gcc/ChangeLog

   1 2018-05-22  Jackson Woodruff  <jackson.woodruff@arm.com>
   2             Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
   3
   4         * config/aarch64/aarch64.md: New patterns to generate stp
   5         and ldp.
   6         (store_pair_sw, store_pair_dw): New patterns to generate stp for
   7         single words and double words.
   8         (load_pair_sw, load_pair_dw): Likewise.
   9         (store_pair_sf, store_pair_df, store_pair_si, store_pair_di):
  10         Delete.
  11         (load_pair_sf, load_pair_df, load_pair_si, load_pair_di):
  12         Delete.
  13         * config/aarch64/aarch64-ldpstp.md: Modify peephole
  14         for different mode ldpstp and add peephole for merged zero stores.
  15         Likewise for loads.
  16         * config/aarch64/aarch64.c (aarch64_operands_ok_for_ldpstp):
  17         Add size check.
  18         (aarch64_gen_store_pair): Rename calls to match new patterns.
  19         (aarch64_gen_load_pair): Rename calls to match new patterns.
  20         * config/aarch64/aarch64-simd.md (load_pair<mode>): Rename to...
  21         (load_pair<DREG:mode><DREG2:mode>): ... This.
  22         (store_pair<mode>): Rename to...
  23         (vec_store_pair<DREG:mode><DREG2:mode>): ... This.
  24         * config/aarch64/iterators.md (DREG, DREG2, DX2, SX, SX2, DSX):
  25         New mode iterators.
  26         (V_INT_EQUIV): Handle SImode.
  27         * config/aarch64/predicates.md (aarch64_reg_zero_or_fp_zero):
  28         New predicate.
  29
  30 2018-05-22  Martin Sebor  <msebor@redhat.com>
  31
  32         PR tree-optimization/85826
  33         * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Avoid
  34         assuming that a DECL necesarily has a constant size.
  35
  36 2018-05-22  Richard Sandiford  <richard.sandiford@linaro.org>
  37
  38         PR middle-end/85862
  39         * internal-fn.c (expand_direct_optab_fn): Cope with a null lhs.
  40
  41 2018-05-22  Richard Biener  <rguenther@suse.de>
  42
  43         PR tree-optimization/85834
  44         * tree-ssa-sccvn.c (vn_reference_lookup_3): Properly handle
  45         non-constant and non-zero memset arguments.
  46
  47 2018-05-22  Martin Liska  <mliska@suse.cz>
  48
  49         PR ipa/85607
  50         * ipa-icf.c (sem_item::add_type): Do not ICE for incomplete types.
  51
  52 2018-05-22  Richard Biener  <rguenther@suse.de>
  53
  54         PR tree-optimization/85863
  55         * tree-vect-stmts.c (vect_is_simple_cond): Only widen invariant
  56         comparisons when vectype is specified.
  57         (vectorizable_condition): Do not specify vectype for
  58         vect_is_simple_cond when SLP vectorizing.
  59
  60 2018-05-21  Michael Meissner  <meissner@linux.ibm.com>
  61
  62         PR target/85657
  63         * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): Do not
  64         define __ibm128 as long double.
  65         * config/rs6000/rs6000.c (rs6000_init_builtins): Create __ibm128
  66         as a distinct type when IEEE 128-bit support is enabled.
  67         (init_float128_ieee): Fix up conversions between IFmode and IEEE
  68         128-bit types to use the correct functions.
  69         (rs6000_expand_float128_convert): Use explicit FLOAT_EXTEND to
  70         convert between 128-bit floating point types that have different
  71         modes but the same representation, instead of using gen_lowpart to
  72         makean alias.
  73         * config/rs6000/rs6000.md (IFKF): New iterator for IFmode and
  74         KFmode.
  75         (IFKF_reg): New attributes to give the register constraints for
  76         IFmode and KFmode.
  77         (extend<mode>tf2_internal): New insns to mark an explicit
  78         conversion between 128-bit floating point types that have a
  79         different mode but share the same representation.
  80
  81 2018-05-21  Richard Sandiford  <richard.sandiford@linaro.org>
  82
  83         PR tree-optimization/85814
  84         * tree-ssa-strlen.c (get_stridx_plus_constant): Cope with
  85         a null return from get_strinfo when unsharing the next
  86         strinfo in the chain.
  87
  88 2018-05-21  Vladimir Mezentsev  <vladimir.mezentsev@oracle.com>
  89
  90         PR gcc/84923
  91         * varasm.c (weak_finish): Clean up weak_decls.
  92
  93 2018-05-21  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
  94
  95         * config/aarch64/aarch64.md ("unspec"): Define UNSPEC_SABAL,
  96         UNSPEC_SABDL2, UNSPEC_SADALP, UNSPEC_UABAL, UNSPEC_UABDL2,
  97         UNSPEC_UADALP values.
  98         * config/aarch64/iterators.md (ABAL): New int iterator.
  99         (ABDL2): Likewise.
 100         (ADALP): Likewise.
 101         (sur): Add mappings for the above.
 102         * config/aarch64/aarch64-simd.md (aarch64_<sur>abdl2<mode>_3):
 103         New define_insn.
 104         (aarch64_<sur>abal<mode>_4): Likewise.
 105         (aarch64_<sur>adalp<mode>_3): Likewise.
 106         (<sur>sadv16qi): New define_expand.
 107
 108 2018-05-21  Alexander Nesterovskiy  <alexander.nesterovskiy@intel.com>
 109
 110         * config/i386/i386.md (*movsf_internal): AVX falsedep fix.
 111         (*movdf_internal): Ditto.
 112         (*rcpsf2_sse): Ditto.
 113         (*rsqrtsf2_sse): Ditto.
 114         (*sqrt<mode>2_sse): Ditto.
 115
 116 2018-05-21  Tamar Christina  <tamar.christina@arm.com>
 117
 118         * config/aarch64/aarch64-simd.md (aarch64_eor3qv8hi): Change to
 119         eor3q<mode>4.
 120         (aarch64_bcaxqv8hi): Change to bcaxq<mode>4.
 121         * config/aarch64/aarch64-simd-builtins.def (veor3q_u8, veor3q_u32,
 122         veor3q_u64, veor3q_s8, veor3q_s16, veor3q_s32, veor3q_s64, vbcaxq_u8,
 123         vbcaxq_u32, vbcaxq_u64, vbcaxq_s8, vbcaxq_s16, vbcaxq_s32,
 124         vbcaxq_s64): New.
 125         * config/aarch64/arm_neon.h: Likewise.
 126         * config/aarch64/iterators.md (VQ_I): New.
 127
 128 2018-05-21  Alexey Brodkin <abrodkin@synopsys.com>
 129
 130         * config.gcc: Add arc/t-multilib-linux to tmake_file for
 131         arc*-*-linux*.
 132         * config/arc/t-multilib-linux: Specify MULTILIB_OPTIONS and
 133         MULTILIB_DIRNAMES
 134
 135 2018-05-20  Chung-Ju Wu  <jasonwucj@gmail.com>
 136
 137         * config/nds32/constraints.md (S): New constraint.
 138         * config/nds32/nds32.md (call_internal): Use constraint S.
 139         (call_value_internal): Likewise.
 140         (sibcall_internal): Likewise.
 141         (sibcall_value_internal): Likewise.
 142
 143 2018-05-20  Kito Cheng  <kito.cheng@gmail.com>
 144             Chung-Ju Wu  <jasonwucj@gmail.com>
 145
 146         * config/nds32/nds32.c (nds32_register_move_cost): Take garywolf cpu
 147         into consideration.
 148
 149 2018-05-20  Kito Cheng  <kito.cheng@gmail.com>
 150             Chung-Ju Wu  <jasonwucj@gmail.com>
 151
 152         * config/nds32/nds32-cost.c (rtx_cost_model_t): New structure.
 153         (insn_size_16bit, insn_size_32bit): New variables for cost evaluation.
 154         (nds32_rtx_costs_impl): Simplify.
 155         (nds32_address_cost_impl): Simplify.
 156         (nds32_init_rtx_costs): New function.
 157         (nds32_rtx_costs_speed_prefer): Likewise.
 158         (nds32_rtx_costs_size_prefer): Likewise.
 159         (nds32_address_cost_speed_prefer): Likewise.
 160         (nds32_address_cost_speed_fwprop): Likewise.
 161         (nds32_address_cost_size_prefer): Likewise.
 162         * config/nds32/nds32-protos.h (nds32_init_rtx_costs): Declare.
 163         * config/nds32/nds32.c (nds32_option_override): Use
 164         nds32_init_rtx_costs function.
 165
 166 2018-05-20  Chung-Ju Wu  <jasonwucj@gmail.com>
 167
 168         * config/nds32/nds32.c (nds32_asm_file_start): Output pipeline model.
 169         * config/nds32/nds32.h (TARGET_PIPELINE_N7): Define.
 170         (TARGET_PIPELINE_N8): Likewise.
 171         (TARGET_PIPELINE_N10): Likewise.
 172         (TARGET_PIPELINE_N13): Likewise.
 173         (TARGET_PIPELINE_GRAYWOLF): Likewise.
 174
 175 2018-05-19  Monk Chiang  <sh.chiang04@gmail.com>
 176
 177         * config/nds32/nds32-fpu.md: Update copyright year.
 178
 179 2018-05-19  Chung-Ju Wu  <jasonwucj@gmail.com>
 180
 181         * config/nds32/nds32.h (ASM_SPEC): Adjust spec rule.
 182
 183 2018-05-19  Chung-Ju Wu  <jasonwucj@gmail.com>
 184
 185         * config/nds32/nds32.c
 186         (nds32_md_asm_adjust): Consider flag_inline_asm_r15 variable.
 187         * config/nds32/nds32.opt (minline-asm-r15): New option.
 188
 189 2018-05-19  Chung-Ju Wu  <jasonwucj@gmail.com>
 190
 191         * common/config/nds32/nds32-common.c (TARGET_DEFAULT_TARGET_FLAGS): Add
 192         MASK_HW_ABS.
 193         * config/nds32/nds32.md (abssi2): New pattern.
 194
 195 2018-05-19  Uros Bizjak  <ubizjak@gmail.com>
 196
 197         * config/i386/i386.md (rex64namesuffix): New mode attribute.
 198         * config/i386/sse.md (sse_cvtsi2ss<rex64namesuffix><round_name>):
 199         Merge insn pattern from sse_cvtsi2ss<round_name> and
 200         sse_cvtsi2ssq<round_name> using SWI48 mode iterator.
 201         (sse_cvtss2si<rex64namesuffix><round_name>): Merge insn pattern
 202         from sse_cvtss2si<round_name> and sse_cvtss2siq<round_name>
 203         using SWI48 mode iterator.
 204         (sse_cvtss2si<rex64namesuffix>_2): Merge insn pattern from
 205         sse_cvtss2si_2 and sse_cvtss2siq_2 using SWI48 mode iterator.
 206         (sse_cvttss2si<rex64namesuffix><round_saeonly_name>): Merge insn
 207         pattern from sse_cvttss2si<round_saeonly_name>
 208         and sse_cvttss2siq<round_saeonly_name> using SWI48 mode iterator.
 209         (avx512f_vcvtss2usi<rex64namesuffix><round_name>): Merge insn pattern
 210         from avx512f_vcvtss2usi<round_name> and avx512f_vcvtss2usiq<round_name>
 211         using SWI48 mode iterator.
 212         (avx512f_vcvttss2usi<rex64namesuffix><round_saeonly_name>): Merge
 213         insn pattern from avx512f_vcvttss2usi<round_saeonly_name> and
 214         avx512f_vcvttss2usiq<round_saeonly_name> using SWI48 mode iterator.
 215         (avx512f_vcvtsd2usi<rex64namesuffix><round_name>): Merge insn pattern
 216         from avx512f_vcvtsd2usi<round_name> and avx512f_vcvtsd2usiq<round_name>
 217         using SWI48 mode iterator.
 218         (avx512f_vcvttsd2usi<rex64namesuffix><round_saeonly_name>): Merge
 219         insn pattern from avx512f_vcvttsd2usi<round_saeonly_name> and
 220         avx512f_vcvttsd2usiq<round_saeonly_name> using SWI48 mode iterator.
 221         (sse2_cvtsd2si<rex64namesuffix><round_name>): Merge insn pattern from
 222         sse2_cvtsd2si<round_name> and sse2_cvtsd2siq<round_name> using
 223         SWI48 mode iterator.
 224         (sse2_cvtsd2si<rex64namesuffix>_2): Merge insn pattern from
 225         sse2_cvtsd2si_2 and sse2_cvtsd2siq_2 using SWI48 mode iterator.
 226         (sse_cvttsd2si<rex64namesuffix><round_saeonly_name>): Merge insn
 227         pattern from sse_cvttsd2si<round_saeonly_name>
 228         and sse_cvttsd2siq<round_saeonly_name> using SWI48 mode iterator.
 229
 230 2018-05-19  Chung-Ju Wu  <jasonwucj@gmail.com>
 231
 232         * config/nds32/nds32-md-auxiliary.c
 233         (nds32_valid_smw_lwm_base_p): Refine.
 234         (nds32_output_smw_single_word): Refine.
 235         (nds32_output_smw_double_word): New.
 236         * config/nds32/nds32-protos.h (nds32_output_smw_double_word): New.
 237
 238 2018-05-19  Chung-Ju Wu  <jasonwucj@gmail.com>
 239
 240         * config/nds32/nds32-md-auxiliary.c (nds32_output_stack_push): Refine.
 241         (nds32_output_stack_pop): Refine.
 242         (nds32_expand_unaligned_load): Refine.
 243         (nds32_expand_unaligned_store): Refine.
 244
 245 2018-05-19  Kuan-Lin Chen <kuanlinchentw@gmail.com>
 246             Chung-Ju Wu  <jasonwucj@gmail.com>
 247
 248         * config/nds32/constants.md: Add TP_REGNUM constant.
 249         (unspec_element): Add UNSPEC_GOTINIT, UNSPEC_GOT, UNSPEC_GOTOFF,
 250         UNSPEC_PLT, UNSPEC_TLSGD, UNSPEC_TLSLD, UNSPEC_TLSIE, UNSPEC_TLSLE and
 251         UNSPEC_ADD32.
 252         * config/nds32/nds32-doubleword.md: Consider flag_pic.
 253         * config/nds32/nds32-dspext.md (mov<mode>): Expand TLS and PIC cases.
 254         * config/nds32/nds32-predicates.c (nds32_const_unspec_p): New.
 255         * config/nds32/nds32-md-auxiliary.c: Implementation that support TLS
 256         and PIC code generation.
 257         * config/nds32/nds32-protos.h: Declarations that support TLS and PIC
 258         code generation.
 259         * config/nds32/nds32-relax-opt.c: Consider TLS and PIC for relax
 260         optimization.
 261         * config/nds32/nds32.md: Support TLS and PIC.
 262         * config/nds32/nds32.c: Support TLS and PIC.
 263         * config/nds32/nds32.h (nds32_relax_insn_type): New enum type.
 264         * config/nds32/predicates.md (nds32_nonunspec_symbolic_operand): New
 265         predicate.
 266
 267 2018-05-19  Chung-Ju Wu  <jasonwucj@gmail.com>
 268
 269         * config/nds32/nds32-predicates.c (const_vector_to_hwint): Use machine
 270         mode with E_ prefix.
 271
 272 2018-05-19  Kuan-Lin Chen <kuanlinchentw@gmail.com>
 273             Chung-Ju Wu  <jasonwucj@gmail.com>
 274
 275         * config/nds32/constants.md (unspec_element): Add UNSPEC_ICT.
 276         * config/nds32/nds32-md-auxiliary.c
 277         (symbolic_reference_mentioned_p): New.
 278         (nds32_legitimize_ict_address): New.
 279         (nds32_expand_ict_move): New.
 280         (nds32_indirect_call_referenced_p): New.
 281         (nds32_symbol_binds_local_p): Delete.
 282         (nds32_long_call_p): Modify.
 283         * config/nds32/nds32-opts.h (nds32_ict_model_type): New enum type.
 284         * config/nds32/nds32-protos.h
 285         (symbolic_reference_mentioned_p): Declare.
 286         (nds32_legitimize_ict_address): Declare.
 287         (nds32_expand_ict_move): Declare.
 288         (nds32_indirect_call_referenced_p): Declare.
 289         * config/nds32/nds32-relax-opt.c (nds32_ict_const_p): New.
 290         (nds32_relax_group): Use nds32_ict_const_p as condition.
 291         * config/nds32/nds32.c (nds32_attribute_table): Add "indirect_call".
 292         (nds32_asm_file_start): Output ict_model directive in asm code.
 293         (nds32_legitimate_address_p): Consider indirect call.
 294         (nds32_print_operand): Consider indirect call.
 295         (nds32_print_operand_address): Consider indirect call.
 296         (nds32_insert_attributes): Handle "indirect_call" attribute.
 297         (TARGET_LEGITIMATE_ADDRESS_P): Define.
 298         (TARGET_LEGITIMATE_CONSTANT_P): Define.
 299         (TARGET_CANNOT_FORCE_CONST_MEM): Define.
 300         (TARGET_DELEGITIMIZE_ADDRESS): Define.
 301         (TARGET_ASM_OUTPUT_ADDR_CONST_EXTRA): Define.
 302         * config/nds32/nds32.h (SYMBOLIC_CONST_P): Define.
 303         (TARGET_ICT_MODEL_SMALL): Define.
 304         (TARGET_ICT_MODEL_LARGE): Define.
 305         * config/nds32/nds32.md (movsi): Consider ict model.
 306         (call, call_value): Consider ict model.
 307         (sibcall, sibcall_value): Consider ict model.
 308         * config/nds32/nds32.opt (mict-model): New option.
 309         * config/nds32/predicates.md (nds32_symbolic_operand): Consider ict
 310         model.
 311
 312 2018-05-18  Kito Cheng  <kito.cheng@gmail.com>
 313             Monk Chiang  <sh.chiang04@gmail.com>
 314             Jim Wilson <jimw@sifive.com>
 315
 316         * common/config/riscv/riscv-common.c (riscv_parse_arch_string):
 317         Add support to parse rv32e*.  Clear MASK_RVE for rv32i and rv64i.
 318         * config.gcc (riscv*-*-*): Add support for rv32e* and ilp32e.
 319         * config/riscv/riscv-c.c (riscv_cpu_cpp_builtins): Define
 320         __riscv_32e when TARGET_RVE.  Handle ABI_ILP32E as soft-float ABI.
 321         * config/riscv/riscv-opts.h (riscv_abi_type): Add ABI_ILP32E.
 322         * config/riscv/riscv.c (riscv_compute_frame_info): When TARGET_RVE,
 323         compute save_libcall_adjustment properly.
 324         (riscv_option_override): Call error if TARGET_RVE and not ABI_ILP32E.
 325         (riscv_conditional_register_usage): Handle TARGET_RVE and ABI_ILP32E.
 326         * config/riscv/riscv.h (UNITS_PER_FP_ARG): Handle ABI_ILP32E.
 327         (STACK_BOUNDARY, ABI_STACK_BOUNDARY): Handle TARGET_RVE.
 328         (GP_REG_LAST, MAX_ARGS_IN_REGISTERS): Likewise.
 329         (ABI_SPEC): Handle mabi=ilp32e.
 330         * config/riscv/riscv.opt (abi_type): Add ABI_ILP32E.
 331         (RVE): Add RVE mask.
 332         * doc/invoke.texi (RISC-V options) <-mabi>: Add ilp32e info.
 333         <-march>: Add rv32e as an example.
 334
 335 2018-05-18  Marc Glisse  <marc.glisse@inria.fr>
 336
 337         PR c++/82899
 338         * tree-ssa-structalias.c (create_variable_info_for_1): Extra argument.
 339         (intra_create_variable_infos): Handle C++ constructors.
 340
 341 2018-05-18  Martin Liska  <mliska@suse.cz>
 342
 343         * passes.def: Remove a redundant pass.
 344
 345 2018-05-18  Eric Botcazou  <ebotcazou@adacore.com>
 346
 347         PR bootstrap/85838
 348         * config/sparc/sparc.c (sparc_expand_builtin): Always initialize op[0].
 349
 350 2018-05-18  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
 351
 352         * config/arm/arm-cpus.in (armv3m, mode32): Delete features.
 353         (ARMv4): Update.
 354         (ARMv2, ARMv3, ARMv3m): Delete fgroups.
 355         (ARMv6m): Update.
 356         (armv2, armv2a, armv3, armv3m): Delete architectures.
 357         (arm2, arm250, arm3, arm6, arm60, arm600, arm610, arm620,
 358         arm7, arm7d, arm7di, arm70, arm700, arm700i, arm710, arm720,
 359         arm710c, arm7100, arm7500, arm7500fe, arm7m, arm7dm, arm7dmi):
 360         Delete cpus.
 361         * config/arm/arm.md (maddsidi4): Remove check for arm_arch3m.
 362         (*mulsidi3adddi): Likewise.
 363         (mulsidi3): Likewise.
 364         (*mulsidi3_nov6): Likewise.
 365         (umulsidi3): Likewise.
 366         (umulsidi3_nov6): Likewise.
 367         (umaddsidi4): Likewise.
 368         (*umulsidi3adddi): Likewise.
 369         (smulsi3_highpart): Likewise.
 370         (*smulsi3_highpart_nov6): Likewise.
 371         (umulsi3_highpart): Likewise.
 372         (*umulsi3_highpart_nov6): Likewise.
 373         * config/arm/arm.h (arm_arch3m): Delete.
 374         * config/arm/arm.c (arm_arch3m): Delete.
 375         (arm_option_override_internal): Update armv3-related comment.
 376         (arm_configure_build_target): Delete use of isa_bit_mode32.
 377         (arm_option_reconfigure_globals): Delete set of arm_ach3m.
 378         (arm_rtx_costs_internal): Delete check of arm_arch3m.
 379         * config/arm/arm-fixed.md (mulsq3): Delete check for arm_arch3m.
 380         (mulsa3): Likewise.
 381         (mulusa3): Likewise.
 382         * config/arm/arm-protos.h (arm_arch3m): Delete.
 383         * config/arm/arm-tables.opt: Regenerate.
 384         * config/arm/arm-tune.md: Likewise.
 385         * config/arm/t-arm-elf (all_early_nofp): Delete mentions of
 386         deleted architectures.
 387
 388 2018-05-18  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
 389
 390         * config/arm/arm-cpus.in (armv5, armv5e): Delete features.
 391         (armv5t, armv5te): New features.
 392         (ARMv5, ARMv5e): Delete fgroups.
 393         (ARMv5t, ARMv5te): Adjust for above changes.
 394         (ARMv6m): Likewise.
 395         (armv5, armv5e): Delete arches.
 396         * config/arm/arm.md (*call_reg_armv5): Use arm_arch5t instead of
 397         arm_arch5.
 398         (*call_reg_arm): Likewise.
 399         (*call_value_reg_armv5): Likewise.
 400         (*call_value_reg_arm): Likewise.
 401         (*call_symbol): Likewise.
 402         (*call_value_symbol): Likewise.
 403         (*sibcall_insn): Likewise.
 404         (*sibcall_value_insn): Likewise.
 405         (clzsi2): Likewise.
 406         (prefetch): Likewise.
 407         (define_split and define_peephole2 dependent on arm_arch5):
 408         Likewise.
 409         * config/arm/arm.h (TARGET_LDRD): Use arm_arch5te instead of
 410         arm_arch5e.
 411         (TARGET_ARM_QBIT): Likewise.
 412         (TARGET_DSP_MULTIPLY): Likewise.
 413         (enum base_architecture): Delete BASE_ARCH_5, BASE_ARCH_5E.
 414         (arm_arch5, arm_arch5e): Delete.
 415         (arm_arch5t, arm_arch5te): Declare.
 416         * config/arm/arm.c (arm_arch5, arm_arch5e): Delete.
 417         (arm_arch5t): Declare.
 418         (arm_option_reconfigure_globals): Update for the above.
 419         (arm_options_perform_arch_sanity_checks): Update comment, replace
 420         use of arm_arch5 with arm_arch5t.
 421         (use_return_insn): Likewise.
 422         (arm_emit_call_insn): Likewise.
 423         (output_return_instruction): Likewise.
 424         (arm_final_prescan_insn): Likewise.
 425         (arm_coproc_builtin_available): Likewise.
 426         * config/arm/arm-c.c (arm_cpu_builtins): Replace arm_arch5 and
 427         arm_arch5e with arm_arch5t and arm_arch5te.
 428         * config/arm/arm-protos.h (arm_arch5, arm_arch5e): Delete.
 429         (arm_arch5t, arm_arch5te): Declare.
 430         * config/arm/arm-tables.opt: Regenerate.
 431         * config/arm/t-arm-elf: Remove references to armv5, armv5e.
 432         * config/arm/t-multilib: Likewise.
 433         * config/arm/thumb1.md (*call_reg_thumb1_v5): Check arm_arch5t
 434         instead of arm_arch5.
 435         (*call_reg_thumb1): Likewise.
 436         (*call_value_reg_thumb1_v5): Likewise.
 437         (*call_value_reg_thumb1): Likewise.
 438         * config/arm/vxworks.h (TARGET_OS_CPP_BUILTINS): Remove now
 439         unreachable path.
 440         * doc/invoke.texi (ARM Options): Remove references to armv5, armv5e.
 441
 442 2018-05-18  Martin Liska  <mliska@suse.cz>
 443
 444         PR gcov-profile/84846
 445         * doc/gcov.texi: Document -t option of gcov tool.
 446
 447 2018-05-18  Martin Liska  <mliska@suse.cz>
 448
 449         PR gcov-profile/84846
 450         * gcov.c (print_usage): Add new -t option.
 451         (process_args): Handle the option.
 452         (generate_results): Use stdout as output when requested by
 453         the option.
 454
 455 2018-05-18  Martin Liska  <mliska@suse.cz>
 456
 457         PR gcov-profile/84846
 458         * coverage.c (coverage_init): Write PWD to .gcno file.
 459         * doc/gcov.texi: Document how working directory is printed.
 460         * gcov-dump.c (dump_gcov_file): Print PWD.
 461         * gcov.c (output_intermediate_file): Likewise.
 462         (read_graph_file): Read PWD string.
 463         (output_lines): Print PWD.
 464
 465 2018-05-18  Prathamesh Kulkarni  <prathamesh.kulkarni@linaro.org>
 466
 467         PR middle-end/85817
 468         * ipa-pure-const.c (malloc_candidate_p): Remove the check integer_zerop
 469         for retval and return false if all args to phi are zero.
 470
 471 2018-05-18  Richard Biener  <rguenther@suse.de>
 472
 473         * gimple-ssa-evrp.c (class evrp_folder): Add simplify_stmt_using_ranges
 474         method.
 475         (evrp_dom_walker::before_dom_children): Call it.
 476
 477 2018-05-18  Richard Biener  <rguenther@suse.de>
 478
 479         * tree-dfa.c (get_ref_base_and_extent): Use range-info to refine
 480         results when processing array refs with variable index.
 481
 482 2018-05-18  Toon Moene  <toon@moene.org>
 483
 484         * doc/invoke.texi: Move -floop-unroll-and-jam documentation
 485         directly after that of -floop-interchange. Indicate that both
 486         options are enabled by default when specifying -O3.
 487
 488 2018-05-18  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
 489
 490         * config/aarch64/aarch64-simd.md (vec_set<mode>): Use VALL_F16 mode
 491         iterator.  Delete separate integer-mode vec_set<mode> expander.
 492         (aarch64_simd_vec_setv2di): Delete.
 493         (vec_setv2di): Delete.
 494         (aarch64_simd_vec_set<mode>): Delete all other patterns with that name.
 495         Use VALL_F16 mode iterator.  Add LD1 alternative and use vwcore for
 496         the "w, r" alternative.
 497
 498 2018-05-18  Martin Liska  <mliska@suse.cz>
 499
 500         * passes.def: Add pass_lower_switch and pass_lower_switch_O0.
 501         * tree-pass.h (make_pass_lower_switch_O0): New function.
 502         * tree-switch-conversion.c (node_has_low_bound): Remove.
 503         (node_has_high_bound): Likewise.
 504         (node_is_bounded): Likewise.
 505         (class pass_lower_switch): Make it a template type and create
 506         two instances.
 507         (pass_lower_switch::execute): Add template argument.
 508         (make_pass_lower_switch): New function.
 509         (make_pass_lower_switch_O0): New function.
 510         (do_jump_if_equal): Remove.
 511         (emit_case_nodes): Simplify to just handle all 3 cases and leave
 512         all the hard work to tree optimization passes.
 513
 514 2018-05-18  Martin Liska  <mliska@suse.cz>
 515
 516         * dbgcnt.c (limit_low): Renamed from limit.
 517         (limit_high): New variable.
 518         (dbg_cnt_is_enabled): Check for upper limit.
 519         (dbg_cnt): Adjust dumping.
 520         (dbg_cnt_set_limit_by_index): Add new argument for high
 521         value.
 522         (dbg_cnt_set_limit_by_name): Likewise.
 523         (dbg_cnt_process_single_pair): Parse new format.
 524         (dbg_cnt_process_opt): Use strtok.
 525         (dbg_cnt_list_all_counters): Remove 'value' and add
 526         'limit_high'.
 527         * doc/invoke.texi: Document changes.
 528
 529 2018-05-18  Richard Sandiford  <richard.sandiford@linaro.org>
 530
 531         * doc/sourcebuild.texi (scalar_all_fma): Document.
 532         * tree.def (FMA_EXPR): Delete.
 533         * internal-fn.def (FMA, FMS, FNMA, FNMS): New internal functions.
 534         * internal-fn.c (ternary_direct): New macro.
 535         (expand_ternary_optab_fn): Likewise.
 536         (direct_ternary_optab_supported_p): Likewise.
 537         * Makefile.in (build/genmatch.o): Depend on case-fn-macros.h.
 538         * builtins.c (fold_builtin_fma): Delete.
 539         (fold_builtin_3): Don't call it.
 540         * cfgexpand.c (expand_debug_expr): Remove FMA_EXPR handling.
 541         * expr.c (expand_expr_real_2): Likewise.
 542         * fold-const.c (operand_equal_p): Likewise.
 543         (fold_ternary_loc): Likewise.
 544         * gimple-pretty-print.c (dump_ternary_rhs): Likewise.
 545         * gimple.c (DEFTREECODE): Likewise.
 546         * gimplify.c (gimplify_expr): Likewise.
 547         * optabs-tree.c (optab_for_tree_code): Likewise.
 548         * tree-cfg.c (verify_gimple_assign_ternary): Likewise.
 549         * tree-eh.c (operation_could_trap_p): Likewise.
 550         (stmt_could_throw_1_p): Likewise.
 551         * tree-inline.c (estimate_operator_cost): Likewise.
 552         * tree-pretty-print.c (dump_generic_node): Likewise.
 553         (op_code_prio): Likewise.
 554         * tree-ssa-loop-im.c (stmt_cost): Likewise.
 555         * tree-ssa-operands.c (get_expr_operands): Likewise.
 556         * tree.c (commutative_ternary_tree_code, add_expr): Likewise.
 557         * fold-const-call.h (fold_fma): Delete.
 558         * fold-const-call.c (fold_const_call_ssss): Handle CFN_FMS,
 559         CFN_FNMA and CFN_FNMS.
 560         (fold_fma): Delete.
 561         * genmatch.c (combined_fn): New enum.
 562         (commutative_ternary_tree_code): Remove FMA_EXPR handling.
 563         (commutative_op): New function.
 564         (commutate): Use it.  Handle more than 2 operands.
 565         (dt_operand::gen_gimple_expr): Use commutative_op.
 566         (parser::parse_expr): Allow :c to be used with non-binary
 567         operators if the commutative operand is known.
 568         * gimple-ssa-backprop.c (backprop::process_builtin_call_use): Handle
 569         CFN_FMS, CFN_FNMA and CFN_FNMS.
 570         (backprop::process_assign_use): Remove FMA_EXPR handling.
 571         * hsa-gen.c (gen_hsa_insns_for_operation_assignment): Likewise.
 572         (gen_hsa_fma): New function.
 573         (gen_hsa_insn_for_internal_fn_call): Use it for IFN_FMA, IFN_FMS,
 574         IFN_FNMA and IFN_FNMS.
 575         * match.pd: Add folds for IFN_FMS, IFN_FNMA and IFN_FNMS.
 576         * gimple-fold.h (follow_all_ssa_edges): Declare.
 577         * gimple-fold.c (follow_all_ssa_edges): New function.
 578         * tree-ssa-math-opts.c (convert_mult_to_fma_1): Use the
 579         gimple_build interface and use follow_all_ssa_edges to fold the result.
 580         (convert_mult_to_fma): Use direct_internal_fn_suppoerted_p
 581         instead of checking for optabs directly.
 582         * config/i386/i386.c (ix86_add_stmt_cost): Recognize FMAs as calls
 583         rather than FMA_EXPRs.
 584         * config/rs6000/rs6000.c (rs6000_gimple_fold_builtin): Create a
 585         call to IFN_FMA instead of an FMA_EXPR.
 586
 587 2018-05-17  Jim Wilson  <jimw@sifive.com>
 588
 589         * expr.c (do_tablejump): When converting index to Pmode, if we have a
 590         sign extended promoted subreg, and the range does not have the sign bit
 591         set, then do a sign extend.
 592
 593         * config/riscv/riscv.c (riscv_extend_comparands): In unsigned QImode
 594         test, check for sign extended subreg and/or constant operands, and
 595         do a sign extend in that case.
 596
 597 2018-05-17  Steve Ellcey  <sellcey@cavium.com>
 598
 599         * config/aarch64/thunderx2t99.md (thunderx2t99_ls_both): Delete.
 600         (thunderx2t99_multiple): Delete psuedo-units from used cpus.
 601         Add untyped.
 602         (thunderx2t99_alu_shift): Remove alu_shift_reg, alus_shift_reg.
 603         Change logics_shift_reg to logics_shift_imm.
 604         (thunderx2t99_fp_loadpair_basic): Delete.
 605         (thunderx2t99_fp_storepair_basic): Delete.
 606         (thunderx2t99_asimd_int): Add neon_sub and neon_sub_q types.
 607         (thunderx2t99_asimd_polynomial): Delete.
 608         (thunderx2t99_asimd_fp_simple): Add neon_fp_mul_s_scalar_q
 609         and neon_fp_mul_d_scalar_q.
 610         (thunderx2t99_asimd_fp_conv): Add *int_to_fp* types.
 611         (thunderx2t99_asimd_misc): Delete neon_dup and neon_dup_q.
 612         (thunderx2t99_asimd_recip_step): Add missing *sqrt* types.
 613         (thunderx2t99_asimd_lut): Add missing tbl types.
 614         (thunderx2t99_asimd_ext): Delete.
 615         (thunderx2t99_asimd_load1_1_mult): Delete.
 616         (thunderx2t99_asimd_load1_2_mult): Delete.
 617         (thunderx2t99_asimd_load1_ldp): New.
 618         (thunderx2t99_asimd_load1): New.
 619         (thunderx2t99_asimd_load2): Add missing *load2* types.
 620         (thunderx2t99_asimd_load3): New.
 621         (thunderx2t99_asimd_load4): New.
 622         (thunderx2t99_asimd_store1_1_mult): Delete.
 623         (thunderx2t99_asimd_store1_2_mult): Delete.
 624         (thunderx2t99_asimd_store2_mult): Delete.
 625         (thunderx2t99_asimd_store2_onelane): Delete.
 626         (thunderx2t99_asimd_store_stp): New.
 627         (thunderx2t99_asimd_store1): New.
 628         (thunderx2t99_asimd_store2): New.
 629         (thunderx2t99_asimd_store3): New.
 630         (thunderx2t99_asimd_store4): New.
 631
 632 2018-05-17  Jerome Lambourg  <lambourg@adacore.com>
 633
 634         * config/arm/arm_cmse.h (cmse_nsfptr_create, cmse_is_nsfptr): Remove
 635         #include <stdint.h>.  Replace intptr_t with __INTPTR_TYPE__.
 636
 637 2018-05-17  Pat Haugen  <pthaugen@us.ibm.com>
 638             Segher Boessenkool  <segher@kernel.crashing.org>
 639
 640         PR target/85698
 641         * config/rs6000/rs6000.c (rs6000_output_move_128bit): Check dest
 642         operand.
 643
 644 2018-05-17  Richard Biener  <rguenther@suse.de>
 645
 646         * tree-ssa-dse.c (dse_classify_store): Fix iterator increment
 647         for pruning loop and prune defs feeding only already visited PHIs.
 648
 649 2018-05-17  Richard Biener  <rguenther@suse.de>
 650
 651         * tree-ssa-sccvn.c (vn_reference_lookup_3): Improve memset handling.
 652
 653 2018-05-17  Bin Cheng  <bin.cheng@arm.com>
 654             Richard Biener  <rguenther@suse.de>
 655
 656         PR tree-optimization/85793
 657         * tree-vect-stmts.c (vectorizable_load): Handle 1 element-wise load
 658         for VMAT_ELEMENTWISE.
 659
 660 2018-05-17  Richard Sandiford  <richard.sandiford@linaro.org>
 661
 662         * internal-fn.h (lookup_internal_fn): Declare
 663         * internal-fn.c (lookup_internal_fn): New function.
 664         * gimple.c (gimple_build_call_from_tree): Handle calls to
 665         internal functions.
 666         * gimple-pretty-print.c (dump_gimple_call): Print "." before
 667         internal function names.
 668         * tree-pretty-print.c (dump_generic_node): Likewise.
 669         * tree-ssa-scopedtables.c (expr_hash_elt::print): Likewise.
 670
 671 2018-05-17  Richard Sandiford  <richard.sandiford@linaro.org>
 672
 673         * gimple-fold.h (gimple_build): Make the function forms take
 674         combined_fn rather than built_in_function.
 675         (gimple_simplify): Likewise.
 676         * gimple-match-head.c (gimple_simplify): Likewise.
 677         * gimple-fold.c (gimple_build): Likewise.
 678         * tree-vect-loop.c (get_initial_def_for_reduction): Use gimple_build
 679         rather than gimple_build_call_internal.
 680         (get_initial_defs_for_reduction): Likewise.
 681         (vect_create_epilog_for_reduction): Likewise.
 682         (vectorizable_live_operation): Likewise.
 683
 684 2018-05-17  Martin Liska  <mliska@suse.cz>
 685
 686         * gimple-ssa-sprintf.c (format_directive): Do not use
 687         space in between 'G_' and '('.
 688
 689 2018-05-17  Jakub Jelinek  <jakub@redhat.com>
 690
 691         PR target/85323
 692         * config/i386/i386.c (ix86_fold_builtin): Handle masked shifts
 693         even if the mask is not all ones.
 694
 695         PR target/85323
 696         * config/i386/i386.c (ix86_fold_builtin): Fold shift builtins by
 697         vector.
 698         (ix86_gimple_fold_builtin): Likewise.
 699
 700         PR target/85323
 701         * config/i386/i386.c: Include tree-vector-builder.h.
 702         (ix86_vector_shift_count): New function.
 703         (ix86_fold_builtin): Fold shift builtins by scalar count.
 704         (ix86_gimple_fold_builtin): Likewise.
 705
 706         * config/i386/avx512fintrin.h (_mm512_set_epi16, _mm512_set_epi8,
 707         _mm512_setzero): New intrinsics.
 708
 709 2018-05-17  James Greenhalgh  <james.greenhalgh@arm.com>
 710             Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
 711
 712         * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify
 713         code generation for cases where splatting a value is not useful.
 714         * simplify-rtx.c (simplify_ternary_operation): Simplify
 715         vec_merge across a vec_duplicate and a paradoxical subreg forming a vector
 716         mode to a vec_concat.
 717
 718 2018-05-17  Olga Makhotina  <olga.makhotina@intel.com>
 719
 720         * config.gcc: Support "goldmont-plus".
 721         * config/i386/driver-i386.c (host_detect_local_cpu): Detect
 722         "goldmont-plus".
 723         * config/i386/i386-c.c (ix86_target_macros_internal): Handle
 724         PROCESSOR_GOLDMONT_PLUS.
 725         * config/i386/i386.c (m_GOLDMONT_PLUS): Define.
 726         (processor_target_table): Add "goldmont-plus".
 727         (PTA_GOLDMONT_PLUS): Define.
 728         (ix86_lea_outperforms): Add TARGET_GOLDMONT_PLUS.
 729         (get_builtin_code_for_version): Handle PROCESSOR_GOLDMONT_PLUS.
 730         (fold_builtin_cpu): Add M_INTEL_GOLDMONT_PLUS.
 731         (fold_builtin_cpu): Add "goldmont-plus".
 732         (ix86_add_stmt_cost): Add TARGET_GOLDMONT_PLUS.
 733         (ix86_option_override_internal): Add "goldmont-plus".
 734         * config/i386/i386.h (processor_costs): Define TARGET_GOLDMONT_PLUS.
 735         (processor_type): Add PROCESSOR_GOLDMONT_PLUS.
 736         * config/i386/x86-tune.def: Add m_GOLDMONT_PLUS.
 737         * doc/invoke.texi: Add goldmont-plus as x86 -march=/-mtune= CPU type.
 738
 739 2018-05-17  Richard Biener  <rguenther@suse.de>
 740
 741         PR tree-optimization/85757
 742         * tree-ssa-dse.c (dse_classify_store): Record a PHI def and
 743         remove defs that only feed that PHI from further processing.
 744
 745 2018-05-16  Jim Wilson  <jimw@sifive.com>
 746
 747         * config/riscv/riscv.md (<optab>si3_mask, <optab>si3_mask_1): Prepend
 748         asterisk to name.
 749         (<optab>di3_mask, <optab>di3_mask_1): Likewise.
 750
 751 2018-05-16  Mark Wielaard  <mark@klomp.org>
 752
 753         * dwarf2out.c (count_index_strings): New function.
 754         (output_indirect_strings): Call count_index_strings and generate
 755         header for dwarf_version >= 5.
 756
 757 2018-05-16  Mark Wielaard  <mark@klomp.org>
 758
 759         * dwarf2out.c (dwarf_FORM): New function.
 760         (set_indirect_string): Use dwarf_FORM.
 761         (reset_indirect_string): Likewise.
 762         (size_of_die): Likewise.
 763         (value_format): Likewise.
 764         (output_die): Likewise.
 765         (add_skeleton_AT_string): Likewise.
 766         (output_macinfo_op): Likewise.
 767         (index_string): Likewise.
 768         (output_index_string_offset): Likewise.
 769         (output_index_string): Likewise.
 770         (count_index_strings): Likewise.
 771
 772 2018-05-16  Carl Love  <cel@us.ibm.com>
 773
 774         * config/rs6000/rs6000.md (prefetch): Generate ISA 2.06 instructions
 775         dcbt and dcbtstt with TH=16 if operands[2] is 0 and Power 8 or newer.
 776
 777 2018-05-16  Martin Jambor  <mjambor@suse.cz>
 778
 779         * ipa-prop.c (ipa_free_all_edge_args): Remove.
 780         * ipa-prop.h (ipa_free_all_edge_args): Likewise.
 781
 782 2018-05-16  Wilco Dijkstra  <wdijkstr@arm.com>
 783
 784         * config/aarch64/aarch64.md (fma<mode>4): Change into expand pattern.
 785         (fnma<mode>4): Likewise.
 786         (fms<mode>4): Likewise.
 787         (fnms<mode>4): Likewise.
 788         (aarch64_fma<mode>4): Rename insn, reorder accumulator operand.
 789         (aarch64_fnma<mode>4): Likewise.
 790         (aarch64_fms<mode>4): Likewise.
 791         (aarch64_fnms<mode>4): Likewise.
 792         (aarch64_fnmadd<mode>4): Likewise.
 793
 794 2018-05-16  Jason Merrill  <jason@redhat.com>
 795
 796         * tree.c (warn_deprecated_use): Return bool.  Simplify logic.
 797
 798 2018-05-16  Richard Biener  <rguenther@suse.de>
 799
 800         * tree-vectorizer.h (struct stmt_info_for_cost): Add where member.
 801         (dump_stmt_cost): Declare.
 802         (add_stmt_cost): Dump cost we add.
 803         (add_stmt_costs): New function.
 804         (vect_model_simple_cost, vect_model_store_cost, vect_model_load_cost):
 805         No longer exported.
 806         (vect_analyze_stmt): Adjust prototype.
 807         (vectorizable_condition): Likewise.
 808         (vectorizable_live_operation): Likewise.
 809         (vectorizable_reduction): Likewise.
 810         (vectorizable_induction): Likewise.
 811         * tree-vect-loop.c (vect_analyze_loop_operations): Create local
 812         cost vector to pass to vectorizable_ and record afterwards.
 813         (vect_model_reduction_cost): Take cost vector argument and adjust.
 814         (vect_model_induction_cost): Likewise.
 815         (vectorizable_reduction): Likewise.
 816         (vectorizable_induction): Likewise.
 817         (vectorizable_live_operation): Likewise.
 818         * tree-vect-slp.c (vect_create_new_slp_node): Initialize
 819         SLP_TREE_NUMBER_OF_VEC_STMTS.
 820         (vect_analyze_slp_cost_1): Remove.
 821         (vect_analyze_slp_cost): Likewise.
 822         (vect_slp_analyze_node_operations): Take visited args and
 823         a target cost vector.  Avoid processing already visited stmt sets.
 824         (vect_slp_analyze_operations): Use a local cost vector to gather
 825         costs and register those of non-discarded instances.
 826         (vect_bb_vectorization_profitable_p): Use add_stmt_costs.
 827         (vect_schedule_slp_instance): Remove copying of
 828         SLP_TREE_NUMBER_OF_VEC_STMTS.  Instead assert that it is not
 829         zero.
 830         * tree-vect-stmts.c (record_stmt_cost): Remove path directly
 831         adding cost.  Record cost entry location.
 832         (vect_prologue_cost_for_slp_op): Function to compute cost of
 833         a constant or invariant generated for SLP vect in the prologue,
 834         split out from vect_analyze_slp_cost_1.
 835         (vect_model_simple_cost): Make static.  Adjust for SLP costing.
 836         (vect_model_promotion_demotion_cost): Likewise.
 837         (vect_model_store_cost): Likewise, make static.
 838         (vect_model_load_cost): Likewise.
 839         (vectorizable_bswap): Add cost vector arg and adjust.
 840         (vectorizable_call): Likewise.
 841         (vectorizable_simd_clone_call): Likewise.
 842         (vectorizable_conversion): Likewise.
 843         (vectorizable_assignment): Likewise.
 844         (vectorizable_shift): Likewise.
 845         (vectorizable_operation): Likewise.
 846         (vectorizable_store): Likewise.
 847         (vectorizable_load): Likewise.
 848         (vectorizable_condition): Likewise.
 849         (vectorizable_comparison): Likewise.
 850         (can_vectorize_live_stmts): Likewise.
 851         (vect_analyze_stmt): Likewise.
 852         (vect_transform_stmt): Adjust calls to vectorizable_*.
 853         * tree-vectorizer.c: Include gimple-pretty-print.h.
 854         (dump_stmt_cost): New function.
 855
 856 2018-05-16  Richard Biener  <rguenther@suse.de>
 857
 858         * params.def (PARAM_DSE_MAX_ALIAS_QUERIES_PER_STORE): New param.
 859         * doc/invoke.texi (dse-max-alias-queries-per-store): Document.
 860         * tree-ssa-dse.c: Include tree-ssa-loop.h.
 861         (check_name): New callback.
 862         (dse_classify_store): Track cycles via a visited bitmap of PHI
 863         defs and simplify handling of in-loop and across loop dead stores
 864         and properly fail for loop-variant refs.  Handle byte-tracking with
 865         multiple defs.  Use PARAM_DSE_MAX_ALIAS_QUERIES_PER_STORE for
 866         limiting the walk.
 867
 868 2018-05-16  Richard Sandiford  <richard.sandiford@linaro.org>
 869
 870         * tree-vectorizer.h (vect_get_vector_types_for_stmt): Declare.
 871         (vect_get_mask_type_for_stmt): Likewise.
 872         * tree-vect-slp.c (vect_two_operations_perm_ok_p): New function,
 873         split out from...
 874         (vect_build_slp_tree_1): ...here.  Use vect_get_vector_types_for_stmt
 875         to determine the statement's vector type and the vector type that
 876         should be used for calculating nunits.  Deal with cases in which
 877         the type has to be deferred.
 878         (vect_slp_analyze_node_operations): Use vect_get_vector_types_for_stmt
 879         and vect_get_mask_type_for_stmt to calculate STMT_VINFO_VECTYPE.
 880         * tree-vect-loop.c (vect_determine_vf_for_stmt_1)
 881         (vect_determine_vf_for_stmt): New functions, split out from...
 882         (vect_determine_vectorization_factor): ...here.
 883         * tree-vect-stmts.c (vect_get_vector_types_for_stmt)
 884         (vect_get_mask_type_for_stmt): New functions, split out from
 885         vect_determine_vectorization_factor.
 886
 887 2018-05-16  Richard Biener  <rguenther@suse.de>
 888
 889         * tree-cfg.c (verify_gimple_assign_ternary): Properly
 890         verify the [VEC_]COND_EXPR embedded comparison.
 891
 892 2018-05-15  Martin Sebor  <msebor@redhat.com>
 893
 894         PR tree-optimization/85753
 895         * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Handle
 896         RECORD_TYPE in addition to ARRAY_TYPE.
 897
 898 2018-05-15  Martin Sebor  <msebor@redhat.com>
 899
 900         PR middle-end/85643
 901         * calls.c (get_attr_nonstring_decl): Handle MEM_REF.
 902
 903 2018-05-15  Richard Biener  <rguenther@suse.de>
 904
 905         * tree-ssa-dse.c (dse_classify_store): Remove use_stmt parameter,
 906         add by_clobber_p one.  Change algorithm to collect all defs
 907         representing uses we need to walk and try reducing them to
 908         a single one before failing.
 909         (dse_dom_walker::dse_optimize_stmt): Adjust.
 910
 911 2018-05-13  Mark Wielaard  <mark@klomp.org>
 912
 913         * dwarf2out.c (dwarf_OP): Handle DW_OP_addrx and DW_OP_constx.
 914         (size_of_loc_descr): Likewise.
 915         (output_loc_operands): Likewise.
 916         (output_loc_operands_raw): Likewise.
 917         (dw_addr_op): Use dwarf_OP () for DW_OP_constx and DW_OP_addrx.
 918         (resolve_addr_in_expr): Handle DW_OP_addrx and DW_OP_constx.
 919         (hash_loc_operands): Likewise.
 920         (compare_loc_operands): Likewise.
 921
 922 2018-05-14  Mark Wielaard  <mark@klomp.org>
 923
 924         * dwarf2out.c (count_index_addrs): New function.
 925         (dwarf2out_finish): Use count_index_addrs to calculate addrs_length.
 926
 927 2018-05-15  Prathamesh Kulkarni  <prathamesh.kulkarni@linaro.org>
 928
 929         PR tree-optimization/83648
 930         * ipa-pure-const.c (malloc_candidate_p): Allow function with NULL
 931         return value as malloc candidate.
 932
 933 2018-05-15  Prathamesh Kulkarni  <prathamesh.kulkarni@linaro.org>
 934
 935         PR ipa/85734
 936         * ipa-pure-const.c (warn_function_malloc): Pass value of known_finite param
 937         as true in call to suggest_attribute.
 938
 939 2018-05-14  Segher Boessenkool  <segher@kernel.crashing.org>
 940
 941         * doc/invoke.texi (RS/6000 and PowerPC Options): Delete @itemx for
 942         -mreadonly-in-sdata.
 943
 944 2018-05-14  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
 945
 946         * config/aarch64/aarch64-simd.md (*aarch64_crypto_aese_fused):
 947         New pattern.
 948         (aarch64_crypto_aesd_fused): Likewise.
 949
 950 2018-05-14  Wilco Dijkstra  <wdijkstr@arm.com>
 951
 952         * config/aarch64/aarch64.md (mov<mode>): Remove '*' in alternatives.
 953         (movsi_aarch64): Likewise.
 954         (load_pairsi): Likewise.
 955         (load_pairdi): Likewise.
 956         (store_pairsi): Likewise.
 957         (store_pairdi): Likewise.
 958         (load_pairsf): Likewise.
 959         (load_pairdf): Likewise.
 960         (store_pairsf): Likewise.
 961         (store_pairdf): Likewise.
 962         (zero_extend): Likewise.
 963         (trunc): Swap alternatives.
 964         (fcvt_target): Add '?' to prefer w over r.
 965
 966 2018-05-14  Jakub Jelinek  <jakub@redhat.com>
 967
 968         PR target/85756
 969         * config/i386/i386.md: Disallow non-commutative arithmetics in
 970         last twpeephole for mem {+,-,&,|,^}= x; mem != 0 after cmpelim
 971         optimization.  Use COMMUTATIVE_ARITH_P test rather than != MINUS
 972         in the peephole2 before it.
 973
 974 2018-05-14  Sebastian Peryt  <sebastian.peryt@intel.com>
 975
 976         * common/config/i386/i386-common.c (OPTION_MASK_ISA_CLDEMOTE_SET,
 977         OPTION_MASK_ISA_CLDEMOTE_UNSET): New defines.
 978         (ix86_handle_option): Handle -mcldemote.
 979         * config.gcc: New header.
 980         * config/i386/cldemoteintrin.h: New file.
 981         * config/i386/cpuid.h (bit_CLDEMOTE): New bit.
 982         * config/i386/driver-i386.c (host_detect_local_cpu): Detect
 983         -mcldemote.
 984         * config/i386/i386-c.c (ix86_target_macros_internal): Handle
 985         OPTION_MASK_ISA_CLDEMOTE.
 986         * config/i386/i386.c (ix86_target_string): Add -mcldemote.
 987         (ix86_valid_target_attribute_inner_p): Ditto.
 988         (enum ix86_builtins): Add IX86_BUILTIN_CLDEMOTE.
 989         (ix86_init_mmx_sse_builtins): Define __builtin_ia32_cldemote.
 990         (ix86_expand_builtin): Expand IX86_BUILTIN_CLDEMOTE.
 991         * config/i386/i386.h (TARGET_CLDEMOTE, TARGET_CLDEMOTE_P): New.
 992         * config/i386/i386.md (UNSPECV_CLDEMOTE): New.
 993         (cldemote): New.
 994         * config/i386/i386.opt: Add -mcldemote.
 995         * config/i386/x86intrin.h: New header.
 996         * doc/invoke.texi: Add -mcldemote.
 997
 998 2018-05-14  Richard Biener  <rguenther@suse.de>
 999
1000         * doc/match-and-simplify.texi: Adjust :s documentation.
1001
1002 2018-05-14  Alexander Monakov  <amonakov@ispras.ru>
1003
1004         * sort.cc (REORDER_23): Pass the type for the temporaries instead of
1005         intended memcpy size.
1006         (REORDER_45): Likewise.
1007
1008 2018-05-13  Alexander Monakov  <amonakov@ispras.ru>
1009
1010         * sort.cc: New file.
1011         * system.h [!CHECKING_P] (qsort): Redirect to gcc_qsort.
1012         * vec.c (qsort_chk): Use gcc_qsort.
1013         * Makefile.in (OBJS-libcommon): Add sort.o.
1014         (build/sort.o): New target.  Use it...
1015         (BUILD_RTL): ... here, and...
1016         (build/gencfn-macros): ... here, and...
1017         (build/genmatch): ... here.
1018
1019 2018-05-13  Kito Cheng  <kito.cheng@gmail.com>
1020             Chung-Ju Wu  <jasonwucj@gmail.com>
1021
1022         * config.gcc (nds32*-*-*): Check that n15 is valid to --with-cpu.
1023         * config/nds32/nds32-graywolf.md: New file.
1024         * config/nds32/nds32-opts.h (nds32_cpu_type): Add CPU_GRAYWOLF.
1025         * config/nds32/nds32-pipelines-auxiliary.c: Implementation for n15
1026         pipeline.
1027         * config/nds32/nds32-protos.h: More declarations for n15 pipeline.
1028         * config/nds32/nds32-utils.c: More implementations for n15 pipeline.
1029         * config/nds32/nds32.md (pipeline_model): Add graywolf.
1030         * config/nds32/nds32.opt (mcpu): Support n15 pipeline cpus.
1031         * config/nds32/pipelines.md: Include n15 settings.
1032
1033 2018-05-13  Kito Cheng  <kito.cheng@gmail.com>
1034             Chung-Ju Wu  <jasonwucj@gmail.com>
1035
1036         * config.gcc (nds32*-*-*): Check that n12/n13 are valid to --with-cpu.
1037         * config/nds32/nds32-n13.md: New file.
1038         * config/nds32/nds32-opts.h (nds32_cpu_type): Add CPU_N12 and CPU_N13.
1039         * config/nds32/nds32-pipelines-auxiliary.c: Implementation for n12/n13
1040         pipeline.
1041         * config/nds32/nds32-protos.h: More declarations for n12/n13 pipeline.
1042         * config/nds32/nds32.md (pipeline_model): Add n13.
1043         * config/nds32/nds32.opt (mcpu): Support n12/n13 pipeline cpus.
1044         * config/nds32/pipelines.md: Include n13 settings.
1045
1046 2018-05-13  Kito Cheng  <kito.cheng@gmail.com>
1047             Chung-Ju Wu  <jasonwucj@gmail.com>
1048
1049         * config.gcc (nds32*-*-*): Check that n10/d10 are valid to --with-cpu.
1050         * config/nds32/nds32-n10.md: New file.
1051         * config/nds32/nds32-opts.h (nds32_cpu_type): Add CPU_N10.
1052         * config/nds32/nds32-pipelines-auxiliary.c: Implementation for n10
1053         pipeline.
1054         * config/nds32/nds32-protos.h: More declarations for n10 pipeline.
1055         * config/nds32/nds32-utils.c: More implementations for n10 pipeline.
1056         * config/nds32/nds32.md (pipeline_model): Add n10.
1057         * config/nds32/nds32.opt (mcpu): Support n10 pipeline cpus.
1058         * config/nds32/pipelines.md: Include n10 settings.
1059
1060 2018-05-13  Monk Chiang  <sh.chiang04@gmail.com>
1061             Kito Cheng  <kito.cheng@gmail.com>
1062             Chung-Ju Wu  <jasonwucj@gmail.com>
1063
1064         * config.gcc (nds32be-*-*): Handle --with-ext-dsp.
1065         * config/nds32/constants.md (unspec_element, unspec_volatile_element):
1066         Add enum values for DSP extension instructions.
1067         * config/nds32/constraints.md (Iu06, IU06, CVp5, CVs5, CVs2, CVhi):
1068         New constraints.
1069         * config/nds32/iterators.md (shifts, shiftrt, sat_plus, all_plus,
1070         sat_minus, all_minus, plus_minus, extend, sumax, sumin, sumin_max):
1071         New code iterators.
1072         (su, zs, uk, opcode, add_rsub, add_sub): New code attributes.
1073         * config/nds32/nds32-dspext.md: New file for DSP implementation.
1074         * config/nds32/nds32-intrinsic.c: Implementation of DSP extension.
1075         * config/nds32/nds32-intrinsic.md: Likewise.
1076         * config/nds32/nds32_intrinsic.h: Likewise.
1077         * config/nds32/nds32-md-auxiliary.c: Likewise.
1078         * config/nds32/nds32-memory-manipulation.c: Consider DSP extension.
1079         * config/nds32/nds32-predicates.c (const_vector_to_hwint): New.
1080         (nds32_valid_CVp5_p, nds32_valid_CVs5_p): New.
1081         (nds32_valid_CVs2_p, nds32_valid_CVhi_p): New.
1082         * config/nds32/nds32-protos.h: New declarations for DSP extension.
1083         * config/nds32/nds32-utils.c (extract_mac_non_acc_rtx): New case
1084         TYPE_DMAC in switch statement.
1085         * config/nds32/nds32.c: New checking and implementation for DSP
1086         extension instructions.
1087         * config/nds32/nds32.h: Likewise.
1088         * config/nds32/nds32.md: Likewise.
1089         * config/nds32/nds32.opt (mhw-abs, mext-dsp): New options.
1090         * config/nds32/predicates.md: Implement new predicates for DSP
1091         extension.
1092
1093 2018-05-11  Michael Meissner  <meissner@linux.vnet.ibm.com>
1094
1095         * config/rs6000/rs6000.md (mov<mode>_softfloat, FMOVE32):
1096         Reformat alternatives and attributes so it is easier to identify
1097         which constraints/attributes go with which instruction.
1098         (mov<mode>_hardfloat32, FMOVE64): Likewise.
1099         (mov<mode>_softfloat32, FMOVE64): Likewise.
1100         (mov<mode>_hardfloat64, FMOVE64): Likewise.
1101         (mov<mode>_softfloat64, FMOVE64): Likewise.
1102
1103 2018-05-11  Kelvin Nilsen  <kelvin@gcc.gnu.org>
1104
1105         * doc/extend.texi (PowerPC Built-in Functions): Rename this
1106         subsection.
1107         (Basic PowerPC Built-in Functions): The new name of the
1108         subsection previously known as "PowerPC Built-in Functions".
1109         (Basic PowerPC Built-in Functions Available on all Configurations):
1110         New subsubsection.
1111         (Basic PowerPC Built-in Functions Available on ISA 2.05): Likewise.
1112         (Basic PowerPC Built-in Functions Available on ISA 2.06): Likewise.
1113         (Basic PowerPC Built-in Functions Available on ISA 2.07): Likewise.
1114         (Basic PowerPC Built-in Functions Available on ISA 3.0): Likewise.
1115
1116 2018-05-11  Martin Jambor  <mjambor@suse.cz>
1117
1118         PR ipa/85655
1119         * ipa-cp.c (intersect_with_plats): Check that the lattice contains
1120         single const.
1121
1122 2018-05-11  Richard Earnshaw  <rearnsha@arm.com>
1123
1124         PR target/85733
1125         * config/arm/arm-cpus.in (fgroup ARMv6m): Add be8 feature.
1126
1127 2018-05-11  Sebastian Peryt  <sebastian.peryt@intel.com>
1128
1129         * common/config/i386/i386-common.c (OPTION_MASK_ISA_WAITPKG_SET,
1130         OPTION_MASK_ISA_WAITPKG_UNSET): New defines.
1131         (ix86_handle_option): Handle -mwaitpkg.
1132         * config.gcc: New header.
1133         * config/i386/cpuid.h (bit_WAITPKG): New bit.
1134         * config/i386/driver-i386.c (host_detect_local_cpu): Detect -mwaitpkg.
1135         * config/i386/i386-builtin-types.def ((UINT8, UNSIGNED, UINT64)): New
1136         function type.
1137         * config/i386/i386-c.c (ix86_target_macros_internal): Handle
1138         OPTION_MASK_ISA_WAITPKG.
1139         * config/i386/i386.c (ix86_target_string): Add -mwaitpkg.
1140         (ix86_option_override_internal): Add PTA_WAITPKG.
1141         (ix86_valid_target_attribute_inner_p): Add -mwaitpkg.
1142         (enum ix86_builtins): Add IX86_BUILTIN_UMONITOR, IX86_BUILTIN_UMWAIT,
1143         IX86_BUILTIN_TPAUSE.
1144         (ix86_init_mmx_sse_builtins): Define __builtin_ia32_umonitor,
1145         __builtin_ia32_umwait and __builtin_ia32_tpause.
1146         (ix86_expand_builtin): Expand IX86_BUILTIN_UMONITOR,
1147         IX86_BUILTIN_UMWAIT, IX86_BUILTIN_TPAUSE.
1148         * config/i386/i386.h (TARGET_WAITPKG, TARGET_WAITPKG_P): New.
1149         * config/i386/i386.md (UNSPECV_UMWAIT, UNSPECV_UMONITOR,
1150         UNSPECV_TPAUSE): New.
1151         (umwait, umwait_rex64, umonitor_<mode>, tpause, tpause_rex64): New.
1152         * config/i386/i386.opt: Add -mwaitpkg.
1153         * config/i386/waitpkgintrin.h: New file.
1154         * config/i386/x86intrin.h: New header.
1155         * doc/invoke.texi: Add -mwaitpkg.
1156
1157 2018-05-11  Richard Earnshaw  <rearnsha@arm.com>
1158
1159         PR target/85606
1160         * config/arm/arm-cpus.in: Add comment that ARMv6-m and ARMv6S-m are now
1161         equivalent.
1162         (cortex-m0): Use armv6s-m isa.
1163         (cortex-m0plus): Likewise.
1164         (cortex-m1): Likewise.
1165         (cortex-m0.small-multiply): Likewise.
1166         (cortex-m0plus.small-multiply): Likewise.
1167         (cortex-m1.small-multiply): Likewise.
1168
1169 2018-05-11  Allan Sandfeld Jensen  <allan.jensen@qt.io>
1170             Jakub Jelinek  <jakub@redhat.com>
1171
1172         PR tree-optimization/85692
1173         * tree-ssa-forwprop.c (simplify_vector_constructor): Try two
1174         source permute as well.
1175
1176 2018-05-11  Martin Liska  <mliska@suse.cz>
1177
1178         PR sanitizer/85556
1179         * doc/extend.texi: Document LLVM style format for no_sanitize
1180         attribute.
1181
1182 2018-05-10  Michael Meissner  <meissner@linux.vnet.ibm.com>
1183
1184         * config/rs6000/rs6000.c (mode_supports_dq_form): Rename
1185         mode_supports_vsx_dform_quad to mode_supports_dq_form.
1186         (mode_supports_vsx_dform_quad): Likewise.
1187         (mode_supports_vmx_dform): Move these functions to be next to the
1188         other mode_supports functions.
1189         (mode_supports_dq_form): Likewise.
1190         (quad_address_p): Change calls of mode_supports_vsx_dform_quad to
1191         mode_supports_dq_form.
1192         (reg_offset_addressing_ok_p): Likewise.
1193         (offsettable_ok_by_alignment): Likewise.
1194         (rs6000_legitimate_offset_address_p): Likewise.
1195         (legitimate_lo_sum_address_p): Likewise.
1196         (rs6000_legitimize_address): Likewise.
1197         (rs6000_legitimize_reload_address): Likewise.
1198         (rs6000_secondary_reload_inner): Likewise.
1199         (rs6000_preferred_reload_class): Likewise.
1200         (rs6000_output_move_128bit): Likewise.
1201
1202 2018-05-10  Uros Bizjak  <ubizjak@gmail.com>
1203
1204         * config/i386/i386.c (ix86_expand_builtin) <case IX86_BUILTIN_RDPID>:
1205         Generate SImode target register for null target.
1206         <case IX86_BUILTIN_XGETBV>: Ditto.
1207         <case IX86_BUILTIN_XSETBV>: Optimize LSHIFTRT generation.
1208         * config/i386/xsaveintrin.h (_xgetbv): Add missing return.
1209
1210 2018-05-10  Carl Love  <cel@us.ibm.com>
1211
1212         * config/rs6000/rs6000.md (prefetch): Generate ISA 2.06 instructions
1213         dcbtt and dcbtstt if operands[2] is 0.
1214
1215 2018-05-10  Uros Bizjak  <ubizjak@gmail.com>
1216
1217         PR target/85693
1218         * config/i386/sse.md (usadv64qi): New expander.
1219
1220 2018-05-10  Segher Boessenkool  <segher@kernel.crashing.org>
1221
1222         * config/rs6000/altivec.md (altivec_vmrghb, altivec_vmrghh,
1223         altivec_vmrghw, altivec_vmrglb, altivec_vmrglh, altivec_vmrglw): Remove
1224         -maltivec=be support.
1225         (vec_widen_umult_even_v16qi, vec_widen_smult_even_v16qi,
1226         vec_widen_umult_even_v8hi, vec_widen_smult_even_v8hi,
1227         vec_widen_umult_even_v4si, vec_widen_smult_even_v4si,
1228         vec_widen_umult_odd_v16qi, vec_widen_smult_odd_v16qi,
1229         vec_widen_umult_odd_v8hi, vec_widen_smult_odd_v8hi,
1230         vec_widen_umult_odd_v4si, vec_widen_smult_odd_v4si, altivec_vpkpx,
1231         altivec_vpks<VI_char>ss, altivec_vpks<VI_char>us,
1232         altivec_vpku<VI_char>us, altivec_vpku<VI_char>um, altivec_vsum2sws,
1233         altivec_vsumsws): Adjust.
1234         (altivec_vspltb *altivec_vspltb_internal, altivec_vsplth,
1235         *altivec_vsplth_internal, altivec_vspltw, *altivec_vspltw_internal,
1236         altivec_vspltsf, *altivec_vspltsf_internal): Remove -maltivec=be
1237         support.
1238         (altivec_vperm_<mode>, altivec_vperm_<mode>_uns,
1239         altivec_vupkhs<VU_char>, altivec_vupkls<VU_char>, altivec_vupkhpx,
1240         altivec_vupklpx, altivec_lvsl, altivec_lvsr): Adjust.
1241         (altivec_lve<VI_char>x): Delete expand.
1242         (*altivec_lve<VI_char>x_internal): Rename to...
1243         (altivec_lve<VI_char>x): ... this.
1244         (altivec_lvxl_<mode>): Delete expand.
1245         (*altivec_lvxl_<mode>_internal): Rename to ...
1246         (altivec_lvxl_<mode>): ... this.
1247         (altivec_stvxl_<mode>): Delete expand.
1248         (*altivec_stvxl_<mode>_internal): Rename to ...
1249         (altivec_stvxl_<mode>): ... this.
1250         (altivec_stve<VI_char>x): Delete expand.
1251         (*altivec_stve<VI_char>x_internal): Rename to ...
1252         (altivec_stve<VI_char>x): ... this.
1253         (doublee<mode>2, unsdoubleev4si2, doubleo<mode>2, unsdoubleov4si2,
1254         doubleh<mode>2, unsdoublehv4si2, doublel<mode>2, unsdoublelv4si2,
1255         reduc_plus_scal_<mode>): Adjust.
1256         * config/rs6000/rs6000-c.c (rs6000_target_modify_macros): Adjust
1257         comment.
1258         (rs6000_cpu_cpp_builtins): Adjust.
1259         (altivec_resolve_overloaded_builtin): Remove -maltivec=be support.
1260         * config/rs6000/rs6000-protos.h (altivec_expand_lvx_be,
1261         altivec_expand_stvx_be, altivec_expand_stvex_be): Delete.
1262         * config/rs6000/rs6000.c (rs6000_option_override_internal): Remove
1263         -maltivec=be support.
1264         (rs6000_split_vec_extract_var): Adjust.
1265         (rs6000_split_v4si_init): Adjust.
1266         (swap_selector_for_mode): Delete.
1267         (altivec_expand_lvx_be, altivec_expand_stvx_be,
1268         altivec_expand_stvex_be): Delete.
1269         (altivec_expand_lv_builtin, altivec_expand_stv_builtin): Remove
1270         -maltivec=be support.
1271         (rs6000_gimple_fold_builtin): Ditto.
1272         (rs6000_generate_float2_double_code, rs6000_generate_float2_code):
1273         Adjust.
1274         * config/rs6000/rs6000.h (VECTOR_ELT_ORDER_BIG): Delete.
1275         (TARGET_DIRECT_MOVE_64BIT): Adjust.
1276         * config/rs6000/rs6000.md (split for extendsidi2 for vectors): Adjust.
1277         * config/rs6000/rs6000.opt (maltivec=le, maltivec=be): Delete.
1278         * config/rs6000/vsx.md (floate<mode>, unsfloatev2di, floato<mode>,
1279         unsfloatov2di, vsignedo_v2df, vsignede_v2df, vunsignedo_v2df,
1280         vunsignede_v2df, vsx_extract_<mode>_p9, *vsx_extract_si,
1281         *vsx_extract_<mode>_p8, *vsx_extract_si_<uns>float_df,
1282         *vsx_extract_si_<uns>float_<mode>, vsx_set_<mode>_p9, vsx_set_v4sf_p9,
1283         *vsx_insert_extract_v4sf_p9, *vsx_insert_extract_v4sf_p9_2, and an
1284         anonymous split): Adjust.
1285         (vsx_mergel_<mode>, vsx_mergeh_<mode>): Remove -maltivec=be support.
1286         (vsx_xxspltd_<mode>, extract4b, insert4b): Adjust.
1287
1288 2018-05-10  Eric Botcazou  <ebotcazou@adacore.com>
1289
1290         * configure.ac (gcc_gxx_include_dir_add_sysroot): Set it to 1 only
1291         when --with-gxx-include-dir is also specified.
1292         * configure: Regenerate.
1293
1294 2018-05-09  Jim Wilson  <jimw@sifive.com>
1295
1296         PR target/84797
1297         * config.gcc (riscv*-*-*): Handle --with-multilib-list.
1298         * config/riscv/t-withmultilib: New.
1299         * config/riscv/withmultilib.h: New.
1300         * doc/install.texi: Document RISC-V --with-multilib-list support.
1301
1302 2018-05-09  Richard Biener  <rguenther@suse.de>
1303
1304         * tree-vect-slp.c (vect_bb_slp_scalar_cost): Fill a cost
1305         vector.
1306         (vect_bb_vectorization_profitable_p): Adjust.  Compute
1307         actual scalar cost using the cost vector and the add_stmt_cost
1308         machinery.
1309
1310 2018-05-09  Segher Boessenkool  <segher@kernel.crashing.org>
1311
1312         PR rtl-optimization/85645
1313         * config/rs6000/rs6000.c (rs6000_emit_prologue_components): Put a SET
1314         in the REG_CFA_REGISTER note for LR, don't leave it empty.
1315
1316 2018-05-09  Segher Boessenkool  <segher@kernel.crashing.org>
1317
1318         PR rtl-optimization/85645
1319         * shrink-wrap.c (spread_components): Return a boolean saying if
1320         anything was changed.
1321         (try_shrink_wrapping_separate): Iterate spread_components until
1322         nothing changes anymore.
1323
1324 2018-05-09  Segher Boessenkool  <segher@kernel.crashing.org>
1325
1326         PR rtl-optimization/85645
1327         * regrename.c (build_def_use): Also kill the chains that include the
1328         destination of a REG_CFA_REGISTER note.
1329
1330 2018-05-09  Segher Boessenkool  <segher@kernel.crashing.org>
1331
1332         PR rtl-optimization/85645
1333         *  regcprop.c (copyprop_hardreg_forward_1): Don't propagate into an
1334         insn that has a REG_CFA_REGISTER note.
1335
1336 2018-05-09  Richard Sandiford  <richard.sandiford@linaro.org>
1337
1338         * cfgexpand.c (expand_clobber): New function.
1339         (expand_gimple_stmt_1): Use it.
1340         * tree-vect-stmts.c (vect_clobber_variable): New function,
1341         split out from...
1342         (vectorizable_simd_clone_call): ...here.
1343         (vectorizable_store): Emit a clobber either side of an
1344         IFN_STORE_LANES sequence.
1345         (vectorizable_load): Emit a clobber after an IFN_LOAD_LANES sequence.
1346
1347 2018-05-09  Tom de Vries  <tom@codesourcery.com>
1348
1349         PR target/85626
1350         * config/nvptx/nvptx.md (define_insn "trap", define_insn "trap_if_true")
1351         (define_insn "trap_if_false"): Add exit after trap.
1352
1353 2018-05-09  Eric Botcazou  <ebotcazou@adacore.com>
1354
1355         PR rtl-optimization/85638
1356         * bb-reorder.c: Include common/common-target.h.
1357         (create_forwarder_block): New function extracted from...
1358         (fix_up_crossing_landing_pad): ...here.  Rename into...
1359         (dw2_fix_up_crossing_landing_pad): ...this.
1360         (sjlj_fix_up_crossing_landing_pad): New function.
1361         (find_rarely_executed_basic_blocks_and_crossing_edges): In SJLJ mode,
1362         call sjlj_fix_up_crossing_landing_pad if there are incoming EH edges
1363         from both partitions and exit the loop after one iteration.
1364
1365 2018-05-08  Kelvin Nilsen  <kelvin@gcc.gnu.org>
1366
1367         Revert:
1368         * doc/extend.texi (PowerPC Built-in Functions): Rename this
1369         subsection.
1370         (Basic PowerPC Built-in Functions): The new name of the
1371         subsection previously known as "PowerPC Built-in Functions".
1372         (Basic PowerPC Built-in Functions Available on all Configurations):
1373         New subsubsection.
1374         (Basic PowerPC Built-in Functions Available on ISA 2.05): New
1375         subsubsection.
1376         (Basic PowerPC Built-in Functions Available on ISA 2.06): New
1377         subsubsection.
1378         (Basic PowerPC Built-in Functions Available on ISA 2.07): New
1379         subsubsection.
1380         (Basic PowerPC Built-in Functions Available on ISA 3.0): New
1381         subsubsection.
1382
1383 2018-05-08  Jim Wilson  <jimw@sifive.com>
1384
1385         * config/riscv/linux.h (MUSL_ABI_SUFFIX): Delete unnecessary backslash.
1386         (LD_EMUL_SUFFIX): New.
1387         (LINK_SPEC): Use it.
1388
1389 2018-05-08  Kelvin Nilsen  <kelvin@gcc.gnu.org>
1390
1391         * doc/extend.texi (PowerPC Built-in Functions): Rename this
1392         subsection.
1393         (Basic PowerPC Built-in Functions): The new name of the
1394         subsection previously known as "PowerPC Built-in Functions".
1395         (Basic PowerPC Built-in Functions Available on all Configurations):
1396         New subsubsection.
1397         (Basic PowerPC Built-in Functions Available on ISA 2.05): New
1398         subsubsection.
1399         (Basic PowerPC Built-in Functions Available on ISA 2.06): New
1400         subsubsection.
1401         (Basic PowerPC Built-in Functions Available on ISA 2.07): New
1402         subsubsection.
1403         (Basic PowerPC Built-in Functions Available on ISA 3.0): New
1404         subsubsection.
1405
1406 2018-05-08  Jakub Jelinek  <jakub@redhat.com>
1407
1408         PR target/85683
1409         * config/i386/i386.md: Add peepholes for mem {+,-,&,|,^}= x; mem != 0
1410         after cmpelim optimization.
1411
1412 2018-05-08  Olga Makhotina  <olga.makhotina@intel.com>
1413
1414         * config.gcc: Support "goldmont".
1415         * config/i386/driver-i386.c (host_detect_local_cpu): Detect "goldmont".
1416         * config/i386/i386-c.c (ix86_target_macros_internal): Handle
1417         PROCESSOR_GOLDMONT.
1418         * config/i386/i386.c (m_GOLDMONT): Define.
1419         (processor_target_table): Add "goldmont".
1420         (PTA_GOLDMONT): Define.
1421         (ix86_lea_outperforms): Add TARGET_GOLDMONT.
1422         (get_builtin_code_for_version): Handle PROCESSOR_GOLDMONT.
1423         (fold_builtin_cpu): Add M_INTEL_GOLDMONT.
1424         (fold_builtin_cpu): Add "goldmont".
1425         (ix86_add_stmt_cost): Add TARGET_GOLDMONT.
1426         (ix86_option_override_internal): Add "goldmont".
1427         * config/i386/i386.h (processor_costs): Define TARGET_GOLDMONT.
1428         (processor_type): Add PROCESSOR_GOLDMONT.
1429         * config/i386/i386.md: Add CPU "glm".
1430         * config/i386/glm.md: New file.
1431         * config/i386/x86-tune.def: Add m_GOLDMONT.
1432         * doc/invoke.texi: Add goldmont as x86 -march=/-mtune= CPU type.
1433
1434 2018-05-08  Jakub Jelinek  <jakub@redhat.com>
1435
1436         PR target/85572
1437         * config/i386/i386.c (ix86_expand_sse2_abs): Handle E_V2DImode and
1438         E_V4DImode.
1439         * config/i386/sse.md (abs<mode>2): Use VI_AVX2 iterator instead of
1440         VI1248_AVX512VL_AVX512BW.  Handle V2DImode and V4DImode if not
1441         TARGET_AVX512VL using ix86_expand_sse2_abs.  Formatting fixes.
1442
1443         PR target/85317
1444         * config/i386/i386.c (ix86_fold_builtin): Handle
1445         IX86_BUILTIN_{,P}MOVMSK{PS,PD,B}{,128,256}.
1446
1447         PR target/85480
1448         * config/i386/sse.md (ssequaterinsnmode): New mode attribute.
1449         (*<extract_type>_vinsert<shuffletype><extract_suf>_0): New pattern.
1450
1451 2018-05-08  Richard Earnshaw  <rearnsha@arm.com>
1452
1453         PR target/85658
1454         * config/arm/parsecpu.awk (check_cpu): Fix operator precedence.
1455         (check_arch): Likewise.
1456         (check_fpu): Return the result rather than printing it.
1457         (end arch): Fix operator precedence.
1458         (end cpu): Likewise.
1459         (END): Print the result from check_fpu.
1460
1461 2018-05-08  Richard Sandiford  <richard.sandiford@linaro.org>
1462             Alan Hayward  <alan.hayward@arm.com>
1463             David Sherwood  <david.sherwood@arm.com>
1464
1465         * config/aarch64/aarch64-sve.md (*pred_cmp<cmp_op><mode>_combine)
1466         (*pred_cmp<cmp_op><mode>, *fcm<cmp_op><mode>_and_combine)
1467         (*fcmuo<mode>_and_combine, *fcm<cmp_op><mode>_and)
1468         (*fcmuo<mode>_and): New patterns.
1469
1470 2018-05-08  Richard Sandiford  <richard.sandiford@linaro.org>
1471
1472         * config/aarch64/iterators.md (UNSPEC_COND_LO, UNSPEC_COND_LS)
1473         (UNSPEC_COND_HI, UNSPEC_COND_HS, UNSPEC_COND_UO): Delete.
1474         (SVE_INT_CMP, SVE_FP_CMP): New code iterators.
1475         (cmp_op, sve_imm_con): New code attributes.
1476         (SVE_COND_INT_CMP, imm_con): Delete.
1477         (cmp_op): Remove above unspecs from int attribute.
1478         * config/aarch64/aarch64-sve.md (*vec_cmp<cmp_op>_<mode>): Rename
1479         to...
1480         (*cmp<cmp_op><mode>): ...this.  Use UNSPEC_MERGE_PTRUE instead of
1481         comparison-specific unspecs.
1482         (*vec_cmp<cmp_op>_<mode>_ptest): Rename to...
1483         (*cmp<cmp_op><mode>_ptest): ...this and adjust likewise.
1484         (*vec_cmp<cmp_op>_<mode>_cc): Rename to...
1485         (*cmp<cmp_op><mode>_cc): ...this and adjust likewise.
1486         (*vec_fcm<cmp_op><mode>): Rename to...
1487         (*fcm<cmp_op><mode>): ...this and adjust likewise.
1488         (*vec_fcmuo<mode>): Rename to...
1489         (*fcmuo<mode>): ...this and adjust likewise.
1490         (*pred_fcm<cmp_op><mode>): New pattern.
1491         * config/aarch64/aarch64.c (aarch64_emit_unop, aarch64_emit_binop)
1492         (aarch64_emit_sve_ptrue_op, aarch64_emit_sve_ptrue_op_cc): New
1493         functions.
1494         (aarch64_unspec_cond_code): Remove handling of LTU, GTU, LEU, GEU
1495         and UNORDERED.
1496         (aarch64_gen_unspec_cond, aarch64_emit_unspec_cond): Delete.
1497         (aarch64_emit_sve_predicated_cond): New function.
1498         (aarch64_expand_sve_vec_cmp_int): Use aarch64_emit_sve_ptrue_op_cc.
1499         (aarch64_emit_unspec_cond_or): Replace with...
1500         (aarch64_emit_sve_or_conds): ...this new function.  Use
1501         aarch64_emit_sve_ptrue_op for the individual comparisons and
1502         aarch64_emit_binop to OR them together.
1503         (aarch64_emit_inverted_unspec_cond): Replace with...
1504         (aarch64_emit_sve_inverted_cond): ...this new function.  Use
1505         aarch64_emit_sve_ptrue_op for the comparison and
1506         aarch64_emit_unop to invert the result.
1507         (aarch64_expand_sve_vec_cmp_float): Update after the above
1508         changes.  Use aarch64_emit_sve_ptrue_op for native comparisons.
1509
1510 2018-05-07  Nathan Sidwell  <nathan@acm.org>
1511
1512         * doc/invoke.texi (C++ Dialect Options): Remove -ffor-scope.
1513         * doc/extend.texi (Deprecated Features): Remove -fno-for-scope
1514         (Backwards Compatibility): Likewise.
1515
1516 2018-05-07  Luis Machado  <luis.machado@linaro.org>
1517
1518         PR bootstrap/85681
1519         Revert:
1520         2018-05-07  Luis Machado  <luis.machado@linaro.org>
1521
1522         * config/aarch64/aarch64-protos.h (cpu_prefetch_tune)
1523         <prefetch_dynamic_strides>: New const bool field.
1524         * config/aarch64/aarch64.c (generic_prefetch_tune): Update to include
1525         prefetch_dynamic_strides.
1526         (exynosm1_prefetch_tune): Likewise.
1527         (thunderxt88_prefetch_tune): Likewise.
1528         (thunderx_prefetch_tune): Likewise.
1529         (thunderx2t99_prefetch_tune): Likewise.
1530         (qdf24xx_prefetch_tune): Likewise. Set prefetch_dynamic_strides to false.
1531         (aarch64_override_options_internal): Update to set
1532         PARAM_PREFETCH_DYNAMIC_STRIDES.
1533         * doc/invoke.texi (prefetch-dynamic-strides): Document new option.
1534         * params.def (PARAM_PREFETCH_DYNAMIC_STRIDES): New.
1535         * params.h (PARAM_PREFETCH_DYNAMIC_STRIDES): Define.
1536         * tree-ssa-loop-prefetch.c (should_issue_prefetch_p): Account for
1537         prefetch-dynamic-strides setting.
1538
1539         2018-05-07  Luis Machado  <luis.machado@linaro.org>
1540
1541         * config/aarch64/aarch64-protos.h (cpu_prefetch_tune)
1542         <minimum_stride>: New const int field.
1543         * config/aarch64/aarch64.c (generic_prefetch_tune): Update to include
1544         minimum_stride field.
1545         (exynosm1_prefetch_tune): Likewise.
1546         (thunderxt88_prefetch_tune): Likewise.
1547         (thunderx_prefetch_tune): Likewise.
1548         (thunderx2t99_prefetch_tune): Likewise.
1549         (qdf24xx_prefetch_tune): Likewise. Set minimum_stride to 2048.
1550         (aarch64_override_options_internal): Update to set
1551         PARAM_PREFETCH_MINIMUM_STRIDE.
1552         * doc/invoke.texi (prefetch-minimum-stride): Document new option.
1553         * params.def (PARAM_PREFETCH_MINIMUM_STRIDE): New.
1554         * params.h (PARAM_PREFETCH_MINIMUM_STRIDE): Define.
1555         * tree-ssa-loop-prefetch.c (should_issue_prefetch_p): Return false if
1556         stride is constant and is below the minimum stride threshold.
1557
1558 2018-05-07  Luis Machado  <luis.machado@linaro.org>
1559
1560         * config/aarch64/aarch64.c (qdf24xx_prefetch_tune) <l2_cache_size>: Set
1561         to 512.
1562
1563 2018-05-07  Luis Machado  <luis.machado@linaro.org>
1564
1565         * config/aarch64/aarch64-protos.h (cpu_prefetch_tune)
1566         <prefetch_dynamic_strides>: New const bool field.
1567         * config/aarch64/aarch64.c (generic_prefetch_tune): Update to include
1568         prefetch_dynamic_strides.
1569         (exynosm1_prefetch_tune): Likewise.
1570         (thunderxt88_prefetch_tune): Likewise.
1571         (thunderx_prefetch_tune): Likewise.
1572         (thunderx2t99_prefetch_tune): Likewise.
1573         (qdf24xx_prefetch_tune): Likewise. Set prefetch_dynamic_strides to false.
1574         (aarch64_override_options_internal): Update to set
1575         PARAM_PREFETCH_DYNAMIC_STRIDES.
1576         * doc/invoke.texi (prefetch-dynamic-strides): Document new option.
1577         * params.def (PARAM_PREFETCH_DYNAMIC_STRIDES): New.
1578         * params.h (PARAM_PREFETCH_DYNAMIC_STRIDES): Define.
1579         * tree-ssa-loop-prefetch.c (should_issue_prefetch_p): Account for
1580         prefetch-dynamic-strides setting.
1581
1582 2018-05-07  Luis Machado  <luis.machado@linaro.org>
1583
1584         * config/aarch64/aarch64-protos.h (cpu_prefetch_tune)
1585         <minimum_stride>: New const int field.
1586         * config/aarch64/aarch64.c (generic_prefetch_tune): Update to include
1587         minimum_stride field.
1588         (exynosm1_prefetch_tune): Likewise.
1589         (thunderxt88_prefetch_tune): Likewise.
1590         (thunderx_prefetch_tune): Likewise.
1591         (thunderx2t99_prefetch_tune): Likewise.
1592         (qdf24xx_prefetch_tune): Likewise. Set minimum_stride to 2048.
1593         (aarch64_override_options_internal): Update to set
1594         PARAM_PREFETCH_MINIMUM_STRIDE.
1595         * doc/invoke.texi (prefetch-minimum-stride): Document new option.
1596         * params.def (PARAM_PREFETCH_MINIMUM_STRIDE): New.
1597         * params.h (PARAM_PREFETCH_MINIMUM_STRIDE): Define.
1598         * tree-ssa-loop-prefetch.c (should_issue_prefetch_p): Return false if
1599         stride is constant and is below the minimum stride threshold.
1600
1601 2018-05-06  Jakub Jelinek  <jakub@redhat.com>
1602
1603         PR c++/85659
1604         * cfgexpand.c (expand_asm_stmt): Don't create a temporary if
1605         the type is addressable.  Don't force op into register if it has
1606         BLKmode.
1607
1608 2018-05-05  Roland McGrath  <mcgrathr@google.com>
1609
1610         PR other/77609
1611         * varasm.c (default_section_type_flags): Set SECTION_NOTYPE for
1612         any section for which we don't know a specific type it should have,
1613         regardless of name.  Previously this was done only for the exact
1614         names ".init_array", ".fini_array", and ".preinit_array".
1615         (default_elf_asm_named_section): Add comment about
1616         relationship with default_section_type_flags and SECTION_NOTYPE.
1617         (get_section): Don't consider it a type conflict if one side has
1618         SECTION_NOTYPE and the other doesn't, as long as neither has the
1619         SECTION_BSS et al used in the default_section_type_flags logic.
1620
1621 2018-05-05  Tom de Vries  <tom@codesourcery.com>
1622
1623         PR target/85653
1624         * config/nvptx/nvptx.c (WORKAROUND_PTXJIT_BUG_3): Define.
1625         (workaround_barsyncs): New function.
1626         (nvptx_reorg): Use workaround_barsyncs.
1627         * config/nvptx/nvptx.md (define_c_enum "unspecv"): Add UNSPECV_MEMBAR.
1628         (define_expand "nvptx_membar_cta"): New define_expand.
1629         (define_insn "*nvptx_membar_cta"): New insn.
1630
1631 2018-05-04  Pekka Jääskeläinen  <pekka.jaaskelainen@parmance.com>
1632
1633         * brig-builtins.def: Add consts to ptrs etc. in BRIG builtin defs.
1634         To improve optimization opportunities.
1635         * builtin-types.def: The new needed builtin types for the above.
1636
1637 2018-05-04  Richard Biener  <rguenther@suse.de>
1638
1639         * bb-reorder.c (sanitize_hot_paths): Release hot_bbs_to_check.
1640         * gimple-ssa-store-merging.c
1641         (imm_store_chain_info::output_merged_store): Remove redundant create,
1642         release split_store vector contents on failure.
1643         * tree-vect-slp.c (vect_schedule_slp_instance): Avoid leaking
1644         scalar stmt vector on cache hit.
1645
1646 2018-05-04  Segher Boessenkool  <segher@kernel.crashing.org>
1647
1648         * common/config/rs6000/rs6000-common.c (rs6000_handle_option): Remove
1649         Xilinx FP support.
1650         * config.gcc (powerpc-xilinx-eabi*): Remove.
1651         * config/rs6000/predicates.md (easy_fp_constant): Remove Xilinx FP
1652         support.
1653         (fusion_addis_mem_combo_load): Ditto.
1654         * config/rs6000/rs6000-c.c (rs6000_target_modify_macros): Remove Xilinx
1655         FP support.
1656         (rs6000_cpu_cpp_builtins): Ditto.
1657         * config/rs6000/rs6000-linux.c
1658         (rs6000_linux_float_exceptions_rounding_supported_p): Ditto.
1659         * config/rs6000/rs6000-opts.h (enum fpu_type_t): Delete.
1660         * config/rs6000/rs6000.c (rs6000_debug_reg_global): Remove Xilinx FP
1661         support.
1662         (rs6000_setup_reg_addr_masks): Ditto.
1663         (rs6000_init_hard_regno_mode_ok): Ditto.
1664         (rs6000_option_override_internal): Ditto.
1665         (legitimate_lo_sum_address_p): Ditto.
1666         (rs6000_legitimize_address): Ditto.
1667         (rs6000_legitimize_reload_address): Ditto.
1668         (rs6000_legitimate_address_p): Ditto.
1669         (abi_v4_pass_in_fpr): Ditto.
1670         (setup_incoming_varargs): Ditto.
1671         (rs6000_gimplify_va_arg): Ditto.
1672         (rs6000_split_multireg_move): Ditto.
1673         (rs6000_savres_strategy): Ditto.
1674         (rs6000_emit_prologue_components): Ditto.
1675         (rs6000_emit_epilogue_components): Ditto.
1676         (rs6000_emit_prologue): Ditto.
1677         (rs6000_emit_epilogue): Ditto.
1678         (rs6000_elf_file_end): Ditto.
1679         (rs6000_function_value): Ditto.
1680         (rs6000_libcall_value): Ditto.
1681         * config/rs6000/rs6000.h: Ditto.
1682         (TARGET_MINMAX_SF, TARGET_MINMAX_DF): Delete, merge to ...
1683         (TARGET_MINMAX): ... this.  New.
1684         (TARGET_SF_FPR, TARGET_DF_FPR, TARGET_SF_INSN, TARGET_DF_INSN): Delete.
1685         * config/rs6000/rs6000.md: Remove Xilinx FP support.
1686         (*movsi_internal1_single): Delete.
1687         * config/rs6000/rs6000.opt (msingle-float, mdouble-float, msimple-fpu,
1688         mfpu=, mxilinx-fpu): Delete.
1689         * config/rs6000/singlefp.h: Delete.
1690         * config/rs6000/sysv4.h: Remove Xilinx FP support.
1691         * config/rs6000/t-rs6000: Ditto.
1692         * config/rs6000/t-xilinx: Delete.
1693         * gcc/config/rs6000/titan.md: Adjust for fp_type removal.
1694         * gcc/config/rs6000/vsx.md: Remove Xilinx FP support.
1695         (VStype_simple): Delete.
1696         (VSfptype_simple, VSfptype_mul, VSfptype_div, VSfptype_sqrt): Delete.
1697         * config/rs6000/xfpu.h: Delete.
1698         * config/rs6000/xfpu.md: Delete.
1699         * config/rs6000/xilinx.h: Delete.
1700         * config/rs6000/xilinx.opt: Delete.
1701         * gcc/doc/invoke.texi (RS/6000 and PowerPC Options): Remove
1702         -msingle-float, -mdouble-float, -msimple-fpu, -mfpu=, and -mxilinx-fpu.
1703
1704 2018-05-04  Tom de Vries  <tom@codesourcery.com>
1705
1706         PR libgomp/85639
1707         * builtins.c (expand_builtin_goacc_parlevel_id_size): Handle null target
1708         if ignore == 0.
1709
1710 2018-05-04  Richard Biener  <rguenther@suse.de>
1711
1712         PR middle-end/85627
1713         * tree-complex.c (update_complex_assignment): We are always in SSA form.
1714         (expand_complex_div_wide): Likewise.
1715         (expand_complex_operations_1): Likewise.
1716         (expand_complex_libcall): Preserve EH info of the original stmt.
1717         (tree_lower_complex): Handle removed blocks.
1718         * tree.c (build_common_builtin_nodes): Do not set ECF_NOTRHOW
1719         on complex multiplication and division libcall builtins.
1720
1721 2018-05-04  Richard Biener  <rguenther@suse.de>
1722
1723         PR middle-end/85574
1724         * fold-const.c (negate_expr_p): Restrict negation of operand
1725         zero of a division to when we know that can happen without
1726         overflow.
1727         (fold_negate_expr_1): Likewise.
1728
1729 2018-05-04  Jakub Jelinek  <jakub@redhat.com>
1730
1731         PR libstdc++/85466
1732         * real.h (real_nextafter): Declare.
1733         * real.c (real_nextafter): New function.
1734         * fold-const-call.c (fold_const_nextafter): New function.
1735         (fold_const_call_sss): Call it for CASE_CFN_NEXTAFTER and
1736         CASE_CFN_NEXTTOWARD.
1737         (fold_const_call_1): For CASE_CFN_NEXTTOWARD call fold_const_call_sss
1738         even when arg1_mode is different from arg0_mode.
1739
1740 2018-05-03  Nathan Sidwell  <nathan@acm.org>
1741
1742         * doc/extend.texi (Deprecated Features): Remove
1743         -ffriend-injection.
1744         (Backwards Compatibility): Likewise.
1745         * doc/invoke.texi (C++ Language Options): Likewise.
1746         (C++ Dialect Options): Likewise.
1747
1748 2018-05-03  Jakub Jelinek  <jakub@redhat.com>
1749
1750         PR target/85530
1751         * config/i386/avx512fintrin.h (_mm512_mullox_epi64,
1752         _mm512_mask_mullox_epi64): New intrinsics.
1753
1754 2018-05-03  Tom de Vries  <tom@codesourcery.com>
1755
1756         PR testsuite/85106
1757         * doc/sourcebuild.texi (Commands for use in dg-final, Scan optimization
1758         dump files): Add offload-tree.
1759
1760 2018-05-03  Richard Biener  <rguenther@suse.de>
1761
1762         PR tree-optimization/85615
1763         * tree-ssa-threadupdate.c (thread_block_1): Only allow exits
1764         to loops not nested in BBs loop father to avoid creating multi-entry
1765         loops.
1766
1767 2018-05-03  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
1768
1769         PR tree-optimization/70291
1770         * tree-complex.c (expand_complex_libcall): Add type, inplace_p
1771         arguments.  Change return type to tree.  Emit libcall as a new
1772         statement rather than replacing existing one when inplace_p is true.
1773         (expand_complex_multiplication_components): New function.
1774         (expand_complex_multiplication): Expand floating-point complex
1775         multiplication using the above.
1776         (expand_complex_division): Rename inner_type parameter to type.
1777         Update expand_complex_libcall call-site.
1778         (expand_complex_operations_1): Update expand_complex_multiplication
1779         and expand_complex_division call-sites.
1780
1781 2018-05-02  Jakub Jelinek  <jakub@redhat.com>
1782
1783         PR target/85582
1784         * config/i386/i386.md (*ashl<dwi>3_doubleword_mask,
1785         *ashl<dwi>3_doubleword_mask_1, *<shift_insn><dwi>3_doubleword_mask,
1786         *<shift_insn><dwi>3_doubleword_mask_1): In condition require that
1787         the highest significant bit of the shift count mask is clear.  In
1788         check whether and[sq]i3 is needed verify that all significant bits
1789         of the shift count other than the highest are set.
1790
1791 2018-05-02  Tom de Vries  <tom@codesourcery.com>
1792
1793         PR libgomp/82428
1794         * builtins.def (DEF_GOACC_BUILTIN_ONLY): Define.
1795         * omp-builtins.def (BUILT_IN_GOACC_PARLEVEL_ID)
1796         (BUILT_IN_GOACC_PARLEVEL_SIZE): New builtin.
1797         * builtins.c (expand_builtin_goacc_parlevel_id_size): New function.
1798         (expand_builtin): Call expand_builtin_goacc_parlevel_id_size.
1799         * doc/extend.texi (Other Builtins): Add __builtin_goacc_parlevel_id and
1800         __builtin_goacc_parlevel_size.
1801
1802 2018-05-02  Richard Biener  <rguenther@suse.de>
1803
1804         PR tree-optimization/85597
1805         * tree-vect-stmts.c (vectorizable_operation): For ternary SLP
1806         do not use split vect_get_vec_defs call but call vect_get_slp_defs
1807         directly.
1808
1809 2018-05-02  Tom de Vries  <tom@codesourcery.com>
1810
1811         PR testsuite/85106
1812         * doc/sourcebuild.texi (Commands for use in dg-final, Scan optimization
1813         dump files): Add ltrans-tree.
1814
1815 2018-05-02  Tom de Vries  <tom@codesourcery.com>
1816
1817         PR testsuite/85106
1818         * doc/sourcebuild.texi (Commands for use in dg-final, Scan optimization
1819         dump files): Add wpa-ipa.
1820
1821 2018-05-02  Segher Boessenkool  <segher@kernel.crashing.org>
1822
1823         * config.gcc (powerpc*-*-*): Remove paired.h.  Unsupport the
1824         powerpc*-*-linux*paired* target.
1825         * config/rs6000/750cl.h: Delete.
1826         * config/rs6000/paired.h: Delete.
1827         * config/rs6000/paired.md: Delete.
1828         * config/rs6000/predicates.md (easy_vector_constant): Remove paired
1829         float support.
1830         * config/rs6000/rs6000-builtin.def: Remove paired float support.
1831         * config/rs6000/rs6000-c.c (rs6000_target_modify_macros): Update
1832         comment.  Remove paired float support.
1833         * config/rs6000/rs6000-modes.def: Remove V2SF and V2SI.
1834         * config/rs6000/rs6000-opts.h (enum rs6000_vector): Delete
1835         VECTOR_PAIRED.
1836         * config/rs6000/rs6000-protos.h (paired_expand_vector_init,
1837         paired_emit_vector_cond_expr, paired_expand_vector_move): Delete
1838         declarations.
1839         * config/rs6000/rs6000.c: Remove paired float support.
1840         (paired_expand_vector_init, paired_expand_vector_move,
1841         paired_emit_vector_compare, paired_emit_vector_cond_expr,
1842         (paired_expand_lv_builtin, paired_expand_stv_builtin,
1843         paired_expand_builtin, paired_expand_predicate_builtin,
1844         paired_init_builtins): Delete.
1845         * config/rs6000/rs6000.h: Remove paired float support.
1846         * config/rs6000/rs6000.md: Remove paired float support.
1847         (move_from_CR_ov_bit): Delete.
1848         * config/rs6000/rs6000.opt (mpaired): Delete.
1849         * config/rs6000/t-rs6000: Remove paired.md from MD_INCLUDES.
1850         * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mpaired.
1851
1852 2018-05-02  Richard Biener  <rguenther@suse.de>
1853
1854         PR middle-end/85567
1855         * gimplify.c (gimplify_save_expr): When in SSA form allow
1856         SAVE_EXPRs to compute to SSA vars.
1857
1858 2018-05-02  Jakub Jelinek  <jakub@redhat.com>
1859
1860         PR target/85582
1861         * config/i386/i386.md (*ashl<dwi>3_doubleword_mask,
1862         *ashl<dwi>3_doubleword_mask_1, *<shift_insn><dwi>3_doubleword_mask,
1863         *<shift_insn><dwi>3_doubleword_mask_1): If and[sq]i3 is needed, don't
1864         clobber operands[2], instead use a new pseudo.  Formatting fixes.
1865
1866 2018-05-02  Richard Sandiford  <richard.sandiford@linaro.org>
1867
1868         PR tree-optimization/85586
1869         * tree-vect-data-refs.c (vect_analyze_data_ref_dependence): Only
1870         exit early for statements in the same group if the accesses are
1871         not strided.
1872
1873 2018-05-02  Tom de Vries  <tom@codesourcery.com>
1874
1875         PR lto/85451
1876         * lto-wrapper.c (compile_offload_image): Add "could not find mkoffload"
1877         error message.
1878
1879 2018-05-01  Marc Glisse  <marc.glisse@inria.fr>
1880
1881         PR tree-optimization/85143
1882         * match.pd (A<B&A<C): Extend to BIT_IOR_EXPR.
1883
1884 2018-05-01  Tom de Vries  <tom@codesourcery.com>
1885
1886         PR lto/85451
1887         * config/nvptx/mkoffload.c (main): Suggest using -B in "offload compiler
1888         not found" error message.
1889
1890 2018-05-01  Tom de Vries  <tom@codesourcery.com>
1891
1892         PR other/83786
1893         * vec.h (VEC_ORDERED_REMOVE_IF, VEC_ORDERED_REMOVE_IF_FROM_TO): Define.
1894         * vec.c (test_ordered_remove_if): New function.
1895         (vec_c_tests): Call test_ordered_remove_if.
1896         * dwarf2cfi.c (connect_traces): Use VEC_ORDERED_REMOVE_IF_FROM_TO.
1897         * lto-streamer-out.c (prune_offload_funcs): Use VEC_ORDERED_REMOVE_IF.
1898         * tree-vect-patterns.c (vect_pattern_recog_1): Use
1899         VEC_ORDERED_REMOVE_IF.
1900
1901 2018-05-01  Prathamesh Kulkarni  <prathamesh.kulkarni@linaro.org>
1902
1903         PR tree-optimization/82665
1904         * vr-values.c (vr_values::extract_range_from_binary_expr): Handle
1905         pointer subtraction where arguments come from a memchr call.
1906
1907 2018-05-01  Jakub Jelinek  <jakub@redhat.com>
1908
1909         * configure.ac (LD_AS_NEEDED_OPTION, LD_NO_AS_NEEDED_OPTION): Use
1910         --push-state --as-needed and --pop-state instead of --as-needed and
1911         --no-as-needed if ld supports it.
1912         * configure: Regenerated.
1913
1914         PR web/85578
1915         * doc/install.texi2html: Replace _002d with - and _002a with * in
1916         generated html files using sed.
1917
1918 2018-04-30  David Malcolm  <dmalcolm@redhat.com>
1919
1920         PR c++/85523
1921         * gcc-rich-location.c (blank_line_before_p): New function.
1922         (use_new_line): New function.
1923         (gcc_rich_location::add_fixit_insert_formatted): New function.
1924         * gcc-rich-location.h
1925         (gcc_rich_location::add_fixit_insert_formatted): New function.
1926
1927 2018-04-30  David Malcolm  <dmalcolm@redhat.com>
1928
1929         * selftest.c (assert_streq): Rename "expected" and "actual" to
1930         "val1" and "val2".  Extend NULL-handling to cover both inputs
1931         symmetrically, while still requiring both to be non-NULL for a pass.
1932         * selftest.h (assert_streq): Rename "expected" and "actual" to
1933         "val1" and "val2".
1934         (ASSERT_EQ): Likewise.
1935         (ASSERT_EQ_AT): Likewise.
1936         (ASSERT_KNOWN_EQ): Likewise.
1937         (ASSERT_KNOWN_EQ_AT): Likewise.
1938         (ASSERT_NE): Likewise.
1939         (ASSERT_MAYBE_NE): Likewise.
1940         (ASSERT_MAYBE_NE_AT): Likewise.
1941         (ASSERT_STREQ): Likewise.  Clarify that both must be non-NULL for
1942         the assertion to pass.
1943         (ASSERT_STREQ_AT): Likewise.
1944
1945 2018-04-30  Jonathan Wakely  <jwakely@redhat.com>
1946
1947         * doc/invoke.texi (-fpie, -fPIE): Fix grammar and clarify
1948         interaction with -pie.
1949
1950 2018-04-30  David Malcolm  <dmalcolm@redhat.com>
1951
1952         * selftest.h: Fix alphabetization of per-source-file selftest
1953         declarations.
1954
1955 2018-04-30  Jason Merrill  <jason@redhat.com>
1956
1957         PR c++/61982 - dead stores to destroyed objects.
1958         * gimplify.c (gimplify_modify_expr): Simplify complex lvalue on LHS
1959         of clobber.
1960
1961 2018-04-30  Jason Merrill  <jason@redhat.com>
1962
1963         * tree.c (build_clobber): New.
1964         * tree.h: Declare it.
1965         * gimplify.c (gimplify_bind_expr, gimplify_target_expr): Use it.
1966
1967 2018-04-30  David Malcolm  <dmalcolm@redhat.com>
1968
1969         * diagnostic-show-locus.c (layout::layout): Update for
1970         location_get_source_line returning a char_span.
1971         (struct char_span): Move to input.h.
1972         (struct correction): Update for fields in char_span becoming
1973         private.
1974         (struct source_line): Update for location_get_source_line
1975         returning a char_span.
1976         (layout::print_line): Likewise.
1977         * edit-context.c (edited_file::print_content): Likewise.
1978         (edited_file::print_diff_hunk): Likewise.
1979         (edited_file::print_run_of_changed_lines): Likewise.
1980         (edited_file::get_num_lines): Likewise.
1981         (edited_line::edited_line): Likewise.
1982         * final.c (asm_show_source): Likewise.
1983         * input.c (location_get_source_line): Convert return type
1984         from const char * to char_span, losing the final "line_len"
1985         param.
1986         (dump_location_info): Update for the above.
1987         (get_substring_ranges_for_loc): Likewise.  Use a char_span
1988         when handling the literal within the line.
1989         (test_reading_source_line): Update for location_get_source_line
1990         returning a char_span.
1991         * input.h (class char_span): Move here from
1992         diagnostic-show-locus.c, converting from a struct to a class.
1993         Make data members private.
1994         (char_span::operator bool): New.
1995         (char_span::length): New.
1996         (char_span::get_buffer): New.
1997         (char_span::operator[]): New.
1998         (char_span::subspan): Make const.
1999         (char_span::xstrdup): New.
2000         (location_get_source_line): Convert return type from const char *
2001         to char_span, losing the final "line_size" param.
2002
2003 2018-04-30  Jan Hubicka  <jh@suse.cz>
2004
2005         * lto-wrapper.c (ltrans_priorities): New static var.
2006         (cmp_priority): New.
2007         (run_gcc): Read priorities and if doing parallel build order
2008         the Makefile by them.
2009
2010 2018-04-30  David Malcolm  <dmalcolm@redhat.com>
2011
2012         * input.h (builtins_location_check): Convert to a STATIC_ASSERT.
2013
2014 2018-04-30  Richard Biener  <rguenther@suse.de>
2015
2016         * tree-cfg.c (verify_address): Remove base argument, add
2017         flag whether to check TREE_ADDRESSABLE and do that.
2018         (verify_expr): Remove.
2019         (verify_types_in_gimple_reference): Add pieces from verify_expr.
2020         (verify_gimple_assign_single): Likewise.
2021         (verify_gimple_switch): Likewise.
2022         (verify_expr_location_1): Dereference tp once.  Add (disabled)
2023         piece from verify_expr.
2024         (verify_gimple_in_cfg): Do not call verify_expr on all ops.
2025
2026 2018-04-30  Claudiu Zissulescu  <claziss@synopsys.com>
2027
2028         * config/arc/linux.h (CLEAR_INSN_CACHE): Define.
2029
2030 2018-04-30  Claudiu Zissulescu  <claziss@synopsys.com>
2031
2032         * config/arc/arc-protos.h (prepare_extend_operands): Remove.
2033         (small_data_pattern): Likewise.
2034         (arc_rewrite_small_data): Likewise.
2035         * config/arc/arc.c (LEGITIMATE_SMALL_DATA_OFFSET_P): Remove.
2036         (LEGITIMATE_SMALL_DATA_ADDRESS_P): Likewise.
2037         (get_symbol_alignment): New function.
2038         (legitimate_small_data_address_p): Likewise.
2039         (legitimate_scaled_address): Update, call
2040         legitimate_small_data_address_p.
2041         (output_sdata): New static variable.
2042         (arc_print_operand): Update how we handle small data operands.
2043         (arc_print_operand_address): Likewise.
2044         (arc_legitimate_address_p): Update, use
2045         legitimate_small_data_address_p.
2046         (arc_rewrite_small_data_p): Remove.
2047         (arc_rewrite_small_data_1): Likewise.
2048         (arc_rewrite_small_data): Likewise.
2049         (small_data_pattern): Likewise.
2050         (compact_sda_memory_operand): Update to use
2051         legitimate_small_data_address_p and get_symbol_alignment.
2052         (prepare_move_operands): Don't rewite sdata pattern.
2053         (prepare_extend_operands): Remove.
2054         * config/arc/arc.md (zero_extendqihi2): Don't rewrite sdata
2055         pattern.
2056         (zero_extendqisi2): Likewise.
2057         (zero_extendhisi2): Likewise.
2058         (extendqihi2): Likewise.
2059         (extendqisi2): Likewise.
2060         (extendhisi2): Likewise.
2061         (addsi3): Likewise.
2062         (subsi3): Likewise.
2063         (andsi3): Likewise.
2064         * config/arc/constraints.md (Usd): Change it to memory constraint.
2065
2066 2018-04-30  Claudiu Zissulescu  <claziss@synopsys.com>
2067
2068         * config/arc/arc.c (arc_split_move): Allow signed 6-bit constants
2069         as source of std instructions.
2070         * config/arc/arc.md (movsi_insn): Update pattern predicate to
2071         allow 6-bit constants as source for store instructions.
2072         (movdi_insn): Update instruction pattern to allow 6-bit constants
2073         as source for store instructions.
2074
2075 2018-04-30  Jonathan Wakely  <jwakely@redhat.com>
2076
2077         * doc/invoke.texi (-fdebug-types-section): Fix grammar.
2078
2079 2018-04-30  Nathan Sidwell  <nathan@acm.org>
2080             Sandra Loosemore <sandra@codesourcery.com>
2081
2082         * dumpfile.c (dump_open): Allow '-' for stdout.
2083         * doc/invoke.texi (Developer Options): Document dump filename
2084         determination early.  Document stdin/stdout selection.
2085
2086 2018-04-30 Andrew Sadek  <andrew.sadek.se@gmail.com>
2087
2088         Microblaze Target: PIC data text relative
2089
2090         * config/microblaze/microblaze.opt: add new option -mpic-data-text-rel.
2091         * config/microblaze/microblaze-protos.h (microblaze_constant_address_p):
2092         Add declaration.
2093         * gcc/config/microblaze/microblaze.h (microblaze_constant_address_p):
2094         CONSTANT_ADDRESS_P definition to microblaze_constant_address_p.
2095         * config/microblaze/microblaze.c (TARGET_PIC_DATA_TEXT_REL):
2096         New addressing mode for data-text relative position indepenedent code.
2097         (microblaze_classify_unspec): add 'UNSPEC_TEXT' case ->
2098         'ADDRESS_SYMBOLIC_TXT_REL'.
2099         (microblaze_classify_address): Add handling for UNSPEC + CONST_INT.
2100         (microblaze_legitimate_pic_operand): Exclude function calls from
2101         pic operands in case of TARGET_PIC_DATA_TEXT_REL option.
2102         (microblaze_legitimize_address): Generate 'UNSPEC_TEXT' for all possible
2103         addresses cases.
2104         (microblaze_address_insns): Add 'ADDRESS_SYMBOLIC_TXT_REL' case.
2105         (print_operand): Add 'ADDRESS_SYMBOLIC_TXT_REL' case.
2106         (print_operand_address): Add 'ADDRESS_SYMBOLIC_TXT_REL' case + handling
2107         for 'address + offset'.
2108         (microblaze_expand_prologue): Add new function prologue call for
2109         'r20' assignation.
2110         (microblaze_asm_generate_pic_addr_dif_vec): Override new target hook
2111         'TARGET_ASM_GENERATE_PIC_ADDR_DIFF_VEC' to disable address diff vector
2112         table in case of TARGET_PIC_DATA_TEXT_REL.
2113         (expand_pic_symbol_ref): Add handling for 'UNSPEC_TEXT'.
2114         * gcc/config/microblaze/microblaze.md (TARGET_PIC_DATA_TEXT_REL):
2115         Add new macros 'UNSPEC_TEXT',
2116         'UNSPEC_SET_TEXT' + add rule for setting r20 in function prologue
2117         + exclude function calls from 'UNSPEC_PLT' in case of data text
2118         relative mode.
2119         * doc/tm.texi.in (TARGET_ASM_GENERATE_PIC_ADDR_DIFF_VEC): Add
2120         new target hook for generating address diff vector tables in case of
2121         flag_pic.
2122         * doc/tm.texi : Regenerate.
2123         * stmt.c (TARGET_ASM_GENERATE_PIC_ADDR_DIFF_VEC): Append new condition
2124         'targetm.asm_out.generate_pic_addr_diff_vec' to flag_pic in case
2125         of addr diff vector generation.
2126         * target.def (TARGET_ASM_GENERATE_PIC_ADDR_DIFF_VEC): Add
2127         target hook definition.
2128         * targhooks.h, gcc/targhooks.c (TARGET_ASM_GENERATE_PIC_ADDR_DIFF_VEC):
2129         Add default function for generate_pic_addr_diff_vec -> flag_pic.
2130         * doc/invoke.texi (Add new pic option): Add new microblaze pic
2131         option for data text relative.
2132
2133 2018-04-30  Richard Biener  <rguenther@suse.de>
2134
2135         * tree-chrec.h (evolution_function_is_constant_p): Remove
2136         redundant check.
2137         * tree-cfg.c (tree_node_can_be_shared): Re-order checks.
2138
2139 2018-04-30  Richard Biener  <rguenther@suse.de>
2140
2141         PR bootstrap/85571
2142         * dwarf2out.c (gen_producer_string): Ignore -fchecking[=].
2143
2144 2018-04-30  Richard Biener  <rguenther@suse.de>
2145
2146         PR tree-optimization/28364
2147         PR tree-optimization/85275
2148         * tree-ssa-loop-ch.c (ch_base::copy_headers): Stop after
2149         copying first exit test.
2150
2151 2018-04-28  Mark Wielaard  <mark@klomp.org>
2152
2153         * dwarf2out.c (dwarf2out_finish): Add .debug_addr table header for
2154         dwarf_version >= 5.
2155         (dwarf_AT): Handle DW_AT_addr_base.
2156         (add_top_level_skeleton_die_attrs): Use dwarf_AT for DW_AT_addr_base.
2157
2158 2018-04-28  Uros Bizjak  <ubizjak@gmail.com>
2159
2160         PR target/84431
2161         * config/i386/i386.md (*ashl<dwi>3_doubleword_mask): New pattern.
2162         (*ashl<dwi>3_doubleword_mask_1): Ditto.
2163         (*<shift_insn><dwi>3_doubleword_mask): Ditto.
2164         (*<shift_insn><dwi>3_doubleword_mask_1): Ditto.
2165
2166 2018-04-28  Richard Biener  <rguenther@suse.de>
2167
2168         * tree-cfg.c (verify_gimple_phi): Take a gphi * argument.
2169         (verify_gimple_in_cfg): Rename visited_stmts to visited_throwing_stmts
2170         to reflect use.  Only add interesting stmts.
2171
2172 2018-04-27  Martin Jambor  <mjambor@suse.cz>
2173
2174         PR ipa/85549
2175         * ipa-cp.c (find_aggregate_values_for_callers_subset): Make sure
2176         the jump function allows for passing through aggregate values.
2177
2178 2018-04-27  David Malcolm  <dmalcolm@redhat.com>
2179
2180         * input.h (in_system_header_at): Convert from macro to inline
2181         function.
2182         (from_macro_expansion_at): Likewise.
2183         (from_macro_definition_at): Likewise.
2184
2185 2018-04-27  Jeff Law  <law@redhat.com>
2186
2187         * config.gcc: Mark tile* targets as deprecated/obsolete.
2188
2189 2018-04-27  Richard Biener  <rguenther@suse.de>
2190
2191         * config/aarch64/aarch64.c: Simplify ap.__stack advance and
2192         fix for ILP32.
2193
2194 2018-04-27  Richard Biener  <rguenther@suse.de>
2195
2196         * tree-cfg.c (verify_expr): Make dead code hit gcc_unreachable.
2197
2198 2018-04-27  Uros Bizjak  <ubizjak@gmail.com>
2199
2200         * config/i386/i386.md (*movti_internal): Substitute Ye constraint
2201         with Yd constraint. Set "preferred_for_speed" attribute from
2202         TARGET_INTER_UNIT_MOVES_{FROM,TO}_VEC for alternatives
2203         with Yd constraint.
2204         (*movdi_internal): Ditto.
2205         (movti_interunit splitters): Remove
2206         TARGET_INTER_UNIT_MOVES_{FROM,TO}_VEC from insn condition.
2207         (movdi_interunit splitters): Ditto.
2208         * config/i386/constraints.md (Ye): Remove.
2209         (Yd): Do not depend on TARGET_INTER_UNIT_MOVES_{FROM,TO}_VEC.
2210
2211 2018-04-27  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
2212
2213         PR target/85512
2214         * config/aarch64/constraints.md (Usg): Limit to 31.
2215         (Usj): Limit to 63.
2216
2217 2018-04-27  Jakub Jelinek  <jakub@redhat.com>
2218
2219         PR tree-optimization/85529
2220         * tree-ssa-reassoc.c (optimize_range_tests_var_bound): Add FIRST_BB
2221         argument.  Don't call get_nonzero_bits if opcode is ERROR_MARK_NODE,
2222         rhs2 def stmt's bb is dominated by first_bb and it isn't an obvious
2223         zero extension or masking of the MSB bit.
2224         (optimize_range_tests): Add FIRST_BB argument, pass it through
2225         to optimize_range_tests_var_bound.
2226         (maybe_optimize_range_tests, reassociate_bb): Adjust
2227         optimize_range_tests callers.
2228
2229 2018-04-26  Richard Biener  <rguenther@suse.de>
2230             Jakub Jelinek  <jakub@redhat.com>
2231
2232         * cgraph.h (symbol_table): Just declare debug method here.
2233         * symtab.c (symbol_table::debug): Define.
2234
2235 2018-04-26  Eric Botcazou  <ebotcazou@adacore.com>
2236
2237         * loop-invariant.c (may_assign_reg_p): Return false for frame pointer.
2238
2239 2018-04-26  Uros Bizjak  <ubizjak@gmail.com>
2240
2241         * config/i386/i386.md ("isa" attribute): Add x64_sse2.
2242         ("enabled" attribute): Handle x64_sse2 "isa" attribute.
2243         (*movdi_internal): Substitute Yi and Yj constraint with x
2244         and Ym and Yn constraint with y constraint.  Update "isa"
2245         attribute and set "preferred_for_speed" attribute from
2246         TARGET_INTER_UNIT_MOVES_{FROM,TO}_VEC for updated alternatives.
2247         (*movsi_internal): Ditto.
2248         (*movdf_internal): Ditto.
2249         (*movsf_internal): Ditto.
2250         (*zero_extendsidi2): Ditto.
2251         * config/i386/sse.md (vec_set<mode>_0): Ditto.
2252         (sse2_loadld): Ditto.
2253         (*vec_extract<ssevecmodelower>_0): Ditto.
2254         (*vec_extractv4si_0_zext_sse4): Ditto.
2255         (vec_concatv2di): Ditto.
2256         (*vec_dup<mode>): Ditto.
2257         * config/i386/mmx.md (*mov<mode>_internal): Ditto.
2258         * config/i386/constraints.md (Yi): Remove.
2259         (Yj): Remove.
2260         (Ym): Remove.
2261         (Yn): Remove.
2262
2263 2018-04-26  Nathan Sidwell  <nathan@acm.org>
2264
2265         * dumpfile.c (dump_open): New.
2266         (dump_open_alternate_stream, dump_start, dump_begin): Call it.
2267         (dump_finish): Detect stdio/stderr by value not name.
2268
2269 2018-04-26  Jonathan Wakely  <jwakely@redhat.com>
2270
2271         * doc/invoke.texi (-Wreturn-type): Document default status for C++.
2272
2273 2018-04-26  Tom de Vries  <tom@codesourcery.com>
2274
2275         PR target/84952
2276         * config/nvptx/nvptx.c (verify_neutering_jumps)
2277         (verify_neutering_labels): New function
2278         (nvptx_single): Use verify_neutering_jumps and verify_neutering_labels.
2279
2280 2018-04-26  Tom de Vries  <tom@codesourcery.com>
2281
2282         PR target/84025
2283         * config/nvptx/nvptx.c (needs_neutering_p): New function.
2284         (nvptx_single): Use needs_neutering_p to skip over insns that do not
2285         need neutering.
2286
2287 2018-04-26  Richard Biener <rguenther@suse.de>
2288             Tom de Vries  <tom@codesourcery.com>
2289
2290         PR lto/85422
2291         * lto-streamer-out.c (output_function): Fixup loops if required to match
2292         discovery done in the reader.
2293
2294 2018-04-26  Richard Biener  <rguenther@suse.de>
2295
2296         PR tree-optimization/85116
2297         * tree-ssa-loop-ch.c (do_while_loop_p): A do-while loop should
2298         have a loop exit from the single latch predecessor.  Remove
2299         case of header with just condition.
2300         (ch_base::copy_headers): Exclude infinite loops from any
2301         processing.
2302         (pass_ch::execute): Record exits.
2303
2304 2018-04-26  Richard Biener  <rguenther@suse.de>
2305
2306         * tree-vect-data-refs.c (vect_get_data_access_cost): Get
2307         prologue cost vector and pass it to vect_get_load_cost.
2308         (vect_get_peeling_costs_all_drs): Likewise.
2309         (vect_peeling_hash_get_lowest_cost): Likewise.
2310         (vect_enhance_data_refs_alignment): Likewise.
2311
2312 2018-04-26  Richard Biener  <rguenther@suse.de>
2313
2314         PR middle-end/85450
2315         * tree-cfg.c (verify_gimple_assign_unary): Restore proper
2316         checking of integer<->pointer conversions.
2317         * omp-expand.c (expand_omp_for_static_nochunk): Avoid
2318         sign-/zero-extending pointer types.
2319         (expand_omp_for_static_chunk): Likewise.
2320
2321 2018-03-22  Hans-Peter Nilsson  <hp@axis.com>
2322             Jean Lee  <xiaoyur347@gmail.com>
2323
2324         * config/mips/mips.c (mips_asan_shadow_offset): New function.
2325         (TARGET_ASAN_SHADOW_OFFSET): Define.
2326         * config/mips/mips.h (FRAME_GROWS_DOWNWARD): Augment to also be
2327         true for -fsanitize=address.
2328
2329 2018-04-25  Mark Wielaard  <mark@klomp.org>
2330
2331         * dwarf2out.c (file_info_cmp): Sort longer dir prefixes before
2332         shorter ones.
2333
2334 2018-04-25  Jakub Jelinek  <jakub@redhat.com>
2335
2336         * config/i386/i386.md (*x86_mov<mode>cc_0_m1): Use type "alu1" rather
2337         than "alu", remove explicit "memory" and "imm_disp" attributes.
2338         (*x86_mov<mode>cc_0_m1_se, *x86_mov<mode>cc_0_m1_neg): Likewise.
2339
2340         PR middle-end/85414
2341         * simplify-rtx.c (simplify_unary_operation_1) <case SIGN_EXTEND,
2342         case ZERO_EXTEND>: Pass SUBREG_REG (op) rather than op to
2343         gen_lowpart_no_emit.
2344
2345 2018-04-25  Sebastian Peryt  <sebastian.peryt@intel.com>
2346
2347         PR target/85473
2348         * config/i386/i386.c (ix86_expand_builtin): Change memory
2349         operand to XI, extend p0 to Pmode.
2350         * config/i386/i386.md: Change unspec volatile and operand
2351         1 mode to XI, change operand 0 mode to P.
2352
2353 2018-04-25  Chung-Ju Wu  <jasonwucj@gmail.com>
2354
2355         * config/nds32/nds32-predicates.c (nds32_can_use_bclr_p): Mask with
2356         GET_MODE_MASK before any checking.
2357         (nds32_can_use_bset_p): Likewise.
2358         (nds32_can_use_btgl_p): Likewise.
2359
2360 2018-04-25  Chung-Ju Wu  <jasonwucj@gmail.com>
2361
2362         * config/nds32/nds32-doubleword.md: New define_split pattern for
2363         illegal register number.
2364
2365 2018-04-25  Chung-Ju Wu  <jasonwucj@gmail.com>
2366
2367         * config/nds32/nds32.c (nds32_print_operand): Set op_value ealier.
2368
2369 2018-04-25  Chung-Ju Wu  <jasonwucj@gmail.com>
2370
2371         * config/nds32/nds32.h (ASM_APP_ON): Add missing newline character.
2372
2373 2018-04-25  Richard Biener  <rguenther@suse.de>
2374
2375         * lto-streamer.h (LTO_major_version): Bump to 8.
2376
2377 2018-04-25  Jakub Jelinek  <jakub@redhat.com>
2378
2379         * BASE-VER: Set to 9.0.0.
2380
2381 2018-04-24  Segher Boessenkool  <segher@kernel.crashing.org>
2382
2383         * config/rs6000/rs6000.c (init_float128_ieee): Fix spelling mistakes
2384         in __abskf2 and __powikf2.
2385
2386 2018-04-24  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
2387
2388         PR target/85512
2389         * config/aarch64/constraints.md (Usg, Usj): New constraints.
2390         * config/aarch64/iterators.md (cmode_simd): New mode attribute.
2391         * config/aarch64/aarch64.md (*aarch64_ashr_sisd_or_int_<mode>3):
2392         Use the above on operand 2.  Reindent.
2393         (*aarch64_lshr_sisd_or_int_<mode>3): Likewise.
2394
2395 2018-04-24  H.J. Lu  <hongjiu.lu@intel.com>
2396
2397         PR target/85485
2398         * common/config/i386/i386-common.c (ix86_handle_option): Don't
2399         handle OPT_mcet.
2400         * config/i386/i386.opt (mcet): Removed.
2401         * doc/install.texi: Remove -mcet documentation.
2402         * doc/invoke.texi: Likewise.
2403
2404 2018-04-24  H.J. Lu  <hongjiu.lu@intel.com>
2405
2406         PR target/85485
2407         * doc/install.texi: Remove -mcet from bootstrap-cet.
2408
2409 2018-04-24  Jakub Jelinek  <jakub@redhat.com>
2410
2411         PR target/85511
2412         * config/i386/i386.c (ix86_init_mmx_sse_builtins): Don't define
2413         __builtin_ia32_readeflags_u32 and __builtin_ia32_writeeflags_u32
2414         if TARGET_64BIT.
2415
2416         PR target/85503
2417         * config/rs6000/rs6000-p8swap.c (const_load_sequence_p): Punt if
2418         const_vector is not CONST_VECTOR or SYMBOL_REF for a constant pool
2419         containing a CONST_VECTOR.
2420
2421 2018-04-24  Cesar Philippidis  <cesar@codesourcery.com>
2422
2423         * doc/install.texi: Update newlib dependency for nvptx.
2424
2425 2018-04-24  Jakub Jelinek  <jakub@redhat.com>
2426
2427         PR target/85508
2428         * config/i386/i386.c (ix86_expand_vector_init_one_var): Use UINTVAL
2429         instead of INTVAL when shifting x left.
2430
2431 2018-04-24  Andreas Krebbel  <krebbel@linux.ibm.com>
2432
2433         PR tree-optimization/85478
2434         * tree-vect-loop.c (vect_analyze_loop_2): Do not call
2435         vect_grouped_store_supported for single element vectors.
2436
2437 2018-04-24  Richard Biener  <rguenther@suse.de>
2438
2439         PR target/85491
2440         * config/i386/i386.c (ix86_add_stmt_cost): Restrict strided
2441         load cost increase to the case of non-constant step.
2442
2443 2018-04-24  Jakub Jelinek  <jakub@redhat.com>
2444
2445         PR target/84828
2446         * reg-stack.c (move_for_stack_reg): Don't fail assertion about dead
2447         destination if any_malformed_asm.
2448
2449 2018-04-23  Eric Botcazou  <ebotcazou@adacore.com>
2450
2451         PR middle-end/85496
2452         * expr.c (store_field): In the bitfield case, if the value comes from
2453         a function call and is returned in registers by means of a PARALLEL,
2454         do not change the mode of the temporary unless BLKmode and VOIDmode.
2455
2456 2018-04-23  Andrey Belevantsev  <abel@ispras.ru>
2457
2458         PR rtl-optimization/85423
2459         * sel-sched-ir.c (has_dependence_note_mem_dep): Only discard
2460         dependencies to debug insns when the previous insn is non-debug.
2461
2462 2018-04-23  Claudiu Zissulescu  <claziss@synopsys.com>
2463
2464         * config/arc/arc.md ("vunspec"): Delete it, unify all the unspec
2465         enums into a single definition.
2466         (fls): Fix predicates and printing.
2467         (seti): Likewise.
2468
2469 2018-04-23  Claudiu Zissulescu  <claziss@synopsys.com>
2470
2471         * config/arc/arc-protos.h (check_if_valid_sleep_operand): Remove.
2472         * config/arc/arc.c (arc_expand_builtin): Sleep accepts registers
2473         and short u6 immediate.
2474         (check_if_valid_sleep_operand): Remove.
2475         * config/arc/arc.md (Sleep): Accepts registers and u6 immediates.
2476
2477 2018-04-22  Chung-Ju Wu  <jasonwucj@gmail.com>
2478
2479         * config/nds32/nds32.c (nds32_compute_stack_frame): Consider
2480         flag_always_save_lp condition.
2481         * config/nds32/nds32.opt (malways-save-lp): New option.
2482
2483 2018-04-22  Shiva Chen  <shiva0217@gmail.com>
2484
2485         * config/nds32/nds32-protos.h (nds32_use_load_post_increment): Declare.
2486         * config/nds32/nds32.c (nds32_use_load_post_increment): New.
2487         * config/nds32/nds32.h
2488         (USE_LOAD_POST_INCREMENT, USE_LOAD_POST_DECREMENT): Define.
2489         (USE_STORE_POST_INCREMENT, USE_STORE_POST_DECREMENT): Define.
2490
2491 2018-04-22  Shiva Chen  <shiva0217@gmail.com>
2492
2493         * config/nds32/nds32-protos.h (nds32_ls_333_p): Remove.
2494         * config/nds32/nds32.c (nds32_ls_333_p): Remove.
2495
2496 2018-04-22  Shiva Chen  <shiva0217@gmail.com>
2497             Chung-Ju Wu  <jasonwucj@gmail.com>
2498
2499         * config/nds32/nds32-protos.h (nds32_case_vector_shorten_mode):
2500         Declare.
2501         * config/nds32/nds32.c (nds32_case_vector_shorten_mode): New function.
2502         * config/nds32/nds32.h (CASE_VECTOR_SHORTEN_MODE): Modify.
2503
2504 2018-04-22  Chung-Ju Wu  <jasonwucj@gmail.com>
2505
2506         * config/nds32/nds32.c (nds32_compute_stack_frame): Fix wrong value.
2507
2508 2018-04-22  Chung-Ju Wu  <jasonwucj@gmail.com>
2509
2510         * config/nds32/nds32-protos.h (nds32_data_alignment,
2511         nds32_local_alignment): Declare.
2512         * config/nds32/nds32.c (nds32_data_alignment, nds32_constant_alignment,
2513         nds32_local_alignment): New functions.
2514         (TARGET_CONSTANT_ALIGNMENT): Define.
2515         * config/nds32/nds32.h (DATA_ALIGNMENT, LOCAL_ALIGNMENT): Define.
2516
2517 2018-04-22  Chung-Ju Wu  <jasonwucj@gmail.com>
2518
2519         * config/nds32/nds32.c
2520         (TARGET_HARD_REGNO_MODE_OK): Move to the bottom of file.
2521         (TARGET_MODES_TIEABLE_P): Likewise.
2522
2523 2018-04-22  Chung-Ju Wu  <jasonwucj@gmail.com>
2524
2525         * config/nds32/nds32.c (nds32_asm_file_start): Display optimization
2526         level Ofast and Og.
2527
2528 2018-04-22  Monk Chiang  <sh.chiang04@gmail.com>
2529             Chung-Ju Wu  <jasonwucj@gmail.com>
2530
2531         * config/nds32/constants.md (unspec_volatile_element): Add enum values
2532         for unaligned access.
2533         * config/nds32/nds32-intrinsic.c: Implementation of expanding
2534         unaligned access.
2535         * config/nds32/nds32-intrinsic.md: Likewise.
2536         * config/nds32/nds32_intrinsic.h: Likewise.
2537         * config/nds32/nds32.h (nds32_builtins): Likewise.
2538         * config/nds32/nds32.opt (munaligned-access): New option.
2539         * config/nds32/nds32.c (nds32_asm_file_start): Display
2540         flag_unaligned_access status.
2541
2542 2018-04-20  Kito Cheng  <kito.cheng@gmail.com>
2543
2544         * config/riscv/elf.h (LINK_SPEC): Pass --no-relax if
2545         -mno-relax is present.
2546         * config/riscv/linux.h (LINK_SPEC): Ditto.
2547
2548 2018-04-20  Martin Sebor  <msebor@redhat.com>
2549
2550         PR c/85365
2551         * gimple-fold.c (gimple_fold_builtin_strcpy): Suppress -Wrestrict
2552         for null pointers.
2553         (gimple_fold_builtin_stxcpy_chk): Same.
2554         * gimple-ssa-warn-restrict.c (check_bounds_or_overlap): Same.
2555
2556 2018-04-20  Michael Meissner  <meissner@linux.ibm.com>
2557
2558         PR target/85456
2559         * config/rs6000/rs6000.c (init_float128_ieee): Add support to call
2560         __powikf2 when long double is IEEE 128-bit.
2561
2562 2018-04-20  Kito Cheng  <kito.cheng@gmail.com>
2563
2564         * config/riscv/riscv.c (riscv_first_stack_step): Round up min
2565         step to make sure stack always aligned.
2566
2567 2018-04-20  Carl Love  <cel@us.ibm.com>
2568
2569         PR target/83402
2570         * config/rs6000/rs6000-c.c (rs6000_gimple_fold_builtin): Add
2571         size check for arg0.
2572
2573 2018-04-20  Nathan Sidwell  <nathan@codesourcery.com>
2574             Tom de Vries  <tom@codesourcery.com>
2575
2576         PR target/85445
2577         * config/nvptx/nvptx.c (nvptx_emit_forking, nvptx_emit_joining):
2578         Emit insns for calls too.
2579         (nvptx_find_par): Always look for worker-level predecessor insn.
2580         (nvptx_propagate): Add is_call parm, return bool.  Copy frame for
2581         calls.
2582         (nvptx_vpropagate, nvptx_wpropagate): Adjust.
2583         (nvptx_process_pars): Propagate frames for calls.
2584
2585 2018-04-20  H.J. Lu  <hongjiu.lu@intel.com>
2586
2587         PR target/85469
2588         * common/config/i386/i386-common.c (OPTION_MASK_ISA_IBT_SET):
2589         Removed.
2590         (OPTION_MASK_ISA_IBT_UNSET): Likewise.
2591         (ix86_handle_option): Don't handle OPT_mibt.
2592         * config/i386/cet.h: Check __CET__ instead of __IBT__ and
2593         __SHSTK__.
2594         * config/i386/driver-i386.c (host_detect_local_cpu): Remove
2595         has_ibt and ibt.
2596         * config/i386/i386-c.c (ix86_target_macros_internal): Don't
2597         check OPTION_MASK_ISA_IBT nor flag_cf_protection.
2598         (ix86_target_macros): Define __CET__ with flag_cf_protection
2599         for -fcf-protection.
2600         * config/i386/i386.c (isa2_opts): Remove -mibt.
2601         * config/i386/i386.h (TARGET_IBT): Removed.
2602         (TARGET_IBT_P): Likewise.
2603         (ix86_valid_target_attribute_inner_p): Don't check OPT_mibt.
2604         * config/i386/i386.md (nop_endbr): Don't check TARGET_IBT.
2605         * config/i386/i386.opt (mcet): Update help message.
2606         (mshstk): Likewise.
2607         (mibt): Removed.
2608         * doc/invoke.texi: Remove -mibt.  Document __CET__.  Document
2609         -mcet as an alias for -mshstk.
2610
2611 2018-04-20  Richard Biener <rguenther@suse.de>
2612
2613         PR middle-end/85475
2614         * match.pd ((X * CST) * Y -> (X * Y) * CST): Avoid exponential
2615         complexity by forcing a single use of the multiply operand.
2616
2617 2018-04-20  Martin Jambor  <mjambor@suse.cz>
2618
2619         ipa/85449
2620         * ipa-cp.c (cgraph_edge_brings_value_p): Move check for self-feeding
2621         recursion dependency to only apply to non-clones.
2622
2623 2018-04-20  Martin Jambor  <mjambor@suse.cz>
2624
2625         ipa/85447
2626         * ipa-cp.c (create_specialized_node): Check that clones of
2627         self-recursive edges exist during IPA-CP.
2628
2629 2018-04-19  Toon Moene  <toon@moene.org>
2630
2631         * doc/invoke.texi: Add -floop-unroll-and-jam to options enabled
2632         by -O3.
2633
2634 2018-04-19  Jakub Jelinek  <jakub@redhat.com>
2635
2636         PR tree-optimization/85467
2637         * fold-const.c (fold_ternary_loc) <case BIT_FIELD_REF>: Use
2638         VECTOR_TYPE_P macro.  If type is vector type, VIEW_CONVERT_EXPR the
2639         VECTOR_CST element to type.
2640
2641 2018-04-19  H.J. Lu  <hongjiu.lu@intel.com>
2642
2643         PR target/85397
2644         * config/i386/i386.h (STACK_SAVEAREA_MODE): New.
2645         * config/i386/i386.md (builtin_setjmp_setup): Removed.
2646         (builtin_longjmp): Likewise.
2647         (save_stack_nonlocal): New pattern.
2648         (restore_stack_nonlocal): Likewise.
2649
2650 2018-04-19  H.J. Lu  <hongjiu.lu@intel.com>
2651
2652         PR target/85404
2653         * config/i386/cet.c (file_end_indicate_exec_stack_and_cet):
2654         Replace ASM_OUTPUT_LABEL with fprintf.
2655
2656 2018-04-19  H.J. Lu  <hongjiu.lu@intel.com>
2657
2658         PR target/85417
2659         * config/i386/cet.c (file_end_indicate_exec_stack_and_cet):
2660         Check flag_cf_protection instead of TARGET_IBT and TARGET_SHSTK.
2661         * config/i386/i386-c.c (ix86_target_macros_internal): Also
2662         define __IBT__ and __SHSTK__ for -fcf-protection.
2663         * config/i386/i386.c (pass_insert_endbranch::gate): Don't check
2664         TARGET_IBT.
2665         (ix86_trampoline_init): Likewise.
2666         (x86_output_mi_thunk): Likewise.
2667         (ix86_notrack_prefixed_insn_p): Likewise.
2668         (ix86_option_override_internal): Don't disallow -fcf-protection.
2669         * config/i386/i386.md (rdssp<mode>): Also enable for
2670         -fcf-protection.
2671         (incssp<mode>): Likewise.
2672         (nop_endbr): Likewise.
2673         * config/i386/i386.opt (mcet): Change help message to built-in
2674         functions only.
2675         (mibt): Likewise.
2676         (mshstk): Likewise.
2677         * doc/invoke.texi: Remove -mcet, -mibt and -mshstk condition
2678         on -fcf-protection.  Change -mcet, -mibt and -mshstk to only
2679         enable CET built-in functions.
2680
2681 2018-04-19  Sebastian Peryt  <sebastian.peryt@intel.com>
2682
2683         * common/config/i386/i386-common.c
2684         (OPTION_MASK_ISA_MOVDIRI_SET, OPTION_MASK_ISA_MOVDIR64B_SET,
2685         OPTION_MASK_ISA_MOVDIRI_UNSET,
2686         OPTION_MASK_ISA_MOVDIR64B_UNSET): New defines.
2687         (ix86_handle_option): Handle -mmovdiri and -mmovdir64b.
2688         * config.gcc (movdirintrin.h): New header.
2689         * config/i386/cpuid.h (bit_MOVDIRI,
2690         bit_MOVDIR64B): New bits.
2691         * config/i386/driver-i386.c (host_detect_local_cpu): Detect -mmovdiri
2692         and -mmvodir64b.
2693         * config/i386/i386-builtin-types.def ((VOID, PUNSIGNED, UNSIGNED),
2694         (VOID, PVOID, PCVOID)): New function types.
2695         * config/i386/i386-builtin.def (__builtin_ia32_directstoreu_u32,
2696         __builtin_ia32_directstoreu_u64,
2697          __builtin_ia32_movdir64b): New builtins.
2698         * config/i386/i386-c.c (__MOVDIRI__, __MOVDIR64B__): New.
2699         * config/i386/i386.c (ix86_target_string): Added -mmovdir64b
2700         and -mmovdiri.
2701         (ix86_valid_target_attribute_inner_p): Ditto.
2702         (ix86_expand_special_args_builtin): Added VOID_FTYPE_PUNSIGNED_UNSIGNED
2703         and VOID_FTYPE_PUNSIGNED_UNSIGNED.
2704         (ix86_expand_builtin): Expand IX86_BUILTIN_MOVDIR64B.
2705         * config/i386/i386.h (TARGET_MOVDIRI, TARGET_MOVDIRI_P,
2706         TARGET_MOVDIR64B, TARGET_MOVDIR64B_P): New.
2707         * config/i386/i386.md (UNSPECV_MOVDIRI, UNSPECV_MOVDIR64B): New.
2708         (movdiri<mode>, movdir64b_<mode>): New.
2709         * config/i386/i386.opt: Add -mmovdiri and -mmovdir64b.
2710         * config/i386/immintrin.h: Include movdirintrin.h.
2711         * config/i386/movdirintrin.h: New file.
2712         * doc/invoke.texi: Added -mmovdiri and -mmovdir64b.
2713
2714 2018-04-19  Richard Biener  <rguenther@suse.de>
2715
2716         PR middle-end/85455
2717         * cfg.c (clear_bb_flags): When loop state says we have
2718         marked irreducible regions also preserve BB_IRREDUCIBLE_LOOP.
2719
2720 2018-04-19  Richard Biener  <rguenther@suse.de>
2721
2722         PR tree-optimization/84737
2723         * tree-vect-data-refs.c (vect_copy_ref_info): New function
2724         copying restrict info.
2725         (vect_setup_realignment): Use it.
2726         * tree-vectorizer.h (vect_copy_ref_info): Declare.
2727         * tree-vect-stmts.c (vectorizable_store): Copy ref info from
2728         the first DR to all generated stores.
2729         (vectorizable_load): Likewise for loads.
2730
2731 2018-04-19  Jakub Jelinek  <jakub@redhat.com>
2732
2733         PR tree-optimization/85446
2734         * match.pd ((intptr_t) x eq/ne CST to x eq/ne (typeof x) cst): Require
2735         the integral and pointer types to have the same precision.
2736
2737         * doc/install.texi: Document --disable-cet being the default and
2738         --enable-cet=auto.
2739
2740 2018-04-18  Martin Liska  <mliska@suse.cz>
2741
2742         * ipa-devirt.c (odr_subtypes_equivalent_p): Fix GNU coding
2743         style.
2744
2745 2018-04-18  Martin Liska  <mliska@suse.cz>
2746
2747         Revert
2748         2018-03-02  Eric Botcazou  <ebotcazou@adacore.com>
2749
2750         PR ipa/83983
2751         * ipa-devirt.c (odr_subtypes_equivalent_p): Get the ODR type of both
2752         arguments if they are comparable.
2753
2754 2018-04-18  Martin Liska  <mliska@suse.cz>
2755
2756         Revert
2757         2018-03-13  Eric Botcazou  <ebotcazou@adacore.com>
2758
2759         PR lto/84805
2760         * ipa-devirt.c (odr_subtypes_equivalent_p): Do not get the ODR type of
2761         incomplete types.
2762
2763 2018-04-18  H.J. Lu  <hongjiu.lu@intel.com>
2764
2765         PR target/85388
2766         * config/i386/i386.c (ix86_expand_split_stack_prologue): Insert
2767         ENDBR after calling __morestack.
2768
2769 2018-04-18  David Malcolm  <dmalcolm@redhat.com>
2770
2771         PR jit/85384
2772         * configure.ac (gcc-driver-name.h): Honor --with-gcc-major-version
2773         by using gcc_base_ver to generate a gcc_driver_version, and use
2774         it when generating GCC_DRIVER_NAME.
2775         * configure: Regenerate.
2776
2777 2018-04-18  Jakub Jelinek  <jakub@redhat.com>
2778
2779         PR target/81084
2780         * config.gcc: Obsolete powerpc*-*-*spe*.
2781
2782 2018-04-17  Jakub Jelinek  <jakub@redhat.com>
2783
2784         PR debug/84637
2785         * dbxout.c (dbxout_int): Perform negation in unsigned int type.
2786         (stabstr_D): Change type of unum from unsigned int to
2787         unsigned HOST_WIDE_INT.  Perform negation in unsigned HOST_WIDE_INT
2788         type.
2789
2790 2018-04-17  Jim Wilson  <jimw@sifive.com>
2791
2792         PR 84856
2793         * config/riscv/riscv.c (riscv_compute_frame_info): Add calls to
2794         RISCV_STACK_ALIGN when using outgoing_args_size and pretend_args_size.
2795         Set arg_pointer_offset after using pretend_args_size.
2796
2797 2018-04-17  Jakub Jelinek  <jakub@redhat.com>
2798
2799         PR rtl-optimization/85431
2800         * dse.c (record_store): Ignore zero width stores.
2801
2802         PR sanitizer/85230
2803         * asan.c (handle_builtin_stack_restore): Adjust comment.  Emit
2804         __asan_allocas_unpoison call and last_alloca_addr = new_sp before
2805         __builtin_stack_restore rather than after it.
2806         * builtins.c (expand_asan_emit_allocas_unpoison): Pass
2807         arg1 + (virtual_dynamic_stack_rtx - stack_pointer_rtx) as second
2808         argument instead of virtual_dynamic_stack_rtx.
2809
2810 2018-04-17  Kelvin Nilsen  <kelvin@gcc.gnu.org>
2811
2812         * config/rs6000/rs6000-protos.h (rs6000_builtin_is_supported_p):
2813         New prototype.
2814         * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin):
2815         Add note to error message to explain internal mapping of overloaded
2816         built-in function name to non-overloaded built-in function name.
2817         * config/rs6000/rs6000.c (rs6000_builtin_is_supported_p): New
2818         function.
2819
2820 2018-04-17  Michael Meissner  <meissner@linux.vnet.ibm.com>
2821
2822         PR target/85424
2823         * config/rs6000/rs6000.md (pack<mode>): Do not try handle a pack
2824         where the inputs overlap with the output.
2825
2826 2018-04-17  Jakub Jelinek  <jakub@redhat.com>
2827
2828         * config/i386/sse.md (vec_extract_lo_<mode><mask_name>): Add
2829         (=v, v) alternative and explicit "memory" attribute.
2830         (vec_extract_lo_<mode><mask_name>): Likewise.  Also add
2831         "type", "prefix", "prefix_extra", "length_immediate" and "mode"
2832         attributes.
2833         (vec_extract_lo_<mode><mask_name>): Add (=v, v) alternative and use
2834         "sselog1" type instead of "sselog".
2835         (vec_extract_hi_<mode><mask_name>): Use "sselog1" type instead of
2836         "sselog".  Remove explicit "memory" attribute.
2837         (vec_extract_lo_v32hi): Add (=v, v) alternative and explicit "memory",
2838         "type", "prefix", "prefix_extra", "length_immediate" and "mode"
2839         attributes.
2840         (vec_extract_hi_v32hi): Merge all alternatives into one, use
2841         "sselog1" type instead of "sselog".  Remove explicit "memory"
2842         attribute.
2843         (vec_extract_hi_v16hi): Merge each pair of alternatives into one,
2844         use "sselog1" type instead of "sselog".  Remove explicit "memory"
2845         attribute.
2846         (vec_extract_lo_v64qi): Add (=v, v) alternative and explicit "memory",
2847         "type", "prefix", "prefix_extra", "length_immediate" and "mode"
2848         attributes.
2849         (vec_extract_hi_v64qi): Merge all alternatives into one, use
2850         "sselog1" type instead of "sselog".  Remove explicit "memory"
2851         attribute.
2852         (vec_extract_hi_v32qi): Merge each pair of alternatives into one,
2853         use "sselog1" type instead of "sselog".  Remove explicit "memory"
2854         attribute.
2855
2856         PR target/85430
2857         * config/i386/i386.md (*ashlqi3_1_slp): Use alu1 type instead of alu.
2858
2859         PR middle-end/85414
2860         * rtlhooks.c (gen_lowpart_if_possible): Don't call gen_lowpart_SUBREG
2861         on a SUBREG.
2862
2863 2018-04-17  Martin Jambor  <mjambor@suse.cz>
2864
2865         PR ipa/85421
2866         * ipa-cp.c (create_specialized_node): Call
2867         expand_all_artificial_thunks if necessary.
2868
2869 2018-04-17  Martin Liska  <mliska@suse.cz>
2870
2871         PR lto/85405
2872         * ipa-devirt.c (odr_types_equivalent_p): Remove trailing
2873         in message, remote space in between '_G' and '('.
2874
2875 2018-04-17  Jakub Jelinek  <jakub@redhat.com>
2876
2877         PR target/85281
2878         * config/i386/sse.md (reduces<mode><mask_scalar_name>,
2879         avx512f_vmcmp<mode>3<round_saeonly_name>,
2880         avx512f_vmcmp<mode>3_mask<round_saeonly_name>,
2881         avx512f_sgetexp<mode><mask_scalar_name><round_saeonly_scalar_name>,
2882         avx512f_rndscale<mode><round_saeonly_name>,
2883         avx512dq_ranges<mode><mask_scalar_name><round_saeonly_scalar_name>,
2884         avx512f_vgetmant<mode><mask_scalar_name><round_saeonly_scalar_name>):
2885         Use %<iptr>2 instead of %2 for -masm=intel.
2886         (avx512f_vcvtss2usi<round_name>, avx512f_vcvtss2usiq<round_name>,
2887         avx512f_vcvttss2usi<round_saeonly_name>,
2888         avx512f_vcvttss2usiq<round_saeonly_name>): Use %k1 instead of %1 for
2889         -masm=intel.
2890         (avx512f_vcvtsd2usi<round_name>, avx512f_vcvtsd2usiq<round_name>,
2891         avx512f_vcvttsd2usi<round_saeonly_name>,
2892         avx512f_vcvttsd2usiq<round_saeonly_name>, ufloatv2siv2df2<mask_name>):
2893         Use %q1 instead of %1 for -masm=intel.
2894         (avx512f_sfixupimm<mode><sd_maskz_name><round_saeonly_name>,
2895         avx512f_sfixupimm<mode>_mask<round_saeonly_name>): Use %<iptr>3 instead
2896         of %3 for -masm=intel.
2897         (sse2_shufpd_v2df_mask): Fix a typo, change %{6%} to %{%6%} for
2898         -masm=intel.
2899         (*avx512vl_<code>v2div2qi2_store): Use %w0 instead of %0 for
2900         -masm=intel.
2901         (*avx512vl_<code><mode>v4qi2_store): Use %k0 instead of %0 for
2902         -masm=intel.
2903         (avx512vl_<code><mode>v4qi2_mask_store): Use a single pattern with
2904         %k0 and %1 for -masm=intel rather than two patterns, one with %0 and
2905         %g1.
2906         (*avx512vl_<code><mode>v8qi2_store): Use %q0 instead of %0 for
2907         -masm=intel.
2908         (avx512vl_<code><mode>v8qi2_mask_store): Use a single pattern with
2909         %q0 and %1 for -masm=intel rather than two patterns, one with %0 and
2910         %g1 and one with %0 and %1.
2911         (avx512er_vmrcp28<mode><round_saeonly_name>,
2912         avx512er_vmrsqrt28<mode><round_saeonly_name>): Use %<iptr>1 instead of
2913         %1 for -masm=intel.
2914         (avx5124fmaddps_4fmaddps_mask, avx5124fmaddps_4fmaddss_mask,
2915         avx5124fmaddps_4fnmaddps_mask, avx5124fmaddps_4fnmaddss_mask,
2916         avx5124vnniw_vp4dpwssd_mask, avx5124vnniw_vp4dpwssds_mask): Swap order
2917         of %0 and %{%4%} for -masm=intel.
2918         (avx5124fmaddps_4fmaddps_maskz, avx5124fmaddps_4fmaddss_maskz,
2919         avx5124fmaddps_4fnmaddps_maskz, avx5124fmaddps_4fnmaddss_maskz,
2920         avx5124vnniw_vp4dpwssd_maskz, avx5124vnniw_vp4dpwssds_maskz): Swap
2921         order of %0 and %{%5%}%{z%} for -masm=intel.
2922
2923 2018-04-17  Jan Hubicka  <jh@suse.cz>
2924
2925         PR lto/85405
2926         * ipa-devirt.c (odr_types_equivalent_p): Handle bit fields.
2927
2928 2018-04-17  Martin Liska  <mliska@suse.cz>
2929
2930         PR ipa/85329
2931         * multiple_target.c (create_dispatcher_calls): Set apostrophes
2932         for target_clone error message.  Make default implementation
2933         clone to be a local declaration.
2934         (separate_attrs): Add new argument and check for an empty
2935         string.
2936         (expand_target_clones): Handle it.
2937         (ipa_target_clone): Make redirection just for target_clones
2938         functions.
2939
2940 2018-04-16  Cesar Philippidis  <cesar@codesourcery.com>
2941             Tom de Vries  <tom@codesourcery.com>
2942
2943         PR middle-end/84955
2944         * omp-expand.c (expand_oacc_for): Add dummy false branch for
2945         tiled basic blocks without omp continue statements.
2946
2947 2018-04-16  Aaron Sawdey  <acsawdey@linux.ibm.com>
2948
2949         PR target/83660
2950         * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Mark
2951         vec_extract expression as having side effects to make sure it gets
2952         a cleanup point.
2953
2954 2018-04-16  H.J. Lu  <hongjiu.lu@intel.com>
2955
2956         PR target/85403
2957         * config/i386/i386.c (get_builtin_code_for_version): Check
2958         error_mark_node.
2959
2960 2018-04-16  Olga Makhotina  <olga.makhotina@intel.com>
2961
2962         PR target/84331
2963         * gcc/config.gcc: Support "skylake".
2964         * gcc/config/i386/i386-c.c (ix86_target_macros_internal): Handle
2965         PROCESSOR_SKYLAKE.
2966         * gcc/config/i386/i386.c (m_SKYLAKE): Define.
2967         (processor_target_table): Add "skylake".
2968         (ix86_option_override_internal): Add "skylake".
2969         (get_builtin_code_for_version): Handle PROCESSOR_SKYLAKE,
2970         PROCESSOR_CANNONLAKE.
2971         (get_builtin_code_for_version): Fix priority for
2972         PROCESSOR_ICELAKE_CLIENT, PROCESSOR_ICELAKE_SERVER,
2973         PROCESSOR_SKYLAKE-AVX512.
2974         * gcc/config/i386/i386.h (processor_costs): Define TARGET_SKYLAKE.
2975         (processor_type): Add PROCESSOR_SKYLAKE.
2976
2977 2018-04-16  Paolo Carlini  <paolo.carlini@oracle.com>
2978             Jason Merrill  <jason@redhat.com>
2979
2980         PR c++/85112
2981         * convert.c (convert_to_integer_1): Use direct recursion for
2982         enumeral types and types with a precision less than the number
2983         of bits in their mode.
2984
2985 2018-04-16  Julia Koval  <julia.koval@intel.com>
2986
2987         PR target/84413
2988         * config/i386/x86-tune.def (X86_TUNE_SSE_UNALIGNED_LOAD_OPTIMAL,
2989         X86_TUNE_SSE_UNALIGNED_STORE_OPTIMAL): Add m_SKYLAKE_AVX512
2990
2991 2018-04-14  Segher Boessenkool  <segher@kernel.crashing.org>
2992
2993         PR target/85293
2994         * config/rs6000/rs6000.opt (mdirect-move): Make deprecated.
2995         * doc/invoke.texi (RS/6000 and PowerPC Options): Remove -mdirect-move
2996         and -mno-direct-move.
2997
2998 2018-04-13  Paul A. Clarke  <pc@us.ibm.com>
2999
3000         PR target/83402
3001         * config/rs6000/emmintrin.h (_mm_slli_epi{16,32,64}):
3002         Ensure that vec_splat_s32 is only called with 0 <= shift < 16.
3003         Ensure negative shifts result in {0}.
3004
3005 2018-04-13  Vladimir Makarov  <vmakarov@redhat.com>
3006
3007         PR rtl-optimization/79916
3008         * config/rs6000/rs6000.c (rs6000_emit_move): Use assigned hard
3009         regs (if any) to define how to gnerate SD moves when LRA is in
3010         progress.
3011
3012 2018-04-13  Jakub Jelinek  <jakub@redhat.com>
3013
3014         PR rtl-optimization/85393
3015         * except.h (expand_dw2_landing_pad_for_region): Remove declaration.
3016         * except.c (expand_dw2_landing_pad_for_region): Make static.
3017         * bb-reorder.c (fix_up_crossing_landing_pad): In new_bb emit just
3018         a label and unconditional jump to old_bb, rather than
3019         expand_dw2_landing_pad_for_region insn(s) and jump to single_succ
3020         basic block.
3021
3022         PR rtl-optimization/85376
3023         * simplify-rtx.c (simplify_const_unary_operation): For CLZ and CTZ and
3024         zero op0, if C?Z_DEFINED_VALUE_AT_ZERO is false, return NULL_RTX
3025         instead of a specific value.
3026
3027 2018-04-13  Jan Hubicka  <hubicka@ucw.cz>
3028             Bin Cheng  <bin.cheng@arm.com>
3029
3030         PR tree-optimization/82965
3031         PR tree-optimization/83991
3032         * cfgloopanal.c (expected_loop_iterations_unbounded): Add
3033         by_profile_only parameter.
3034         * cfgloopmanip.c (scale_loop_profile): Further scale loop's profile
3035         information if the loop was predicted to iterate too many times.
3036         * cfgloop.h (expected_loop_iterations_unbounded): Update prototype
3037
3038 2018-04-13  Jan Hubicka  <hubicka@ucw.cz>
3039
3040         PR lto/71991
3041         * config/i386/i386.c (ix86_can_inline_p): Allow safe transitions for
3042         always inline.
3043
3044 2018-04-13  Martin Liska  <mliska@suse.cz>
3045             Jakub Jelinek  <jakub@redhat.com>
3046
3047         PR middle-end/81657
3048         * expr.h (enum block_op_methods): Add BLOCK_OP_NO_LIBCALL_RET.
3049         * expr.c (emit_block_move_hints): Handle BLOCK_OP_NO_LIBCALL_RET.
3050         * builtins.c (expand_builtin_memory_copy_args): Use
3051         BLOCK_OP_NO_LIBCALL_RET method for mempcpy with non-ignored target,
3052         handle dest_addr == pc_rtx.
3053
3054 2018-04-12  Segher Boessenkool  <segher@kernel.crashing.org>
3055
3056         PR target/85291
3057         * config/rs6000/rs6000.md (fix_trunc<mode>si2): Use legacy code if
3058         asked to not generate direct moves.
3059         (fix_trunc<mode>si2_stfiwx): Similar.
3060         (fix_trunc<mode>si2_internal): Similar.
3061
3062 2018-04-12  Jakub Jelinek  <jakub@redhat.com>
3063
3064         PR debug/83157
3065         * var-tracking.c (add_stores): Handle STRICT_LOW_PART SET_DEST.
3066         * cselib.c (cselib_record_sets): For STRICT_LOW_PART dest,
3067         lookup if dest in some wider mode is known to be const0_rtx and
3068         if so, record permanent equivalence for it to be ZERO_EXTEND of
3069         the narrower mode destination.
3070
3071 2018-04-12  Cesar Philippidis  <cesar@codesourcery.com>
3072
3073         * lto-streamer-out.c (output_function): Revert 259346.
3074         * omp-expand.c (expand_oacc_for): Likewise.
3075
3076 2018-04-12  Alexander Monakov  <amonakov@ispras.ru>
3077
3078         PR rtl-optimization/85354
3079         * sel-sched-ir.c (sel_init_pipelining): Move cfg_cleanup call...
3080         * sel-sched.c (sel_global_init): ... here.
3081
3082 2018-04-12  Eric Botcazou  <ebotcazou@adacore.com>
3083
3084         PR target/85238
3085         * lto-wrapper.c (debug_objcopy): Open the files in binary mode.
3086         * dwarf2out.c (dwarf2out_early_finish): Do not generate assembly in LTO
3087         mode for PE-COFF targets.
3088         * config/i386/i386-protos.h (i386_pe_asm_lto_start): Declare.
3089         (i386_pe_asm_lto_end): Likewise.
3090         * config/i386/cygming.h (TARGET_ASM_LTO_START): Define.
3091         (TARGET_ASM_LTO_END): Likewise.
3092         * config/i386/winnt.c (saved_debug_info_level): New static variable.
3093         (i386_pe_asm_lto_start): New function.
3094         (i386_pe_asm_lto_end): Likewise.
3095
3096 2018-04-12  Cesar Philippidis  <cesar@codesourcery.com>
3097             Richard Biener  <rguenther@suse.de>
3098
3099         PR middle-end/84955
3100         * lto-streamer-out.c (output_function): Fix CFG loop state before
3101         streaming out.
3102         * omp-expand.c (expand_oacc_for): Handle calls to internal
3103         functions like regular functions.
3104
3105 2018-04-12  Richard Biener  <rguenther@suse.de>
3106
3107         PR lto/85371
3108         * dwarf2out.c (init_sections_and_labels): Use debug_line_section[_label]
3109         for the early LTO debug to properly generate references to it
3110         during DIE emission.  Do not re-use that for the skeleton for
3111         split-dwarf.
3112         (dwarf2out_early_finish): Likewise.
3113
3114 2018-04-12  Jakub Jelinek  <jakub@redhat.com>
3115
3116         PR target/85328
3117         * config/i386/sse.md
3118         (<mask_codefor>avx512dq_vextract<shuffletype>64x2_1<mask_name> split,
3119         <mask_codefor>avx512f_vextract<shuffletype>32x4_1<mask_name> split,
3120         vec_extract_lo_<mode><mask_name> split, vec_extract_lo_v32hi,
3121         vec_extract_lo_v64qi): For non-AVX512VL if input is xmm16+ reg
3122         and output is a reg, avoid creating invalid lowpart subreg, but
3123         instead split into a 512-bit move.  Don't split if not AVX512VL,
3124         input is xmm16+ reg and output is a mem.
3125         (vec_extract_lo_<mode><mask_name>, vec_extract_lo_v32hi,
3126         vec_extract_lo_v64qi): Don't require split if not AVX512VL, input is
3127         xmm16+ reg and output is a mem.
3128
3129 2018-04-12  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>
3130
3131         * config/s390/s390.c (s390_output_indirect_thunk_function): Check
3132         also for flag_dwarf2_cfi_asm.
3133
3134 2018-04-12  Jakub Jelinek  <jakub@redhat.com>
3135
3136         PR rtl-optimization/85342
3137         * regcprop.c (copyprop_hardreg_forward_1): Remove replaced array, use
3138         a bool scalar var inside of the loop instead.  Don't try to update
3139         recog_data.operand after failed apply_change_group.
3140
3141 2018-04-12  Tom de Vries  <tom@codesourcery.com>
3142
3143         PR target/85296
3144         * config/nvptx/nvptx.c (flexible_array_member_type_p): New function.
3145         (nvptx_assemble_decl_begin): Add undefined param.  Declare undefined
3146         array with flexible array member as array without given dimension.
3147         (nvptx_assemble_undefined_decl): Set nvptx_assemble_decl_begin call
3148         argument for undefined param to true.
3149
3150 2018-04-11  Aaron Sawdey  <acsawdey@linux.ibm.com>
3151
3152         PR target/85321
3153         * doc/invoke.texi (RS/6000 and PowerPC Options): Document options
3154         -mcall- and -mtraceback=. Remove options -mabi=spe and -mabi=no-spe
3155         from PowerPC section.
3156         * config/rs6000/sysv4.opt (mcall-): Improve help text.
3157         * config/rs6000/rs6000.opt (mblock-compare-inline-limit=): Trim
3158         help text that is too long.
3159         * config/rs6000/rs6000.opt (mblock-compare-inline-loop-limit=): Trim
3160         help text that is too long.
3161         * config/rs6000/rs6000.opt (mstring-compare-inline-limit=): Trim
3162         help text that is too long.
3163
3164 2018-04-11  Uros Bizjak  <ubizjak@gmail.com>
3165
3166         * config/alpha/alpha.md (stack_probe_internal): Rename
3167         from "probe_stack".  Update all callers.
3168
3169 2018-04-11  Alexander Monakov  <amonakov@ispras.ru>
3170
3171         PR rtl-optimization/84566
3172         * sched-deps.c (sched_analyze_insn): Check deps->readonly when invoking
3173         sched_macro_fuse_insns.
3174
3175 2018-04-11  Alexander Monakov  <amonakov@ispras.ru>
3176
3177         PR target/84301
3178         * sched-rgn.c (add_branch_dependences): Move sel_sched_p check here...
3179         (compute_block_dependences): ... from here.
3180
3181 2018-04-11  Jakub Jelinek  <jakub@redhat.com>
3182
3183         PR tree-optimization/85331
3184         * vec-perm-indices.h (vec_perm_indices::clamp): Change input type
3185         from int to HOST_WIDE_INT.
3186
3187 2018-04-11  Martin Jambor  <mjambor@suse.cz>
3188
3189         PR ipa/84149
3190         * ipa-cp.c (propagate_vals_across_pass_through): Expand comment.
3191         (cgraph_edge_brings_value_p): New parameter dest_val, check if it is
3192         not the same as the source val.
3193         (cgraph_edge_brings_value_p): New parameter.
3194         (gather_edges_for_value): Pass destination value to
3195         cgraph_edge_brings_value_p.
3196         (perhaps_add_new_callers): Likewise.
3197         (get_info_about_necessary_edges): Likewise and exclude values brought
3198         only by self-recursive edges.
3199         (create_specialized_node): Redirect only clones of self-calling edges.
3200         (+self_recursive_pass_through_p): New function.
3201         (find_more_scalar_values_for_callers_subset): Use it.
3202         (find_aggregate_values_for_callers_subset): Likewise.
3203         (known_aggs_to_agg_replacement_list): Removed.
3204         (decide_whether_version_node): Re-calculate known constants for all
3205         remaining context clones.
3206
3207 2018-04-11  Richard Biener  <rguenther@suse.de>
3208
3209         PR lto/85339
3210         * dwarf2out.c (dwarf2out_finish): Remove DW_AT_stmt_list attribute
3211         from early DWARF output.
3212         (dwarf2out_early_finish): Output line info unconditionally into
3213         early DWARF and add reference to it.
3214
3215 2018-04-11  Jakub Jelinek  <jakub@redhat.com>
3216
3217         PR target/85281
3218         * config/i386/sse.md (iptr): Add V16SFmode and V8DFmode cases.
3219         (<avx512>_vec_dup<mode><mask_name>): Use a single pattern for modes
3220         other than V2DFmode using iptr mode attribute.
3221         (<avx512>_vec_dup<mode><mask_name>): Use iptr mode attribute.
3222
3223 2018-04-11  Alexander Monakov  <amonakov@ispras.ru>
3224
3225         PR rtl-optimization/84659
3226         * sel-sched-ir.c (sel_init_pipelining): Invoke cleanup_cfg.
3227
3228 2018-04-11  Jakub Jelinek  <jakub@redhat.com>
3229
3230         PR debug/85302
3231         * dwarf2out.c (skip_loc_list_entry): Don't call size_of_locs if
3232         SIZEP is NULL.
3233         (output_loc_list): Pass address of a dummy size variable even in the
3234         locview handling loop.
3235         (index_location_lists): Add comment on why skip_loc_list_entry can't
3236         call size_of_locs.
3237
3238 2018-04-11  Thomas Preud'homme  <thomas.preudhomme@arm.com>
3239
3240         PR target/85261
3241         * config/arm/arm-builtins.c (arm_expand_builtin): Force input operand
3242         into register.
3243
3244 2018-04-10  Aaron Sawdey  <acsawdey@linux.ibm.com>
3245
3246         PR target/85321
3247         * doc/invoke.texi (RS/6000 and PowerPC Options): Document options
3248         -mblock-compare-inline-limit, -mblock-compare-inline-loop-limit,
3249         and -mstring-compare-inline-limit.
3250
3251 2018-04-10  Segher Boessenkool  <segher@kernel.crashing.org>
3252
3253         PR target/85287
3254         * gcc/config/rs6000/rs6000.md (allocate_stack): Put the residual size
3255         for stack clash protection in a register whenever we need it to be in
3256         a register.
3257
3258 2018-04-10  Segher Boessenkool  <segher@kernel.crashing.org>
3259
3260         * common/config/rs6000/rs6000-common.c (rs6000_option_init_struct):
3261         Enable -fasynchronous-unwind-tables by default if OBJECT_FORMAT_ELF.
3262
3263 2018-04-10  Segher Boessenkool  <segher@kernel.crashing.org>
3264
3265         PR target/85321
3266         * config/rs6000/rs6000.opt (mtraceback=): Show the allowed values in
3267         the help text.
3268         (mlong-double-): Ditto.
3269         * config/rs6000/sysv4.opt (msdata=): Ditto.
3270         (mtls-size=): Ditto.
3271
3272 2018-04-10  Kelvin Nilsen  <kelvin@gcc.gnu.org>
3273
3274         * config/rs6000/rs6000-c.c (altivec_overloaded_builtins): Remove
3275         erroneous entries for
3276         "vector int vec_ldl (int, long int *)", and
3277         "vector unsigned int vec_ldl (int, unsigned long int *)".
3278         Add comments and entries for
3279         "vector bool char vec_ldl (int, bool char *)",
3280         "vector bool short vec_ldl (int, bool short *)",
3281         "vector bool int vec_ldl (int, bool int *)",
3282         "vector bool long long vec_ldl (int, bool long long *)",
3283         "vector pixel vec_ldl (int, pixel *)",
3284         "vector long long vec_ldl (int, long long *)",
3285         "vector unsigned long long vec_ldl (int, unsigned long long *)".
3286         * config/rs6000/rs6000.c (rs6000_init_builtins): Initialize new
3287         type tree bool_long_long_type_node and correct definition of
3288         bool_V2DI_type_node to make reference to this new type tree.
3289         (rs6000_mangle_type): Replace erroneous reference to
3290         bool_long_type_node with bool_long_long_type_node.
3291         * config/rs6000/rs6000.h (enum rs6000_builtin_type_index): Add
3292         comments to emphasize sign distinctions for char and int types and
3293         replace RS6000_BTI_bool_long constant with
3294         RS6000_BTI_bool_long_long constant.  Also add comment to restrict
3295         use of RS6000_BTI_pixel.
3296         (bool_long_type_node): Remove this macro definition.
3297         (bool_long_long_type_node): New macro definition
3298
3299 2018-04-10  Jakub Jelinek  <jakub@redhat.com>
3300
3301         PR rtl-optimization/85300
3302         * combine.c (subst): Handle subst of CONST_SCALAR_INT_P new_rtx also
3303         into FLOAT and UNSIGNED_FLOAT like ZERO_EXTEND, return a CLOBBER if
3304         simplify_unary_operation fails.
3305
3306 2018-04-10  Martin Liska  <mliska@suse.cz>
3307
3308         * gdbhooks.py: Add pretty-printers for varpool_node, symtab_node,
3309         cgraph_edge and ipa_ref.
3310
3311 2018-04-10  Jakub Jelinek  <jakub@redhat.com>
3312
3313         PR target/85177
3314         PR target/85255
3315         * config/i386/sse.md
3316         (<extract_type>_vinsert<shuffletype><extract_suf>_mask): Fix
3317         computation of the VEC_MERGE selector from mask.
3318         (<extract_type>_vinsert<shuffletype><extract_suf>_1<mask_name>):
3319         Fix decoding of the VEC_MERGE selector into mask.
3320
3321 2018-04-10  Richard Sandiford  <richard.sandiford@linaro.org>
3322
3323         PR tree-optimization/85286
3324         * tree-vect-data-refs.c (vect_get_smallest_scalar_type):
3325
3326 2018-04-10  Richard Sandiford  <richard.sandiford@linaro.org>
3327
3328         * final.c (final_1): Set insn_last_address as well as
3329         insn_current_address.
3330
3331 2018-04-10  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
3332
3333         PR target/85173
3334         * explow.c (emit_stack_probe): Call validize_mem on memory location
3335         before passing it to gen_probe_stack.  Create address operand and
3336         legitimize it for the probe_stack_address case.
3337
3338 2018-04-09  Jan Hubicka  <jh@suse.cz>
3339
3340         PR lto/85078
3341         * ipa-devirt.c (rebuild_type_inheritance-hash): New.
3342         * ipa-utils.h (rebuild_type_inheritance-hash): Declare.
3343         * tree.c (free_lang_data_in_type): Fix handling of binfos;
3344         walk basetypes.
3345         (free_lang_data): Rebuild type inheritance graph.
3346
3347 2018-04-09  Martin Sebor  <msebor@redhat.com>
3348
3349         * invoke.texi (-finline-small-functions): Mention other optimization
3350         options.
3351         (-findirect-inlining, -fpartial-inlining): Same.
3352         (-finline-functions-called-once): Same.
3353         (-freorder-blocks-and-partition): Same.
3354
3355 2018-04-09  Jan Hubicka  <jh@suse.cz>
3356
3357         PR rtl/84058
3358         * cfgcleanup.c (try_forward_edges): Do not give up on crossing
3359         jumps; choose last target that matches the criteria (i.e.
3360         no partition changes for non-crossing jumps).
3361         * cfgrtl.c (cfg_layout_redirect_edge_and_branch): Add basic
3362         support for redirecting crossing jumps to non-crossing.
3363
3364 2018-04-09  Alexey Brodkin  <abrodkin@synopsys.com>
3365
3366         * config/arc/arc.c (arc_expand_prologue): Set stack usage info
3367         also for naked functions.
3368
3369 2018-04-09  Claudiu Zissulescu  <claziss@synopsys.com>
3370
3371         * config/arc/arc.md (add_shift): New pattern.
3372         (add_shift2): Likewise.
3373         (sub_shift): Likewise.
3374         (sub_shift_cmp0_noout): Likewise.
3375         (compare_si_ashiftsi): Likewise.
3376         (xbfu_cmp0_noout): New combine pattern.
3377         (xbfu_cmp0"): Likewise.
3378         (movsi_set_cc_insn): Place the predicable variant first.
3379         (commutative_binary_cmp0_noout): Remove clobber.
3380         (commutative_binary_cmp0): New pattern.
3381         (noncommutative_binary_cmp0): Likewise.
3382         (noncommutative_binary_cmp0_noout): Likewise.
3383         (noncommutative_binary_comparison_result_used): Removed.
3384         (rsub_cmp0): New pattern.
3385         (rsub_cmp0_noout): Likewise.
3386         (extzvsi): Changed, keep only meaningful variants.
3387         (SQH, SEZ): New iterators.
3388         (SQH_postfix): New mode attribute.
3389         (SEZ_prefix): New code attribute.
3390         (<SEZ_prefix>xt<SQH_postfix>_cmp0_noout): New instruction pattern.
3391         (<SEZ_prefix>xt<SQH_postfix>_cmp0): Likewise.
3392         * config/arc/predicates.md (cc_set_register): Use CC_REG instead
3393         of numerical value.
3394         (noncommutative_operator): Check the availability of barrel
3395         shifter option.
3396
3397 2018-04-09  Richard Biener  <rguenther@suse.de>
3398
3399         PR tree-optimization/85284
3400         * tree-ssa-loop-niter.c (number_of_iterations_exit_assumptions):
3401         Only use the niter constraining form of simple_iv when the exit
3402         is always executed.
3403
3404 2018-04-09  Tom de Vries  <tom@codesourcery.com>
3405
3406         PR target/84041
3407         * config/nvptx/nvptx.md (define_c_enum "unspecv"): Add UNSPECV_MEMBAR.
3408         (define_expand "*memory_barrier"): New define_expand.
3409         (define_insn "memory_barrier"): New insn.
3410
3411 2018-04-09  Andrey Belevantsev  <abel@ispras.ru>
3412
3413         PR rtl-optimization/80463
3414         PR rtl-optimization/83972
3415         PR rtl-optimization/83480
3416
3417         * sel-sched-ir.c (has_dependence_note_mem_dep): Take into account the
3418         correct producer for the insn.
3419         (tidy_control_flow): Fixup seqnos in case of debug insns.
3420
3421 2018-04-09  Andrey Belevantsev  <abel@ispras.ru>
3422
3423         PR rtl-optimization/83913
3424
3425         * sel-sched-ir.c (merge_expr_data): Choose the middle between two
3426         different sched-times when merging exprs.
3427
3428 2018-04-09  Andrey Belevantsev  <abel@ispras.ru>
3429
3430         PR rtl-optimization/83962
3431
3432         * sel-sched-ir.c (tidy_control_flow): Correct the order in which we call
3433         tidy_fallthru_edge and tidy_control_flow.
3434
3435 2018-04-09  Andrey Belevantsev  <abel@ispras.ru>
3436
3437         PR rtl-optimization/83530
3438
3439         * sel-sched.c (force_next_insn): New global variable.
3440         (remove_insn_for_debug): When force_next_insn is true, also leave only
3441         next insn in the ready list.
3442         (sel_sched_region): When the region wasn't scheduled, make another pass
3443         over it with force_next_insn set to 1.
3444
3445 2018-04-08  Monk Chiang  <sh.chiang04@gmail.com>
3446
3447         * config.gcc (nds32le-*-*, nds32be-*-*): Add nds32/nds32_intrinsic.h
3448         into tm_file.
3449         * config/nds32/constants.md (unspec_volatile_element): Add enum values
3450         for interrupt control.
3451         * config/nds32/nds32-intrinsic.c: Implementation of intrinsic
3452         functions for interrupt control.
3453         * config/nds32/nds32-intrinsic.md: Likewise.
3454         * config/nds32/nds32_intrinsic.h: Likewise.
3455         * config/nds32/nds32.h (nds32_builtins): Likewise.
3456
3457 2018-04-08  Chung-Ju Wu  <jasonwucj@gmail.com>
3458
3459         * config/nds32/nds32.c (nds32_init_machine_status,
3460         nds32_legitimate_index_p, nds32_legitimate_address_p): Consider
3461         strict_aligned_p field.
3462         (nds32_expand_to_rtl_hook): New function.
3463         (TARGET_EXPAND_TO_RTL_HOOK): Define.
3464         * config/nds32/nds32.h (machine_function): Add strict_aligned_p field.
3465
3466 2018-04-08  Kito Cheng  <kito.cheng@gmail.com>
3467             Chung-Ju Wu  <jasonwucj@gmail.com>
3468
3469         * config.gcc (nds32*-*-*): Check that n7 is valid to --with-cpu.
3470         * config/nds32/nds32-n7.md: New file.
3471         * config/nds32/nds32-opts.h (nds32_cpu_type): Add CPU_N7.
3472         * config/nds32/nds32-pipelines-auxiliary.c: Implementation for n7
3473         pipeline.
3474         * config/nds32/nds32-protos.h: More declarations for n7 pipeline.
3475         * config/nds32/nds32.md (pipeline_model): Add n7.
3476         * config/nds32/nds32.opt (mcpu): Support n7 pipeline cpus.
3477         * config/nds32/pipelines.md: Include n7 settings.
3478
3479 2018-04-08  Kito Cheng  <kito.cheng@gmail.com>
3480             Chung-Ju Wu  <jasonwucj@gmail.com>
3481
3482         * config.gcc (nds32*-*-*): Check that e8 is valid to --with-cpu.
3483         * config/nds32/nds32-e8.md: New file.
3484         * config/nds32/nds32-opts.h (nds32-cpu_type): Add CPU_E8.
3485         * config/nds32/nds32-pipelines-auxiliary.c: Implementation for e8
3486         pipeline.
3487         * config/nds32/nds32-protos.h: More declarations for e8 pipeline.
3488         * config/nds32/nds32.md (pipeline_model): Add e8.
3489         * config/nds32/nds32.opt (mcpu): Support e8 pipeline cpus.
3490         * config/nds32/pipelines.md: Include e8 settings.
3491
3492 2018-04-08  Kito Cheng  <kito.cheng@gmail.com>
3493             Chung-Ju Wu  <jasonwucj@gmail.com>
3494
3495         * config.gcc (nds32*-*-*): Check that n6/n8/s8 are valid to --with-cpu.
3496         * config/nds32/nds32-n8.md: New file.
3497         * config/nds32/nds32-opts.h (nds32_cpu_type): Add CPU_N6 and CPU_N8.
3498         * config/nds32/nds32-pipelines-auxiliary.c: Implementation for n8
3499         pipeline.
3500         * config/nds32/nds32-protos.h: More declarations for n8 pipeline.
3501         * config/nds32/nds32-utils.c: More implementations for n8 pipeline.
3502         * config/nds32/nds32.md (pipeline_model): Add n8.
3503         * config/nds32/nds32.opt (mcpu): Support n8 pipeline cpus.
3504         * config/nds32/pipelines.md: Include n8 settings.
3505
3506 2018-04-08  Kito Cheng  <kito.cheng@gmail.com>
3507             Chung-Ju Wu  <jasonwucj@gmail.com>
3508
3509         * config.gcc (nds32*): Add nds32-utils.o into extra_objs.
3510         * config/nds32/nds32-n9-2r1w.md: New file.
3511         * config/nds32/nds32-n9-3r2w.md: New file.
3512         * config/nds32/nds32-opts.h (nds32_cpu_type, nds32_mul_type,
3513         nds32_register_ports): New or modify for cpu n9.
3514         * config/nds32/nds32-pipelines-auxiliary.c: Implementation for n9
3515         pipeline.
3516         * config/nds32/nds32-protos.h: More declarations for n9 pipeline.
3517         * config/nds32/nds32-utils.c: New file.
3518         * config/nds32/nds32.h (TARGET_PIPELINE_N9, TARGET_PIPELINE_SIMPLE,
3519         TARGET_MUL_SLOW): Define.
3520         * config/nds32/nds32.md (pipeline_model): New attribute.
3521         * config/nds32/nds32.opt (mcpu, mconfig-mul, mconfig-register-ports):
3522         New options that support cpu n9.
3523         * config/nds32/pipelines.md: Include n9 settings.
3524         * config/nds32/t-nds32 (nds32-utils.o): Add dependency.
3525
3526 2018-04-08  Chung-Ju Wu  <jasonwucj@gmail.com>
3527
3528         * config/nds32/nds32-md-auxiliary.c (output_cond_branch): Output align
3529         information if necessary.
3530         (output_cond_branch_compare_zero): Likewise.
3531         * config/nds32/nds32.c (nds32_adjust_insn_length): Consider align case.
3532         (nds32_target_alignment): Refine for alignment.
3533         * config/nds32/nds32.h (NDS32_ALIGN_P): Define.
3534         (FUNCTION_BOUNDARY): Modify.
3535         * config/nds32/nds32.md (call_internal, call_value_internal): Consider
3536         align case.
3537         * config/nds32/nds32.opt (malways-align, malign-functions): New.
3538
3539 2018-04-08  Monk Chiang  <sh.chiang04@gmail.com>
3540
3541         * config/nds32/constants.md (unspec_volatile_element): Add values for
3542         TLB operation and data prefetch.
3543         * config/nds32/nds32-intrinsic.c: Implementation of intrinsic
3544         functions for TLB operation and data prefetch.
3545         * config/nds32/nds32-intrinsic.md: Likewise.
3546         * config/nds32/nds32_intrinsic.h: Likewise.
3547         * config/nds32/nds32.c (nds32_dpref_names): Likewise.
3548         (nds32_print_operand): Likewise.
3549         * config/nds32/nds32.h (nds32_builtins): Likewise.
3550
3551 2018-04-07  Thomas Koenig  <tkoenig@gcc.gnu.org>
3552         Andrew Pinski <pinsika@gcc.gnu.org>
3553
3554         PR middle-end/82976
3555         * match.pd: Use constant_boolean_node of correct type instead of
3556         boolean_true_node or boolean_false_node for simplifying
3557         pointer comparisons to zero.
3558
3559 2018-04-07  Jakub Jelinek  <jakub@redhat.com>
3560
3561         PR tree-optimization/80021
3562         * tree.c (verify_type_variant): Make error call in verify_variant_match
3563         translatable and remove final full stop.
3564
3565 2018-04-07  Chung-Ju Wu  <jasonwucj@gmail.com>
3566
3567         * config/nds32/constants.md (unspec_volatile_element): Add
3568         UNSPEC_VOLATILE_EH_RETURN.
3569         * config/nds32/nds32-md-auxiliary.c (nds32_output_stack_push,
3570         nds32_output_stack_pop): Support dwarf exception handling process.
3571         * config/nds32/nds32-protos.h (nds32_dynamic_chain_address): Declare.
3572         * config/nds32/nds32.c (nds32_init_machine_status): Support dwarf
3573         exception handling process.
3574         (nds32_compute_stack_frame): Likewise.
3575         (nds32_return_addr_rtx): Likewise.
3576         (nds32_initial_elimination_offset): Likewise.
3577         (nds32_expand_prologue): Likewise.
3578         (nds32_expand_epilogue): Likewise.
3579         (nds32_dynamic_chain_address): New function.
3580         * config/nds32/nds32.h (machine_function): Add fields for dwarf
3581         exception handling.
3582         (DYNAMIC_CHAIN_ADDRESS): Define.
3583         (EH_RETURN_DATA_REGNO): Define.
3584         (EH_RETURN_STACKADJ_RTX): Define.
3585         * config/nds32/nds32.md (eh_return, nds32_eh_return): Implement
3586         patterns for dwarf exception handling.
3587
3588 2018-04-07  Chung-Ju Wu  <jasonwucj@gmail.com>
3589
3590         * config/nds32/nds32.h: Clean up obsolete macros.
3591
3592 2018-04-07  Monk Chiang  <sh.chiang04@gmail.com>
3593
3594         * config/nds32/constants.md (unspec_element, unspec_volatile_element):
3595         Add enum values for particular instructions.
3596         * config/nds32/nds32-intrinsic.c: Implementation of expanding
3597         particular intrinsic functions.
3598         * config/nds32/nds32-intrinsic.md: Likewise.
3599         * config/nds32/nds32_intrinsic.h: Likewise.
3600         * config/nds32/nds32.h (nds32_builtins): Likewise.
3601         * config/nds32/nds32.md (type): Add pbsad and pbsada.
3602         (btst, ave): New patterns for particular instructions.
3603
3604 2018-04-07  Monk Chiang  <sh.chiang04@gmail.com>
3605
3606         * config/nds32/constants.md (unspec_element, unspec_volatile_element):
3607         Add enum values for atomic load/store and memory sync.
3608         * config/nds32/nds32-intrinsic.c: Implementation for atomic load/store
3609         and memory sync.
3610         * config/nds32/nds32-intrinsic.md: Likewise.
3611         * config/nds32/nds32_intrinsic.h: Likewise.
3612         * config/nds32/nds32.h (nds32_builtins): Likewise.
3613
3614 2018-04-07  Jakub Jelinek  <jakub@redhat.com>
3615
3616         PR tree-optimization/85257
3617         * fold-const.c (native_encode_vector): If not all elts could fit
3618         and off is -1, return 0 rather than offset.
3619         * tree-ssa-sccvn.c (vn_reference_lookup_3): Pass
3620         (offseti - offset2) / BITS_PER_UNIT as 4th argument to
3621         native_encode_expr.  Verify len * BITS_PER_UNIT >= maxsizei.  Don't
3622         adjust buffer in native_interpret_expr call.
3623
3624 2018-04-07  Monk Chiang  <sh.chiang04@gmail.com>
3625
3626         * config/nds32/constants.md (unspec_volatile_element): Add cache
3627         control enum values.
3628         * config/nds32/nds32-intrinsic.c: Add cache control expand functions.
3629         * config/nds32/nds32-intrinsic.md: Add cache control patterns.
3630         * config/nds32/nds32.c (nds32_cctl_names): New.
3631         (nds32_print_operand): Handle cache control register names.
3632         * config/nds32/nds32.h (nds32_builtins): New enum values.
3633         * config/nds32/nds32_intrinsic.h: Add cache control enum types and
3634         macros.
3635         * config/nds32/nds32.md (type): Add mmu.
3636         * config/nds32/pipelines.md (simple_insn): Add mmu.
3637
3638 2018-04-07  Chung-Ju Wu  <jasonwucj@gmail.com>
3639
3640         * config/nds32/nds32.md (type): Remove call.
3641         * config/nds32/pipelines.md (simple_insn): Likewise.
3642
3643 2018-04-07  Monk Chiang  <sh.chiang04@gmail.com>
3644
3645         * config/nds32/constants.md (unspec_volatile_element): Add
3646         UNSPEC_VOLATILE_FMFCSR, UNSPEC_VOLATILE_FMTCSR and
3647         UNSPEC_VOLATILE_FMFCFG.
3648         * config/nds32/nds32-intrinsic.c (bdesc_noarg): New builtin
3649         description for fmfcfg and fmfcsr.
3650         (bdesc_1arg): Add fmtcsr.
3651         (bdesc_2arg): Add fcpynss, fcpyss, fcpynsd and fcpysd.
3652         (nds32_expand_builtin_impl): Deal with FPU intrinsic functions.
3653         * config/nds32/nds32-intrinsic.md (unspec_fcpynsd, unspec_fcpysd,
3654         unspec_fcpynss, unspec_fcpysd, unspec_fcpyss, unspec_fmfcsr,
3655         unspec_fmfcfg): New patterns.
3656         * config/nds32/nds32.h (nds32_builtins): Add NDS32_BUILTIN_FMFCFG,
3657         NDS32_BUILTIN_FMFCSR, NDS32_BUILTIN_FMTCSR, NDS32_BUILTIN_FCPYNSS,
3658         NDS32_BUILTIN_FCPYSS,NDS32_BUILTIN_FCPYNSD and NDS32_BUILTIN_FCPYSD.
3659         * config/nds32/nds32_intrinsic.h (__nds32__fcpynsd, __nds32__fcpynss,
3660         __nds32__fcpysd, __nds32__fcpyss, __nds32__fmfcsr, __nds32__fmtcsr,
3661         __nds32__fmfcfg): Define.
3662
3663 2018-04-07  Monk Chiang  <sh.chiang04@gmail.com>
3664
3665         * config/nds32/nds32.c (nds32_intrinsic_register_names): Add more
3666         intrinsic register names.
3667         * config/nds32/nds32_intrinsic.h (nds32_intrinsic_registers): Add more
3668         intrinsic register enum values and macros.
3669
3670 2018-04-07  Chung-Ju Wu  <jasonwucj@gmail.com>
3671
3672         * config/nds32/nds32.c (nds32_legitimate_index_p): Modify condition
3673         for load/store addressing form.
3674         (nds32_print_operand_address): Likewise.
3675
3676 2018-04-06  Eric Botcazou  <ebotcazou@adacore.com>
3677
3678         PR target/85196
3679         * config/sparc/sparc.c (sparc_expand_move): Deal with symbolic operands
3680         based on LABEL_REF.  Remove useless assertion.
3681         (pic_address_needs_scratch): Fix formatting.
3682         (sparc_legitimize_pic_address): Minor tweaks.
3683         (sparc_delegitimize_address): Adjust assertion accordingly.
3684         * config/sparc/sparc.md (movsi_pic_label_ref): Change label_ref_operand
3685         into symbolic_operand.
3686         (movsi_high_pic_label_ref): Likewise.
3687         (movsi_lo_sum_pic_label_ref): Likewise.
3688         (movdi_pic_label_ref): Likewise.
3689         (movdi_high_pic_label_ref): Likewise.
3690         (movdi_lo_sum_pic_label_ref): Likewise.
3691
3692 2018-04-06  Amaan Cheval  <amaan.cheval@gmail.com>
3693
3694         * config.gcc (x86_64-*-rtems*): Add rtems.h to tm_file for
3695         custom LIB_SPEC setup.
3696
3697 2018-04-06  Ruslan Bukin  <br@bsdpad.com>
3698             Kito Cheng  <kito.cheng@gmail.com>
3699
3700         * config.gcc (riscv*-*-freebsd*): Add RISC-V FreeBSD support.
3701         * config/riscv/freebsd.h: New.
3702
3703 2018-04-06  Chung-Ju Wu  <jasonwucj@gmail.com>
3704
3705         * config/nds32/nds32.c (nds32_adjust_insn_length): Refine.
3706         * config/nds32/nds32.h (ADJUST_INSN_LENGTH): Change the location in
3707         file.
3708
3709 2018-04-06  Chung-Ju Wu  <jasonwucj@gmail.com>
3710             Kito Cheng  <kito.cheng@gmail.com>
3711
3712         * config/nds32/nds32-md-auxiliary.c (nds32_output_return,
3713         nds32_output_call, nds32_symbol_binds_local_p): New functions.
3714         * config/nds32/nds32-protos.h (nds32_output_call,
3715         nds32_output_return): Declare.
3716         * config/nds32/nds32.md: Refine all the call and return patterns.
3717
3718 2018-04-06  Jakub Jelinek  <jakub@redhat.com>
3719
3720         PR debug/85252
3721         * dwarf2out.c (rtl_for_decl_init): For STRING_CST initializer only
3722         build CONST_STRING if TYPE_MAX_VALUE is non-NULL and is INTEGER_CST.
3723
3724         PR rtl-optimization/84872
3725         * cfgloopmanip.c (create_preheader): Use make_forwarder_block even if
3726         nentry == 1 when CP_FALLTHRU_PREHEADERS and single_entry is
3727         EDGE_CROSSING edge.
3728
3729 2018-04-06  Tamar Christina  <tamar.christina@arm.com>
3730
3731         * expr.c (copy_blkmode_to_reg): Revert 254862.
3732         * doc/sourcebuild.texi (word_mode_no_slow_unalign): Likewise.
3733
3734 2018-04-06  Richard Biener  <rguenther@suse.de>
3735
3736         PR middle-end/85244
3737         * tree-dfa.c (get_ref_base_and_extent): Reset seen_variable_array_ref
3738         after seeing a component reference with an adjacent field.  Treat
3739         refs to arrays at struct end of external decls similar to
3740         refs to unconstrained commons.
3741
3742 2018-04-06  Jakub Jelinek  <jakub@redhat.com>
3743
3744         PR sanitizer/85213
3745         * fold-const.c (twoval_comparison_p): Remove SAVE_P argument and don't
3746         look through SAVE_EXPRs with non-side-effects argument.  Adjust
3747         recursive calls.
3748         (fold_comparison): Adjust twoval_comparison_p caller, don't handle
3749         save_p here.
3750
3751 2018-04-06  Richard Biener  <rguenther@suse.de>
3752
3753         PR middle-end/85180
3754         * alias.c (find_base_term): New wrapper around find_base_term
3755         unwinding CSELIB_VAL_PTR changes.
3756         (find_base_term): Do not restore CSELIB_VAL_PTR during the
3757         recursion.
3758
3759 2018-04-06  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>
3760
3761         * config/s390/s390.c (s390_z10_optimize_cmp): Expand dedicated NOP
3762         instructions.
3763         * config/s390/s390.md (UNSPECV_NOP_LR_0, UNSPECV_NOP_LR_1): New
3764         constant definitions.
3765         ("nop"): lr 0,0 -> nopr r0
3766         ("nop_lr0", "nop_lr1"): New insn definitions.
3767
3768 2018-04-06  Chung-Ju Wu  <jasonwucj@gmail.com>
3769
3770         * config/nds32/nds32.md (*stack_push, *stack_pop): Use
3771         NDS32_V3PUSH_AVAILABLE_P macro.
3772
3773 2018-04-06  Monk Chiang  <sh.chiang04@gmail.com>
3774             Chung-Ju Wu  <jasonwucj@gmail.com>
3775
3776         * config.gcc (nds32*-*-*): Add v2j v3f v3s checking.
3777         (nds32*-*-*): Add float and fpu_config into supported_defaults.
3778         * common/config/nds32/nds32-common.c (TARGET_DEFAULT_TARGET_FLAGS):
3779         Include TARGET_DEFAULT_FPU_ISA and TARGET_DEFAULT_FPU_FMA.
3780         * config/nds32/constants.md (unspec_element): Add UNSPEC_COPYSIGN,
3781         UNSPEC_FCPYNSD, UNSPEC_FCPYNSS, UNSPEC_FCPYSD and UNSPEC_FCPYSS.
3782         * config/nds32/constraints.md: New constraints and checking for hard
3783         float configuration.
3784         * config/nds32/iterators.md: New mode iterator and attribute for hard
3785         float configuration.
3786         * config/nds32/nds32-doubleword.md: Use hard float alternatives and
3787         patterns.
3788         * config/nds32/nds32-fpu.md: New file.
3789         * config/nds32/nds32-md-auxiliary.c: New functions and checkings to
3790         deal with hard float code generation.
3791         * config/nds32/nds32-opts.h (nds32_arch_type): Add ARCH_V3F and
3792         ARCH_V3S.
3793         (abi_type, float_reg_number): New enum type.
3794         * config/nds32/nds32-predicates.c: New predicates for hard float.
3795         * config/nds32/nds32-protos.h: Declare functions for hard float.
3796         * config/nds32/nds32.c: Implementation for hard float configuration.
3797         * config/nds32/nds32.h: Definitions for hard float configuration.
3798         * config/nds32/nds32.md: Include hard float machine description and
3799         modify patterns for hard float configuration.
3800         * config/nds32/nds32.opt: New options for hard float configuration.
3801         * config/nds32/predicates.md: New predicates for hard float
3802         configuration.
3803
3804 2018-04-06  Kuan-Lin Chen  <kuanlinchentw@gmail.com>
3805
3806         * common/config/nds32/nds32-common.c
3807         (nds32_option_optimization_table): Enable -mreleax-hint by default.
3808
3809 2018-04-05  Jakub Jelinek  <jakub@redhat.com>
3810
3811         PR middle-end/85195
3812         * match.pd (BIT_FIELD_REF CONSTRUCTOR@0 @1 @2): Use view_convert around
3813         CONSTRUCTOR_ELT (ctor, ...)->value.
3814
3815 2018-04-05  Uros Bizjak  <ubizjak@gmail.com>
3816
3817         PR target/85193
3818         * config/i386/i386.md (define_attr "memory"): Handle rotate1 type.
3819
3820 2018-04-05  Tom de Vries  <tom@codesourcery.com>
3821
3822         PR target/85204
3823         * config/nvptx/nvptx.c (nvptx_single): Fix neutering of bb with only
3824         cond jump.
3825
3826 2018-04-05  Shiva Chen  <shiva0217@gmail.com>
3827             Kito Cheng  <kito.cheng@gmail.com>
3828
3829         * config/nds32/constraints.md (U33): Fine-tune checking condition.
3830         * config/nds32/nds32-md-auxiliary.c (nds32_mem_format): Ditto.
3831         * config/nds32/nds32.h (nds32_16bit_address_type): Add
3832         ADDRESS_POST_MODIFY_LO_REG_IMM3U.
3833
3834 2018-04-05  Shiva Chen  <shiva0217@gmail.com>
3835             Kito Cheng  <kito.cheng@gmail.com>
3836
3837         * config/nds32/constraints.md (Ufe): New memory constraint.
3838         * config/nds32/nds32-md-auxiliary.c (nds32_mem_format,
3839         nds32_output_16bit_load): Consider r8 register for lwi45.fe format.
3840         * config/nds32/nds32.c (nds32_print_operand): Output lwi45.fe
3841         operands.
3842         * config/nds32/nds32.h (nds32_16bit_address_type): Add ADDRESS_R8_IMM7U.
3843         * config/nds32/nds32.md (*mov<mode>): Adjust pattern.
3844
3845 2018-04-05  Chung-Ju Wu  <jasonwucj@gmail.com>
3846
3847         * config/nds32/nds32.md: Use optimize_size in the condition for
3848         alu-shift instructions.
3849
3850 2018-04-05  Chung-Ju Wu  <jasonwucj@gmail.com>
3851
3852         * config/nds32/nds32.md (divsi4, udivsi4): New patterns.
3853
3854 2018-04-05  Chung-Ju Wu  <jasonwucj@gmail.com>
3855
3856         * config/nds32/nds32.md (negsi2): Refine pattern.
3857
3858 2018-04-05  Kito Cheng  <kito.cheng@gmail.com>
3859             Chung-Ju Wu  <jasonwucj@gmail.com>
3860
3861         * config/nds32/iterators.md (shift_rotate): New code iterator.
3862         (shift): New code attribute.
3863         * config/nds32/nds32-md-auxiliary.c (nds32_expand_constant): New.
3864         * config/nds32/nds32-protos.h (nds32_expand_constant): Declare.
3865         * config/nds32/nds32.c (nds32_print_operand): Deal with more cases.
3866         * config/nds32/nds32.md (addsi3, *add_srli): Refine implementation for
3867         bit-wise operations.
3868         (andsi3, *andsi3): Ditto.
3869         (iorsi3, *iorsi3, *or_slli, *or_srli): Ditto.
3870         (xorsi3, *xorsi3, *xor_slli, *xor_srli): Ditto.
3871         (<shift>si3, *ashlsi3, *ashrsi3, *lshrsi3, *rotrsi3): Ditto.
3872         * config/nds32/predicates.md (nds32_rimm5u_operand, nds32_and_operand,
3873         nds32_ior_operand, nds32_xor_operand): New predicates.
3874
3875 2018-04-05  Chung-Ju Wu  <jasonwucj@gmail.com>
3876
3877         * config/nds32/nds32.md (add<mode>3, sub<mode>3): Rename to ...
3878         (addsi3, subsi3): ... this.
3879
3880 2018-04-05  Chung-Ju Wu  <jasonwucj@gmail.com>
3881
3882         * config/nds32/nds32.md (*sub_srli, *and_slli): Fine-tune predicator.
3883
3884 2018-04-05  Chung-Ju Wu  <jasonwucj@gmail.com>
3885
3886         * config/nds32/nds32.md: Adjust indention.
3887
3888 2018-04-05  Kito Cheng  <kito.cheng@gmail.com>
3889
3890         * config/nds32/nds32.md (feature): New attribute.
3891
3892 2018-04-05  Chung-Ju Wu  <jasonwucj@gmail.com>
3893
3894         * config/nds32/nds32.md (subtype): New attribute.
3895
3896 2018-04-04  Thomas Preud'homme  <thomas.preudhomme@arm.com>
3897
3898         PR target/85203
3899         * config/arm/arm-builtins.c (arm_expand_builtin): Change
3900         expansion to perform a bitwise AND of the argument followed by a
3901         boolean negation of the result.
3902
3903 2018-04-04  Peter Bergner  <bergner@vnet.ibm.com>
3904
3905         PR rtl-optimization/84878
3906         * ddg.c (add_cross_iteration_register_deps): Use DF_REF_BB to determine
3907         the basic block.  Assert the use reference is not artificial and that
3908         it has an associated insn.
3909
3910 2018-04-04  Michael Matz  <matz@suse.de>
3911
3912         * builtins.c (compute_objsize): Pass correct operand
3913         to array_at_struct_end_p.
3914
3915 2018-04-04  Richard Biener  <rguenther@suse.de>
3916
3917         PR lto/85176
3918         * dwarf2out.c (dwarf2out_register_external_die): Peel namespaces
3919         from contexts for DINFO_LEVEL_TERSE and below.
3920
3921 2018-04-04  Kito Cheng  <kito.cheng@gmail.com>
3922
3923         * config/nds32/nds32-doubleword.md (move_<mode>): Require
3924         resiter_operand condition.
3925         * config/nds32/nds32.md (*move<mode>): Ditto.
3926
3927 2018-04-04  Kito Cheng  <kito.cheng@gmail.com>
3928             Monk Chiang  <sh.chiang04@gmail.com>
3929
3930         * config/nds32/nds32.md (movmisalign<mode>): New pattern.
3931
3932 2018-04-04  Chung-Ju Wu  <jasonwucj@gmail.com>
3933
3934         * config/nds32/nds32.md (movqi, movhi): Merge into mov<mode>.
3935
3936 2018-04-04  Chung-Ju Wu  <jasonwucj@gmail.com>
3937             Kito Cheng  <kito.cheng@gmail.com>
3938
3939         * config/nds32/nds32-md-auxiliary.c (nds32_inverse_cond_code,
3940         nds32_cond_code_str, output_cond_branch,
3941         output_cond_branch_compare_zero, nds32_expand_cbranch,
3942         nds32_expand_cstore, nds32_expand_movcc,
3943         nds32_output_cbranchsi4_equality_zero,
3944         nds32_output_cbranchsi4_equality_reg,
3945         nds32_output_cbranchsi4_equality_reg_or_const_int,
3946         nds32_output_cbranchsi4_greater_less_zero: New functions.
3947         * config/nds32/nds32-protos.h (nds32_expand_cbranch,
3948         nds32_expand_cstore, nds32_expand_movcc,
3949         nds32_output_cbranchsi4_equality_zero,
3950         nds32_output_cbranchsi4_equality_reg,
3951         nds32_output_cbranchsi4_equality_reg_or_const_int,
3952         nds32_output_cbranchsi4_greater_less_zero): Declare.
3953         * config/nds32/predicates.md (nds32_movecc_comparison_operator,
3954         nds32_rimm11s_operand): New predicates.
3955         * config/nds32/nds32.h (nds32_expand_result_type): New enum type.
3956         * config/nds32/nds32.md: Rewrite all the branch and conditional move
3957         patterns.
3958
3959 2018-04-04  Kito Cheng  <kito.cheng@gmail.com>
3960
3961         * config/nds32/nds32-doubleword.md: Refine all the instruction type.
3962         * config/nds32/nds32.md: Ditto.
3963         * config/nds32/pipelines.md: Ditto.
3964
3965 2018-04-04  Richard Biener  <rguenther@suse.de>
3966
3967         PR tree-optimization/85168
3968         * tree-ssa-sccvn.c (vn_reference_maybe_forwprop_address): Avoid
3969         propagating abnormals.
3970
3971 2018-04-04  Chung-Ju Wu  <jasonwucj@gmail.com>
3972
3973         * config/nds32/nds32.md (enabled): Use yes/no for this attribute.
3974
3975 2018-04-04  Chung-Ju Wu  <jasonwucj@gmail.com>
3976             Kito Cheng  <kito.cheng@gmail.com>
3977
3978         * config/nds32/nds32-md-auxiliary.c (nds32_long_call_p): New function.
3979         * config/nds32/nds32-protos.h (nds32_long_call_p): Declare.
3980         * config/nds32/nds32.c (nds32_function_ok_for_sibcall): New function.
3981         (TARGET_FUNCTION_OK_FOR_SIBCALL): Define.
3982         * config/nds32/nds32.md (sibcall_internal): New.
3983         (sibcall_register): Remove.
3984         (sibcall_immediate): Remove.
3985         (sibcall_value_internal): New.
3986         (sibcall_value_register): Remove.
3987         (sibcall_value_immediate): Remove.
3988         * config/nds32/predicates.md (nds32_general_register_operand): New.
3989         (nds32_call_address_operand): New.
3990
3991 2018-04-03  Jakub Jelinek  <jakub@redhat.com>
3992
3993         PR rtl-optimization/85167
3994         * shrink-wrap.c (move_insn_for_shrink_wrap): Don't set bb_uses and
3995         bb_defs if *split_p, instead preinitialize it to NULL.
3996
3997         PR tree-optimization/85156
3998         * builtins.c (fold_builtin_expect): Use save_expr on arg1 to avoid
3999         evaluating the argument multiple times.
4000
4001 2018-04-03  Bill Schmidt  <wschmidt@linux.ibm.com>
4002
4003         * config/rs6000/emmintrin.h (_mm_cvtpd_epi32): Use __vector rather
4004         than vector.
4005         (_mm_cvtpd_ps): Likewise.
4006         (_mm_cvttpd_epi32): Likewise.
4007         * config/rs6000/mmintrin.h (_mm_unpacklo_pi8): Likewise.
4008         * config/rs6000/xmmintrin.h: For strict-ANSI C++ or C11, undefine
4009         vector, pixel, and bool following altivec.h include.
4010
4011 2018-04-03  Martin Sebor  <msebor@redhat.com>
4012
4013         * doc/extend.texi (Common Function Attributes): Clarify.
4014         (const attribute): Likewise.
4015         (pure attribute): Likewise.
4016
4017 2018-04-03  Jakub Jelinek  <jakub@redhat.com>
4018
4019         PR target/85169
4020         * config/i386/i386.c (ix86_expand_vector_set): Use
4021         HOST_WIDE_INT_1U << elt instead of 1 << elt.  Formatting fix.
4022
4023 2018-04-03  Uros Bizjak  <ubizjak@gmail.com>
4024
4025         * config/i386/i386.c (emit_i387_cw_initialization): Always use logic
4026         instructions when changing rounding bits to preserve precision bits
4027         in the x87 control word.
4028
4029 2018-04-03  Martin Liska  <mliska@suse.cz>
4030
4031         PR tree-optimization/82491
4032         * rtl.h (strip_offset_and_add): Replace += suboffset with
4033         poly_uint64 () + suboffset.
4034
4035 2018-03-29  Martin Liska  <mliska@suse.cz>
4036             Martin Jambor  <mjambor@suse.cz>
4037
4038         PR ipa/84947
4039         * ipa-cp.c (propagate_bits_across_jump_function): Bail out if
4040         param_type is not an integral or pointer type.
4041
4042 2018-04-03  Richard Biener  <rguenther@suse.de>
4043
4044         * sese.h (recompute_all_dominators): Remove.
4045
4046 2018-04-02  Martin Sebor  <msebor@redhat.com>
4047
4048         * doc/invoke.texi (-Wrestrict): Fix typos.
4049
4050 2018-04-02  Jim Wilson  <jimw@sifive.com>
4051
4052         * config/riscv/riscv.h (SHIFT_COUNT_TRUNCATED): Set to zero.
4053         * config/riscv/riscv.md (<optab>si3): Use QImode shift count.
4054         (<optab>di3, <optab>si3_extend): Likewise.
4055         (<optab>si3_mask, <optab>si3_mask_1): New.
4056         (<optab>di3_mask, <optab>di3_mask_1): New.
4057         (<optab>si3_extend_mask, <optab>si3_extend_mask_1): New.
4058         (lshrsi3_zero_extend_1): Use VOIDmode shift count.
4059         * config/riscv/sync.md (atomic_test_and_set): Emit QImode shift count.
4060
4061 2018-04-02  Gerald Pfeifer  <gerald@pfeifer.com>
4062
4063         * doc/cpp.texi (Variadic Macros): Fix line continuation in an
4064         example.
4065
4066 2018-04-02  Chung-Ju Wu  <jasonwucj@gmail.com>
4067
4068         * config/nds32/nds32.c (TARGET_CANONICALIZE_COMPARISON): Define.
4069         (nds32_canonicalize_comparison): New function.
4070
4071 2018-04-01  Chung-Ju Wu  <jasonwucj@gmail.com>
4072             Kito Cheng  <kito.cheng@gmail.com>
4073             Kuan-Lin Chen  <kuanlinchentw@gmail.com>
4074
4075         * config.gcc (nds32): Add nds32-relax-opt.o into extra_objs.
4076         * config/nds32/constants.md (unspec_volatile_element): Add
4077         UNSPEC_VOLATILE_RELAX_GROUP.
4078         * config/nds32/nds32-relax-opt.c: New file.
4079         * config/nds32/nds32-predicates.c
4080         (nds32_symbol_load_store_p): New function.
4081         * config/nds32/nds32-protos.h
4082         (nds32_symbol_load_store_p): Declare function.
4083         (make_pass_nds32_relax_opt): Declare new rtl pass function.
4084         * config/nds32/nds32.c
4085         (nds32_register_pass): New function to register pass.
4086         (nds32_register_passes): New function to register passes.
4087         * config/nds32/nds32.md (relax_group): New pattern.
4088         * config/nds32/nds32.opt (mrelax-hint): New option.
4089         * config/nds32/t-nds32 (nds32-relax-opt.o): New dependency.
4090
4091 2018-04-01  Kito Cheng  <kito.cheng@gmail.com>
4092
4093         * config/nds32/t-nds32: Modify files dependency.
4094
4095 2018-04-01  Chung-Ju Wu  <jasonwucj@gmail.com>
4096
4097         * config/nds32/nds32.h (FUNCTION_PROFILER): Output newline character.
4098         (PROFILE_HOOK): Define its implementation.
4099
4100 2018-04-01  Chung-Ju Wu  <jasonwucj@gmail.com>
4101
4102         * config/nds32/nds32.h (WCHAR_TYPE, WCHAR_TYPE_SIZE): Use unsigned int
4103         type and 32-bit size.
4104
4105 2018-04-01  Jakub Jelinek  <jakub@redhat.com>
4106
4107         PR middle-end/85090
4108         * config/i386/sse.md (V): Add V64QI and V32HI for TARGET_AVX512F.
4109         (V_128_256): New mode iterator.
4110         (*avx512dq_vextract<shuffletype>64x2_1 splitter): New define_split.
4111         (*avx512f_vextract<shuffletype>32x4_1 splitter): Likewise.
4112         (xop_pcmov_<mode><avxsizesuffix>): Use V_128_256 mode iterator instead
4113         of V.
4114         * config/i386/i386.c (ix86_expand_vector_set): Improve V32HImode and
4115         V64QImode expansion for !TARGET_AVX512BW && TARGET_AVX512F.
4116
4117 2018-03-31  Segher Boessenkool  <segher@kernel.crashing.org>
4118
4119         PR target/83315
4120         * config/rs6000/xmmintrin.h (_mm_set_ps, _mm_max_ps): Handle (quiet)
4121         NaN inputs correctly.
4122
4123 2018-03-30  Peter Bergner  <bergner@vnet.ibm.com>
4124
4125         PR target/80546
4126         * config/rs6000/vsx.md (??r): New mode attribute.
4127         (*vsx_mov<mode>_64bit): Use it.
4128         (*vsx_mov<mode>_32bit): Likewise.
4129
4130 2018-03-30  Martin Sebor  <msebor@redhat.com>
4131
4132         PR tree-optimization/84818
4133         * builtins.c (check_access): Use warning_n.
4134
4135 2018-03-30  Aaron Sawdey  <acsawdey@linux.vnet.ibm.com>
4136
4137         PR target/83822
4138         * config/rs6000/rs6000-string.c (expand_compare_loop): Fix redundant
4139         condition.
4140         * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): Fix redundant
4141         condition.
4142
4143 2018-03-30  Julia Koval  <julia.koval@intel.com>
4144
4145         PR target/84413
4146         * x86-tune.def (movx, partial_reg_dependency): Enable for
4147         m_SKYLAKE_AVX512.
4148
4149 2018-03-29  Vladimir Makarov  <vmakarov@redhat.com>
4150
4151         PR inline-asm/84985
4152         * lra-constraints.c (process_alt_operands): Move setting
4153         this_alternative_matches below.
4154
4155 2018-03-29  Martin Liska  <mliska@suse.cz>
4156
4157         PR lto/84995.
4158         * doc/invoke.texi: Document how LTO works with debug info.
4159         Describe auto-load support of binutils.  Mention 'x86-64'
4160         as valid option value of -march option.
4161
4162 2018-03-29  Jakub Jelinek  <jakub@redhat.com>
4163
4164         * config/i386/sse.md (<avx512>_blendm<mode>): Use <sseintprefix>.
4165
4166         PR c/85094
4167         * fold-const.c (operand_equal_p): Handle DEBUG_BEGIN_STMT.
4168         For STATEMENT_LIST, pass down OEP_LEXICOGRAPHIC and maybe
4169         OEP_NO_HASH_CHECK for recursive call, to avoid exponential
4170         checking.
4171
4172 2018-03-28  Peter Bergner  <bergner@vnet.ibm.com>
4173
4174         PR target/84912
4175         * config/rs6000/rs6000.h (RS6000_BTM_POWERPC64): New define.
4176         (RS6000_BTM_COMMON): Add RS6000_BTM_POWERPC64.
4177         * config/rs6000/rs6000.c (rs6000_builtin_mask_calculate): Add support
4178         for RS6000_BTM_POWERPC64.
4179         (rs6000_invalid_builtin): Add handling for RS6000_BTM_POWERPC64
4180         (rs6000_builtin_mask_names): Add RS6000_BTM_POWERPC64.
4181         * config/rs6000/rs6000-builtin.def (BU_P7_POWERPC64_MISC_2): New macro
4182         definition.
4183         (DIVDE): Use it.
4184         (DIVDEU): Likewise.
4185
4186 2018-03-28 Carl Love  <cel@us.ibm.com>
4187
4188         Revert
4189         2017-09-27  Carl Love  <cel@us.ibm.com>
4190
4191         * config/rs6000/rs6000-builtin.def (BU_FP_1MISC_1): Add define macro.
4192         (FCTID, FCTIW): Add BU_FP_MISC_1 macro expansion for builtins.
4193         * config/rs6000/rs6000.md (lrintsfsi2): Add define_insn for the
4194         fctiw instruction.
4195
4196 2018-03-28  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
4197
4198         * config/rs6000/xmmintrin.h (_mm_max_pi16): Use __vector __bool
4199         instead of __vector bool.
4200         (_mm_max_pu8): Likewise.
4201         (_mm_min_pi16): Likewise.
4202
4203 2018-03-28  Peter Bergner  <bergner@vnet.ibm.com>
4204
4205         PR target/84912
4206         * config/rs6000/rs6000-builtin.def (DIVWEO): Delete macro expansion.
4207         (DIVWEUO): Likewise.
4208         (DIVDEO): Likewise.
4209         (DIVDEUO): Likewise.
4210         * config/rs6000/rs6000.c (builtin_function_type): Remove support for
4211         DIVWEUO and DIVDEUO.
4212         * config/rs6000/rs6000.md (UNSPEC_DIVEO, UNSPEC_DIVEUO): Delete unspecs.
4213         (UNSPEC_DIV_EXTEND): Remove deleted unspecs.
4214         (div_extend): Likewise.
4215         * doc/extend.texi (__builtin_divweo): Remove documentation for deleted
4216         builtin function.
4217         (__builtin_divweuo): Likewise.
4218         (__builtin_divdeo): Likewise.
4219         (__builtin_divdeuo): Likewise.
4220
4221 2018-03-28  Jakub Jelinek  <jakub@redhat.com>
4222
4223         PR target/85095
4224         * config/i386/i386.md (*add<mode>3_carry_0, *addsi3_carry_zext_0,
4225         *sub<mode>3_carry_0, *subsi3_carry_zext_0): New patterns.
4226
4227         PR tree-optimization/82004
4228         * gimple-match-head.c (optimize_pow_to_exp): New function.
4229         * match.pd (pow(C,x) -> exp(log(C)*x)): Wrap with #if GIMPLE.
4230         Don't fold to exp if optimize_pow_to_exp is false.
4231
4232 2018-03-28  Martin Liska  <mliska@suse.cz>
4233
4234         PR other/84819
4235         * calls.c (initialize_argument_information): Fix trailing space.
4236         * common.opt: Fix typo and provide better explanation for
4237         -fsanitize-coverage option.
4238         * config/i386/i386.opt: Fix typo.
4239
4240 2018-03-28  Jakub Jelinek  <jakub@redhat.com>
4241             Martin Liska  <mliska@suse.cz>
4242
4243         PR sanitizer/85081
4244         * gimplify.c (asan_poison_variable): Don't do the check for
4245         gimplify_omp_ctxp here.
4246         (gimplify_decl_expr): Do it here.
4247         (gimplify_target_expr): Likewise.
4248
4249 2018-03-28  Martin Liska  <mliska@suse.cz>
4250
4251         PR target/84988
4252         * config/i386/i386.c (ix86_function_arg_advance): Do not call
4253         chkp_type_bounds_count if MPX is not enabled.
4254
4255 2018-03-27  Chung-Ju Wu  <jasonwucj@gmail.com>
4256
4257         * config/nds32/nds32.h (BRANCH_COST): Adjust cost.
4258
4259 2018-03-27  Michael Meissner  <meissner@linux.vnet.ibm.com>
4260
4261         PR target/84914
4262         * config/rs6000/rs6000.c (create_complex_muldiv): New helper
4263         function to create the function decl for complex long double
4264         multiply and divide for -mabi=ieeelongdouble.
4265         (init_float128_ieee): Call it.
4266
4267 2018-03-27  H.J. Lu  <hongjiu.lu@intel.com>
4268
4269         PR target/85044
4270         * config/i386/i386.c (ix86_trampoline_init): Insert ENDBR for
4271         -fcf-protection=branch -mibt.
4272         * config/i386/i386.h (TRAMPOLINE_SIZE): Increased by 4 bytes.
4273
4274 2018-03-27  Ramana Radhakrishnan  <ramana.radhakrishnan@arm.com>
4275
4276         PR target/81863
4277         * config/arm/arm.c (arm_valid_symbolic_address): Handle arm_word_relocations
4278
4279 2018-03-27  Cesar Philippidis  <cesar@codesourcery.com>
4280
4281         PR target/85056
4282         * config/nvptx/nvptx.c (nvptx_assemble_decl_begin): Add '[]' to
4283         extern array declarations.
4284
4285 2018-03-27  Richard Biener  <rguenther@suse.de>
4286
4287         PR middle-end/84067
4288         * match.pd ((A * C) +- (B * C) -> (A+-B) * C): Guard with
4289         explicit single_use checks.
4290
4291 2018-03-27  Richard Biener  <rguenther@suse.de>
4292
4293         PR tree-optimization/85082
4294         * tree-ssa-sccvn.c (vn_reference_lookup_or_insert_for_pieces):
4295         Valueize the VUSE.
4296
4297 2018-03-27  Ramana Radhakrishnan  <ramana.radhakrishnan@arm.com>
4298
4299         * config.gcc (aarch64*-*-linux*): New TARGET_DEFAULT_ASYNC_UNWIND_TABLES
4300         * common/config/aarch64/aarch64-common.c (aarch64_optimization_table[]):
4301         Turn on fasynchronous-unwind-tables and funwind-tables.
4302
4303 2018-03-26  Uros Bizjak  <ubizjak@gmail.com>
4304
4305         PR target/85073
4306         * config/i386/i386.md (*bmi_blsr_<mode>_cmp): New insn pattern.
4307         (*bmi_blsr_<mode>_ccz): Ditto.
4308
4309 2018-03-26  Tom de Vries  <tom@codesourcery.com>
4310
4311         PR tree-optimization/85063
4312         * omp-general.c (offloading_function_p): New function.  Factor out
4313         of ...
4314         * omp-offload.c (pass_omp_target_link::gate): ... here.
4315         * omp-general.h (offloading_function_p): Declare.
4316         * tree-switch-conversion.c (build_one_array): Mark CSWTCH.x variable
4317         with attribute omp declare target for offloading functions.
4318
4319 2018-03-24  Richard Sandiford  <richard.sandiford@linaro.org>
4320
4321         PR tree-optimization/84005
4322         * tree-data-ref.h (get_base_for_alignment): Declare.
4323         * tree-data-ref.c (get_base_for_alignment_1): New function.
4324         (get_base_for_alignment): Likewise.
4325         * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Use
4326         get_base_for_alignment to find a suitable base object, instead
4327         of always using drb->base_address.
4328
4329 2018-03-23  Jakub Jelinek  <jakub@redhat.com>
4330
4331         PR inline-asm/85022
4332         * emit-rtl.c (init_emit_regs): Indicate that VOIDmode MEMs don't have
4333         known size by default.
4334
4335 2018-03-23  Vladimir Makarov  <vmakarov@redhat.com>
4336
4337         PR inline-asm/85030
4338         * lra-constraints.c (process_alt_operands): Don't match BLKmode
4339         and non BLKmode operands.
4340
4341 2018-03-23  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
4342
4343         PR target/85026
4344         * config/arm/arm.md (unaligned_loadhis): Remove first alternative.
4345         Clean up attributes.
4346
4347 2018-03-23  Richard Biener  <rguenther@suse.de>
4348
4349         PR debug/85020
4350         * dwarf2out.c (rtl_for_decl_location): Do not generate RTL early when
4351         we are going to emit early debug for LTO.
4352
4353 2018-03-23  Jakub Jelinek  <jakub@redhat.com>
4354
4355         PR inline-asm/85034
4356         * function.c (match_asm_constraints_1): Don't optimize if input
4357         doesn't satisfy general_operand predicate for output's mode.
4358
4359         PR inline-asm/85022
4360         * alias.c (write_dependence_p): Don't require for x_canonicalized
4361         non-VOIDmode if x has VOIDmode.
4362
4363         PR sanitizer/85029
4364         * sanopt.c (maybe_optimize_ubsan_ptr_ifn): If DECL_REGISTER (base),
4365         just don't try to optimize it rather than assert it never happens.
4366
4367 2018-03-22  Kelvin Nilsen  <kelvin@gcc.gnu.org>
4368
4369         * config/rs6000/rs6000-builtin.def: Remove various BU_ALTIVEC_X
4370         macro expansions for definition of ST_INTERNAL_<mode> and
4371         LD_INTERNAL_<mode> builtins.
4372         * config/rs6000/rs6000-protos.h (rs6000_address_for_altivec):
4373         Remove prototype.
4374         * config/rs6000/rs6000.c (altivec_expand_ld_builtin): Delete this
4375         function.
4376         (altivec_expand_st_builtin): Likewise.
4377         (altivec_expand_builtin): Remove calls to deleted functions.
4378         (rs6000_address_for_altivec): Delete this function.
4379         * config/rs6000/vector.md: Remove expands for
4380         vector_altivec_load_<mode> and vector_altivec_store_<mode>.
4381
4382 2018-03-22  Sudakshina Das  <sudi.das@arm.com>
4383
4384         PR target/84826
4385         * config/arm/arm.h (machine_function): Add static_chain_stack_bytes.
4386         * config/arm/arm.c (arm_compute_static_chain_stack_bytes): Avoid
4387         re-computing once computed.
4388         (arm_expand_prologue): Compute machine->static_chain_stack_bytes.
4389         (arm_init_machine_status): Initialize
4390         machine->static_chain_stack_bytes.
4391
4392 2018-03-22  Kelvin Nilsen  <kelvin@gcc.gnu.org>
4393
4394         PR target/84760
4395         * doc/extend.texi: Add four new prototypes for vec_ld.
4396         * config/rs6000/rs6000-builtin.def (LVX_V1TI): Reorder symbol
4397         definitions for more logical presentation.
4398         * config/rs6000/rs6000-c.c: (altivec_overloaded_builtins): Add
4399         entries for V1TI variants of __builtin_altivec_ld builtin.
4400         * config/rs6000/rs6000.c: (altivec_expand_lv_builtin): Add test and
4401         handling of V1TI variant of LVX icode pattern.
4402         (altivec_expand_builtin): Add case for ALTIVEC_BUILTIN_LVX_V1TI.
4403         (rs6000_gimple_fold_builtin): Likewise.
4404         (altivec_init_builtins): Add code to define
4405         __builtin_altivec_lvx_v1ti function.
4406
4407 2018-03-22  Jakub Jelinek  <jakub@redhat.com>
4408
4409         PR inline-asm/84941
4410         * function.c (match_asm_constraints_1): Don't do the optimization
4411         if input isn't a REG, SUBREG, MEM or constant.
4412
4413 2018-03-22  Tom de Vries  <tom@codesourcery.com>
4414
4415         PR tree-optimization/84956
4416         * tree-ssa-tail-merge.c (find_clusters_1): Skip bbs with
4417         bb_has_abnormal_pred.
4418
4419 2018-03-22  Jakub Jelinek  <jakub@redhat.com>
4420
4421         PR sanitizer/85018
4422         * dwarf2asm.c (dw2_output_indirect_constant_1): Set
4423         DECL_INITIAL (decl) to decl at the end.
4424         * varasm.c (use_blocks_for_decl_p): Revert the 2018-03-20 change,
4425         adjust the comment.
4426
4427 2018-03-21  Joseph Myers  <joseph@codesourcery.com>
4428
4429         * doc/extend.texi (__builtin_tgmath): Document when complex
4430         integer types are treated as _Complex _Float64.
4431
4432 2018-03-21  Tom de Vries  <tom@codesourcery.com>
4433
4434         * doc/extend.texi (__builtin_extend_pointer): Remove pasto.
4435
4436 2018-03-21  Jakub Jelinek  <jakub@redhat.com>
4437
4438         PR tree-optimization/84960
4439         * tree-cfg.c (remove_bb): Don't move forced labels into bb->prev_bb
4440         if it is ENTRY block, move them into single succ of ENTRY in that case.
4441
4442 2018-03-21  Richard Sandiford  <richard.sandiford@linaro.org>
4443
4444         PR tree-optimization/84811
4445         * poly-int.h (poly_span_traits): Remove the T3 parameter and
4446         promote HOST_WIDE_INT T2 - T1 results to unsigned HOST_WIDE_INT.
4447         (maybe_in_range_p, known_in_range_p, ranges_known_overlap_p):
4448         (known_subrange_p): Update accordingly.  Cast each value involved
4449         in the size comparison, rather than casting the result of the
4450         subtraction.
4451
4452 2018-03-21  Jakub Jelinek  <jakub@redhat.com>
4453
4454         PR tree-optimization/84982
4455         * gimple-ssa-store-merging.c (invert_op): Handle boolean inversion
4456         by flipping the least significant bit rather than all bits from
4457         bitpos to bitpos + bitsize - 1.
4458
4459 2018-03-21  Nathan Sidwell  <nathan@acm.org>
4460
4461         * doc/extend.texi (Deprecated Features): Remove mention of
4462         long-deleted deprecations.
4463
4464 2018-03-21  Rainer Orth  <ro@CeBiTec.Uni-Bielefeld.DE>
4465
4466         PR jit/84288
4467         * configure.ac (gcc_cv_ld_soname) <*-*-solaris2*>: Set.
4468         * configure: Regenerate.
4469
4470 2018-03-21  Tom de Vries  <tom@codesourcery.com>
4471
4472         PR tree-optimization/83126
4473         * tree-parloops.c (num_phis): New function.
4474         (gen_parallel_loop): Detect and handle canonicalize_loop_ivs failure.
4475
4476 2018-03-21  Nathan Sidwell  <nathan@acm.org>
4477
4478         * doc/extend.texi (Deprecated Features): Update deprecated flags,
4479         mention anon-struct/union members and trailing attributes.
4480
4481 2018-03-21  Bin Cheng  <bin.cheng@arm.com>
4482
4483         PR tree-optimization/84969
4484         * tree-loop-distribution.c (fuse_memset_builtins): Don't reorder
4485         builtin memset partitions if they set different rhs values.
4486
4487 2018-03-21  Jakub Jelinek  <jakub@redhat.com>
4488
4489         PR rtl-optimization/84989
4490         * simplify-rtx.c (simplify_unary_operation_1): Don't try to simplify
4491         VEC_DUPLICATE with scalar result mode.
4492
4493 2018-03-21  Martin Liska  <mliska@suse.cz>
4494
4495         PR ipa/84963
4496         * ipa-icf.c (sem_item_optimizer::fixup_points_to_sets): Remove
4497         not intended return statement.
4498
4499 2018-03-21  Martin Liska  <mliska@suse.cz>
4500
4501         PR target/84988
4502         * tree-chkp.c (CHKP_ARRAY_MAX_CHECK_STEPS): Define a new macro.
4503         (chkp_find_bound_slots_1): Limit number of iterations.
4504
4505 2018-03-20  David H. Gutteridge  <dhgutteridge@sympatico.ca>
4506
4507         PR target/84838
4508         * Minor grammar fixes for x86 options.
4509
4510 2018-03-20  Jakub Jelinek  <jakub@redhat.com>
4511
4512         PR debug/84875
4513         * dce.c (delete_unmarked_insns): Don't remove frame related noop moves
4514         holding REG_CFA_RESTORE notes, instead turn them into a USE.
4515
4516 2018-03-20  Peter Bergner  <bergner@vnet.ibm.com>
4517
4518         PR target/83789
4519         * config/rs6000/altivec.md (altivec_lvx_<mode>_2op): Delete define_insn.
4520         (altivec_lvx_<mode>_1op): Likewise.
4521         (altivec_stvx_<mode>_2op): Likewise.
4522         (altivec_stvx_<mode>_1op): Likewise.
4523         (altivec_lvx_<VM2:mode>): New define_expand.
4524         (altivec_stvx_<VM2:mode>): Likewise.
4525         (altivec_lvx_<VM2:mode>_2op_<P:mptrsize>): New define_insn.
4526         (altivec_lvx_<VM2:mode>_1op_<P:mptrsize>): Likewise.
4527         (altivec_stvx_<VM2:mode>_2op_<P:mptrsize>): Likewise.
4528         (altivec_stvx_<VM2:mode>_1op_<P:mptrsize>): Likewise.
4529         * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Use new expanders.
4530         (rs6000_gen_lvx): Likewise.
4531         * config/rs6000/rs6000.c (altivec_expand_lv_builtin): Likewise.
4532         (altivec_expand_stv_builtin): Likewise.
4533         (altivec_expand_builtin): Likewise.
4534         * config/rs6000/vector.md: Likewise.
4535
4536 2018-03-20  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
4537
4538         PR target/82518
4539         * config/arm/arm.c (arm_array_mode_supported_p): Return false for
4540         BYTES_BIG_ENDIAN.
4541
4542 2018-03-20  Richard Biener  <rguenther@suse.de>
4543
4544         PR target/84986
4545         * config/i386/i386.c (ix86_add_stmt_cost): Only cost
4546         sign-conversions as zero, fall back to standard scalar_stmt
4547         cost for the rest.
4548
4549 2018-03-20  Martin Liska  <mliska@suse.cz>
4550
4551         PR ipa/84825
4552         * predict.c (rebuild_frequencies): Handle case when we have
4553         PROFILE_ABSENT, but flag_guess_branch_prob is false.
4554
4555 2018-03-20  Jakub Jelinek  <jakub@redhat.com>
4556
4557         PR target/84990
4558         * dwarf2asm.c (dw2_output_indirect_constant_1): Temporarily turn off
4559         flag_section_anchors.
4560         * varasm.c (use_blocks_for_decl_p): Remove hack for
4561         dw2_force_const_mem.
4562
4563         PR target/84845
4564         * config/aarch64/aarch64.md (*aarch64_reg_<mode>3_neg_mask2): Rename
4565         to ...
4566         (*aarch64_<optab>_reg_<mode>3_neg_mask2): ... this.  If pseudos can't
4567         be created, use lowpart_subreg of operands[0] rather than operands[0]
4568         itself.
4569         (*aarch64_reg_<mode>3_minus_mask): Rename to ...
4570         (*aarch64_ashl_reg_<mode>3_minus_mask): ... this.
4571         (*aarch64_<optab>_reg_di3_mask2): Use const_int_operand predicate
4572         and n constraint instead of aarch64_shift_imm_di and Usd.
4573         (*aarch64_reg_<optab>_minus<mode>3): Rename to ...
4574         (*aarch64_<optab>_reg_minus<mode>3): ... this.
4575
4576 2018-03-20  Sudakshina Das  <sudi.das@arm.com>
4577
4578         PR target/82989
4579         * config/arm/neon.md (ashldi3_neon): Update ?s for constraints
4580         to favor GPR over NEON registers.
4581         (<shift>di3_neon): Likewise.
4582
4583 2018-03-20  Tom de Vries  <tom@codesourcery.com>
4584
4585         PR target/84952
4586         * config/nvptx/nvptx.c (nvptx_single): Don't neuter bar.sync.
4587         (nvptx_process_pars): Emit bar.sync asap and alap.
4588
4589 2018-03-20  Tom de Vries  <tom@codesourcery.com>
4590
4591         PR target/84954
4592         * config/nvptx/nvptx.c (prevent_branch_around_nothing): Also update
4593         seen_label if seen_label is already set.
4594
4595 2018-03-20  Jakub Jelinek  <jakub@redhat.com>
4596
4597         PR target/84945
4598         * config/i386/i386.c (fold_builtin_cpu): For features above 31
4599         use __cpu_features2 variable instead of __cpu_model.__cpu_features[0].
4600         Use 1U instead of 1.  Formatting fixes.
4601
4602         PR c/84953
4603         * builtins.c (fold_builtin_strpbrk): For strpbrk(x, "") use type
4604         instead of TREE_TYPE (s1) for the return value.
4605
4606 2018-03-19  Jakub Jelinek  <jakub@redhat.com>
4607
4608         PR tree-optimization/84946
4609         * gimple-ssa-store-merging.c (mem_valid_for_store_merging): Compute
4610         bitsize + bitsize in poly_uint64 rather than poly_int64.
4611
4612         PR sanitizer/78651
4613         * dwarf2asm.c: Include fold-const.c.
4614         (dw2_output_indirect_constant_1): Set DECL_INITIAL (decl) to ADDR_EXPR
4615         of decl rather than decl itself.
4616
4617         PR rtl-optimization/84643
4618         * memmodel.h (enum memmodel): Add MEMMODEL_MAX enumerator.
4619
4620 2018-03-19  Maxim Ostapenko  <m.ostapenko@samsung.com>
4621
4622         PR sanitizer/78651
4623         * dwarf2asm.c (dw2_output_indirect_constant_1): Disable ASan before
4624         calling assemble_variable.
4625
4626 2018-03-19  Sudakshina Das  <sudi.das@arm.com>
4627
4628         PR target/81647
4629         * config/aarch64/aarch64-simd.md (vec_cmp<mode><v_int_equiv>): Modify
4630         instructions for UNLT, UNLE, UNGT, UNGE, UNEQ, UNORDERED and ORDERED.
4631
4632 2018-03-19  Jim Wilson  <jimw@sifive.com>
4633
4634         PR bootstrap/84856
4635         * config/riscv/riscv.c (riscv_function_arg_boundary): Use
4636         PREFERRED_STACK_BOUNDARY instead of STACK_BOUNDARY.
4637         (riscv_first_stack_step): Likewise.
4638         (riscv_option_override): Use STACK_BOUNDARY instead of
4639         MIN_STACK_BOUNDARY.
4640         * config/riscv/riscv.h (STACK_BOUNDARY): Renamed from
4641         MIN_STACK_BOUNDARY.
4642         (BIGGEST_ALIGNMENT): Set to 128.
4643         (PREFERRED_STACK_BOUNDARY): Renamed from STACK_BOUNDARY.
4644         (RISCV_STACK_ALIGN): Use PREFERRED_STACK_BOUNDARY instead of
4645         STACK_BOUNDARY.
4646
4647 2018-03-19  Richard Biener  <rguenther@suse.de>
4648
4649         PR tree-optimization/84933
4650         * tree-vrp.c (set_and_canonicalize_value_range): Treat out-of-bound
4651         values as -INF/INF when canonicalizing an ANTI_RANGE to a RANGE.
4652
4653 2018-03-19  Richard Biener  <rguenther@suse.de>
4654
4655         PR tree-optimization/84859
4656         * tree-ssa-phiopt.c (single_trailing_store_in_bb): New function.
4657         (cond_if_else_store_replacement): Perform sinking operation on
4658         single-store BBs regardless of MAX_STORES_TO_SINK setting.
4659         Generalize what a BB with a single eligible store is.
4660
4661 2018-03-19  Richard Biener  <rguenther@suse.de>
4662
4663         PR tree-optimization/84929
4664         * tree-data-ref.c (analyze_siv_subscript_cst_affine): Guard
4665         chrec_is_positive against non-chrec arg.
4666
4667 2018-03-19  Tamar Christina  <tamar.christina@arm.com>
4668
4669         PR target/84711
4670         * config/arm/arm.c (arm_can_change_mode_class): revert r258554.
4671
4672 2018-03-18  Martin Liska  <mliska@suse.cz>
4673
4674         PR rtl-optimization/84635
4675         * regrename.c (build_def_use): Use matches_mode only when
4676         matches >= 0.
4677
4678 2018-03-18  Richard Sandiford  <richard.sandiford@linaro.org>
4679
4680         PR tree-optimization/84913
4681         * tree-vect-loop.c (vectorizable_reduction): Don't try to
4682         vectorize chains of COND_EXPRs.
4683
4684 2018-03-18  Chung-Ju Wu  <jasonwucj@gmail.com>
4685
4686         * config/nds32/nds32.h (MAX_REGS_PER_ADDRESS): Fix the value.
4687
4688 2018-03-18  Chung-Ju Wu  <jasonwucj@gmail.com>
4689
4690         * config/nds32/nds32.h (LOGICAL_OP_NON_SHORT_CIRCUIT): Define.
4691
4692 2018-03-18  Chung-Ju Wu  <jasonwucj@gmail.com>
4693
4694         * config/nds32/nds32.h (CLZ_DEFINED_VALUE_AT_ZERO): Define.
4695
4696 2018-03-17  Chung-Ju Wu  <jasonwucj@gmail.com>
4697             Kito Cheng  <kito.cheng@gmail.com>
4698
4699         * config/nds32/nds32-protos.h (nds32_adjust_reg_alloc_order): Declare.
4700         * config/nds32/nds32.c (nds32_reg_alloc_order_for_speed): New array.
4701         (nds32_adjust_reg_alloc_order): New function.
4702         * config/nds32/nds32.h (ADJUST_REG_ALLOC_ORDER): Define.
4703
4704 2018-03-17  Kito Cheng  <kito.cheng@gmail.com>
4705
4706         * config/nds32/nds32.c (nds32_asm_output_mi_thunk,
4707         nds32_print_operand, nds32_print_operand_address): Use
4708         HOST_WIDE_INT_PRINT_DEC instead.
4709
4710 2018-03-17  Chung-Ju Wu  <jasonwucj@gmail.com>
4711
4712         * config/nds32/nds32.c (nds32_register_priority): Modify cost.
4713
4714 2018-03-17  Jakub Jelinek  <jakub@redhat.com>
4715
4716         PR target/84902
4717         * config/i386/i386.c (initial_ix86_tune_features,
4718         initial_ix86_arch_features): Use unsigned HOST_WIDE_INT rather than
4719         unsigned long long.
4720         (set_ix86_tune_features): Change ix86_tune_mask from unsigned int
4721         to unsigned HOST_WIDE_INT, initialize to HOST_WIDE_INT_1U << ix86_tune
4722         rather than 1u << ix86_tune.  Formatting fix.
4723         (ix86_option_override_internal): Change ix86_arch_mask from
4724         unsigned int to unsigned HOST_WIDE_INT, initialize to
4725         HOST_WIDE_INT_1U << ix86_arch rather than 1u << ix86_arch.
4726         (ix86_function_specific_restore): Likewise.
4727
4728 2018-03-16  Jakub Jelinek  <jakub@redhat.com>
4729
4730         PR target/84899
4731         * postreload.c (reload_combine_recognize_pattern): Perform
4732         INTVAL addition in unsigned HOST_WIDE_INT type to avoid UB and
4733         truncate_int_for_mode the result for the destination's mode.
4734
4735         PR c/84909
4736         * hsa-gen.c (mem_type_for_type): Fix comment typo.
4737         * tree-vect-loop-manip.c (vect_create_cond_for_niters_checks):
4738         Likewise.
4739         * gimple-ssa-warn-restrict.c (builtin_memref::set_base_and_offset):
4740         Likewise.
4741
4742 2018-03-16  Vladimir Makarov  <vmakarov@redhat.com>
4743
4744         PR target/84876
4745         * lra-assigns.c (lra_split_hard_reg_for): Don't use
4746         regno_allocno_class_array and sorted_pseudos.
4747         * lra-constraints.c (spill_hard_reg_in_range): Ignore hard regs in
4748         insns where regno is used.
4749
4750 2018-03-16  Martin Liska  <mliska@suse.cz>
4751
4752         PR ipa/84833
4753         * multiple_target.c (create_dispatcher_calls): Redirect
4754         reference in the symbol table.
4755
4756 2018-03-16  Martin Liska  <mliska@suse.cz>
4757
4758         PR ipa/84722
4759         * multiple_target.c (create_dispatcher_calls): Redirect also
4760         an alias.
4761
4762 2018-03-16  Jakub Jelinek  <jakub@redhat.com>
4763
4764         PR c++/79937
4765         PR c++/82410
4766         * tree.h (TARGET_EXPR_NO_ELIDE): Define.
4767         * gimplify.c (gimplify_modify_expr_rhs): Don't elide TARGET_EXPRs with
4768         TARGET_EXPR_NO_ELIDE flag set unless *expr_p is INIT_EXPR.
4769
4770 2018-03-16  Julia Koval  <julia.koval@intel.com>
4771
4772         * doc/invoke.texi (Skylake Server): Add CLWB.
4773         Cannonlake): Remove CLWB.
4774
4775 2018-03-16  Jakub Jelinek  <jakub@redhat.com>
4776
4777         PR tree-optimization/84841
4778         * tree-ssa-reassoc.c (INTEGER_CONST_TYPE): Change to 1 << 4 from
4779         1 << 3.
4780         (FLOAT_ONE_CONST_TYPE): Define.
4781         (constant_type): Return FLOAT_ONE_CONST_TYPE for -1.0 and 1.0.
4782         (sort_by_operand_rank): Put entries with higher constant_type last
4783         rather than first to match comments.
4784
4785 2018-03-15  Sandra Loosemore  <sandra@codesourcery.com>
4786
4787         * config/nios2/nios2.md (movsi_internal): Fix thinko in
4788         split predicate.
4789
4790 2018-03-15  Jakub Jelinek  <jakub@redhat.com>
4791
4792         PR c++/79085
4793         * calls.c (expand_call): For TREE_ADDRESSABLE rettype ignore alignment
4794         check and use address of target always.
4795
4796 2018-03-15  H.J. Lu  <hongjiu.lu@intel.com>
4797
4798         PR target/84574
4799         * config/i386/i386.c (indirect_thunk_needed): Update comments.
4800         (indirect_thunk_bnd_needed): Likewise.
4801         (indirect_thunks_used): Likewise.
4802         (indirect_thunks_bnd_used): Likewise.
4803         (indirect_return_needed): New.
4804         (indirect_return_bnd_needed): Likewise.
4805         (output_indirect_thunk_function): Add a bool argument for
4806         function return.
4807         (output_indirect_thunk_function): Don't generate alias for
4808         function return thunk.
4809         (ix86_code_end): Call output_indirect_thunk_function to generate
4810         function return thunks.
4811         (ix86_output_function_return): Set indirect_return_bnd_needed
4812         and indirect_return_needed instead of indirect_thunk_bnd_needed
4813         and indirect_thunk_needed.
4814
4815 2018-03-15  Olga Makhotina  <olga.makhotina@intel.com>
4816
4817         * config/i386/sgxintrin.h (_enclv_u32): New intrinsic.
4818         (__enclv_bc, __enclv_cd, __enclv_generic): New definitions.
4819         (ERDINFO, ETRACKC, ELDBC, ELDUC): New leaves.
4820
4821 2018-03-15  David Malcolm  <dmalcolm@redhat.com>
4822             Paul Hua <paul.hua.gm@gmail.com>
4823
4824         PR c/84852
4825         * gcc.dg/fixits-pr84852-1.c: Fix filename in dg-regexp.
4826
4827 2018-03-15  Segher Boessenkool  <segher@kernel.crashing.org>
4828
4829         * config/rs6000/rs6000.c (abi_v4_pass_in_fpr): Add back the
4830         TARGET_DOUBLE_FLOAT and TARGET_SINGLE_FLOAT conditions on the DFmode
4831         resp. SFmode cases.
4832
4833 2018-03-15  Tamar Christina  <tamar.christina@arm.com>
4834
4835         PR target/84711
4836         * config/arm/arm.c (arm_can_change_mode_class): Use GET_MODE_UNIT_SIZE
4837         instead of GET_MODE_SIZE when comparing Units.
4838
4839 2018-03-15  Vladimir Mezentsev  <vladimir.mezentsev@oracle.com>
4840
4841         PR target/68256
4842         * varasm.c (hash_section): Return an unchangeble hash value
4843         * config/aarch64/aarch64.c (aarch64_use_blocks_for_constant_p):
4844         Return !aarch64_can_use_per_function_literal_pools_p ().
4845
4846 2018-03-15  Jakub Jelinek  <jakub@redhat.com>
4847
4848         PR target/84860
4849         * optabs.c (emit_conditional_move): Pass address of cmode's copy
4850         rather than address of cmode as last argument to prepare_cmp_insn.
4851
4852 2018-03-15  Julia Koval  <julia.koval@intel.com>
4853
4854         * config/i386/i386.c (F_AVX512VBMI2, F_GFNI, F_VPCLMULQDQ,
4855         F_AVX512VNNI, F_AVX512BITALG): New.
4856
4857 2018-03-14  John David Anglin  <danglin@gcc.gnu.org>
4858
4859         PR target/83451
4860         * config/pa/pa.c (pa_emit_move_sequence):  Always emit secondary reload
4861         insn for floating-point loads and stores.
4862
4863 2018-03-14  Carl Love  <cel@us.ibm.com>
4864
4865         * config/rs6000/rs6000-c.c: Add macro definitions for
4866         ALTIVEC_BUILTIN_VEC_PERMXOR.
4867         * config/rs6000/rs6000.h: Add #define for vec_permxor builtin.
4868         * config/rs6000/rs6000-builtin.def: Add macro expansions for VPERMXOR.
4869         * config/rs6000/altivec.md (altivec_vpermxor): New define expand.
4870         * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Add case
4871         UNSPEC_VPERMXOR.
4872         * config/doc/extend.texi: Add prototypes for vec_permxor.
4873
4874 2018-03-14  David Malcolm  <dmalcolm@redhat.com>
4875
4876         PR c/84852
4877         * diagnostic-show-locus.c (class layout_point): Convert m_line
4878         from int to linenum_type.
4879         (line_span::comparator): Use linenum "compare" function when
4880         comparing line numbers.
4881         (test_line_span): New function.
4882         (layout_range::contains_point): Convert param "row" from int to
4883         linenum_type.
4884         (layout_range::intersects_line_p): Likewise.
4885         (layout::will_show_line_p): Likewise.
4886         (layout::print_source_line): Likewise.
4887         (layout::should_print_annotation_line_p): Likewise.
4888         (layout::print_annotation_line): Likewise.
4889         (layout::print_leading_fixits): Likewise.
4890         (layout::annotation_line_showed_range_p): Likewise.
4891         (struct line_corrections): Likewise for field m_row.
4892         (line_corrections::line_corrections): Likewise for param "row".
4893         (layout::print_trailing_fixits): Likewise.
4894         (layout::get_state_at_point): Likewise.
4895         (layout::get_x_bound_for_row): Likewise.
4896         (layout::print_line): Likewise.
4897         (diagnostic_show_locus): Likewise for locals "last_line" and
4898         "row".
4899         (selftest::diagnostic_show_locus_c_tests): Call test_line_span.
4900         * input.c (selftest::test_linenum_comparisons): New function.
4901         (selftest::input_c_tests): Call it.
4902         * selftest.c (selftest::test_assertions): Test ASSERT_GT,
4903         ASSERT_GT_AT, ASSERT_LT, and ASSERT_LT_AT.
4904         * selftest.h (ASSERT_GT): New macro.
4905         (ASSERT_GT_AT): New macro.
4906         (ASSERT_LT): New macro.
4907         (ASSERT_LT_AT): New macro.
4908
4909 2018-03-14  Segher Boessenkool  <segher@kernel.crashing.org>
4910
4911         PR rtl-optimization/84780
4912         * combine.c (distribute_links): Don't make a link based on pc_rtx.
4913
4914 2018-03-14  Martin Liska  <mliska@suse.cz>
4915
4916         * tree.c (record_node_allocation_statistics): Use
4917         get_stats_node_kind.
4918         (get_stats_node_kind): New function extracted from
4919         record_node_allocation_statistics.
4920         (free_node): Use get_stats_node_kind.
4921
4922 2018-03-14  Richard Biener  <rguenther@suse.de>
4923
4924         * tree-ssa-pre.c (compute_antic_aux): Remove code that asserts
4925         that the value-set of ANTIC_IN doesn't grow.
4926
4927         Revert
4928         * tree-ssa-pre.c (struct bb_bitmap_sets): Add visited_with_visited_succs
4929         member.
4930         (BB_VISITED_WITH_VISITED_SUCCS): New define.
4931         (compute_antic): Initialize BB_VISITED_WITH_VISITED_SUCCS.
4932
4933 2018-03-14  Julia Koval  <julia.koval@intel.com>
4934
4935         * config.gcc (icelake-client, icelake-server): New.
4936         (icelake): Remove.
4937         * config/i386/i386.c (initial_ix86_tune_features): Extend to 64 bit.
4938         (initial_ix86_arch_features): Ditto.
4939         (PTA_SKYLAKE): Add SGX.
4940         (PTA_ICELAKE): Remove.
4941         (PTA_ICELAKE_CLIENT): New.
4942         (PTA_ICELAKE_SERVER): New.
4943         (ix86_option_override_internal): Split up icelake on icelake client and
4944         icelake server.
4945         (get_builtin_code_for_version): Ditto.
4946         (fold_builtin_cpu): Ditto.
4947         * config/i386/driver-i386.c (config/i386/driver-i386.c): Ditto.
4948         * config/i386/i386-c.c (ix86_target_macros_internal): Ditto
4949         * config/i386/i386.h (processor_type): Ditto.
4950         * doc/invoke.texi: Ditto.
4951
4952 2018-03-14  Jakub Jelinek  <jakub@redhat.com>
4953
4954         PR sanitizer/83392
4955         * sanopt.c (maybe_optimize_ubsan_ptr_ifn): Handle also
4956         INTEGER_CST offset, add it together with bitpos / 8 and
4957         sign extend based on POINTER_SIZE.
4958
4959         PR target/84844
4960         Revert
4961         2017-04-20  Uros Bizjak  <ubizjak@gmail.com>
4962
4963         PR target/78090
4964         * config/i386/constraints.md (Yc): New register constraint.
4965         * config/i386/i386.md (*float<SWI48:mode><MODEF:mode>2_mixed):
4966         Use Yc constraint for alternative 2 of operand 0.  Remove
4967         preferred_for_speed attribute.
4968
4969 2018-03-14  Richard Biener  <rguenther@suse.de>
4970
4971         PR tree-optimization/84830
4972         * tree-ssa-pre.c (compute_antic_aux): Intersect the new ANTIC_IN
4973         with the old one to avoid oscillations.
4974
4975 2018-03-13  Vladimir Makarov  <vmakarov@redhat.com>
4976
4977         PR target/83712
4978         * lra-assigns.c (find_all_spills_for): Ignore uninteresting
4979         pseudos.
4980         (assign_by_spills): Return a flag of reload assignment failure.
4981         Do not process the reload assignment failures.  Do not spill other
4982         reload pseudos if they has the same reg class.  Update n if
4983         necessary.
4984         (lra_assign): Add a return arg.  Set up from the result of
4985         assign_by_spills call.
4986         (find_reload_regno_insns, lra_split_hard_reg_for): New functions.
4987         * lra-constraints.c (split_reg): Add a new arg.  Use it instead of
4988         usage_insns if it is not NULL.
4989         (spill_hard_reg_in_range): New function.
4990         (split_if_necessary, inherit_in_ebb): Pass a new arg to split_reg.
4991         * lra-int.h (spill_hard_reg_in_range, lra_split_hard_reg_for): New
4992         function prototypes.
4993         (lra_assign): Change prototype.
4994         * lra.c (lra): Add code to deal with fails by splitting hard reg
4995         live ranges.
4996
4997 2018-03-01  Palmer Dabbelt  <palmer@sifive.com>
4998
4999         * config/riscv/riscv.opt (mrelax): New option.
5000         * config/riscv/riscv.c (riscv_file_start): Emit ".option
5001         "norelax" when riscv_mrelax is disabled.
5002         * doc/invoke.texi (RISC-V): Document "-mrelax" and "-mno-relax".
5003
5004 2018-03-13  Aaron Sawdey  <acsawdey@linux.vnet.ibm.com>
5005
5006         PR target/84743
5007         * config/rs6000/rs6000.c (rs6000_reassociation_width): Disable parallel
5008         reassociation for int modes.
5009
5010 2018-03-13  Richard Sandiford  <richard.sandiford@linaro.org>
5011
5012         * tree-vect-loop-manip.c (vect_maybe_permute_loop_masks):
5013         Reverse the choice between VEC_UNPACK_LO_EXPR and VEC_UNPACK_HI_EXPR
5014         for big-endian.
5015         * config/aarch64/iterators.md (hi_lanes_optab): New int attribute.
5016         * config/aarch64/aarch64-sve.md
5017         (*aarch64_sve_<perm_insn><perm_hilo><mode>): Rename to...
5018         (aarch64_sve_<perm_insn><perm_hilo><mode>): ...this.
5019         (*extend<mode><Vwide>2): Rename to...
5020         (aarch64_sve_extend<mode><Vwide>2): ...this.
5021         (vec_unpack<su>_<perm_hilo>_<mode>): Turn into a define_expand,
5022         renaming the old pattern to...
5023         (aarch64_sve_punpk<perm_hilo>_<mode>): ...this.  Only define
5024         unsigned packs.
5025         (vec_unpack<su>_<perm_hilo>_<SVE_BHSI:mode>): Turn into a
5026         define_expand, renaming the old pattern to...
5027         (aarch64_sve_<su>unpk<perm_hilo>_<SVE_BHSI:mode>): ...this.
5028         (*vec_unpacku_<perm_hilo>_<mode>_no_convert): Delete.
5029         (vec_unpacks_<perm_hilo>_<mode>): Take BYTES_BIG_ENDIAN into
5030         account when deciding which SVE instruction the optab should use.
5031         (vec_unpack<su_optab>_float_<perm_hilo>_vnx4si): Likewise.
5032
5033 2018-03-13  Richard Sandiford  <richard.sandiford@linaro.org>
5034
5035         * config/aarch64/aarch64.md (V4_REGNUM, V8_REGNUM, V12_REGNUM)
5036         (V20_REGNUM, V24_REGNUM, V28_REGNUM, P1_REGNUM, P2_REGNUM, P3_REGNUM)
5037         (P4_REGNUM, P5_REGNUM, P6_REGNUM, P8_REGNUM, P9_REGNUM, P10_REGNUM)
5038         (P11_REGNUM, P12_REGNUM, P13_REGNUM, P14_REGNUM): New define_constants.
5039         (tlsdesc_small_<mode>): Turn a define_expand and use
5040         tlsdesc_small_sve_<mode> for SVE.  Rename original define_insn to...
5041         (tlsdesc_small_advsimd_<mode>): ...this.
5042         (tlsdesc_small_sve_<mode>): New pattern.
5043
5044 2018-03-13  Richard Sandiford  <richard.sandiford@linaro.org>
5045
5046         * config/aarch64/iterators.md (UNSPEC_SMUL_HIGHPART)
5047         (UNSPEC_UMUL_HIGHPART): New constants.
5048         (MUL_HIGHPART): New int iteraor.
5049         (su): Handle UNSPEC_SMUL_HIGHPART and UNSPEC_UMUL_HIGHPART.
5050         * config/aarch64/aarch64-sve.md (<su>mul<mode>3_highpart): New
5051         define_expand.
5052         (*<su>mul<mode>3_highpart): New define_insn.
5053
5054 2018-03-13  Eric Botcazou  <ebotcazou@adacore.com>
5055
5056         PR lto/84805
5057         * ipa-devirt.c (odr_subtypes_equivalent_p): Do not get the ODR type of
5058         incomplete types.
5059
5060 2018-03-13  Martin Liska  <mliska@suse.cz>
5061
5062         PR ipa/84658.
5063         * (sem_item_optimizer::sem_item_optimizer): Initialize new
5064         vector.
5065         (sem_item_optimizer::~sem_item_optimizer): Release it.
5066         (sem_item_optimizer::merge_classes): Register variable aliases.
5067         (sem_item_optimizer::fixup_pt_set): New function.
5068         (sem_item_optimizer::fixup_points_to_sets): Likewise.
5069         * ipa-icf.h: Declare new variables and functions.
5070
5071 2018-03-13  Jakub Jelinek  <jakub@redhat.com>
5072
5073         PR middle-end/84834
5074         * match.pd ((A & C) != 0 ? D : 0): Use INTEGER_CST@2 instead of
5075         integer_pow2p@2 and test integer_pow2p in condition.
5076         (A < 0 ? C : 0): Similarly for @1.
5077
5078         PR middle-end/84831
5079         * stmt.c (parse_output_constraint): If the CONSTRAINT_LEN (*p, p)
5080         characters starting at p contain '\0' character, don't look beyond
5081         that.
5082
5083         PR target/84827
5084         * config/i386/i386.md (round<mode>2): For 387 fancy math, disable
5085         pattern if -ftrapping-math -fno-fp-int-builtin-inexact.
5086
5087         PR target/84828
5088         * reg-stack.c (change_stack): Change update_end var from int to
5089         rtx_insn *, if non-NULL don't update just BB_END (current_block), but
5090         also call set_block_for_insn on the newly added insns and rescan.
5091
5092         PR target/84786
5093         * config/i386/sse.md (sse2_loadhpd): Use Yv constraint rather than v
5094         on the last operand.
5095
5096         PR c++/84704
5097         * tree.c (stabilize_reference_1): Return save_expr (e) for
5098         STATEMENT_LIST even if it doesn't have side-effects.
5099
5100 2018-03-12  Jonathan Wakely  <jwakely@redhat.com>
5101
5102         * doc/invoke.texi (-mclflushopt): Fix spelling of option.
5103
5104 2018-03-12  Renlin Li  <renlin.li@arm.com>
5105
5106         * config/aarch64/aarch64.md (movhf_aarch64): Fix mode argument to
5107         aarch64_output_scalar_simd_mov_immediate.
5108
5109 2018-03-12  Martin Sebor  <msebor@redhat.com>
5110
5111         PR tree-optimization/83456
5112         * gimple-fold.c (gimple_fold_builtin_memory_op): Avoid warning
5113         for perfectly overlapping calls to memcpy.
5114         (gimple_fold_builtin_memory_chk): Same.
5115         (gimple_fold_builtin_strcpy): Handle no-warning.
5116         (gimple_fold_builtin_stxcpy_chk): Same.
5117         * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Handle no-warning.
5118
5119 2018-03-12  Segher Boessenkool  <segher@kernel.crashing.org>
5120
5121         * config/rs6000/rs6000.c (abi_v4_pass_in_fpr): Add bool "named"
5122         parameter.  Use it for SFmode.
5123         (rs6000_function_arg_advance_1): Adjust.
5124         (rs6000_function_arg): Adjust.
5125         (rs6000_gimplify_va_arg): Pass false for that new parameter.
5126
5127 2018-03-12  Segher Boessenkool  <segher@kernel.crashing.org>
5128
5129         PR rtl-optimization/84169
5130         PR rtl-optimization/84780
5131         * combine.c (can_combine_p): Check for a 2-insn combination whether
5132         the destination register is used between the two insns, too.
5133
5134 2018-03-12  Richard Biener  <rguenther@suse.de>
5135
5136         PR tree-optimization/84803
5137         * tree-if-conv.c (ifcvt_memrefs_wont_trap): Don't do anything
5138         for refs DR analysis didn't process.
5139
5140 2018-03-12  Richard Biener  <rguenther@suse.de>
5141
5142         PR tree-optimization/84777
5143         * tree-ssa-loop-ch.c (should_duplicate_loop_header_p): For
5144         force-vectorize loops ignore whether we are optimizing for size.
5145
5146 2018-03-12  Chung-Ju Wu  <jasonwucj@gmail.com>
5147
5148         * config/nds32/nds32.c (nds32_md_asm_adjust): New function.
5149         (TARGET_MD_ASM_ADJUST): Define.
5150
5151 2018-03-12  Monk Chiang  <sh.chiang04@gmail.com>
5152             Kito Cheng  <kito.cheng@gmail.com>
5153             Chung-Ju Wu  <jasonwucj@gmail.com>
5154
5155         * config/nds32/nds32.c (nds32_compute_stack_frame,
5156         nds32_emit_stack_push_multiple, nds32_emit_stack_pop_multiple,
5157         nds32_emit_stack_v3push, nds32_emit_stack_v3pop,
5158         nds32_emit_adjust_frame, nds32_expand_prologue, nds32_expand_epilogue,
5159         nds32_expand_prologue_v3push, nds32_expand_epilogue_v3pop): Refine.
5160         * config/nds32/nds32.h (NDS32_FIRST_CALLEE_SAVE_GPR_REGNUM,
5161         NDS32_LAST_CALLEE_SAVE_GPR_REGNUM, NDS32_V3PUSH_AVAILABLE_P): New.
5162         * config/nds32/nds32.md (prologue, epilogue): Use macro
5163         NDS32_V3PUSH_AVAILABLE_P to do checking.
5164
5165 2018-03-11  Jakub Jelinek  <jakub@redhat.com>
5166
5167         PR debug/58150
5168         * dwarf2out.c (gen_enumeration_type_die): Don't guard adding
5169         DW_AT_declaration for ENUM_IS_OPAQUE on -gdwarf-4 or -gno-strict-dwarf,
5170         but on TYPE_SIZE.  Don't do anything for ENUM_IS_OPAQUE if not creating
5171         a new die.  Don't set TREE_ASM_WRITTEN if ENUM_IS_OPAQUE.  Guard
5172         addition of most attributes on !orig_type_die or the attribute not
5173         being present already.  Assert TYPE_VALUES is NULL for ENUM_IS_OPAQUE.
5174
5175 2018-03-11  Kito Cheng  <kito.cheng@gmail.com>
5176             Chung-Ju Wu  <jasonwucj@gmail.com>
5177
5178         * config/nds32/nds32.c (nds32_cpu_cpp_builtins): Modify to define
5179         __NDS32_VH__ macro.
5180         * config/nds32/nds32.opt (mvh): New option.
5181
5182 2018-03-11  Kito Cheng  <kito.cheng@gmail.com>
5183             Chung-Ju Wu  <jasonwucj@gmail.com>
5184
5185         * config/nds32/nds32-protos.h (nds32_cpu_cpp_builtins): Declare
5186         function.
5187         * config/nds32/nds32.c (nds32_cpu_cpp_builtins): New function.
5188         * config/nds32/nds32.h (TARGET_CPU_CPP_BUILTINS): Modify its
5189         definition.
5190
5191 2018-03-11  Kito Cheng  <kito.cheng@gmail.com>
5192             Chung-Ju Wu  <jasonwucj@gmail.com>
5193
5194         * config/nds32/nds32-memory-manipulation.c (nds32_expand_strlen): New
5195         function.
5196         * config/nds32/nds32-multiple.md (strlensi): New pattern.
5197         * config/nds32/nds32-protos.h (nds32_expand_strlen): Declare function.
5198
5199 2018-03-11  Monk Chiang  <sh.chiang04@gmail.com>
5200             Kito Cheng  <kito.cheng@gmail.com>
5201             Chung-Ju Wu  <jasonwucj@gmail.com>
5202
5203         * config/nds32/constants.md (unspec_element): Add UNSPEC_FFB,
5204         UNSPEC_FFMISM and UNSPEC_FLMISM.
5205         * config/nds32/nds32-intrinsic.c (bdesc_2arg): Add builtin description
5206         for ffb, ffmism and flmism.
5207         * config/nds32/nds32-intrinsic.md (unspec_ffb): Define new pattern.
5208         (unspec_ffmism): Ditto.
5209         (unspec_flmism): Ditto.
5210         (nds32_expand_builtin_impl): Check if string extension is available.
5211         * config/nds32/nds32.h (nds32_builtins): Add NDS32_BUILTIN_FFB,
5212         NDS32_BUILTIN_FFMISM and NDS32_BUILTIN_FLMISM.
5213
5214 2018-03-10  Vladimir Makarov  <vmakarov@redhat.com>
5215
5216         Reverting patch:
5217         2018-03-09  Vladimir Makarov  <vmakarov@redhat.com>
5218
5219         PR target/83712
5220         * lra-assigns.c (assign_by_spills): Return a flag of reload
5221         assignment failure.  Do not process the reload assignment
5222         failures.  Do not spill other reload pseudos if they has the same
5223         reg class.
5224         (lra_assign): Add a return arg.  Set up from the result of
5225         assign_by_spills call.
5226         (find_reload_regno_insns, lra_split_hard_reg_for): New functions.
5227         * lra-constraints.c (split_reg): Add a new arg.  Use it instead of
5228         usage_insns if it is not NULL.
5229         (spill_hard_reg_in_range): New function.
5230         (split_if_necessary, inherit_in_ebb): Pass a new arg to split_reg.
5231         * lra-int.h (spill_hard_reg_in_range, lra_split_hard_reg_for): New
5232         function prototypes.
5233         (lra_assign): Change prototype.
5234         * lra.c (lra): Add code to deal with fails by splitting hard reg
5235         live ranges.
5236
5237 2018-03-10  H.J. Lu  <hongjiu.lu@intel.com>
5238
5239         PR target/84807
5240         * config/i386/i386.opt: Replace Enforcment with Enforcement.
5241
5242 2018-03-10  Alexandre Oliva  <aoliva@redhat.com>
5243
5244         PR debug/84620
5245         * dwarf2out.h (dw_val_class): Add dw_val_class_symview.
5246         (dw_val_node): Add val_symbolic_view.
5247         * dwarf2out.c (dw_line_info_table): Add symviews_since_reset.
5248         (symview_upper_bound): New.
5249         (new_line_info_table): Initialize symviews_since_reset.
5250         (dwarf2out_source_line): Count symviews_since_reset and set
5251         symview_upper_bound.
5252         (dw_val_equal_p): Handle symview.
5253         (add_AT_symview): New.
5254         (print_dw_val): Handle symview.
5255         (attr_checksum, attr_checksum_ordered): Likewise.
5256         (same_dw_val_p, size_of_die): Likewise.
5257         (value_format, output_die): Likewise.
5258         (add_high_low_attributes): Use add_AT_symview for entry_view.
5259         (dwarf2out_finish): Reset symview_upper_bound, clear
5260         zero_view_p.
5261
5262 2018-03-09  Peter Bergner  <bergner@vnet.ibm.com>
5263
5264         PR target/83969
5265         * config/rs6000/rs6000.c (rs6000_offsettable_memref_p): New prototype.
5266         Add strict argument and use it.
5267         (rs6000_split_multireg_move): Update for new strict argument.
5268         (mem_operand_gpr): Disallow all non-offsettable addresses.
5269         * config/rs6000/rs6000.md (*movdi_internal64): Use YZ constraint.
5270
5271 2018-03-09  Jakub Jelinek  <jakub@redhat.com>
5272
5273         PR target/84772
5274         * config/rs6000/rs6000.c (rs6000_gimplify_va_arg): Mark va_arg_tmp
5275         temporary TREE_ADDRESSABLE before gimplification of BUILT_IN_MEMCPY.
5276         * config/powerpcspe/powerpcspe.c (rs6000_gimplify_va_arg): Likewise.
5277
5278         PR c++/84767
5279         * tree-inline.c (copy_tree_body_r): For INDIRECT_REF of a remapped
5280         decl, use remap_type if we want to use the type.
5281
5282 2018-03-09  Martin Sebor  <msebor@redhat.com>
5283
5284         PR tree-optimization/84526
5285         * gimple-ssa-warn-restrict.c (builtin_memref::set_base_and_offset):
5286         Remove dead code.
5287         (builtin_access::generic_overlap): Be prepared to handle non-array
5288         base objects.
5289
5290 2018-03-09  Alexandre Oliva  <aoliva@redhat.com>
5291
5292         PR rtl-optimization/84682
5293         * lra-constraints.c (process_address_1): Check is_address flag
5294         for address constraints.
5295         (process_alt_operands): Likewise.
5296         * lra.c (lra_set_insn_recog_data): Pass asm operand locs to
5297         preprocess_constraints.
5298         * recog.h (preprocess_constraints): Add oploc parameter.
5299         Adjust callers.
5300         * recog.c (preprocess_constraints): Test address_operand for
5301         CT_ADDRESS constraints.
5302
5303 2018-03-09  Vladimir Makarov  <vmakarov@redhat.com>
5304
5305         PR target/83712
5306         * lra-assigns.c (assign_by_spills): Return a flag of reload
5307         assignment failure.  Do not process the reload assignment
5308         failures.  Do not spill other reload pseudos if they has the same
5309         reg class.
5310         (lra_assign): Add a return arg.  Set up from the result of
5311         assign_by_spills call.
5312         (find_reload_regno_insns, lra_split_hard_reg_for): New functions.
5313         * lra-constraints.c (split_reg): Add a new arg.  Use it instead of
5314         usage_insns if it is not NULL.
5315         (spill_hard_reg_in_range): New function.
5316         (split_if_necessary, inherit_in_ebb): Pass a new arg to split_reg.
5317         * lra-int.h (spill_hard_reg_in_range, lra_split_hard_reg_for): New
5318         function prototypes.
5319         (lra_assign): Change prototype.
5320         * lra.c (lra): Add code to deal with fails by splitting hard reg
5321         live ranges.
5322
5323 2018-03-09  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
5324
5325         PR target/83193
5326         * common/config/arm/arm-common.c (arm_parse_arch_option_name):
5327         Accept complain bool parameter.  Only emit errors if it is true.
5328         (arm_parse_cpu_option_name): Likewise.
5329         (arm_target_thumb_only): Adjust callers of the above.
5330         * config/arm/arm-protos.h (arm_parse_cpu_option_name): Adjust
5331         prototype to take a default true bool parameter.
5332         (arm_parse_arch_option_name): Likewise.
5333
5334 2018-03-09  David Malcolm  <dmalcolm@redhat.com>
5335             Francois-Xavier Coudert  <fxcoudert@gcc.gnu.org>
5336
5337         PR jit/64089
5338         PR jit/84288
5339         * Makefile.in (LD_VERSION_SCRIPT_OPTION, LD_SONAME_OPTION): New.
5340         * configure: Regenerate.
5341         * configure.ac ("linker --version-script option"): New.
5342         ("linker soname option"): New.
5343
5344 2018-03-09  Richard Biener  <rguenther@suse.de>
5345
5346         PR tree-optimization/84775
5347         * tree-if-conv.c (add_bb_predicate_gimplified_stmts): Delink
5348         immediate uses of predicate stmts and mark them modified.
5349
5350         Revert
5351         PR tree-optimization/84178
5352         * tree-if-conv.c (combine_blocks): Move insert_gimplified_predicates
5353         to caller.
5354         (version_loop_for_if_conversion): Delay update_ssa call.
5355         (tree_if_conversion): Delay update_ssa until after predicate
5356         insertion.
5357
5358 2018-03-09  Eric Botcazou  <ebotcazou@adacore.com>
5359
5360         PR target/84763
5361         * config/i386/winnt.c (i386_pe_seh_cold_init): Use small pre-allocation
5362         when the function accesses prior frames.
5363
5364 2018-03-08  Jakub Jelinek  <jakub@redhat.com>
5365
5366         PR debug/84456
5367         * dwarf2out.c (dw_loc_list): If list && loc_list->first->next, call
5368         gen_llsym, otherwise call maybe_gen_llsym.
5369
5370         PR inline-asm/84742
5371         * recog.c (asm_operand_ok): Return 0 if multi-character constraint
5372         has ',' character inside of it.
5373
5374 2018-03-08  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
5375
5376         PR target/84748
5377         * config/aarch64/aarch64.md (*compare_cstore<mode>_insn): Mark pattern
5378         as clobbering CC_REGNUM.
5379
5380 2018-03-08  Richard Biener  <rguenther@suse.de>
5381
5382         PR middle-end/84552
5383         * tree-scalar-evolution.c: Include tree-into-ssa.h.
5384         (follow_copies_to_constant): Do not follow SSA names registered
5385         for update.
5386
5387 2018-03-08  Richard Biener  <rguenther@suse.de>
5388
5389         PR tree-optimization/84178
5390         * tree-if-conv.c (combine_blocks): Move insert_gimplified_predicates
5391         to caller.
5392         (version_loop_for_if_conversion): Delay update_ssa call.
5393         (tree_if_conversion): Delay update_ssa until after predicate
5394         insertion.
5395
5396 2018-03-08  David Malcolm  <dmalcolm@redhat.com>
5397
5398         PR tree-optimization/84178
5399         * tree-if-conv.c (release_bb_predicate): Remove the
5400         the assertion that the stmts have NULL use_ops.
5401         Discard the statements, asserting that they haven't
5402         yet been added to a BB.
5403
5404 2018-03-08  Richard Biener  <rguenther@suse.de>
5405
5406         PR tree-optimization/84746
5407         * tree-ssa-pre.c (find_leader_in_sets): Deal with SET1 being NULL.
5408         (phi_translate): Pass in destination ANTIC_OUT set.
5409         (phi_translate_1): Likewise.  For a simplified result lookup
5410         a leader in ANTIC_OUT and AVAIL_OUT, not the ANTIC_IN sets.
5411         (phi_translate_set): Adjust.
5412         (do_pre_regular_insertion): Likewise.
5413         (do_pre_partial_partial_insertion): Likewise.
5414
5415 2018-03-08  Martin Liska  <mliska@suse.cz>
5416
5417         PR gcov-profile/84735
5418         * doc/gcov.texi: Document usage of profile files.
5419         * gcov-io.h: Document changes in the format.
5420
5421 2018-03-08  Alexandre Oliva  <aoliva@redhat.com>
5422
5423         PR debug/84404
5424         PR debug/84408
5425         * dwarf2out.c (struct dw_line_info_table): Update comments for
5426         view == -1.
5427         (FORCE_RESET_NEXT_VIEW): New.
5428         (FORCE_RESETTING_VIEW_P): New.
5429         (RESETTING_VIEW_P): Check for -1 too.
5430         (ZERO_VIEW_P): Likewise.
5431         (new_line_info_table): Force-reset next view.
5432         (dwarf2out_begin_function): Likewise.
5433         (dwarf2out_source_line): Simplify zero_view_p initialization.
5434         Test FORCE_RESETTING_VIEW_P and RESETTING_VIEW_P instead of
5435         view directly.  Omit view when omitting .loc at line 0.
5436
5437 2018-03-08  Jakub Jelinek  <jakub@redhat.com>
5438
5439         PR tree-optimization/84740
5440         * tree-switch-conversion.c (process_switch): Call build_constructors
5441         only if info.phi_count is non-zero.
5442
5443         PR tree-optimization/84739
5444         * tree-tailcall.c (find_tail_calls): Check call arguments against
5445         DECL_ARGUMENTS (current_function_decl) rather than
5446         DECL_ARGUMENTS (func) when checking for tail recursion.
5447
5448 2018-03-07  Jakub Jelinek  <jakub@redhat.com>
5449
5450         * doc/contrib.texi: Add entries for Martin Liska, David Malcolm,
5451         Marek Polacek, extend Vladimir Makarov's, Jonathan Wakely's and
5452         Volker Reichelt's entry and add entries for people that perform
5453         GCC fuzzy testing and report numerous bugs.
5454
5455 2018-03-07  Segher Boessenkool  <segher@kernel.crashing.org>
5456
5457         PR target/82411
5458         * config/rs6000/rs6000.c (rs6000_elf_in_small_data_p): Don't put
5459         readonly data in sdata, if that is disabled.
5460         * config/rs6000/sysv4.opt (mreadonly-in-sdata): New option.
5461         * doc/invoke.texi (RS/6000 and PowerPC Options): Document
5462         -mreadonly-in-sdata option.
5463
5464 2018-03-07  Martin Sebor  <msebor@redhat.com>
5465
5466         PR tree-optimization/84468
5467         * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Consider successor
5468         basic block when looking for nul assignment.
5469
5470 2018-03-07  Eric Botcazou  <ebotcazou@adacore.com>
5471
5472         PR target/84277
5473         * except.h (output_function_exception_table): Adjust prototype.
5474         * except.c (output_function_exception_table): Remove FNNAME parameter
5475         and add SECTION parameter.  Ouput one part of the table at a time.
5476         * final.c (final_scan_insn_1) <NOTE_INSN_SWITCH_TEXT_SECTIONS>: Output
5477         the first part of the exception table and emit unwind directives.
5478         * config/i386/i386-protos.h (i386_pe_end_cold_function): Declare.
5479         (i386_pe_seh_cold_init): Likewise.
5480         * config/i386/cygming.h (ASM_DECLARE_COLD_FUNCTION_NAME): New macro.
5481         (ASM_DECLARE_COLD_FUNCTION_SIZE): Likewise.
5482         * config/i386/i386.c (x86_expand_epilogue): Fix wording in comment.
5483         (ix86_output_call_insn): Emit a nop in one more case for SEH.
5484         * config/i386/winnt.c: Include except.h.
5485         (struct seh_frame_state): Add reg_offset, after_prologue and
5486         in_cold_section fields.
5487         (i386_pe_seh_end_prologue): Set seh->after_prologue.
5488         (i386_pe_seh_cold_init): New function.
5489         (i386_pe_seh_fini): Add COLD parameter and bail out if it is not equal
5490         to seh->in_cold_section.
5491         (seh_emit_push): Record the offset of the push.
5492         (seh_emit_save): Record the offet of the save.
5493         (i386_pe_seh_unwind_emit): Deal with NOTE_INSN_SWITCH_TEXT_SECTIONS.
5494         Test seh->after_prologue to disregard the epilogue.
5495         (i386_pe_end_function): Pass FALSE to i386_pe_seh_fini.
5496         (i386_pe_end_cold_function): New function.
5497
5498 2018-03-07  Jakub Jelinek  <jakub@redhat.com>
5499
5500         PR fortran/84565
5501         * config/aarch64/predicates.md (aarch64_simd_reg_or_zero): Use
5502         aarch64_simd_or_scalar_imm_zero rather than aarch64_simd_imm_zero.
5503
5504         PR c++/84704
5505         * gimple-expr.c (create_tmp_var_raw): Set DECL_NAMELESS flag
5506         on tmp_var.
5507         * tree-pretty-print.c (dump_decl_name): For TDF_COMPARE_DEBUG,
5508         don't print names of DECL_NAMELESS DECL_IGNORED_P decls.
5509
5510         PR middle-end/84723
5511         * multiple_target.c: Include tree-inline.h and intl.h.
5512         (expand_target_clones): Diagnose and fail if node->definition and
5513         !tree_versionable_function_p (node->decl).
5514
5515 2018-03-06  John David Anglin  <danglin@gcc.gnu.org>
5516
5517         * config/pa/pa.h (ASM_GENERATE_INTERNAL_LABEL): Revise to use
5518         sprint_ul.
5519         (ASM_OUTPUT_ADDR_VEC_ELT): Revise for above change.
5520         (ASM_OUTPUT_ADDR_DIFF_ELT): Likewise.
5521         * config/pa/pa64-hpux.h (ASM_GENERATE_INTERNAL_LABEL): Revise as above.
5522
5523 2018-03-06  Jakub Jelinek  <jakub@redhat.com>
5524
5525         PR target/84710
5526         * combine.c (try_combine): Use reg_or_subregno instead of handling
5527         just paradoxical SUBREGs and REGs.
5528
5529 2018-03-06  Claudiu Zissulescu  <claziss@synopsys.com>
5530
5531          * config/arc/arc.c (arc_finalize_pic): Remove function.
5532          (arc_must_save_register): We use single base PIC register, remove
5533          checks to save/restore the PIC register.
5534          (arc_expand_prologue): Likewise.
5535          * config/arc/arc-protos.h (arc_set_default_type_attributes):
5536          Remove.
5537          (arc_verify_short): Likewise.
5538          (arc_attr_type): Likewise.
5539          * config/arc/arc.c (arc_set_default_type_attributes): Remove.
5540          (walk_stores): Likewise.
5541          (arc_address_cost): Make it static.
5542          (arc_verify_short): Likewise.
5543          (branch_dest): Likewise.
5544          (arc_attr_type): Likewise.
5545          * config/arc/arc.c (TARGET_ADJUST_INSN_LENGTH): Remove.
5546          (TARGET_INSN_LENGTH_PARAMETERS): Likewise.
5547          (arc_final_prescan_insn): Remove inserting the nops due to
5548          hardware hazards.  It is done in reorg step.
5549          (insn_length_variant_t): Remove.
5550          (insn_length_parameters_t): Likewise.
5551          (arc_insn_length_parameters): Likewise.
5552          (arc_get_insn_variants): Likewise.
5553          * config/arc/arc.h (TARGET_UPSIZE_DBR): Remove.
5554
5555 2018-03-06  Jakub Jelinek  <jakub@redhat.com>
5556
5557         PR inline-asm/84683
5558         * reg-stack.c (move_for_stack_reg): If any_malformed_asm, avoid
5559         assertion failure.
5560
5561         PR tree-optimization/84687
5562         * omp-simd-clone.c (simd_clone_create): Clear DECL_BUILT_IN_CLASS
5563         on new_node->decl.
5564         * match.pd (pow(C,x)*expN(y) -> expN(logN(C)*x+y)): New optimization.
5565
5566 2018-03-05  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
5567
5568         * config/rs6000/rs6000-builtin.def (rs6000_speculation_barrier):
5569         Rename to ppc_speculation_barrier.
5570         * config/rs6000/rs6000.c (rs6000_init_builtins): Rename builtin to
5571         __builtin_ppc_speculation_barrier.
5572
5573 2018-03-05  Jakub Jelinek  <jakub@redhat.com>
5574
5575         PR target/84700
5576         * combine.c (combine_simplify_rtx): Don't try to simplify if
5577         if_then_else_cond returned non-NULL, but either true_rtx or false_rtx
5578         are equal to x.
5579
5580 2018-03-05  Segher Boessenkool  <segher@kernel.crashing.org>
5581
5582         * config/rs6000/rs6000.c (rs6000_loop_align): Don't align tiny loops
5583         to 32 bytes when compiling for POWER9.
5584
5585 2018-03-05  Jakub Jelinek  <jakub@redhat.com>
5586
5587         PR target/84564
5588         * config/i386/i386.c (ix86_function_ok_for_sibcall): Check for
5589         regparm >= 3 with no arg reg available also for calls with
5590         flag_force_indirect_call.  Pass decl to ix86_function_regparm.
5591
5592         PR target/84524
5593         * config/i386/sse.md (*<code><mode>3): Replace <mask_prefix3> with
5594         orig,vex.
5595         (*<plusminus_insn><mode>3): Likewise.  Remove <mask_operand3> uses.
5596
5597 2018-03-05  Peter Bergner  <bergner@vnet.ibm.com>
5598
5599         PR target/84264
5600         * config/rs6000/vector.md (mov<mode>): Disallow altivec memory operands.
5601
5602 2018-03-05  Richard Biener  <rguenther@suse.de>
5603
5604         PR tree-optimization/84486
5605         * tree-ssa-pre.c (create_expression_by_pieces): Remove dead code.
5606         When inserting a __builtin_assume_aligned call set the LHS
5607         SSA name alignment info accordingly.
5608
5609 2018-03-05  Wilco Dijkstra  <wdijkstr@arm.com>
5610
5611         PR tree-optimization/84114
5612         * config/aarch64/aarch64.c (aarch64_reassociation_width)
5613         Avoid reassociation of FLOAT_MODE addition.
5614
5615 2018-03-05  Olga Makhotina  <olga.makhotina@intel.com>
5616
5617         * common/config/i386/i386-common.c (OPTION_MASK_ISA_PCONFIG_SET,
5618         OPTION_MASK_ISA_PCONFIG_UNSET, OPTION_MASK_ISA_WBNOINVD_SET,
5619         OPTION_MASK_ISA_WBNOINVD_UNSET): New definitions.
5620         (ix86_handle_option): Handle -mpconfig and -mwbnoinvd.
5621         * config.gcc (pconfigintrin.h, wbnoinvdintrin.h) : Add headers.
5622         * config/i386/cpuid.h (bit_PCONFIG, bit_WBNOINVD): New.
5623         * config/i386/driver-i386.c (host_detect_local_cpu): Detect -mpconfig
5624         and -mwbnoinvd.
5625         * config/i386/i386-builtin.def (__builtin_ia32_wbnoinvd,
5626         __builtin_ia32_wbinvd): New builtins.
5627         (SPECIAL_ARGS2): New.
5628         * config/i386/i386-c.c (__WBNOINVD__, __PCONFIG__): New.
5629         (SPECIAL_ARGS2): New.
5630         * config/i386/i386.c (ix86_target_string): Add -mpconfig and -mwbnoinvd.
5631         (ix86_valid_target_attribute_inner_p): Ditto.
5632         (ix86_init_mmx_sse_builtins): Add special_args2.
5633         * config/i386/i386.h (TARGET_PCONFIG, TARGET_PCONFIG_P, TARGET_WBNOINVD,
5634         TARGET_WBNOINVD_P): New.
5635         * config/i386/i386.md (UNSPECV_WBINVD, UNSPECV_WBNOINVD): New.
5636         (define_insn "wbinvd", define_insn "wbnoinvd"): New.
5637         * config/i386/i386.opt: Add -mpconfig and -mwbnoinvd.
5638         * config/i386/immintrin.h (_wbinvd): New intrinsic.
5639         * config/i386/pconfigintrin.h: New file.
5640         * config/i386/wbnoinvdintrin.h: Ditto.
5641         * config/i386/x86intrin.h: Add headers pconfigintrin.h and wbnoinvdintrin.h.
5642         * doc/invoke.texi (-mpconfig, -mwbnoinvd): New.
5643
5644 2018-03-05  Richard Biener  <rguenther@suse.de>
5645
5646         PR tree-optimization/84670
5647         * tree-ssa-pre.c (struct bb_bitmap_sets): Add visited_with_visited_succs
5648         member.
5649         (BB_VISITED_WITH_VISITED_SUCCS): New define.
5650         (compute_antic): Initialize BB_VISITED_WITH_VISITED_SUCCS.
5651         (compute_antic_aux): Only assert the number of values in ANTIC_IN
5652         doesn't grow if all successors (recursively) were visited at least
5653         once.
5654
5655 2018-03-05  Richard Biener  <rguenther@suse.de>
5656
5657         PR tree-optimization/84650
5658         * tree-ssa-loop-im.c (pass_lim::execute): Reset the SCEV cache
5659         if executed in the loop pipeline.
5660
5661 2018-03-05  Sandra Loosemore  <sandra@codesourcery.com>
5662
5663         * doc/configfiles.texi (Configuration Files): Move info about
5664         conditionalizing $target-protos.h to...
5665         * doc/sourcebuild.texi (Back End): Here.  Explain how $target.h
5666         differs from $target-protos.h.
5667
5668 2018-03-05  Kito Cheng  <kito.cheng@gmail.com>
5669             Chung-Ju Wu  <jasonwucj@gmail.com>
5670
5671         * config/nds32/nds32-protos.h (nds32_expand_setmem): Declare.
5672         * config/nds32/nds32-multiple.md (setmemsi): Define.
5673         * config/nds32/nds32-memory-manipulation.c
5674         (nds32_gen_dup_4_byte_to_word_value): New.
5675         (emit_setmem_word_loop): New.
5676         (emit_setmem_byte_loop): New.
5677         (nds32_expand_setmem_loop): New.
5678         (nds32_expand_setmem_loop_v3m): New.
5679         (nds32_expand_setmem_unroll): New.
5680         (nds32_expand_setmem): New.
5681
5682 2018-03-04  Kito Cheng  <kito.cheng@gmail.com>
5683             Chung-Ju Wu  <jasonwucj@gmail.com>
5684
5685         * config/nds32/nds32-memory-manipulation.c
5686         (nds32_emit_load_store): New.
5687         (nds32_emit_post_inc_load_store): New.
5688         (nds32_emit_mem_move): New.
5689         (nds32_emit_mem_move_block): New.
5690         (nds32_expand_movmemsi_loop_unknown_size): New.
5691         (nds32_expand_movmemsi_loop_known_size): New.
5692         (nds32_expand_movmemsi_loop): New.
5693         (nds32_expand_movmemsi_unroll): New.
5694         (nds32_expand_movmemqi): Rename ...
5695         (nds32_expand_movmemsi): ... to this.
5696         * config/nds32/nds32-multiple.md (movmemqi): Rename ...
5697         (movmemsi): ... to this.
5698         * config/nds32/nds32-protos.h (nds32_expand_movmemqi): Rename ...
5699         (nds32_expand_movmemsi): ... to this.
5700
5701 2018-03-04  Kito Cheng  <kito.cheng@gmail.com>
5702             Monk Chiang  <sh.chiang04@gmail.com>
5703             Chung-Ju Wu  <jasonwucj@gmail.com>
5704
5705         * config/nds32/nds32-protos.h
5706         (nds32_expand_load_multiple): New arguments.
5707         (nds32_expand_store_multiple): Ditto.
5708         (nds32_valid_multiple_load_store): Rename ...
5709         (nds32_valid_multiple_load_store_p): ... to this.
5710         * config/nds32/nds32-memory-manipulation.c
5711         (nds32_expand_load_multiple): Refine implementation.
5712         (nds32_expand_store_multiple): Ditto.
5713         * config/nds32/nds32-multiple.md
5714         (load_multiple): Update nds32_expand_load_multiple interface.
5715         (store_multiple): Update nds32_expand_store_multiple interface.
5716         * config/nds32/nds32-predicates.c
5717         (nds32_valid_multiple_load_store): Rename ...
5718         (nds32_valid_multiple_load_store_p): ... to this and refine
5719         implementation.
5720         * config/nds32/predicates.md
5721         (nds32_load_multiple_and_update_address_operation): New predicate.
5722         (nds32_store_multiple_and_update_address_operation): New predicate.
5723
5724 2018-03-04  Kito Cheng  <kito.cheng@gmail.com>
5725             Chung-Ju Wu  <jasonwucj@gmail.com>
5726
5727         * config/nds32/nds32.md (type): Add load_multiple and store_multiple.
5728         (combo): New attribute.
5729         * config/nds32/nds32-multiple.md: Refine patterns with new attributes.
5730
5731 2018-03-03  Chung-Ju Wu  <jasonwucj@gmail.com>
5732
5733         * config/nds32/nds32.opt: Change -mcmodel= default value.
5734
5735 2018-03-03  Kito Cheng  <kito.cheng@gmail.com>
5736             Monk Chiang  <sh.chiang04@gmail.com>
5737             Chung-Ju Wu  <jasonwucj@gmail.com>
5738
5739         * config/nds32/constants.md (unspec_element): New enum.
5740         * config/nds32/constraints.md (Umw): New constraint.
5741         * config/nds32/nds32-intrinsic.c: Add more builtin functions.
5742         * config/nds32/nds32-intrinsic.md: Likewise.
5743         * config/nds32/nds32-md-auxiliary.c (nds32_regno_to_enable4): New.
5744         (nds32_valid_smw_lwm_base_p): New.
5745         (nds32_output_smw_single_word): New.
5746         (nds32_output_lmw_single_word): New.
5747         (nds32_expand_unaligned_load): New.
5748         (nds32_expand_unaligned_store): New.
5749         * config/nds32/nds32-protos.h (nds32_valid_smw_lwm_base_p): Declare.
5750         (nds32_output_smw_single_word): Declare.
5751         (nds32_output_lmw_single_word): Declare.
5752         (nds32_expand_unaligned_load): Declare.
5753         (nds32_expand_unaligned_store): Declare.
5754         * config/nds32/nds32.h (nds32_builtins): Add NDS32_BUILTIN_UALOAD_HW,
5755         NDS32_BUILTIN_UALOAD_W, NDS32_BUILTIN_UALOAD_DW,
5756         NDS32_BUILTIN_UASTORE_HW, NDS32_BUILTIN_UASTORE_W,
5757         NDS32_BUILTIN_UASTORE_DW.
5758         * config/nds32/predicates.md (nds32_lmw_smw_base_operand): New
5759         predicate.
5760
5761 2018-03-03  Monk Chiang  <sh.chiang04@gmail.com>
5762             Kito Cheng  <kito.cheng@gmail.com>
5763             Chung-Ju Wu  <jasonwucj@gmail.com>
5764
5765         * config/nds32/nds32-intrinsic.c
5766         (nds32_expand_builtin_null_ftype_reg): Delete.
5767         (nds32_expand_builtin_reg_ftype_imm): Ditto.
5768         (nds32_expand_builtin_null_ftype_reg_imm): Ditto.
5769         (nds32_read_argument): New.
5770         (nds32_legitimize_target): Ditto.
5771         (nds32_legitimize_argument): Ditto.
5772         (nds32_check_constant_argument): Ditto.
5773         (nds32_expand_unop_builtin): Ditto.
5774         (nds32_expand_unopimm_builtin): Ditto.
5775         (nds32_expand_binop_builtin): Ditto.
5776         (nds32_builtin_decl_impl): Ditto.
5777         (builtin_description): Ditto.
5778         (nds32_expand_builtin_impl): Rewrite with new infrastructure.
5779         (nds32_init_builtins_impl): Ditto.
5780         * config/nds32/nds32.c (TARGET_BUILTIN_DECL): Define.
5781         (nds32_builtin_decl): New.
5782         * config/nds32/nds32.h (nds32_builtins): Add NDS32_BUILTIN_COUNT.
5783         * config/nds32/nds32-protos.h (nds32_builtin_decl_impl): Declare.
5784
5785 2018-03-02  Jeff Law  <law@redhat.com>
5786
5787         * reorg.c (stop_search_p): Handle DEBUG_INSN.
5788         (redundant_insn, fill_simple_delay_slots): Likewise.
5789         (fill_slots_from_thread): Likewise.
5790         * resource.c (mark_referenced_resources): Likewise.
5791         (mark_set_resources, find_dead_or_set_registers): Likewise.
5792
5793 2018-03-02  Jakub Jelinek  <jakub@redhat.com>
5794
5795         * substring-locations.h (format_warning_va): Formatting fix for
5796         ATTRIBUTE_GCC_DIAG.
5797         (format_warning_at_substring): Fix up ATTRIBUTE_GCC_DIAG second
5798         argument.
5799         (format_warning_n_va, format_warning_at_substring_n): New prototypes.
5800         * substring-locations.c: Include intl.h.
5801         (format_warning_va): Turned into small wrapper around
5802         format_warning_n_va, renamed to ...
5803         (format_warning_n_va): ... this, add N and PLURAL_GMSGID arguments,
5804         rename GMSGID to SINGULAR_GMSGID, if SINGULAR_GMSGID != PLURAL_GMSGID,
5805         use ngettext.
5806         (format_warning_at_substring_n): New function.
5807         * gimple-ssa-sprintf.c: Remove GCC diagnostic ignored pragma.
5808         (fmtwarn): Add ATTRIBUTE_GCC_DIAG.  Turn into a copy of
5809         format_warning_at_substring with just a shorter name instead of
5810         const function pointer.
5811         (fmtwarn_n): New function.
5812         (maybe_warn, format_directive, parse_directive): Use fmtwarn_n where
5813         appropriate, get rid of all the fmtstr temporaries, move conditionals
5814         with G_() wrapped string literals directly into fmtwarn arguments,
5815         cast dir.len to (int), formatting fixes.
5816
5817 2018-03-02  Thomas Schwinge  <thomas@codesourcery.com>
5818
5819         * doc/invoke.texi: Remove "Cilk Plus" references.
5820
5821 2018-03-02  Jakub Jelinek  <jakub@redhat.com>
5822             Richard Biener  <rguenther@suse.de>
5823
5824         PR ipa/84628
5825         * expr.c (expand_expr_real_1) <case CALL_EXPR>: Don't emit diagnostics
5826         for error or warning attributes if CALL_FROM_THUNK_P is set.
5827         Formatting fixes.
5828
5829 2018-03-02  Jakub Jelinek  <jakub@redhat.com>
5830
5831         PR target/56540
5832         * config/pa/pa.h (TARGET_CPU_CPP_BUILTINS): Predefine
5833         __SIZEOF_128__ macro if HPUX_LONG_DOUBLE_LIBRARY.
5834
5835         PR target/56540
5836         * config/ia64/ia64.h (TARGET_CPU_CPP_BUILTINS): Predefine
5837         __SIZEOF_{FPREG,FLOAT{80,128}}__ macros.
5838
5839         * predict.c (test_prediction_value_range): Use PROB_UNINITIALIZED
5840         instead of -1U in last predictors element's probability member.
5841
5842 2018-03-02  Eric Botcazou  <ebotcazou@adacore.com>
5843
5844         PR ipa/83983
5845         * ipa-devirt.c (odr_subtypes_equivalent_p): Get the ODR type of both
5846         arguments if they are comparable.
5847
5848 2018-03-02  Richard Sandiford  <richard.sandiford@linaro.org>
5849
5850         PR tree-optimization/84634
5851         * tree-vect-stmts.c (vectorizable_store, vectorizable_load): Replace
5852         masks and masked_loop_p with a single loop_masks, making sure it's
5853         null for bb vectorization.
5854
5855 2018-03-02  Richard Sandiford  <richard.sandiford@linaro.org>
5856
5857         * tree-vect-data-refs.c (vect_analyze_data_ref_dependence)
5858         (vect_analyze_data_ref_access): Use loop->safe_len rather than
5859         loop->force_vectorize to check whether there is no alias.
5860
5861 2018-03-02  Jakub Jelinek  <jakub@redhat.com>
5862
5863         PR target/84614
5864         * rtl.h (prev_real_nondebug_insn, next_real_nondebug_insn): New
5865         prototypes.
5866         * emit-rtl.c (next_real_insn, prev_real_insn): Fix up function
5867         comments.
5868         (next_real_nondebug_insn, prev_real_nondebug_insn): New functions.
5869         * cfgcleanup.c (try_head_merge_bb): Use prev_real_nondebug_insn
5870         instead of a loop around prev_real_insn.
5871         * combine.c (move_deaths): Use prev_real_nondebug_insn instead of
5872         prev_real_insn.
5873
5874         PR inline-asm/84625
5875         * config/i386/i386.c (ix86_print_operand): Use conditional
5876         output_operand_lossage instead of gcc_assert if CONST_VECTOR is not
5877         zero vector.
5878
5879 2018-03-02  Richard Biener  <rguenther@suse.de>
5880
5881         PR tree-optimization/84427
5882         * tree-ssa-pre.c (bitmap_remove_expr_from_set): Remove.
5883         (bitmap_set_subtract_values): Rewrite to handle multiple
5884         exprs per value.
5885         (clean): Likewise.
5886         (prune_clobbered_mems): Likewise.
5887         (phi_translate): Take edge instead of pred/phiblock.
5888         (phi_translate_1): Likewise.
5889         (phi_translate_set): Likewise.  Insert all translated
5890         exprs for a value into the set, keeping possibly multiple
5891         expressions per value.
5892         (compute_antic_aux): Adjust for phi_translate changes.
5893         When intersecting union the expressions and prune those
5894         not in the final value set, keeping possibly multiple
5895         expressions per value.  Do not use value-insertion
5896         for unioning ANTIC_OUT U EXP_GEN - TMP_GEN but merge
5897         all expressions.  Add verification that the value-sets
5898         only shrink during iteration.
5899         (compute_partial_antic_aux): Adjust for the phi_translate changes.
5900         (do_pre_regular_insertion): Likewise.
5901         (do_pre_partial_partial_insertion): Likewise.
5902
5903 2018-03-02  Richard Biener  <rguenther@suse.de>
5904
5905         PR target/82005
5906         * config/darwin.c (saved_debug_info_level): New static global.
5907         (darwin_asm_lto_start): Disable debug info generation for LTO out.
5908         (darwin_asm_lto_end): Restore debug info generation settings.
5909
5910 2018-03-01  Martin Liska  <mliska@suse.cz>
5911
5912         PR sanitizer/82484
5913         * sanopt.c (sanitize_rewrite_addressable_params): Do not handle
5914         volatile arguments.
5915
5916 2018-03-01  Richard Biener  <rguenther@suse.de>
5917
5918         PR debug/84645
5919         * dwarf2out.c (gen_variable_die): Properly handle late VLA
5920         type annotation with LTO when debug was disabled at compile-time.
5921
5922 2018-03-01  Matthew Fortune  <mfortune@gmail.com>
5923
5924         * config/mips/mips.c (mips_final_prescan_insn): Fix incorrect
5925         XINT with INTVAL.
5926         (mips_final_postscan_insn): Likewise.
5927
5928 2018-03-01  Richard Sandiford  <richard.sandiford@linaro.org>
5929
5930         PR rtl-optimization/84528
5931         * alias.c (init_alias_target): Add commentary.
5932         (init_alias_analysis): Only give HARD_FRAME_POINTER_REGNUM
5933         a unique base value if the frame pointer is not eliminated
5934         to the stack pointer.
5935
5936 2018-03-01  Tom de Vries  <tom@codesourcery.com>
5937
5938         PR rtl-optimization/83327
5939         * lra-int.h (hard_regs_spilled_into): Declare.
5940         * lra.c (hard_regs_spilled_into): Define.
5941         (init_reg_info): Init hard_regs_spilled_into.
5942         * lra-spills.c (assign_spill_hard_regs): Update hard_regs_spilled_into.
5943         * lra-lives.c (make_hard_regno_born, make_hard_regno_dead)
5944         (process_bb_lives): Handle hard_regs_spilled_into.
5945         (lra_create_live_ranges_1): Before doing liveness propagation, clear
5946         regs in all_hard_regs_bitmap if set in hard_regs_spilled_into.
5947
5948 2018-02-28  David Edelsohn  <dje.gcc@gmail.com>
5949
5950         * config.gcc (powerpc-ibm-aix7.1.*): New stanza.
5951         (powerpc-ibm-aix[789]*): Default to AIX 7.2.
5952         * config/rs6000/aix71.h (TARGET_DEFAULT): Revert to Power4 ISA.
5953         * config/rs6000/aix72.h: New file.
5954
5955 2018-02-28  Jakub Jelinek  <jakub@redhat.com>
5956
5957         * gimple-ssa-warn-restrict.c (maybe_diag_overlap): Use warning_n
5958         instead of warning_at with conditional singular and plural messages
5959         where possible.
5960
5961         PR target/52991
5962         * stor-layout.c (update_alignment_for_field): For
5963         targetm.ms_bitfield_layout_p (rli->t), if !is_bitfield
5964         && !DECL_PACKED (field), do the alignment update, just use
5965         only desired_align instead of MAX (type_align, desired_align)
5966         as the alignment.
5967         (place_field): Don't do known_align < desired_align handling
5968         early if targetm.ms_bitfield_layout_p (rli->t) and rli->prev_field
5969         is non-NULL, instead do it after rli->prev_field handling and
5970         only if not within a bitfield word.  For DECL_PACKED (field)
5971         use type_align of BITS_PER_UNIT.
5972
5973 2018-02-28  Eric Botcazou  <ebotcazou@adacore.com>
5974
5975         * config/aarch64/aarch64.c (aarch64_emit_probe_stack_range): Remove
5976         superfluous parentheses and trailing spaces.
5977
5978 2018-02-28  Richard Biener  <rguenther@suse.de>
5979
5980         PR tree-optimization/84584
5981         * graphite-scop-detection.c (scop_detection::add_scop): Discard
5982         SCoPs with fake exit edge.
5983
5984 2018-02-28  Martin Liska  <mliska@suse.cz>
5985
5986         PR testsuite/84597
5987         * timevar.c (timer::print): Fix format to properly print 100%
5988         values.
5989
5990 2018-02-28  Richard Biener  <rguenther@suse.de>
5991
5992         PR middle-end/84607
5993         * genmatch.c (capture_info::walk_match): Do not mark
5994         captured expressions without operands as expr_p given
5995         they act more like predicates and should be subject to
5996         "lost tail" side-effect preserving.
5997
5998 2018-02-28  Alexandre Oliva  <aoliva@redhat.com>
5999
6000         PR rtl-optimization/81611
6001         * auto-inc-dec.c (attempt_change): Move dead note from
6002         mem_insn if it's the next use of regno
6003         (find_address): Take address use of reg holding
6004         non-incremented value.  Add parm to limit search to the named
6005         reg only.
6006         (merge_in_block): Attempt to use a mem insn that is the next
6007         use of the original regno.
6008
6009 2018-02-27  Martin Sebor  <msebor@redhat.com>
6010
6011         PR c++/83871
6012         * gcc/doc/invoke.texi (-Wmissing-attributes): New option.
6013         * gcc/print-tree.c (print_node): Handle DECL_UNINLINABLE.
6014
6015 2018-02-27  Martin Sebor  <msebor@redhat.com>
6016
6017         PR translation/84207
6018         * diagnostic-core.h (warning_n, error_n, inform_n): Change
6019         n argument to unsigned HOST_WIDE_INT.
6020         * diagnostic.c (warning_n, error_n, inform_n): Ditto.
6021         (diagnostic_n_impl): Ditto.  Handle arguments in excess of LONG_MAX.
6022         * gimple-ssa-sprintf.c (format_directive): Simplify inform_n call.
6023         * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Use warning_n.
6024
6025 2018-02-27  Richard Biener  <rguenther@suse.de>
6026
6027         PR tree-optimization/84512
6028         * tree-vect-loop.c (vect_compute_single_scalar_iteration_cost):
6029         Do not use the estimate returned from record_stmt_cost for
6030         the scalar iteration cost but sum properly using add_stmt_cost.
6031
6032 2018-02-27  Richard Biener  <rguenther@suse.de>
6033
6034         PR tree-optimization/84466
6035         * graphite-scop-detection.c (scop_detection::stmt_simple_for_scop_p):
6036         Adjust last change to less strictly validate use operands.
6037
6038 2018-02-27  Martin Liska  <mliska@suse.cz>
6039
6040         PR gcov-profile/84548
6041         * gcov.c (process_file): Allow partial overlap and consider it
6042         also as group functions.
6043         (output_lines): Properly calculate range of lines for a group.
6044
6045 2018-02-27  Martin Liska  <mliska@suse.cz>
6046
6047         * timevar.c (timer::print_row): Remove 'usr', 'sys', 'wall' and
6048         'ggc' suffixes.  Change first column width.
6049         (timer::print): Fix formatting of the column.
6050
6051 2018-02-27  Alexandre Oliva  <aoliva@redhat.com>
6052
6053         * tree-ssa-live.c (remove_unused_scope_block_p): Do not
6054         preserve inline entry blocks for the sake of debug inline
6055         entry point markers alone.
6056         (remove_unused_locals): Suggest in comments a better place to
6057         force the preservation of inline entry blocks that are
6058         otherwise unused, but do not preserve them.
6059
6060 2018-02-26  H.J. Lu  <hongjiu.lu@intel.com>
6061
6062         * config/i386/i386.c (ix86_output_indirect_jmp): Update comments.
6063
6064 2018-02-26  H.J. Lu  <hongjiu.lu@intel.com>
6065
6066         PR target/84039
6067         * config/i386/constraints.md (Bs): Replace
6068         ix86_indirect_branch_register with
6069         TARGET_INDIRECT_BRANCH_REGISTER.
6070         (Bw): Likewise.
6071         * config/i386/i386.md (indirect_jump): Likewise.
6072         (tablejump): Likewise.
6073         (*sibcall_memory): Likewise.
6074         (*sibcall_value_memory): Likewise.
6075         Peepholes of indirect call and jump via memory: Likewise.
6076         (*sibcall_GOT_32): Disallowed for TARGET_INDIRECT_BRANCH_REGISTER.
6077         (*sibcall_value_GOT_32): Likewise.
6078         * config/i386/predicates.md (indirect_branch_operand): Likewise.
6079         (GOT_memory_operand): Likewise.
6080         (call_insn_operand): Likewise.
6081         (sibcall_insn_operand): Likewise.
6082         (GOT32_symbol_operand): Likewise.
6083         * config/i386/i386.h (TARGET_INDIRECT_BRANCH_REGISTER): New.
6084
6085 2018-02-26  Eric Botcazou  <ebotcazou@adacore.com>
6086
6087         PR rtl-optimization/83496
6088         * reorg.c (steal_delay_list_from_target): Change REDUNDANT array from
6089         booleans to RTXes.  Call fix_reg_dead_note on every non-null element.
6090         (steal_delay_list_from_fallthrough): Call fix_reg_dead_note on a
6091         redundant insn, if any.
6092         (relax_delay_slots): Likewise.
6093         (update_reg_unused_notes): Rename REDUNDANT_INSN to OTHER_INSN.
6094
6095 2018-02-26  Richard Sandiford  <richard.sandiford@linaro.org>
6096
6097         PR tree-optimization/83965
6098         * tree-vect-patterns.c (vect_reassociating_reduction_p): Assume
6099         that grouped statements are part of a reduction chain.  Return
6100         true if the statement is not marked as a reduction itself but
6101         is part of a group.
6102         (vect_recog_dot_prod_pattern): Don't check whether the statement
6103         is part of a group here.
6104         (vect_recog_sad_pattern): Likewise.
6105         (vect_recog_widen_sum_pattern): Likewise.
6106
6107 2018-02-26  Eric Botcazou  <ebotcazou@adacore.com>
6108
6109         PR debug/84545
6110         * final.c (rest_of_clean_state): Also look for calls inside sequences.
6111
6112 2018-02-26  H.J. Lu  <hongjiu.lu@intel.com>
6113
6114         PR target/84530
6115         * config/i386/i386-protos.h (ix86_output_indirect_jmp): Remove
6116         the bool argument.
6117         (ix86_output_indirect_function_return): New prototype.
6118         (ix86_split_simple_return_pop_internal): Likewise.
6119         * config/i386/i386.c (indirect_return_via_cx): New.
6120         (indirect_return_via_cx_bnd): Likewise.
6121         (indirect_thunk_name): Handle return va CX_REG.
6122         (output_indirect_thunk_function): Create alias for
6123         __x86_return_thunk_[re]cx and __x86_return_thunk_[re]cx_bnd.
6124         (ix86_output_indirect_jmp): Remove the bool argument.
6125         (ix86_output_indirect_function_return): New function.
6126         (ix86_split_simple_return_pop_internal): Likewise.
6127         * config/i386/i386.md (*indirect_jump): Don't pass false
6128         to ix86_output_indirect_jmp.
6129         (*tablejump_1): Likewise.
6130         (simple_return_pop_internal): Change it to define_insn_and_split.
6131         Call ix86_split_simple_return_pop_internal to split it for
6132         -mfunction-return=.
6133         (simple_return_indirect_internal): Call
6134         ix86_output_indirect_function_return instead of
6135         ix86_output_indirect_jmp.
6136
6137 2018-02-26  Jakub Jelinek  <jakub@redhat.com>
6138
6139         PR bootstrap/84405
6140         * vec.h (vec_default_construct): For BROKEN_VALUE_INITIALIZATION use
6141         memset and value initialization afterwards.
6142
6143 2018-02-26  Christophe Lyon  <christophe.lyon@linaro.org>
6144
6145         * Makefile.in (lto-wrapper): Use ALL_LINKERFLAGS.
6146
6147 2018-02-26  Ramana Radhakrishnan  <ramana.radhakrishnan@arm.com>
6148
6149         PR target/84521
6150         * common/config/aarch64/aarch64-common.c
6151         (aarch_option_optimization_table[]): Switch
6152         off fomit-frame-pointer
6153
6154 2018-02-26  Kito Cheng  <kito.cheng@gmail.com>
6155             Chung-Ju Wu  <jasonwucj@gmail.com>
6156
6157         * config/nds32/nds32-multiple.md (load_multiple): Disallow
6158         volatile memory.
6159         (store_multiple): Ditto.
6160
6161 2018-02-26  Kito Cheng  <kito.cheng@gmail.com>
6162
6163         * config.gcc: Add --with-cpu support for nds32 target.
6164         * config/nds32/nds32-opts.h (nds32_cpu_type): New.
6165         * config/nds32/nds32.opt: Add -mcpu= option.
6166
6167 2018-02-25  Segher Boessenkool  <segher@kernel.crashing.org>
6168
6169         * config/rs6000/rs6000.opt (mvrsave=no, mvrsave=yes, isel=no,
6170         isel=yes): Warn for these deprecated options.
6171
6172 2018-02-23  David Edelsohn  <dje.gcc@gmail.com>
6173
6174         * config/rs6000/aix71.h (TARGET_DEFAULT): Change to
6175         ISA_2_5_MASKS_EMBEDDED.
6176
6177 2018-02-23  Jakub Jelinek  <jakub@redhat.com>
6178
6179         * ipa-prop.c (ipa_vr_ggc_hash_traits::hash): Hash p->min and
6180         p->max as pointers rather than using iterative_hash_expr.
6181
6182 2018-02-23  Carl Love  <cel@us.ibm.com>
6183
6184         * config/rs6000/rs6000-builtin.def: Change VSIGNED2 and VUNSIGNED2
6185         macro expansions from BU_VSX_2 to BU_P8V_VSX_2 and BU_VSX_OVERLOAD_2 to
6186         BU_P8V_OVERLOAD_2.
6187         * config/rs6000/rs6000-c.c: Change VSX_BUILTIN_VEC_VSIGNED2 to
6188         P8V_BUILTIN_VEC_VSIGNED2.  Change VSX_BUILTIN_VEC_VUNSIGNED2 to
6189         P8V_BUILTIN_VEC_VUNSIGNED2.
6190
6191 2018-02-22  Vladimir Makarov  <vmakarov@redhat.com>
6192
6193         PR target/81572
6194         * lra-int.h (LRA_UNKNOWN_ALT, LRA_NON_CLOBBERED_ALT): New macros.
6195         * lra.c (lra_set_insn_recog_data, lra_update_insn_recog_data): Use
6196         LRA_UNKNOWN_ALT.
6197         * lra-constraints.c (curr_insn_transform): Set up
6198         LRA_NON_CLOBBERED_ALT for moves processed on the fast path.  Use
6199         LRA_UNKNOWN_ALT.
6200         (remove_inheritance_pseudos): Use LRA_UNKNOWN_ALT.
6201         * lra-eliminations.c (spill_pseudos): Ditto.
6202         (process_insn_for_elimination): Ditto.
6203         * lra-lives.c (reg_early_clobber_p): Use the new macros.
6204         * lra-spills.c (spill_pseudos): Use LRA_UNKNOWN_ALT and
6205         LRA_NON_CLOBBERED_ALT.
6206
6207 2018-02-22  Martin Sebor  <msebor@redhat.com>
6208
6209         PR tree-optimization/84480
6210         * gimple-fold.c (gimple_fold_builtin_strcpy): Move warnings
6211         to maybe_diag_stxncpy_trunc.  Call it.
6212         * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Integrate warnings
6213         from gimple_fold_builtin_strcpy.  Print inlining stack.
6214         (handle_builtin_stxncpy): Print inlining stack.
6215         * tree-ssa-strlen.h (maybe_diag_stxncpy_trunc): Declare.
6216
6217 2018-02-22  H.J. Lu  <hongjiu.lu@intel.com>
6218
6219         PR target/84176
6220         * config/i386/i386.c (ix86_set_indirect_branch_type): Issue an
6221         error when -mindirect-branch=thunk-extern, -fcf-protection=branch
6222         and -fcheck-pointer-bounds are used together.
6223         (indirect_thunk_prefix): New enum.
6224         (indirect_thunk_need_prefix): New function.
6225         (indirect_thunk_name): Replace need_bnd_p with need_prefix.  Use
6226         "_nt" instead of "_bnd" for NOTRACK prefix.
6227         (output_indirect_thunk): Replace need_bnd_p with need_prefix.
6228         (output_indirect_thunk_function): Likewise.
6229         (): Likewise.
6230         (ix86_code_end): Update output_indirect_thunk_function calls.
6231         (ix86_output_indirect_branch_via_reg): Replace
6232         ix86_bnd_prefixed_insn_p with indirect_thunk_need_prefix.
6233         (ix86_output_indirect_branch_via_push): Likewise.
6234         (ix86_output_function_return): Likewise.
6235         * doc/invoke.texi: Document -mindirect-branch=thunk-extern is
6236         incompatible with -fcf-protection=branch and
6237         -fcheck-pointer-bounds.
6238
6239 2018-02-22  Steve Ellcey  <sellcey@cavium.com>
6240
6241         PR target/83335
6242         * config/aarch64/aarch64.c (aarch64_print_address_internal):
6243         Change gcc_assert call to output_operand_lossage.
6244
6245 2018-02-22  Steve Ellcey  <sellcey@cavium.com>
6246
6247         * doc/extend.texi (__builtin_extend_pointer): Document builtin.
6248
6249 2018-02-22  DJ Delorie  <dj@redhat.com>
6250             Sebastian Perta  <sebastian.perta@renesas.com>
6251             Oleg Endo  <olegendo@gcc.gnu.org>
6252
6253         * config/rx/rx.c (rx_rtx_costs): New function.
6254         (TARGET_RTX_COSTS): Override to use rx_rtx_costs.
6255
6256 2018-02-22  Thomas Preud'homme  <thomas.preudhomme@arm.com>
6257
6258         * config/arm/t-multilib: Map Armv8-R to Armv7 multilibs.
6259
6260 2018-02-22  Martin Liska  <mliska@suse.cz>
6261
6262         PR driver/83193
6263         * common/config/arm/arm-common.c (arm_print_hint_for_cpu_option):
6264         Add "native" as a possible value.
6265
6266 2018-02-22  Martin Liska  <mliska@suse.cz>
6267
6268         PR driver/83193
6269         * config/i386/i386.c (ix86_option_override_internal):
6270         Add "native" as a possible value for -march and -mtune.
6271
6272 2018-02-22  Jakub Jelinek  <jakub@redhat.com>
6273
6274         PR target/84502
6275         * stor-layout.c (finalize_type_size): Propagate TYPE_EMPTY_P flag
6276         to all type variants.
6277
6278         PR tree-optimization/84503
6279         * gimple-ssa-store-merging.c (merged_store_group::merge_into): Compute
6280         width as info->bitpos + info->bitsize - start.
6281         (merged_store_group::merge_overlapping): Simplify width computation.
6282         (check_no_overlap): New function.
6283         (imm_store_chain_info::try_coalesce_bswap): Compute expected
6284         start + width and last_order of the group, fail if check_no_overlap
6285         fails.
6286         (imm_store_chain_info::coalesce_immediate_stores): Don't merge info
6287         to group if check_no_overlap fails.
6288
6289 2018-02-21  Segher Boessenkool  <segher@kernel.crashing.org>
6290
6291         * config/rs6000/altivec.md: Delete contraint arguments to
6292         define_expand, define_split, and define_peephole2, and in
6293         define_insn_and_split if always unused.
6294         * config/rs6000/darwin.md: Ditto.
6295         * config/rs6000/dfp.md: Ditto.
6296         * config/rs6000/rs6000.md: Ditto.
6297         * config/rs6000/sync.md: Ditto.
6298         * config/rs6000/vector.md: Ditto.
6299         * config/rs6000/vsx.md: Ditto.
6300
6301 2018-02-21  Segher Boessenkool  <segher@kernel.crashing.org>
6302
6303         * config/rs6000/altivec.md: Write output control strings as braced
6304         blocks instead of double-quoted strings.
6305         * config/rs6000/darwin.md: Ditto.
6306         * config/rs6000/rs6000.md: Ditto.
6307         * config/rs6000/vector.md: Ditto.
6308         * config/rs6000/vsx.md: Ditto.
6309
6310 2018-02-21  Jason Merrill  <jason@redhat.com>
6311
6312         PR c++/84314 - ICE with templates and fastcall attribute.
6313         * attribs.c (build_type_attribute_qual_variant): Remove assert.
6314
6315 2018-02-21  Jan Hubicka  <hubicka@ucw.cz>
6316
6317         * ipa-cp.c (determine_versionability): Fix comment typos.
6318
6319 2018-02-21  Jan Hubicka  <hubicka@ucw.cz>
6320
6321         PR c/84229
6322         * ipa-cp.c (determine_versionability): Do not version functions caling
6323         va_arg_pack.
6324
6325 2018-02-21  Martin Liska  <mliska@suse.cz>
6326
6327         PR driver/83193
6328         * config/aarch64/aarch64.c (aarch64_print_hint_for_core_or_arch):
6329         Add "native" as a possible value.
6330         * config/aarch64/aarch64.h (HAVE_LOCAL_CPU_DETECT):  Define
6331         the macro when native cpu detection is available.
6332
6333 2018-02-21  Martin Liska  <mliska@suse.cz>
6334
6335         PR driver/83193
6336         * common/config/arm/arm-common.c (arm_print_hint_for_arch_option):
6337         Add "native" as a possible value.
6338         * config/arm/arm.h (HAVE_LOCAL_CPU_DETECT): Define the macro
6339         when native cpu detection is available.
6340
6341 2018-02-21  Jakub Jelinek  <jakub@redhat.com>
6342             Martin Sebor  <msebor@redhat.com>
6343
6344         PR tree-optimization/84478
6345         * gimple-fold.h (get_range_strlen): Add a bool argument defaulted to
6346         false.
6347         * gimple-fold.c (get_range_strlen): Make minlen const and assume it
6348         can't be NULL.  Change FUZZY from bool to int, for 1 add PHI/COND_EXPR
6349         support which is conservatively correct, for 2 only stay conservative
6350         for maxlen.  Formatting and comment capitalization fixes.  Add STRICT
6351         argument to the 2 argument get_range_strlen, adjust 6 arg
6352         get_range_strlen caller and clear minmaxlen[0] and [1] if it returned
6353         false.
6354         (get_maxval_strlen): Adjust 6 arg get_range_strlen caller.
6355         (gimple_fold_builtin_strlen): Pass true as last argument to
6356         get_range_strlen.
6357
6358 2018-02-20  Martin Sebor  <msebor@redhat.com>
6359
6360         PR middle-end/84095
6361         * gimple-ssa-warn-restrict.c (builtin_memref::extend_offset_range): New.
6362         (builtin_memref::set_base_and_offset): Same.  Handle inner references.
6363         (builtin_memref::builtin_memref): Factor out parts into
6364         set_base_and_offset and call it.
6365
6366 2018-02-20  Richard Sandiford  <richard.sandiford@linaro.org>
6367
6368         PR middle-end/84406
6369         * optabs-query.c (find_widening_optab_handler_and_mode): If from_mode
6370         is a scalar_int_mode, assert that to_mode is a scalar_int_mode with
6371         greater precision.  If to_mode is a MODE_PARTIAL_INT, stop the
6372         search at the associated MODE_INT.
6373
6374 2018-02-20  Jeff Law  <law@redhat.com>
6375
6376         PR middle-end/82123
6377         PR tree-optimization/81592
6378         PR middle-end/79257
6379         * gimple-ssa-sprintf.c (format_integer): Query EVRP range analyzer
6380         for range data rather than using global data.
6381         * gimple-ssa-sprintf.c (get_int_range): Query EVRP range analyzer for
6382         range data rather than using global data.
6383         * gimple-ssa-sprintf.c (get_int_range): Accept vr_values parameter
6384         pass it to children as needed.
6385         (struct directive::fmtresult): Similarly.
6386         (struct directive::set_width): Similarly.
6387         (struct directive::set_precision): Similarly.
6388         (format_integer, format_directive, parse_directive): Similarly.
6389         (format_none): Accept unnamed vr_values parameter.
6390         (format_percent, format_floating, format_character): Similarly.
6391         (format_string, format_plain): Similarly.
6392         * gimple-ssa-sprintf.c (sprintf_dom_walker::handle_gimple_call): Query
6393         the EVRP range analyzer for range data rather than using global data.
6394         * gimple-ssa-sprintf.c: Include alloc-pool.h, vr-values.h and
6395         gimple-ssa-evrp-analyze.h
6396         (class sprintf_dom_walker): Add after_dom_children member function.
6397         Add evrp_range_analyzer member.
6398         (sprintf_dom_walker::before_dom_children): Call into the EVRP
6399         range analyzer as needed.
6400         (sprintf_dom_walker::after_dom_children): New member function.
6401         * gimple-ssa-evrp-analyze.c (evrp_range_analyzer::enter): Do nothing
6402         if not optimizing.
6403         (evrp_range_analyzer::record_ranges_from_stmt): Likewise.
6404         (evrp_range_analyzer::pop_to_marker): Likewise.
6405
6406 2018-02-20  Richard Sandiford  <richard.sandiford@linaro.org>
6407
6408         PR tree-optimization/84419
6409         * internal-fn.c (expand_call_mem_ref): Create a TARGET_MEM_REF
6410         with the required type if its current type is compatible but
6411         different.
6412
6413 2018-02-20  Jakub Jelinek  <jakub@redhat.com>
6414
6415         PR middle-end/82004
6416         * match.pd (pow(C,x) -> exp(log(C)*x)): Delay all folding until
6417         after vectorization.
6418
6419 2018-02-20  Martin Liska  <mliska@suse.cz>
6420
6421         PR driver/83193
6422         * config/aarch64/aarch64.c (aarch64_print_hint_for_core_or_arch): Print
6423         possible values if we don't have a hint.
6424
6425 2018-02-20  Martin Liska  <mliska@suse.cz>
6426
6427         PR c/84310
6428         PR target/79747
6429         * final.c (shorten_branches): Build align_tab array with one
6430         more element.
6431         * opts.c (finish_options): Add alignment option limit check.
6432         (MAX_CODE_ALIGN): Likewise.
6433         (MAX_CODE_ALIGN_VALUE): Likewise.
6434         * doc/invoke.texi: Document maximum allowed option value for
6435         all -falign-* options.
6436
6437 2018-02-19  Jakub Jelinek  <jakub@redhat.com>
6438
6439         PR target/84146
6440         * reg-notes.def (REG_CALL_ARG_LOCATION): New reg note.
6441         * insn-notes.def (NOTE_INSN_CALL_ARG_LOCATION): Remove.
6442         * var-tracking.c (emit_note_insn_var_location): Remove all references
6443         to NOTE_INSN_CALL_ARG_LOCATION.
6444         (emit_notes_in_bb): Emit arguments as REG_CALL_ARG_LOCATION note on
6445         the CALL_INSN rather than separate NOTE_INSN_CALL_ARG_LOCATION note.
6446         Use copy_rtx_if_shared.
6447         * dwarf2out.c (gen_subprogram_die): Use XEXP with 0 instead of
6448         NOTE_VAR_LOCATION on ca_loc->call_arg_loc_note.
6449         (dwarf2out_var_location): Remove handling of
6450         NOTE_INSN_CALL_ARG_LOCATION, instead handle REG_CALL_ARG_LOCATION note
6451         on call_insn.
6452         * final.c (final_scan_insn): Remove all references to
6453         NOTE_INSN_CALL_ARG_LOCATION.
6454         (rest_of_clean_state): Likewise.  Remove REG_CALL_ARG_LOCATION notes
6455         before dumping final insns.
6456         * except.c (emit_note_eh_region_end): Remove all references to
6457         NOTE_INSN_CALL_ARG_LOCATION.
6458         * config/alpha/alpha.c (alpha_pad_function_end): Likewise.
6459         * config/c6x/c6x.c (c6x_gen_bundles): Likewise.
6460         * config/arc/arc.c (hwloop_optimize): Likewise.
6461         * config/arm/arm.c (create_fix_barrier): Likewise.
6462         * config/s390/s390.c (s390_chunkify_start): Likewise.
6463         * config/sh/sh.c (find_barrier): Likewise.
6464         * config/i386/i386.c (rest_of_insert_endbranch,
6465         ix86_seh_fixup_eh_fallthru): Likewise.
6466         * config/xtensa/xtensa.c (hwloop_optimize): Likewise.
6467         * config/iq2000/iq2000.c (final_prescan_insn): Likewise.
6468         * config/frv/frv.c (frv_function_prologue): Likewise.
6469         * emit-rtl.c (try_split): Likewise.  Copy over REG_CALL_ARG_LOCATION
6470         reg note.
6471         (note_outside_basic_block_p): Remove all references to
6472         NOTE_INSN_CALL_ARG_LOCATION.
6473         * gengtype.c (adjust_field_rtx_def): Likewise.
6474         * print-rtl.c (rtx_writer::print_rtx_operand_code_0, print_insn):
6475         Likewise.
6476         * jump.c (cleanup_barriers, delete_related_insns): Likewise.
6477         * cfgrtl.c (force_nonfallthru_and_redirect): Likewise.
6478
6479         PR c++/84444
6480         * builtins.c (builtin_mathfn_code): Don't check if CALL_EXPR_FN (t)
6481         is ADDR_EXPR.
6482
6483         PR tree-optimization/84452
6484         * tree-vect-patterns.c (vect_recog_pow_pattern): Don't call
6485         expand_simd_clones if targetm.simd_clone.compute_vecsize_and_simdlen
6486         is NULL.
6487
6488 2018-02-19  Martin Liska  <mliska@suse.cz>
6489
6490         PR sanitizer/82183
6491         * passes.def: Put pass_sancov_O0 before pass_lower_switch with -O0.
6492
6493 2018-02-19  Martin Liska  <mliska@suse.cz>
6494             Richard Sandiford  <richard.sandiford@linaro.org>
6495
6496         PR tree-optimization/82491
6497         * gimple-fold.c (get_base_constructor): Make earlier bail out
6498         to prevent ubsan.
6499
6500 2018-02-19  Carl Love  <cel@us.ibm.com>
6501
6502         * config/rs6000/rs6000-builtin.def: Change NEG macro expansions from
6503         BU_ALTIVEC_A to BU_P8V_AV_1 and BU_ALTIVEC_OVERLOAD_1 to
6504         BU_P8V_OVERLOAD_1.
6505         * config/rs6000/rs6000-c.c: Change ALTIVEC_BUILTIN_VEC_NEG to
6506         P8V_BUILTIN_VEC_NEG.
6507
6508 2018-02-19  Sebastian Perta  <sebastian.perta@renesas.com>
6509
6510         * config/rl78/rl78.md (movdf): New define expand.
6511
6512 2018-02-19  Martin Liska  <mliska@suse.cz>
6513
6514         PR other/80589
6515         * doc/invoke.texi: Fix typo.
6516         * params.def (PARAM_MAX_LOOP_HEADER_INSNS): Likewise.
6517
6518 2018-02-18  Segher Boessenkool  <segher@kernel.crashing.org>
6519
6520         * config/rs6000/rs6000.c (rs6000_option_override_internal): Don't
6521         handle rs6000_single_float and rs6000_double_float specially for
6522         e500 family CPUs.
6523
6524 2018-02-16  Jeff Law  <law@redhat.com>
6525
6526         * config/rx/rx.c (add_pop_cfi_notes): New function.;
6527         (pop_regs): Use it.
6528
6529 2018-02-16  Jakub Jelinek  <jakub@redhat.com>
6530
6531         PR ipa/84425
6532         * ipa-inline.c (inline_small_functions): Fix a typo.
6533
6534 2018-02-16  Nathan Sidwell  <nathan@acm.org>
6535
6536         * doc/extend.texi (Backwards Compatibility): Americanize 'behaviour'.
6537
6538 2018-02-16  Carl Love  <cel@us.ibm.com>
6539
6540         * config/rs6000/rs6000-builtin.def: Add BU_P8V_VSX_2 macro definition.
6541         Change FLOAT2 expansions from BU_VSX_2 to BU_P8V_VSX_2 and
6542         from BU_VSX_OVERLOAD_2 to BU_P8V_OVERLOAD_2.
6543         * config/rs6000/rs6000-c.c: Changed macro VSX_BUILTIN_VEC_FLOAT2
6544         expansion to P8V_BUILTIN_VEC_FLOAT2.
6545
6546 2018-02-16  Vladimir Makarov  <vmakarov@redhat.com>
6547
6548         PR rtl-optimization/70023
6549         * lra-constraints.c (inherit_in_ebb): Take hard reg mode of
6550         src_regno into account.
6551
6552 2018-02-16  Carl Love  <cel@us.ibm.com>
6553
6554         * config/rs6000/altivec.h: Remove vec_vextract4b and vec_vinsert4b.
6555         * config/rs6000/rs6000-builtin.def: Remove macro expansion for
6556         VEXTRACT4B, VINSERT4B, VINSERT4B_DI and VEXTRACT4B.
6557         * config/rs6000/rs6000.c: Remove case statements for
6558         P9V_BUILTIN_VEXTRACT4B, P9V_BUILTIN_VEC_VEXTRACT4B,
6559         P9V_BUILTIN_VINSERT4B, P9V_BUILTIN_VINSERT4B_DI,
6560         and P9V_BUILTIN_VEC_VINSERT4B.
6561         * config/rs6000/rs6000-c.c (altivec_expand_builtin): Remove entries for
6562         P9V_BUILTIN_VEC_VEXTRACT4B and P9V_BUILTIN_VEC_VINSERT4B.
6563         * config/rs6000/vsx.md:
6564         * doc/extend.texi: Remove vec_vextract4b, non ABI definitions for
6565         vec_insert4b.
6566
6567 2018-02-16  Carl Love  <cel@us.ibm.com>
6568
6569         * config/rs6000/altivec.h: Add builtin names vec_extract4b
6570         vec_insert4b.
6571         * config/rs6000/rs6000-builtin.def: Add INSERT4B and EXTRACT4B
6572         definitions.
6573         * config/rs6000/rs6000-c.c: Add the definitions for
6574         P9V_BUILTIN_VEC_EXTRACT4B and P9V_BUILTIN_VEC_INSERT4B.
6575         * config/rs6000/rs6000.c (altivec_expand_builtin): Add
6576         P9V_BUILTIN_EXTRACT4B and P9V_BUILTIN_INSERT4B case statements.
6577         * config/rs6000/vsx.md: Add define_insn extract4b.  Add define_expand
6578         definition for insert4b and define insn *insert3b_internal.
6579         * doc/extend.texi: Add documentation for vec_extract4b.
6580
6581 2018-02-16  Nathan Sidwell  <nathan@acm.org>
6582
6583         * doc/extend.texi (Backwards Compatibility): Mention friend
6584         injection.  Note for-scope is deprecated.
6585         * doc/invoke.texi (-ffriend-injection): Deprecate.
6586
6587 2018-02-16  Segher Boessenkool  <segher@kernel.crashing.org>
6588
6589         * combine.c (try_combine): When adjusting LOG_LINKS for the destination
6590         that moved to I2, also allow destinations that are a paradoxical
6591         subreg (instead of a normal reg).
6592
6593 2018-02-16  Oleg Endo  <olegendo@gcc.gnu.org>
6594
6595         PR target/83831
6596         * config/rx/rx.c (rx_fuse_in_memory_bitop): Convert shift operand
6597         to QImode.
6598
6599 2018-02-16  Richard Biener  <rguenther@suse.de>
6600
6601         PR tree-optimization/84037
6602         PR tree-optimization/84016
6603         PR target/82862
6604         * config/i386/i386.c (ix86_builtin_vectorization_cost):
6605         Adjust vec_construct for the fact we need additional higher latency
6606         128bit inserts for AVX256 and AVX512 vector builds.
6607         (ix86_add_stmt_cost): Scale vector construction cost for
6608         elementwise loads.
6609
6610 2018-02-16  Richard Biener  <rguenther@suse.de>
6611
6612         PR tree-optimization/84417
6613         * tree-ssa.c (non_rewritable_mem_ref_base): Properly constrain
6614         the MEM_REF offset when conversion to BIT_FIELD_REF is desired.
6615         (non_rewritable_lvalue_p): Likewise, use poly-ints.
6616
6617 2018-02-16  Martin Liska  <mliska@suse.cz>
6618
6619         PR sanitizer/84307
6620         * internal-fn.def (ASAN_CHECK): Set proper flags.
6621         (ASAN_MARK): Likewise.
6622
6623 2018-02-16  Julia Koval  <julia.koval@intel.com>
6624
6625         * config/i386/i386.c (ix86_option_override_internal): Remove PTA_CLWB
6626         from PTA_CANNONLAKE.
6627
6628 2018-02-16  Jakub Jelinek  <jakub@redhat.com>
6629
6630         PR target/84272
6631         * config/aarch64/cortex-a57-fma-steering.c (fma_forest::merge_forest):
6632         Use ++iter rather than iter++ for std::list iterators.
6633         (func_fma_steering::dfs): Likewise.  Don't delete nodes right away,
6634         defer deleting them until all nodes in the forest are processed.  Do
6635         free even leaf nodes.  Change to_process into auto_vec.
6636
6637         PR bootstrap/84405
6638         * system.h (BROKEN_VALUE_INITIALIZATION): Define for GCC < 4.3.
6639         * vec.h (vec_default_construct): Use memset instead of placement new
6640         if BROKEN_VALUE_INITIALIZATION is defined.
6641         * hash-table.h (hash_table<Descriptor, Allocator>::empty_slow): Use
6642         memset instead of value initialization if BROKEN_VALUE_INITIALIZATION
6643         is defined.
6644
6645         PR rtl-optimization/83723
6646         * lra-int.h (lra_substitute_pseudo): Add DEBUG_P argument.
6647         * lra.c (lra_substitute_pseudo): Likewise.  If true, use
6648         gen_rtx_raw_SUBREG instead of gen_rtx_SUBREG.  Pass DEBUG_P to
6649         recursive calls.
6650         (lra_substitute_pseudo_within_insn): Adjust lra_substitute_pseudo
6651         callers.
6652         * lra-constraints.c (inherit_reload_reg, split_reg): Likewise.
6653
6654 2018-02-16  Eric Botcazou  <ebotcazou@adacore.com>
6655
6656         PR rtl-optimization/81443
6657         * rtlanal.c (num_sign_bit_copies1) <SUBREG>: Do not propagate results
6658         from inner REGs to paradoxical SUBREGs.
6659
6660 2018-02-16  Richard Biener  <rguenther@suse.de>
6661
6662         PR tree-optimization/84399
6663         * graphite-scop-detection.c (scop_detection::stmt_simple_for_scop_p):
6664         For operands we can analyze at their definition make sure we can
6665         analyze them at each use as well.
6666
6667 2018-02-16  Richard Biener  <rguenther@suse.de>
6668
6669         PR tree-optimization/84190
6670         * tree-ssa.c (non_rewritable_mem_ref_base): Do not touch
6671         volatile accesses if the decl isn't volatile.
6672
6673 2018-02-15  Jason Merrill  <jason@redhat.com>
6674
6675         PR c++/84314 - ICE with templates and fastcall attribute.
6676         * attribs.c (build_type_attribute_qual_variant): Don't clobber
6677         TYPE_CANONICAL on an existing type.
6678
6679 2018-02-15  Jakub Jelinek  <jakub@redhat.com>
6680
6681         PR tree-optimization/84383
6682         * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Don't look at
6683         dstoff nor call operand_equal_p if dstbase is NULL.
6684
6685         PR tree-optimization/84334
6686         * match.pd ((A +- CST1) +- CST2 -> A + CST3): If A is
6687         also a CONSTANT_CLASS_P, punt.
6688
6689 2018-02-14  Jim Wilson  <jimw@sifive.com>
6690
6691         * config/riscv/riscv.c (riscv_first_stack_step): Move locals after
6692         first SMALL_OPERAND check.  New local min_second_step.  Move assert
6693         to where locals are set.  Add TARGET_RVC support.
6694         * config/riscv/riscv.h (C_SxSP_BITS, SWSP_REACH, SDSP_REACH): New.
6695
6696 2018-02-14  Indu Bhagat  <indu.bhagat@oracle.com>
6697
6698         * doc/invoke.texi: Correct -Wformat-overflow code sample.
6699
6700 2018-02-14  Martin Sebor  <msebor@redhat.com>
6701
6702         PR tree-optimization/83698
6703         * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For
6704         arrays constrain the offset range to their bounds.
6705         (builtin_access::strcat_overlap): Adjust the bounds of overlap offset.
6706         (builtin_access::overlap): Avoid setting the size of overlap if it's
6707         already been set.
6708         (maybe_diag_overlap): Also consider arrays when deciding what values
6709         of offsets to include in diagnostics.
6710
6711 2018-02-14  Martin Sebor  <msebor@redhat.com>
6712
6713         PR c/84108
6714         * attribs.c (diag_attr_exclusions): Consider the exclusion(s)
6715         that correspond to the kind of a declaration.
6716
6717 2018-02-14  John David Anglin  <danglin@gcc.gnu.org>
6718
6719         PR target/83984
6720         * config/pa/pa.md: Load address of PIC label using the linkage table
6721         if the label is nonlocal.
6722
6723 2018-02-14  Kelvin Nilsen  <kelvin@gcc.gnu.org>
6724
6725         * config/rs6000/rs6000.c (rs6000_option_override_internal): Issue
6726         warning message if user requests -maltivec=be.
6727         * doc/invoke.texi: Document deprecation of -maltivec=be.
6728
6729 2018-02-14  Will Schmidt  <will_schmidt@vnet.ibm.com>
6730
6731         PR target/84220
6732         * config/rs6000/rs6000-c.c: Update definitions for
6733         ALTIVEC_BUILTIN_VEC_SLD, ALTIVEC_BUILTIN_VEC_SLDW,
6734         VEC_XXSLDWI and ALTIVEC_BUILTIN_VEC_XXPERMDI builtins.
6735
6736 2018-02-14  Igor Tsimbalist  <igor.v.tsimbalist@intel.com>
6737
6738         PR target/84239
6739         * config/i386/cetintrin.h: Remove _rdssp[d|q] and
6740         add _get_ssp intrinsics. Remove argument from
6741         __builtin_ia32_rdssp[d|q].
6742         * config/i386/i386-builtin-types.def: Add UINT_FTYPE_VOID.
6743         * config/i386/i386-builtin.def: Remove argument from
6744         __builtin_ia32_rdssp[d|q].
6745         * config/i386/i386.c: Use UINT_FTYPE_VOID. Use
6746         ix86_expand_special_args_builtin for _rdssp[d|q].
6747         * config/i386/i386.md: Remove argument from rdssp[si|di] insn.
6748         Clear register before usage.
6749         * doc/extend.texi: Remove argument from __builtin_ia32_rdssp[d|q].
6750         Add documentation for new _get_ssp and _inc_ssp intrinsics.
6751
6752 2018-02-14  Richard Sandiford  <richard.sandiford@linaro.org>
6753
6754         PR tree-optimization/84357
6755         * tree-data-ref.c (object_address_invariant_in_loop_p): Check
6756         operand 1 of an ARRAY_REF too.
6757
6758 2018-02-14  Oleg Endo  <olegendo@gcc.gnu.org>
6759
6760         PR target/83831
6761         * config/rx/rx-protos.h (rx_reg_dead_or_unused_after_insn,
6762         rx_copy_reg_dead_or_unused_notes, rx_fuse_in_memory_bitop): New
6763         declarations.
6764         (set_of_reg): New struct.
6765         (rx_find_set_of_reg, rx_find_use_of_reg): New functions.
6766         * config/rx/rx.c (rx_reg_dead_or_unused_after_insn,
6767         rx_copy_reg_dead_or_unused_notes, rx_fuse_in_memory_bitop): New
6768         functions.
6769         * config/rx/rx.md (andsi3, iorsi3, xorsi3): Convert to insn_and_split.
6770         Split into bitclr, bitset, bitinvert patterns if appropriate.
6771         (*bitset, *bitinvert, *bitclr): Convert to named insn_and_split and
6772         use rx_fuse_in_memory_bitop.
6773         (*bitset_in_memory, *bitinvert_in_memory, *bitclr_in_memory): Convert
6774         to named insn, correct maximum insn length.
6775
6776 2018-02-14  Jozef Lawrynowicz  <jozefl.gcc@gmail.com>
6777
6778         PR target/79242
6779         * machmode.def: Define a complex mode for PARTIAL_INT.
6780         * genmodes.c (complex_class): Return MODE_COMPLEX_INT for
6781         MODE_PARTIAL_INT.
6782         * doc/rtl.texi: Document CSPImode.
6783         * config/msp430/msp430.c (msp430_hard_regno_nregs): Add CPSImode
6784         handling.
6785         (msp430_hard_regno_nregs_with_padding): Likewise.
6786
6787 2018-02-13  Peter Bergner  <bergner@vnet.ibm.com>
6788
6789         PR target/84279
6790         * config/rs6000/rs6000.c (mem_operand_gpr): Disallow altivec addresses.
6791
6792 2018-02-13  Segher Boessenkool  <segher@kernel.crashing.org>
6793
6794         PR rtl-optimization/84169
6795         * combine.c (try_combine): New variable split_i2i3.  Set it to true if
6796         we generated a parallel as new i3 and we split that to new i2 and i3
6797         instructions.  Handle split_i2i3 similar to swap_i2i3: scan the
6798         LOG_LINKs of i3 to see which of those need to link to i2 now.  Link
6799         those to i2, not i1.  Partially rewrite this scan code.
6800
6801 2018-02-13  Jakub Jelinek  <jakub@redhat.com>
6802
6803         PR c/82210
6804         * stor-layout.c (place_field): For variable length fields, adjust
6805         offset_align afterwards not just based on the field's alignment,
6806         but also on the size.
6807
6808         PR middle-end/84309
6809         * match.pd (pow(C,x) -> exp(log(C)*x)): Use exp2s and log2s instead
6810         of exps and logs in the use_exp2 case.
6811
6812 2018-02-13  Jeff Law  <law@redhat.com>
6813
6814         * config/rl/rl78.c (rl78_attribute_table): Fix terminator and
6815         entry for "vector".
6816
6817         * config/rl78/rl78.c (rl78_handle_func_attribute): Mark
6818         ARGS as unused.
6819
6820 2018-02-13  Alexandre Oliva  <aoliva@redhat.com>
6821
6822         PR debug/84342
6823         PR debug/84319
6824         * common.opt (gas-loc-support, gas-locview-support): New.
6825         (ginline-points, ginternal-reset-location-views): New.
6826         * doc/invoke.texi: Document them.  Use @itemx where intended.
6827         (gvariable-location-views): Adjust.
6828         * target.def (reset_location_view): New.
6829         * doc/tm.texi.in (DWARF2_ASM_VIEW_DEBUG_INFO): New.
6830         (TARGET_RESET_LOCATION_VIEW): New.
6831         * doc/tm.texi: Rebuilt.
6832         * dwarf2out.c (dwarf2out_default_as_loc_support): New.
6833         (dwarf2out_default_as_locview_support): New.
6834         (output_asm_line_debug_info): Use option variables.
6835         (dwarf2out_maybe_output_loclist_view_pair): Likewise.
6836         (output_loc_list): Likewise.
6837         (add_high_low_attributes): Check option variables.
6838         Don't output entry view attribute in strict mode.
6839         (gen_inlined_subroutine_die): Check option variables.
6840         (dwarf2out_inline_entry): Likewise.
6841         (init_sections_and_labels): Likewise.
6842         (dwarf2out_early_finish): Likewise.
6843         (maybe_reset_location_view): New, from...
6844         (dwarf2out_var_location): ... here.  Call it.
6845         * debug.h (dwarf2out_default_as_loc_support): Declare.
6846         (dwarf2out_default_as_locview_support): Declare.
6847         * hooks.c (hook_int_rtx_insn_0): New.
6848         * hooks.h (hook_int_rtx_insn_0): Declare.
6849         * toplev.c (process_options): Take -gas-loc-support and
6850         -gas-locview-support from dwarf2out.  Enable
6851         -gvariable-location-views by default only with locview
6852         assembler support.  Enable -ginternal-reset-location-views by
6853         default only if the target defines the corresponding hook.
6854         Enable -ginline-points by default if location views are
6855         enabled; force it disabled if statement frontiers are
6856         disabled.
6857         * tree-inline.c (expand_call_inline): Check option variables.
6858         * tree-ssa-live.c (remove_unused_scope_block_p): Likewise.
6859
6860 2018-02-13  Richard Sandiford  <richard.sandiford@linaro.org>
6861
6862         PR tree-optimization/84321
6863         * tree-vrp.c (intersect_range_with_nonzero_bits): Fix VR_ANTI_RANGE
6864         handling.  Also check whether the anti-range contains any values
6865         that satisfy the mask; switch to a VR_RANGE if not.
6866
6867 2018-02-13  Paolo Bonzini  <bonzini@gnu.org>
6868
6869         PR sanitizer/84340
6870         * internal-fn.def (ASAN_CHECK, ASAN_MARK): Revert changes to fnspec.
6871
6872 2018-02-13  Martin Jambor  <mjambor@suse.cz>
6873
6874         PR c++/83990
6875         * ipa-param-manipulation.c (ipa_modify_call_arguments): Use location
6876         of call statements, also set location of a load to a temporary.
6877
6878 2018-02-13  Sebastian Perta  <sebastian.perta@renesas.com>
6879
6880         * config/rl78/rl78.c (add_vector_labels): New function.
6881         * config/rl78/rl78.c (rl78_handle_vector_attribute): New function.
6882         * config/rl78/rl78.c (rl78_start_function): Call add_vector_labels.
6883         * config/rl78/rl78.c (rl78_handle_func_attribute): Removed the assert
6884         which checks that no arguments are passed.
6885         * config/rl78/rl78.c (rl78_attribute_table): Add "vector" attribute.
6886         * doc/extend.texi: Documentation for the new attribute.
6887
6888 2018-02-13  Andreas Schwab  <schwab@suse.de>
6889
6890         * config/riscv/linux.h (CPP_SPEC): Define.
6891
6892 2018-02-13  Jakub Jelinek  <jakub@redhat.com>
6893
6894         PR target/84335
6895         * config/i386/i386.c (ix86_init_mmx_sse_builtins): Pass
6896         OPTION_MASK_ISA_AES | OPTION_MASK_ISA_SSE2 instead of
6897         OPTION_MASK_ISA_AES as first argument to def_builtin_const
6898         for AES builtins.  Pass OPTION_MASK_ISA_PCLMUL | OPTION_MASK_ISA_SSE2
6899         instead of OPTION_MASK_ISA_PCLMUL as first argument to
6900         def_builtin_const for __builtin_ia32_pclmulqdq128 builtin.
6901         * config/i386/wmmintrin.h: If __SSE2__ is not defined, enable it
6902         temporarily for AES and PCLMUL builtins.
6903
6904         PR tree-optimization/84339
6905         * gimple-fold.c (get_range_strlen): Set *FLEXP to true when handling
6906         ARRAY_REF where first operand is array_at_struct_end_p COMPONENT_REF.
6907         Formatting fixes.
6908
6909         PR middle-end/84309
6910         * match.pd (pow(C,x) -> exp(log(C)*x)): Optimize instead into
6911         exp2(log2(C)*x) if C is a power of 2 and c99 runtime is available.
6912         * generic-match-head.c (canonicalize_math_after_vectorization_p): New
6913         inline function.
6914         * gimple-match-head.c (canonicalize_math_after_vectorization_p): New
6915         inline function.
6916         * omp-simd-clone.h: New file.
6917         * omp-simd-clone.c: Include omp-simd-clone.h.
6918         (expand_simd_clones): No longer static.
6919         * tree-vect-patterns.c: Include fold-const-call.h, attribs.h,
6920         cgraph.h and omp-simd-clone.h.
6921         (vect_recog_pow_pattern): Optimize pow(C,x) to exp(log(C)*x).
6922         (vect_recog_widen_shift_pattern): Formatting fix.
6923         (vect_pattern_recog_1): Don't check optab for calls.
6924
6925         PR target/84336
6926         * config/i386/sse.md (<avx512>_vpermi2var<mode>3_mask): Force
6927         operands[2] into a REG before using gen_lowpart on it.
6928
6929 2018-02-12  Jeff Law  <law@redhat.com>
6930
6931         PR target/83760
6932         * config/sh/sh.c (find_barrier): Consider a sibling call
6933         a barrier as well.
6934
6935         * cse.c (try_back_substitute_reg): Move any REG_ARGS_SIZE note when
6936         successfully back substituting a reg.
6937
6938 2018-02-12  Richard Biener  <rguenther@suse.de>
6939
6940         PR tree-optimization/84037
6941         * tree-vect-slp.c (vect_analyze_slp_cost): Add visited
6942         parameter, move visited init to caller.
6943         (vect_slp_analyze_operations): Separate cost from validity
6944         check, initialize visited once for all instances.
6945         (vect_schedule_slp): Analyze map to CSE vectorized nodes once
6946         for all instances.
6947         * tree-vect-stmts.c (vect_model_simple_cost): Make early
6948         out an assert.
6949         (vect_model_promotion_demotion_cost): Likewise.
6950         (vectorizable_bswap): Guard cost modeling with !slp_node
6951         instead of !PURE_SLP_STMT to avoid double-counting on hybrid
6952         SLP stmts.
6953         (vectorizable_call): Likewise.
6954         (vectorizable_conversion): Likewise.
6955         (vectorizable_assignment): Likewise.
6956         (vectorizable_shift): Likewise.
6957         (vectorizable_operation): Likewise.
6958         (vectorizable_store): Likewise.
6959         (vectorizable_load): Likewise.
6960         (vectorizable_condition): Likewise.
6961         (vectorizable_comparison): Likewise.
6962
6963 2018-02-12  Paolo Bonzini  <bonzini@gnu.org>
6964
6965         PR sanitizer/84307
6966         * internal-fn.def (ASAN_CHECK): Fix fnspec to account for return value.
6967         (ASAN_MARK): Fix fnspec to account for return value, change pointer
6968         argument from 'R' to 'W' so that the pointed-to datum is clobbered.
6969
6970 2018-02-08  Jan Hubicka  <hubicka@ucw.cz>
6971
6972         PR middle-end/83665
6973         * params.def (inline-min-speedup): Increase from 8 to 15.
6974         (max-inline-insns-auto): Decrease from 40 to 30.
6975         * ipa-split.c (consider_split): Add some buffer for function to
6976         be considered inlining candidate.
6977         * invoke.texi (max-inline-insns-auto, inline-min-speedup): UPdate
6978         default values.
6979
6980 2018-02-12  Richard Biener  <rguenther@suse.de>
6981
6982         PR tree-optimization/84037
6983         * tree-vect-slp.c (vect_build_slp_tree_2): Try swapping the
6984         matched stmts if we cannot swap the non-matched ones.
6985
6986 2018-02-12  Olga Makhotina  <olga.makhotina@intel.com>
6987
6988         * config/i386/avx512fintrin.h (_mm_mask_scalef_round_sd,
6989         _mm_maskz_scalef_round_sd, _mm_mask_scalef_round_ss,
6990         _mm_maskz_scalef_round_ss): New intrinsics.
6991         (__builtin_ia32_scalefsd_round, __builtin_ia32_scalefss_round): Fix.
6992         * config/i386/i386-builtin.def (__builtin_ia32_scalefsd_round,
6993         __builtin_ia32_scalefss_round): Remove.
6994         (__builtin_ia32_scalefsd_mask_round,
6995         __builtin_ia32_scalefss_mask_round): New intrinsics.
6996         * config/i386/sse.md (vmscalef<mode><round_name>): Renamed to ...
6997         (vmscalef<mode><mask_scalar_name><round_scalar_name>): ... this.
6998         ((match_operand:VF_128 2 "<round_nimm_predicate>"
6999         "<round_constraint>")): Changed to ...
7000         ((match_operand:VF_128 2 "<round_scalar_nimm_predicate>"
7001         "<round_scalar_constraint>")): ... this.
7002         ("vscalef<ssescalarmodesuffix>\t{<round_op3>%2, %1, %0|
7003         %0, %1, %2<round_op3>}"): Changed to ...
7004         ("vscalef<ssescalarmodesuffix>\t{<round_scalar_mask_op3>%2, %1,
7005         %0<mask_scalar_operand3>|%0<mask_scalar_operand3>, %1,
7006         %2<round_scalar_mask_op3>}"): ... this.
7007         * config/i386/subst.md (round_scalar_nimm_predicate): New.
7008
7009 2018-02-12  Olga Makhotina  <olga.makhotina@intel.com>
7010
7011         * config/i386/avx512fintrin.h (_mm_mask_sqrt_round_sd)
7012         (_mm_maskz_sqrt_round_sd, _mm_mask_sqrt_round_ss)
7013         (_mm_maskz_sqrt_round_ss): New intrinsics.
7014         (__builtin_ia32_sqrtsd_round, __builtin_ia32_sqrtss_round): Remove.
7015         (__builtin_ia32_sqrtsd_mask_round)
7016         (__builtin_ia32_sqrtss_mask_round): New builtins.
7017         * config/i386/i386-builtin.def (__builtin_ia32_sqrtsd_round)
7018         (__builtin_ia32_sqrtss_round): Remove.
7019         (__builtin_ia32_sqrtsd_mask_round)
7020         (__builtin_ia32_sqrtss_mask_round): New builtins.
7021         * config/i386/sse.md (vmsqrt<mode>2<round_name>): Renamed to ...
7022         (vmsqrt<mode>2<mask_scalar_name><round_scalar_name>): ... this.
7023         ((match_operand:VF_128 1 "vector_operand"
7024         "xBm,<round_constraint>")): Changed to ...
7025         ((match_operand:VF_128 1 "vector_operand"
7026         "xBm,<round_scalar_constraint>")): ... this.
7027         (vsqrt<ssescalarmodesuffix>\t{<round_op3>%1, %2, %0|
7028         %0, %2, %<iptr>1<round_op3>}): Changed to ...
7029         (vsqrt<ssescalarmodesuffix>\t{<round_scalar_mask_op3>%1, %2,
7030         %0<mask_scalar_operand3>|%0<mask_scalar_operand3>, %2,
7031         %<iptr>1<round_scalar_mask_op3>}): ... this.
7032         ((set_attr "prefix" "<round_prefix>")): Changed to ...
7033         ((set_attr "prefix" "<round_scalar_prefix>")): ... this.
7034
7035 2018-02-11  Steven Munroe  <munroesj@gcc.gnu.org>
7036
7037         PR target/84266
7038         * config/rs6000/mmintrin.h (_mm_cmpeq_pi32 [_ARCH_PWR9]):
7039         Cast vec_cmpeq result to correct type.
7040         * config/rs6000/mmintrin.h (_mm_cmpgt_pi32 [_ARCH_PWR9]):
7041         Cast vec_cmpgt result to correct type.
7042
7043 2018-02-11  Alexandre Oliva  <aoliva@redhat.com>
7044
7045         * final.c (final_scan_insn_1): Renamed from...
7046         (final_scan_insn): ... this.  New wrapper, to recover
7047         seen from the outermost call in recursive ones.
7048         * config/sparc/sparc.c (output_return): Drop seen from call.
7049         (output_sibcall): Likewise.
7050         * config/visium/visium.c (output_branch): Likewise.
7051
7052 2018-02-10  John David Anglin  <danglin@gcc.gnu.org>
7053
7054         * config/pa/pa.c (hppa_profile_hook): Mark SYMBOL_REF for _mcount as
7055         function label.
7056
7057 2018-02-10  Alan Modra  <amodra@gmail.com>
7058
7059         PR target/84300
7060         * config/rs6000/rs6000.md (split_stack_return): Remove (use ..).
7061         Specify LR as an input.
7062
7063 2018-02-10  Jakub Jelinek  <jakub@redhat.com>
7064
7065         PR sanitizer/83987
7066         * omp-low.c (maybe_remove_omp_member_access_dummy_vars,
7067         remove_member_access_dummy_vars): New functions.
7068         (lower_omp_for, lower_omp_taskreg, lower_omp_target,
7069         lower_omp_1, execute_lower_omp): Use them.
7070
7071         PR rtl-optimization/84308
7072         * shrink-wrap.c (spread_components): Release todo vector.
7073
7074 2018-02-09  Vladimir Makarov  <vmakarov@redhat.com>
7075
7076         PR rtl-optimization/57193
7077         * ira-color.c (struct allocno_color_data): Add member
7078         conflict_allocno_hard_prefs.
7079         (update_conflict_allocno_hard_prefs): New.
7080         (bucket_allocno_compare_func): Add a preference based on
7081         conflict_allocno_hard_prefs.
7082         (push_allocno_to_stack): Update conflict_allocno_hard_prefs.
7083         (color_allocnos): Remove a dead code.  Initiate
7084         conflict_allocno_hard_prefs.  Call update_costs_from_prefs.
7085
7086 2018-02-09  Jakub Jelinek  <jakub@redhat.com>
7087
7088         PR target/84226
7089         * config/rs6000/vsx.md (p9_xxbrq_v16qi): Change input operand
7090         constraint from =wa to wa.  Avoid a subreg on the output operand,
7091         instead use a pseudo and subreg it in a move.
7092         (p9_xxbrd_<mode>): Changed to ...
7093         (p9_xxbrd_v2di): ... this insn, without VSX_D iterator.
7094         (p9_xxbrd_v2df): New expander.
7095         (p9_xxbrw_<mode>): Changed to ...
7096         (p9_xxbrw_v4si): ... this insn, without VSX_W iterator.
7097         (p9_xxbrw_v4sf): New expander.
7098
7099 2018-02-09  Sebastian Perta  <sebastian.perta@renesas.com>
7100
7101         * config/rx/rx.md (movsicc): Update expander to be matched by GCC.
7102
7103 2018-02-09  Peter Bergner  <bergner@vnet.ibm.com>
7104
7105         PR target/83926
7106         * config/rs6000/vsx.md (vsx_mul_v2di): Handle generating a 64-bit
7107         multiply in 32-bit mode.
7108         (vsx_div_v2di): Handle generating a 64-bit signed divide in 32-bit mode.
7109         (vsx_udiv_v2di): Handle generating a 64-bit unsigned divide in 32-bit
7110         mode.
7111
7112 2018-02-09  Sebastian Perta  <sebastian.perta@renesas.com>
7113
7114         * config/rx/constraints.md (CALL_OP_SYMBOL_REF): Added new constraint
7115         to allow or block "symbol_ref" depending on the value of TARGET_JSR.
7116         * config/rx/rx.md (call_internal): Use CALL_OP_SYMBOL_REF.
7117         * config/rx/rx.md (call_value_internal): Use CALL_OP_SYMBOL_REF.
7118
7119 2018-02-09  Pierre-Marie de Rodat  <derodat@adacore.com>
7120
7121         PR lto/84213
7122         * dwarf2out.c (is_trivial_indirect_ref): New function.
7123         (dwarf2out_late_global_decl): Do not generate a location
7124         attribute for variables that have a non-trivial DECL_VALUE_EXPR
7125         and that are not defined in the current unit.
7126
7127 2018-02-09  Eric Botcazou  <ebotcazou@adacore.com>
7128
7129         * optabs.c (prepare_cmp_insn): Try harder to emit a direct comparison
7130         instead of a libcall for UNORDERED.
7131
7132 2018-02-09  Tamar Christina  <tamar.christina@arm.com>
7133
7134         PR target/82641
7135         * config/arm/arm-c.c (arm_cpu_builtins): Un-define __ARM_FEATURE_LDREX,
7136         __ARM_ARCH_PROFILE, __ARM_ARCH_ISA_THUMB, __ARM_FP and __ARM_NEON_FP.
7137
7138 2018-02-09  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>
7139
7140         PR target/PR84295
7141         * config/s390/s390.c (s390_set_current_function): Invoke
7142         s390_indirect_branch_settings also if fndecl didn't change.
7143
7144 2018-02-09  Alexandre Oliva  <aoliva@redhat.com>
7145
7146         * config/rs6000/rs6000.md (blockage): Set length to zero.
7147
7148 2018-02-09  Eric Botcazou  <ebotcazou@adacore.com>
7149
7150         * expr.c (optimize_bitfield_assignment_op): Remove obsolete assertion.
7151
7152 2018-02-09  Jakub Jelinek  <jakub@redhat.com>
7153
7154         PR sanitizer/84285
7155         * gcc.c (STATIC_LIBASAN_LIBS, STATIC_LIBTSAN_LIBS,
7156         STATIC_LIBLSAN_LIBS, STATIC_LIBUBSAN_LIBS): Handle -static like
7157         -static-lib*san.
7158
7159         PR debug/84252
7160         * var-tracking.c (vt_add_function_parameter): Punt for non-onepart
7161         PARALLEL incoming that failed vt_get_decl_and_offset check.
7162
7163         PR middle-end/84237
7164         * output.h (bss_initializer_p): Add NAMED argument, defaulted to false.
7165         * varasm.c (bss_initializer_p): Add NAMED argument, if true, ignore
7166         TREE_READONLY bit.
7167         (get_variable_section): For decls in named .bss* sections pass true as
7168         second argument to bss_initializer_p.
7169
7170 2018-02-09  Marek Polacek  <polacek@redhat.com>
7171             Jakub Jelinek  <jakub@redhat.com>
7172
7173         PR c++/83659
7174         * fold-const.c (fold_indirect_ref_1): Use VECTOR_TYPE_P macro.
7175         Formatting fixes.  Verify first that tree_fits_poly_int64_p (op01).
7176         Sync some changes from cxx_fold_indirect_ref.
7177
7178 2018-02-09  Alexandre Oliva  <aoliva@redhat.com>
7179
7180         * cfgexpand.c (expand_gimple_basic_block): Handle inline entry
7181         markers.
7182         * dwarf2out.c (dwarf2_debug_hooks): Enable inline_entry hook.
7183         (BLOCK_INLINE_ENTRY_LABEL): New.
7184         (dwarf2out_var_location): Disregard inline entry markers.
7185         (inline_entry_data): New struct.
7186         (inline_entry_data_hasher): New hashtable type.
7187         (inline_entry_data_hasher::hash): New.
7188         (inline_entry_data_hasher::equal): New.
7189         (inline_entry_data_table): New variable.
7190         (add_high_low_attributes): Add DW_AT_entry_pc and
7191         DW_AT_GNU_entry_view attributes if a pending entry is found
7192         in inline_entry_data_table.  Add old entry_pc attribute only
7193         if debug nonbinding markers are disabled.
7194         (gen_inlined_subroutine_die): Set BLOCK_DIE if nonbinding
7195         markers are enabled.
7196         (block_within_block_p, dwarf2out_inline_entry): New.
7197         (dwarf2out_finish): Check that no entries remained in
7198         inline_entry_data_table.
7199         * final.c (reemit_insn_block_notes): Handle inline entry notes.
7200         (final_scan_insn, notice_source_line): Likewise.
7201         (rest_of_clean_state): Skip inline entry markers.
7202         * gimple-pretty-print.c (dump_gimple_debug): Handle inline entry
7203         markers.
7204         * gimple.c (gimple_build_debug_inline_entry): New.
7205         * gimple.h (enum gimple_debug_subcode): Add
7206         GIMPLE_DEBUG_INLINE_ENTRY.
7207         (gimple_build_debug_inline_entry): Declare.
7208         (gimple_debug_inline_entry_p): New.
7209         (gimple_debug_nonbind_marker_p): Adjust.
7210         * insn-notes.def (INLINE_ENTRY): New.
7211         * print-rtl.c (rtx_writer::print_rtx_operand_code_0): Handle
7212         inline entry marker notes.
7213         (print_insn): Likewise.
7214         * rtl.h (NOTE_MARKER_P): Add INLINE_ENTRY support.
7215         (INSN_DEBUG_MARKER_KIND): Likewise.
7216         (GEN_RTX_DEBUG_MARKER_INLINE_ENTRY_PAT): New.
7217         * tree-inline.c (expand_call_inline): Build and insert
7218         debug_inline_entry stmt.
7219         * tree-ssa-live.c (remove_unused_scope_block_p): Preserve
7220         inline entry blocks early, if nonbind markers are enabled.
7221         (dump_scope_block): Dump fragment info.
7222         * var-tracking.c (reemit_marker_as_note): Handle inline entry note.
7223         * doc/gimple.texi (gimple_debug_inline_entry_p): New.
7224         (gimple_build_debug_inline_entry): New.
7225         * doc/invoke.texi (gstatement-frontiers, gno-statement-frontiers):
7226         Enable/disable inline entry points too.
7227         * doc/rtl.texi (NOTE_INSN_INLINE_ENTRY): New.
7228         (DEBUG_INSN): Describe inline entry markers.
7229
7230         * common.opt (gvariable-location-views): New.
7231         (gvariable-location-views=incompat5): New.
7232         * config.in: Rebuilt.
7233         * configure: Rebuilt.
7234         * configure.ac: Test assembler for view support.
7235         * dwarf2asm.c (dw2_asm_output_symname_uleb128): New.
7236         * dwarf2asm.h (dw2_asm_output_symname_uleb128): Declare.
7237         * dwarf2out.c (var_loc_view): New typedef.
7238         (struct dw_loc_list_struct): Add vl_symbol, vbegin, vend.
7239         (dwarf2out_locviews_in_attribute): New.
7240         (dwarf2out_locviews_in_loclist): New.
7241         (dw_val_equal_p): Compare val_view_list of dw_val_class_view_lists.
7242         (enum dw_line_info_opcode): Add LI_adv_address.
7243         (struct dw_line_info_table): Add view.
7244         (RESET_NEXT_VIEW, RESETTING_VIEW_P): New macros.
7245         (DWARF2_ASM_VIEW_DEBUG_INFO): Define default.
7246         (zero_view_p): New variable.
7247         (ZERO_VIEW_P): New macro.
7248         (output_asm_line_debug_info): New.
7249         (struct var_loc_node): Add view.
7250         (add_AT_view_list, AT_loc_list): New.
7251         (add_var_loc_to_decl): Add view param.  Test it against last.
7252         (new_loc_list): Add view params.  Record them.
7253         (AT_loc_list_ptr): Handle loc and view lists.
7254         (view_list_to_loc_list_val_node): New.
7255         (print_dw_val): Handle dw_val_class_view_list.
7256         (size_of_die): Likewise.
7257         (value_format): Likewise.
7258         (loc_list_has_views): New.
7259         (gen_llsym): Set vl_symbol too.
7260         (maybe_gen_llsym, skip_loc_list_entry): New.
7261         (dwarf2out_maybe_output_loclist_view_pair): New.
7262         (output_loc_list): Output view list or entries too.
7263         (output_view_list_offset): New.
7264         (output_die): Handle dw_val_class_view_list.
7265         (output_dwarf_version): New.
7266         (output_compilation_unit_header): Use it.
7267         (output_skeleton_debug_sections): Likewise.
7268         (output_rnglists, output_line_info): Likewise.
7269         (output_pubnames, output_aranges): Update version comments.
7270         (output_one_line_info_table): Output view numbers in asm comments.
7271         (dw_loc_list): Determine current endview, pass it to new_loc_list.
7272         Call maybe_gen_llsym.
7273         (loc_list_from_tree_1): Adjust.
7274         (add_AT_location_description): Create view list attribute if
7275         needed, check it's absent otherwise.
7276         (convert_cfa_to_fb_loc_list): Adjust.
7277         (maybe_emit_file): Call output_asm_line_debug_info for test.
7278         (dwarf2out_var_location): Reset views as needed.  Precompute
7279         add_var_loc_to_decl args.  Call get_attr_min_length only if we have the
7280         attribute.  Set view.
7281         (new_line_info_table): Reset next view.
7282         (set_cur_line_info_table): Call output_asm_line_debug_info for test.
7283         (dwarf2out_source_line): Likewise.  Output view resets and labels to
7284         the assembler, or select appropriate line info opcodes.
7285         (prune_unused_types_walk_attribs): Handle dw_val_class_view_list.
7286         (optimize_string_length): Catch it.  Adjust.
7287         (resolve_addr): Copy vl_symbol along with ll_symbol.  Handle
7288         dw_val_class_view_list, and remove it if no longer needed.
7289         (hash_loc_list): Hash view numbers.
7290         (loc_list_hasher::equal): Compare them.
7291         (optimize_location_lists): Check whether a view list symbol is
7292         needed, and whether the locview attribute is present, and
7293         whether they match.  Remove the locview attribute if no longer
7294         needed.
7295         (index_location_lists): Call skip_loc_list_entry for test.
7296         (dwarf2out_finish): Call output_asm_line_debug_info for test.
7297         Use output_dwarf_version.
7298         * dwarf2out.h (enum dw_val_class): Add dw_val_class_view_list.
7299         (struct dw_val_node): Add val_view_list.
7300         * final.c (SEEN_NEXT_VIEW): New.
7301         (set_next_view_needed): New.
7302         (clear_next_view_needed): New.
7303         (maybe_output_next_view): New.
7304         (final_start_function): Rename to...
7305         (final_start_function_1): ... this.  Take pointer to FIRST,
7306         add SEEN parameter.  Emit param bindings in the initial view.
7307         (final_start_function): Reintroduce SEEN-less interface.
7308         (final): Rename to...
7309         (final_1): ... this.  Take SEEN parameter.  Output final pending
7310         next view at the end.
7311         (final): Reintroduce seen-less interface.
7312         (final_scan_insn): Output pending next view before switching
7313         sections or ending a block.  Mark the next view as needed when
7314         outputting variable locations.  Notify debug backend of section
7315         changes, and of location view changes.
7316         (rest_of_handle_final): Adjust.
7317         * toplev.c (process_options): Autodetect value for debug variable
7318         location views option.  Warn on incompat5 without -gdwarf-5.
7319         * doc/invoke.texi (gvariable-location-views): New.
7320         (gvariable-location-views=incompat5): New.
7321         (gno-variable-location-views): New.
7322
7323 2018-02-08  David Malcolm  <dmalcolm@redhat.com>
7324
7325         PR tree-optimization/84136
7326         * tree-cfg.c (find_taken_edge_computed_goto): Remove assertion
7327         that the result of find_edge is non-NULL.
7328
7329 2018-02-08  Sergey Shalnov  <sergey.shalnov@intel.com>
7330
7331         PR target/83008
7332         * config/i386/x86-tune-costs.h (skylake_cost): Fix cost of
7333         storing integer register in SImode.  Fix cost of 256 and 512
7334         byte aligned SSE register store.
7335
7336 2018-02-08  Sergey Shalnov  <sergey.shalnov@intel.com>
7337
7338         * config/i386/i386.c (ix86_multiplication_cost): Fix
7339         multiplication cost for TARGET_AVX512DQ.
7340
7341 2018-02-08  Marek Polacek  <polacek@redhat.com>
7342
7343         PR tree-optimization/84238
7344         * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Verify the result of
7345         get_range_strlen.
7346
7347 2018-02-08  Richard Sandiford  <richard.sandiford@linaro.org>
7348
7349         PR tree-optimization/84265
7350         * tree-vect-stmts.c (vectorizable_store): Don't treat
7351         VMAT_CONTIGUOUS accesses as grouped.
7352         (vectorizable_load): Likewise.
7353
7354 2018-02-08  Richard Sandiford  <richard.sandiford@linaro.org>
7355
7356         PR tree-optimization/81635
7357         * wide-int.h (wi::round_down_for_mask, wi::round_up_for_mask): Declare.
7358         * wide-int.cc (wi::round_down_for_mask, wi::round_up_for_mask)
7359         (test_round_for_mask): New functions.
7360         (wide_int_cc_tests): Call test_round_for_mask.
7361         * tree-vrp.h (intersect_range_with_nonzero_bits): Declare.
7362         * tree-vrp.c (intersect_range_with_nonzero_bits): New function.
7363         * tree-data-ref.c (split_constant_offset_1): Use it to refine the
7364         range returned by get_range_info.
7365
7366 2018-02-08  Jan Hubicka  <hubicka@ucw.cz>
7367
7368         PR ipa/81360
7369         * cgraph.h (symtab_node::output_to_lto_symbol_table_p): Declare
7370         * symtab.c: Include builtins.h
7371         (symtab_node::output_to_lto_symbol_table_p): Move here
7372         from lto-streamer-out.c:output_symbol_p.
7373         * lto-streamer-out.c (write_symbol): Turn early exit to assert.
7374         (output_symbol_p): Move all logic to symtab.c
7375         (produce_symtab): Update.
7376
7377 2018-02-08  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>
7378
7379         * config/s390/s390-opts.h (enum indirect_branch): Define.
7380         * config/s390/s390-protos.h (s390_return_addr_from_memory)
7381         (s390_indirect_branch_via_thunk)
7382         (s390_indirect_branch_via_inline_thunk): Add function prototypes.
7383         (enum s390_indirect_branch_type): Define.
7384         * config/s390/s390.c (struct s390_frame_layout, struct
7385         machine_function): Remove.
7386         (indirect_branch_prez10thunk_mask, indirect_branch_z10thunk_mask)
7387         (indirect_branch_table_label_no, indirect_branch_table_name):
7388         Define variables.
7389         (INDIRECT_BRANCH_NUM_OPTIONS): Define macro.
7390         (enum s390_indirect_branch_option): Define.
7391         (s390_return_addr_from_memory): New function.
7392         (s390_handle_string_attribute): New function.
7393         (s390_attribute_table): Add new attribute handler.
7394         (s390_execute_label): Handle UNSPEC_EXECUTE_JUMP patterns.
7395         (s390_indirect_branch_via_thunk): New function.
7396         (s390_indirect_branch_via_inline_thunk): New function.
7397         (s390_function_ok_for_sibcall): When jumping via thunk disallow
7398         sibling call optimization for non z10 compiles.
7399         (s390_emit_call): Force indirect branch target to be a single
7400         register.  Add r1 clobber for non-z10 compiles.
7401         (s390_emit_epilogue): Emit return jump via return_use expander.
7402         (s390_reorg): Handle JUMP_INSNs as execute targets.
7403         (s390_option_override_internal): Perform validity checks for the
7404         new command line options.
7405         (s390_indirect_branch_attrvalue): New function.
7406         (s390_indirect_branch_settings): New function.
7407         (s390_set_current_function): Invoke s390_indirect_branch_settings.
7408         (s390_output_indirect_thunk_function):  New function.
7409         (s390_code_end): Implement target hook.
7410         (s390_case_values_threshold): Implement target hook.
7411         (TARGET_ASM_CODE_END, TARGET_CASE_VALUES_THRESHOLD): Define target
7412         macros.
7413         * config/s390/s390.h (struct s390_frame_layout)
7414         (struct machine_function): Move here from s390.c.
7415         (TARGET_INDIRECT_BRANCH_NOBP_RET)
7416         (TARGET_INDIRECT_BRANCH_NOBP_JUMP)
7417         (TARGET_INDIRECT_BRANCH_NOBP_JUMP_THUNK)
7418         (TARGET_INDIRECT_BRANCH_NOBP_JUMP_INLINE_THUNK)
7419         (TARGET_INDIRECT_BRANCH_NOBP_CALL)
7420         (TARGET_DEFAULT_INDIRECT_BRANCH_TABLE)
7421         (TARGET_INDIRECT_BRANCH_THUNK_NAME_EXRL)
7422         (TARGET_INDIRECT_BRANCH_THUNK_NAME_EX)
7423         (TARGET_INDIRECT_BRANCH_TABLE): Define macros.
7424         * config/s390/s390.md (UNSPEC_EXECUTE_JUMP)
7425         (INDIRECT_BRANCH_THUNK_REGNUM): Define constants.
7426         (mnemonic attribute): Add values which aren't recognized
7427         automatically.
7428         ("*cjump_long", "*icjump_long", "*basr", "*basr_r"): Disable
7429         pattern for branch conversion.  Fix mnemonic attribute.
7430         ("*c<code>", "*sibcall_br", "*sibcall_value_br", "*return"): Emit
7431         indirect branch via thunk if requested.
7432         ("indirect_jump", "<code>"): Expand patterns for branch conversion.
7433         ("*indirect_jump"): Disable for branch conversion using out of
7434         line thunks.
7435         ("indirect_jump_via_thunk<mode>_z10")
7436         ("indirect_jump_via_thunk<mode>")
7437         ("indirect_jump_via_inlinethunk<mode>_z10")
7438         ("indirect_jump_via_inlinethunk<mode>", "*casesi_jump")
7439         ("casesi_jump_via_thunk<mode>_z10", "casesi_jump_via_thunk<mode>")
7440         ("casesi_jump_via_inlinethunk<mode>_z10")
7441         ("casesi_jump_via_inlinethunk<mode>", "*basr_via_thunk<mode>_z10")
7442         ("*basr_via_thunk<mode>", "*basr_r_via_thunk_z10")
7443         ("*basr_r_via_thunk", "return<mode>_prez10"): New pattern.
7444         ("*indirect2_jump"): Disable for branch conversion.
7445         ("casesi_jump"): Turn into expander and expand patterns for branch
7446         conversion.
7447         ("return_use"): New expander.
7448         ("*return"): Emit return via thunk and rename it to ...
7449         ("*return<mode>"): ... this one.
7450         * config/s390/s390.opt: Add new options and and enum for the
7451         option values.
7452
7453 2018-02-08  Richard Sandiford  <richard.sandiford@linaro.org>
7454
7455         * lra-constraints.c (match_reload): Unconditionally use
7456         gen_lowpart_SUBREG, rather than selecting between that
7457         and equivalent gen_rtx_SUBREG code.
7458
7459 2018-02-08  Richard Biener  <rguenther@suse.de>
7460
7461         PR tree-optimization/84233
7462         * tree-ssa-phiprop.c (propagate_with_phi): Use separate
7463         changed flag instead of boguously re-using phi_inserted.
7464
7465 2018-02-08  Martin Jambor  <mjambor@suse.cz>
7466
7467         * hsa-gen.c (get_symbol_for_decl): Set program allocation for
7468         static local variables.
7469
7470 2018-02-08  Richard Biener  <rguenther@suse.de>
7471
7472         PR tree-optimization/84278
7473         * tree-vect-stmts.c (vectorizable_store): When looking for
7474         smaller vector types to perform grouped strided loads/stores
7475         make sure the mode is supported by the target.
7476         (vectorizable_load): Likewise.
7477
7478 2018-02-08  Wilco Dijkstra  <wdijkstr@arm.com>
7479
7480         * config/aarch64/aarch64.c (aarch64_components_for_bb):
7481         Increase LDP/STP opportunities by adding adjacent callee-saves.
7482
7483 2018-02-08  Wilco Dijkstra  <wdijkstr@arm.com>
7484
7485         PR rtl-optimization/84068
7486         PR rtl-optimization/83459
7487         * haifa-sched.c (rank_for_schedule): Fix SCHED_PRESSURE_MODEL sorting.
7488
7489 2018-02-08  Aldy Hernandez  <aldyh@redhat.com>
7490
7491         PR tree-optimization/84224
7492         * gimple-ssa-warn-alloca.c (pass_walloca::execute): Remove assert.
7493         * calls.c (gimple_alloca_call_p): Only return TRUE when we have
7494         non-zero arguments.
7495
7496 2018-02-07  Iain Sandoe  <iain@codesourcery.com>
7497
7498         PR target/84113
7499         * config/rs6000/altivec.md (*restore_world): Remove LR use.
7500         * config/rs6000/predicates.md (restore_world_operation): Adjust op
7501         count, remove one USE.
7502
7503 2018-02-07  Michael Meissner  <meissner@linux.vnet.ibm.com>
7504
7505         * doc/install.texi (Configuration): Document the
7506         --with-long-double-format={ibm,ieee} PowerPC configuration
7507         options.
7508
7509         PR target/84154
7510         * config/rs6000/rs6000.md (fix_trunc<SFDF:mode><QHI:mode>2):
7511         Convert from define_expand to be define_insn_and_split.  Rework
7512         float/double/_Float128 conversions to QI/HI/SImode to work with
7513         both ISA 2.07 (power8) or ISA 3.0 (power9).  Fix regression where
7514         conversions to QI/HImode types did a store and then a load to
7515         truncate the value.  For conversions to VSX registers, don't split
7516         the insn, instead emit the code directly.  Use the code iterator
7517         any_fix to combine signed and unsigned conversions.
7518         (fix<uns>_trunc<SFDF:mode>si2_p8): Likewise.
7519         (fixuns_trunc<SFDF:mode><QHI:mode>2): Likewise.
7520         (fix_trunc<IEEE128:mode><QHI:mode>2): Likewise.
7521         (fix<uns>_trunc<SFDF:mode><QHI:mode>2): Likewise.
7522         (fix_<mode>di2_hw): Likewise.
7523         (fixuns_<mode>di2_hw): Likewise.
7524         (fix_<mode>si2_hw): Likewise.
7525         (fixuns_<mode>si2_hw): Likewise.
7526         (fix<uns>_<IEEE128:mode><SDI:mode>2_hw): Likewise.
7527         (fix<uns>_trunc<IEEE128:mode><QHI:mode>2): Likewise.
7528         (fctiw<u>z_<mode>_smallint): Rename fctiw<u>z_<mode>_smallint to
7529         fix<uns>_trunc<SFDF:mode>si2_p8.
7530         (fix_trunc<SFDF:mode><QHI:mode>2_internal): Delete, no longer
7531         used.
7532         (fixuns_trunc<SFDF:mode><QHI:mode>2_internal): Likewise.
7533         (fix<uns>_<mode>_mem): Likewise.
7534         (fctiw<u>z_<mode>_mem): Likewise.
7535         (fix<uns>_<mode>_mem): Likewise.
7536         (fix<uns>_trunc<SFDF:mode><QHSI:mode>2_mem): On ISA 3.0, prevent
7537         the register allocator from doing a direct move to the GPRs to do
7538         a store, and instead use the ISA 3.0 store byte/half-word from
7539         vector register instruction.  For IEEE 128-bit floating point,
7540         also optimize stores of 32-bit ints.
7541         (fix<uns>_trunc<IEEE128:mode><QHSI:mode>2_mem): Likewise.
7542
7543 2018-02-07  Alan Hayward  <alan.hayward@arm.com>
7544
7545         * genextract.c (push_pathstr_operand): New function to support
7546         [a-zA-Z].
7547         (walk_rtx): Call push_pathstr_operand.
7548         (print_path): Support [a-zA-Z].
7549
7550 2018-02-07  Richard Biener  <rguenther@suse.de>
7551
7552         PR tree-optimization/84037
7553         * tree-vectorizer.h (struct _loop_vec_info): Add ivexpr_map member.
7554         (cse_and_gimplify_to_preheader): Declare.
7555         (vect_get_place_in_interleaving_chain): Likewise.
7556         * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
7557         ivexpr_map.
7558         (_loop_vec_info::~_loop_vec_info): Delete it.
7559         (cse_and_gimplify_to_preheader): New function.
7560         * tree-vect-slp.c (vect_get_place_in_interleaving_chain): Export.
7561         * tree-vect-stmts.c (vectorizable_store): CSE base and steps.
7562         (vectorizable_load): Likewise.  For grouped stores always base
7563         the IV on the first element.
7564         * tree-vect-loop-manip.c (vect_loop_versioning): Unshare versioning
7565         condition before gimplifying.
7566
7567 2018-02-07  Jakub Jelinek  <jakub@redhat.com>
7568
7569         * tree-eh.c (operation_could_trap_helper_p): Ignore honor_trapv for
7570         *DIV_EXPR and *MOD_EXPR.
7571
7572 2018-02-07  H.J. Lu  <hongjiu.lu@intel.com>
7573
7574         PR target/84248
7575         * config/i386/i386.c (ix86_option_override_internal): Mask out
7576         the CF_SET bit when checking -fcf-protection.
7577
7578 2018-02-07  Tom de Vries  <tom@codesourcery.com>
7579
7580         PR libgomp/84217
7581         * omp-expand.c (expand_oacc_collapse_init): Ensure diff_type is large
7582         enough.
7583
7584 2018-02-07  Richard Biener  <rguenther@suse.de>
7585
7586         PR tree-optimization/84204
7587         * tree-chrec.c (chrec_fold_plus_1): Remove size limiting in
7588         this place.
7589
7590         PR tree-optimization/84205
7591         * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Also
7592         special-case isl_ast_op_zdiv_r.
7593
7594         PR tree-optimization/84223
7595         * graphite-scop-detection.c (gather_bbs::before_dom_children):
7596         Only add conditions from within the region.
7597         (gather_bbs::after_dom_children): Adjust.
7598
7599 2018-02-07  Georg-Johann Lay  <avr@gjlay.de>
7600
7601         PR target/84209
7602         * config/avr/avr.h (GENERAL_REGNO_P, GENERAL_REG_P): New macros.
7603         * config/avr/avr.md: Only post-reload split REG-REG moves if
7604         either register is GENERAL_REG_P.
7605
7606 2018-02-07  Jakub Jelinek  <jakub@redhat.com>
7607
7608         PR tree-optimization/84235
7609         * tree-ssa-scopedtables.c
7610         (avail_exprs_stack::simplify_binary_operation): Fir MINUS_EXPR, punt
7611         if the subtraction is performed in floating point type where NaNs are
7612         honored.  For *DIV_EXPR, punt for ALL_FRACT_MODE_Ps where we can't
7613         build 1.  Formatting fix.
7614
7615 2018-02-06  Jakub Jelinek  <jakub@redhat.com>
7616
7617         PR target/84146
7618         * config/i386/i386.c (rest_of_insert_endbranch): Only skip
7619         NOTE_INSN_CALL_ARG_LOCATION after a call, not anything else,
7620         and skip it regardless of bb boundaries.  Use CALL_P macro,
7621         don't test INSN_P (insn) together with CALL_P or JUMP_P check
7622         unnecessarily, formatting fix.
7623
7624 2018-02-06  Michael Collison  <michael.collison@arm.com>
7625
7626         * config/arm/thumb2.md:
7627         (*thumb2_mov_negscc): Split only if TARGET_THUMB2 && !arm_restrict_it.
7628         (*thumb_mov_notscc): Ditto.
7629
7630 2018-02-06  Michael Meissner  <meissner@linux.vnet.ibm.com>
7631
7632         PR target/84154
7633         * config/rs6000/rs6000.md (su code attribute): Use "u" for
7634         unsigned_fix, not "s".
7635
7636 2018-02-06  Rainer Orth  <ro@CeBiTec.Uni-Bielefeld.DE>
7637
7638         * configure.ac (gcc_fn_eh_frame_ro): New function.
7639         (gcc_cv_as_cfi_directive): Check both 32 and 64-bit assembler for
7640         correct .eh_frame permissions.
7641         * configure: Regenerate.
7642
7643 2018-02-06  Andrew Jenner  <andrew@codeourcery.com>
7644
7645         * doc/invoke.texi: Add section for the PowerPC SPE backend. Remove
7646         irrelevant options.
7647
7648 2018-02-06  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
7649
7650         * config/rs6000/rs6000.c (rs6000_option_override_internal):
7651         Display warning message for -mno-speculate-indirect-jumps.
7652
7653 2018-02-06  Andrew Jenner  <andrew@codesourcery.com>
7654
7655         * config/powerpcspe/powerpcspe.opt: (msimple-fpu, mfpu) Add
7656         Undocumented.
7657         * config/powerpcspe/sysv4.opt (mbit-align): Likewise.
7658
7659 2018-02-06  Aldy Hernandez  <aldyh@redhat.com>
7660
7661         PR tree-optimization/84225
7662         * tree-eh.c (find_trapping_overflow): Only call
7663         operation_no_trapping_overflow when ANY_INTEGRAL_TYPE_P.
7664
7665 2018-02-06  Igor Tsimbalist  <igor.v.tsimbalist@intel.com>
7666
7667         PR target/84145
7668         * config/i386/i386.c: Reimplement the check of possible options
7669         -mibt/-mshstk conbination. Change error messages.
7670         * doc/invoke.texi: Fix a typo: remove extra '='.
7671
7672 2018-02-06  Marek Polacek  <polacek@redhat.com>
7673
7674         PR tree-optimization/84228
7675         * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Skip debug statements.
7676
7677 2018-02-06  Tamar Christina  <tamar.christina@arm.com>
7678
7679         PR target/82641
7680         * config/arm/arm.c (arm_print_asm_arch_directives): Record already
7681         emitted arch directives.
7682         * config/arm/arm-c.c (arm_cpu_builtins): Undefine __ARM_ARCH and
7683         __ARM_FEATURE_COPROC before changing architectures.
7684
7685 2018-02-06  Richard Biener  <rguenther@suse.de>
7686
7687         * config/i386/i386.c (print_reg): Fix typo.
7688         (ix86_loop_unroll_adjust): Do not unroll beyond the original nunroll.
7689
7690 2018-02-06  Eric Botcazou  <ebotcazou@adacore.com>
7691
7692         * configure: Regenerate.
7693
7694 2018-02-05  Martin Sebor  <msebor@redhat.com>
7695
7696         PR tree-optimization/83369
7697         * tree-ssa-ccp.c (pass_post_ipa_warn::execute): Use %G to print
7698         inlining context.
7699
7700 2018-02-05  Martin Liska  <mliska@suse.cz>
7701
7702         * doc/invoke.texi: Cherry-pick upstream r323995.
7703
7704 2018-02-05  Richard Sandiford  <richard.sandiford@linaro.org>
7705
7706         * ira.c (ira_init_register_move_cost): Adjust comment.
7707
7708 2018-02-05  Martin Liska  <mliska@suse.cz>
7709
7710         PR gcov-profile/84137
7711         * doc/gcov.texi: Fix typo in documentation.
7712
7713 2018-02-05  Martin Liska  <mliska@suse.cz>
7714
7715         PR gcov-profile/83879
7716         * doc/gcov.texi: Document necessity of --dynamic-list-data when
7717         using dlopen functionality.
7718
7719 2018-02-05  Olga Makhotina  <olga.makhotina@intel.com>
7720
7721         * config/i386/avx512dqintrin.h (_mm_mask_range_sd, _mm_maskz_range_sd,
7722         _mm_mask_range_round_sd, _mm_maskz_range_round_sd, _mm_mask_range_ss,
7723         _mm_maskz_range_ss, _mm_mask_range_round_ss,
7724         _mm_maskz_range_round_ss): New intrinsics.
7725         (__builtin_ia32_rangesd128_round)
7726         (__builtin_ia32_rangess128_round): Remove.
7727         (__builtin_ia32_rangesd128_mask_round,
7728         __builtin_ia32_rangess128_mask_round): New builtins.
7729         * config/i386/i386-builtin.def (__builtin_ia32_rangesd128_round,
7730         __builtin_ia32_rangess128_round): Remove.
7731         (__builtin_ia32_rangesd128_mask_round,
7732         __builtin_ia32_rangess128_mask_round): New builtins.
7733         * config/i386/sse.md (ranges<mode><round_saeonly_name>): Renamed to ...
7734         (ranges<mode><mask_scalar_name><round_saeonly_scalar_name>): ... this.
7735         ((match_operand:VF_128 2 "<round_saeonly_nimm_predicate>"
7736         "<round_saeonly_constraint>")): Changed to ...
7737         ((match_operand:VF_128 2 "<round_saeonly_scalar_nimm_predicate>"
7738         "<round_saeonly_scalar_constraint>")): ... this.
7739         ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_op4>%2, %1, %0|
7740         %0, %1, %2<round_saeonly_op4>, %3}"): Changed to ...
7741         ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_scalar_mask_op4>%2,
7742         %1, %0<mask_scalar_operand4>|%0<mask_scalar_operand4>, %1,
7743         %2<round_saeonly_scalar_mask_op4>, %3}"): ... this.
7744
7745 2018-02-02  Andrew Jenner  <andrew@codesourcery.com>
7746
7747         * config/powerpcspe/powerpcspe.opt: Add Undocumented to irrelevant
7748         options.
7749         * config/powerpcspe/powerpcspe-tables.opt (rs6000_cpu_opt_value):
7750         Remove all values except native, 8540 and 8548.
7751
7752 2018-02-02  H.J. Lu  <hongjiu.lu@intel.com>
7753
7754         * config/i386/i386.c (ix86_output_function_return): Pass
7755         INVALID_REGNUM, instead of -1, as invalid register number to
7756         indirect_thunk_name and output_indirect_thunk.
7757
7758 2018-02-02  Julia Koval  <julia.koval@intel.com>
7759
7760         * config.gcc: Add -march=icelake.
7761         * config/i386/driver-i386.c (host_detect_local_cpu): Detect icelake.
7762         * config/i386/i386-c.c (ix86_target_macros_internal): Handle icelake.
7763         * config/i386/i386.c (processor_costs): Add m_ICELAKE.
7764         (PTA_ICELAKE, PTA_AVX512VNNI, PTA_GFNI, PTA_VAES, PTA_AVX512VBMI2,
7765         PTA_VPCLMULQDQ, PTA_RDPID, PTA_AVX512BITALG): New.
7766         (processor_target_table): Add icelake.
7767         (ix86_option_override_internal): Handle new PTAs.
7768         (get_builtin_code_for_version): Handle icelake.
7769         (M_INTEL_COREI7_ICELAKE): New.
7770         (fold_builtin_cpu): Handle icelake.
7771         * config/i386/i386.h (TARGET_ICELAKE, PROCESSOR_ICELAKE): New.
7772         * doc/invoke.texi: Add -march=icelake.
7773
7774 2018-02-02  Julia Koval  <julia.koval@intel.com>
7775
7776         * config/i386/i386.c (ix86_option_override_internal): Change flags type
7777         to wide_int_bitmask.
7778         * wide-int-bitmask.h: New.
7779
7780 2018-02-02  Igor Tsimbalist  <igor.v.tsimbalist@intel.com>
7781
7782         PR target/84066
7783         * config/i386/i386.md: Replace Pmode with word_mode in
7784         builtin_setjmp_setup and builtin_longjmp to support x32.
7785
7786 2018-02-01  Peter Bergner  <bergner@vnet.ibm.com>
7787
7788         PR target/56010
7789         PR target/83743
7790         * config/rs6000/driver-rs6000.c: #include "diagnostic.h".
7791         #include "opts.h".
7792         (rs6000_supported_cpu_names): New static variable.
7793         (linux_cpu_translation_table): Likewise.
7794         (elf_platform) <cpu>: Define new static variable and use it.
7795         Translate kernel AT_PLATFORM name to canonical name if needed.
7796         Error if platform name is unknown.
7797
7798 2018-02-01  Aldy Hernandez  <aldyh@redhat.com>
7799
7800         PR target/84089
7801         * config/pa/predicates.md (base14_operand): Handle E_VOIDmode.
7802
7803 2018-02-01  Jeff Law  <law@redhat.com>
7804
7805         PR target/84128
7806         * config/i386/i386.c (release_scratch_register_on_entry): Add new
7807         OFFSET and RELEASE_VIA_POP arguments.  Use SP+OFFSET to restore
7808         the scratch if RELEASE_VIA_POP is false.
7809         (ix86_adjust_stack_and_probe_stack_clash): Un-constify SIZE.
7810         If we have to save a temporary register, decrement SIZE appropriately.
7811         Pass new arguments to release_scratch_register_on_entry.
7812         (ix86_adjust_stack_and_probe): Likewise.
7813         (ix86_emit_probe_stack_range): Pass new arguments to
7814         release_scratch_register_on_entry.
7815
7816 2018-02-01  Uros Bizjak  <ubizjak@gmail.com>
7817
7818         PR rtl-optimization/84157
7819         * combine.c (change_zero_ext): Use REG_P predicate in
7820         front of HARD_REGISTER_P predicate.
7821
7822 2018-02-01  Georg-Johann Lay  <avr@gjlay.de>
7823
7824         * config/avr/avr.c (avr_option_override): Move disabling of
7825         -fdelete-null-pointer-checks to...
7826         * common/config/avr/avr-common.c (avr_option_optimization_table):
7827         ...here.
7828
7829 2018-02-01  Richard Sandiford  <richard.sandiford@linaro.org>
7830
7831         PR tree-optimization/81635
7832         * tree-data-ref.c (split_constant_offset_1): For types that
7833         wrap on overflow, try to use range info to prove that wrapping
7834         cannot occur.
7835
7836 2018-02-01  Renlin Li  <renlin.li@arm.com>
7837
7838         PR target/83370
7839         * config/aarch64/aarch64.c (aarch64_class_max_nregs): Handle
7840         TAILCALL_ADDR_REGS.
7841         (aarch64_register_move_cost): Likewise.
7842         * config/aarch64/aarch64.h (reg_class): Rename CALLER_SAVE_REGS to
7843         TAILCALL_ADDR_REGS.
7844         (REG_CLASS_NAMES): Likewise.
7845         (REG_CLASS_CONTENTS): Rename CALLER_SAVE_REGS to
7846         TAILCALL_ADDR_REGS. Remove IP registers.
7847         * config/aarch64/aarch64.md (Ucs): Update register constraint.
7848
7849 2018-02-01  Richard Biener  <rguenther@suse.de>
7850
7851         * domwalk.h (dom_walker::dom_walker): Add additional constructor
7852         for specifying RPO order and allow NULL for that.
7853         * domwalk.c (dom_walker::dom_walker): Likewise.
7854         (dom_walker::walk): Handle NULL RPO order.
7855         * tree-into-ssa.c (rewrite_dom_walker): Do not walk dom children
7856         in RPO order.
7857         (rewrite_update_dom_walker): Likewise.
7858         (mark_def_dom_walker): Likewise.
7859
7860 2018-02-01  Richard Sandiford  <richard.sandiford@linaro.org>
7861
7862         * config/aarch64/aarch64-protos.h (aarch64_split_sve_subreg_move)
7863         (aarch64_maybe_expand_sve_subreg_move): Declare.
7864         * config/aarch64/aarch64.md (UNSPEC_REV_SUBREG): New unspec.
7865         * config/aarch64/predicates.md (aarch64_any_register_operand): New
7866         predicate.
7867         * config/aarch64/aarch64-sve.md (mov<mode>): Optimize subreg moves
7868         that are semantically a reverse operation.
7869         (*aarch64_sve_mov<mode>_subreg_be): New pattern.
7870         * config/aarch64/aarch64.c (aarch64_maybe_expand_sve_subreg_move):
7871         (aarch64_replace_reg_mode, aarch64_split_sve_subreg_move): New
7872         functions.
7873         (aarch64_can_change_mode_class): For big-endian, forbid changes
7874         between two SVE modes if they have different element sizes.
7875
7876 2018-02-01  Richard Sandiford  <richard.sandiford@linaro.org>
7877
7878         * config/aarch64/aarch64.c (aarch64_expand_sve_const_vector): Prefer
7879         the TImode handling for big-endian targets.
7880
7881 2018-02-01  Richard Sandiford  <richard.sandiford@linaro.org>
7882
7883         * config/aarch64/aarch64-sve.md (sve_ld1rq): Replace with...
7884         (*sve_ld1rq<Vesize>): ... this new pattern.  Handle all element sizes,
7885         not just bytes.
7886         * config/aarch64/aarch64.c (aarch64_expand_sve_widened_duplicate):
7887         Remove BSWAP handing for big-endian targets and use the form of
7888         LD1RQ appropariate for the mode.
7889
7890 2018-02-01  Richard Sandiford  <richard.sandiford@linaro.org>
7891
7892         * config/aarch64/aarch64.c (aarch64_simd_valid_immediate): Handle
7893         all CONST_VECTOR_DUPLICATE_P vectors, not just those with a single
7894         duplicated element.
7895
7896 2018-02-01  Richard Sandiford  <richard.sandiford@linaro.org>
7897
7898         PR tearget/83845
7899         * config/aarch64/aarch64.c (aarch64_secondary_reload): Tighten
7900         check for operands that need to go through aarch64_sve_reload_be.
7901
7902 2018-02-01  Jakub Jelinek  <jakub@redhat.com>
7903
7904         PR tree-optimization/81661
7905         PR tree-optimization/84117
7906         * tree-eh.h (rewrite_to_non_trapping_overflow): Declare.
7907         * tree-eh.c: Include gimplify.h.
7908         (find_trapping_overflow, replace_trapping_overflow,
7909         rewrite_to_non_trapping_overflow): New functions.
7910         * tree-vect-loop.c: Include tree-eh.h.
7911         (vect_get_loop_niters): Use rewrite_to_non_trapping_overflow.
7912         * tree-data-ref.c: Include tree-eh.h.
7913         (get_segment_min_max): Use rewrite_to_non_trapping_overflow.
7914
7915 2018-01-31  Uros Bizjak  <ubizjak@gmail.com>
7916
7917         PR rtl-optimization/84123
7918         * combine.c (change_zero_ext): Check if hard register satisfies
7919         can_change_dest_mode before calling gen_lowpart_SUBREG.
7920
7921 2018-01-31  Vladimir Makarov  <vmakarov@redhat.com>
7922
7923         PR target/82444
7924         * ira.c (ira_init_register_move_cost): Remove assert.
7925
7926 2018-01-31  Eric Botcazou  <ebotcazou@adacore.com>
7927
7928         PR rtl-optimization/84071
7929         * doc/tm.texi.in (WORD_REGISTER_OPERATIONS): Add explicit case.
7930         * doc/tm.texi: Regenerate.
7931
7932 2018-01-31  Richard Biener  <rguenther@suse.de>
7933
7934         PR tree-optimization/84132
7935         * tree-data-ref.c (analyze_miv_subscript): Properly
7936         check whether evolution_function_is_affine_multivariate_p
7937         before calling gcd_of_steps_may_divide_p.
7938
7939 2018-01-31  Julia Koval  <julia.koval@intel.com>
7940
7941         PR target/83618
7942         * config/i386/i386.c (ix86_expand_builtin): Handle IX86_BUILTIN_RDPID.
7943         * config/i386/i386.md (rdpid_rex64) New.
7944         (rdpid): Make 32bit only.
7945
7946 2018-01-29  Aldy Hernandez  <aldyh@redhat.com>
7947
7948         PR lto/84105
7949         * tree-pretty-print.c (dump_generic_node): Handle a TYPE_NAME with
7950         an IDENTIFIER_NODE for FUNCTION_TYPE's.
7951
7952 2018-01-31  Eric Botcazou  <ebotcazou@adacore.com>
7953
7954         Revert
7955         2018-01-12  Eric Botcazou  <ebotcazou@adacore.com>
7956
7957         * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
7958
7959 2018-01-31  Eric Botcazou  <ebotcazou@adacore.com>
7960
7961         PR rtl-optimization/84071
7962         * combine.c (record_dead_and_set_regs_1): Record the source unmodified
7963         for a paradoxical SUBREG on a WORD_REGISTER_OPERATIONS target.
7964
7965 2018-01-31  Claudiu Zissulescu  <claziss@synopsys.com>
7966
7967         * config/arc/arc.c (arc_handle_aux_attribute): New function.
7968         (arc_attribute_table): Add 'aux' attribute.
7969         (arc_in_small_data_p): Consider aux like variables.
7970         (arc_is_aux_reg_p): New function.
7971         (arc_asm_output_aligned_decl_local): Ignore 'aux' like variables.
7972         (arc_get_aux_arg): New function.
7973         (prepare_move_operands): Handle aux-register access.
7974         (arc_handle_aux_attribute): New function.
7975         * doc/extend.texi (ARC Variable attributes): Add subsection.
7976
7977 2018-01-31  Claudiu Zissulescu  <claziss@synopsys.com>
7978
7979         * config/arc/arc-protos.h (arc_is_uncached_mem_p): Function proto.
7980         * config/arc/arc.c (arc_handle_uncached_attribute): New function.
7981         (arc_attribute_table): Add 'uncached' attribute.
7982         (arc_print_operand): Print '.di' flag for uncached memory
7983         accesses.
7984         (arc_in_small_data_p): Do not consider for small data the uncached
7985         types.
7986         (arc_is_uncached_mem_p): New function.
7987         * config/arc/predicates.md (compact_store_memory_operand): Check
7988         for uncached memory accesses.
7989         (nonvol_nonimm_operand): Likewise.
7990         * gcc/doc/extend.texi (ARC Type Attribute): New subsection.
7991
7992 2018-01-31  Jakub Jelinek  <jakub@redhat.com>
7993
7994         PR c/84100
7995         * common.opt (falign-functions=, falign-jumps=, falign-labels=,
7996         falign-loops=): Add Optimization flag.
7997
7998 2018-01-30  Jeff Law  <law@redhat.com>
7999
8000         PR target/84064
8001         * i386.c (ix86_adjust_stack_and_probe_stack_clash): New argument
8002         INT_REGISTERS_SAVED.  Check it prior to calling
8003         get_scratch_register_on_entry.
8004         (ix86_adjust_stack_and_probe): Similarly.
8005         (ix86_emit_probe_stack_range): Similarly.
8006         (ix86_expand_prologue): Corresponding changes.
8007
8008 2018-01-30  Rainer Orth  <ro@CeBiTec.Uni-Bielefeld.DE>
8009
8010         PR target/40411
8011         * config/sol2.h (STARTFILE_ARCH_SPEC): Use -std=c*,
8012         -std=iso9899:199409 instead of -pedantic to select values-Xc.o.
8013
8014 2018-01-30  Vladimir Makarov  <vmakarov@redhat.com>
8015
8016         PR target/84112
8017         * lra-constraints.c (curr_insn_transform): Process AND in the
8018         address.
8019
8020 2018-01-30  Jakub Jelinek  <jakub@redhat.com>
8021
8022         PR rtl-optimization/83986
8023         * sched-deps.c (sched_analyze_insn): For frame related insns, add anti
8024         dependence against last_pending_memory_flush in addition to
8025         pending_jump_insns.
8026
8027 2018-01-30  Alexandre Oliva  <aoliva@redhat.com>
8028
8029         PR tree-optimization/81611
8030         * tree-ssa-dom.c (simple_iv_increment_p): Skip intervening
8031         copies.
8032
8033 2018-01-30  Aaron Sawdey  <acsawdey@linux.vnet.ibm.com>
8034
8035         PR target/83758
8036         * config/rs6000/rs6000.c (rs6000_internal_arg_pointer): Only return
8037         a reg rtx.
8038
8039 2018-01-30  Richard Biener  <rguenther@suse.de>
8040             Jakub Jelinek  <jakub@redhat.com>
8041
8042         PR tree-optimization/84111
8043         * tree-ssa-loop-ivcanon.c (tree_unroll_loops_completely_1): Skip
8044         inner loops added during recursion, as they don't have up-to-date
8045         SSA form.
8046
8047 2018-01-30  Jan Hubicka  <hubicka@ucw.cz>
8048
8049         PR ipa/81360
8050         * ipa-inline.c (can_inline_edge_p): Break out late tests to...
8051         (can_inline_edge_by_limits_p): ... here.
8052         (can_early_inline_edge_p, check_callers,
8053         update_caller_keys, update_callee_keys, recursive_inlining,
8054         add_new_edges_to_heap, speculation_useful_p,
8055         inline_small_functions,
8056         inline_small_functions, flatten_function,
8057         inline_to_all_callers_1): Update.
8058
8059 2018-01-30  Jan Hubicka  <hubicka@ucw.cz>
8060
8061         * profile-count.c (profile_count::combine_with_ipa_count): Handle
8062         zeros correctly.
8063
8064 2018-01-30  Richard Biener  <rguenther@suse.de>
8065
8066         PR tree-optimization/83008
8067         * tree-vect-slp.c (vect_analyze_slp_cost_1): Properly cost
8068         invariant and constant vector uses in stmts when they need
8069         more than one stmt.
8070
8071 2018-01-30  Rainer Orth  <ro@CeBiTec.Uni-Bielefeld.DE>
8072
8073         PR bootstrap/84017
8074         * configure.ac (gcc_cv_as_shf_merge): Disable on Solaris 10/x86.
8075         * configure: Regenerate.
8076
8077 2018-01-30  Richard Sandiford  <richard.sandiford@linaro.org>
8078
8079         * config/aarch64/aarch64-sve.md (*vec_extract<mode><Vel>_0): New
8080         pattern.
8081         (*vec_extract<mode><Vel>_v128): Require a nonzero lane number.
8082         Use gen_rtx_REG rather than gen_lowpart.
8083
8084 2018-01-30  Richard Sandiford  <richard.sandiford@linaro.org>
8085
8086         * lra-constraints.c (match_reload): Use subreg_lowpart_offset
8087         rather than 0 when creating partial subregs.
8088
8089 2018-01-30  Richard Sandiford  <richard.sandiford@linaro.org>
8090
8091         * vec-perm-indices.c (vec_perm_indices::series_p): Give examples
8092         of usage.
8093
8094 2018-01-29  Michael Meissner  <meissner@linux.vnet.ibm.com>
8095
8096         PR target/81550
8097         * config/rs6000/rs6000.c (rs6000_setup_reg_addr_masks): If DFmode
8098         and SFmode can go in Altivec registers (-mcpu=power7 for DFmode,
8099         -mcpu=power8 for SFmode) don't set the PRE_INCDEC or PRE_MODIFY
8100         flags.  This restores the settings used before the 2017-07-24.
8101         Turning off pre increment/decrement/modify allows IVOPTS to
8102         optimize DF/SF loops where the index is an int.
8103
8104 2018-01-29  Richard Biener  <rguenther@suse.de>
8105             Kelvin Nilsen  <kelvin@gcc.gnu.org>
8106
8107         PR bootstrap/80867
8108         * tree-vect-stmts.c (vectorizable_call): Don't call
8109         targetm.vectorize_builtin_md_vectorized_function if callee is
8110         NULL.
8111
8112 2018-01-22  Carl Love  <cel@us.ibm.com>
8113
8114         * doc/extend.tex: Fix typo in second arg in
8115         __builtin_bcdadd_{lt|eq|gt|ov} and __builtin_bcdsub_{lt|eq|gt|ov}.
8116
8117 2018-01-29  Richard Biener  <rguenther@suse.de>
8118
8119         PR tree-optimization/84086
8120         * tree-ssanames.c: Include cfgloop.h and tree-scalar-evolution.h.
8121         (flush_ssaname_freelist): When SSA names were released reset
8122         the SCEV hash table.
8123
8124 2018-01-29  Richard Biener  <rguenther@suse.de>
8125
8126         PR tree-optimization/84057
8127         * tree-ssa-loop-ivcanon.c (unloop_loops): Deal with already
8128         removed paths when removing edges.
8129
8130 2018-01-27  H.J. Lu  <hongjiu.lu@intel.com>
8131
8132         * doc/invoke.texi: Replace -mfunction-return==@var{choice} with
8133         -mfunction-return=@var{choice}.
8134
8135 2018-01-27  Bernd Edlinger  <bernd.edlinger@hotmail.de>
8136
8137         PR diagnostic/84034
8138         * diagnostic-show-locus.c (get_line_width_without_trailing_whitespace):
8139         Handle CR like TAB.
8140         (layout::print_source_line): Likewise.
8141         (test_get_line_width_without_trailing_whitespace): Add test cases.
8142
8143 2018-01-27  Jakub Jelinek  <jakub@redhat.com>
8144
8145         PR middle-end/84040
8146         * sched-deps.c (sched_macro_fuse_insns): Return immediately for
8147         debug insns.
8148
8149 2018-01-26  Jim Wilson  <jimw@sifive.com>
8150
8151         * config/riscv/riscv.h (MAX_FIXED_MODE_SIZE): New.
8152
8153         * config/riscv/elf.h (LIB_SPEC): Don't include -lgloss when nosys.specs
8154         specified.
8155
8156 2018-01-26  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
8157
8158         * config/aarch64/aarch64.md: Add peepholes for CMP + SUB -> SUBS
8159         and CMP + SUB-immediate -> SUBS.
8160
8161 2018-01-26  Martin Sebor  <msebor@redhat.com>
8162
8163         PR tree-optimization/83896
8164         * tree-ssa-strlen.c (get_string_len): Rename...
8165         (get_string_cst_length): ...to this.  Return HOST_WIDE_INT.
8166         Avoid assuming length is constant.
8167         (handle_char_store): Use HOST_WIDE_INT for string length.
8168
8169 2018-01-26  Uros Bizjak  <ubizjak@gmail.com>
8170
8171         PR target/81763
8172         * config/i386/i386.md (*andndi3_doubleword): Add earlyclobber
8173         to (=&r,r,rm) alternative. Add (=r,0,rm) and (=r,r,0) alternatives.
8174
8175 2018-01-26  Richard Biener  <rguenther@suse.de>
8176
8177         PR rtl-optimization/84003
8178         * dse.c (record_store): Only record redundant stores when
8179         the earlier store aliases at least all accesses the later one does.
8180
8181 2018-01-26  Jakub Jelinek  <jakub@redhat.com>
8182
8183         PR rtl-optimization/83985
8184         * dce.c (deletable_insn_p): Return false for separate shrink wrapping
8185         REG_CFA_RESTORE insns.
8186         (delete_unmarked_insns): Don't ignore separate shrink wrapping
8187         REG_CFA_RESTORE insns here.
8188
8189         PR c/83989
8190         * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Don't
8191         use SSA_NAME_VAR as base for SSA_NAMEs with non-NULL SSA_NAME_VAR.
8192
8193 2018-01-26  Claudiu Zissulescu  <claziss@synopsys.com>
8194
8195         * config/arc/arc-arch.h (arc_tune_attr): Add ARC_TUNE_CORE_3.
8196         * config/arc/arc.c (arc_sched_issue_rate): Use ARC_TUNE_... .
8197         (arc_init): Likewise.
8198         (arc_override_options): Likewise.
8199         (arc_file_start): Choose Tag_ARC_CPU_variation based on arc_tune
8200         value.
8201         (hwloop_fail): Use TARGET_DBNZ when we want to check for dbnz insn
8202         support.
8203         * config/arc/arc.h (TARGET_DBNZ): Define.
8204         * config/arc/arc.md (attr tune): Add core_3, use ARC_TUNE_... to
8205         properly set the tune attribute.
8206         (dbnz): Use TARGET_DBNZ guard.
8207         * config/arc/arc.opt (mtune): Add core3 option.
8208
8209 2018-01-26  Claudiu Zissulescu  <claziss@synopsys.com>
8210
8211         * config/arc/arc.c (arc_delegitimize_address_0): Refactored to
8212         recognize new pic like addresses.
8213         (arc_delegitimize_address): Clean up.
8214
8215 2018-01-26  Claudiu Zissulescu  <claziss@synopsys.com>
8216
8217         * config/arc/arc-arches.def: Option mrf16 valid for all
8218         architectures.
8219         * config/arc/arc-c.def (__ARC_RF16__): New predefined macro.
8220         * config/arc/arc-cpus.def (em_mini): New cpu with rf16 on.
8221         * config/arc/arc-options.def (FL_RF16): Add mrf16 option.
8222         * config/arc/arc-tables.opt: Regenerate.
8223         * config/arc/arc.c (arc_conditional_register_usage): Handle
8224         reduced register file case.
8225         (arc_file_start): Set must have build attributes.
8226         * config/arc/arc.h (MAX_ARC_PARM_REGS): Conditional define using
8227         mrf16 option value.
8228         * config/arc/arc.opt (mrf16): Add new option.
8229         * config/arc/elf.h (ATTRIBUTE_PCS): Define.
8230         * config/arc/genmultilib.awk: Handle new mrf16 option.
8231         * config/arc/linux.h (ATTRIBUTE_PCS): Define.
8232         * config/arc/t-multilib: Regenerate.
8233         * doc/invoke.texi (ARC Options): Document mrf16 option.
8234
8235 2018-01-26  Claudiu Zissulescu  <claziss@synopsys.com>
8236
8237         * config/arc/arc-protos.h: Add arc_is_secure_call_p proto.
8238         * config/arc/arc.c (arc_handle_secure_attribute): New function.
8239         (arc_attribute_table): Add 'secure_call' attribute.
8240         (arc_print_operand): Print secure call operand.
8241         (arc_function_ok_for_sibcall): Don't optimize tail calls when
8242         secure.
8243         (arc_is_secure_call_p): New function.  * config/arc/arc.md
8244         (call_i): Add support for sjli instruction.
8245         (call_value_i): Likewise.
8246         * config/arc/constraints.md (Csc): New constraint.
8247
8248 2018-01-26  Claudiu Zissulescu  <claziss@synopsys.com>
8249             John Eric Martin  <John.Martin@emmicro-us.com>
8250
8251         * config/arc/arc-protos.h: Add arc_is_jli_call_p proto.
8252         * config/arc/arc.c (_arc_jli_section): New struct.
8253         (arc_jli_section): New type.
8254         (rc_jli_sections): New static variable.
8255         (arc_handle_jli_attribute): New function.
8256         (arc_attribute_table): Add jli_always and jli_fixed attribute.
8257         (arc_file_end): New function.
8258         (TARGET_ASM_FILE_END): Define.
8259         (arc_print_operand): Reuse 'S' letter for JLI output instruction.
8260         (arc_add_jli_section): New function.
8261         (jli_call_scan): Likewise.
8262         (arc_reorg): Call jli_call_scan.
8263         (arc_output_addsi): Remove 'S' from printing asm operand.
8264         (arc_is_jli_call_p): New function.
8265         * config/arc/arc.md (movqi_insn): Remove 'S' from printing asm
8266         operand.
8267         (movhi_insn): Likewise.
8268         (movsi_insn): Likewise.
8269         (movsi_set_cc_insn): Likewise.
8270         (loadqi_update): Likewise.
8271         (load_zeroextendqisi_update): Likewise.
8272         (load_signextendqisi_update): Likewise.
8273         (loadhi_update): Likewise.
8274         (load_zeroextendhisi_update): Likewise.
8275         (load_signextendhisi_update): Likewise.
8276         (loadsi_update): Likewise.
8277         (loadsf_update): Likewise.
8278         (movsicc_insn): Likewise.
8279         (bset_insn): Likewise.
8280         (bxor_insn): Likewise.
8281         (bclr_insn): Likewise.
8282         (bmsk_insn): Likewise.
8283         (bicsi3_insn): Likewise.
8284         (cmpsi_cc_c_insn): Likewise.
8285         (movsi_ne): Likewise.
8286         (movsi_cond_exec): Likewise.
8287         (clrsbsi2): Likewise.
8288         (norm_f): Likewise.
8289         (normw): Likewise.
8290         (swap): Likewise.
8291         (divaw): Likewise.
8292         (flag): Likewise.
8293         (sr): Likewise.
8294         (kflag): Likewise.
8295         (ffs): Likewise.
8296         (ffs_f): Likewise.
8297         (fls): Likewise.
8298         (call_i): Remove 'S' asm letter, add jli instruction.
8299         (call_value_i): Likewise.
8300         * config/arc/arc.op (mjli-always): New option.
8301         * config/arc/constraints.md (Cji): New constraint.
8302         * config/arc/fpx.md (addsf3_fpx): Remove 'S' from printing asm
8303         operand.
8304         (subsf3_fpx): Likewise.
8305         (mulsf3_fpx): Likewise.
8306         * config/arc/simdext.md (vendrec_insn): Remove 'S' from printing
8307         asm operand.
8308         * doc/extend.texi (ARC): Document 'jli-always' and 'jli-fixed'
8309         function attrbutes.
8310         * doc/invoke.texi (ARC): Document mjli-always option.
8311
8312 2018-01-26  Sebastian Perta  <sebastian.perta@renesas.com>
8313
8314         * config/rl78/rl78.c (rl78_addsi3_internal): If operand 2 is const
8315         avoid addition with 0 and use incw and decw where possible.
8316
8317 2018-01-26  Richard Biener  <rguenther@suse.de>
8318
8319         PR tree-optimization/81082
8320         * fold-const.c (fold_plusminus_mult_expr): Do not perform the
8321         association if it requires casting to unsigned.
8322         * match.pd ((A * C) +- (B * C) -> (A+-B)): New patterns derived
8323         from fold_plusminus_mult_expr to catch important cases late when
8324         range info is available.
8325
8326 2018-01-26  Rainer Orth  <ro@CeBiTec.Uni-Bielefeld.DE>
8327
8328         * config/i386/sol2.h (USE_HIDDEN_LINKONCE): Remove.
8329         * configure.ac (hidden_linkonce): New test.
8330         * configure: Regenerate.
8331         * config.in: Regenerate.
8332
8333 2018-01-26  Julia Koval  <julia.koval@intel.com>
8334
8335         * config/i386/avx512bitalgintrin.h (_mm512_bitshuffle_epi64_mask,
8336         _mm512_mask_bitshuffle_epi64_mask, _mm256_bitshuffle_epi64_mask,
8337         _mm256_mask_bitshuffle_epi64_mask, _mm_bitshuffle_epi64_mask,
8338         _mm_mask_bitshuffle_epi64_mask): Fix type.
8339         * config/i386/i386-builtin-types.def (UHI_FTYPE_V2DI_V2DI_UHI,
8340         USI_FTYPE_V4DI_V4DI_USI): Remove.
8341         * config/i386/i386-builtin.def (__builtin_ia32_vpshufbitqmb512_mask,
8342         __builtin_ia32_vpshufbitqmb256_mask,
8343         __builtin_ia32_vpshufbitqmb128_mask): Fix types.
8344         * config/i386/i386.c (ix86_expand_args_builtin): Remove old types.
8345         * config/i386/sse.md (VI1_AVX512VLBW): Change types.
8346
8347 2018-01-26  Alan Modra  <amodra@gmail.com>
8348
8349         PR target/84033
8350         * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Exclude
8351         UNSPEC_VBPERMQ.  Sort other unspecs.
8352
8353 2018-01-25  David Edelsohn  <dje.gcc@gmail.com>
8354
8355         * doc/invoke.texi (PowerPC Options): Document 'native' cpu type.
8356
8357 2018-01-25  Jan Hubicka  <hubicka@ucw.cz>
8358
8359         PR middle-end/83055
8360         * predict.c (drop_profile): Do not push/pop cfun; update also
8361         node->count.
8362         (handle_missing_profiles): Fix logic looking for zero profiles.
8363
8364 2018-01-25  Jakub Jelinek  <jakub@redhat.com>
8365
8366         PR middle-end/83977
8367         * ipa-fnsummary.c (compute_fn_summary): Clear can_change_signature
8368         on functions with #pragma omp declare simd or functions with simd
8369         attribute.
8370         * omp-simd-clone.c (expand_simd_clones): Revert 2018-01-24 change.
8371         * config/i386/i386.c (ix86_simd_clone_compute_vecsize_and_simdlen):
8372         Remove trailing \n from warning_at calls.
8373
8374 2018-01-25  Tom de Vries  <tom@codesourcery.com>
8375
8376         PR target/84028
8377         * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
8378         for neutered workers.
8379
8380 2018-01-24  Joseph Myers  <joseph@codesourcery.com>
8381
8382         PR target/68467
8383         * config/m68k/m68k.c (m68k_promote_function_mode): New function.
8384         (TARGET_PROMOTE_FUNCTION_MODE): New macro.
8385
8386 2018-01-24  Jeff Law  <law@redhat.com>
8387
8388         PR target/83994
8389         * i386.c (get_probe_interval): Move to earlier point.
8390         (ix86_compute_frame_layout): If -fstack-clash-protection and
8391         the frame is larger than the probe interval, then use pushes
8392         to save registers rather than reg->mem moves.
8393         (ix86_expand_prologue): Remove conditional for int_registers_saved
8394         assertion.
8395
8396 2018-01-24  Vladimir Makarov  <vmakarov@redhat.com>
8397
8398         PR target/84014
8399         * ira-build.c (setup_min_max_allocno_live_range_point): Set up
8400         min/max for never referenced object.
8401
8402 2018-01-24  Jakub Jelinek  <jakub@redhat.com>
8403
8404         PR middle-end/83977
8405         * tree.c (free_lang_data_in_decl): Don't clear DECL_ABSTRACT_ORIGIN
8406         here.
8407         * omp-low.c (create_omp_child_function): Remove "omp declare simd"
8408         attributes from DECL_ATTRIBUTES (decl) without affecting
8409         DECL_ATTRIBUTES (current_function_decl).
8410         * omp-simd-clone.c (expand_simd_clones): Ignore DECL_ARTIFICIAL
8411         functions with non-NULL DECL_ABSTRACT_ORIGIN.
8412
8413 2018-01-24  Richard Sandiford  <richard.sandiford@linaro.org>
8414
8415         PR tree-optimization/83979
8416         * fold-const.c (fold_comparison): Use constant_boolean_node
8417         instead of boolean_{true,false}_node.
8418
8419 2018-01-24  Jan Hubicka  <hubicka@ucw.cz>
8420
8421         * ipa-profile.c (ipa_propagate_frequency_1): Fix logic skipping calls
8422         with zero counts.
8423
8424 2018-01-24  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
8425
8426         * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
8427         Simplify the clause that sets the length attribute.
8428         (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
8429         (*sibcall_nonlocal_sysv<mode>): Clean up code block; simplify the
8430         clause that sets the length attribute.
8431         (*sibcall_value_nonlocal_sysv<mode>): Likewise.
8432
8433 2018-01-24  Tom de Vries  <tom@codesourcery.com>
8434
8435         PR target/83589
8436         * config/nvptx/nvptx.c (WORKAROUND_PTXJIT_BUG_2): Define to 1.
8437         (nvptx_pc_set, nvptx_condjump_label): New function. Copy from jump.c.
8438         Add strict parameter.
8439         (prevent_branch_around_nothing): Insert dummy insn between branch to
8440         label and label with no ptx insn inbetween.
8441         * config/nvptx/nvptx.md (define_insn "fake_nop"): New insn.
8442
8443 2018-01-24  Tom de Vries  <tom@codesourcery.com>
8444
8445         PR target/81352
8446         * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
8447         for neutered threads in warp.
8448         * config/nvptx/nvptx.md (define_insn "exit"): New insn.
8449
8450 2018-01-24  Richard Biener  <rguenther@suse.de>
8451
8452         PR tree-optimization/83176
8453         * tree-chrec.c (chrec_fold_plus_1): Handle (signed T){(T) .. }
8454         operands.
8455
8456 2018-01-24  Richard Biener  <rguenther@suse.de>
8457
8458         PR tree-optimization/82819
8459         * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Avoid
8460         code generating pluses that are no-ops in the target precision.
8461
8462 2018-01-24  Richard Biener  <rguenther@suse.de>
8463
8464         PR middle-end/84000
8465         * tree-cfg.c (replace_loop_annotate): Handle annot_expr_parallel_kind.
8466
8467 2018-01-23  Jan Hubicka  <hubicka@ucw.cz>
8468
8469         * cfgcleanup.c (try_crossjump_to_edge): Use combine_with_count
8470         to merge probabilities.
8471         * predict.c (probably_never_executed): Also mark as cold functions
8472         with global 0 profile and guessed local profile.
8473         * profile-count.c (profile_probability::combine_with_count): New
8474         member function.
8475         * profile-count.h (profile_probability::operator*,
8476         profile_probability::operator*=, profile_probability::operator/,
8477         profile_probability::operator/=): Reduce precision to adjusted
8478         and set value to guessed on contradictory divisions.
8479         (profile_probability::combine_with_freq): Remove.
8480         (profile_probability::combine_wiht_count): Declare.
8481         (profile_count::force_nonzero):: Set to adjusted.
8482         (profile_count::probability_in):: Set quality to adjusted.
8483         * tree-ssa-tail-merge.c (replace_block_by): Use
8484         combine_with_count.
8485
8486 2018-01-23  Andrew Waterman  <andrew@sifive.com>
8487             Jim Wilson  <jimw@sifive.com>
8488
8489         * config/riscv/riscv.c (riscv_stack_boundary): New.
8490         (riscv_option_override): Set riscv_stack_boundary.  Handle
8491         riscv_preferred_stack_boundary_arg.
8492         * config/riscv/riscv.h (MIN_STACK_BOUNDARY, ABI_STACK_BOUNDARY): New.
8493         (BIGGEST_ALIGNMENT): Set to STACK_BOUNDARY.
8494         (STACK_BOUNDARY): Set to riscv_stack_boundary.
8495         (RISCV_STACK_ALIGN): Use STACK_BOUNDARY.
8496         * config/riscv/riscv.opt (mpreferred-stack-boundary): New.
8497         * doc/invoke.tex (RISC-V Options): Add -mpreferred-stack-boundary.
8498
8499 2018-01-23  H.J. Lu  <hongjiu.lu@intel.com>
8500
8501         PR target/83905
8502         * config/i386/i386.c (ix86_expand_prologue): Use cost reference
8503         of struct ix86_frame.
8504         (ix86_expand_epilogue): Likewise.  Add a local variable for
8505         the reg_save_offset field in struct ix86_frame.
8506
8507 2018-01-23  Bin Cheng  <bin.cheng@arm.com>
8508
8509         PR tree-optimization/82604
8510         * tree-loop-distribution.c (enum partition_kind): New enum item
8511         PKIND_PARTIAL_MEMSET.
8512         (partition_builtin_p): Support above new enum item.
8513         (generate_code_for_partition): Ditto.
8514         (compute_access_range): Differentiate cases that equality can be
8515         proven at all loops, the innermost loops or no loops.
8516         (classify_builtin_st, classify_builtin_ldst): Adjust call to above
8517         function.  Set PKIND_PARTIAL_MEMSET for partition appropriately.
8518         (finalize_partitions, distribute_loop): Don't fuse partition of
8519         PKIND_PARTIAL_MEMSET kind when distributing 3-level loop nest.
8520         (prepare_perfect_loop_nest): Distribute 3-level loop nest only if
8521         parloop is enabled.
8522
8523 2018-01-23  Martin Liska  <mliska@suse.cz>
8524
8525         * predict.def (PRED_INDIR_CALL): Set probability to PROB_EVEN in
8526         order to ignore the predictor.
8527         (PRED_POLYMORPHIC_CALL): Likewise.
8528         (PRED_RECURSIVE_CALL): Likewise.
8529
8530 2018-01-23  Martin Liska  <mliska@suse.cz>
8531
8532         * tree-profile.c (tree_profiling): Print function header to
8533         aware reader which function we are working on.
8534         * value-prof.c (gimple_find_values_to_profile): Do not print
8535         not interesting value histograms.
8536
8537 2018-01-23  Martin Liska  <mliska@suse.cz>
8538
8539         * profile-count.h (enum profile_quality): Add
8540         profile_uninitialized as the first value. Do not number values
8541         as they are zero based.
8542         (profile_count::verify): Update sanity check.
8543         (profile_probability::verify): Likewise.
8544
8545 2018-01-23  Nathan Sidwell  <nathan@acm.org>
8546
8547         * doc/invoke.texi (ffor-scope): Deprecate.
8548
8549 2018-01-23  David Malcolm  <dmalcolm@redhat.com>
8550
8551         PR tree-optimization/83510
8552         * domwalk.c (set_all_edges_as_executable): New function.
8553         (dom_walker::dom_walker): Convert bool param
8554         "skip_unreachable_blocks" to enum reachability.  Move setup of
8555         edge flags to set_all_edges_as_executable and only do it when
8556         reachability is REACHABLE_BLOCKS.
8557         * domwalk.h (enum dom_walker::reachability): New enum.
8558         (dom_walker::dom_walker): Convert bool param
8559         "skip_unreachable_blocks" to enum reachability.
8560         (set_all_edges_as_executable): New decl.
8561         * graphite-scop-detection.c  (gather_bbs::gather_bbs): Convert
8562         from false for "skip_unreachable_blocks" to ALL_BLOCKS for
8563         "reachability".
8564         * tree-ssa-dom.c (dom_opt_dom_walker::dom_opt_dom_walker): Likewise,
8565         but converting true to REACHABLE_BLOCKS.
8566         * tree-ssa-sccvn.c (sccvn_dom_walker::sccvn_dom_walker): Likewise.
8567         * tree-vrp.c
8568         (check_array_bounds_dom_walker::check_array_bounds_dom_walker):
8569         Likewise, but converting it to REACHABLE_BLOCKS_PRESERVING_FLAGS.
8570         (vrp_dom_walker::vrp_dom_walker): Likewise, but converting it to
8571         REACHABLE_BLOCKS.
8572         (vrp_prop::vrp_finalize): Call set_all_edges_as_executable
8573         if check_all_array_refs will be called.
8574
8575 2018-01-23  David Malcolm  <dmalcolm@redhat.com>
8576
8577         * tree.c (selftest::test_location_wrappers): Add more test
8578         coverage.
8579
8580 2018-01-23  David Malcolm  <dmalcolm@redhat.com>
8581
8582         * sbitmap.c (selftest::test_set_range): Fix memory leaks.
8583         (selftest::test_bit_in_range): Likewise.
8584
8585 2018-01-23  Richard Sandiford  <richard.sandiford@linaro.org>
8586
8587         PR testsuite/83888
8588         * doc/sourcebuild.texi (vect_float): Say that the selector
8589         only describes the situation when -funsafe-math-optimizations is on.
8590         (vect_float_strict): Document.
8591
8592 2018-01-23  Richard Sandiford  <richard.sandiford@linaro.org>
8593
8594         PR tree-optimization/83965
8595         * tree-vect-patterns.c (vect_reassociating_reduction_p): New function.
8596         (vect_recog_dot_prod_pattern, vect_recog_sad_pattern): Use it
8597         instead of checking only for a reduction.
8598         (vect_recog_widen_sum_pattern): Likewise.
8599
8600 2018-01-23  Jan Hubicka  <hubicka@ucw.cz>
8601
8602         * predict.c (probably_never_executed): Only use precise profile info.
8603         (compute_function_frequency): Skip after inlining hack since we now
8604         have quality checking.
8605
8606 2018-01-23  Jan Hubicka  <hubicka@ucw.cz>
8607
8608         * profile-count.h (profile_probability::very_unlikely,
8609         profile_probability::unlikely, profile_probability::even): Set
8610         precision to guessed.
8611
8612 2018-01-23  Richard Biener  <rguenther@suse.de>
8613
8614         PR tree-optimization/83963
8615         * graphite-scop-detection.c (scop_detection::harmful_loop_in_region):
8616         Properly terminate dominator walk when crossing the exit edge not
8617         when visiting its source block.
8618
8619 2018-01-23  Jakub Jelinek  <jakub@redhat.com>
8620
8621         PR c++/83918
8622         * tree.c (maybe_wrap_with_location): Use NON_LVALUE_EXPR rather than
8623         VIEW_CONVERT_EXPR to wrap CONST_DECLs.
8624
8625 2018-01-22  Jakub Jelinek  <jakub@redhat.com>
8626
8627         PR tree-optimization/83957
8628         * omp-expand.c (expand_omp_for_generic): Ignore virtual PHIs.  Remove
8629         semicolon after for body surrounded by braces.
8630
8631         PR tree-optimization/83081
8632         * profile-count.h (profile_probability::split): New method.
8633         * dojump.c (do_jump_1) <case TRUTH_ANDIF_EXPR, case TRUTH_ORIF_EXPR>:
8634         Use profile_probability::split.
8635         (do_compare_rtx_and_jump): Fix adjustment of probabilities
8636         when splitting a single conditional jump into 2.
8637
8638 2018-01-22  David Malcolm  <dmalcolm@redhat.com>
8639
8640         PR tree-optimization/69452
8641         * tree-ssa-loop-im.c (class move_computations_dom_walker): Remove
8642         decl.
8643
8644 2018-01-22  Sebastian Perta  <sebastian.perta@renesas.com>
8645
8646         * config/rl78/rl78-expand.md (bswaphi2): New define_expand.
8647         * config/rl78/rl78-virt.md (*bswaphi2_virt): New define_insn.
8648         * config/rl78/rl78-real.md (*bswaphi2_real): New define_insn.
8649
8650 2018-01-22  Sebastian Perta  <sebastian.perta@renesas.com>
8651
8652         * config/rl78/rl78-protos.h (rl78_split_movdi): New function declaration.
8653         * config/rl78/rl78.md (movdi): New define_expand.
8654         * config/rl78/rl78.c (rl78_split_movdi): New function.
8655
8656 2018-01-22  Michael Meissner  <meissner@linux.vnet.ibm.com>
8657
8658         PR target/83862
8659         * config/rs6000/rs6000-protos.h (rs6000_split_signbit): Delete,
8660         no longer used.
8661         * config/rs6000/rs6000.c (rs6000_split_signbit): Likewise.
8662         * config/rs6000/rs6000.md (signbit<mode>2): Change code for IEEE
8663         128-bit to produce an UNSPEC move to get the double word with the
8664         signbit and then a shift directly to do signbit.
8665         (signbit<mode>2_dm): Replace old IEEE 128-bit signbit
8666         implementation with a new version that just does either a direct
8667         move or a regular move.  Move memory interface to separate insns.
8668         Move insns so they are next to the expander.
8669         (signbit<mode>2_dm_mem_be): New combiner insns to combine load
8670         with signbit move.  Split big and little endian case.
8671         (signbit<mode>2_dm_mem_le): Likewise.
8672         (signbit<mode>2_dm_<su>ext): Delete, no longer used.
8673         (signbit<mode>2_dm2): Likewise.
8674
8675 2018-01-22  Sebastian Perta  <sebastian.perta@renesas.com>
8676
8677         * config/rl78/rl78.md (anddi3): New define_expand.
8678
8679 2018-01-22  Sebastian Perta  <sebastian.perta@renesas.com>
8680
8681         * config/rl78/rl78.md (umindi3): New define_expand.
8682
8683 2018-01-22  Sebastian Perta  <sebastian.perta@renesas.com>
8684
8685         * config/rl78/rl78.md (smindi3): New define_expand.
8686
8687 2018-01-22  Sebastian Perta  <sebastian.perta@renesas.com>
8688
8689         * config/rl78/rl78.md (smaxdi3): New define_expand.
8690
8691 2018-01-22  Carl Love  <cel@us.ibm.com>
8692
8693         * config/rs6000/rs6000-builtin.def (ST_ELEMREV_V1TI, LD_ELEMREV_V1TI,
8694         LVX_V1TI): Add macro expansion.
8695         * config/rs6000/rs6000-c.c (altivec_builtin_types): Add argument
8696         definitions for VSX_BUILTIN_VEC_XST_BE, VSX_BUILTIN_VEC_ST,
8697         VSX_BUILTIN_VEC_XL, LD_ELEMREV_V1TI builtins.
8698         * config/rs6000/rs6000-p8swap.c (insn_is_swappable_p);
8699         Change check to determine if the instruction is a byte reversing
8700         entry.  Fix typo in comment.
8701         * config/rs6000/rs6000.c (altivec_expand_builtin): Add case entry
8702         for VSX_BUILTIN_ST_ELEMREV_V1TI and VSX_BUILTIN_LD_ELEMREV_V1TI.
8703         Add def_builtin calls for new builtins.
8704         * config/rs6000/vsx.md (vsx_st_elemrev_v1ti, vsx_ld_elemrev_v1ti):
8705         Add define_insn expansion.
8706
8707 2018-01-22  Sebastian Perta  <sebastian.perta@renesas.com>
8708
8709         * config/rl78/rl78.md (umaxdi3): New define_expand.
8710
8711 2018-01-22  Sebastian Perta  <sebastian.perta@renesas.com>
8712
8713         * config/rl78/rl78.c (rl78_note_reg_set): Fixed dead reg check
8714         for non-QImode registers.
8715
8716 2018-01-22  Richard Biener  <rguenther@suse.de>
8717
8718         PR tree-optimization/83963
8719         * graphite-scop-detection.c (scop_detection::get_sese): Delay
8720         including the loop exit block.
8721         (scop_detection::merge_sese): Likewise.
8722         (scop_detection::add_scop): Do it here instead.
8723
8724 2018-01-22  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
8725
8726         * doc/sourcebuild.texi (arm_softfloat): Document.
8727
8728 2018-01-21  John David Anglin  <danglin@gcc.gnu.org>
8729
8730         PR gcc/77734
8731         * config/pa/pa.c (pa_function_ok_for_sibcall): Use
8732         targetm.binds_local_p instead of TREE_PUBLIC to check local binding.
8733         Move TARGET_PORTABLE_RUNTIME check after TARGET_64BIT check.
8734
8735 2018-01-21  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
8736             David Edelsohn  <dje.gcc@gmail.com>
8737
8738         PR target/83946
8739         * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
8740         Change "crset eq" to "crset 2".
8741         (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
8742         (*call_indirect_aix<mode>_nospec): Likewise.
8743         (*call_value_indirect_aix<mode>_nospec): Likewise.
8744         (*call_indirect_elfv2<mode>_nospec): Likewise.
8745         (*call_value_indirect_elfv2<mode>_nospec): Likewise.
8746         (*sibcall_nonlocal_sysv<mode>): Change "crset eq" to "crset 2";
8747         change assembly output from . to $.
8748         (*sibcall_value_nonlocal_sysv<mode>): Likewise.
8749         (indirect_jump<mode>_nospec): Change assembly output from . to $.
8750         (*tablejump<mode>_internal1_nospec): Likewise.
8751
8752 2018-01-21  Oleg Endo  <olegendo@gcc.gnu.org>
8753
8754         PR target/80870
8755         * config/sh/sh_optimize_sett_clrt.cc:
8756         Use INCLUDE_ALGORITHM and INCLUDE_VECTOR instead of direct includes.
8757
8758 2018-01-20  Richard Sandiford  <richard.sandiford@linaro.org>
8759
8760         PR tree-optimization/83940
8761         * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): Set
8762         offset_dt to vect_constant_def rather than vect_unknown_def_type.
8763         (vect_check_load_store_mask): Add a mask_dt_out parameter and
8764         use it to pass back the definition type.
8765         (vect_check_store_rhs): Likewise rhs_dt_out.
8766         (vect_build_gather_load_calls): Add a mask_dt argument and use
8767         it instead of a call to vect_is_simple_use.
8768         (vectorizable_store): Update calls to vect_check_load_store_mask
8769         and vect_check_store_rhs.  Use the dt returned by the latter instead
8770         of scatter_src_dt.  Use the cached mask_dt and gs_info.offset_dt
8771         instead of calls to vect_is_simple_use.  Pass the scalar rather
8772         than the vector operand to vect_is_simple_use when handling
8773         second and subsequent copies of an rhs value.
8774         (vectorizable_load): Update calls to vect_check_load_store_mask
8775         and vect_build_gather_load_calls.  Use the cached mask_dt and
8776         gs_info.offset_dt instead of calls to vect_is_simple_use.
8777
8778 2018-01-20  Jakub Jelinek  <jakub@redhat.com>
8779
8780         PR middle-end/83945
8781         * tree-emutls.c: Include gimplify.h.
8782         (lower_emutls_2): New function.
8783         (lower_emutls_1): If ADDR_EXPR is a gimple invariant and walk_tree
8784         with lower_emutls_2 callback finds some TLS decl in it, unshare_expr
8785         it before further processing.
8786
8787         PR target/83930
8788         * simplify-rtx.c (simplify_binary_operation_1) <case UMOD>: Use
8789         UINTVAL (trueop1) instead of INTVAL (op1).
8790
8791 2018-01-19  Jakub Jelinek  <jakub@redhat.com>
8792
8793         PR debug/81570
8794         PR debug/83728
8795         * dwarf2cfi.c (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define to
8796         INCOMING_FRAME_SP_OFFSET if not defined.
8797         (scan_trace): Add ENTRY argument.  If true and
8798         DEFAULT_INCOMING_FRAME_SP_OFFSET != INCOMING_FRAME_SP_OFFSET,
8799         emit a note to adjust the CFA offset.
8800         (create_cfi_notes): Adjust scan_trace callers.
8801         (create_cie_data): Use DEFAULT_INCOMING_FRAME_SP_OFFSET rather than
8802         INCOMING_FRAME_SP_OFFSET in the CIE.
8803         * config/i386/i386.h (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define.
8804         * config/stormy16/stormy16.h (DEFAULT_INCOMING_FRAME_SP_OFFSET):
8805         Likewise.
8806         * doc/tm.texi.in (DEFAULT_INCOMING_FRAME_SP_OFFSET): Document.
8807         * doc/tm.texi: Regenerated.
8808
8809 2018-01-19  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>
8810
8811         PR rtl-optimization/83147
8812         * lra-constraints.c (remove_inheritance_pseudos): Use
8813         lra_substitute_pseudo_within_insn.
8814
8815 2018-01-19  Tom de Vries  <tom@codesourcery.com>
8816             Cesar Philippidis  <cesar@codesourcery.com>
8817
8818         PR target/83920
8819         * config/nvptx/nvptx.c (nvptx_single): Fix jit workaround.
8820
8821 2018-01-19  Cesar Philippidis  <cesar@codesourcery.com>
8822
8823         PR target/83790
8824         * config/nvptx/nvptx.c (output_init_frag): Don't use generic address
8825         spaces for function labels.
8826
8827 2018-01-19  Martin Liska  <mliska@suse.cz>
8828
8829         * predict.def (PRED_LOOP_EXIT): Change from 85 to 89.
8830         (PRED_LOOP_EXIT_WITH_RECURSION): Change from 72 to 78.
8831         (PRED_LOOP_EXTRA_EXIT): Change from 83 to 67.
8832         (PRED_OPCODE_POSITIVE): Change from 64 to 59.
8833         (PRED_TREE_OPCODE_POSITIVE): Change from 64 to 59.
8834         (PRED_CONST_RETURN): Change from 69 to 65.
8835         (PRED_NULL_RETURN): Change from 91 to 71.
8836         (PRED_LOOP_IV_COMPARE_GUESS): Change from 98 to 64.
8837         (PRED_LOOP_GUARD): Change from 66 to 73.
8838
8839 2018-01-19  Martin Liska  <mliska@suse.cz>
8840
8841         * predict.c (predict_insn_def): Add new assert.
8842         (struct branch_predictor): Change type to signed integer.
8843         (test_prediction_value_range): Amend test to cover
8844         PROB_UNINITIALIZED.
8845         * predict.def (PRED_LOOP_ITERATIONS): Use the new constant.
8846         (PRED_LOOP_ITERATIONS_GUESSED): Likewise.
8847         (PRED_LOOP_ITERATIONS_MAX): Likewise.
8848         (PRED_LOOP_IV_COMPARE): Likewise.
8849         * predict.h (PROB_UNINITIALIZED): Define new constant.
8850
8851 2018-01-19  Martin Liska  <mliska@suse.cz>
8852
8853         * predict.c (dump_prediction): Add new format for
8854         analyze_brprob.py script which is enabled with -details
8855         suboption.
8856         * profile-count.h (precise_p): New function.
8857
8858 2018-01-19  Richard Sandiford  <richard.sandiford@linaro.org>
8859
8860         PR tree-optimization/83922
8861         * tree-vect-loop.c (vect_verify_full_masking): Return false if
8862         there are no statements that need masking.
8863         (vect_active_double_reduction_p): New function.
8864         (vect_analyze_loop_operations): Use it when handling phis that
8865         are not in the loop header.
8866
8867 2018-01-19  Richard Sandiford  <richard.sandiford@linaro.org>
8868
8869         PR tree-optimization/83914
8870         * tree-vect-loop.c (vectorizable_induction): Don't convert
8871         init_expr or apply the peeling adjustment for inductions
8872         that are nested within the vectorized loop.
8873
8874 2018-01-19  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
8875
8876         * config/arm/thumb2.md (*thumb2_negsi2_short): Use RSB mnemonic
8877         instead of NEG.
8878
8879 2018-01-18  Jakub Jelinek  <jakub@redhat.com>
8880
8881         PR sanitizer/81715
8882         PR testsuite/83882
8883         * function.h (gimplify_parameters): Add gimple_seq * argument.
8884         * function.c: Include gimple.h and options.h.
8885         (gimplify_parameters): Add cleanup argument, add CLOBBER stmts
8886         for the added local temporaries if needed.
8887         * gimplify.c (gimplify_body): Adjust gimplify_parameters caller,
8888         if there are any parameter cleanups, wrap whole body into a
8889         try/finally with the cleanups.
8890
8891 2018-01-18  Wilco Dijkstra  <wdijkstr@arm.com>
8892
8893         PR target/82964
8894         * config/aarch64/aarch64.c (aarch64_legitimate_constant_p):
8895         Use GET_MODE_CLASS for scalar floating point.
8896
8897 2018-01-18  Jan Hubicka  <hubicka@ucw.cz>
8898
8899         PR ipa/82256
8900         patch by PaX Team
8901         * cgraphclones.c (cgraph_node::create_version_clone_with_body):
8902         Fix call of call_cgraph_insertion_hooks.
8903
8904 2018-01-18  Martin Sebor  <msebor@redhat.com>
8905
8906         * doc/invoke.texi (-Wclass-memaccess): Tweak text.
8907
8908 2018-01-18  Jan Hubicka  <hubicka@ucw.cz>
8909
8910         PR ipa/83619
8911         * cgraph.c (cgraph_edge::redirect_call_stmt_to_callee): Update edge
8912         frequencies.
8913
8914 2018-01-18  Boris Kolpackov  <boris@codesynthesis.com>
8915
8916         PR other/70268
8917         * common.opt: (-ffile-prefix-map): New option.
8918         * opts.c (common_handle_option): Defer it.
8919         * opts-global.c (handle_common_deferred_options): Handle it.
8920         * debug.h (remap_debug_filename, add_debug_prefix_map): Move to...
8921         * file-prefix-map.h: New file.
8922         (remap_debug_filename, add_debug_prefix_map): ...here.
8923         (add_macro_prefix_map, add_file_prefix_map, remap_macro_filename): New.
8924         * final.c (debug_prefix_map, add_debug_prefix_map
8925         remap_debug_filename): Move to...
8926         * file-prefix-map.c: New file.
8927         (file_prefix_map, add_prefix_map, remap_filename) ...here and rename,
8928         generalize, get rid of alloca(), use strrchr() instead of strchr().
8929         (add_macro_prefix_map, add_debug_prefix_map, add_file_prefix_map):
8930         Implement in terms of add_prefix_map().
8931         (remap_macro_filename, remap_debug_filename): Implement in term of
8932         remap_filename().
8933         * Makefile.in (OBJS, PLUGIN_HEADERS): Add new files.
8934         * builtins.c (fold_builtin_FILE): Call remap_macro_filename().
8935         * dbxout.c: Include file-prefix-map.h.
8936         * varasm.c: Likewise.
8937         * vmsdbgout.c: Likewise.
8938         * xcoffout.c: Likewise.
8939         * dwarf2out.c: Likewise plus omit new options from DW_AT_producer.
8940         * doc/cppopts.texi (-fmacro-prefix-map): Document.
8941         * doc/invoke.texi (-ffile-prefix-map): Document.
8942         (-fdebug-prefix-map): Update description.
8943
8944 2018-01-18  Martin Liska  <mliska@suse.cz>
8945
8946         * config/i386/i386.c (indirect_thunk_name): Document that also
8947         lfence is emitted.
8948         (output_indirect_thunk): Document why both instructions
8949         (pause and lfence) are generated.
8950
8951 2018-01-18  Richard Biener  <rguenther@suse.de>
8952
8953         PR tree-optimization/83887
8954         * graphite-scop-detection.c
8955         (scop_detection::get_nearest_dom_with_single_entry): Remove.
8956         (scop_detection::get_nearest_pdom_with_single_exit): Likewise.
8957         (scop_detection::merge_sese): Re-implement with a flood-fill
8958         algorithm that properly finds a SESE region if it exists.
8959
8960 2018-01-18  Jakub Jelinek  <jakub@redhat.com>
8961
8962         PR c/61240
8963         * match.pd ((P + A) - P, P - (P + A), (P + A) - (P + B)): For
8964         pointer_diff optimizations use view_convert instead of convert.
8965
8966 2018-01-17  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
8967
8968         * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
8969         Generate different code for -mno-speculate-indirect-jumps.
8970         (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
8971         (*call_indirect_aix<mode>): Disable for
8972         -mno-speculate-indirect-jumps.
8973         (*call_indirect_aix<mode>_nospec): New define_insn.
8974         (*call_value_indirect_aix<mode>): Disable for
8975         -mno-speculate-indirect-jumps.
8976         (*call_value_indirect_aix<mode>_nospec): New define_insn.
8977         (*sibcall_nonlocal_sysv<mode>): Generate different code for
8978         -mno-speculate-indirect-jumps.
8979         (*sibcall_value_nonlocal_sysv<mode>): Likewise.
8980
8981 2018-01-17  Michael Meissner  <meissner@linux.vnet.ibm.com>
8982
8983         * config/rs6000/rs6000.c (rs6000_emit_move): If we load or store a
8984         long double type, set the flags for noting the default long double
8985         type, even if we don't pass or return a long double type.
8986
8987 2018-01-17  Jan Hubicka  <hubicka@ucw.cz>
8988
8989         PR ipa/83051
8990         * ipa-inline.c (flatten_function): Do not overwrite final inlining
8991         failure.
8992
8993 2018-01-17  Will Schmidt  <will_schmidt@vnet.ibm.com>
8994
8995         * config/rs6000/rs6000.c (rs6000_gimple_builtin): Add gimple folding
8996         support for merge[hl].
8997         (fold_mergehl_helper): New helper function.
8998         (tree-vector-builder.h): New #include for tree_vector_builder usage.
8999         * config/rs6000/altivec.md (altivec_vmrghw_direct): Add xxmrghw insn.
9000         (altivec_vmrglw_direct): Add xxmrglw insn.
9001
9002 2018-01-17  Andrew Waterman  <andrew@sifive.com>
9003
9004         * config/riscv/riscv.c (riscv_conditional_register_usage): If
9005         UNITS_PER_FP_ARG is 0, set call_used_regs to 1 for all FP regs.
9006
9007 2018-01-17  David Malcolm  <dmalcolm@redhat.com>
9008
9009         PR lto/83121
9010         * ipa-devirt.c (add_type_duplicate): When comparing memory layout,
9011         call the lto_location_cache before reading the
9012         DECL_SOURCE_LOCATION of the types.
9013
9014 2018-01-17  Wilco Dijkstra  <wdijkstr@arm.com>
9015             Richard Sandiford  <richard.sandiford@linaro.org>
9016
9017         * config/aarch64/aarch64.md (movti_aarch64): Use Uti constraint.
9018         * config/aarch64/aarch64.c (aarch64_mov128_immediate): New function.
9019         (aarch64_legitimate_constant_p): Just support CONST_DOUBLE
9020         SF/DF/TF mode to avoid creating illegal CONST_WIDE_INT immediates.
9021         * config/aarch64/aarch64-protos.h (aarch64_mov128_immediate):
9022         Add declaration.
9023         * config/aarch64/constraints.md (aarch64_movti_operand):
9024         Limit immediates.
9025         * config/aarch64/predicates.md (Uti): Add new constraint.
9026
9027 2018-01-17 Carl Love  <cel@us.ibm.com>
9028
9029         * config/rs6000/vsx.md (define_expand xl_len_r,
9030         define_expand stxvl, define_expand *stxvl): Add match_dup argument.
9031         (define_insn): Add, match_dup 1 argument to define_insn stxvll and
9032         lxvll.
9033         (define_expand, define_insn): Move the shift left from  the
9034         define_insn to the define_expand for lxvl and stxvl instructions.
9035         * config/rs6000/rs6000-builtin.def (BU_P9V_64BIT_VSX_2): Change LXVL
9036         and XL_LEN_R definitions to PURE.
9037
9038 2018-01-17  Uros Bizjak  <ubizjak@gmail.com>
9039
9040         * config/i386/i386.c (indirect_thunk_name): Declare regno
9041         as unsigned int.  Compare regno with INVALID_REGNUM.
9042         (output_indirect_thunk): Ditto.
9043         (output_indirect_thunk_function): Ditto.
9044         (ix86_code_end): Declare regno as unsigned int.  Use INVALID_REGNUM
9045         in the call to output_indirect_thunk_function.
9046
9047 2018-01-17  Richard Sandiford  <richard.sandiford@linaro.org>
9048
9049         PR middle-end/83884
9050         * expr.c (expand_expr_real_1): Use the size of GET_MODE (op0)
9051         rather than the size of inner_type to determine the stack slot size
9052         when handling VIEW_CONVERT_EXPRs on strict-alignment targets.
9053
9054 2018-01-16  Sebastian Peryt  <sebastian.peryt@intel.com>
9055
9056         PR target/83546
9057         * config/i386/i386.c (ix86_option_override_internal): Add PTA_RDRND
9058         to PTA_SILVERMONT.
9059
9060 2018-01-16  Michael Meissner  <meissner@linux.vnet.ibm.com>
9061
9062         * config.gcc (powerpc*-linux*-*): Add support for 64-bit little
9063         endian Linux systems to optionally enable multilibs for selecting
9064         the long double type if the user configured an explicit type.
9065         * config/rs6000/rs6000.h (TARGET_IEEEQUAD_MULTILIB): Indicate we
9066         have no long double multilibs if not defined.
9067         * config/rs6000/rs6000.c (rs6000_option_override_internal): Do not
9068         warn if the user used -mabi={ieee,ibm}longdouble and we built
9069         multilibs for long double.
9070         * config/rs6000/linux64.h (MULTILIB_DEFAULTS_IEEE): Define as the
9071         appropriate multilib option.
9072         (MULTILIB_DEFAULTS): Add MULTILIB_DEFAULTS_IEEE to the default
9073         multilib options.
9074         * config/rs6000/t-ldouble-linux64le-ibm: New configuration files
9075         for building long double multilibs.
9076         * config/rs6000/t-ldouble-linux64le-ieee: Likewise.
9077
9078 2018-01-16  John David Anglin  <danglin@gcc.gnu.org>
9079
9080         * config.gcc (hppa*-*-linux*): Change callee copies ABI to caller
9081         copies.
9082
9083         * config/pa.h (MALLOC_ABI_ALIGNMENT): Set 32-bit alignment default to
9084         64 bits.
9085         * config/pa/pa32-linux.h (MALLOC_ABI_ALIGNMENT): Set alignment to
9086         128 bits.
9087
9088         * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Cleanup type and mode
9089         variables.
9090
9091         * config/pa/pa.c (pa_function_arg_size): Apply CEIL to GET_MODE_SIZE
9092         return value.
9093
9094 2018-01-16  Eric Botcazou  <ebotcazou@adacore.com>
9095
9096         * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For an
9097         ADDR_EXPR, do not count the offset of a COMPONENT_REF twice.
9098
9099 2018-01-16  Kelvin Nilsen  <kelvin@gcc.gnu.org>
9100
9101         * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Generate
9102         different rtl trees depending on TARGET_64BIT.
9103         (rs6000_gen_lvx): Likewise.
9104
9105 2018-01-16  Eric Botcazou  <ebotcazou@adacore.com>
9106
9107         * config/visium/visium.md (nop): Tweak comment.
9108         (hazard_nop): Likewise.
9109
9110 2018-01-16  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
9111
9112         * config/rs6000/rs6000.c (rs6000_opt_vars): Add entry for
9113         -mspeculate-indirect-jumps.
9114         * config/rs6000/rs6000.md (*call_indirect_elfv2<mode>): Disable
9115         for -mno-speculate-indirect-jumps.
9116         (*call_indirect_elfv2<mode>_nospec): New define_insn.
9117         (*call_value_indirect_elfv2<mode>): Disable for
9118         -mno-speculate-indirect-jumps.
9119         (*call_value_indirect_elfv2<mode>_nospec): New define_insn.
9120         (indirect_jump): Emit different RTL for
9121         -mno-speculate-indirect-jumps.
9122         (*indirect_jump<mode>): Disable for
9123         -mno-speculate-indirect-jumps.
9124         (*indirect_jump<mode>_nospec): New define_insn.
9125         (tablejump): Emit different RTL for
9126         -mno-speculate-indirect-jumps.
9127         (tablejumpsi): Disable for -mno-speculate-indirect-jumps.
9128         (tablejumpsi_nospec): New define_expand.
9129         (tablejumpdi): Disable for -mno-speculate-indirect-jumps.
9130         (tablejumpdi_nospec): New define_expand.
9131         (*tablejump<mode>_internal1): Disable for
9132         -mno-speculate-indirect-jumps.
9133         (*tablejump<mode>_internal1_nospec): New define_insn.
9134         * config/rs6000/rs6000.opt (mspeculate-indirect-jumps): New
9135         option.
9136
9137 2018-01-16  Artyom Skrobov tyomitch@gmail.com
9138
9139         * caller-save.c (insert_save): Drop unnecessary parameter.  All
9140         callers updated.
9141
9142 2018-01-16  Jakub Jelinek  <jakub@redhat.com>
9143             Richard Biener  <rguenth@suse.de>
9144
9145         PR libgomp/83590
9146         * gimplify.c (gimplify_one_sizepos): For is_gimple_constant (expr)
9147         return early, inline manually is_gimple_sizepos.  Make sure if we
9148         call gimplify_expr we don't end up with a gimple constant.
9149         * tree.c (variably_modified_type_p): Don't return true for
9150         is_gimple_constant (_t).  Inline manually is_gimple_sizepos.
9151         * gimplify.h (is_gimple_sizepos): Remove.
9152
9153 2018-01-16  Richard Sandiford  <richard.sandiford@linaro.org>
9154
9155         PR tree-optimization/83857
9156         * tree-vect-loop.c (vect_analyze_loop_operations): Don't call
9157         vectorizable_live_operation for pure SLP statements.
9158         (vectorizable_live_operation): Handle PHIs.
9159
9160 2018-01-16  Richard Biener  <rguenther@suse.de>
9161
9162         PR tree-optimization/83867
9163         * tree-vect-stmts.c (vect_transform_stmt): Precompute
9164         nested_in_vect_loop_p since the scalar stmt may get invalidated.
9165
9166 2018-01-16  Jakub Jelinek  <jakub@redhat.com>
9167
9168         PR c/83844
9169         * stor-layout.c (handle_warn_if_not_align): Use byte_position and
9170         multiple_of_p instead of unchecked tree_to_uhwi and UHWI check.
9171         If off is not INTEGER_CST, issue a may not be aligned warning
9172         rather than isn't aligned.  Use isn%'t rather than isn't.
9173         * fold-const.c (multiple_of_p) <case BIT_AND_EXPR>: Don't fall through
9174         into MULT_EXPR.
9175         <case MULT_EXPR>: Improve the case when bottom and one of the
9176         MULT_EXPR operands are INTEGER_CSTs and bottom is multiple of that
9177         operand, in that case check if the other operand is multiple of
9178         bottom divided by the INTEGER_CST operand.
9179
9180 2018-01-16  Richard Sandiford  <richard.sandiford@linaro.org>
9181
9182         PR target/83858
9183         * config/pa/pa.h (FUNCTION_ARG_SIZE): Delete.
9184         * config/pa/pa-protos.h (pa_function_arg_size): Declare.
9185         * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Use
9186         pa_function_arg_size instead of FUNCTION_ARG_SIZE.
9187         * config/pa/pa.c (pa_function_arg_advance): Likewise.
9188         (pa_function_arg, pa_arg_partial_bytes): Likewise.
9189         (pa_function_arg_size): New function.
9190
9191 2018-01-16  Richard Sandiford  <richard.sandiford@linaro.org>
9192
9193         * fold-const.c (fold_ternary_loc): Construct the vec_perm_indices
9194         in a separate statement.
9195
9196 2018-01-16  Richard Sandiford  <richard.sandiford@linaro.org>
9197
9198         PR tree-optimization/83847
9199         * tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Don't
9200         group gathers and scatters.
9201
9202 2018-01-16  Jakub Jelinek  <jakub@redhat.com>
9203
9204         PR rtl-optimization/86620
9205         * params.def (max-sched-ready-insns): Bump minimum value to 1.
9206
9207         PR rtl-optimization/83213
9208         * recog.c (peep2_attempt): Copy over CROSSING_JUMP_P from peepinsn
9209         to last if both are JUMP_INSNs.
9210
9211         PR tree-optimization/83843
9212         * gimple-ssa-store-merging.c
9213         (imm_store_chain_info::output_merged_store): Handle bit_not_p on
9214         store_immediate_info for bswap/nop orig_stores.
9215
9216 2018-01-15  Andrew Waterman  <andrew@sifive.com>
9217
9218         * config/riscv/riscv.c (riscv_rtx_costs) <MULT>: Increase cost if
9219         !TARGET_MUL.
9220         <UDIV>: Increase cost if !TARGET_DIV.
9221
9222 2018-01-15  Segher Boessenkool  <segher@kernel.crashing.org>
9223
9224         * config/rs6000/rs6000.md (define_attr "type"): Remove delayed_cr.
9225         (define_attr "cr_logical_3op"): New.
9226         (cceq_ior_compare): Adjust.
9227         (cceq_ior_compare_complement): Adjust.
9228         (*cceq_rev_compare): Adjust.
9229         * config/rs6000/rs6000.c (rs6000_adjust_cost): Adjust.
9230         (is_cracked_insn): Adjust.
9231         (insn_must_be_first_in_group): Adjust.
9232         * config/rs6000/40x.md: Adjust.
9233         * config/rs6000/440.md: Adjust.
9234         * config/rs6000/476.md: Adjust.
9235         * config/rs6000/601.md: Adjust.
9236         * config/rs6000/603.md: Adjust.
9237         * config/rs6000/6xx.md: Adjust.
9238         * config/rs6000/7450.md: Adjust.
9239         * config/rs6000/7xx.md: Adjust.
9240         * config/rs6000/8540.md: Adjust.
9241         * config/rs6000/cell.md: Adjust.
9242         * config/rs6000/e300c2c3.md: Adjust.
9243         * config/rs6000/e500mc.md: Adjust.
9244         * config/rs6000/e500mc64.md: Adjust.
9245         * config/rs6000/e5500.md: Adjust.
9246         * config/rs6000/e6500.md: Adjust.
9247         * config/rs6000/mpc.md: Adjust.
9248         * config/rs6000/power4.md: Adjust.
9249         * config/rs6000/power5.md: Adjust.
9250         * config/rs6000/power6.md: Adjust.
9251         * config/rs6000/power7.md: Adjust.
9252         * config/rs6000/power8.md: Adjust.
9253         * config/rs6000/power9.md: Adjust.
9254         * config/rs6000/rs64.md: Adjust.
9255         * config/rs6000/titan.md: Adjust.
9256
9257 2018-01-15  H.J. Lu  <hongjiu.lu@intel.com>
9258
9259         * config/i386/predicates.md (indirect_branch_operand): Rewrite
9260         ix86_indirect_branch_register logic.
9261
9262 2018-01-15  H.J. Lu  <hongjiu.lu@intel.com>
9263
9264         * config/i386/constraints.md (Bs): Update
9265         ix86_indirect_branch_register check.  Don't check
9266         ix86_indirect_branch_register with GOT_memory_operand.
9267         (Bw): Likewise.
9268         * config/i386/predicates.md (GOT_memory_operand): Don't check
9269         ix86_indirect_branch_register here.
9270         (GOT32_symbol_operand): Likewise.
9271
9272 2018-01-15  H.J. Lu  <hongjiu.lu@intel.com>
9273
9274         * config/i386/predicates.md (constant_call_address_operand):
9275         Rewrite ix86_indirect_branch_register logic.
9276         (sibcall_insn_operand): Likewise.
9277
9278 2018-01-15  H.J. Lu  <hongjiu.lu@intel.com>
9279
9280         * config/i386/constraints.md (Bs): Replace
9281         ix86_indirect_branch_thunk_register with
9282         ix86_indirect_branch_register.
9283         (Bw): Likewise.
9284         * config/i386/i386.md (indirect_jump): Likewise.
9285         (tablejump): Likewise.
9286         (*sibcall_memory): Likewise.
9287         (*sibcall_value_memory): Likewise.
9288         Peepholes of indirect call and jump via memory: Likewise.
9289         * config/i386/i386.opt: Likewise.
9290         * config/i386/predicates.md (indirect_branch_operand): Likewise.
9291         (GOT_memory_operand): Likewise.
9292         (call_insn_operand): Likewise.
9293         (sibcall_insn_operand): Likewise.
9294         (GOT32_symbol_operand): Likewise.
9295
9296 2018-01-15  Jakub Jelinek  <jakub@redhat.com>
9297
9298         PR middle-end/83837
9299         * omp-expand.c (expand_omp_atomic_pipeline): Use loaded_val
9300         type rather than type addr's type points to.
9301         (expand_omp_atomic_mutex): Likewise.
9302         (expand_omp_atomic): Likewise.
9303
9304 2018-01-15  H.J. Lu  <hongjiu.lu@intel.com>
9305
9306         PR target/83839
9307         * config/i386/i386.c (output_indirect_thunk_function): Use
9308         ASM_OUTPUT_LABEL, instead of ASM_OUTPUT_DEF, for TARGET_MACHO
9309         for  __x86_return_thunk.
9310
9311 2018-01-15  Richard Biener  <rguenther@suse.de>
9312
9313         PR middle-end/83850
9314         * expmed.c (extract_bit_field_1): Fix typo.
9315
9316 2018-01-15  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
9317
9318         PR target/83687
9319         * config/arm/iterators.md (VF): New mode iterator.
9320         * config/arm/neon.md (neon_vabd<mode>_2): Use the above.
9321         Remove integer-related logic from pattern.
9322         (neon_vabd<mode>_3): Likewise.
9323
9324 2018-01-15  Jakub Jelinek  <jakub@redhat.com>
9325
9326         PR middle-end/82694
9327         * common.opt (fstrict-overflow): No longer an alias.
9328         (fwrapv-pointer): New option.
9329         * tree.h (TYPE_OVERFLOW_WRAPS, TYPE_OVERFLOW_UNDEFINED): Define
9330         also for pointer types based on flag_wrapv_pointer.
9331         * opts.c (common_handle_option) <case OPT_fstrict_overflow>: Set
9332         opts->x_flag_wrap[pv] to !value, clear opts->x_flag_trapv if
9333         opts->x_flag_wrapv got set.
9334         * fold-const.c (fold_comparison, fold_binary_loc): Revert 2017-08-01
9335         changes, just use TYPE_OVERFLOW_UNDEFINED on pointer type instead of
9336         POINTER_TYPE_OVERFLOW_UNDEFINED.
9337         * match.pd: Likewise in address comparison pattern.
9338         * doc/invoke.texi: Document -fwrapv and -fstrict-overflow.
9339
9340 2018-01-15  Richard Biener  <rguenther@suse.de>
9341
9342         PR lto/83804
9343         * tree.c (free_lang_data_in_type): Always unlink TYPE_DECLs
9344         from TYPE_FIELDS.  Free TYPE_BINFO if not used by devirtualization.
9345         Reset type names to their identifier if their TYPE_DECL doesn't
9346         have linkage (and thus is used for ODR and devirt).
9347         (save_debug_info_for_decl): Remove.
9348         (save_debug_info_for_type): Likewise.
9349         (add_tree_to_fld_list): Adjust.
9350         * tree-pretty-print.c (dump_generic_node): Make dumping of
9351         type names more robust.
9352
9353 2018-01-15  Richard Biener  <rguenther@suse.de>
9354
9355         * BASE-VER: Bump to 8.0.1.
9356
9357 2018-01-14  Martin Sebor  <msebor@redhat.com>
9358
9359         PR other/83508
9360         * builtins.c (check_access): Avoid warning when the no-warning bit
9361         is set.
9362
9363 2018-01-14  Cory Fields  <cory-nospam-@coryfields.com>
9364
9365         * tree-ssa-loop-im.c (sort_bbs_in_loop_postorder_cmp): Stabilize sort.
9366         * ira-color (allocno_hard_regs_compare): Likewise.
9367
9368 2018-01-14  Nathan Rossi  <nathan@nathanrossi.com>
9369
9370         PR target/83013
9371         * config/microblaze/microblaze.c (microblaze_asm_output_ident):
9372         Use .pushsection/.popsection.
9373
9374 2018-01-14  Martin Sebor  <msebor@redhat.com>
9375
9376         PR c++/81327
9377         * doc/invoke.texi (-Wlass-memaccess): Document suppression by casting.
9378
9379 2018-01-14  Jakub Jelinek  <jakub@redhat.com>
9380
9381         * config.gcc (i[34567]86-*-*): Remove one duplicate gfniintrin.h
9382         entry from extra_headers.
9383         (x86_64-*-*): Remove two duplicate gfniintrin.h entries from
9384         extra_headers, make the list bitwise identical to the i?86-*-* one.
9385
9386 2018-01-14  H.J. Lu  <hongjiu.lu@intel.com>
9387
9388         * config/i386/i386.c (ix86_set_indirect_branch_type): Disallow
9389         -mcmodel=large with -mindirect-branch=thunk,
9390         -mindirect-branch=thunk-extern, -mfunction-return=thunk and
9391         -mfunction-return=thunk-extern.
9392         * doc/invoke.texi: Document -mcmodel=large is incompatible with
9393         -mindirect-branch=thunk, -mindirect-branch=thunk-extern,
9394         -mfunction-return=thunk and -mfunction-return=thunk-extern.
9395
9396 2018-01-14  H.J. Lu  <hongjiu.lu@intel.com>
9397
9398         * config/i386/i386.c (print_reg): Print the name of the full
9399         integer register without '%'.
9400         (ix86_print_operand): Handle 'V'.
9401          * doc/extend.texi: Document 'V' modifier.
9402
9403 2018-01-14  H.J. Lu  <hongjiu.lu@intel.com>
9404
9405         * config/i386/constraints.md (Bs): Disallow memory operand for
9406         -mindirect-branch-register.
9407         (Bw): Likewise.
9408         * config/i386/predicates.md (indirect_branch_operand): Likewise.
9409         (GOT_memory_operand): Likewise.
9410         (call_insn_operand): Likewise.
9411         (sibcall_insn_operand): Likewise.
9412         (GOT32_symbol_operand): Likewise.
9413         * config/i386/i386.md (indirect_jump): Call convert_memory_address
9414         for -mindirect-branch-register.
9415         (tablejump): Likewise.
9416         (*sibcall_memory): Likewise.
9417         (*sibcall_value_memory): Likewise.
9418         Disallow peepholes of indirect call and jump via memory for
9419         -mindirect-branch-register.
9420         (*call_pop): Replace m with Bw.
9421         (*call_value_pop): Likewise.
9422         (*sibcall_pop_memory): Replace m with Bs.
9423         * config/i386/i386.opt (mindirect-branch-register): New option.
9424         * doc/invoke.texi: Document -mindirect-branch-register option.
9425
9426 2018-01-14  H.J. Lu  <hongjiu.lu@intel.com>
9427
9428         * config/i386/i386-protos.h (ix86_output_function_return): New.
9429         * config/i386/i386.c (ix86_set_indirect_branch_type): Also
9430         set function_return_type.
9431         (indirect_thunk_name): Add ret_p to indicate thunk for function
9432         return.
9433         (output_indirect_thunk_function): Pass false to
9434         indirect_thunk_name.
9435         (ix86_output_indirect_branch_via_reg): Likewise.
9436         (ix86_output_indirect_branch_via_push): Likewise.
9437         (output_indirect_thunk_function): Create alias for function
9438         return thunk if regno < 0.
9439         (ix86_output_function_return): New function.
9440         (ix86_handle_fndecl_attribute): Handle function_return.
9441         (ix86_attribute_table): Add function_return.
9442         * config/i386/i386.h (machine_function): Add
9443         function_return_type.
9444         * config/i386/i386.md (simple_return_internal): Use
9445         ix86_output_function_return.
9446         (simple_return_internal_long): Likewise.
9447         * config/i386/i386.opt (mfunction-return=): New option.
9448         (indirect_branch): Mention -mfunction-return=.
9449         * doc/extend.texi: Document function_return function attribute.
9450         * doc/invoke.texi: Document -mfunction-return= option.
9451
9452 2018-01-14  H.J. Lu  <hongjiu.lu@intel.com>
9453
9454         * config/i386/i386-opts.h (indirect_branch): New.
9455         * config/i386/i386-protos.h (ix86_output_indirect_jmp): Likewise.
9456         * config/i386/i386.c (ix86_using_red_zone): Disallow red-zone
9457         with local indirect jump when converting indirect call and jump.
9458         (ix86_set_indirect_branch_type): New.
9459         (ix86_set_current_function): Call ix86_set_indirect_branch_type.
9460         (indirectlabelno): New.
9461         (indirect_thunk_needed): Likewise.
9462         (indirect_thunk_bnd_needed): Likewise.
9463         (indirect_thunks_used): Likewise.
9464         (indirect_thunks_bnd_used): Likewise.
9465         (INDIRECT_LABEL): Likewise.
9466         (indirect_thunk_name): Likewise.
9467         (output_indirect_thunk): Likewise.
9468         (output_indirect_thunk_function): Likewise.
9469         (ix86_output_indirect_branch_via_reg): Likewise.
9470         (ix86_output_indirect_branch_via_push): Likewise.
9471         (ix86_output_indirect_branch): Likewise.
9472         (ix86_output_indirect_jmp): Likewise.
9473         (ix86_code_end): Call output_indirect_thunk_function if needed.
9474         (ix86_output_call_insn): Call ix86_output_indirect_branch if
9475         needed.
9476         (ix86_handle_fndecl_attribute): Handle indirect_branch.
9477         (ix86_attribute_table): Add indirect_branch.
9478         * config/i386/i386.h (machine_function): Add indirect_branch_type
9479         and has_local_indirect_jump.
9480         * config/i386/i386.md (indirect_jump): Set has_local_indirect_jump
9481         to true.
9482         (tablejump): Likewise.
9483         (*indirect_jump): Use ix86_output_indirect_jmp.
9484         (*tablejump_1): Likewise.
9485         (simple_return_indirect_internal): Likewise.
9486         * config/i386/i386.opt (mindirect-branch=): New option.
9487         (indirect_branch): New.
9488         (keep): Likewise.
9489         (thunk): Likewise.
9490         (thunk-inline): Likewise.
9491         (thunk-extern): Likewise.
9492         * doc/extend.texi: Document indirect_branch function attribute.
9493         * doc/invoke.texi: Document -mindirect-branch= option.
9494
9495 2018-01-14  Jan Hubicka  <hubicka@ucw.cz>
9496
9497         PR ipa/83051
9498         * ipa-inline.c (edge_badness): Tolerate roundoff errors.
9499
9500 2018-01-14  Richard Sandiford  <richard.sandiford@linaro.org>
9501
9502         * ipa-inline.c (want_inline_small_function_p): Return false if
9503         inlining has already failed with CIF_FINAL_ERROR.
9504         (update_caller_keys): Call want_inline_small_function_p before
9505         can_inline_edge_p.
9506         (update_callee_keys): Likewise.
9507
9508 2018-01-10  Kelvin Nilsen  <kelvin@gcc.gnu.org>
9509
9510         * config/rs6000/rs6000-p8swap.c (rs6000_sum_of_two_registers_p):
9511         New function.
9512         (rs6000_quadword_masked_address_p): Likewise.
9513         (quad_aligned_load_p): Likewise.
9514         (quad_aligned_store_p): Likewise.
9515         (const_load_sequence_p): Add comment to describe the outer-most loop.
9516         (mimic_memory_attributes_and_flags): New function.
9517         (rs6000_gen_stvx): Likewise.
9518         (replace_swapped_aligned_store): Likewise.
9519         (rs6000_gen_lvx): Likewise.
9520         (replace_swapped_aligned_load): Likewise.
9521         (replace_swapped_load_constant): Capitalize argument name in
9522         comment describing this function.
9523         (rs6000_analyze_swaps): Add a third pass to search for vector loads
9524         and stores that access quad-word aligned addresses and replace
9525         with stvx or lvx instructions when appropriate.
9526         * config/rs6000/rs6000-protos.h (rs6000_sum_of_two_registers_p):
9527         New function prototype.
9528         (rs6000_quadword_masked_address_p): Likewise.
9529         (rs6000_gen_lvx): Likewise.
9530         (rs6000_gen_stvx): Likewise.
9531         * config/rs6000/vsx.md (*vsx_le_perm_load_<mode>): For modes
9532         VSX_D (V2DF, V2DI), modify this split to select lvx instruction
9533         when memory address is aligned.
9534         (*vsx_le_perm_load_<mode>): For modes VSX_W (V4SF, V4SI), modify
9535         this split to select lvx instruction when memory address is aligned.
9536         (*vsx_le_perm_load_v8hi): Modify this split to select lvx
9537         instruction when memory address is aligned.
9538         (*vsx_le_perm_load_v16qi): Likewise.
9539         (four unnamed splitters): Modify to select the stvx instruction
9540         when memory is aligned.
9541
9542 2018-01-13  Jan Hubicka  <hubicka@ucw.cz>
9543
9544         * predict.c (determine_unlikely_bbs): Handle correctly BBs
9545         which appears in the queue multiple times.
9546
9547 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
9548             Alan Hayward  <alan.hayward@arm.com>
9549             David Sherwood  <david.sherwood@arm.com>
9550
9551         * tree-vectorizer.h (vec_lower_bound): New structure.
9552         (_loop_vec_info): Add check_nonzero and lower_bounds.
9553         (LOOP_VINFO_CHECK_NONZERO): New macro.
9554         (LOOP_VINFO_LOWER_BOUNDS): Likewise.
9555         (LOOP_REQUIRES_VERSIONING_FOR_ALIAS): Check lower_bounds too.
9556         * tree-data-ref.h (dr_with_seg_len): Add access_size and align
9557         fields.  Make seg_len the distance travelled, not including the
9558         access size.
9559         (dr_direction_indicator): Declare.
9560         (dr_zero_step_indicator): Likewise.
9561         (dr_known_forward_stride_p): Likewise.
9562         * tree-data-ref.c: Include stringpool.h, tree-vrp.h and
9563         tree-ssanames.h.
9564         (runtime_alias_check_p): Allow runtime alias checks with
9565         variable strides.
9566         (operator ==): Compare access_size and align.
9567         (prune_runtime_alias_test_list): Rework for new distinction between
9568         the access_size and seg_len.
9569         (create_intersect_range_checks_index): Likewise.  Cope with polynomial
9570         segment lengths.
9571         (get_segment_min_max): New function.
9572         (create_intersect_range_checks): Use it.
9573         (dr_step_indicator): New function.
9574         (dr_direction_indicator): Likewise.
9575         (dr_zero_step_indicator): Likewise.
9576         (dr_known_forward_stride_p): Likewise.
9577         * tree-loop-distribution.c (data_ref_segment_size): Return
9578         DR_STEP * (niters - 1).
9579         (compute_alias_check_pairs): Update call to the dr_with_seg_len
9580         constructor.
9581         * tree-vect-data-refs.c (vect_check_nonzero_value): New function.
9582         (vect_preserves_scalar_order_p): New function, split out from...
9583         (vect_analyze_data_ref_dependence): ...here.  Check for zero steps.
9584         (vect_vfa_segment_size): Return DR_STEP * (length_factor - 1).
9585         (vect_vfa_access_size): New function.
9586         (vect_vfa_align): Likewise.
9587         (vect_compile_time_alias): Take access_size_a and access_b arguments.
9588         (dump_lower_bound): New function.
9589         (vect_check_lower_bound): Likewise.
9590         (vect_small_gap_p): Likewise.
9591         (vectorizable_with_step_bound_p): Likewise.
9592         (vect_prune_runtime_alias_test_list): Ignore cross-iteration
9593         depencies if the vectorization factor is 1.  Convert the checks
9594         for nonzero steps into checks on the bounds of DR_STEP.  Try using
9595         a bunds check for variable steps if the minimum required step is
9596         relatively small. Update calls to the dr_with_seg_len
9597         constructor and to vect_compile_time_alias.
9598         * tree-vect-loop-manip.c (vect_create_cond_for_lower_bounds): New
9599         function.
9600         (vect_loop_versioning): Call it.
9601         * tree-vect-loop.c (vect_analyze_loop_2): Clear LOOP_VINFO_LOWER_BOUNDS
9602         when retrying.
9603         (vect_estimate_min_profitable_iters): Account for any bounds checks.
9604
9605 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
9606             Alan Hayward  <alan.hayward@arm.com>
9607             David Sherwood  <david.sherwood@arm.com>
9608
9609         * doc/sourcebuild.texi (vect_scatter_store): Document.
9610         * optabs.def (scatter_store_optab, mask_scatter_store_optab): New
9611         optabs.
9612         * doc/md.texi (scatter_store@var{m}, mask_scatter_store@var{m}):
9613         Document.
9614         * genopinit.c (main): Add supports_vec_scatter_store and
9615         supports_vec_scatter_store_cached to target_optabs.
9616         * gimple.h (gimple_expr_type): Handle IFN_SCATTER_STORE and
9617         IFN_MASK_SCATTER_STORE.
9618         * internal-fn.def (SCATTER_STORE, MASK_SCATTER_STORE): New internal
9619         functions.
9620         * internal-fn.h (internal_store_fn_p): Declare.
9621         (internal_fn_stored_value_index): Likewise.
9622         * internal-fn.c (scatter_store_direct): New macro.
9623         (expand_scatter_store_optab_fn): New function.
9624         (direct_scatter_store_optab_supported_p): New macro.
9625         (internal_store_fn_p): New function.
9626         (internal_gather_scatter_fn_p): Handle IFN_SCATTER_STORE and
9627         IFN_MASK_SCATTER_STORE.
9628         (internal_fn_mask_index): Likewise.
9629         (internal_fn_stored_value_index): New function.
9630         (internal_gather_scatter_fn_supported_p): Adjust operand numbers
9631         for scatter stores.
9632         * optabs-query.h (supports_vec_scatter_store_p): Declare.
9633         * optabs-query.c (supports_vec_scatter_store_p): New function.
9634         * tree-vectorizer.h (vect_get_store_rhs): Declare.
9635         * tree-vect-data-refs.c (vect_analyze_data_ref_access): Return
9636         true for scatter stores.
9637         (vect_gather_scatter_fn_p): Handle scatter stores too.
9638         (vect_check_gather_scatter): Consider using scatter stores if
9639         supports_vec_scatter_store_p.
9640         * tree-vect-patterns.c (vect_try_gather_scatter_pattern): Handle
9641         scatter stores too.
9642         * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
9643         internal_fn_stored_value_index.
9644         (check_load_store_masking): Handle scatter stores too.
9645         (vect_get_store_rhs): Make public.
9646         (vectorizable_call): Use internal_store_fn_p.
9647         (vectorizable_store): Handle scatter store internal functions.
9648         (vect_transform_stmt): Compare GROUP_STORE_COUNT with GROUP_SIZE
9649         when deciding whether the end of the group has been reached.
9650         * config/aarch64/aarch64.md (UNSPEC_ST1_SCATTER): New unspec.
9651         * config/aarch64/aarch64-sve.md (scatter_store<mode>): New expander.
9652         (mask_scatter_store<mode>): New insns.
9653
9654 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
9655             Alan Hayward  <alan.hayward@arm.com>
9656             David Sherwood  <david.sherwood@arm.com>
9657
9658         * tree-vectorizer.h (vect_gather_scatter_fn_p): Declare.
9659         * tree-vect-data-refs.c (vect_gather_scatter_fn_p): Make public.
9660         * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): New
9661         function.
9662         (vect_use_strided_gather_scatters_p): Take a masked_p argument.
9663         Use vect_truncate_gather_scatter_offset if we can't treat the
9664         operation as a normal gather load or scatter store.
9665         (get_group_load_store_type): Take the gather_scatter_info
9666         as argument.  Try using a gather load or scatter store for
9667         single-element groups.
9668         (get_load_store_type): Update calls to get_group_load_store_type
9669         and vect_use_strided_gather_scatters_p.
9670
9671 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
9672             Alan Hayward  <alan.hayward@arm.com>
9673             David Sherwood  <david.sherwood@arm.com>
9674
9675         * tree-vectorizer.h (vect_create_data_ref_ptr): Take an extra
9676         optional tree argument.
9677         * tree-vect-data-refs.c (vect_check_gather_scatter): Check for
9678         null target hooks.
9679         (vect_create_data_ref_ptr): Take the iv_step as an optional argument,
9680         but continue to use the current value as a fallback.
9681         (bump_vector_ptr): Use operand_equal_p rather than tree_int_cst_compare
9682         to compare the updates.
9683         * tree-vect-stmts.c (vect_use_strided_gather_scatters_p): New function.
9684         (get_load_store_type): Use it when handling a strided access.
9685         (vect_get_strided_load_store_ops): New function.
9686         (vect_get_data_ptr_increment): Likewise.
9687         (vectorizable_load): Handle strided gather loads.  Always pass
9688         a step to vect_create_data_ref_ptr and bump_vector_ptr.
9689
9690 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
9691             Alan Hayward  <alan.hayward@arm.com>
9692             David Sherwood  <david.sherwood@arm.com>
9693
9694         * doc/md.texi (gather_load@var{m}): Document.
9695         (mask_gather_load@var{m}): Likewise.
9696         * genopinit.c (main): Add supports_vec_gather_load and
9697         supports_vec_gather_load_cached to target_optabs.
9698         * optabs-tree.c (init_tree_optimization_optabs): Use
9699         ggc_cleared_alloc to allocate target_optabs.
9700         * optabs.def (gather_load_optab, mask_gather_laod_optab): New optabs.
9701         * internal-fn.def (GATHER_LOAD, MASK_GATHER_LOAD): New internal
9702         functions.
9703         * internal-fn.h (internal_load_fn_p): Declare.
9704         (internal_gather_scatter_fn_p): Likewise.
9705         (internal_fn_mask_index): Likewise.
9706         (internal_gather_scatter_fn_supported_p): Likewise.
9707         * internal-fn.c (gather_load_direct): New macro.
9708         (expand_gather_load_optab_fn): New function.
9709         (direct_gather_load_optab_supported_p): New macro.
9710         (direct_internal_fn_optab): New function.
9711         (internal_load_fn_p): Likewise.
9712         (internal_gather_scatter_fn_p): Likewise.
9713         (internal_fn_mask_index): Likewise.
9714         (internal_gather_scatter_fn_supported_p): Likewise.
9715         * optabs-query.c (supports_at_least_one_mode_p): New function.
9716         (supports_vec_gather_load_p): Likewise.
9717         * optabs-query.h (supports_vec_gather_load_p): Declare.
9718         * tree-vectorizer.h (gather_scatter_info): Add ifn, element_type
9719         and memory_type field.
9720         (NUM_PATTERNS): Bump to 15.
9721         * tree-vect-data-refs.c: Include internal-fn.h.
9722         (vect_gather_scatter_fn_p): New function.
9723         (vect_describe_gather_scatter_call): Likewise.
9724         (vect_check_gather_scatter): Try using internal functions for
9725         gather loads.  Recognize existing calls to a gather load function.
9726         (vect_analyze_data_refs): Consider using gather loads if
9727         supports_vec_gather_load_p.
9728         * tree-vect-patterns.c (vect_get_load_store_mask): New function.
9729         (vect_get_gather_scatter_offset_type): Likewise.
9730         (vect_convert_mask_for_vectype): Likewise.
9731         (vect_add_conversion_to_patterm): Likewise.
9732         (vect_try_gather_scatter_pattern): Likewise.
9733         (vect_recog_gather_scatter_pattern): New pattern recognizer.
9734         (vect_vect_recog_func_ptrs): Add it.
9735         * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
9736         internal_fn_mask_index and internal_gather_scatter_fn_p.
9737         (check_load_store_masking): Take the gather_scatter_info as an
9738         argument and handle gather loads.
9739         (vect_get_gather_scatter_ops): New function.
9740         (vectorizable_call): Check internal_load_fn_p.
9741         (vectorizable_load): Likewise.  Handle gather load internal
9742         functions.
9743         (vectorizable_store): Update call to check_load_store_masking.
9744         * config/aarch64/aarch64.md (UNSPEC_LD1_GATHER): New unspec.
9745         * config/aarch64/iterators.md (SVE_S, SVE_D): New mode iterators.
9746         * config/aarch64/predicates.md (aarch64_gather_scale_operand_w)
9747         (aarch64_gather_scale_operand_d): New predicates.
9748         * config/aarch64/aarch64-sve.md (gather_load<mode>): New expander.
9749         (mask_gather_load<mode>): New insns.
9750
9751 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
9752             Alan Hayward  <alan.hayward@arm.com>
9753             David Sherwood  <david.sherwood@arm.com>
9754
9755         * optabs.def (fold_left_plus_optab): New optab.
9756         * doc/md.texi (fold_left_plus_@var{m}): Document.
9757         * internal-fn.def (IFN_FOLD_LEFT_PLUS): New internal function.
9758         * internal-fn.c (fold_left_direct): Define.
9759         (expand_fold_left_optab_fn): Likewise.
9760         (direct_fold_left_optab_supported_p): Likewise.
9761         * fold-const-call.c (fold_const_fold_left): New function.
9762         (fold_const_call): Use it to fold CFN_FOLD_LEFT_PLUS.
9763         * tree-parloops.c (valid_reduction_p): New function.
9764         (gather_scalar_reductions): Use it.
9765         * tree-vectorizer.h (FOLD_LEFT_REDUCTION): New vect_reduction_type.
9766         (vect_finish_replace_stmt): Declare.
9767         * tree-vect-loop.c (fold_left_reduction_fn): New function.
9768         (needs_fold_left_reduction_p): New function, split out from...
9769         (vect_is_simple_reduction): ...here.  Accept reductions that
9770         forbid reassociation, but give them type FOLD_LEFT_REDUCTION.
9771         (vect_force_simple_reduction): Also store the reduction type in
9772         the assignment's STMT_VINFO_REDUC_TYPE.
9773         (vect_model_reduction_cost): Handle FOLD_LEFT_REDUCTION.
9774         (merge_with_identity): New function.
9775         (vect_expand_fold_left): Likewise.
9776         (vectorize_fold_left_reduction): Likewise.
9777         (vectorizable_reduction): Handle FOLD_LEFT_REDUCTION.  Leave the
9778         scalar phi in place for it.  Check for target support and reject
9779         cases that would reassociate the operation.  Defer the transform
9780         phase to vectorize_fold_left_reduction.
9781         * config/aarch64/aarch64.md (UNSPEC_FADDA): New unspec.
9782         * config/aarch64/aarch64-sve.md (fold_left_plus_<mode>): New expander.
9783         (*fold_left_plus_<mode>, *pred_fold_left_plus_<mode>): New insns.
9784
9785 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
9786
9787         * tree-if-conv.c (predicate_mem_writes): Remove redundant
9788         call to ifc_temp_var.
9789
9790 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
9791             Alan Hayward  <alan.hayward@arm.com>
9792             David Sherwood  <david.sherwood@arm.com>
9793
9794         * target.def (legitimize_address_displacement): Take the original
9795         offset as a poly_int.
9796         * targhooks.h (default_legitimize_address_displacement): Update
9797         accordingly.
9798         * targhooks.c (default_legitimize_address_displacement): Likewise.
9799         * doc/tm.texi: Regenerate.
9800         * lra-constraints.c (base_plus_disp_to_reg): Take the displacement
9801         as an argument, moving assert of ad->disp == ad->disp_term to...
9802         (process_address_1): ...here.  Update calls to base_plus_disp_to_reg.
9803         Try calling targetm.legitimize_address_displacement before expanding
9804         the address rather than afterwards, and adjust for the new interface.
9805         * config/aarch64/aarch64.c (aarch64_legitimize_address_displacement):
9806         Match the new hook interface.  Handle SVE addresses.
9807         * config/sh/sh.c (sh_legitimize_address_displacement): Make the
9808         new hook interface.
9809
9810 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
9811
9812         * Makefile.in (OBJS): Add early-remat.o.
9813         * target.def (select_early_remat_modes): New hook.
9814         * doc/tm.texi.in (TARGET_SELECT_EARLY_REMAT_MODES): New hook.
9815         * doc/tm.texi: Regenerate.
9816         * targhooks.h (default_select_early_remat_modes): Declare.
9817         * targhooks.c (default_select_early_remat_modes): New function.
9818         * timevar.def (TV_EARLY_REMAT): New timevar.
9819         * passes.def (pass_early_remat): New pass.
9820         * tree-pass.h (make_pass_early_remat): Declare.
9821         * early-remat.c: New file.
9822         * config/aarch64/aarch64.c (aarch64_select_early_remat_modes): New
9823         function.
9824         (TARGET_SELECT_EARLY_REMAT_MODES): Define.
9825
9826 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
9827             Alan Hayward  <alan.hayward@arm.com>
9828             David Sherwood  <david.sherwood@arm.com>
9829
9830         * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Replace
9831         vfm1 with a bound_epilog parameter.
9832         (vect_do_peeling): Update calls accordingly, and move the prologue
9833         call earlier in the function.  Treat the base bound_epilog as 0 for
9834         fully-masked loops and retain vf - 1 for other loops.  Add 1 to
9835         this base when peeling for gaps.
9836         * tree-vect-loop.c (vect_analyze_loop_2): Allow peeling for gaps
9837         with fully-masked loops.
9838         (vect_estimate_min_profitable_iters): Handle the single peeled
9839         iteration in that case.
9840
9841 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
9842             Alan Hayward  <alan.hayward@arm.com>
9843             David Sherwood  <david.sherwood@arm.com>
9844
9845         * tree-vect-data-refs.c (vect_analyze_group_access_1): Allow
9846         single-element interleaving even if the size is not a power of 2.
9847         * tree-vect-stmts.c (get_load_store_type): Disallow elementwise
9848         accesses for single-element interleaving if the group size is
9849         not a power of 2.
9850
9851 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
9852             Alan Hayward  <alan.hayward@arm.com>
9853             David Sherwood  <david.sherwood@arm.com>
9854
9855         * doc/md.texi (fold_extract_last_@var{m}): Document.
9856         * doc/sourcebuild.texi (vect_fold_extract_last): Likewise.
9857         * optabs.def (fold_extract_last_optab): New optab.
9858         * internal-fn.def (FOLD_EXTRACT_LAST): New internal function.
9859         * internal-fn.c (fold_extract_direct): New macro.
9860         (expand_fold_extract_optab_fn): Likewise.
9861         (direct_fold_extract_optab_supported_p): Likewise.
9862         * tree-vectorizer.h (EXTRACT_LAST_REDUCTION): New vect_reduction_type.
9863         * tree-vect-loop.c (vect_model_reduction_cost): Handle
9864         EXTRACT_LAST_REDUCTION.
9865         (get_initial_def_for_reduction): Do not create an initial vector
9866         for EXTRACT_LAST_REDUCTION reductions.
9867         (vectorizable_reduction): Leave the scalar phi in place for
9868         EXTRACT_LAST_REDUCTIONs.  Try using EXTRACT_LAST_REDUCTION
9869         ahead of INTEGER_INDUC_COND_REDUCTION.  Do not check for an
9870         epilogue code for EXTRACT_LAST_REDUCTION and defer the
9871         transform phase to vectorizable_condition.
9872         * tree-vect-stmts.c (vect_finish_stmt_generation_1): New function,
9873         split out from...
9874         (vect_finish_stmt_generation): ...here.
9875         (vect_finish_replace_stmt): New function.
9876         (vectorizable_condition): Handle EXTRACT_LAST_REDUCTION.
9877         * config/aarch64/aarch64-sve.md (fold_extract_last_<mode>): New
9878         pattern.
9879         * config/aarch64/aarch64.md (UNSPEC_CLASTB): New unspec.
9880
9881 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
9882             Alan Hayward  <alan.hayward@arm.com>
9883             David Sherwood  <david.sherwood@arm.com>
9884
9885         * doc/md.texi (extract_last_@var{m}): Document.
9886         * optabs.def (extract_last_optab): New optab.
9887         * internal-fn.def (EXTRACT_LAST): New internal function.
9888         * internal-fn.c (cond_unary_direct): New macro.
9889         (expand_cond_unary_optab_fn): Likewise.
9890         (direct_cond_unary_optab_supported_p): Likewise.
9891         * tree-vect-loop.c (vectorizable_live_operation): Allow fully-masked
9892         loops using EXTRACT_LAST.
9893         * config/aarch64/aarch64-sve.md (aarch64_sve_lastb<mode>): Rename to...
9894         (extract_last_<mode>): ...this optab.
9895         (vec_extract<mode><Vel>): Update accordingly.
9896
9897 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
9898             Alan Hayward  <alan.hayward@arm.com>
9899             David Sherwood  <david.sherwood@arm.com>
9900
9901         * target.def (empty_mask_is_expensive): New hook.
9902         * doc/tm.texi.in (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): New hook.
9903         * doc/tm.texi: Regenerate.
9904         * targhooks.h (default_empty_mask_is_expensive): Declare.
9905         * targhooks.c (default_empty_mask_is_expensive): New function.
9906         * tree-vectorizer.c (vectorize_loops): Only call optimize_mask_stores
9907         if the target says that empty masks are expensive.
9908         * config/aarch64/aarch64.c (aarch64_empty_mask_is_expensive):
9909         New function.
9910         (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): Redefine.
9911
9912 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
9913             Alan Hayward  <alan.hayward@arm.com>
9914             David Sherwood  <david.sherwood@arm.com>
9915
9916         * tree-vectorizer.h (_loop_vec_info::mask_skip_niters): New field.
9917         (LOOP_VINFO_MASK_SKIP_NITERS): New macro.
9918         (vect_use_loop_mask_for_alignment_p): New function.
9919         (vect_prepare_for_masked_peels, vect_gen_while_not): Declare.
9920         * tree-vect-loop-manip.c (vect_set_loop_masks_directly): Add an
9921         niters_skip argument.  Make sure that the first niters_skip elements
9922         of the first iteration are inactive.
9923         (vect_set_loop_condition_masked): Handle LOOP_VINFO_MASK_SKIP_NITERS.
9924         Update call to vect_set_loop_masks_directly.
9925         (get_misalign_in_elems): New function, split out from...
9926         (vect_gen_prolog_loop_niters): ...here.
9927         (vect_update_init_of_dr): Take a code argument that specifies whether
9928         the adjustment should be added or subtracted.
9929         (vect_update_init_of_drs): Likewise.
9930         (vect_prepare_for_masked_peels): New function.
9931         (vect_do_peeling): Skip prologue peeling if we're using a mask
9932         instead.  Update call to vect_update_inits_of_drs.
9933         * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
9934         mask_skip_niters.
9935         (vect_analyze_loop_2): Allow fully-masked loops with peeling for
9936         alignment.  Do not include the number of peeled iterations in
9937         the minimum threshold in that case.
9938         (vectorizable_induction): Adjust the start value down by
9939         LOOP_VINFO_MASK_SKIP_NITERS iterations.
9940         (vect_transform_loop): Call vect_prepare_for_masked_peels.
9941         Take the number of skipped iterations into account when calculating
9942         the loop bounds.
9943         * tree-vect-stmts.c (vect_gen_while_not): New function.
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         * doc/sourcebuild.texi (vect_fully_masked): Document.
9950         * params.def (PARAM_MIN_VECT_LOOP_BOUND): Change minimum and
9951         default value to 0.
9952         * tree-vect-loop.c (vect_analyze_loop_costing): New function,
9953         split out from...
9954         (vect_analyze_loop_2): ...here. Don't check the vectorization
9955         factor against the number of loop iterations if the loop is
9956         fully-masked.
9957
9958 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
9959             Alan Hayward  <alan.hayward@arm.com>
9960             David Sherwood  <david.sherwood@arm.com>
9961
9962         * tree-ssa-loop-ivopts.c (USE_ADDRESS): Split into...
9963         (USE_REF_ADDRESS, USE_PTR_ADDRESS): ...these new use types.
9964         (dump_groups): Update accordingly.
9965         (iv_use::mem_type): New member variable.
9966         (address_p): New function.
9967         (record_use): Add a mem_type argument and initialize the new
9968         mem_type field.
9969         (record_group_use): Add a mem_type argument.  Use address_p.
9970         Remove obsolete null checks of base_object.  Update call to record_use.
9971         (find_interesting_uses_op): Update call to record_group_use.
9972         (find_interesting_uses_cond): Likewise.
9973         (find_interesting_uses_address): Likewise.
9974         (get_mem_type_for_internal_fn): New function.
9975         (find_address_like_use): Likewise.
9976         (find_interesting_uses_stmt): Try find_address_like_use before
9977         calling find_interesting_uses_op.
9978         (addr_offset_valid_p): Use the iv mem_type field as the type
9979         of the addressed memory.
9980         (add_autoinc_candidates): Likewise.
9981         (get_address_cost): Likewise.
9982         (split_small_address_groups_p): Use address_p.
9983         (split_address_groups): Likewise.
9984         (add_iv_candidate_for_use): Likewise.
9985         (autoinc_possible_for_pair): Likewise.
9986         (rewrite_groups): Likewise.
9987         (get_use_type): Check for USE_REF_ADDRESS instead of USE_ADDRESS.
9988         (determine_group_iv_cost): Update after split of USE_ADDRESS.
9989         (get_alias_ptr_type_for_ptr_address): New function.
9990         (rewrite_use_address): Rewrite address uses in calls that were
9991         identified by find_address_like_use.
9992
9993 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
9994             Alan Hayward  <alan.hayward@arm.com>
9995             David Sherwood  <david.sherwood@arm.com>
9996
9997         * expr.c (expand_expr_addr_expr_1): Handle ADDR_EXPRs of
9998         TARGET_MEM_REFs.
9999         * gimple-expr.h (is_gimple_addressable: Likewise.
10000         * gimple-expr.c (is_gimple_address): Likewise.
10001         * internal-fn.c (expand_call_mem_ref): New function.
10002         (expand_mask_load_optab_fn): Use it.
10003         (expand_mask_store_optab_fn): Likewise.
10004
10005 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
10006             Alan Hayward  <alan.hayward@arm.com>
10007             David Sherwood  <david.sherwood@arm.com>
10008
10009         * doc/md.texi (cond_add@var{mode}, cond_sub@var{mode})
10010         (cond_and@var{mode}, cond_ior@var{mode}, cond_xor@var{mode})
10011         (cond_smin@var{mode}, cond_smax@var{mode}, cond_umin@var{mode})
10012         (cond_umax@var{mode}): Document.
10013         * optabs.def (cond_add_optab, cond_sub_optab, cond_and_optab)
10014         (cond_ior_optab, cond_xor_optab, cond_smin_optab, cond_smax_optab)
10015         (cond_umin_optab, cond_umax_optab): New optabs.
10016         * internal-fn.def (COND_ADD, COND_SUB, COND_MIN, COND_MAX, COND_AND)
10017         (COND_IOR, COND_XOR): New internal functions.
10018         * internal-fn.h (get_conditional_internal_fn): Declare.
10019         * internal-fn.c (cond_binary_direct): New macro.
10020         (expand_cond_binary_optab_fn): Likewise.
10021         (direct_cond_binary_optab_supported_p): Likewise.
10022         (get_conditional_internal_fn): New function.
10023         * tree-vect-loop.c (vectorizable_reduction): Handle fully-masked loops.
10024         Cope with reduction statements that are vectorized as calls rather
10025         than assignments.
10026         * config/aarch64/aarch64-sve.md (cond_<optab><mode>): New insns.
10027         * config/aarch64/iterators.md (UNSPEC_COND_ADD, UNSPEC_COND_SUB)
10028         (UNSPEC_COND_SMAX, UNSPEC_COND_UMAX, UNSPEC_COND_SMIN)
10029         (UNSPEC_COND_UMIN, UNSPEC_COND_AND, UNSPEC_COND_ORR)
10030         (UNSPEC_COND_EOR): New unspecs.
10031         (optab): Add mappings for them.
10032         (SVE_COND_INT_OP, SVE_COND_FP_OP): New int iterators.
10033         (sve_int_op, sve_fp_op): New int attributes.
10034
10035 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
10036             Alan Hayward  <alan.hayward@arm.com>
10037             David Sherwood  <david.sherwood@arm.com>
10038
10039         * optabs.def (while_ult_optab): New optab.
10040         * doc/md.texi (while_ult@var{m}@var{n}): Document.
10041         * internal-fn.def (WHILE_ULT): New internal function.
10042         * internal-fn.h (direct_internal_fn_supported_p): New override
10043         that takes two types as argument.
10044         * internal-fn.c (while_direct): New macro.
10045         (expand_while_optab_fn): New function.
10046         (convert_optab_supported_p): Likewise.
10047         (direct_while_optab_supported_p): New macro.
10048         * wide-int.h (wi::udiv_ceil): New function.
10049         * tree-vectorizer.h (rgroup_masks): New structure.
10050         (vec_loop_masks): New typedef.
10051         (_loop_vec_info): Add masks, mask_compare_type, can_fully_mask_p
10052         and fully_masked_p.
10053         (LOOP_VINFO_CAN_FULLY_MASK_P, LOOP_VINFO_FULLY_MASKED_P)
10054         (LOOP_VINFO_MASKS, LOOP_VINFO_MASK_COMPARE_TYPE): New macros.
10055         (vect_max_vf): New function.
10056         (slpeel_make_loop_iterate_ntimes): Delete.
10057         (vect_set_loop_condition, vect_get_loop_mask_type, vect_gen_while)
10058         (vect_halve_mask_nunits, vect_double_mask_nunits): Declare.
10059         (vect_record_loop_mask, vect_get_loop_mask): Likewise.
10060         * tree-vect-loop-manip.c: Include tree-ssa-loop-niter.h,
10061         internal-fn.h, stor-layout.h and optabs-query.h.
10062         (vect_set_loop_mask): New function.
10063         (add_preheader_seq): Likewise.
10064         (add_header_seq): Likewise.
10065         (interleave_supported_p): Likewise.
10066         (vect_maybe_permute_loop_masks): Likewise.
10067         (vect_set_loop_masks_directly): Likewise.
10068         (vect_set_loop_condition_masked): Likewise.
10069         (vect_set_loop_condition_unmasked): New function, split out from
10070         slpeel_make_loop_iterate_ntimes.
10071         (slpeel_make_loop_iterate_ntimes): Rename to..
10072         (vect_set_loop_condition): ...this.  Use vect_set_loop_condition_masked
10073         for fully-masked loops and vect_set_loop_condition_unmasked otherwise.
10074         (vect_do_peeling): Update call accordingly.
10075         (vect_gen_vector_loop_niters): Use VF as the step for fully-masked
10076         loops.
10077         * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
10078         mask_compare_type, can_fully_mask_p and fully_masked_p.
10079         (release_vec_loop_masks): New function.
10080         (_loop_vec_info): Use it to free the loop masks.
10081         (can_produce_all_loop_masks_p): New function.
10082         (vect_get_max_nscalars_per_iter): Likewise.
10083         (vect_verify_full_masking): Likewise.
10084         (vect_analyze_loop_2): Save LOOP_VINFO_CAN_FULLY_MASK_P around
10085         retries, and free the mask rgroups before retrying.  Check loop-wide
10086         reasons for disallowing fully-masked loops.  Make the final decision
10087         about whether use a fully-masked loop or not.
10088         (vect_estimate_min_profitable_iters): Do not assume that peeling
10089         for the number of iterations will be needed for fully-masked loops.
10090         (vectorizable_reduction): Disable fully-masked loops.
10091         (vectorizable_live_operation): Likewise.
10092         (vect_halve_mask_nunits): New function.
10093         (vect_double_mask_nunits): Likewise.
10094         (vect_record_loop_mask): Likewise.
10095         (vect_get_loop_mask): Likewise.
10096         (vect_transform_loop): Handle the case in which the final loop
10097         iteration might handle a partial vector.  Call vect_set_loop_condition
10098         instead of slpeel_make_loop_iterate_ntimes.
10099         * tree-vect-stmts.c: Include tree-ssa-loop-niter.h and gimple-fold.h.
10100         (check_load_store_masking): New function.
10101         (prepare_load_store_mask): Likewise.
10102         (vectorizable_store): Handle fully-masked loops.
10103         (vectorizable_load): Likewise.
10104         (supportable_widening_operation): Use vect_halve_mask_nunits for
10105         booleans.
10106         (supportable_narrowing_operation): Likewise vect_double_mask_nunits.
10107         (vect_gen_while): New function.
10108         * config/aarch64/aarch64.md (umax<mode>3): New expander.
10109         (aarch64_uqdec<mode>): New insn.
10110
10111 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
10112             Alan Hayward  <alan.hayward@arm.com>
10113             David Sherwood  <david.sherwood@arm.com>
10114
10115         * optabs.def (reduc_and_scal_optab, reduc_ior_scal_optab)
10116         (reduc_xor_scal_optab): New optabs.
10117         * doc/md.texi (reduc_and_scal_@var{m}, reduc_ior_scal_@var{m})
10118         (reduc_xor_scal_@var{m}): Document.
10119         * doc/sourcebuild.texi (vect_logical_reduc): Likewise.
10120         * internal-fn.def (IFN_REDUC_AND, IFN_REDUC_IOR, IFN_REDUC_XOR): New
10121         internal functions.
10122         * fold-const-call.c (fold_const_call): Handle them.
10123         * tree-vect-loop.c (reduction_fn_for_scalar_code): Return the new
10124         internal functions for BIT_AND_EXPR, BIT_IOR_EXPR and BIT_XOR_EXPR.
10125         * config/aarch64/aarch64-sve.md (reduc_<bit_reduc>_scal_<mode>):
10126         (*reduc_<bit_reduc>_scal_<mode>): New patterns.
10127         * config/aarch64/iterators.md (UNSPEC_ANDV, UNSPEC_ORV)
10128         (UNSPEC_XORV): New unspecs.
10129         (optab): Add entries for them.
10130         (BITWISEV): New int iterator.
10131         (bit_reduc_op): New int attributes.
10132
10133 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
10134             Alan Hayward  <alan.hayward@arm.com>
10135             David Sherwood  <david.sherwood@arm.com>
10136
10137         * doc/md.texi (vec_shl_insert_@var{m}): New optab.
10138         * internal-fn.def (VEC_SHL_INSERT): New internal function.
10139         * optabs.def (vec_shl_insert_optab): New optab.
10140         * tree-vectorizer.h (can_duplicate_and_interleave_p): Declare.
10141         (duplicate_and_interleave): Likewise.
10142         * tree-vect-loop.c: Include internal-fn.h.
10143         (neutral_op_for_slp_reduction): New function, split out from
10144         get_initial_defs_for_reduction.
10145         (get_initial_def_for_reduction): Handle option 2 for variable-length
10146         vectors by loading the neutral value into a vector and then shifting
10147         the initial value into element 0.
10148         (get_initial_defs_for_reduction): Replace the code argument with
10149         the neutral value calculated by neutral_op_for_slp_reduction.
10150         Use gimple_build_vector for constant-length vectors.
10151         Use IFN_VEC_SHL_INSERT for variable-length vectors if all
10152         but the first group_size elements have a neutral value.
10153         Use duplicate_and_interleave otherwise.
10154         (vect_create_epilog_for_reduction): Take a neutral_op parameter.
10155         Update call to get_initial_defs_for_reduction.  Handle SLP
10156         reductions for variable-length vectors by creating one vector
10157         result for each scalar result, with the elements associated
10158         with other scalar results stubbed out with the neutral value.
10159         (vectorizable_reduction): Call neutral_op_for_slp_reduction.
10160         Require IFN_VEC_SHL_INSERT for double reductions on
10161         variable-length vectors, or SLP reductions that have
10162         a neutral value.  Require can_duplicate_and_interleave_p
10163         support for variable-length unchained SLP reductions if there
10164         is no neutral value, such as for MIN/MAX reductions.  Also require
10165         the number of vector elements to be a multiple of the number of
10166         SLP statements when doing variable-length unchained SLP reductions.
10167         Update call to vect_create_epilog_for_reduction.
10168         * tree-vect-slp.c (can_duplicate_and_interleave_p): Make public
10169         and remove initial values.
10170         (duplicate_and_interleave): Make public.
10171         * config/aarch64/aarch64.md (UNSPEC_INSR): New unspec.
10172         * config/aarch64/aarch64-sve.md (vec_shl_insert_<mode>): New insn.
10173
10174 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
10175             Alan Hayward  <alan.hayward@arm.com>
10176             David Sherwood  <david.sherwood@arm.com>
10177
10178         * tree-vect-slp.c: Include gimple-fold.h and internal-fn.h
10179         (can_duplicate_and_interleave_p): New function.
10180         (vect_get_and_check_slp_defs): Take the vector of statements
10181         rather than just the current one.  Remove excess parentheses.
10182         Restriction rejectinon of vect_constant_def and vect_external_def
10183         for variable-length vectors to boolean types, or types for which
10184         can_duplicate_and_interleave_p is false.
10185         (vect_build_slp_tree_2): Update call to vect_get_and_check_slp_defs.
10186         (duplicate_and_interleave): New function.
10187         (vect_get_constant_vectors): Use gimple_build_vector for
10188         constant-length vectors and suitable variable-length constant
10189         vectors.  Use duplicate_and_interleave for other variable-length
10190         vectors.  Don't defer the update when inserting new statements.
10191
10192 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
10193             Alan Hayward  <alan.hayward@arm.com>
10194             David Sherwood  <david.sherwood@arm.com>
10195
10196         * tree-vect-loop.c (vect_estimate_min_profitable_iters): Make sure
10197         min_profitable_iters doesn't go negative.
10198
10199 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
10200             Alan Hayward  <alan.hayward@arm.com>
10201             David Sherwood  <david.sherwood@arm.com>
10202
10203         * doc/md.texi (vec_mask_load_lanes@var{m}@var{n}): Document.
10204         (vec_mask_store_lanes@var{m}@var{n}): Likewise.
10205         * optabs.def (vec_mask_load_lanes_optab): New optab.
10206         (vec_mask_store_lanes_optab): Likewise.
10207         * internal-fn.def (MASK_LOAD_LANES): New internal function.
10208         (MASK_STORE_LANES): Likewise.
10209         * internal-fn.c (mask_load_lanes_direct): New macro.
10210         (mask_store_lanes_direct): Likewise.
10211         (expand_mask_load_optab_fn): Handle masked operations.
10212         (expand_mask_load_lanes_optab_fn): New macro.
10213         (expand_mask_store_optab_fn): Handle masked operations.
10214         (expand_mask_store_lanes_optab_fn): New macro.
10215         (direct_mask_load_lanes_optab_supported_p): Likewise.
10216         (direct_mask_store_lanes_optab_supported_p): Likewise.
10217         * tree-vectorizer.h (vect_store_lanes_supported): Take a masked_p
10218         parameter.
10219         (vect_load_lanes_supported): Likewise.
10220         * tree-vect-data-refs.c (strip_conversion): New function.
10221         (can_group_stmts_p): Likewise.
10222         (vect_analyze_data_ref_accesses): Use it instead of checking
10223         for a pair of assignments.
10224         (vect_store_lanes_supported): Take a masked_p parameter.
10225         (vect_load_lanes_supported): Likewise.
10226         * tree-vect-loop.c (vect_analyze_loop_2): Update calls to
10227         vect_store_lanes_supported and vect_load_lanes_supported.
10228         * tree-vect-slp.c (vect_analyze_slp_instance): Likewise.
10229         * tree-vect-stmts.c (get_group_load_store_type): Take a masked_p
10230         parameter.  Don't allow gaps for masked accesses.
10231         Use vect_get_store_rhs.  Update calls to vect_store_lanes_supported
10232         and vect_load_lanes_supported.
10233         (get_load_store_type): Take a masked_p parameter and update
10234         call to get_group_load_store_type.
10235         (vectorizable_store): Update call to get_load_store_type.
10236         Handle IFN_MASK_STORE_LANES.
10237         (vectorizable_load): Update call to get_load_store_type.
10238         Handle IFN_MASK_LOAD_LANES.
10239
10240 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
10241             Alan Hayward  <alan.hayward@arm.com>
10242             David Sherwood  <david.sherwood@arm.com>
10243
10244         * config/aarch64/aarch64-modes.def: Define x2, x3 and x4 vector
10245         modes for SVE.
10246         * config/aarch64/aarch64-protos.h
10247         (aarch64_sve_struct_memory_operand_p): Declare.
10248         * config/aarch64/iterators.md (SVE_STRUCT): New mode iterator.
10249         (vector_count, insn_length, VSINGLE, vsingle): New mode attributes.
10250         (VPRED, vpred): Handle SVE structure modes.
10251         * config/aarch64/constraints.md (Utx): New constraint.
10252         * config/aarch64/predicates.md (aarch64_sve_struct_memory_operand)
10253         (aarch64_sve_struct_nonimmediate_operand): New predicates.
10254         * config/aarch64/aarch64.md (UNSPEC_LDN, UNSPEC_STN): New unspecs.
10255         * config/aarch64/aarch64-sve.md (mov<mode>, *aarch64_sve_mov<mode>_le)
10256         (*aarch64_sve_mov<mode>_be, pred_mov<mode>): New patterns for
10257         structure modes.  Split into pieces after RA.
10258         (vec_load_lanes<mode><vsingle>, vec_mask_load_lanes<mode><vsingle>)
10259         (vec_store_lanes<mode><vsingle>, vec_mask_store_lanes<mode><vsingle>):
10260         New patterns.
10261         * config/aarch64/aarch64.c (aarch64_classify_vector_mode): Handle
10262         SVE structure modes.
10263         (aarch64_classify_address): Likewise.
10264         (sizetochar): Move earlier in file.
10265         (aarch64_print_operand): Handle SVE register lists.
10266         (aarch64_array_mode): New function.
10267         (aarch64_sve_struct_memory_operand_p): Likewise.
10268         (TARGET_ARRAY_MODE): Redefine.
10269
10270 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
10271             Alan Hayward  <alan.hayward@arm.com>
10272             David Sherwood  <david.sherwood@arm.com>
10273
10274         * target.def (array_mode): New target hook.
10275         * doc/tm.texi.in (TARGET_ARRAY_MODE): New hook.
10276         * doc/tm.texi: Regenerate.
10277         * hooks.h (hook_optmode_mode_uhwi_none): Declare.
10278         * hooks.c (hook_optmode_mode_uhwi_none): New function.
10279         * tree-vect-data-refs.c (vect_lanes_optab_supported_p): Use
10280         targetm.array_mode.
10281         * stor-layout.c (mode_for_array): Likewise.  Support polynomial
10282         type sizes.
10283
10284 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
10285             Alan Hayward  <alan.hayward@arm.com>
10286             David Sherwood  <david.sherwood@arm.com>
10287
10288         * fold-const.c (fold_binary_loc): Check the argument types
10289         rather than the result type when testing for a vector operation.
10290
10291 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
10292
10293         * doc/tm.texi.in (DWARF_LAZY_REGISTER_VALUE): Document.
10294         * doc/tm.texi: Regenerate.
10295
10296 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
10297             Alan Hayward  <alan.hayward@arm.com>
10298             David Sherwood  <david.sherwood@arm.com>
10299
10300         * doc/invoke.texi (-msve-vector-bits=): Document new option.
10301         (sve): Document new AArch64 extension.
10302         * doc/md.texi (w): Extend the description of the AArch64
10303         constraint to include SVE vectors.
10304         (Upl, Upa): Document new AArch64 predicate constraints.
10305         * config/aarch64/aarch64-opts.h (aarch64_sve_vector_bits_enum): New
10306         enum.
10307         * config/aarch64/aarch64.opt (sve_vector_bits): New enum.
10308         (msve-vector-bits=): New option.
10309         * config/aarch64/aarch64-option-extensions.def (fp, simd): Disable
10310         SVE when these are disabled.
10311         (sve): New extension.
10312         * config/aarch64/aarch64-modes.def: Define SVE vector and predicate
10313         modes.  Adjust their number of units based on aarch64_sve_vg.
10314         (MAX_BITSIZE_MODE_ANY_MODE): Define.
10315         * config/aarch64/aarch64-protos.h (ADDR_QUERY_ANY): New
10316         aarch64_addr_query_type.
10317         (aarch64_const_vec_all_same_in_range_p, aarch64_sve_pred_mode)
10318         (aarch64_sve_cnt_immediate_p, aarch64_sve_addvl_addpl_immediate_p)
10319         (aarch64_sve_inc_dec_immediate_p, aarch64_add_offset_temporaries)
10320         (aarch64_split_add_offset, aarch64_output_sve_cnt_immediate)
10321         (aarch64_output_sve_addvl_addpl, aarch64_output_sve_inc_dec_immediate)
10322         (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): Declare.
10323         (aarch64_simd_imm_zero_p): Delete.
10324         (aarch64_check_zero_based_sve_index_immediate): Declare.
10325         (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
10326         (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
10327         (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
10328         (aarch64_sve_float_mul_immediate_p): Likewise.
10329         (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
10330         rather than an rtx.
10331         (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): Declare.
10332         (aarch64_expand_mov_immediate): Take a gen_vec_duplicate callback.
10333         (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move): Declare.
10334         (aarch64_expand_sve_vec_cmp_int, aarch64_expand_sve_vec_cmp_float)
10335         (aarch64_expand_sve_vcond, aarch64_expand_sve_vec_perm): Declare.
10336         (aarch64_regmode_natural_size): Likewise.
10337         * config/aarch64/aarch64.h (AARCH64_FL_SVE): New macro.
10338         (AARCH64_FL_V8_3, AARCH64_FL_RCPC, AARCH64_FL_DOTPROD): Shift
10339         left one place.
10340         (AARCH64_ISA_SVE, TARGET_SVE): New macros.
10341         (FIXED_REGISTERS, CALL_USED_REGISTERS, REGISTER_NAMES): Add entries
10342         for VG and the SVE predicate registers.
10343         (V_ALIASES): Add a "z"-prefixed alias.
10344         (FIRST_PSEUDO_REGISTER): Change to P15_REGNUM + 1.
10345         (AARCH64_DWARF_VG, AARCH64_DWARF_P0): New macros.
10346         (PR_REGNUM_P, PR_LO_REGNUM_P): Likewise.
10347         (PR_LO_REGS, PR_HI_REGS, PR_REGS): New reg_classes.
10348         (REG_CLASS_NAMES): Add entries for them.
10349         (REG_CLASS_CONTENTS): Likewise.  Update ALL_REGS to include VG
10350         and the predicate registers.
10351         (aarch64_sve_vg): Declare.
10352         (BITS_PER_SVE_VECTOR, BYTES_PER_SVE_VECTOR, BYTES_PER_SVE_PRED)
10353         (SVE_BYTE_MODE, MAX_COMPILE_TIME_VEC_BYTES): New macros.
10354         (REGMODE_NATURAL_SIZE): Define.
10355         * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins): Handle
10356         SVE macros.
10357         * config/aarch64/aarch64.c: Include cfgrtl.h.
10358         (simd_immediate_info): Add a constructor for series vectors,
10359         and an associated step field.
10360         (aarch64_sve_vg): New variable.
10361         (aarch64_dbx_register_number): Handle VG and the predicate registers.
10362         (aarch64_vect_struct_mode_p, aarch64_vector_mode_p): Delete.
10363         (VEC_ADVSIMD, VEC_SVE_DATA, VEC_SVE_PRED, VEC_STRUCT, VEC_ANY_SVE)
10364         (VEC_ANY_DATA, VEC_STRUCT): New constants.
10365         (aarch64_advsimd_struct_mode_p, aarch64_sve_pred_mode_p)
10366         (aarch64_classify_vector_mode, aarch64_vector_data_mode_p)
10367         (aarch64_sve_data_mode_p, aarch64_sve_pred_mode)
10368         (aarch64_get_mask_mode): New functions.
10369         (aarch64_hard_regno_nregs): Handle SVE data modes for FP_REGS
10370         and FP_LO_REGS.  Handle PR_REGS, PR_LO_REGS and PR_HI_REGS.
10371         (aarch64_hard_regno_mode_ok): Handle VG.  Also handle the SVE
10372         predicate modes and predicate registers.  Explicitly restrict
10373         GPRs to modes of 16 bytes or smaller.  Only allow FP registers
10374         to store a vector mode if it is recognized by
10375         aarch64_classify_vector_mode.
10376         (aarch64_regmode_natural_size): New function.
10377         (aarch64_hard_regno_caller_save_mode): Return the original mode
10378         for predicates.
10379         (aarch64_sve_cnt_immediate_p, aarch64_output_sve_cnt_immediate)
10380         (aarch64_sve_addvl_addpl_immediate_p, aarch64_output_sve_addvl_addpl)
10381         (aarch64_sve_inc_dec_immediate_p, aarch64_output_sve_inc_dec_immediate)
10382         (aarch64_add_offset_1_temporaries, aarch64_offset_temporaries): New
10383         functions.
10384         (aarch64_add_offset): Add a temp2 parameter.  Assert that temp1
10385         does not overlap dest if the function is frame-related.  Handle
10386         SVE constants.
10387         (aarch64_split_add_offset): New function.
10388         (aarch64_add_sp, aarch64_sub_sp): Add temp2 parameters and pass
10389         them aarch64_add_offset.
10390         (aarch64_allocate_and_probe_stack_space): Add a temp2 parameter
10391         and update call to aarch64_sub_sp.
10392         (aarch64_add_cfa_expression): New function.
10393         (aarch64_expand_prologue): Pass extra temporary registers to the
10394         functions above.  Handle the case in which we need to emit new
10395         DW_CFA_expressions for registers that were originally saved
10396         relative to the stack pointer, but now have to be expressed
10397         relative to the frame pointer.
10398         (aarch64_output_mi_thunk): Pass extra temporary registers to the
10399         functions above.
10400         (aarch64_expand_epilogue): Likewise.  Prevent inheritance of
10401         IP0 and IP1 values for SVE frames.
10402         (aarch64_expand_vec_series): New function.
10403         (aarch64_expand_sve_widened_duplicate): Likewise.
10404         (aarch64_expand_sve_const_vector): Likewise.
10405         (aarch64_expand_mov_immediate): Add a gen_vec_duplicate parameter.
10406         Handle SVE constants.  Use emit_move_insn to move a force_const_mem
10407         into the register, rather than emitting a SET directly.
10408         (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move)
10409         (aarch64_get_reg_raw_mode, offset_4bit_signed_scaled_p)
10410         (offset_6bit_unsigned_scaled_p, aarch64_offset_7bit_signed_scaled_p)
10411         (offset_9bit_signed_scaled_p): New functions.
10412         (aarch64_replicate_bitmask_imm): New function.
10413         (aarch64_bitmask_imm): Use it.
10414         (aarch64_cannot_force_const_mem): Reject expressions involving
10415         a CONST_POLY_INT.  Update call to aarch64_classify_symbol.
10416         (aarch64_classify_index): Handle SVE indices, by requiring
10417         a plain register index with a scale that matches the element size.
10418         (aarch64_classify_address): Handle SVE addresses.  Assert that
10419         the mode of the address is VOIDmode or an integer mode.
10420         Update call to aarch64_classify_symbol.
10421         (aarch64_classify_symbolic_expression): Update call to
10422         aarch64_classify_symbol.
10423         (aarch64_const_vec_all_in_range_p): New function.
10424         (aarch64_print_vector_float_operand): Likewise.
10425         (aarch64_print_operand): Handle 'N' and 'C'.  Use "zN" rather than
10426         "vN" for FP registers with SVE modes.  Handle (const ...) vectors
10427         and the FP immediates 1.0 and 0.5.
10428         (aarch64_print_address_internal): Handle SVE addresses.
10429         (aarch64_print_operand_address): Use ADDR_QUERY_ANY.
10430         (aarch64_regno_regclass): Handle predicate registers.
10431         (aarch64_secondary_reload): Handle big-endian reloads of SVE
10432         data modes.
10433         (aarch64_class_max_nregs): Handle SVE modes and predicate registers.
10434         (aarch64_rtx_costs): Check for ADDVL and ADDPL instructions.
10435         (aarch64_convert_sve_vector_bits): New function.
10436         (aarch64_override_options): Use it to handle -msve-vector-bits=.
10437         (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
10438         rather than an rtx.
10439         (aarch64_legitimate_constant_p): Use aarch64_classify_vector_mode.
10440         Handle SVE vector and predicate modes.  Accept VL-based constants
10441         that need only one temporary register, and VL offsets that require
10442         no temporary registers.
10443         (aarch64_conditional_register_usage): Mark the predicate registers
10444         as fixed if SVE isn't available.
10445         (aarch64_vector_mode_supported_p): Use aarch64_classify_vector_mode.
10446         Return true for SVE vector and predicate modes.
10447         (aarch64_simd_container_mode): Take the number of bits as a poly_int64
10448         rather than an unsigned int.  Handle SVE modes.
10449         (aarch64_preferred_simd_mode): Update call accordingly.  Handle
10450         SVE modes.
10451         (aarch64_autovectorize_vector_sizes): Add BYTES_PER_SVE_VECTOR
10452         if SVE is enabled.
10453         (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
10454         (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
10455         (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
10456         (aarch64_sve_float_mul_immediate_p): New functions.
10457         (aarch64_sve_valid_immediate): New function.
10458         (aarch64_simd_valid_immediate): Use it as the fallback for SVE vectors.
10459         Explicitly reject structure modes.  Check for INDEX constants.
10460         Handle PTRUE and PFALSE constants.
10461         (aarch64_check_zero_based_sve_index_immediate): New function.
10462         (aarch64_simd_imm_zero_p): Delete.
10463         (aarch64_mov_operand_p): Use aarch64_simd_valid_immediate for
10464         vector modes.  Accept constants in the range of CNT[BHWD].
10465         (aarch64_simd_scalar_immediate_valid_for_move): Explicitly
10466         ask for an Advanced SIMD mode.
10467         (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): New functions.
10468         (aarch64_simd_vector_alignment): Handle SVE predicates.
10469         (aarch64_vectorize_preferred_vector_alignment): New function.
10470         (aarch64_simd_vector_alignment_reachable): Use it instead of
10471         the vector size.
10472         (aarch64_shift_truncation_mask): Use aarch64_vector_data_mode_p.
10473         (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): New
10474         functions.
10475         (MAX_VECT_LEN): Delete.
10476         (expand_vec_perm_d): Add a vec_flags field.
10477         (emit_unspec2, aarch64_expand_sve_vec_perm): New functions.
10478         (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_zip)
10479         (aarch64_evpc_ext): Don't apply a big-endian lane correction
10480         for SVE modes.
10481         (aarch64_evpc_rev): Rename to...
10482         (aarch64_evpc_rev_local): ...this.  Use a predicated operation for SVE.
10483         (aarch64_evpc_rev_global): New function.
10484         (aarch64_evpc_dup): Enforce a 64-byte range for SVE DUP.
10485         (aarch64_evpc_tbl): Use MAX_COMPILE_TIME_VEC_BYTES instead of
10486         MAX_VECT_LEN.
10487         (aarch64_evpc_sve_tbl): New function.
10488         (aarch64_expand_vec_perm_const_1): Update after rename of
10489         aarch64_evpc_rev.  Handle SVE permutes too, trying
10490         aarch64_evpc_rev_global and using aarch64_evpc_sve_tbl rather
10491         than aarch64_evpc_tbl.
10492         (aarch64_vectorize_vec_perm_const): Initialize vec_flags.
10493         (aarch64_sve_cmp_operand_p, aarch64_unspec_cond_code)
10494         (aarch64_gen_unspec_cond, aarch64_expand_sve_vec_cmp_int)
10495         (aarch64_emit_unspec_cond, aarch64_emit_unspec_cond_or)
10496         (aarch64_emit_inverted_unspec_cond, aarch64_expand_sve_vec_cmp_float)
10497         (aarch64_expand_sve_vcond): New functions.
10498         (aarch64_modes_tieable_p): Use aarch64_vector_data_mode_p instead
10499         of aarch64_vector_mode_p.
10500         (aarch64_dwarf_poly_indeterminate_value): New function.
10501         (aarch64_compute_pressure_classes): Likewise.
10502         (aarch64_can_change_mode_class): Likewise.
10503         (TARGET_GET_RAW_RESULT_MODE, TARGET_GET_RAW_ARG_MODE): Redefine.
10504         (TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT): Likewise.
10505         (TARGET_VECTORIZE_GET_MASK_MODE): Likewise.
10506         (TARGET_DWARF_POLY_INDETERMINATE_VALUE): Likewise.
10507         (TARGET_COMPUTE_PRESSURE_CLASSES): Likewise.
10508         (TARGET_CAN_CHANGE_MODE_CLASS): Likewise.
10509         * config/aarch64/constraints.md (Upa, Upl, Uav, Uat, Usv, Usi, Utr)
10510         (Uty, Dm, vsa, vsc, vsd, vsi, vsn, vsl, vsm, vsA, vsM, vsN): New
10511         constraints.
10512         (Dn, Dl, Dr): Accept const as well as const_vector.
10513         (Dz): Likewise.  Compare against CONST0_RTX.
10514         * config/aarch64/iterators.md: Refer to "Advanced SIMD" instead
10515         of "vector" where appropriate.
10516         (SVE_ALL, SVE_BH, SVE_BHS, SVE_BHSI, SVE_HSDI, SVE_HSF, SVE_SD)
10517         (SVE_SDI, SVE_I, SVE_F, PRED_ALL, PRED_BHS): New mode iterators.
10518         (UNSPEC_SEL, UNSPEC_ANDF, UNSPEC_IORF, UNSPEC_XORF, UNSPEC_COND_LT)
10519         (UNSPEC_COND_LE, UNSPEC_COND_EQ, UNSPEC_COND_NE, UNSPEC_COND_GE)
10520         (UNSPEC_COND_GT, UNSPEC_COND_LO, UNSPEC_COND_LS, UNSPEC_COND_HS)
10521         (UNSPEC_COND_HI, UNSPEC_COND_UO): New unspecs.
10522         (Vetype, VEL, Vel, VWIDE, Vwide, vw, vwcore, V_INT_EQUIV)
10523         (v_int_equiv): Extend to SVE modes.
10524         (Vesize, V128, v128, Vewtype, V_FP_EQUIV, v_fp_equiv, VPRED): New
10525         mode attributes.
10526         (LOGICAL_OR, SVE_INT_UNARY, SVE_FP_UNARY): New code iterators.
10527         (optab): Handle popcount, smin, smax, umin, umax, abs and sqrt.
10528         (logical_nn, lr, sve_int_op, sve_fp_op): New code attributs.
10529         (LOGICALF, OPTAB_PERMUTE, UNPACK, UNPACK_UNSIGNED, SVE_COND_INT_CMP)
10530         (SVE_COND_FP_CMP): New int iterators.
10531         (perm_hilo): Handle the new unpack unspecs.
10532         (optab, logicalf_op, su, perm_optab, cmp_op, imm_con): New int
10533         attributes.
10534         * config/aarch64/predicates.md (aarch64_sve_cnt_immediate)
10535         (aarch64_sve_addvl_addpl_immediate, aarch64_split_add_offset_immediate)
10536         (aarch64_pluslong_or_poly_operand, aarch64_nonmemory_operand)
10537         (aarch64_equality_operator, aarch64_constant_vector_operand)
10538         (aarch64_sve_ld1r_operand, aarch64_sve_ldr_operand): New predicates.
10539         (aarch64_sve_nonimmediate_operand): Likewise.
10540         (aarch64_sve_general_operand): Likewise.
10541         (aarch64_sve_dup_operand, aarch64_sve_arith_immediate): Likewise.
10542         (aarch64_sve_sub_arith_immediate, aarch64_sve_inc_dec_immediate)
10543         (aarch64_sve_logical_immediate, aarch64_sve_mul_immediate): Likewise.
10544         (aarch64_sve_dup_immediate, aarch64_sve_cmp_vsc_immediate): Likewise.
10545         (aarch64_sve_cmp_vsd_immediate, aarch64_sve_index_immediate): Likewise.
10546         (aarch64_sve_float_arith_immediate): Likewise.
10547         (aarch64_sve_float_arith_with_sub_immediate): Likewise.
10548         (aarch64_sve_float_mul_immediate, aarch64_sve_arith_operand): Likewise.
10549         (aarch64_sve_add_operand, aarch64_sve_logical_operand): Likewise.
10550         (aarch64_sve_lshift_operand, aarch64_sve_rshift_operand): Likewise.
10551         (aarch64_sve_mul_operand, aarch64_sve_cmp_vsc_operand): Likewise.
10552         (aarch64_sve_cmp_vsd_operand, aarch64_sve_index_operand): Likewise.
10553         (aarch64_sve_float_arith_operand): Likewise.
10554         (aarch64_sve_float_arith_with_sub_operand): Likewise.
10555         (aarch64_sve_float_mul_operand): Likewise.
10556         (aarch64_sve_vec_perm_operand): Likewise.
10557         (aarch64_pluslong_operand): Include aarch64_sve_addvl_addpl_immediate.
10558         (aarch64_mov_operand): Accept const_poly_int and const_vector.
10559         (aarch64_simd_lshift_imm, aarch64_simd_rshift_imm): Accept const
10560         as well as const_vector.
10561         (aarch64_simd_imm_zero, aarch64_simd_imm_minus_one): Move earlier
10562         in file.  Use CONST0_RTX and CONSTM1_RTX.
10563         (aarch64_simd_or_scalar_imm_zero): Likewise.  Add match_codes.
10564         (aarch64_simd_reg_or_zero): Accept const as well as const_vector.
10565         Use aarch64_simd_imm_zero.
10566         * config/aarch64/aarch64-sve.md: New file.
10567         * config/aarch64/aarch64.md: Include it.
10568         (VG_REGNUM, P0_REGNUM, P7_REGNUM, P15_REGNUM): New register numbers.
10569         (UNSPEC_REV, UNSPEC_LD1_SVE, UNSPEC_ST1_SVE, UNSPEC_MERGE_PTRUE)
10570         (UNSPEC_PTEST_PTRUE, UNSPEC_UNPACKSHI, UNSPEC_UNPACKUHI)
10571         (UNSPEC_UNPACKSLO, UNSPEC_UNPACKULO, UNSPEC_PACK)
10572         (UNSPEC_FLOAT_CONVERT, UNSPEC_WHILE_LO): New unspec constants.
10573         (sve): New attribute.
10574         (enabled): Disable instructions with the sve attribute unless
10575         TARGET_SVE.
10576         (movqi, movhi): Pass CONST_POLY_INT operaneds through
10577         aarch64_expand_mov_immediate.
10578         (*mov<mode>_aarch64, *movsi_aarch64, *movdi_aarch64): Handle
10579         CNT[BHSD] immediates.
10580         (movti): Split CONST_POLY_INT moves into two halves.
10581         (add<mode>3): Accept aarch64_pluslong_or_poly_operand.
10582         Split additions that need a temporary here if the destination
10583         is the stack pointer.
10584         (*add<mode>3_aarch64): Handle ADDVL and ADDPL immediates.
10585         (*add<mode>3_poly_1): New instruction.
10586         (set_clobber_cc): New expander.
10587
10588 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
10589
10590         * simplify-rtx.c (simplify_immed_subreg): Add an inner_bytes
10591         parameter and use it instead of GET_MODE_SIZE (innermode).  Use
10592         inner_bytes * BITS_PER_UNIT instead of GET_MODE_BITSIZE (innermode).
10593         Use CEIL (inner_bytes, GET_MODE_UNIT_SIZE (innermode)) instead of
10594         GET_MODE_NUNITS (innermode).  Also add a first_elem parameter.
10595         Change innermode from fixed_mode_size to machine_mode.
10596         (simplify_subreg): Update call accordingly.  Handle a constant-sized
10597         subreg of a variable-length CONST_VECTOR.
10598
10599 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
10600             Alan Hayward  <alan.hayward@arm.com>
10601             David Sherwood  <david.sherwood@arm.com>
10602
10603         * tree-ssa-address.c (mem_ref_valid_without_offset_p): New function.
10604         (add_offset_to_base): New function, split out from...
10605         (create_mem_ref): ...here.  When handling a scale other than 1,
10606         check first whether the address is valid without the offset.
10607         Add it into the base if so, leaving the index and scale as-is.
10608
10609 2018-01-12  Jakub Jelinek  <jakub@redhat.com>
10610
10611         PR c++/83778
10612         * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Call
10613         fold_for_warn before checking if arg2 is INTEGER_CST.
10614
10615 2018-01-12  Segher Boessenkool  <segher@kernel.crashing.org>
10616
10617         * config/rs6000/predicates.md (load_multiple_operation): Delete.
10618         (store_multiple_operation): Delete.
10619         * config/rs6000/rs6000-cpus.def (601): Remove MASK_STRING.
10620         * config/rs6000/rs6000-protos.h (rs6000_output_load_multiple): Delete.
10621         * config/rs6000/rs6000-string.c (expand_block_move): Delete everything
10622         guarded by TARGET_STRING.
10623         (rs6000_output_load_multiple): Delete.
10624         * config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
10625         OPTION_MASK_STRING / TARGET_STRING handling.
10626         (print_operand) <'N', 'O'>: Add comment that these are unused now.
10627         (const rs6000_opt_masks) <"string">: Change mask to 0.
10628         * config/rs6000/rs6000.h (TARGET_DEFAULT): Remove MASK_STRING.
10629         (MASK_STRING): Delete.
10630         * config/rs6000/rs6000.md (*mov<mode>_string): Delete TARGET_STRING
10631         parts.  Simplify.
10632         (load_multiple): Delete.
10633         (*ldmsi8): Delete.
10634         (*ldmsi7): Delete.
10635         (*ldmsi6): Delete.
10636         (*ldmsi5): Delete.
10637         (*ldmsi4): Delete.
10638         (*ldmsi3): Delete.
10639         (store_multiple): Delete.
10640         (*stmsi8): Delete.
10641         (*stmsi7): Delete.
10642         (*stmsi6): Delete.
10643         (*stmsi5): Delete.
10644         (*stmsi4): Delete.
10645         (*stmsi3): Delete.
10646         (movmemsi_8reg): Delete.
10647         (corresponding unnamed define_insn): Delete.
10648         (movmemsi_6reg): Delete.
10649         (corresponding unnamed define_insn): Delete.
10650         (movmemsi_4reg): Delete.
10651         (corresponding unnamed define_insn): Delete.
10652         (movmemsi_2reg): Delete.
10653         (corresponding unnamed define_insn): Delete.
10654         (movmemsi_1reg): Delete.
10655         (corresponding unnamed define_insn): Delete.
10656         * config/rs6000/rs6000.opt (mno-string): New.
10657         (mstring): Replace by deprecation warning stub.
10658         * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mstring.
10659
10660 2018-01-12  Jakub Jelinek  <jakub@redhat.com>
10661
10662         * regrename.c (regrename_do_replace): If replacing the same
10663         reg multiple times, try to reuse last created gen_raw_REG.
10664
10665         PR debug/81155
10666         * bb-reorder.c (pass_partition_blocks::gate): In lto don't partition
10667         main to workaround a bug in GDB.
10668
10669 2018-01-12  Tom de Vries  <tom@codesourcery.com>
10670
10671         PR target/83737
10672         * config.gcc (nvptx*-*-*): Set use_gcc_stdint=wrap.
10673
10674 2018-01-12  Vladimir Makarov  <vmakarov@redhat.com>
10675
10676         PR rtl-optimization/80481
10677         * ira-color.c (get_cap_member): New function.
10678         (allocnos_conflict_by_live_ranges_p): Use it.
10679         (slot_coalesced_allocno_live_ranges_intersect_p): Add assert.
10680         (setup_slot_coalesced_allocno_live_ranges): Ditto.
10681
10682 2018-01-12  Uros Bizjak  <ubizjak@gmail.com>
10683
10684         PR target/83628
10685         * config/alpha/alpha.md (*saddsi_1): New insn_ans_split pattern.
10686         (*saddl_se_1): Ditto.
10687         (*ssubsi_1): Ditto.
10688         (*ssubl_se_1): Ditto.
10689
10690 2018-01-12  Richard Sandiford  <richard.sandiford@linaro.org>
10691
10692         * tree-predcom.c (aff_combination_dr_offset): Use wi::to_poly_widest
10693         rather than wi::to_widest for DR_INITs.
10694         * tree-vect-data-refs.c (vect_find_same_alignment_drs): Use
10695         wi::to_poly_offset rather than wi::to_offset for DR_INIT.
10696         (vect_analyze_data_ref_accesses): Require both DR_INITs to be
10697         INTEGER_CSTs.
10698         (vect_analyze_group_access_1): Note that here.
10699
10700 2018-01-12  Richard Sandiford  <richard.sandiford@linaro.org>
10701
10702         * tree-vectorizer.c (get_vec_alignment_for_array_type): Handle
10703         polynomial type sizes.
10704
10705 2018-01-12  Richard Sandiford  <richard.sandiford@linaro.org>
10706
10707         * gimplify.c (gimple_add_tmp_var_fn): Allow variables to have a
10708         poly_uint64 size, rather than requiring an unsigned HOST_WIDE_INT size.
10709         (gimple_add_tmp_var): Likewise.
10710
10711 2018-01-12  Martin Liska  <mliska@suse.cz>
10712
10713         * gimple.c (gimple_alloc_counts): Use uint64_t instead of int.
10714         (gimple_alloc_sizes): Likewise.
10715         (dump_gimple_statistics): Use PRIu64 in printf format.
10716         * gimple.h: Change uint64_t to int.
10717
10718 2018-01-12  Martin Liska  <mliska@suse.cz>
10719
10720         * tree-core.h: Use uint64_t instead of int.
10721         * tree.c (tree_node_counts): Likewise.
10722         (tree_node_sizes): Likewise.
10723         (dump_tree_statistics): Use PRIu64 in printf format.
10724
10725 2018-01-12  Martin Liska  <mliska@suse.cz>
10726
10727         * Makefile.in: As qsort_chk is implemented in vec.c, add
10728         vec.o to linkage of gencfn-macros.
10729         * tree.c (build_new_poly_int_cst): Add CXX_MEM_STAT_INFO as it's
10730         passing the info to record_node_allocation_statistics.
10731         (test_vector_cst_patterns): Add CXX_MEM_STAT_INFO to declaration
10732         and pass the info.
10733         * ggc-common.c (struct ggc_usage): Add operator== and use
10734         it in operator< and compare function.
10735         * mem-stats.h (struct mem_usage): Likewise.
10736         * vec.c (struct vec_usage): Remove operator< and compare
10737         function. Can be simply inherited.
10738
10739 2018-01-12  Martin Jambor  <mjambor@suse.cz>
10740
10741         PR target/81616
10742         * params.def: New parameter PARAM_AVOID_FMA_MAX_BITS.
10743         * tree-ssa-math-opts.c: Include domwalk.h.
10744         (convert_mult_to_fma_1): New function.
10745         (fma_transformation_info): New type.
10746         (fma_deferring_state): Likewise.
10747         (cancel_fma_deferring): New function.
10748         (result_of_phi): Likewise.
10749         (last_fma_candidate_feeds_initial_phi): Likewise.
10750         (convert_mult_to_fma): Added deferring logic, split actual
10751         transformation to convert_mult_to_fma_1.
10752         (math_opts_dom_walker): New type.
10753         (math_opts_dom_walker::after_dom_children): New method, body moved
10754         here from pass_optimize_widening_mul::execute, added deferring logic
10755         bits.
10756         (pass_optimize_widening_mul::execute): Moved most of code to
10757         math_opts_dom_walker::after_dom_children.
10758         * config/i386/x86-tune.def (X86_TUNE_AVOID_128FMA_CHAINS): New.
10759         * config/i386/i386.c (ix86_option_override_internal): Added
10760         maybe_setting of PARAM_AVOID_FMA_MAX_BITS.
10761
10762 2018-01-12  Richard Biener  <rguenther@suse.de>
10763
10764         PR debug/83157
10765         * dwarf2out.c (gen_variable_die): Do not reset old_die for
10766         inline instance vars.
10767
10768 2018-01-12  Oleg Endo  <olegendo@gcc.gnu.org>
10769
10770         PR target/81819
10771         * config/rx/rx.c (rx_is_restricted_memory_address):
10772         Handle SUBREG case.
10773
10774 2018-01-12  Richard Biener  <rguenther@suse.de>
10775
10776         PR tree-optimization/80846
10777         * target.def (split_reduction): New target hook.
10778         * targhooks.c (default_split_reduction): New function.
10779         * targhooks.h (default_split_reduction): Declare.
10780         * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
10781         target requests first reduce vectors by combining low and high
10782         parts.
10783         * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
10784         (get_vectype_for_scalar_type_and_size): Export.
10785         * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
10786         * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
10787         * doc/tm.texi: Regenerate.
10788         * config/i386/i386.c (ix86_split_reduction): Implement
10789         TARGET_VECTORIZE_SPLIT_REDUCTION.
10790
10791 2018-01-12  Eric Botcazou  <ebotcazou@adacore.com>
10792
10793         PR target/83368
10794         * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
10795         in PIC mode except for TARGET_VXWORKS_RTP.
10796         * config/sparc/sparc.c: Include cfgrtl.h.
10797         (TARGET_INIT_PIC_REG): Define.
10798         (TARGET_USE_PSEUDO_PIC_REG): Likewise.
10799         (sparc_pic_register_p): New predicate.
10800         (sparc_legitimate_address_p): Use it.
10801         (sparc_legitimize_pic_address): Likewise.
10802         (sparc_delegitimize_address): Likewise.
10803         (sparc_mode_dependent_address_p): Likewise.
10804         (gen_load_pcrel_sym): Remove 4th parameter.
10805         (load_got_register): Adjust call to above.  Remove obsolete stuff.
10806         (sparc_expand_prologue): Do not call load_got_register here.
10807         (sparc_flat_expand_prologue): Likewise.
10808         (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
10809         (sparc_use_pseudo_pic_reg): New function.
10810         (sparc_init_pic_reg): Likewise.
10811         * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
10812         (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
10813
10814 2018-01-12  Christophe Lyon  <christophe.lyon@linaro.org>
10815
10816         * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
10817         Add item for branch_cost.
10818
10819 2018-01-12  Eric Botcazou  <ebotcazou@adacore.com>
10820
10821         PR rtl-optimization/83565
10822         * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
10823         not extend the result to a larger mode for rotate operations.
10824         (num_sign_bit_copies1): Likewise.
10825
10826 2018-01-12  Rainer Orth  <ro@CeBiTec.Uni-Bielefeld.DE>
10827
10828         PR target/40411
10829         * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
10830         -symbolic.
10831         Use values-Xc.o for -pedantic.
10832         Link with values-xpg4.o for C90, values-xpg6.o otherwise.
10833
10834 2018-01-12  Martin Liska  <mliska@suse.cz>
10835
10836         PR ipa/83054
10837         * ipa-devirt.c (final_warning_record::grow_type_warnings):
10838         New function.
10839         (possible_polymorphic_call_targets): Use it.
10840         (ipa_devirt): Likewise.
10841
10842 2018-01-12  Martin Liska  <mliska@suse.cz>
10843
10844         * profile-count.h (enum profile_quality): Use 0 as invalid
10845         enum value of profile_quality.
10846
10847 2018-01-12  Chung-Ju Wu  <jasonwucj@gmail.com>
10848
10849         * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
10850         -mext-string options.
10851
10852 2018-01-12  Richard Biener  <rguenther@suse.de>
10853
10854         * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
10855         DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
10856         * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
10857         Likewise.
10858         * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
10859
10860 2018-01-11  Michael Meissner  <meissner@linux.vnet.ibm.com>
10861
10862         * configure.ac (--with-long-double-format): Add support for the
10863         configuration option to change the default long double format on
10864         PowerPC systems.
10865         * config.gcc (powerpc*-linux*-*): Likewise.
10866         * configure: Regenerate.
10867         * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
10868         double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
10869         used without modification.
10870
10871 2018-01-11  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
10872
10873         * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
10874         (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
10875         * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
10876         MISC_BUILTIN_SPEC_BARRIER.
10877         (rs6000_init_builtins): Likewise.
10878         * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
10879         enum value.
10880         (speculation_barrier): New define_insn.
10881         * doc/extend.texi: Document __builtin_speculation_barrier.
10882
10883 2018-01-11  Jakub Jelinek  <jakub@redhat.com>
10884
10885         PR target/83203
10886         * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
10887         is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
10888         * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
10889         iterators.
10890         (ssescalarmodesuffix): Add 512-bit vectors.  Use "d" or "q" for
10891         integral modes instead of "ss" and "sd".
10892         (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
10893         vectors with 32-bit and 64-bit elements.
10894         (vecdupssescalarmodesuffix): New mode attribute.
10895         (vec_dup<mode>): Use it.
10896
10897 2018-01-11  H.J. Lu  <hongjiu.lu@intel.com>
10898
10899         PR target/83330
10900         * config/i386/i386.c (ix86_compute_frame_layout): Align stack
10901         frame if argument is passed on stack.
10902
10903 2018-01-11  Jakub Jelinek  <jakub@redhat.com>
10904
10905         PR target/82682
10906         * ree.c (combine_reaching_defs): Optimize also
10907         reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
10908         reg2=any_extend(exp); reg1=reg2;, formatting fix.
10909
10910 2018-01-11  Jan Hubicka  <hubicka@ucw.cz>
10911
10912         PR middle-end/83189
10913         * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
10914
10915 2018-01-11  Jan Hubicka  <hubicka@ucw.cz>
10916
10917         PR middle-end/83718
10918         * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
10919         after they are computed.
10920
10921 2018-01-11  Bin Cheng  <bin.cheng@arm.com>
10922
10923         PR tree-optimization/83695
10924         * gimple-loop-linterchange.cc
10925         (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
10926         reset cached scev information after interchange.
10927         (pass_linterchange::execute): Remove call to scev_reset_htab.
10928
10929 2018-01-11  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
10930
10931         * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
10932         vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
10933         vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
10934         vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
10935         vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
10936         vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
10937         * config/arm/arm_neon_builtins.def (vfmal_lane_low,
10938         vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
10939         vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
10940         vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
10941         vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
10942         * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
10943         (V_lane_reg): Likewise.
10944         * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
10945         New define_expand.
10946         (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
10947         (vfmal_lane_low<mode>_intrinsic,
10948         vfmal_lane_low<vfmlsel2><mode>_intrinsic,
10949         vfmal_lane_high<vfmlsel2><mode>_intrinsic,
10950         vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
10951         vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
10952         vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
10953         vfmsl_lane_high<mode>_intrinsic): New define_insns.
10954
10955 2018-01-11  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
10956
10957         * config/arm/arm-cpus.in (fp16fml): New feature.
10958         (ALL_SIMD): Add fp16fml.
10959         (armv8.2-a): Add fp16fml as an option.
10960         (armv8.3-a): Likewise.
10961         (armv8.4-a): Add fp16fml as part of fp16.
10962         * config/arm/arm.h (TARGET_FP16FML): Define.
10963         * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
10964         when appropriate.
10965         * config/arm/arm-modes.def (V2HF): Define.
10966         * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
10967         vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
10968         vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
10969         * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
10970         vfmsl_low, vfmsl_high): New set of builtins.
10971         * config/arm/iterators.md (PLUSMINUS): New code iterator.
10972         (vfml_op): New code attribute.
10973         (VFMLHALVES): New int iterator.
10974         (VFML, VFMLSEL): New mode attributes.
10975         (V_reg): Define mapping for V2HF.
10976         (V_hi, V_lo): New mode attributes.
10977         (VF_constraint): Likewise.
10978         (vfml_half, vfml_half_selector): New int attributes.
10979         * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
10980         define_expand.
10981         (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
10982         vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
10983         New define_insn.
10984         * config/arm/t-arm-elf (v8_fps): Add fp16fml.
10985         * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
10986         * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
10987         * doc/invoke.texi (ARM Options): Document fp16fml.  Update armv8.4-a
10988         documentation.
10989         * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
10990         Document new effective target and option set.
10991
10992 2018-01-11  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
10993
10994         * config/arm/arm-cpus.in (armv8_4): New feature.
10995         (ARMv8_4a): New fgroup.
10996         (armv8.4-a): New arch.
10997         * config/arm/arm-tables.opt: Regenerate.
10998         * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
10999         * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
11000         * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
11001         Add matching rules for -march=armv8.4-a and extensions.
11002         * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
11003
11004 2018-01-11  Oleg Endo  <olegendo@gcc.gnu.org>
11005
11006         PR target/81821
11007         * config/rx/rx.md (BW): New mode attribute.
11008         (sync_lock_test_and_setsi): Add mode suffix to insn output.
11009
11010 2018-01-11  Richard Biener  <rguenther@suse.de>
11011
11012         PR tree-optimization/83435
11013         * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
11014         * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
11015         * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
11016
11017 2018-01-11  Richard Sandiford  <richard.sandiford@linaro.org>
11018             Alan Hayward  <alan.hayward@arm.com>
11019             David Sherwood  <david.sherwood@arm.com>
11020
11021         * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
11022         field.
11023         (aarch64_classify_address): Initialize it.  Track polynomial offsets.
11024         (aarch64_print_address_internal): Use it to check for a zero offset.
11025
11026 2018-01-11  Richard Sandiford  <richard.sandiford@linaro.org>
11027             Alan Hayward  <alan.hayward@arm.com>
11028             David Sherwood  <david.sherwood@arm.com>
11029
11030         * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
11031         * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
11032         Return a poly_int64 rather than a HOST_WIDE_INT.
11033         (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
11034         rather than a HOST_WIDE_INT.
11035         * config/aarch64/aarch64.h (aarch64_frame): Protect with
11036         HAVE_POLY_INT_H rather than HOST_WIDE_INT.  Change locals_offset,
11037         hard_fp_offset, frame_size, initial_adjust, callee_offset and
11038         final_offset from HOST_WIDE_INT to poly_int64.
11039         * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
11040         to_constant when getting the number of units in an Advanced SIMD
11041         mode.
11042         (aarch64_builtin_vectorized_function): Check for a constant number
11043         of units.
11044         * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
11045         GET_MODE_SIZE.
11046         (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
11047         attribute instead of GET_MODE_NUNITS.
11048         * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
11049         (aarch64_class_max_nregs): Use the constant_lowest_bound of the
11050         GET_MODE_SIZE for fixed-size registers.
11051         (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
11052         (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
11053         (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
11054         (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
11055         (aarch64_print_operand, aarch64_print_address_internal)
11056         (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
11057         (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
11058         (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
11059         Handle polynomial GET_MODE_SIZE.
11060         (aarch64_hard_regno_caller_save_mode): Likewise.  Return modes
11061         wider than SImode without modification.
11062         (tls_symbolic_operand_type): Use strip_offset instead of split_const.
11063         (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
11064         (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
11065         passing and returning SVE modes.
11066         (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
11067         rather than GEN_INT.
11068         (aarch64_emit_probe_stack_range): Take the size as a poly_int64
11069         rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
11070         (aarch64_allocate_and_probe_stack_space): Likewise.
11071         (aarch64_layout_frame): Cope with polynomial offsets.
11072         (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
11073         start_offset as a poly_int64 rather than a HOST_WIDE_INT.  Track
11074         polynomial offsets.
11075         (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
11076         (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
11077         poly_int64 rather than a HOST_WIDE_INT.
11078         (aarch64_get_separate_components, aarch64_process_components)
11079         (aarch64_expand_prologue, aarch64_expand_epilogue)
11080         (aarch64_use_return_insn_p): Handle polynomial frame offsets.
11081         (aarch64_anchor_offset): New function, split out from...
11082         (aarch64_legitimize_address): ...here.
11083         (aarch64_builtin_vectorization_cost): Handle polynomial
11084         TYPE_VECTOR_SUBPARTS.
11085         (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
11086         GET_MODE_NUNITS.
11087         (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
11088         number of elements from the PARALLEL rather than the mode.
11089         (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
11090         rather than GET_MODE_BITSIZE.
11091         (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
11092         (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
11093         (aarch64_expand_vec_perm_const_1): Handle polynomial
11094         d->perm.length () and d->perm elements.
11095         (aarch64_evpc_tbl): Likewise.  Use nelt rather than GET_MODE_NUNITS.
11096         Apply to_constant to d->perm elements.
11097         (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
11098         polynomial CONST_VECTOR_NUNITS.
11099         (aarch64_move_pointer): Take amount as a poly_int64 rather
11100         than an int.
11101         (aarch64_progress_pointer): Avoid temporary variable.
11102         * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
11103         the mode attribute instead of GET_MODE.
11104
11105 2018-01-11  Richard Sandiford  <richard.sandiford@linaro.org>
11106             Alan Hayward  <alan.hayward@arm.com>
11107             David Sherwood  <david.sherwood@arm.com>
11108
11109         * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
11110         x exists before using it.
11111         (aarch64_add_constant_internal): Rename to...
11112         (aarch64_add_offset_1): ...this.  Replace regnum with separate
11113         src and dest rtxes.  Handle the case in which they're different,
11114         including when the offset is zero.  Replace scratchreg with an rtx.
11115         Use 2 additions if there is no spare register into which we can
11116         move a 16-bit constant.
11117         (aarch64_add_constant): Delete.
11118         (aarch64_add_offset): Replace reg with separate src and dest
11119         rtxes.  Take a poly_int64 offset instead of a HOST_WIDE_INT.
11120         Use aarch64_add_offset_1.
11121         (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
11122         an rtx rather than an int.  Take the delta as a poly_int64
11123         rather than a HOST_WIDE_INT.  Use aarch64_add_offset.
11124         (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
11125         (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
11126         aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
11127         (aarch64_expand_epilogue): Update calls to aarch64_add_offset
11128         and aarch64_add_sp.
11129         (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
11130         aarch64_add_constant.
11131
11132 2018-01-11  Richard Sandiford  <richard.sandiford@linaro.org>
11133
11134         * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
11135         Use scalar_float_mode.
11136
11137 2018-01-11  Richard Sandiford  <richard.sandiford@linaro.org>
11138
11139         * config/aarch64/aarch64-simd.md
11140         (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
11141         (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
11142         (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
11143         (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
11144         (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
11145         (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
11146         (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
11147         (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
11148         (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
11149         (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
11150
11151 2018-01-11  Prathamesh Kulkarni  <prathamesh.kulkarni@linaro.org>
11152
11153         PR target/83514
11154         * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
11155         targ_options->x_arm_arch_string is non NULL.
11156
11157 2018-01-11  Tamar Christina  <tamar.christina@arm.com>
11158
11159         * config/aarch64/aarch64.h
11160         (AARCH64_FL_FOR_ARCH8_4): Add  AARCH64_FL_DOTPROD.
11161
11162 2018-01-11  Sudakshina Das  <sudi.das@arm.com>
11163
11164         PR target/82096
11165         * expmed.c (emit_store_flag_force): Swap if const op0
11166         and change VOIDmode to mode of op0.
11167
11168 2018-01-11  Richard Sandiford  <richard.sandiford@linaro.org>
11169
11170         PR rtl-optimization/83761
11171         * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
11172         than bytes to mode_for_size.
11173
11174 2018-01-10  Jan Hubicka  <hubicka@ucw.cz>
11175
11176         PR middle-end/83189
11177         * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
11178         * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
11179         profile.
11180
11181 2018-01-10  Jan Hubicka  <hubicka@ucw.cz>
11182
11183         PR middle-end/83575
11184         * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
11185         when in layout mode.
11186         (cfg_layout_finalize): Do not verify cfg before we are out of layout.
11187         * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
11188         partition fixup.
11189
11190 2018-01-10  Michael Collison  <michael.collison@arm.com>
11191
11192         * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
11193         * config/aarch64/aarch64-option-extension.def: Add
11194         AARCH64_OPT_EXTENSION of 'fp16fml'.
11195         * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
11196         (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
11197         * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
11198         * config/aarch64/constraints.md (Ui7): New constraint.
11199         * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
11200         (VFMLA_SEL_W): Ditto.
11201         (f16quad): Ditto.
11202         (f16mac1): Ditto.
11203         (VFMLA16_LOW): New int iterator.
11204         (VFMLA16_HIGH): Ditto.
11205         (UNSPEC_FMLAL): New unspec.
11206         (UNSPEC_FMLSL): Ditto.
11207         (UNSPEC_FMLAL2): Ditto.
11208         (UNSPEC_FMLSL2): Ditto.
11209         (f16mac): New code attribute.
11210         * config/aarch64/aarch64-simd-builtins.def
11211         (aarch64_fmlal_lowv2sf): Ditto.
11212         (aarch64_fmlsl_lowv2sf): Ditto.
11213         (aarch64_fmlalq_lowv4sf): Ditto.
11214         (aarch64_fmlslq_lowv4sf): Ditto.
11215         (aarch64_fmlal_highv2sf): Ditto.
11216         (aarch64_fmlsl_highv2sf): Ditto.
11217         (aarch64_fmlalq_highv4sf): Ditto.
11218         (aarch64_fmlslq_highv4sf): Ditto.
11219         (aarch64_fmlal_lane_lowv2sf): Ditto.
11220         (aarch64_fmlsl_lane_lowv2sf): Ditto.
11221         (aarch64_fmlal_laneq_lowv2sf): Ditto.
11222         (aarch64_fmlsl_laneq_lowv2sf): Ditto.
11223         (aarch64_fmlalq_lane_lowv4sf): Ditto.
11224         (aarch64_fmlsl_lane_lowv4sf): Ditto.
11225         (aarch64_fmlalq_laneq_lowv4sf): Ditto.
11226         (aarch64_fmlsl_laneq_lowv4sf): Ditto.
11227         (aarch64_fmlal_lane_highv2sf): Ditto.
11228         (aarch64_fmlsl_lane_highv2sf): Ditto.
11229         (aarch64_fmlal_laneq_highv2sf): Ditto.
11230         (aarch64_fmlsl_laneq_highv2sf): Ditto.
11231         (aarch64_fmlalq_lane_highv4sf): Ditto.
11232         (aarch64_fmlsl_lane_highv4sf): Ditto.
11233         (aarch64_fmlalq_laneq_highv4sf): Ditto.
11234         (aarch64_fmlsl_laneq_highv4sf): Ditto.
11235         * config/aarch64/aarch64-simd.md:
11236         (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
11237         (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
11238         (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
11239         (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
11240         (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
11241         (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
11242         (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
11243         (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
11244         (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
11245         (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
11246         (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
11247         (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
11248         (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
11249         (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
11250         (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
11251         (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
11252         (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
11253         (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
11254         (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
11255         (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
11256         * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
11257         (vfmlsl_low_u32): Ditto.
11258         (vfmlalq_low_u32): Ditto.
11259         (vfmlslq_low_u32): Ditto.
11260         (vfmlal_high_u32): Ditto.
11261         (vfmlsl_high_u32): Ditto.
11262         (vfmlalq_high_u32): Ditto.
11263         (vfmlslq_high_u32): Ditto.
11264         (vfmlal_lane_low_u32): Ditto.
11265         (vfmlsl_lane_low_u32): Ditto.
11266         (vfmlal_laneq_low_u32): Ditto.
11267         (vfmlsl_laneq_low_u32): Ditto.
11268         (vfmlalq_lane_low_u32): Ditto.
11269         (vfmlslq_lane_low_u32): Ditto.
11270         (vfmlalq_laneq_low_u32): Ditto.
11271         (vfmlslq_laneq_low_u32): Ditto.
11272         (vfmlal_lane_high_u32): Ditto.
11273         (vfmlsl_lane_high_u32): Ditto.
11274         (vfmlal_laneq_high_u32): Ditto.
11275         (vfmlsl_laneq_high_u32): Ditto.
11276         (vfmlalq_lane_high_u32): Ditto.
11277         (vfmlslq_lane_high_u32): Ditto.
11278         (vfmlalq_laneq_high_u32): Ditto.
11279         (vfmlslq_laneq_high_u32): Ditto.
11280         * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
11281         (AARCH64_FL_FOR_ARCH8_4): New.
11282         (AARCH64_ISA_F16FML): New ISA flag.
11283         (TARGET_F16FML): New feature flag for fp16fml.
11284         (doc/invoke.texi): Document new fp16fml option.
11285
11286 2018-01-10  Michael Collison  <michael.collison@arm.com>
11287
11288         * config/aarch64/aarch64-builtins.c:
11289         (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
11290         * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
11291         (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
11292         * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
11293         (AARCH64_ISA_SHA3): New ISA flag.
11294         (TARGET_SHA3): New feature flag for sha3.
11295         * config/aarch64/iterators.md (sha512_op): New int attribute.
11296         (CRYPTO_SHA512): New int iterator.
11297         (UNSPEC_SHA512H): New unspec.
11298         (UNSPEC_SHA512H2): Ditto.
11299         (UNSPEC_SHA512SU0): Ditto.
11300         (UNSPEC_SHA512SU1): Ditto.
11301         * config/aarch64/aarch64-simd-builtins.def
11302         (aarch64_crypto_sha512hqv2di): New builtin.
11303         (aarch64_crypto_sha512h2qv2di): Ditto.
11304         (aarch64_crypto_sha512su0qv2di): Ditto.
11305         (aarch64_crypto_sha512su1qv2di): Ditto.
11306         (aarch64_eor3qv8hi): Ditto.
11307         (aarch64_rax1qv2di): Ditto.
11308         (aarch64_xarqv2di): Ditto.
11309         (aarch64_bcaxqv8hi): Ditto.
11310         * config/aarch64/aarch64-simd.md:
11311         (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
11312         (aarch64_crypto_sha512su0qv2di): Ditto.
11313         (aarch64_crypto_sha512su1qv2di): Ditto.
11314         (aarch64_eor3qv8hi): Ditto.
11315         (aarch64_rax1qv2di): Ditto.
11316         (aarch64_xarqv2di): Ditto.
11317         (aarch64_bcaxqv8hi): Ditto.
11318         * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
11319         (vsha512h2q_u64): Ditto.
11320         (vsha512su0q_u64): Ditto.
11321         (vsha512su1q_u64): Ditto.
11322         (veor3q_u16): Ditto.
11323         (vrax1q_u64): Ditto.
11324         (vxarq_u64): Ditto.
11325         (vbcaxq_u16): Ditto.
11326         * config/arm/types.md (crypto_sha512): New type attribute.
11327         (crypto_sha3): Ditto.
11328         (doc/invoke.texi): Document new sha3 option.
11329
11330 2018-01-10  Michael Collison  <michael.collison@arm.com>
11331
11332         * config/aarch64/aarch64-builtins.c:
11333         (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
11334         * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
11335         (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
11336         (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
11337         * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
11338         (AARCH64_ISA_SM4): New ISA flag.
11339         (TARGET_SM4): New feature flag for sm4.
11340         * config/aarch64/aarch64-simd-builtins.def
11341         (aarch64_sm3ss1qv4si): Ditto.
11342         (aarch64_sm3tt1aq4si): Ditto.
11343         (aarch64_sm3tt1bq4si): Ditto.
11344         (aarch64_sm3tt2aq4si): Ditto.
11345         (aarch64_sm3tt2bq4si): Ditto.
11346         (aarch64_sm3partw1qv4si): Ditto.
11347         (aarch64_sm3partw2qv4si): Ditto.
11348         (aarch64_sm4eqv4si): Ditto.
11349         (aarch64_sm4ekeyqv4si): Ditto.
11350         * config/aarch64/aarch64-simd.md:
11351         (aarch64_sm3ss1qv4si): Ditto.
11352         (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
11353         (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
11354         (aarch64_sm4eqv4si): Ditto.
11355         (aarch64_sm4ekeyqv4si): Ditto.
11356         * config/aarch64/iterators.md (sm3tt_op): New int iterator.
11357         (sm3part_op): Ditto.
11358         (CRYPTO_SM3TT): Ditto.
11359         (CRYPTO_SM3PART): Ditto.
11360         (UNSPEC_SM3SS1): New unspec.
11361         (UNSPEC_SM3TT1A): Ditto.
11362         (UNSPEC_SM3TT1B): Ditto.
11363         (UNSPEC_SM3TT2A): Ditto.
11364         (UNSPEC_SM3TT2B): Ditto.
11365         (UNSPEC_SM3PARTW1): Ditto.
11366         (UNSPEC_SM3PARTW2): Ditto.
11367         (UNSPEC_SM4E): Ditto.
11368         (UNSPEC_SM4EKEY): Ditto.
11369         * config/aarch64/constraints.md (Ui2): New constraint.
11370         * config/aarch64/predicates.md (aarch64_imm2): New predicate.
11371         * config/arm/types.md (crypto_sm3): New type attribute.
11372         (crypto_sm4): Ditto.
11373         * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
11374         (vsm3tt1aq_u32): Ditto.
11375         (vsm3tt1bq_u32): Ditto.
11376         (vsm3tt2aq_u32): Ditto.
11377         (vsm3tt2bq_u32): Ditto.
11378         (vsm3partw1q_u32): Ditto.
11379         (vsm3partw2q_u32): Ditto.
11380         (vsm4eq_u32): Ditto.
11381         (vsm4ekeyq_u32): Ditto.
11382         (doc/invoke.texi): Document new sm4 option.
11383
11384 2018-01-10  Michael Collison  <michael.collison@arm.com>
11385
11386         * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
11387         * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
11388         (AARCH64_FL_FOR_ARCH8_4): New.
11389         (AARCH64_FL_V8_4): New flag.
11390         (doc/invoke.texi): Document new armv8.4-a option.
11391
11392 2018-01-10  Michael Collison  <michael.collison@arm.com>
11393
11394         * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
11395         (__ARM_FEATURE_AES): Define if TARGET_AES is true.
11396         (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
11397         * config/aarch64/aarch64-option-extension.def: Add
11398         AARCH64_OPT_EXTENSION of 'sha2'.
11399         (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
11400         (crypto): Disable sha2 and aes if crypto disabled.
11401         (crypto): Enable aes and sha2 if enabled.
11402         (simd): Disable sha2 and aes if simd disabled.
11403         * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
11404         New flags.
11405         (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
11406         (TARGET_SHA2): New feature flag for sha2.
11407         (TARGET_AES): New feature flag for aes.
11408         * config/aarch64/aarch64-simd.md:
11409         (aarch64_crypto_aes<aes_op>v16qi): Make pattern
11410         conditional on TARGET_AES.
11411         (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
11412         (aarch64_crypto_sha1hsi): Make pattern conditional
11413         on TARGET_SHA2.
11414         (aarch64_crypto_sha1hv4si): Ditto.
11415         (aarch64_be_crypto_sha1hv4si): Ditto.
11416         (aarch64_crypto_sha1su1v4si): Ditto.
11417         (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
11418         (aarch64_crypto_sha1su0v4si): Ditto.
11419         (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
11420         (aarch64_crypto_sha256su0v4si): Ditto.
11421         (aarch64_crypto_sha256su1v4si): Ditto.
11422         (doc/invoke.texi): Document new aes and sha2 options.
11423
11424 2018-01-10  Martin Sebor  <msebor@redhat.com>
11425
11426         PR tree-optimization/83781
11427         * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
11428         as string arrays.
11429
11430 2018-01-11  Martin Sebor  <msebor@gmail.com>
11431             Prathamesh Kulkarni  <prathamesh.kulkarni@linaro.org>
11432
11433         PR tree-optimization/83501
11434         PR tree-optimization/81703
11435
11436         * tree-ssa-strlen.c (get_string_cst): Rename...
11437         (get_string_len): ...to this.  Handle global constants.
11438         (handle_char_store): Adjust.
11439
11440 2018-01-10  Kito Cheng  <kito.cheng@gmail.com>
11441             Jim Wilson  <jimw@sifive.com>
11442
11443         * config/riscv/riscv-protos.h (riscv_output_return): New.
11444         * config/riscv/riscv.c (struct machine_function): New naked_p field.
11445         (riscv_attribute_table, riscv_output_return),
11446         (riscv_handle_fndecl_attribute, riscv_naked_function_p),
11447         (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
11448         (riscv_compute_frame_info): Only compute frame->mask if not a naked
11449         function.
11450         (riscv_expand_prologue): Add early return for naked function.
11451         (riscv_expand_epilogue): Likewise.
11452         (riscv_function_ok_for_sibcall): Return false for naked function.
11453         (riscv_set_current_function): New.
11454         (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
11455         (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
11456         * config/riscv/riscv.md (simple_return): Call riscv_output_return.
11457         * doc/extend.texi (RISC-V Function Attributes): New.
11458
11459 2018-01-10  Michael Meissner  <meissner@linux.vnet.ibm.com>
11460
11461         * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
11462         check for 128-bit long double before checking TCmode.
11463         * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
11464         128-bit long doubles before checking TFmode or TCmode.
11465         (FLOAT128_IBM_P): Likewise.
11466
11467 2018-01-10  Martin Sebor  <msebor@redhat.com>
11468
11469         PR tree-optimization/83671
11470         * builtins.c (c_strlen): Unconditionally return zero for the empty
11471         string.
11472         Use -Warray-bounds for warnings.
11473         * gimple-fold.c (get_range_strlen): Handle non-constant lengths
11474         for non-constant array indices with COMPONENT_REF, arrays of
11475         arrays, and pointers to arrays.
11476         (gimple_fold_builtin_strlen): Determine and set length range for
11477         non-constant character arrays.
11478
11479 2018-01-10  Aldy Hernandez  <aldyh@redhat.com>
11480
11481         PR middle-end/81897
11482         * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
11483         empty blocks.
11484
11485 2018-01-10  Eric Botcazou  <ebotcazou@adacore.com>
11486
11487         * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
11488
11489 2018-01-10  Peter Bergner  <bergner@vnet.ibm.com>
11490
11491         PR target/83399
11492         * config/rs6000/rs6000.c (print_operand) <'y'>: Use
11493         VECTOR_MEM_ALTIVEC_OR_VSX_P.
11494         * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
11495         indexed_or_indirect_operand predicate.
11496         (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
11497         (*vsx_le_perm_load_v8hi): Likewise.
11498         (*vsx_le_perm_load_v16qi): Likewise.
11499         (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
11500         (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
11501         (*vsx_le_perm_store_v8hi): Likewise.
11502         (*vsx_le_perm_store_v16qi): Likewise.
11503         (eight unnamed splitters): Likewise.
11504
11505 2018-01-10  Peter Bergner  <bergner@vnet.ibm.com>
11506
11507         * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
11508         * config/rs6000/emmintrin.h: Likewise.
11509         * config/rs6000/mmintrin.h: Likewise.
11510         * config/rs6000/xmmintrin.h: Likewise.
11511
11512 2018-01-10  David Malcolm  <dmalcolm@redhat.com>
11513
11514         PR c++/43486
11515         * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
11516         "public_flag".
11517         * tree.c (tree_nop_conversion): Return true for location wrapper
11518         nodes.
11519         (maybe_wrap_with_location): New function.
11520         (selftest::check_strip_nops): New function.
11521         (selftest::test_location_wrappers): New function.
11522         (selftest::tree_c_tests): Call it.
11523         * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
11524         (maybe_wrap_with_location): New decl.
11525         (EXPR_LOCATION_WRAPPER_P): New macro.
11526         (location_wrapper_p): New inline function.
11527         (tree_strip_any_location_wrapper): New inline function.
11528
11529 2018-01-10  H.J. Lu  <hongjiu.lu@intel.com>
11530
11531         PR target/83735
11532         * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
11533         stack_realign_offset for the largest alignment of stack slot
11534         actually used.
11535         (ix86_find_max_used_stack_alignment): New function.
11536         (ix86_finalize_stack_frame_flags): Use it.  Set
11537         max_used_stack_alignment if we don't realign stack.
11538         * config/i386/i386.h (machine_function): Add
11539         max_used_stack_alignment.
11540
11541 2018-01-10  Christophe Lyon  <christophe.lyon@linaro.org>
11542
11543         * config/arm/arm.opt (-mbranch-cost): New option.
11544         * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
11545         account.
11546
11547 2018-01-10  Segher Boessenkool  <segher@kernel.crashing.org>
11548
11549         PR target/83629
11550         * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
11551         load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
11552
11553 2018-01-10  Richard Biener  <rguenther@suse.de>
11554
11555         PR debug/83765
11556         * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
11557         early out so it also covers the case where we have a non-NULL
11558         origin.
11559
11560 2018-01-10  Richard Sandiford  <richard.sandiford@linaro.org>
11561
11562         PR tree-optimization/83753
11563         * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
11564         for non-strided grouped accesses if the number of elements is 1.
11565
11566 2018-01-10  Jan Hubicka  <hubicka@ucw.cz>
11567
11568         PR target/81616
11569         * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
11570         * i386.h (TARGET_USE_GATHER): Define.
11571         * x86-tune.def (X86_TUNE_USE_GATHER): New.
11572
11573 2018-01-10  Martin Liska  <mliska@suse.cz>
11574
11575         PR bootstrap/82831
11576         * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
11577         * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
11578         partitioning.
11579         * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
11580         CLEANUP_NO_PARTITIONING is not set.
11581
11582 2018-01-10  Richard Sandiford  <richard.sandiford@linaro.org>
11583
11584         * doc/rtl.texi: Remove documentation of (const ...) wrappers
11585         for vectors, as a partial revert of r254296.
11586         * rtl.h (const_vec_p): Delete.
11587         (const_vec_duplicate_p): Don't test for vector CONSTs.
11588         (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
11589         * expmed.c (make_tree): Likewise.
11590
11591         Revert:
11592         * common.md (E, F): Use CONSTANT_P instead of checking for
11593         CONST_VECTOR.
11594         * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
11595         checking for CONST_VECTOR.
11596
11597 2018-01-09  Jan Hubicka  <hubicka@ucw.cz>
11598
11599         PR middle-end/83575
11600         * predict.c (force_edge_cold): Handle in more sane way edges
11601         with no prediction.
11602
11603 2018-01-09  Carl Love  <cel@us.ibm.com>
11604
11605         * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
11606         V4SI, V4SF types.
11607         (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
11608         * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
11609         VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
11610         VMRGOW_V2DI, VMRGOW_V2DF.  Remove definition for VMRGOW.
11611         * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
11612         P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW):  Add definitions.
11613         * config/rs6000/rs6000-protos.h: Add extern defition for
11614         rs6000_generate_float2_double_code.
11615         * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
11616         function.
11617         * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
11618         (float2_v2df): Add define_expand.
11619
11620 2018-01-09  Uros Bizjak  <ubizjak@gmail.com>
11621
11622         PR target/83628
11623         * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
11624         op_mode in the force_to_mode call.
11625
11626 2018-01-09  Richard Sandiford  <richard.sandiford@linaro.org>
11627
11628         * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
11629         instead of checking each element individually.
11630         (aarch64_evpc_uzp): Likewise.
11631         (aarch64_evpc_zip): Likewise.
11632         (aarch64_evpc_ext): Likewise.
11633         (aarch64_evpc_rev): Likewise.
11634         (aarch64_evpc_dup): Test the encoding for a single duplicated element,
11635         instead of checking each element individually.  Return true without
11636         generating rtl if
11637         (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
11638         whether all selected elements come from the same input, instead of
11639         checking each element individually.  Remove calls to gen_rtx_REG,
11640         start_sequence and end_sequence and instead assert that no rtl is
11641         generated.
11642
11643 2018-01-09  Richard Sandiford  <richard.sandiford@linaro.org>
11644
11645         * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
11646         order of HIGH and CONST checks.
11647
11648 2018-01-09  Richard Sandiford  <richard.sandiford@linaro.org>
11649
11650         * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
11651         if the destination isn't an SSA_NAME.
11652
11653 2018-01-09  Richard Biener  <rguenther@suse.de>
11654
11655         PR tree-optimization/83668
11656         * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
11657         move prologue...
11658         (canonicalize_loop_form): ... here, renamed from ...
11659         (canonicalize_loop_closed_ssa_form): ... this and amended to
11660         swap successor edges for loop exit blocks to make us use
11661         the RPO order we need for initial schedule generation.
11662
11663 2018-01-09  Joseph Myers  <joseph@codesourcery.com>
11664
11665         PR tree-optimization/64811
11666         * match.pd: When optimizing comparisons with Inf, avoid
11667         introducing or losing exceptions from comparisons with NaN.
11668
11669 2018-01-09  Martin Liska  <mliska@suse.cz>
11670
11671         PR sanitizer/82517
11672         * asan.c (shadow_mem_size): Add gcc_assert.
11673
11674 2018-01-09  Georg-Johann Lay  <avr@gjlay.de>
11675
11676         Don't save registers in main().
11677
11678         PR target/83738
11679         * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
11680         * config/avr/avr.opt (-mmain-is-OS_task): New target option.
11681         * config/avr/avr.c (avr_set_current_function): Don't error if
11682         naked, OS_task or OS_main are specified at the same time.
11683         (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
11684         OS_main.
11685         (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
11686         attribute.
11687         * common/config/avr/avr-common.c (avr_option_optimization_table):
11688         Switch on -mmain-is-OS_task for optimizing compilations.
11689
11690 2018-01-09  Richard Biener  <rguenther@suse.de>
11691
11692         PR tree-optimization/83572
11693         * graphite.c: Include cfganal.h.
11694         (graphite_transform_loops): Connect infinite loops to exit
11695         and remove fake edges at the end.
11696
11697 2018-01-09  Jan Hubicka  <hubicka@ucw.cz>
11698
11699         * ipa-inline.c (edge_badness): Revert accidental checkin.
11700
11701 2018-01-09  Jan Hubicka  <hubicka@ucw.cz>
11702
11703         PR ipa/80763
11704         * ipa-comdats.c (set_comdat_group): Only set comdat group of real
11705         symbols; not inline clones.
11706
11707 2018-01-09  Jakub Jelinek  <jakub@redhat.com>
11708
11709         PR target/83507
11710         * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
11711         hard registers.  Formatting fixes.
11712
11713         PR preprocessor/83722
11714         * gcc.c (try_generate_repro): Pass
11715         &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
11716         &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
11717         do_report_bug.
11718
11719 2018-01-08  Monk Chiang  <sh.chiang04@gmail.com>
11720             Kito Cheng  <kito.cheng@gmail.com>
11721
11722         * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
11723         (riscv_leaf_function_p): Delete.
11724         (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
11725
11726 2018-01-08  Aaron Sawdey  <acsawdey@linux.vnet.ibm.com>
11727
11728         * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
11729         function.
11730         (do_ifelse): New function.
11731         (do_isel): New function.
11732         (do_sub3): New function.
11733         (do_add3): New function.
11734         (do_load_mask_compare): New function.
11735         (do_overlap_load_compare): New function.
11736         (expand_compare_loop): New function.
11737         (expand_block_compare): Call expand_compare_loop() when appropriate.
11738         * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
11739         option description.
11740         (-mblock-compare-inline-loop-limit): New option.
11741
11742 2018-01-08  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
11743
11744         PR target/83677
11745         * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
11746         Reverse order of second and third operands in first alternative.
11747         * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
11748         of first and second elements in UNSPEC_VPERMR vector.
11749         (altivec_expand_vec_perm_le): Likewise.
11750
11751 2018-01-08  Jeff Law  <law@redhat.com>
11752
11753         PR rtl-optimizatin/81308
11754         * tree-switch-conversion.c (cfg_altered): New file scoped static.
11755         (process_switch): If group_case_labels makes a change, then set
11756         cfg_altered.
11757         (pass_convert_switch::execute): If a switch is converted, then
11758         set cfg_altered.  Return TODO_cfg_cleanup if cfg_altered is true.
11759
11760         PR rtl-optimization/81308
11761         * recog.c (split_all_insns): Conditionally cleanup the CFG after
11762         splitting insns.
11763
11764 2018-01-08  Vidya Praveen  <vidyapraveen@arm.com>
11765
11766         PR target/83663 - Revert r255946
11767         * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
11768         generation for cases where splatting a value is not useful.
11769         * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
11770         across a vec_duplicate and a paradoxical subreg forming a vector
11771         mode to a vec_concat.
11772
11773 2018-01-08  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
11774
11775         * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
11776         -march=armv8.3-a variants.
11777         * config/arm/t-multilib: Likewise.
11778         * config/arm/t-arm-elf: Likewise.  Handle dotprod extension.
11779
11780 2018-01-08  Aaron Sawdey  <acsawdey@linux.vnet.ibm.com>
11781
11782         * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
11783         to generate rtl.
11784         (cceq_ior_compare_complement): Give it a name so I can use it, and
11785         change boolean_or_operator predicate to boolean_operator so it can
11786         be used to generate a crand.
11787         (eqne): New code iterator.
11788         (bd/bd_neg): New code_attrs.
11789         (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
11790         a single define_insn.
11791         (<bd>tf_<mode>): A new insn pattern for the conditional form branch
11792         decrement (bdnzt/bdnzf/bdzt/bdzf).
11793         * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
11794         with the new names of the branch decrement patterns, and added the
11795         names of the branch decrement conditional patterns.
11796
11797 2018-01-08  Richard Biener  <rguenther@suse.de>
11798
11799         PR tree-optimization/83563
11800         * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
11801         cache.
11802
11803 2018-01-08  Richard Biener  <rguenther@suse.de>
11804
11805         PR middle-end/83713
11806         * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
11807
11808 2018-01-08  Richard Biener  <rguenther@suse.de>
11809
11810         PR tree-optimization/83685
11811         * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
11812         references to abnormals.
11813
11814 2018-01-08  Richard Biener  <rguenther@suse.de>
11815
11816         PR lto/83719
11817         * dwarf2out.c (output_indirect_strings): Handle empty
11818         skeleton_debug_str_hash.
11819         (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
11820
11821 2018-01-08  Claudiu Zissulescu  <claziss@synopsys.com>
11822
11823         * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
11824         (emit_store_direct): Likewise.
11825         (arc_trampoline_adjust_address): Likewise.
11826         (arc_asm_trampoline_template): New function.
11827         (arc_initialize_trampoline): Use asm_trampoline_template.
11828         (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
11829         * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
11830         * config/arc/arc.md (flush_icache): Delete pattern.
11831
11832 2018-01-08  Claudiu Zissulescu  <claziss@synopsys.com>
11833
11834         * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
11835         * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
11836         munaligned-access.
11837
11838 2018-01-08  Sebastian Huber  <sebastian.huber@embedded-brains.de>
11839
11840         PR target/83681
11841         * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
11842         by not USED_FOR_TARGET.
11843         (make_pass_resolve_sw_modes): Likewise.
11844
11845 2018-01-08  Sebastian Huber  <sebastian.huber@embedded-brains.de>
11846
11847         * config/nios2/nios2.h (nios2_section_threshold): Guard by not
11848         USED_FOR_TARGET.
11849
11850 2018-01-08  Richard Biener  <rguenther@suse.de>
11851
11852         PR middle-end/83580
11853         * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
11854
11855 2018-01-08  Richard Biener  <rguenther@suse.de>
11856
11857         PR middle-end/83517
11858         * match.pd ((t * 2) / 2) -> t): Add missing :c.
11859
11860 2018-01-06  Aldy Hernandez  <aldyh@redhat.com>
11861
11862         PR middle-end/81897
11863         * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
11864         basic blocks with a small number of successors.
11865         (convert_control_dep_chain_into_preds): Improve handling of
11866         forwarder blocks.
11867         (dump_predicates): Split apart into...
11868         (dump_pred_chain): ...here...
11869         (dump_pred_info): ...and here.
11870         (can_one_predicate_be_invalidated_p): Add debugging printfs.
11871         (can_chain_union_be_invalidated_p): Improve check for invalidation
11872         of paths.
11873         (uninit_uses_cannot_happen): Avoid unnecessary if
11874         convert_control_dep_chain_into_preds yielded nothing.
11875
11876 2018-01-06  Martin Sebor  <msebor@redhat.com>
11877
11878         PR tree-optimization/83640
11879         * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
11880         subtracting negative offset from size.
11881         (builtin_access::overlap): Adjust offset bounds of the access to fall
11882         within the size of the object if possible.
11883
11884 2018-01-06  Richard Sandiford  <richard.sandiford@linaro.org>
11885
11886         PR rtl-optimization/83699
11887         * expmed.c (extract_bit_field_1): Restrict the vector usage of
11888         extract_bit_field_as_subreg to cases in which the extracted
11889         value is also a vector.
11890
11891         * lra-constraints.c (process_alt_operands): Test for the equivalence
11892         substitutions when detecting a possible reload cycle.
11893
11894 2018-01-06  Jakub Jelinek  <jakub@redhat.com>
11895
11896         PR debug/83480
11897         * toplev.c (process_options): Don't enable debug_nonbind_markers_p
11898         by default if flag_selective_schedling{,2}.  Formatting fixes.
11899
11900         PR rtl-optimization/83682
11901         * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
11902         if it has non-VECTOR_MODE element mode.
11903         (vec_duplicate_p): Likewise.
11904
11905         PR middle-end/83694
11906         * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
11907         and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
11908
11909 2018-01-05  Jakub Jelinek  <jakub@redhat.com>
11910
11911         PR target/83604
11912         * config/i386/i386-builtin.def
11913         (__builtin_ia32_vgf2p8affineinvqb_v64qi,
11914         __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
11915         Require also OPTION_MASK_ISA_AVX512F in addition to
11916         OPTION_MASK_ISA_GFNI.
11917         (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
11918         __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
11919         OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
11920         to OPTION_MASK_ISA_GFNI.
11921         (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
11922         OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
11923         OPTION_MASK_ISA_AVX512BW.
11924         (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
11925         OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
11926         addition to OPTION_MASK_ISA_GFNI.
11927         (__builtin_ia32_vgf2p8affineinvqb_v16qi,
11928         __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
11929         Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
11930         to OPTION_MASK_ISA_GFNI.
11931         * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
11932         a requirement for all ISAs rather than any of them with a few
11933         exceptions.
11934         (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
11935         processing.
11936         (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
11937         bitmasks to be enabled with 3 exceptions, instead of requiring any
11938         enabled ISA with lots of exceptions.
11939         * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
11940         vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
11941         Change avx512bw in isa attribute to avx512f.
11942         * config/i386/sgxintrin.h: Add license boilerplate.
11943         * config/i386/vaesintrin.h: Likewise.  Fix macro spelling __AVX512F
11944         to __AVX512F__ and __AVX512VL to __AVX512VL__.
11945         (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
11946         _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
11947         defined.
11948         * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
11949         _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
11950         temporarily sse2 rather than sse if not enabled already.
11951
11952         PR target/83604
11953         * config/i386/sse.md (VI248_VLBW): Rename to ...
11954         (VI248_AVX512VL): ... this.  Don't guard V32HI with TARGET_AVX512BW.
11955         (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
11956         vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
11957         vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
11958         vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
11959         mode iterator instead of VI248_VLBW.
11960
11961 2018-01-05  Jan Hubicka  <hubicka@ucw.cz>
11962
11963         * ipa-fnsummary.c (record_modified_bb_info): Add OP.
11964         (record_modified): Skip clobbers; add debug output.
11965         (param_change_prob): Use sreal frequencies.
11966
11967 2018-01-05  Richard Sandiford  <richard.sandiford@linaro.org>
11968
11969         * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
11970         punt for user-aligned variables.
11971
11972 2018-01-05  Richard Sandiford  <richard.sandiford@linaro.org>
11973
11974         * tree-chrec.c (chrec_contains_symbols): Return true for
11975         POLY_INT_CST.
11976
11977 2018-01-05  Sudakshina Das  <sudi.das@arm.com>
11978
11979         PR target/82439
11980         * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
11981         of (x|y) == x for BICS pattern.
11982
11983 2018-01-05  Jakub Jelinek  <jakub@redhat.com>
11984
11985         PR tree-optimization/83605
11986         * gimple-ssa-strength-reduction.c: Include tree-eh.h.
11987         (find_candidates_dom_walker::before_dom_children): Ignore stmts that
11988         can throw.
11989
11990 2018-01-05  Sebastian Huber  <sebastian.huber@embedded-brains.de>
11991
11992         * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
11993         * config/epiphany/rtems.h: New file.
11994
11995 2018-01-04  Jakub Jelinek  <jakub@redhat.com>
11996             Uros Bizjak  <ubizjak@gmail.com>
11997
11998         PR target/83554
11999         * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
12000         QIreg_operand instead of register_operand predicate.
12001         * config/i386/i386.c (ix86_rop_should_change_byte_p,
12002         set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
12003         comments instead of -fmitigate[-_]rop.
12004
12005 2018-01-04  Rainer Orth  <ro@CeBiTec.Uni-Bielefeld.DE>
12006
12007         PR bootstrap/81926
12008         * cgraphunit.c (symbol_table::compile): Switch to text_section
12009         before calling assembly_start debug hook.
12010         * run-rtl-passes.c (run_rtl_passes): Likewise.
12011         Include output.h.
12012
12013 2018-01-04  Richard Sandiford  <richard.sandiford@linaro.org>
12014
12015         * tree-vrp.c (extract_range_from_binary_expr_1): Check
12016         range_int_cst_p rather than !symbolic_range_p before calling
12017         extract_range_from_multiplicative_op_1.
12018
12019 2018-01-04  Jeff Law  <law@redhat.com>
12020
12021         * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
12022         redundant test in assertion.
12023
12024 2018-01-04  Richard Sandiford  <richard.sandiford@linaro.org>
12025
12026         * doc/rtl.texi: Document machine_mode wrapper classes.
12027
12028 2018-01-04  Richard Sandiford  <richard.sandiford@linaro.org>
12029
12030         * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
12031         using tree_to_uhwi.
12032
12033 2018-01-04  Richard Sandiford  <richard.sandiford@linaro.org>
12034
12035         * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
12036         the VEC_PERM_EXPR fold to fail.
12037
12038 2018-01-04  Jakub Jelinek  <jakub@redhat.com>
12039
12040         PR debug/83585
12041         * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
12042         to switched_sections.
12043
12044 2018-01-04  Richard Sandiford  <richard.sandiford@linaro.org>
12045
12046         PR target/83680
12047         * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
12048         test for d.testing.
12049
12050 2018-01-04  Peter Bergner  <bergner@vnet.ibm.com>
12051
12052         PR target/83387
12053         * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
12054         allow arguments in FP registers if TARGET_HARD_FLOAT is false.
12055
12056 2018-01-04  Jakub Jelinek  <jakub@redhat.com>
12057
12058         PR debug/83666
12059         * cfgexpand.c (expand_debug_expr) <case BIT_FIELD_REF>: Punt if mode
12060         is BLKmode and bitpos not zero or mode change is needed.
12061
12062 2018-01-04  Richard Sandiford  <richard.sandiford@linaro.org>
12063
12064         PR target/83675
12065         * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
12066         TARGET_VIS2.
12067
12068 2018-01-04  Uros Bizjak  <ubizjak@gmail.com>
12069
12070         PR target/83628
12071         * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
12072         instead of MULT rtx.  Update all corresponding splitters.
12073         (*saddl_se): Ditto.
12074         (*ssub<modesuffix>): Ditto.
12075         (*ssubl_se): Ditto.
12076         (*cmp_sadd_di): Update split patterns.
12077         (*cmp_sadd_si): Ditto.
12078         (*cmp_sadd_sidi): Ditto.
12079         (*cmp_ssub_di): Ditto.
12080         (*cmp_ssub_si): Ditto.
12081         (*cmp_ssub_sidi): Ditto.
12082         * config/alpha/predicates.md (const23_operand): New predicate.
12083         * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
12084         Look for ASHIFT, not MULT inner operand.
12085         (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
12086
12087 2018-01-04  Martin Liska  <mliska@suse.cz>
12088
12089         PR gcov-profile/83669
12090         * gcov.c (output_intermediate_file): Add version to intermediate
12091         gcov file.
12092         * doc/gcov.texi: Document new field 'version' in intermediate
12093         file format. Fix location of '-k' option of gcov command.
12094
12095 2018-01-04  Martin Liska  <mliska@suse.cz>
12096
12097         PR ipa/82352
12098         * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
12099
12100 2018-01-04  Jakub Jelinek  <jakub@redhat.com>
12101
12102         * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
12103
12104 2018-01-03  Martin Sebor  <msebor@redhat.com>
12105
12106         PR tree-optimization/83655
12107         * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
12108         checking calls with invalid arguments.
12109
12110 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12111
12112         * tree-vect-stmts.c (vect_get_store_rhs): New function.
12113         (vectorizable_mask_load_store): Delete.
12114         (vectorizable_call): Return false for masked loads and stores.
12115         (vectorizable_store): Handle IFN_MASK_STORE.  Use vect_get_store_rhs
12116         instead of gimple_assign_rhs1.
12117         (vectorizable_load): Handle IFN_MASK_LOAD.
12118         (vect_transform_stmt): Don't set is_store for call_vec_info_type.
12119
12120 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12121
12122         * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
12123         split out from..,
12124         (vectorizable_mask_load_store): ...here.
12125         (vectorizable_load): ...and here.
12126
12127 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12128
12129         * tree-vect-stmts.c (vect_build_all_ones_mask)
12130         (vect_build_zero_merge_argument): New functions, split out from...
12131         (vectorizable_load): ...here.
12132
12133 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12134
12135         * tree-vect-stmts.c (vect_check_store_rhs): New function,
12136         split out from...
12137         (vectorizable_mask_load_store): ...here.
12138         (vectorizable_store): ...and here.
12139
12140 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12141
12142         * tree-vect-stmts.c (vect_check_load_store_mask): New function,
12143         split out from...
12144         (vectorizable_mask_load_store): ...here.
12145
12146 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12147
12148         * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
12149         (vect_model_store_cost): Take a vec_load_store_type instead of a
12150         vect_def_type.
12151         * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
12152         (vect_model_store_cost): Take a vec_load_store_type instead of a
12153         vect_def_type.
12154         (vectorizable_mask_load_store): Update accordingly.
12155         (vectorizable_store): Likewise.
12156         * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
12157
12158 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12159
12160         * tree-vect-loop.c (vect_transform_loop): Stub out scalar
12161         IFN_MASK_LOAD calls here rather than...
12162         * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
12163
12164 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12165             Alan Hayward  <alan.hayward@arm.com>
12166             David Sherwood  <david.sherwood@arm.com>
12167
12168         * expmed.c (extract_bit_field_1): For vector extracts,
12169         fall back to extract_bit_field_as_subreg if vec_extract
12170         isn't available.
12171
12172 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12173             Alan Hayward  <alan.hayward@arm.com>
12174             David Sherwood  <david.sherwood@arm.com>
12175
12176         * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
12177         they are variable or constant sized.
12178         (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
12179         slots for constant-sized data.
12180
12181 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12182             Alan Hayward  <alan.hayward@arm.com>
12183             David Sherwood  <david.sherwood@arm.com>
12184
12185         * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
12186         handling COND_EXPRs with boolean comparisons, try to find a better
12187         basis for the mask type than the boolean itself.
12188
12189 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12190
12191         * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
12192         is calculated and how it can be overridden.
12193         * genmodes.c (max_bitsize_mode_any_mode): New variable.
12194         (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
12195         if defined.
12196         (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
12197         if nonzero.
12198
12199 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12200             Alan Hayward  <alan.hayward@arm.com>
12201             David Sherwood  <david.sherwood@arm.com>
12202
12203         * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
12204         Remove the mode argument.
12205         (aarch64_simd_valid_immediate): Remove the mode and inverse
12206         arguments.
12207         * config/aarch64/iterators.md (bitsize): New iterator.
12208         * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
12209         (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
12210         * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
12211         aarch64_simd_valid_immediate.
12212         * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
12213         (aarch64_reg_or_bic_imm): Likewise.
12214         * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
12215         with an insn_type enum and msl with a modifier_type enum.
12216         Replace element_width with a scalar_mode.  Change the shift
12217         to unsigned int.  Add constructors for scalar_float_mode and
12218         scalar_int_mode elements.
12219         (aarch64_vect_float_const_representable_p): Delete.
12220         (aarch64_can_const_movi_rtx_p)
12221         (aarch64_simd_scalar_immediate_valid_for_move)
12222         (aarch64_simd_make_constant): Update call to
12223         aarch64_simd_valid_immediate.
12224         (aarch64_advsimd_valid_immediate_hs): New function.
12225         (aarch64_advsimd_valid_immediate): Likewise.
12226         (aarch64_simd_valid_immediate): Remove mode and inverse
12227         arguments.  Rewrite to use the above.  Use const_vec_duplicate_p
12228         to detect duplicated constants and use aarch64_float_const_zero_rtx_p
12229         and aarch64_float_const_representable_p on the result.
12230         (aarch64_output_simd_mov_immediate): Remove mode argument.
12231         Update call to aarch64_simd_valid_immediate and use of
12232         simd_immediate_info.
12233         (aarch64_output_scalar_simd_mov_immediate): Update call
12234         accordingly.
12235
12236 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12237             Alan Hayward  <alan.hayward@arm.com>
12238             David Sherwood  <david.sherwood@arm.com>
12239
12240         * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
12241         (mode_nunits): Likewise CONST_MODE_NUNITS.
12242         * machmode.def (ADJUST_NUNITS): Document.
12243         * genmodes.c (mode_data::need_nunits_adj): New field.
12244         (blank_mode): Update accordingly.
12245         (adj_nunits): New variable.
12246         (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
12247         parameter.
12248         (emit_mode_size_inline): Set need_bytesize_adj for all modes
12249         listed in adj_nunits.
12250         (emit_mode_nunits_inline): Set need_nunits_adj for all modes
12251         listed in adj_nunits.  Don't emit case statements for such modes.
12252         (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
12253         and CONST_MODE_PRECISION.  Make CONST_MODE_SIZE expand to
12254         nothing if adj_nunits is nonnull.
12255         (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
12256         (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
12257         (emit_mode_fbit): Update use of print_maybe_const_decl.
12258         (emit_move_size): Likewise.  Treat the array as non-const
12259         if adj_nunits.
12260         (emit_mode_adjustments): Handle adj_nunits.
12261
12262 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12263
12264         * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
12265         * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
12266         (VECTOR_MODES): Use it.
12267         (make_vector_modes): Take the prefix as an argument.
12268
12269 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12270             Alan Hayward  <alan.hayward@arm.com>
12271             David Sherwood  <david.sherwood@arm.com>
12272
12273         * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
12274         * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
12275         for MODE_VECTOR_BOOL.
12276         * machmode.def (VECTOR_BOOL_MODE): Document.
12277         * genmodes.c (VECTOR_BOOL_MODE): New macro.
12278         (make_vector_bool_mode): New function.
12279         (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
12280         MODE_VECTOR_BOOL.
12281         * lto-streamer-in.c (lto_input_mode_table): Likewise.
12282         * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
12283         Likewise.
12284         * stor-layout.c (int_mode_for_mode): Likewise.
12285         * tree.c (build_vector_type_for_mode): Likewise.
12286         * varasm.c (output_constant_pool_2): Likewise.
12287         * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
12288         CONSTM1_RTX (BImode) are the same thing.  Initialize const_tiny_rtx
12289         for MODE_VECTOR_BOOL.
12290         * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
12291         of mode class checks.
12292         * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
12293         instead of a list of mode class checks.
12294         (expand_vector_scalar_condition): Likewise.
12295         (type_for_widest_vector_mode): Handle BImode as an inner mode.
12296
12297 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12298             Alan Hayward  <alan.hayward@arm.com>
12299             David Sherwood  <david.sherwood@arm.com>
12300
12301         * machmode.h (mode_size): Change from unsigned short to
12302         poly_uint16_pod.
12303         (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
12304         (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
12305         or if measurement_type is not polynomial.
12306         (fixed_size_mode::includes_p): Check for constant-sized modes.
12307         * genmodes.c (emit_mode_size_inline): Make mode_size_inline
12308         return a poly_uint16 rather than an unsigned short.
12309         (emit_mode_size): Change the type of mode_size from unsigned short
12310         to poly_uint16_pod.  Use ZERO_COEFFS for the initializer.
12311         (emit_mode_adjustments): Cope with polynomial vector sizes.
12312         * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
12313         for GET_MODE_SIZE.
12314         * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
12315         for GET_MODE_SIZE.
12316         * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
12317         * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
12318         * caller-save.c (setup_save_areas): Likewise.
12319         (replace_reg_with_saved_mem): Likewise.
12320         * calls.c (emit_library_call_value_1): Likewise.
12321         * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
12322         * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
12323         (gen_lowpart_for_combine): Likewise.
12324         * convert.c (convert_to_integer_1): Likewise.
12325         * cse.c (equiv_constant, cse_insn): Likewise.
12326         * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
12327         (cselib_subst_to_values): Likewise.
12328         * dce.c (word_dce_process_block): Likewise.
12329         * df-problems.c (df_word_lr_mark_ref): Likewise.
12330         * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
12331         * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
12332         (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
12333         (rtl_for_decl_location): Likewise.
12334         * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
12335         * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
12336         * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
12337         (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
12338         (expand_expr_real_1): Likewise.
12339         * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
12340         (pad_below): Likewise.
12341         * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
12342         * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
12343         * ira.c (get_subreg_tracking_sizes): Likewise.
12344         * ira-build.c (ira_create_allocno_objects): Likewise.
12345         * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
12346         (ira_sort_regnos_for_alter_reg): Likewise.
12347         * ira-costs.c (record_operand_costs): Likewise.
12348         * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
12349         (resolve_simple_move): Likewise.
12350         * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
12351         (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
12352         (lra_constraints): Likewise.
12353         (CONST_POOL_OK_P): Reject variable-sized modes.
12354         * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
12355         (add_pseudo_to_slot, lra_spill): Likewise.
12356         * omp-low.c (omp_clause_aligned_alignment): Likewise.
12357         * optabs-query.c (get_best_extraction_insn): Likewise.
12358         * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
12359         * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
12360         (expand_mult_highpart, valid_multiword_target_p): Likewise.
12361         * recog.c (offsettable_address_addr_space_p): Likewise.
12362         * regcprop.c (maybe_mode_change): Likewise.
12363         * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
12364         * regrename.c (build_def_use): Likewise.
12365         * regstat.c (dump_reg_info): Likewise.
12366         * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
12367         (find_reloads, find_reloads_subreg_address): Likewise.
12368         * reload1.c (eliminate_regs_1): Likewise.
12369         * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
12370         * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
12371         (simplify_binary_operation_1, simplify_subreg): Likewise.
12372         * targhooks.c (default_function_arg_padding): Likewise.
12373         (default_hard_regno_nregs, default_class_max_nregs): Likewise.
12374         * tree-cfg.c (verify_gimple_assign_binary): Likewise.
12375         (verify_gimple_assign_ternary): Likewise.
12376         * tree-inline.c (estimate_move_cost): Likewise.
12377         * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
12378         * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
12379         (get_address_cost_ainc): Likewise.
12380         * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
12381         (vect_supportable_dr_alignment): Likewise.
12382         * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
12383         (vectorizable_reduction): Likewise.
12384         * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
12385         (vectorizable_operation, vectorizable_load): Likewise.
12386         * tree.c (build_same_sized_truth_vector_type): Likewise.
12387         * valtrack.c (cleanup_auto_inc_dec): Likewise.
12388         * var-tracking.c (emit_note_insn_var_location): Likewise.
12389         * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
12390         (ADDR_VEC_ALIGN): Likewise.
12391
12392 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12393             Alan Hayward  <alan.hayward@arm.com>
12394             David Sherwood  <david.sherwood@arm.com>
12395
12396         * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
12397         unsigned short.
12398         (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
12399         or if measurement_type is polynomial.
12400         * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
12401         * combine.c (make_extraction): Likewise.
12402         * dse.c (find_shift_sequence): Likewise.
12403         * dwarf2out.c (mem_loc_descriptor): Likewise.
12404         * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
12405         (extract_bit_field, extract_low_bits): Likewise.
12406         * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
12407         (optimize_bitfield_assignment_op, expand_assignment): Likewise.
12408         (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
12409         * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
12410         * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
12411         * reload.c (find_reloads): Likewise.
12412         * reload1.c (alter_reg): Likewise.
12413         * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
12414         * targhooks.c (default_secondary_memory_needed_mode): Likewise.
12415         * tree-if-conv.c (predicate_mem_writes): Likewise.
12416         * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
12417         * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
12418         * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
12419         * valtrack.c (dead_debug_insert_temp): Likewise.
12420         * varasm.c (mergeable_constant_section): Likewise.
12421         * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
12422
12423 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12424             Alan Hayward  <alan.hayward@arm.com>
12425             David Sherwood  <david.sherwood@arm.com>
12426
12427         * expr.c (expand_assignment): Cope with polynomial mode sizes
12428         when assigning to a CONCAT.
12429
12430 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12431             Alan Hayward  <alan.hayward@arm.com>
12432             David Sherwood  <david.sherwood@arm.com>
12433
12434         * machmode.h (mode_precision): Change from unsigned short to
12435         poly_uint16_pod.
12436         (mode_to_precision): Return a poly_uint16 rather than an unsigned
12437         short.
12438         (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
12439         or if measurement_type is not polynomial.
12440         (HWI_COMPUTABLE_MODE_P): Turn into a function.  Optimize the case
12441         in which the mode is already known to be a scalar_int_mode.
12442         * genmodes.c (emit_mode_precision): Change the type of mode_precision
12443         from unsigned short to poly_uint16_pod.  Use ZERO_COEFFS for the
12444         initializer.
12445         * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
12446         for GET_MODE_PRECISION.
12447         * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
12448         for GET_MODE_PRECISION.
12449         * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
12450         as polynomial.
12451         (try_combine, find_split_point, combine_simplify_rtx): Likewise.
12452         (expand_field_assignment, make_extraction): Likewise.
12453         (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
12454         (get_last_value): Likewise.
12455         * convert.c (convert_to_integer_1): Likewise.
12456         * cse.c (cse_insn): Likewise.
12457         * expr.c (expand_expr_real_1): Likewise.
12458         * lra-constraints.c (simplify_operand_subreg): Likewise.
12459         * optabs-query.c (can_atomic_load_p): Likewise.
12460         * optabs.c (expand_atomic_load): Likewise.
12461         (expand_atomic_store): Likewise.
12462         * ree.c (combine_reaching_defs): Likewise.
12463         * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
12464         * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
12465         * tree.h (type_has_mode_precision_p): Likewise.
12466         * ubsan.c (instrument_si_overflow): Likewise.
12467
12468 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12469             Alan Hayward  <alan.hayward@arm.com>
12470             David Sherwood  <david.sherwood@arm.com>
12471
12472         * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
12473         polynomial numbers of units.
12474         (SET_TYPE_VECTOR_SUBPARTS): Likewise.
12475         (valid_vector_subparts_p): New function.
12476         (build_vector_type): Remove temporary shim and take the number
12477         of units as a poly_uint64 rather than an int.
12478         (build_opaque_vector_type): Take the number of units as a
12479         poly_uint64 rather than an int.
12480         * tree.c (build_vector_from_ctor): Handle polynomial
12481         TYPE_VECTOR_SUBPARTS.
12482         (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
12483         (uniform_vector_p, vector_type_mode, build_vector): Likewise.
12484         (build_vector_from_val): If the number of units is variable,
12485         use build_vec_duplicate_cst for constant operands and
12486         VEC_DUPLICATE_EXPR otherwise.
12487         (make_vector_type): Remove temporary is_constant ().
12488         (build_vector_type, build_opaque_vector_type): Take the number of
12489         units as a poly_uint64 rather than an int.
12490         (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
12491         VECTOR_CST_NELTS.
12492         * cfgexpand.c (expand_debug_expr): Likewise.
12493         * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
12494         (store_constructor, expand_expr_real_1): Likewise.
12495         (const_scalar_mask_from_tree): Likewise.
12496         * fold-const-call.c (fold_const_reduction): Likewise.
12497         * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
12498         (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
12499         (native_encode_vector, vec_cst_ctor_to_array): Likewise.
12500         (fold_relational_const): Likewise.
12501         (native_interpret_vector): Likewise.  Change the size from an
12502         int to an unsigned int.
12503         * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
12504         TYPE_VECTOR_SUBPARTS.
12505         (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
12506         (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
12507         duplicating a non-constant operand into a variable-length vector.
12508         * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
12509         TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
12510         * ipa-icf.c (sem_variable::equals): Likewise.
12511         * match.pd: Likewise.
12512         * omp-simd-clone.c (simd_clone_subparts): Likewise.
12513         * print-tree.c (print_node): Likewise.
12514         * stor-layout.c (layout_type): Likewise.
12515         * targhooks.c (default_builtin_vectorization_cost): Likewise.
12516         * tree-cfg.c (verify_gimple_comparison): Likewise.
12517         (verify_gimple_assign_binary): Likewise.
12518         (verify_gimple_assign_ternary): Likewise.
12519         (verify_gimple_assign_single): Likewise.
12520         * tree-pretty-print.c (dump_generic_node): Likewise.
12521         * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
12522         (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
12523         * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
12524         (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
12525         (vect_shift_permute_load_chain): Likewise.
12526         * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
12527         (expand_vector_condition, optimize_vector_constructor): Likewise.
12528         (lower_vec_perm, get_compute_type): Likewise.
12529         * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
12530         (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
12531         * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
12532         (vect_recog_mask_conversion_pattern): Likewise.
12533         * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
12534         (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
12535         * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
12536         (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
12537         (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
12538         (vectorizable_shift, vectorizable_operation, vectorizable_store)
12539         (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
12540         (supportable_widening_operation): Likewise.
12541         (supportable_narrowing_operation): Likewise.
12542         * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
12543         Likewise.
12544         * varasm.c (output_constant): Likewise.
12545
12546 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12547             Alan Hayward  <alan.hayward@arm.com>
12548             David Sherwood  <david.sherwood@arm.com>
12549
12550         * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
12551         so that both the length == 3 and length != 3 cases set up their
12552         own permute vectors.  Add comments explaining why we know the
12553         number of elements is constant.
12554         (vect_permute_load_chain): Likewise.
12555
12556 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12557             Alan Hayward  <alan.hayward@arm.com>
12558             David Sherwood  <david.sherwood@arm.com>
12559
12560         * machmode.h (mode_nunits): Change from unsigned char to
12561         poly_uint16_pod.
12562         (ONLY_FIXED_SIZE_MODES): New macro.
12563         (pod_mode::measurement_type, scalar_int_mode::measurement_type)
12564         (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
12565         (complex_mode::measurement_type, fixed_size_mode::measurement_type):
12566         New typedefs.
12567         (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
12568         (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
12569         or if measurement_type is not polynomial.
12570         * genmodes.c (ZERO_COEFFS): New macro.
12571         (emit_mode_nunits_inline): Make mode_nunits_inline return a
12572         poly_uint16.
12573         (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
12574         Use ZERO_COEFFS when emitting initializers.
12575         * data-streamer.h (bp_pack_poly_value): New function.
12576         (bp_unpack_poly_value): Likewise.
12577         * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
12578         for GET_MODE_NUNITS.
12579         * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
12580         for GET_MODE_NUNITS.
12581         * tree.c (make_vector_type): Remove temporary shim and make
12582         the real function take the number of units as a poly_uint64
12583         rather than an int.
12584         (build_vector_type_for_mode): Handle polynomial nunits.
12585         * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
12586         * emit-rtl.c (const_vec_series_p_1): Likewise.
12587         (gen_rtx_CONST_VECTOR): Likewise.
12588         * fold-const.c (test_vec_duplicate_folding): Likewise.
12589         * genrecog.c (validate_pattern): Likewise.
12590         * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
12591         * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
12592         * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
12593         (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
12594         (expand_vec_cond_expr, expand_mult_highpart): Likewise.
12595         * rtlanal.c (subreg_get_info): Likewise.
12596         * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
12597         (vect_grouped_load_supported): Likewise.
12598         * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
12599         * tree-vect-loop.c (have_whole_vector_shift): Likewise.
12600         * simplify-rtx.c (simplify_unary_operation_1): Likewise.
12601         (simplify_const_unary_operation, simplify_binary_operation_1)
12602         (simplify_const_binary_operation, simplify_ternary_operation)
12603         (test_vector_ops_duplicate, test_vector_ops): Likewise.
12604         (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
12605         instead of CONST_VECTOR_NUNITS.
12606         * varasm.c (output_constant_pool_2): Likewise.
12607         * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
12608         explicit-encoded elements in the XVEC for variable-length vectors.
12609
12610 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12611
12612         * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
12613
12614 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12615             Alan Hayward  <alan.hayward@arm.com>
12616             David Sherwood  <david.sherwood@arm.com>
12617
12618         * coretypes.h (fixed_size_mode): Declare.
12619         (fixed_size_mode_pod): New typedef.
12620         * builtins.h (target_builtins::x_apply_args_mode)
12621         (target_builtins::x_apply_result_mode): Change type to
12622         fixed_size_mode_pod.
12623         * builtins.c (apply_args_size, apply_result_size, result_vector)
12624         (expand_builtin_apply_args_1, expand_builtin_apply)
12625         (expand_builtin_return): Update accordingly.
12626
12627 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12628
12629         * cse.c (hash_rtx_cb): Hash only the encoded elements.
12630         * cselib.c (cselib_hash_rtx): Likewise.
12631         * expmed.c (make_tree): Build VECTOR_CSTs directly from the
12632         CONST_VECTOR encoding.
12633
12634 2018-01-03  Jakub Jelinek  <jakub@redhat.com>
12635             Jeff Law  <law@redhat.com>
12636
12637         PR target/83641
12638         * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
12639         noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
12640         only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
12641         and add REG_CFA_ADJUST_CFA notes in that case to both insns.
12642
12643         PR target/83641
12644         * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
12645         explicitly probe *sp in a noreturn function if there were any callee
12646         register saves or frame pointer is needed.
12647
12648 2018-01-03  Jakub Jelinek  <jakub@redhat.com>
12649
12650         PR debug/83621
12651         * cfgexpand.c (expand_debug_expr): Return NULL if mode is
12652         BLKmode for ternary, binary or unary expressions.
12653
12654         PR debug/83645
12655         * var-tracking.c (delete_vta_debug_insn): New inline function.
12656         (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
12657         insns from get_insns () to NULL instead of each bb separately.
12658         Use delete_vta_debug_insn.  No longer static.
12659         (vt_debug_insns_local, variable_tracking_main_1): Adjust
12660         delete_vta_debug_insns callers.
12661         * rtl.h (delete_vta_debug_insns): Declare.
12662         * final.c (rest_of_handle_final): Call delete_vta_debug_insns
12663         instead of variable_tracking_main.
12664
12665 2018-01-03  Martin Sebor  <msebor@redhat.com>
12666
12667         PR tree-optimization/83603
12668         * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
12669         arguments past the endof the argument list in functions declared
12670         without a prototype.
12671         * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
12672         Avoid checking when arguments are null.
12673
12674 2018-01-03  Martin Sebor  <msebor@redhat.com>
12675
12676         PR c/83559
12677         * doc/extend.texi (attribute const): Fix a typo.
12678         * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
12679         issuing -Wsuggest-attribute for void functions.
12680
12681 2018-01-03  Martin Sebor  <msebor@redhat.com>
12682
12683         * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
12684         offset_int::from instead of wide_int::to_shwi.
12685         (maybe_diag_overlap): Remove assertion.
12686         Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
12687         * gimple-ssa-sprintf.c (format_directive): Same.
12688         (parse_directive): Same.
12689         (sprintf_dom_walker::compute_format_length): Same.
12690         (try_substitute_return_value): Same.
12691
12692 2018-01-03  Jeff Law  <law@redhat.com>
12693
12694         PR middle-end/83654
12695         * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
12696         non-constant residual for zero at runtime and avoid probing in
12697         that case.  Reorganize code for trailing problem to mirror handling
12698         of the residual.
12699
12700 2018-01-03  Prathamesh Kulkarni  <prathamesh.kulkarni@linaro.org>
12701
12702         PR tree-optimization/83501
12703         * tree-ssa-strlen.c (get_string_cst): New.
12704         (handle_char_store): Call get_string_cst.
12705
12706 2018-01-03  Martin Liska  <mliska@suse.cz>
12707
12708         PR tree-optimization/83593
12709         * tree-ssa-strlen.c: Include tree-cfg.h.
12710         (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
12711         (strlen_dom_walker): Add new member variable m_cleanup_cfg.
12712         (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
12713         to false.
12714         (strlen_dom_walker::before_dom_children): Call
12715         gimple_purge_dead_eh_edges. Dump tranformation with details
12716         dump flags.
12717         (strlen_dom_walker::before_dom_children): Update call by adding
12718         new argument cleanup_eh.
12719         (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
12720
12721 2018-01-03  Martin Liska  <mliska@suse.cz>
12722
12723         PR ipa/83549
12724         * cif-code.def (VARIADIC_THUNK): New enum value.
12725         * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
12726         thunks.
12727
12728 2018-01-03  Jan Beulich  <jbeulich@suse.com>
12729
12730         * sse.md (mov<mode>_internal): Tighten condition for when to use
12731         vmovdqu<ssescalarsize> for TI and OI modes.
12732
12733 2018-01-03  Jakub Jelinek  <jakub@redhat.com>
12734
12735         Update copyright years.
12736
12737 2018-01-03  Martin Liska  <mliska@suse.cz>
12738
12739         PR ipa/83594
12740         * ipa-visibility.c (function_and_variable_visibility): Skip
12741         functions with noipa attribure.
12742
12743 2018-01-03  Jakub Jelinek  <jakub@redhat.com>
12744
12745         * gcc.c (process_command): Update copyright notice dates.
12746         * gcov-dump.c (print_version): Ditto.
12747         * gcov.c (print_version): Ditto.
12748         * gcov-tool.c (print_version): Ditto.
12749         * gengtype.c (create_file): Ditto.
12750         * doc/cpp.texi: Bump @copying's copyright year.
12751         * doc/cppinternals.texi: Ditto.
12752         * doc/gcc.texi: Ditto.
12753         * doc/gccint.texi: Ditto.
12754         * doc/gcov.texi: Ditto.
12755         * doc/install.texi: Ditto.
12756         * doc/invoke.texi: Ditto.
12757
12758 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12759
12760         * vector-builder.h (vector_builder::m_full_nelts): Change from
12761         unsigned int to poly_uint64.
12762         (vector_builder::full_nelts): Update prototype accordingly.
12763         (vector_builder::new_vector): Likewise.
12764         (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
12765         (vector_builder::operator ==): Likewise.
12766         (vector_builder::finalize): Likewise.
12767         * int-vector-builder.h (int_vector_builder::int_vector_builder):
12768         Take the number of elements as a poly_uint64 rather than an
12769         unsigned int.
12770         * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
12771         from unsigned int to poly_uint64.
12772         (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
12773         (vec_perm_indices::new_vector): Likewise.
12774         (vec_perm_indices::length): Likewise.
12775         (vec_perm_indices::nelts_per_input): Likewise.
12776         (vec_perm_indices::input_nelts): Likewise.
12777         * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
12778         number of elements per input as a poly_uint64 rather than an
12779         unsigned int.  Use the original encoding for variable-length
12780         vectors, rather than clamping each individual element.
12781         For the second and subsequent elements in each pattern,
12782         clamp the step and base before clamping their sum.
12783         (vec_perm_indices::series_p): Handle polynomial element counts.
12784         (vec_perm_indices::all_in_range_p): Likewise.
12785         (vec_perm_indices_to_tree): Likewise.
12786         (vec_perm_indices_to_rtx): Likewise.
12787         * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
12788         * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
12789         (tree_vector_builder::new_binary_operation): Handle polynomial
12790         element counts.  Return false if we need to know the number
12791         of elements at compile time.
12792         * fold-const.c (fold_vec_perm): Punt if the number of elements
12793         isn't known at compile time.
12794
12795 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12796
12797         * vec-perm-indices.h (vec_perm_builder): Change element type
12798         from HOST_WIDE_INT to poly_int64.
12799         (vec_perm_indices::element_type): Update accordingly.
12800         (vec_perm_indices::clamp): Handle polynomial element_types.
12801         * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
12802         (vec_perm_indices::all_in_range_p): Likewise.
12803         (tree_to_vec_perm_builder): Check for poly_int64 trees rather
12804         than shwi trees.
12805         * vector-builder.h (vector_builder::stepped_sequence_p): Handle
12806         polynomial vec_perm_indices element types.
12807         * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
12808         * fold-const.c (fold_vec_perm): Likewise.
12809         * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
12810         * tree-vect-generic.c (lower_vec_perm): Likewise.
12811         * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
12812         * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
12813         element type to HOST_WIDE_INT.
12814
12815 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12816             Alan Hayward  <alan.hayward@arm.com>
12817             David Sherwood  <david.sherwood@arm.com>
12818
12819         * alias.c (addr_side_effect_eval): Take the size as a poly_int64
12820         rather than an int.  Use plus_constant.
12821         (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
12822         Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
12823
12824 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12825             Alan Hayward  <alan.hayward@arm.com>
12826             David Sherwood  <david.sherwood@arm.com>
12827
12828         * calls.c (emit_call_1, expand_call): Change struct_value_size from
12829         a HOST_WIDE_INT to a poly_int64.
12830
12831 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12832             Alan Hayward  <alan.hayward@arm.com>
12833             David Sherwood  <david.sherwood@arm.com>
12834
12835         * calls.c (load_register_parameters): Cope with polynomial
12836         mode sizes.  Require a constant size for BLKmode parameters
12837         that aren't described by a PARALLEL.  If BLOCK_REG_PADDING
12838         forces a parameter to be padded at the lsb end in order to
12839         fill a complete number of words, require the parameter size
12840         to be ordered wrt UNITS_PER_WORD.
12841
12842 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12843             Alan Hayward  <alan.hayward@arm.com>
12844             David Sherwood  <david.sherwood@arm.com>
12845
12846         * reload1.c (spill_stack_slot_width): Change element type
12847         from unsigned int to poly_uint64_pod.
12848         (alter_reg): Treat mode sizes as polynomial.
12849
12850 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12851             Alan Hayward  <alan.hayward@arm.com>
12852             David Sherwood  <david.sherwood@arm.com>
12853
12854         * reload.c (complex_word_subreg_p): New function.
12855         (reload_inner_reg_of_subreg, push_reload): Use it.
12856
12857 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12858             Alan Hayward  <alan.hayward@arm.com>
12859             David Sherwood  <david.sherwood@arm.com>
12860
12861         * lra-constraints.c (process_alt_operands): Reject matched
12862         operands whose sizes aren't ordered.
12863         (match_reload): Refer to this check here.
12864
12865 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12866             Alan Hayward  <alan.hayward@arm.com>
12867             David Sherwood  <david.sherwood@arm.com>
12868
12869         * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
12870         that the mode size is in the set {1, 2, 4, 8, 16}.
12871
12872 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12873             Alan Hayward  <alan.hayward@arm.com>
12874             David Sherwood  <david.sherwood@arm.com>
12875
12876         * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
12877         Use plus_constant instead of gen_rtx_PLUS.
12878
12879 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12880             Alan Hayward  <alan.hayward@arm.com>
12881             David Sherwood  <david.sherwood@arm.com>
12882
12883         * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
12884         * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
12885         * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
12886         * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
12887         * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
12888         * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
12889         * config/i386/i386-protos.h (ix86_push_rounding): Declare.
12890         * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
12891         * config/i386/i386.c (ix86_push_rounding): ...this new function.
12892         * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
12893         a poly_int64.
12894         * config/m32c/m32c.c (m32c_push_rounding): Likewise.
12895         * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
12896         * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
12897         * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
12898         * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
12899         * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
12900         * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
12901         * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
12902         * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
12903         * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
12904         function.
12905         * expr.c (emit_move_resolve_push): Treat the input and result
12906         of PUSH_ROUNDING as a poly_int64.
12907         (emit_move_complex_push, emit_single_push_insn_1): Likewise.
12908         (emit_push_insn): Likewise.
12909         * lra-eliminations.c (mark_not_eliminable): Likewise.
12910         * recog.c (push_operand): Likewise.
12911         * reload1.c (elimination_effects): Likewise.
12912         * rtlanal.c (nonzero_bits1): Likewise.
12913         * calls.c (store_one_arg): Likewise.  Require the padding to be
12914         known at compile time.
12915
12916 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12917             Alan Hayward  <alan.hayward@arm.com>
12918             David Sherwood  <david.sherwood@arm.com>
12919
12920         * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
12921         Use plus_constant instead of gen_rtx_PLUS.
12922
12923 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12924             Alan Hayward  <alan.hayward@arm.com>
12925             David Sherwood  <david.sherwood@arm.com>
12926
12927         * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
12928         rather than an int.
12929
12930 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12931             Alan Hayward  <alan.hayward@arm.com>
12932             David Sherwood  <david.sherwood@arm.com>
12933
12934         * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
12935         instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
12936         via stack temporaries.  Treat the mode size as polynomial too.
12937
12938 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12939             Alan Hayward  <alan.hayward@arm.com>
12940             David Sherwood  <david.sherwood@arm.com>
12941
12942         * expr.c (expand_expr_real_2): When handling conversions involving
12943         unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
12944         multiplying int_size_in_bytes by BITS_PER_UNIT.  Treat GET_MODE_BISIZE
12945         as a poly_uint64 too.
12946
12947 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12948             Alan Hayward  <alan.hayward@arm.com>
12949             David Sherwood  <david.sherwood@arm.com>
12950
12951         * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
12952
12953 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12954             Alan Hayward  <alan.hayward@arm.com>
12955             David Sherwood  <david.sherwood@arm.com>
12956
12957         * combine.c (can_change_dest_mode): Handle polynomial
12958         REGMODE_NATURAL_SIZE.
12959         * expmed.c (store_bit_field_1): Likewise.
12960         * expr.c (store_constructor): Likewise.
12961         * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
12962         and polynomial REGMODE_NATURAL_SIZE.
12963         (gen_lowpart_common): Likewise.
12964         * reginfo.c (record_subregs_of_mode): Likewise.
12965         * rtlanal.c (read_modify_subreg_p): Likewise.
12966
12967 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12968             Alan Hayward  <alan.hayward@arm.com>
12969             David Sherwood  <david.sherwood@arm.com>
12970
12971         * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
12972         numbers of elements.
12973
12974 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12975             Alan Hayward  <alan.hayward@arm.com>
12976             David Sherwood  <david.sherwood@arm.com>
12977
12978         * match.pd: Cope with polynomial numbers of vector elements.
12979
12980 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12981             Alan Hayward  <alan.hayward@arm.com>
12982             David Sherwood  <david.sherwood@arm.com>
12983
12984         * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
12985         in a POINTER_PLUS_EXPR.
12986
12987 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12988             Alan Hayward  <alan.hayward@arm.com>
12989             David Sherwood  <david.sherwood@arm.com>
12990
12991         * omp-simd-clone.c (simd_clone_subparts): New function.
12992         (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
12993         (ipa_simd_modify_function_body): Likewise.
12994
12995 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12996             Alan Hayward  <alan.hayward@arm.com>
12997             David Sherwood  <david.sherwood@arm.com>
12998
12999         * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
13000         (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
13001         (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
13002         (expand_vector_condition, vector_element): Likewise.
13003         (subparts_gt): New function.
13004         (get_compute_type): Use subparts_gt.
13005         (count_type_subparts): Delete.
13006         (expand_vector_operations_1): Use subparts_gt instead of
13007         count_type_subparts.
13008
13009 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
13010             Alan Hayward  <alan.hayward@arm.com>
13011             David Sherwood  <david.sherwood@arm.com>
13012
13013         * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
13014         (vect_compile_time_alias): ...this new function.  Do the calculation
13015         on poly_ints rather than trees.
13016         (vect_prune_runtime_alias_test_list): Update call accordingly.
13017
13018 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
13019             Alan Hayward  <alan.hayward@arm.com>
13020             David Sherwood  <david.sherwood@arm.com>
13021
13022         * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
13023         numbers of units.
13024         (vect_schedule_slp_instance): Likewise.
13025
13026 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
13027             Alan Hayward  <alan.hayward@arm.com>
13028             David Sherwood  <david.sherwood@arm.com>
13029
13030         * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
13031         constant and extern definitions for variable-length vectors.
13032         (vect_get_constant_vectors): Note that the number of units
13033         is known to be constant.
13034
13035 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
13036             Alan Hayward  <alan.hayward@arm.com>
13037             David Sherwood  <david.sherwood@arm.com>
13038
13039         * tree-vect-stmts.c (vectorizable_conversion): Treat the number
13040         of units as polynomial.  Choose between WIDE and NARROW based
13041         on multiple_p.
13042
13043 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
13044             Alan Hayward  <alan.hayward@arm.com>
13045             David Sherwood  <david.sherwood@arm.com>
13046
13047         * tree-vect-stmts.c (simd_clone_subparts): New function.
13048         (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
13049
13050 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
13051             Alan Hayward  <alan.hayward@arm.com>
13052             David Sherwood  <david.sherwood@arm.com>
13053
13054         * tree-vect-stmts.c (vectorizable_call): Treat the number of
13055         vectors as polynomial.  Use build_index_vector for
13056         IFN_GOMP_SIMD_LANE.
13057
13058 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
13059             Alan Hayward  <alan.hayward@arm.com>
13060             David Sherwood  <david.sherwood@arm.com>
13061
13062         * tree-vect-stmts.c (get_load_store_type): Treat the number of
13063         units as polynomial.  Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
13064         for variable-length vectors.
13065         (vectorizable_mask_load_store): Treat the number of units as
13066         polynomial, asserting that it is constant if the condition has
13067         already been enforced.
13068         (vectorizable_store, vectorizable_load): Likewise.
13069
13070 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
13071             Alan Hayward  <alan.hayward@arm.com>
13072             David Sherwood  <david.sherwood@arm.com>
13073
13074         * tree-vect-loop.c (vectorizable_live_operation): Treat the number
13075         of units as polynomial.  Punt if we can't tell at compile time
13076         which vector contains the final result.
13077
13078 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
13079             Alan Hayward  <alan.hayward@arm.com>
13080             David Sherwood  <david.sherwood@arm.com>
13081
13082         * tree-vect-loop.c (vectorizable_induction): Treat the number
13083         of units as polynomial.  Punt on SLP inductions.  Use an integer
13084         VEC_SERIES_EXPR for variable-length integer reductions.  Use a
13085         cast of such a series for variable-length floating-point
13086         reductions.
13087
13088 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
13089             Alan Hayward  <alan.hayward@arm.com>
13090             David Sherwood  <david.sherwood@arm.com>
13091
13092         * tree.h (build_index_vector): Declare.
13093         * tree.c (build_index_vector): New function.
13094         * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
13095         of units as polynomial, forcibly converting it to a constant if
13096         vectorizable_reduction has already enforced the condition.
13097         (vect_create_epilog_for_reduction): Likewise.  Use build_index_vector
13098         to create a {1,2,3,...} vector.
13099         (vectorizable_reduction): Treat the number of units as polynomial.
13100         Choose vectype_in based on the largest scalar element size rather
13101         than the smallest number of units.  Enforce the restrictions
13102         relied on above.
13103
13104 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
13105             Alan Hayward  <alan.hayward@arm.com>
13106             David Sherwood  <david.sherwood@arm.com>
13107
13108         * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
13109         number of units as polynomial.
13110
13111 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
13112             Alan Hayward  <alan.hayward@arm.com>
13113             David Sherwood  <david.sherwood@arm.com>
13114
13115         * target.h (vector_sizes, auto_vector_sizes): New typedefs.
13116         * target.def (autovectorize_vector_sizes): Return the vector sizes
13117         by pointer, using vector_sizes rather than a bitmask.
13118         * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
13119         * targhooks.c (default_autovectorize_vector_sizes): Likewise.
13120         * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
13121         Likewise.
13122         * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
13123         * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
13124         * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
13125         * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
13126         * omp-general.c (omp_max_vf): Likewise.
13127         * omp-low.c (omp_clause_aligned_alignment): Likewise.
13128         * optabs-query.c (can_vec_mask_load_store_p): Likewise.
13129         * tree-vect-loop.c (vect_analyze_loop): Likewise.
13130         * tree-vect-slp.c (vect_slp_bb): Likewise.
13131         * doc/tm.texi: Regenerate.
13132         * tree-vectorizer.h (current_vector_size): Change from an unsigned int
13133         to a poly_uint64.
13134         * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
13135         the vector size as a poly_uint64 rather than an unsigned int.
13136         (current_vector_size): Change from an unsigned int to a poly_uint64.
13137         (get_vectype_for_scalar_type): Update accordingly.
13138         * tree.h (build_truth_vector_type): Take the size and number of
13139         units as a poly_uint64 rather than an unsigned int.
13140         (build_vector_type): Add a temporary overload that takes
13141         the number of units as a poly_uint64 rather than an unsigned int.
13142         * tree.c (make_vector_type): Likewise.
13143         (build_truth_vector_type): Take the number of units as a poly_uint64
13144         rather than an unsigned int.
13145
13146 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
13147             Alan Hayward  <alan.hayward@arm.com>
13148             David Sherwood  <david.sherwood@arm.com>
13149
13150         * target.def (get_mask_mode): Take the number of units and length
13151         as poly_uint64s rather than unsigned ints.
13152         * targhooks.h (default_get_mask_mode): Update accordingly.
13153         * targhooks.c (default_get_mask_mode): Likewise.
13154         * config/i386/i386.c (ix86_get_mask_mode): Likewise.
13155         * doc/tm.texi: Regenerate.
13156
13157 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
13158             Alan Hayward  <alan.hayward@arm.com>
13159             David Sherwood  <david.sherwood@arm.com>
13160
13161         * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
13162         * omp-general.c (omp_max_vf): Likewise.
13163         * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
13164         (expand_omp_simd): Handle polynomial safelen.
13165         * omp-low.c (omplow_simd_context): Add a default constructor.
13166         (omplow_simd_context::max_vf): Change from int to poly_uint64.
13167         (lower_rec_simd_input_clauses): Update accordingly.
13168         (lower_rec_input_clauses): Likewise.
13169
13170 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
13171             Alan Hayward  <alan.hayward@arm.com>
13172             David Sherwood  <david.sherwood@arm.com>
13173
13174         * tree-vectorizer.h (vect_nunits_for_cost): New function.
13175         * tree-vect-loop.c (vect_model_reduction_cost): Use it.
13176         * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
13177         (vect_analyze_slp_cost): Likewise.
13178         * tree-vect-stmts.c (vect_model_store_cost): Likewise.
13179         (vect_model_load_cost): Likewise.
13180
13181 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
13182             Alan Hayward  <alan.hayward@arm.com>
13183             David Sherwood  <david.sherwood@arm.com>
13184
13185         * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
13186         (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
13187         from an unsigned int * to a poly_uint64_pod *.
13188         (calculate_unrolling_factor): New function.
13189         (vect_analyze_slp_instance): Use it.  Track polynomial max_nunits.
13190
13191 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
13192             Alan Hayward  <alan.hayward@arm.com>
13193             David Sherwood  <david.sherwood@arm.com>
13194
13195         * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
13196         from an unsigned int to a poly_uint64.
13197         (_loop_vec_info::slp_unrolling_factor): Likewise.
13198         (_loop_vec_info::vectorization_factor): Change from an int
13199         to a poly_uint64.
13200         (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
13201         (vect_get_num_vectors): New function.
13202         (vect_update_max_nunits, vect_vf_for_cost): Likewise.
13203         (vect_get_num_copies): Use vect_get_num_vectors.
13204         (vect_analyze_data_ref_dependences): Change max_vf from an int *
13205         to an unsigned int *.
13206         (vect_analyze_data_refs): Change min_vf from an int * to a
13207         poly_uint64 *.
13208         (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
13209         than an unsigned HOST_WIDE_INT.
13210         * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
13211         (vect_analyze_data_ref_dependence): Change max_vf from an int *
13212         to an unsigned int *.
13213         (vect_analyze_data_ref_dependences): Likewise.
13214         (vect_compute_data_ref_alignment): Handle polynomial vf.
13215         (vect_enhance_data_refs_alignment): Likewise.
13216         (vect_prune_runtime_alias_test_list): Likewise.
13217         (vect_shift_permute_load_chain): Likewise.
13218         (vect_supportable_dr_alignment): Likewise.
13219         (dependence_distance_ge_vf): Take the vectorization factor as a
13220         poly_uint64 rather than an unsigned HOST_WIDE_INT.
13221         (vect_analyze_data_refs): Change min_vf from an int * to a
13222         poly_uint64 *.
13223         * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
13224         vfm1 as a poly_uint64 rather than an int.  Make the same change
13225         for the returned bound_scalar.
13226         (vect_gen_vector_loop_niters): Handle polynomial vf.
13227         (vect_do_peeling): Likewise.  Update call to
13228         vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
13229         (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
13230         be constant.
13231         * tree-vect-loop.c (vect_determine_vectorization_factor)
13232         (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
13233         (vect_get_known_peeling_cost): Likewise.
13234         (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
13235         (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
13236         (vect_transform_loop): Likewise.  Use the lowest possible VF when
13237         updating the upper bounds of the loop.
13238         (vect_min_worthwhile_factor): Make static.  Return an unsigned int
13239         rather than an int.
13240         * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
13241         polynomial unroll factors.
13242         (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
13243         (vect_make_slp_decision): Likewise.
13244         (vect_supported_load_permutation_p): Likewise, and polynomial
13245         vf too.
13246         (vect_analyze_slp_cost): Handle polynomial vf.
13247         (vect_slp_analyze_node_operations): Likewise.
13248         (vect_slp_analyze_bb_1): Likewise.
13249         (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
13250         than an unsigned HOST_WIDE_INT.
13251         * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
13252         (vectorizable_load): Handle polynomial vf.
13253         * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
13254         a poly_uint64.
13255         (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
13256
13257 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
13258             Alan Hayward  <alan.hayward@arm.com>
13259             David Sherwood  <david.sherwood@arm.com>
13260
13261         * match.pd: Handle bit operations involving three constants
13262         and try to fold one pair.
13263
13264 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
13265
13266         * tree-vect-loop-manip.c: Include gimple-fold.h.
13267         (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
13268         niters_maybe_zero parameters.  Handle other cases besides a step of 1.
13269         (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
13270         Add a path that uses a step of VF instead of 1, but disable it
13271         for now.
13272         (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
13273         and niters_no_overflow parameters.  Update calls to
13274         slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
13275         Create a new SSA name if the latter choses to use a ste other
13276         than zero, and return it via niters_vector_mult_vf_var.
13277         * tree-vect-loop.c (vect_transform_loop): Update calls to
13278         vect_do_peeling, vect_gen_vector_loop_niters and
13279         slpeel_make_loop_iterate_ntimes.
13280         * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
13281         (vect_gen_vector_loop_niters): Update declarations after above changes.
13282
13283 2018-01-02  Michael Meissner  <meissner@linux.vnet.ibm.com>
13284
13285         * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
13286         128-bit round to integer instructions.
13287         (ceil<mode>2): Likewise.
13288         (btrunc<mode>2): Likewise.
13289         (round<mode>2): Likewise.
13290
13291 2018-01-02  Aaron Sawdey  <acsawdey@linux.vnet.ibm.com>
13292
13293         * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
13294         unaligned VSX load/store on P8/P9.
13295         (expand_block_clear): Allow the use of unaligned VSX
13296         load/store on P8/P9.
13297
13298 2018-01-02  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
13299
13300         * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
13301         New function.
13302         (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
13303         swap associated with both a load and a store.
13304
13305 2018-01-02  Andrew Waterman  <andrew@sifive.com>
13306
13307         * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
13308         * config/riscv/riscv.md (clear_cache): Use it.
13309
13310 2018-01-02  Artyom Skrobov  <tyomitch@gmail.com>
13311
13312         * web.c: Remove out-of-date comment.
13313
13314 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
13315
13316         * expr.c (fixup_args_size_notes): Check that any existing
13317         REG_ARGS_SIZE notes are correct, and don't try to re-add them.
13318         (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
13319         (emit_single_push_insn): ...here.
13320
13321 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
13322
13323         * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
13324         (const_vector_encoded_nelts): New function.
13325         (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
13326         (const_vector_int_elt, const_vector_elt): Declare.
13327         * emit-rtl.c (const_vector_int_elt_1): New function.
13328         (const_vector_elt): Likewise.
13329         * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
13330         of CONST_VECTOR_ELT.
13331
13332 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
13333
13334         * expr.c: Include rtx-vector-builder.h.
13335         (const_vector_mask_from_tree): Use rtx_vector_builder and operate
13336         directly on the tree encoding.
13337         (const_vector_from_tree): Likewise.
13338         * optabs.c: Include rtx-vector-builder.h.
13339         (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
13340         sequence of "u" values.
13341         * vec-perm-indices.c: Include rtx-vector-builder.h.
13342         (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
13343         directly on the vec_perm_indices encoding.
13344
13345 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
13346
13347         * doc/rtl.texi (const_vector): Describe new encoding scheme.
13348         * Makefile.in (OBJS): Add rtx-vector-builder.o.
13349         * rtx-vector-builder.h: New file.
13350         * rtx-vector-builder.c: Likewise.
13351         * rtl.h (rtx_def::u2): Add a const_vector field.
13352         (CONST_VECTOR_NPATTERNS): New macro.
13353         (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
13354         (CONST_VECTOR_DUPLICATE_P): Likewise.
13355         (CONST_VECTOR_STEPPED_P): Likewise.
13356         (CONST_VECTOR_ENCODED_ELT): Likewise.
13357         (const_vec_duplicate_p): Check for a duplicated vector encoding.
13358         (unwrap_const_vec_duplicate): Likewise.
13359         (const_vec_series_p): Check for a non-duplicated vector encoding.
13360         Say that the function only returns true for integer vectors.
13361         * emit-rtl.c: Include rtx-vector-builder.h.
13362         (gen_const_vec_duplicate_1): Delete.
13363         (gen_const_vector): Call gen_const_vec_duplicate instead of
13364         gen_const_vec_duplicate_1.
13365         (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
13366         (gen_const_vec_duplicate): Use rtx_vector_builder.
13367         (gen_const_vec_series): Likewise.
13368         (gen_rtx_CONST_VECTOR): Likewise.
13369         * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
13370         (swap_const_vector_halves): Take an rtx pointer rather than rtx.
13371         Build a new vector rather than modifying a CONST_VECTOR in-place.
13372         (handle_special_swappables): Update call accordingly.
13373         * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
13374         (swap_const_vector_halves): Take an rtx pointer rather than rtx.
13375         Build a new vector rather than modifying a CONST_VECTOR in-place.
13376         (handle_special_swappables): Update call accordingly.
13377
13378 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
13379
13380         * simplify-rtx.c (simplify_const_binary_operation): Use
13381         CONST_VECTOR_ELT instead of XVECEXP.
13382
13383 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
13384
13385         * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
13386         the selector elements to be different from the data elements
13387         if the selector is a VECTOR_CST.
13388         * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
13389         ssizetype for the selector.
13390
13391 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
13392
13393         * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
13394         before testing each element individually.
13395         * tree-vect-generic.c (lower_vec_perm): Likewise.
13396
13397 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
13398
13399         * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
13400         * selftest-run-tests.c (selftest::run_tests): Call it.
13401         * vector-builder.h (vector_builder::operator ==): New function.
13402         (vector_builder::operator !=): Likewise.
13403         * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
13404         (vec_perm_indices::all_from_input_p): New function.
13405         * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
13406         (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
13407         * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
13408         instead of reading the VECTOR_CST directly.  Detect whether both
13409         vector inputs are the same before constructing the vec_perm_indices,
13410         and update the number of inputs argument accordingly.  Use the
13411         utility functions added above.  Only construct sel2 if we need to.
13412
13413 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
13414
13415         * optabs.c (expand_vec_perm_var): Use an explicit encoding for
13416         the broadcast of the low byte.
13417         (expand_mult_highpart): Use an explicit encoding for the permutes.
13418         * optabs-query.c (can_mult_highpart_p): Likewise.
13419         * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
13420         * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
13421         (vectorizable_bswap): Likewise.
13422         * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
13423         explicit encoding for the power-of-2 permutes.
13424         (vect_permute_store_chain): Likewise.
13425         (vect_grouped_load_supported): Likewise.
13426         (vect_permute_load_chain): Likewise.
13427
13428 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
13429
13430         * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
13431         * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
13432         * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
13433         * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
13434         * tree-vect-stmts.c (vectorizable_bswap): Likewise.
13435         (vect_gen_perm_mask_any): Likewise.
13436
13437 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
13438
13439         * int-vector-builder.h: New file.
13440         * vec-perm-indices.h: Include int-vector-builder.h.
13441         (vec_perm_indices): Redefine as an int_vector_builder.
13442         (auto_vec_perm_indices): Delete.
13443         (vec_perm_builder): Redefine as a stand-alone class.
13444         (vec_perm_indices::vec_perm_indices): New function.
13445         (vec_perm_indices::clamp): Likewise.
13446         * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
13447         (vec_perm_indices::new_vector): New function.
13448         (vec_perm_indices::new_expanded_vector): Update for new
13449         vec_perm_indices class.
13450         (vec_perm_indices::rotate_inputs): New function.
13451         (vec_perm_indices::all_in_range_p): Operate directly on the
13452         encoded form, without computing elided elements.
13453         (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
13454         encoding.  Update for new vec_perm_indices class.
13455         * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
13456         the given vec_perm_builder.
13457         (expand_vec_perm_var): Update vec_perm_builder constructor.
13458         (expand_mult_highpart): Use vec_perm_builder instead of
13459         auto_vec_perm_indices.
13460         * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
13461         vec_perm_indices instead of auto_vec_perm_indices.  Use a single
13462         or double series encoding as appropriate.
13463         * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
13464         vec_perm_indices instead of auto_vec_perm_indices.
13465         * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
13466         * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
13467         (vect_permute_store_chain): Likewise.
13468         (vect_grouped_load_supported): Likewise.
13469         (vect_permute_load_chain): Likewise.
13470         (vect_shift_permute_load_chain): Likewise.
13471         * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
13472         (vect_transform_slp_perm_load): Likewise.
13473         (vect_schedule_slp_instance): Likewise.
13474         * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
13475         (vectorizable_mask_load_store): Likewise.
13476         (vectorizable_bswap): Likewise.
13477         (vectorizable_store): Likewise.
13478         (vectorizable_load): Likewise.
13479         * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
13480         vec_perm_indices instead of auto_vec_perm_indices.  Use
13481         tree_to_vec_perm_builder to read the vector from a tree.
13482         * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
13483         vec_perm_builder instead of a vec_perm_indices.
13484         (have_whole_vector_shift): Use vec_perm_builder and
13485         vec_perm_indices instead of auto_vec_perm_indices.  Leave the
13486         truncation to calc_vec_perm_mask_for_shift.
13487         (vect_create_epilog_for_reduction): Likewise.
13488         * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
13489         from auto_vec_perm_indices to vec_perm_indices.
13490         (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
13491         instead of changing individual elements.
13492         (aarch64_vectorize_vec_perm_const): Use new_vector to install
13493         the vector in d.perm.
13494         * config/arm/arm.c (expand_vec_perm_d::perm): Change
13495         from auto_vec_perm_indices to vec_perm_indices.
13496         (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
13497         instead of changing individual elements.
13498         (arm_vectorize_vec_perm_const): Use new_vector to install
13499         the vector in d.perm.
13500         * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
13501         Update vec_perm_builder constructor.
13502         (rs6000_expand_interleave): Likewise.
13503         * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
13504         (rs6000_expand_interleave): Likewise.
13505
13506 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
13507
13508         * optabs-query.c (can_vec_perm_var_p): Check whether lowering
13509         to qimode could truncate the indices.
13510         * optabs.c (expand_vec_perm_var): Likewise.
13511
13512 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
13513
13514         * Makefile.in (OBJS): Add vec-perm-indices.o.
13515         * vec-perm-indices.h: New file.
13516         * vec-perm-indices.c: Likewise.
13517         * target.h (vec_perm_indices): Replace with a forward class
13518         declaration.
13519         (auto_vec_perm_indices): Move to vec-perm-indices.h.
13520         * optabs.h: Include vec-perm-indices.h.
13521         (expand_vec_perm): Delete.
13522         (selector_fits_mode_p, expand_vec_perm_var): Declare.
13523         (expand_vec_perm_const): Declare.
13524         * target.def (vec_perm_const_ok): Replace with...
13525         (vec_perm_const): ...this new hook.
13526         * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
13527         (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
13528         * doc/tm.texi: Regenerate.
13529         * optabs.def (vec_perm_const): Delete.
13530         * doc/md.texi (vec_perm_const): Likewise.
13531         (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
13532         * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
13533         expand_vec_perm for constant permutation vectors.  Assert that
13534         the mode of variable permutation vectors is the integer equivalent
13535         of the mode that is being permuted.
13536         * optabs-query.h (selector_fits_mode_p): Declare.
13537         * optabs-query.c: Include vec-perm-indices.h.
13538         (selector_fits_mode_p): New function.
13539         (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
13540         is defined, instead of checking whether the vec_perm_const_optab
13541         exists.  Use targetm.vectorize.vec_perm_const instead of
13542         targetm.vectorize.vec_perm_const_ok.  Check whether the indices
13543         fit in the vector mode before using a variable permute.
13544         * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
13545         vec_perm_indices instead of an rtx.
13546         (expand_vec_perm): Replace with...
13547         (expand_vec_perm_const): ...this new function.  Take the selector
13548         as a vec_perm_indices rather than an rtx.  Also take the mode of
13549         the selector.  Update call to shift_amt_for_vec_perm_mask.
13550         Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
13551         Use vec_perm_indices::new_expanded_vector to expand the original
13552         selector into bytes.  Check whether the indices fit in the vector
13553         mode before using a variable permute.
13554         (expand_vec_perm_var): Make global.
13555         (expand_mult_highpart): Use expand_vec_perm_const.
13556         * fold-const.c: Includes vec-perm-indices.h.
13557         * tree-ssa-forwprop.c: Likewise.
13558         * tree-vect-data-refs.c: Likewise.
13559         * tree-vect-generic.c: Likewise.
13560         * tree-vect-loop.c: Likewise.
13561         * tree-vect-slp.c: Likewise.
13562         * tree-vect-stmts.c: Likewise.
13563         * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
13564         Delete.
13565         * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
13566         * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
13567         (aarch64_vectorize_vec_perm_const_ok): Fuse into...
13568         (aarch64_vectorize_vec_perm_const): ...this new function.
13569         (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
13570         (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
13571         * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
13572         * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
13573         * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
13574         (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
13575         (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
13576         into...
13577         (arm_vectorize_vec_perm_const): ...this new function.  Explicitly
13578         check for NEON modes.
13579         * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
13580         * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
13581         * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
13582         (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
13583         into...
13584         (ix86_vectorize_vec_perm_const): ...this new function.  Incorporate
13585         the old VEC_PERM_CONST conditions.
13586         * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
13587         * config/ia64/vect.md (vec_perm_const<mode>): Delete.
13588         * config/ia64/ia64.c (ia64_expand_vec_perm_const)
13589         (ia64_vectorize_vec_perm_const_ok): Merge into...
13590         (ia64_vectorize_vec_perm_const): ...this new function.
13591         * config/mips/loongson.md (vec_perm_const<mode>): Delete.
13592         * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
13593         * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
13594         * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
13595         * config/mips/mips.c (mips_expand_vec_perm_const)
13596         (mips_vectorize_vec_perm_const_ok): Merge into...
13597         (mips_vectorize_vec_perm_const): ...this new function.
13598         * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
13599         * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
13600         * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
13601         * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
13602         * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
13603         (rs6000_expand_vec_perm_const): Delete.
13604         * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
13605         Delete.
13606         (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
13607         (altivec_expand_vec_perm_const_le): Take each operand individually.
13608         Operate on constant selectors rather than rtxes.
13609         (altivec_expand_vec_perm_const): Likewise.  Update call to
13610         altivec_expand_vec_perm_const_le.
13611         (rs6000_expand_vec_perm_const): Delete.
13612         (rs6000_vectorize_vec_perm_const_ok): Delete.
13613         (rs6000_vectorize_vec_perm_const): New function.
13614         (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
13615         an element count and rtx array.
13616         (rs6000_expand_extract_even): Update call accordingly.
13617         (rs6000_expand_interleave): Likewise.
13618         * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
13619         * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
13620         * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
13621         * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
13622         (rs6000_expand_vec_perm_const): Delete.
13623         * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
13624         (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
13625         (altivec_expand_vec_perm_const_le): Take each operand individually.
13626         Operate on constant selectors rather than rtxes.
13627         (altivec_expand_vec_perm_const): Likewise.  Update call to
13628         altivec_expand_vec_perm_const_le.
13629         (rs6000_expand_vec_perm_const): Delete.
13630         (rs6000_vectorize_vec_perm_const_ok): Delete.
13631         (rs6000_vectorize_vec_perm_const): New function.  Remove stray
13632         reference to the SPE evmerge intructions.
13633         (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
13634         an element count and rtx array.
13635         (rs6000_expand_extract_even): Update call accordingly.
13636         (rs6000_expand_interleave): Likewise.
13637         * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
13638         * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
13639         new function.
13640         (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
13641
13642 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
13643
13644         * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
13645         vector mode and that that mode matches the mode of the data
13646         being permuted.
13647         (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
13648         out into expand_vec_perm_var.  Do all CONST_VECTOR handling here,
13649         directly using expand_vec_perm_1 when forcing selectors into
13650         registers.
13651         (expand_vec_perm_var): New function, split out from expand_vec_perm.
13652
13653 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
13654
13655         * optabs-query.h (can_vec_perm_p): Delete.
13656         (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
13657         * optabs-query.c (can_vec_perm_p): Split into...
13658         (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
13659         (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
13660         particular selector is valid.
13661         * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
13662         * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
13663         (vect_grouped_load_supported): Likewise.
13664         (vect_shift_permute_load_chain): Likewise.
13665         * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
13666         (vect_transform_slp_perm_load): Likewise.
13667         * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
13668         (vectorizable_bswap): Likewise.
13669         (vect_gen_perm_mask_checked): Likewise.
13670         * fold-const.c (fold_ternary_loc): Likewise.  Don't take
13671         implementations of variable permutation vectors into account
13672         when deciding which selector to use.
13673         * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
13674         vec_perm_const_optab is supported; instead use can_vec_perm_const_p
13675         with a false third argument.
13676         * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
13677         to test whether the constant selector is valid and can_vec_perm_var_p
13678         to test whether a variable selector is valid.
13679
13680 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
13681
13682         * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
13683         * optabs-query.c (can_vec_perm_p): Likewise.
13684         * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
13685         instead of vec_perm_indices.
13686         * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
13687         (vect_gen_perm_mask_checked): Likewise,
13688         * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
13689         (vect_gen_perm_mask_checked): Likewise,
13690
13691 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
13692
13693         * optabs-query.h (qimode_for_vec_perm): Declare.
13694         * optabs-query.c (can_vec_perm_p): Split out qimode search to...
13695         (qimode_for_vec_perm): ...this new function.
13696         * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
13697
13698 2018-01-02  Aaron Sawdey  <acsawdey@linux.vnet.ibm.com>
13699
13700         * rtlanal.c (canonicalize_condition): Return 0 if final rtx
13701         does not have a conditional at the top.
13702
13703 2018-01-02  Richard Biener  <rguenther@suse.de>
13704
13705         * ipa-inline.c (big_speedup_p): Fix expression.
13706
13707 2018-01-02  Jan Hubicka  <hubicka@ucw.cz>
13708
13709         PR target/81616
13710         * config/i386/x86-tune-costs.h: Increase cost of integer load costs
13711         for generic 4->6.
13712
13713 2018-01-02  Jan Hubicka  <hubicka@ucw.cz>
13714
13715         PR target/81616
13716         Generic tuning.
13717         * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
13718         cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
13719         and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
13720         cond_taken_branch_cost 3->4.
13721
13722 2018-01-01  Jakub Jelinek  <jakub@redhat.com>
13723
13724         PR tree-optimization/83581
13725         * tree-loop-distribution.c (pass_loop_distribution::execute): Return
13726         TODO_cleanup_cfg if any changes have been made.
13727
13728         PR middle-end/83608
13729         * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
13730         convert_modes if target mode has the right side, but different mode
13731         class.
13732
13733         PR middle-end/83609
13734         * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
13735         last argument when extracting from CONCAT.  If either from_real or
13736         from_imag is NULL, use expansion through memory.  If result is not
13737         a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
13738         the parts directly to inner mode, if even that fails, use expansion
13739         through memory.
13740
13741         PR middle-end/83623
13742         * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
13743         check for bswap in mode rather than HImode and use that in expand_unop
13744         too.
13745 \f
13746 Copyright (C) 2018 Free Software Foundation, Inc.
13747
13748 Copying and distribution of this file, with or without modification,
13749 are permitted in any medium without royalty provided the copyright
13750 notice and this notice are preserved.