (omp_max_vf): Likewise.
12934         * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
12935         (expand_omp_simd): Handle polynomial safelen.
12936         * omp-low.c (omplow_simd_context): Add a default constructor.
12937         (omplow_simd_context::max_vf): Change from int to poly_uint64.
12938         (lower_rec_simd_input_clauses): Update accordingly.
12939         (lower_rec_input_clauses): Likewise.
12940
12941 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12942             Alan Hayward  <alan.hayward@arm.com>
12943             David Sherwood  <david.sherwood@arm.com>
12944
12945         * tree-vectorizer.h (vect_nunits_for_cost): New function.
12946         * tree-vect-loop.c (vect_model_reduction_cost): Use it.
12947         * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
12948         (vect_analyze_slp_cost): Likewise.
12949         * tree-vect-stmts.c (vect_model_store_cost): Likewise.
12950         (vect_model_load_cost): Likewise.
12951
12952 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12953             Alan Hayward  <alan.hayward@arm.com>
12954             David Sherwood  <david.sherwood@arm.com>
12955
12956         * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
12957         (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
12958         from an unsigned int * to a poly_uint64_pod *.
12959         (calculate_unrolling_factor): New function.
12960         (vect_analyze_slp_instance): Use it.  Track polynomial max_nunits.
12961
12962 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12963             Alan Hayward  <alan.hayward@arm.com>
12964             David Sherwood  <david.sherwood@arm.com>
12965
12966         * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
12967         from an unsigned int to a poly_uint64.
12968         (_loop_vec_info::slp_unrolling_factor): Likewise.
12969         (_loop_vec_info::vectorization_factor): Change from an int
12970         to a poly_uint64.
12971         (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
12972         (vect_get_num_vectors): New function.
12973         (vect_update_max_nunits, vect_vf_for_cost): Likewise.
12974         (vect_get_num_copies): Use vect_get_num_vectors.
12975         (vect_analyze_data_ref_dependences): Change max_vf from an int *
12976         to an unsigned int *.
12977         (vect_analyze_data_refs): Change min_vf from an int * to a
12978         poly_uint64 *.
12979         (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
12980         than an unsigned HOST_WIDE_INT.
12981         * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
12982         (vect_analyze_data_ref_dependence): Change max_vf from an int *
12983         to an unsigned int *.
12984         (vect_analyze_data_ref_dependences): Likewise.
12985         (vect_compute_data_ref_alignment): Handle polynomial vf.
12986         (vect_enhance_data_refs_alignment): Likewise.
12987         (vect_prune_runtime_alias_test_list): Likewise.
12988         (vect_shift_permute_load_chain): Likewise.
12989         (vect_supportable_dr_alignment): Likewise.
12990         (dependence_distance_ge_vf): Take the vectorization factor as a
12991         poly_uint64 rather than an unsigned HOST_WIDE_INT.
12992         (vect_analyze_data_refs): Change min_vf from an int * to a
12993         poly_uint64 *.
12994         * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
12995         vfm1 as a poly_uint64 rather than an int.  Make the same change
12996         for the returned bound_scalar.
12997         (vect_gen_vector_loop_niters): Handle polynomial vf.
12998         (vect_do_peeling): Likewise.  Update call to
12999         vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
13000         (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
13001         be constant.
13002         * tree-vect-loop.c (vect_determine_vectorization_factor)
13003         (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
13004         (vect_get_known_peeling_cost): Likewise.
13005         (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
13006         (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
13007         (vect_transform_loop): Likewise.  Use the lowest possible VF when
13008         updating the upper bounds of the loop.
13009         (vect_min_worthwhile_factor): Make static.  Return an unsigned int
13010         rather than an int.
13011         * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
13012         polynomial unroll factors.
13013         (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
13014         (vect_make_slp_decision): Likewise.
13015         (vect_supported_load_permutation_p): Likewise, and polynomial
13016         vf too.
13017         (vect_analyze_slp_cost): Handle polynomial vf.
13018         (vect_slp_analyze_node_operations): Likewise.
13019         (vect_slp_analyze_bb_1): Likewise.
13020         (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
13021         than an unsigned HOST_WIDE_INT.
13022         * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
13023         (vectorizable_load): Handle polynomial vf.
13024         * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
13025         a poly_uint64.
13026         (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
13027
13028 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
13029             Alan Hayward  <alan.hayward@arm.com>
13030             David Sherwood  <david.sherwood@arm.com>
13031
13032         * match.pd: Handle bit operations involving three constants
13033         and try to fold one pair.
13034
13035 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
13036
13037         * tree-vect-loop-manip.c: Include gimple-fold.h.
13038         (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
13039         niters_maybe_zero parameters.  Handle other cases besides a step of 1.
13040         (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
13041         Add a path that uses a step of VF instead of 1, but disable it
13042         for now.
13043         (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
13044         and niters_no_overflow parameters.  Update calls to
13045         slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
13046         Create a new SSA name if the latter choses to use a ste other
13047         than zero, and return it via niters_vector_mult_vf_var.
13048         * tree-vect-loop.c (vect_transform_loop): Update calls to
13049         vect_do_peeling, vect_gen_vector_loop_niters and
13050         slpeel_make_loop_iterate_ntimes.
13051         * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
13052         (vect_gen_vector_loop_niters): Update declarations after above changes.
13053
13054 2018-01-02  Michael Meissner  <meissner@linux.vnet.ibm.com>
13055
13056         * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
13057         128-bit round to integer instructions.
13058         (ceil<mode>2): Likewise.
13059         (btrunc<mode>2): Likewise.
13060         (round<mode>2): Likewise.
13061
13062 2018-01-02  Aaron Sawdey  <acsawdey@linux.vnet.ibm.com>
13063
13064         * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
13065         unaligned VSX load/store on P8/P9.
13066         (expand_block_clear): Allow the use of unaligned VSX
13067         load/store on P8/P9.
13068
13069 2018-01-02  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
13070
13071         * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
13072         New function.
13073         (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
13074         swap associated with both a load and a store.
13075
13076 2018-01-02  Andrew Waterman  <andrew@sifive.com>
13077
13078         * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
13079         * config/riscv/riscv.md (clear_cache): Use it.
13080
13081 2018-01-02  Artyom Skrobov  <tyomitch@gmail.com>
13082
13083         * web.c: Remove out-of-date comment.
13084
13085 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
13086
13087         * expr.c (fixup_args_size_notes): Check that any existing
13088         REG_ARGS_SIZE notes are correct, and don't try to re-add them.
13089         (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
13090         (emit_single_push_insn): ...here.
13091
13092 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
13093
13094         * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
13095         (const_vector_encoded_nelts): New function.
13096         (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
13097         (const_vector_int_elt, const_vector_elt): Declare.
13098         * emit-rtl.c (const_vector_int_elt_1): New function.
13099         (const_vector_elt): Likewise.
13100         * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
13101         of CONST_VECTOR_ELT.
13102
13103 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
13104
13105         * expr.c: Include rtx-vector-builder.h.
13106         (const_vector_mask_from_tree): Use rtx_vector_builder and operate
13107         directly on the tree encoding.
13108         (const_vector_from_tree): Likewise.
13109         * optabs.c: Include rtx-vector-builder.h.
13110         (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
13111         sequence of "u" values.
13112         * vec-perm-indices.c: Include rtx-vector-builder.h.
13113         (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
13114         directly on the vec_perm_indices encoding.
13115
13116 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
13117
13118         * doc/rtl.texi (const_vector): Describe new encoding scheme.
13119         * Makefile.in (OBJS): Add rtx-vector-builder.o.
13120         * rtx-vector-builder.h: New file.
13121         * rtx-vector-builder.c: Likewise.
13122         * rtl.h (rtx_def::u2): Add a const_vector field.
13123         (CONST_VECTOR_NPATTERNS): New macro.
13124         (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
13125         (CONST_VECTOR_DUPLICATE_P): Likewise.
13126         (CONST_VECTOR_STEPPED_P): Likewise.
13127         (CONST_VECTOR_ENCODED_ELT): Likewise.
13128         (const_vec_duplicate_p): Check for a duplicated vector encoding.
13129         (unwrap_const_vec_duplicate): Likewise.
13130         (const_vec_series_p): Check for a non-duplicated vector encoding.
13131         Say that the function only returns true for integer vectors.
13132         * emit-rtl.c: Include rtx-vector-builder.h.
13133         (gen_const_vec_duplicate_1): Delete.
13134         (gen_const_vector): Call gen_const_vec_duplicate instead of
13135         gen_const_vec_duplicate_1.
13136         (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
13137         (gen_const_vec_duplicate): Use rtx_vector_builder.
13138         (gen_const_vec_series): Likewise.
13139         (gen_rtx_CONST_VECTOR): Likewise.
13140         * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
13141         (swap_const_vector_halves): Take an rtx pointer rather than rtx.
13142         Build a new vector rather than modifying a CONST_VECTOR in-place.
13143         (handle_special_swappables): Update call accordingly.
13144         * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
13145         (swap_const_vector_halves): Take an rtx pointer rather than rtx.
13146         Build a new vector rather than modifying a CONST_VECTOR in-place.
13147         (handle_special_swappables): Update call accordingly.
13148
13149 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
13150
13151         * simplify-rtx.c (simplify_const_binary_operation): Use
13152         CONST_VECTOR_ELT instead of XVECEXP.
13153
13154 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
13155
13156         * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
13157         the selector elements to be different from the data elements
13158         if the selector is a VECTOR_CST.
13159         * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
13160         ssizetype for the selector.
13161
13162 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
13163
13164         * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
13165         before testing each element individually.
13166         * tree-vect-generic.c (lower_vec_perm): Likewise.
13167
13168 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
13169
13170         * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
13171         * selftest-run-tests.c (selftest::run_tests): Call it.
13172         * vector-builder.h (vector_builder::operator ==): New function.
13173         (vector_builder::operator !=): Likewise.
13174         * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
13175         (vec_perm_indices::all_from_input_p): New function.
13176         * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
13177         (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
13178         * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
13179         instead of reading the VECTOR_CST directly.  Detect whether both
13180         vector inputs are the same before constructing the vec_perm_indices,
13181         and update the number of inputs argument accordingly.  Use the
13182         utility functions added above.  Only construct sel2 if we need to.
13183
13184 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
13185
13186         * optabs.c (expand_vec_perm_var): Use an explicit encoding for
13187         the broadcast of the low byte.
13188         (expand_mult_highpart): Use an explicit encoding for the permutes.
13189         * optabs-query.c (can_mult_highpart_p): Likewise.
13190         * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
13191         * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
13192         (vectorizable_bswap): Likewise.
13193         * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
13194         explicit encoding for the power-of-2 permutes.
13195         (vect_permute_store_chain): Likewise.
13196         (vect_grouped_load_supported): Likewise.
13197         (vect_permute_load_chain): Likewise.
13198
13199 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
13200
13201         * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
13202         * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
13203         * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
13204         * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
13205         * tree-vect-stmts.c (vectorizable_bswap): Likewise.
13206         (vect_gen_perm_mask_any): Likewise.
13207
13208 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
13209
13210         * int-vector-builder.h: New file.
13211         * vec-perm-indices.h: Include int-vector-builder.h.
13212         (vec_perm_indices): Redefine as an int_vector_builder.
13213         (auto_vec_perm_indices): Delete.
13214         (vec_perm_builder): Redefine as a stand-alone class.
13215         (vec_perm_indices::vec_perm_indices): New function.
13216         (vec_perm_indices::clamp): Likewise.
13217         * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
13218         (vec_perm_indices::new_vector): New function.
13219         (vec_perm_indices::new_expanded_vector): Update for new
13220         vec_perm_indices class.
13221         (vec_perm_indices::rotate_inputs): New function.
13222         (vec_perm_indices::all_in_range_p): Operate directly on the
13223         encoded form, without computing elided elements.
13224         (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
13225         encoding.  Update for new vec_perm_indices class.
13226         * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
13227         the given vec_perm_builder.
13228         (expand_vec_perm_var): Update vec_perm_builder constructor.
13229         (expand_mult_highpart): Use vec_perm_builder instead of
13230         auto_vec_perm_indices.
13231         * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
13232         vec_perm_indices instead of auto_vec_perm_indices.  Use a single
13233         or double series encoding as appropriate.
13234         * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
13235         vec_perm_indices instead of auto_vec_perm_indices.
13236         * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
13237         * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
13238         (vect_permute_store_chain): Likewise.
13239         (vect_grouped_load_supported): Likewise.
13240         (vect_permute_load_chain): Likewise.
13241         (vect_shift_permute_load_chain): Likewise.
13242         * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
13243         (vect_transform_slp_perm_load): Likewise.
13244         (vect_schedule_slp_instance): Likewise.
13245         * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
13246         (vectorizable_mask_load_store): Likewise.
13247         (vectorizable_bswap): Likewise.
13248         (vectorizable_store): Likewise.
13249         (vectorizable_load): Likewise.
13250         * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
13251         vec_perm_indices instead of auto_vec_perm_indices.  Use
13252         tree_to_vec_perm_builder to read the vector from a tree.
13253         * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
13254         vec_perm_builder instead of a vec_perm_indices.
13255         (have_whole_vector_shift): Use vec_perm_builder and
13256         vec_perm_indices instead of auto_vec_perm_indices.  Leave the
13257         truncation to calc_vec_perm_mask_for_shift.
13258         (vect_create_epilog_for_reduction): Likewise.
13259         * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
13260         from auto_vec_perm_indices to vec_perm_indices.
13261         (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
13262         instead of changing individual elements.
13263         (aarch64_vectorize_vec_perm_const): Use new_vector to install
13264         the vector in d.perm.
13265         * config/arm/arm.c (expand_vec_perm_d::perm): Change
13266         from auto_vec_perm_indices to vec_perm_indices.
13267         (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
13268         instead of changing individual elements.
13269         (arm_vectorize_vec_perm_const): Use new_vector to install
13270         the vector in d.perm.
13271         * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
13272         Update vec_perm_builder constructor.
13273         (rs6000_expand_interleave): Likewise.
13274         * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
13275         (rs6000_expand_interleave): Likewise.
13276
13277 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
13278
13279         * optabs-query.c (can_vec_perm_var_p): Check whether lowering
13280         to qimode could truncate the indices.
13281         * optabs.c (expand_vec_perm_var): Likewise.
13282
13283 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
13284
13285         * Makefile.in (OBJS): Add vec-perm-indices.o.
13286         * vec-perm-indices.h: New file.
13287         * vec-perm-indices.c: Likewise.
13288         * target.h (vec_perm_indices): Replace with a forward class
13289         declaration.
13290         (auto_vec_perm_indices): Move to vec-perm-indices.h.
13291         * optabs.h: Include vec-perm-indices.h.
13292         (expand_vec_perm): Delete.
13293         (selector_fits_mode_p, expand_vec_perm_var): Declare.
13294         (expand_vec_perm_const): Declare.
13295         * target.def (vec_perm_const_ok): Replace with...
13296         (vec_perm_const): ...this new hook.
13297         * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
13298         (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
13299         * doc/tm.texi: Regenerate.
13300         * optabs.def (vec_perm_const): Delete.
13301         * doc/md.texi (vec_perm_const): Likewise.
13302         (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
13303         * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
13304         expand_vec_perm for constant permutation vectors.  Assert that
13305         the mode of variable permutation vectors is the integer equivalent
13306         of the mode that is being permuted.
13307         * optabs-query.h (selector_fits_mode_p): Declare.
13308         * optabs-query.c: Include vec-perm-indices.h.
13309         (selector_fits_mode_p): New function.
13310         (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
13311         is defined, instead of checking whether the vec_perm_const_optab
13312         exists.  Use targetm.vectorize.vec_perm_const instead of
13313         targetm.vectorize.vec_perm_const_ok.  Check whether the indices
13314         fit in the vector mode before using a variable permute.
13315         * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
13316         vec_perm_indices instead of an rtx.
13317         (expand_vec_perm): Replace with...
13318         (expand_vec_perm_const): ...this new function.  Take the selector
13319         as a vec_perm_indices rather than an rtx.  Also take the mode of
13320         the selector.  Update call to shift_amt_for_vec_perm_mask.
13321         Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
13322         Use vec_perm_indices::new_expanded_vector to expand the original
13323         selector into bytes.  Check whether the indices fit in the vector
13324         mode before using a variable permute.
13325         (expand_vec_perm_var): Make global.
13326         (expand_mult_highpart): Use expand_vec_perm_const.
13327         * fold-const.c: Includes vec-perm-indices.h.
13328         * tree-ssa-forwprop.c: Likewise.
13329         * tree-vect-data-refs.c: Likewise.
13330         * tree-vect-generic.c: Likewise.
13331         * tree-vect-loop.c: Likewise.
13332         * tree-vect-slp.c: Likewise.
13333         * tree-vect-stmts.c: Likewise.
13334         * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
13335         Delete.
13336         * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
13337         * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
13338         (aarch64_vectorize_vec_perm_const_ok): Fuse into...
13339         (aarch64_vectorize_vec_perm_const): ...this new function.
13340         (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
13341         (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
13342         * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
13343         * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
13344         * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
13345         (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
13346         (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
13347         into...
13348         (arm_vectorize_vec_perm_const): ...this new function.  Explicitly
13349         check for NEON modes.
13350         * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
13351         * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
13352         * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
13353         (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
13354         into...
13355         (ix86_vectorize_vec_perm_const): ...this new function.  Incorporate
13356         the old VEC_PERM_CONST conditions.
13357         * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
13358         * config/ia64/vect.md (vec_perm_const<mode>): Delete.
13359         * config/ia64/ia64.c (ia64_expand_vec_perm_const)
13360         (ia64_vectorize_vec_perm_const_ok): Merge into...
13361         (ia64_vectorize_vec_perm_const): ...this new function.
13362         * config/mips/loongson.md (vec_perm_const<mode>): Delete.
13363         * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
13364         * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
13365         * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
13366         * config/mips/mips.c (mips_expand_vec_perm_const)
13367         (mips_vectorize_vec_perm_const_ok): Merge into...
13368         (mips_vectorize_vec_perm_const): ...this new function.
13369         * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
13370         * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
13371         * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
13372         * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
13373         * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
13374         (rs6000_expand_vec_perm_const): Delete.
13375         * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
13376         Delete.
13377         (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
13378         (altivec_expand_vec_perm_const_le): Take each operand individually.
13379         Operate on constant selectors rather than rtxes.
13380         (altivec_expand_vec_perm_const): Likewise.  Update call to
13381         altivec_expand_vec_perm_const_le.
13382         (rs6000_expand_vec_perm_const): Delete.
13383         (rs6000_vectorize_vec_perm_const_ok): Delete.
13384         (rs6000_vectorize_vec_perm_const): New function.
13385         (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
13386         an element count and rtx array.
13387         (rs6000_expand_extract_even): Update call accordingly.
13388         (rs6000_expand_interleave): Likewise.
13389         * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
13390         * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
13391         * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
13392         * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
13393         (rs6000_expand_vec_perm_const): Delete.
13394         * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
13395         (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
13396         (altivec_expand_vec_perm_const_le): Take each operand individually.
13397         Operate on constant selectors rather than rtxes.
13398         (altivec_expand_vec_perm_const): Likewise.  Update call to
13399         altivec_expand_vec_perm_const_le.
13400         (rs6000_expand_vec_perm_const): Delete.
13401         (rs6000_vectorize_vec_perm_const_ok): Delete.
13402         (rs6000_vectorize_vec_perm_const): New function.  Remove stray
13403         reference to the SPE evmerge intructions.
13404         (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
13405         an element count and rtx array.
13406         (rs6000_expand_extract_even): Update call accordingly.
13407         (rs6000_expand_interleave): Likewise.
13408         * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
13409         * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
13410         new function.
13411         (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
13412
13413 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
13414
13415         * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
13416         vector mode and that that mode matches the mode of the data
13417         being permuted.
13418         (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
13419         out into expand_vec_perm_var.  Do all CONST_VECTOR handling here,
13420         directly using expand_vec_perm_1 when forcing selectors into
13421         registers.
13422         (expand_vec_perm_var): New function, split out from expand_vec_perm.
13423
13424 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
13425
13426         * optabs-query.h (can_vec_perm_p): Delete.
13427         (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
13428         * optabs-query.c (can_vec_perm_p): Split into...
13429         (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
13430         (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
13431         particular selector is valid.
13432         * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
13433         * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
13434         (vect_grouped_load_supported): Likewise.
13435         (vect_shift_permute_load_chain): Likewise.
13436         * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
13437         (vect_transform_slp_perm_load): Likewise.
13438         * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
13439         (vectorizable_bswap): Likewise.
13440         (vect_gen_perm_mask_checked): Likewise.
13441         * fold-const.c (fold_ternary_loc): Likewise.  Don't take
13442         implementations of variable permutation vectors into account
13443         when deciding which selector to use.
13444         * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
13445         vec_perm_const_optab is supported; instead use can_vec_perm_const_p
13446         with a false third argument.
13447         * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
13448         to test whether the constant selector is valid and can_vec_perm_var_p
13449         to test whether a variable selector is valid.
13450
13451 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
13452
13453         * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
13454         * optabs-query.c (can_vec_perm_p): Likewise.
13455         * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
13456         instead of vec_perm_indices.
13457         * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
13458         (vect_gen_perm_mask_checked): Likewise,
13459         * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
13460         (vect_gen_perm_mask_checked): Likewise,
13461
13462 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
13463
13464         * optabs-query.h (qimode_for_vec_perm): Declare.
13465         * optabs-query.c (can_vec_perm_p): Split out qimode search to...
13466         (qimode_for_vec_perm): ...this new function.
13467         * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
13468
13469 2018-01-02  Aaron Sawdey  <acsawdey@linux.vnet.ibm.com>
13470
13471         * rtlanal.c (canonicalize_condition): Return 0 if final rtx
13472         does not have a conditional at the top.
13473
13474 2018-01-02  Richard Biener  <rguenther@suse.de>
13475
13476         * ipa-inline.c (big_speedup_p): Fix expression.
13477
13478 2018-01-02  Jan Hubicka  <hubicka@ucw.cz>
13479
13480         PR target/81616
13481         * config/i386/x86-tune-costs.h: Increase cost of integer load costs
13482         for generic 4->6.
13483
13484 2018-01-02  Jan Hubicka  <hubicka@ucw.cz>
13485
13486         PR target/81616
13487         Generic tuning.
13488         * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
13489         cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
13490         and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
13491         cond_taken_branch_cost 3->4.
13492
13493 2018-01-01  Jakub Jelinek  <jakub@redhat.com>
13494
13495         PR tree-optimization/83581
13496         * tree-loop-distribution.c (pass_loop_distribution::execute): Return
13497         TODO_cleanup_cfg if any changes have been made.
13498
13499         PR middle-end/83608
13500         * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
13501         convert_modes if target mode has the right side, but different mode
13502         class.
13503
13504         PR middle-end/83609
13505         * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
13506         last argument when extracting from CONCAT.  If either from_real or
13507         from_imag is NULL, use expansion through memory.  If result is not
13508         a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
13509         the parts directly to inner mode, if even that fails, use expansion
13510         through memory.
13511
13512         PR middle-end/83623
13513         * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
13514         check for bswap in mode rather than HImode and use that in expand_unop
13515         too.
13516 \f
13517 Copyright (C) 2018 Free Software Foundation, Inc.
13518
13519 Copying and distribution of this file, with or without modification,
13520 are permitted in any medium without royalty provided the copyright
13521 notice and this notice are preserved.