gcc/ChangeLog

   1 2018-04-25  Richard Biener  <rguenther@suse.de>
   2
   3         * lto-streamer.h (LTO_major_version): Bump to 8.
   4
   5 2018-04-25  Jakub Jelinek  <jakub@redhat.com>
   6
   7         * BASE-VER: Set to 9.0.0.
   8
   9 2018-04-24  Segher Boessenkool  <segher@kernel.crashing.org>
  10
  11         * config/rs6000/rs6000.c (init_float128_ieee): Fix spelling mistakes
  12         in __abskf2 and __powikf2.
  13
  14 2018-04-24  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
  15
  16         PR target/85512
  17         * config/aarch64/constraints.md (Usg, Usj): New constraints.
  18         * config/aarch64/iterators.md (cmode_simd): New mode attribute.
  19         * config/aarch64/aarch64.md (*aarch64_ashr_sisd_or_int_<mode>3):
  20         Use the above on operand 2.  Reindent.
  21         (*aarch64_lshr_sisd_or_int_<mode>3): Likewise.
  22
  23 2018-04-24  H.J. Lu  <hongjiu.lu@intel.com>
  24
  25         PR target/85485
  26         * common/config/i386/i386-common.c (ix86_handle_option): Don't
  27         handle OPT_mcet.
  28         * config/i386/i386.opt (mcet): Removed.
  29         * doc/install.texi: Remove -mcet documentation.
  30         * doc/invoke.texi: Likewise.
  31
  32 2018-04-24  H.J. Lu  <hongjiu.lu@intel.com>
  33
  34         PR target/85485
  35         * doc/install.texi: Remove -mcet from bootstrap-cet.
  36
  37 2018-04-24  Jakub Jelinek  <jakub@redhat.com>
  38
  39         PR target/85511
  40         * config/i386/i386.c (ix86_init_mmx_sse_builtins): Don't define
  41         __builtin_ia32_readeflags_u32 and __builtin_ia32_writeeflags_u32
  42         if TARGET_64BIT.
  43
  44         PR target/85503
  45         * config/rs6000/rs6000-p8swap.c (const_load_sequence_p): Punt if
  46         const_vector is not CONST_VECTOR or SYMBOL_REF for a constant pool
  47         containing a CONST_VECTOR.
  48
  49 2018-04-24  Cesar Philippidis  <cesar@codesourcery.com>
  50
  51         * doc/install.texi: Update newlib dependency for nvptx.
  52
  53 2018-04-24  Jakub Jelinek  <jakub@redhat.com>
  54
  55         PR target/85508
  56         * config/i386/i386.c (ix86_expand_vector_init_one_var): Use UINTVAL
  57         instead of INTVAL when shifting x left.
  58
  59 2018-04-24  Andreas Krebbel  <krebbel@linux.ibm.com>
  60
  61         PR tree-optimization/85478
  62         * tree-vect-loop.c (vect_analyze_loop_2): Do not call
  63         vect_grouped_store_supported for single element vectors.
  64
  65 2018-04-24  Richard Biener  <rguenther@suse.de>
  66
  67         PR target/85491
  68         * config/i386/i386.c (ix86_add_stmt_cost): Restrict strided
  69         load cost increase to the case of non-constant step.
  70
  71 2018-04-24  Jakub Jelinek  <jakub@redhat.com>
  72
  73         PR target/84828
  74         * reg-stack.c (move_for_stack_reg): Don't fail assertion about dead
  75         destination if any_malformed_asm.
  76
  77 2018-04-23  Eric Botcazou  <ebotcazou@adacore.com>
  78
  79         PR middle-end/85496
  80         * expr.c (store_field): In the bitfield case, if the value comes from
  81         a function call and is returned in registers by means of a PARALLEL,
  82         do not change the mode of the temporary unless BLKmode and VOIDmode.
  83
  84 2018-04-23  Andrey Belevantsev  <abel@ispras.ru>
  85
  86         PR rtl-optimization/85423
  87         * sel-sched-ir.c (has_dependence_note_mem_dep): Only discard
  88         dependencies to debug insns when the previous insn is non-debug.
  89
  90 2018-04-23  Claudiu Zissulescu  <claziss@synopsys.com>
  91
  92         * config/arc/arc.md ("vunspec"): Delete it, unify all the unspec
  93         enums into a single definition.
  94         (fls): Fix predicates and printing.
  95         (seti): Likewise.
  96
  97 2018-04-23  Claudiu Zissulescu  <claziss@synopsys.com>
  98
  99         * config/arc/arc-protos.h (check_if_valid_sleep_operand): Remove.
 100         * config/arc/arc.c (arc_expand_builtin): Sleep accepts registers
 101         and short u6 immediate.
 102         (check_if_valid_sleep_operand): Remove.
 103         * config/arc/arc.md (Sleep): Accepts registers and u6 immediates.
 104
 105 2018-04-22  Chung-Ju Wu  <jasonwucj@gmail.com>
 106
 107         * config/nds32/nds32.c (nds32_compute_stack_frame): Consider
 108         flag_always_save_lp condition.
 109         * config/nds32/nds32.opt (malways-save-lp): New option.
 110
 111 2018-04-22  Shiva Chen  <shiva0217@gmail.com>
 112
 113         * config/nds32/nds32-protos.h (nds32_use_load_post_increment): Declare.
 114         * config/nds32/nds32.c (nds32_use_load_post_increment): New.
 115         * config/nds32/nds32.h
 116         (USE_LOAD_POST_INCREMENT, USE_LOAD_POST_DECREMENT): Define.
 117         (USE_STORE_POST_INCREMENT, USE_STORE_POST_DECREMENT): Define.
 118
 119 2018-04-22  Shiva Chen  <shiva0217@gmail.com>
 120
 121         * config/nds32/nds32-protos.h (nds32_ls_333_p): Remove.
 122         * config/nds32/nds32.c (nds32_ls_333_p): Remove.
 123
 124 2018-04-22  Shiva Chen  <shiva0217@gmail.com>
 125             Chung-Ju Wu  <jasonwucj@gmail.com>
 126
 127         * config/nds32/nds32-protos.h (nds32_case_vector_shorten_mode):
 128         Declare.
 129         * config/nds32/nds32.c (nds32_case_vector_shorten_mode): New function.
 130         * config/nds32/nds32.h (CASE_VECTOR_SHORTEN_MODE): Modify.
 131
 132 2018-04-22  Chung-Ju Wu  <jasonwucj@gmail.com>
 133
 134         * config/nds32/nds32.c (nds32_compute_stack_frame): Fix wrong value.
 135
 136 2018-04-22  Chung-Ju Wu  <jasonwucj@gmail.com>
 137
 138         * config/nds32/nds32-protos.h (nds32_data_alignment,
 139         nds32_local_alignment): Declare.
 140         * config/nds32/nds32.c (nds32_data_alignment, nds32_constant_alignment,
 141         nds32_local_alignment): New functions.
 142         (TARGET_CONSTANT_ALIGNMENT): Define.
 143         * config/nds32/nds32.h (DATA_ALIGNMENT, LOCAL_ALIGNMENT): Define.
 144
 145 2018-04-22  Chung-Ju Wu  <jasonwucj@gmail.com>
 146
 147         * config/nds32/nds32.c
 148         (TARGET_HARD_REGNO_MODE_OK): Move to the bottom of file.
 149         (TARGET_MODES_TIEABLE_P): Likewise.
 150
 151 2018-04-22  Chung-Ju Wu  <jasonwucj@gmail.com>
 152
 153         * config/nds32/nds32.c (nds32_asm_file_start): Display optimization
 154         level Ofast and Og.
 155
 156 2018-04-22  Monk Chiang  <sh.chiang04@gmail.com>
 157             Chung-Ju Wu  <jasonwucj@gmail.com>
 158
 159         * config/nds32/constants.md (unspec_volatile_element): Add enum values
 160         for unaligned access.
 161         * config/nds32/nds32-intrinsic.c: Implementation of expanding
 162         unaligned access.
 163         * config/nds32/nds32-intrinsic.md: Likewise.
 164         * config/nds32/nds32_intrinsic.h: Likewise.
 165         * config/nds32/nds32.h (nds32_builtins): Likewise.
 166         * config/nds32/nds32.opt (munaligned-access): New option.
 167         * config/nds32/nds32.c (nds32_asm_file_start): Display
 168         flag_unaligned_access status.
 169
 170 2018-04-20  Kito Cheng  <kito.cheng@gmail.com>
 171
 172         * config/riscv/elf.h (LINK_SPEC): Pass --no-relax if
 173         -mno-relax is present.
 174         * config/riscv/linux.h (LINK_SPEC): Ditto.
 175
 176 2018-04-20  Martin Sebor  <msebor@redhat.com>
 177
 178         PR c/85365
 179         * gimple-fold.c (gimple_fold_builtin_strcpy): Suppress -Wrestrict
 180         for null pointers.
 181         (gimple_fold_builtin_stxcpy_chk): Same.
 182         * gimple-ssa-warn-restrict.c (check_bounds_or_overlap): Same.
 183
 184 2018-04-20  Michael Meissner  <meissner@linux.ibm.com>
 185
 186         PR target/85456
 187         * config/rs6000/rs6000.c (init_float128_ieee): Add support to call
 188         __powikf2 when long double is IEEE 128-bit.
 189
 190 2018-04-20  Kito Cheng  <kito.cheng@gmail.com>
 191
 192         * config/riscv/riscv.c (riscv_first_stack_step): Round up min
 193         step to make sure stack always aligned.
 194
 195 2018-04-20  Carl Love  <cel@us.ibm.com>
 196
 197         PR target/83402
 198         * config/rs6000/rs6000-c.c (rs6000_gimple_fold_builtin): Add
 199         size check for arg0.
 200
 201 2018-04-20  Nathan Sidwell  <nathan@codesourcery.com>
 202             Tom de Vries  <tom@codesourcery.com>
 203
 204         PR target/85445
 205         * config/nvptx/nvptx.c (nvptx_emit_forking, nvptx_emit_joining):
 206         Emit insns for calls too.
 207         (nvptx_find_par): Always look for worker-level predecessor insn.
 208         (nvptx_propagate): Add is_call parm, return bool.  Copy frame for
 209         calls.
 210         (nvptx_vpropagate, nvptx_wpropagate): Adjust.
 211         (nvptx_process_pars): Propagate frames for calls.
 212
 213 2018-04-20  H.J. Lu  <hongjiu.lu@intel.com>
 214
 215         PR target/85469
 216         * common/config/i386/i386-common.c (OPTION_MASK_ISA_IBT_SET):
 217         Removed.
 218         (OPTION_MASK_ISA_IBT_UNSET): Likewise.
 219         (ix86_handle_option): Don't handle OPT_mibt.
 220         * config/i386/cet.h: Check __CET__ instead of __IBT__ and
 221         __SHSTK__.
 222         * config/i386/driver-i386.c (host_detect_local_cpu): Remove
 223         has_ibt and ibt.
 224         * config/i386/i386-c.c (ix86_target_macros_internal): Don't
 225         check OPTION_MASK_ISA_IBT nor flag_cf_protection.
 226         (ix86_target_macros): Define __CET__ with flag_cf_protection
 227         for -fcf-protection.
 228         * config/i386/i386.c (isa2_opts): Remove -mibt.
 229         * config/i386/i386.h (TARGET_IBT): Removed.
 230         (TARGET_IBT_P): Likewise.
 231         (ix86_valid_target_attribute_inner_p): Don't check OPT_mibt.
 232         * config/i386/i386.md (nop_endbr): Don't check TARGET_IBT.
 233         * config/i386/i386.opt (mcet): Update help message.
 234         (mshstk): Likewise.
 235         (mibt): Removed.
 236         * doc/invoke.texi: Remove -mibt.  Document __CET__.  Document
 237         -mcet as an alias for -mshstk.
 238
 239 2018-04-20  Richard Biener <rguenther@suse.de>
 240
 241         PR middle-end/85475
 242         * match.pd ((X * CST) * Y -> (X * Y) * CST): Avoid exponential
 243         complexity by forcing a single use of the multiply operand.
 244
 245 2018-04-20  Martin Jambor  <mjambor@suse.cz>
 246
 247         ipa/85449
 248         * ipa-cp.c (cgraph_edge_brings_value_p): Move check for self-feeding
 249         recursion dependency to only apply to non-clones.
 250
 251 2018-04-20  Martin Jambor  <mjambor@suse.cz>
 252
 253         ipa/85447
 254         * ipa-cp.c (create_specialized_node): Check that clones of
 255         self-recursive edges exist during IPA-CP.
 256
 257 2018-04-19  Toon Moene  <toon@moene.org>
 258
 259         * doc/invoke.texi: Add -floop-unroll-and-jam to options enabled
 260         by -O3.
 261
 262 2018-04-19  Jakub Jelinek  <jakub@redhat.com>
 263
 264         PR tree-optimization/85467
 265         * fold-const.c (fold_ternary_loc) <case BIT_FIELD_REF>: Use
 266         VECTOR_TYPE_P macro.  If type is vector type, VIEW_CONVERT_EXPR the
 267         VECTOR_CST element to type.
 268
 269 2018-04-19  H.J. Lu  <hongjiu.lu@intel.com>
 270
 271         PR target/85397
 272         * config/i386/i386.h (STACK_SAVEAREA_MODE): New.
 273         * config/i386/i386.md (builtin_setjmp_setup): Removed.
 274         (builtin_longjmp): Likewise.
 275         (save_stack_nonlocal): New pattern.
 276         (restore_stack_nonlocal): Likewise.
 277
 278 2018-04-19  H.J. Lu  <hongjiu.lu@intel.com>
 279
 280         PR target/85404
 281         * config/i386/cet.c (file_end_indicate_exec_stack_and_cet):
 282         Replace ASM_OUTPUT_LABEL with fprintf.
 283
 284 2018-04-19  H.J. Lu  <hongjiu.lu@intel.com>
 285
 286         PR target/85417
 287         * config/i386/cet.c (file_end_indicate_exec_stack_and_cet):
 288         Check flag_cf_protection instead of TARGET_IBT and TARGET_SHSTK.
 289         * config/i386/i386-c.c (ix86_target_macros_internal): Also
 290         define __IBT__ and __SHSTK__ for -fcf-protection.
 291         * config/i386/i386.c (pass_insert_endbranch::gate): Don't check
 292         TARGET_IBT.
 293         (ix86_trampoline_init): Likewise.
 294         (x86_output_mi_thunk): Likewise.
 295         (ix86_notrack_prefixed_insn_p): Likewise.
 296         (ix86_option_override_internal): Don't disallow -fcf-protection.
 297         * config/i386/i386.md (rdssp<mode>): Also enable for
 298         -fcf-protection.
 299         (incssp<mode>): Likewise.
 300         (nop_endbr): Likewise.
 301         * config/i386/i386.opt (mcet): Change help message to built-in
 302         functions only.
 303         (mibt): Likewise.
 304         (mshstk): Likewise.
 305         * doc/invoke.texi: Remove -mcet, -mibt and -mshstk condition
 306         on -fcf-protection.  Change -mcet, -mibt and -mshstk to only
 307         enable CET built-in functions.
 308
 309 2018-04-19  Sebastian Peryt  <sebastian.peryt@intel.com>
 310
 311         * common/config/i386/i386-common.c
 312         (OPTION_MASK_ISA_MOVDIRI_SET, OPTION_MASK_ISA_MOVDIR64B_SET,
 313         OPTION_MASK_ISA_MOVDIRI_UNSET,
 314         OPTION_MASK_ISA_MOVDIR64B_UNSET): New defines.
 315         (ix86_handle_option): Handle -mmovdiri and -mmovdir64b.
 316         * config.gcc (movdirintrin.h): New header.
 317         * config/i386/cpuid.h (bit_MOVDIRI,
 318         bit_MOVDIR64B): New bits.
 319         * config/i386/driver-i386.c (host_detect_local_cpu): Detect -mmovdiri
 320         and -mmvodir64b.
 321         * config/i386/i386-builtin-types.def ((VOID, PUNSIGNED, UNSIGNED),
 322         (VOID, PVOID, PCVOID)): New function types.
 323         * config/i386/i386-builtin.def (__builtin_ia32_directstoreu_u32,
 324         __builtin_ia32_directstoreu_u64,
 325          __builtin_ia32_movdir64b): New builtins.
 326         * config/i386/i386-c.c (__MOVDIRI__, __MOVDIR64B__): New.
 327         * config/i386/i386.c (ix86_target_string): Added -mmovdir64b
 328         and -mmovdiri.
 329         (ix86_valid_target_attribute_inner_p): Ditto.
 330         (ix86_expand_special_args_builtin): Added VOID_FTYPE_PUNSIGNED_UNSIGNED
 331         and VOID_FTYPE_PUNSIGNED_UNSIGNED.
 332         (ix86_expand_builtin): Expand IX86_BUILTIN_MOVDIR64B.
 333         * config/i386/i386.h (TARGET_MOVDIRI, TARGET_MOVDIRI_P,
 334         TARGET_MOVDIR64B, TARGET_MOVDIR64B_P): New.
 335         * config/i386/i386.md (UNSPECV_MOVDIRI, UNSPECV_MOVDIR64B): New.
 336         (movdiri<mode>, movdir64b_<mode>): New.
 337         * config/i386/i386.opt: Add -mmovdiri and -mmovdir64b.
 338         * config/i386/immintrin.h: Include movdirintrin.h.
 339         * config/i386/movdirintrin.h: New file.
 340         * doc/invoke.texi: Added -mmovdiri and -mmovdir64b.
 341
 342 2018-04-19  Richard Biener  <rguenther@suse.de>
 343
 344         PR middle-end/85455
 345         * cfg.c (clear_bb_flags): When loop state says we have
 346         marked irreducible regions also preserve BB_IRREDUCIBLE_LOOP.
 347
 348 2018-04-19  Richard Biener  <rguenther@suse.de>
 349
 350         PR tree-optimization/84737
 351         * tree-vect-data-refs.c (vect_copy_ref_info): New function
 352         copying restrict info.
 353         (vect_setup_realignment): Use it.
 354         * tree-vectorizer.h (vect_copy_ref_info): Declare.
 355         * tree-vect-stmts.c (vectorizable_store): Copy ref info from
 356         the first DR to all generated stores.
 357         (vectorizable_load): Likewise for loads.
 358
 359 2018-04-19  Jakub Jelinek  <jakub@redhat.com>
 360
 361         PR tree-optimization/85446
 362         * match.pd ((intptr_t) x eq/ne CST to x eq/ne (typeof x) cst): Require
 363         the integral and pointer types to have the same precision.
 364
 365         * doc/install.texi: Document --disable-cet being the default and
 366         --enable-cet=auto.
 367
 368 2018-04-18  Martin Liska  <mliska@suse.cz>
 369
 370         * ipa-devirt.c (odr_subtypes_equivalent_p): Fix GNU coding
 371         style.
 372
 373 2018-04-18  Martin Liska  <mliska@suse.cz>
 374
 375         Revert
 376         2018-03-02  Eric Botcazou  <ebotcazou@adacore.com>
 377
 378         PR ipa/83983
 379         * ipa-devirt.c (odr_subtypes_equivalent_p): Get the ODR type of both
 380         arguments if they are comparable.
 381
 382 2018-04-18  Martin Liska  <mliska@suse.cz>
 383
 384         Revert
 385         2018-03-13  Eric Botcazou  <ebotcazou@adacore.com>
 386
 387         PR lto/84805
 388         * ipa-devirt.c (odr_subtypes_equivalent_p): Do not get the ODR type of
 389         incomplete types.
 390
 391 2018-04-18  H.J. Lu  <hongjiu.lu@intel.com>
 392
 393         PR target/85388
 394         * config/i386/i386.c (ix86_expand_split_stack_prologue): Insert
 395         ENDBR after calling __morestack.
 396
 397 2018-04-18  David Malcolm  <dmalcolm@redhat.com>
 398
 399         PR jit/85384
 400         * configure.ac (gcc-driver-name.h): Honor --with-gcc-major-version
 401         by using gcc_base_ver to generate a gcc_driver_version, and use
 402         it when generating GCC_DRIVER_NAME.
 403         * configure: Regenerate.
 404
 405 2018-04-18  Jakub Jelinek  <jakub@redhat.com>
 406
 407         PR target/81084
 408         * config.gcc: Obsolete powerpc*-*-*spe*.
 409
 410 2018-04-17  Jakub Jelinek  <jakub@redhat.com>
 411
 412         PR debug/84637
 413         * dbxout.c (dbxout_int): Perform negation in unsigned int type.
 414         (stabstr_D): Change type of unum from unsigned int to
 415         unsigned HOST_WIDE_INT.  Perform negation in unsigned HOST_WIDE_INT
 416         type.
 417
 418 2018-04-17  Jim Wilson  <jimw@sifive.com>
 419
 420         PR 84856
 421         * config/riscv/riscv.c (riscv_compute_frame_info): Add calls to
 422         RISCV_STACK_ALIGN when using outgoing_args_size and pretend_args_size.
 423         Set arg_pointer_offset after using pretend_args_size.
 424
 425 2018-04-17  Jakub Jelinek  <jakub@redhat.com>
 426
 427         PR rtl-optimization/85431
 428         * dse.c (record_store): Ignore zero width stores.
 429
 430         PR sanitizer/85230
 431         * asan.c (handle_builtin_stack_restore): Adjust comment.  Emit
 432         __asan_allocas_unpoison call and last_alloca_addr = new_sp before
 433         __builtin_stack_restore rather than after it.
 434         * builtins.c (expand_asan_emit_allocas_unpoison): Pass
 435         arg1 + (virtual_dynamic_stack_rtx - stack_pointer_rtx) as second
 436         argument instead of virtual_dynamic_stack_rtx.
 437
 438 2018-04-17  Kelvin Nilsen  <kelvin@gcc.gnu.org>
 439
 440         * config/rs6000/rs6000-protos.h (rs6000_builtin_is_supported_p):
 441         New prototype.
 442         * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin):
 443         Add note to error message to explain internal mapping of overloaded
 444         built-in function name to non-overloaded built-in function name.
 445         * config/rs6000/rs6000.c (rs6000_builtin_is_supported_p): New
 446         function.
 447
 448 2018-04-17  Michael Meissner  <meissner@linux.vnet.ibm.com>
 449
 450         PR target/85424
 451         * config/rs6000/rs6000.md (pack<mode>): Do not try handle a pack
 452         where the inputs overlap with the output.
 453
 454 2018-04-17  Jakub Jelinek  <jakub@redhat.com>
 455
 456         * config/i386/sse.md (vec_extract_lo_<mode><mask_name>): Add
 457         (=v, v) alternative and explicit "memory" attribute.
 458         (vec_extract_lo_<mode><mask_name>): Likewise.  Also add
 459         "type", "prefix", "prefix_extra", "length_immediate" and "mode"
 460         attributes.
 461         (vec_extract_lo_<mode><mask_name>): Add (=v, v) alternative and use
 462         "sselog1" type instead of "sselog".
 463         (vec_extract_hi_<mode><mask_name>): Use "sselog1" type instead of
 464         "sselog".  Remove explicit "memory" attribute.
 465         (vec_extract_lo_v32hi): Add (=v, v) alternative and explicit "memory",
 466         "type", "prefix", "prefix_extra", "length_immediate" and "mode"
 467         attributes.
 468         (vec_extract_hi_v32hi): Merge all alternatives into one, use
 469         "sselog1" type instead of "sselog".  Remove explicit "memory"
 470         attribute.
 471         (vec_extract_hi_v16hi): Merge each pair of alternatives into one,
 472         use "sselog1" type instead of "sselog".  Remove explicit "memory"
 473         attribute.
 474         (vec_extract_lo_v64qi): Add (=v, v) alternative and explicit "memory",
 475         "type", "prefix", "prefix_extra", "length_immediate" and "mode"
 476         attributes.
 477         (vec_extract_hi_v64qi): Merge all alternatives into one, use
 478         "sselog1" type instead of "sselog".  Remove explicit "memory"
 479         attribute.
 480         (vec_extract_hi_v32qi): Merge each pair of alternatives into one,
 481         use "sselog1" type instead of "sselog".  Remove explicit "memory"
 482         attribute.
 483
 484         PR target/85430
 485         * config/i386/i386.md (*ashlqi3_1_slp): Use alu1 type instead of alu.
 486
 487         PR middle-end/85414
 488         * rtlhooks.c (gen_lowpart_if_possible): Don't call gen_lowpart_SUBREG
 489         on a SUBREG.
 490
 491 2018-04-17  Martin Jambor  <mjambor@suse.cz>
 492
 493         PR ipa/85421
 494         * ipa-cp.c (create_specialized_node): Call
 495         expand_all_artificial_thunks if necessary.
 496
 497 2018-04-17  Martin Liska  <mliska@suse.cz>
 498
 499         PR lto/85405
 500         * ipa-devirt.c (odr_types_equivalent_p): Remove trailing
 501         in message, remote space in between '_G' and '('.
 502
 503 2018-04-17  Jakub Jelinek  <jakub@redhat.com>
 504
 505         PR target/85281
 506         * config/i386/sse.md (reduces<mode><mask_scalar_name>,
 507         avx512f_vmcmp<mode>3<round_saeonly_name>,
 508         avx512f_vmcmp<mode>3_mask<round_saeonly_name>,
 509         avx512f_sgetexp<mode><mask_scalar_name><round_saeonly_scalar_name>,
 510         avx512f_rndscale<mode><round_saeonly_name>,
 511         avx512dq_ranges<mode><mask_scalar_name><round_saeonly_scalar_name>,
 512         avx512f_vgetmant<mode><mask_scalar_name><round_saeonly_scalar_name>):
 513         Use %<iptr>2 instead of %2 for -masm=intel.
 514         (avx512f_vcvtss2usi<round_name>, avx512f_vcvtss2usiq<round_name>,
 515         avx512f_vcvttss2usi<round_saeonly_name>,
 516         avx512f_vcvttss2usiq<round_saeonly_name>): Use %k1 instead of %1 for
 517         -masm=intel.
 518         (avx512f_vcvtsd2usi<round_name>, avx512f_vcvtsd2usiq<round_name>,
 519         avx512f_vcvttsd2usi<round_saeonly_name>,
 520         avx512f_vcvttsd2usiq<round_saeonly_name>, ufloatv2siv2df2<mask_name>):
 521         Use %q1 instead of %1 for -masm=intel.
 522         (avx512f_sfixupimm<mode><sd_maskz_name><round_saeonly_name>,
 523         avx512f_sfixupimm<mode>_mask<round_saeonly_name>): Use %<iptr>3 instead
 524         of %3 for -masm=intel.
 525         (sse2_shufpd_v2df_mask): Fix a typo, change %{6%} to %{%6%} for
 526         -masm=intel.
 527         (*avx512vl_<code>v2div2qi2_store): Use %w0 instead of %0 for
 528         -masm=intel.
 529         (*avx512vl_<code><mode>v4qi2_store): Use %k0 instead of %0 for
 530         -masm=intel.
 531         (avx512vl_<code><mode>v4qi2_mask_store): Use a single pattern with
 532         %k0 and %1 for -masm=intel rather than two patterns, one with %0 and
 533         %g1.
 534         (*avx512vl_<code><mode>v8qi2_store): Use %q0 instead of %0 for
 535         -masm=intel.
 536         (avx512vl_<code><mode>v8qi2_mask_store): Use a single pattern with
 537         %q0 and %1 for -masm=intel rather than two patterns, one with %0 and
 538         %g1 and one with %0 and %1.
 539         (avx512er_vmrcp28<mode><round_saeonly_name>,
 540         avx512er_vmrsqrt28<mode><round_saeonly_name>): Use %<iptr>1 instead of
 541         %1 for -masm=intel.
 542         (avx5124fmaddps_4fmaddps_mask, avx5124fmaddps_4fmaddss_mask,
 543         avx5124fmaddps_4fnmaddps_mask, avx5124fmaddps_4fnmaddss_mask,
 544         avx5124vnniw_vp4dpwssd_mask, avx5124vnniw_vp4dpwssds_mask): Swap order
 545         of %0 and %{%4%} for -masm=intel.
 546         (avx5124fmaddps_4fmaddps_maskz, avx5124fmaddps_4fmaddss_maskz,
 547         avx5124fmaddps_4fnmaddps_maskz, avx5124fmaddps_4fnmaddss_maskz,
 548         avx5124vnniw_vp4dpwssd_maskz, avx5124vnniw_vp4dpwssds_maskz): Swap
 549         order of %0 and %{%5%}%{z%} for -masm=intel.
 550
 551 2018-04-17  Jan Hubicka  <jh@suse.cz>
 552
 553         PR lto/85405
 554         * ipa-devirt.c (odr_types_equivalent_p): Handle bit fields.
 555
 556 2018-04-17  Martin Liska  <mliska@suse.cz>
 557
 558         PR ipa/85329
 559         * multiple_target.c (create_dispatcher_calls): Set apostrophes
 560         for target_clone error message.  Make default implementation
 561         clone to be a local declaration.
 562         (separate_attrs): Add new argument and check for an empty
 563         string.
 564         (expand_target_clones): Handle it.
 565         (ipa_target_clone): Make redirection just for target_clones
 566         functions.
 567
 568 2018-04-16  Cesar Philippidis  <cesar@codesourcery.com>
 569             Tom de Vries  <tom@codesourcery.com>
 570
 571         PR middle-end/84955
 572         * omp-expand.c (expand_oacc_for): Add dummy false branch for
 573         tiled basic blocks without omp continue statements.
 574
 575 2018-04-16  Aaron Sawdey  <acsawdey@linux.ibm.com>
 576
 577         PR target/83660
 578         * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Mark
 579         vec_extract expression as having side effects to make sure it gets
 580         a cleanup point.
 581
 582 2018-04-16  H.J. Lu  <hongjiu.lu@intel.com>
 583
 584         PR target/85403
 585         * config/i386/i386.c (get_builtin_code_for_version): Check
 586         error_mark_node.
 587
 588 2018-04-16  Olga Makhotina  <olga.makhotina@intel.com>
 589
 590         PR target/84331
 591         * gcc/config.gcc: Support "skylake".
 592         * gcc/config/i386/i386-c.c (ix86_target_macros_internal): Handle
 593         PROCESSOR_SKYLAKE.
 594         * gcc/config/i386/i386.c (m_SKYLAKE): Define.
 595         (processor_target_table): Add "skylake".
 596         (ix86_option_override_internal): Add "skylake".
 597         (get_builtin_code_for_version): Handle PROCESSOR_SKYLAKE,
 598         PROCESSOR_CANNONLAKE.
 599         (get_builtin_code_for_version): Fix priority for
 600         PROCESSOR_ICELAKE_CLIENT, PROCESSOR_ICELAKE_SERVER,
 601         PROCESSOR_SKYLAKE-AVX512.
 602         * gcc/config/i386/i386.h (processor_costs): Define TARGET_SKYLAKE.
 603         (processor_type): Add PROCESSOR_SKYLAKE.
 604
 605 2018-04-16  Paolo Carlini  <paolo.carlini@oracle.com>
 606             Jason Merrill  <jason@redhat.com>
 607
 608         PR c++/85112
 609         * convert.c (convert_to_integer_1): Use direct recursion for
 610         enumeral types and types with a precision less than the number
 611         of bits in their mode.
 612
 613 2018-04-16  Julia Koval  <julia.koval@intel.com>
 614
 615         PR target/84413
 616         * config/i386/x86-tune.def (X86_TUNE_SSE_UNALIGNED_LOAD_OPTIMAL,
 617         X86_TUNE_SSE_UNALIGNED_STORE_OPTIMAL): Add m_SKYLAKE_AVX512
 618
 619 2018-04-14  Segher Boessenkool  <segher@kernel.crashing.org>
 620
 621         PR target/85293
 622         * config/rs6000/rs6000.opt (mdirect-move): Make deprecated.
 623         * doc/invoke.texi (RS/6000 and PowerPC Options): Remove -mdirect-move
 624         and -mno-direct-move.
 625
 626 2018-04-13  Paul A. Clarke  <pc@us.ibm.com>
 627
 628         PR target/83402
 629         * config/rs6000/emmintrin.h (_mm_slli_epi{16,32,64}):
 630         Ensure that vec_splat_s32 is only called with 0 <= shift < 16.
 631         Ensure negative shifts result in {0}.
 632
 633 2018-04-13  Vladimir Makarov  <vmakarov@redhat.com>
 634
 635         PR rtl-optimization/79916
 636         * config/rs6000/rs6000.c (rs6000_emit_move): Use assigned hard
 637         regs (if any) to define how to gnerate SD moves when LRA is in
 638         progress.
 639
 640 2018-04-13  Jakub Jelinek  <jakub@redhat.com>
 641
 642         PR rtl-optimization/85393
 643         * except.h (expand_dw2_landing_pad_for_region): Remove declaration.
 644         * except.c (expand_dw2_landing_pad_for_region): Make static.
 645         * bb-reorder.c (fix_up_crossing_landing_pad): In new_bb emit just
 646         a label and unconditional jump to old_bb, rather than
 647         expand_dw2_landing_pad_for_region insn(s) and jump to single_succ
 648         basic block.
 649
 650         PR rtl-optimization/85376
 651         * simplify-rtx.c (simplify_const_unary_operation): For CLZ and CTZ and
 652         zero op0, if C?Z_DEFINED_VALUE_AT_ZERO is false, return NULL_RTX
 653         instead of a specific value.
 654
 655 2018-04-13  Jan Hubicka  <hubicka@ucw.cz>
 656             Bin Cheng  <bin.cheng@arm.com>
 657
 658         PR tree-optimization/82965
 659         PR tree-optimization/83991
 660         * cfgloopanal.c (expected_loop_iterations_unbounded): Add
 661         by_profile_only parameter.
 662         * cfgloopmanip.c (scale_loop_profile): Further scale loop's profile
 663         information if the loop was predicted to iterate too many times.
 664         * cfgloop.h (expected_loop_iterations_unbounded): Update prototype
 665
 666 2018-04-13  Jan Hubicka  <hubicka@ucw.cz>
 667
 668         PR lto/71991
 669         * config/i386/i386.c (ix86_can_inline_p): Allow safe transitions for
 670         always inline.
 671
 672 2018-04-13  Martin Liska  <mliska@suse.cz>
 673             Jakub Jelinek  <jakub@redhat.com>
 674
 675         PR middle-end/81657
 676         * expr.h (enum block_op_methods): Add BLOCK_OP_NO_LIBCALL_RET.
 677         * expr.c (emit_block_move_hints): Handle BLOCK_OP_NO_LIBCALL_RET.
 678         * builtins.c (expand_builtin_memory_copy_args): Use
 679         BLOCK_OP_NO_LIBCALL_RET method for mempcpy with non-ignored target,
 680         handle dest_addr == pc_rtx.
 681
 682 2018-04-12  Segher Boessenkool  <segher@kernel.crashing.org>
 683
 684         PR target/85291
 685         * config/rs6000/rs6000.md (fix_trunc<mode>si2): Use legacy code if
 686         asked to not generate direct moves.
 687         (fix_trunc<mode>si2_stfiwx): Similar.
 688         (fix_trunc<mode>si2_internal): Similar.
 689
 690 2018-04-12  Jakub Jelinek  <jakub@redhat.com>
 691
 692         PR debug/83157
 693         * var-tracking.c (add_stores): Handle STRICT_LOW_PART SET_DEST.
 694         * cselib.c (cselib_record_sets): For STRICT_LOW_PART dest,
 695         lookup if dest in some wider mode is known to be const0_rtx and
 696         if so, record permanent equivalence for it to be ZERO_EXTEND of
 697         the narrower mode destination.
 698
 699 2018-04-12  Cesar Philippidis  <cesar@codesourcery.com>
 700
 701         * lto-streamer-out.c (output_function): Revert 259346.
 702         * omp-expand.c (expand_oacc_for): Likewise.
 703
 704 2018-04-12  Alexander Monakov  <amonakov@ispras.ru>
 705
 706         PR rtl-optimization/85354
 707         * sel-sched-ir.c (sel_init_pipelining): Move cfg_cleanup call...
 708         * sel-sched.c (sel_global_init): ... here.
 709
 710 2018-04-12  Eric Botcazou  <ebotcazou@adacore.com>
 711
 712         PR target/85238
 713         * lto-wrapper.c (debug_objcopy): Open the files in binary mode.
 714         * dwarf2out.c (dwarf2out_early_finish): Do not generate assembly in LTO
 715         mode for PE-COFF targets.
 716         * config/i386/i386-protos.h (i386_pe_asm_lto_start): Declare.
 717         (i386_pe_asm_lto_end): Likewise.
 718         * config/i386/cygming.h (TARGET_ASM_LTO_START): Define.
 719         (TARGET_ASM_LTO_END): Likewise.
 720         * config/i386/winnt.c (saved_debug_info_level): New static variable.
 721         (i386_pe_asm_lto_start): New function.
 722         (i386_pe_asm_lto_end): Likewise.
 723
 724 2018-04-12  Cesar Philippidis  <cesar@codesourcery.com>
 725             Richard Biener  <rguenther@suse.de>
 726
 727         PR middle-end/84955
 728         * lto-streamer-out.c (output_function): Fix CFG loop state before
 729         streaming out.
 730         * omp-expand.c (expand_oacc_for): Handle calls to internal
 731         functions like regular functions.
 732
 733 2018-04-12  Richard Biener  <rguenther@suse.de>
 734
 735         PR lto/85371
 736         * dwarf2out.c (init_sections_and_labels): Use debug_line_section[_label]
 737         for the early LTO debug to properly generate references to it
 738         during DIE emission.  Do not re-use that for the skeleton for
 739         split-dwarf.
 740         (dwarf2out_early_finish): Likewise.
 741
 742 2018-04-12  Jakub Jelinek  <jakub@redhat.com>
 743
 744         PR target/85328
 745         * config/i386/sse.md
 746         (<mask_codefor>avx512dq_vextract<shuffletype>64x2_1<mask_name> split,
 747         <mask_codefor>avx512f_vextract<shuffletype>32x4_1<mask_name> split,
 748         vec_extract_lo_<mode><mask_name> split, vec_extract_lo_v32hi,
 749         vec_extract_lo_v64qi): For non-AVX512VL if input is xmm16+ reg
 750         and output is a reg, avoid creating invalid lowpart subreg, but
 751         instead split into a 512-bit move.  Don't split if not AVX512VL,
 752         input is xmm16+ reg and output is a mem.
 753         (vec_extract_lo_<mode><mask_name>, vec_extract_lo_v32hi,
 754         vec_extract_lo_v64qi): Don't require split if not AVX512VL, input is
 755         xmm16+ reg and output is a mem.
 756
 757 2018-04-12  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>
 758
 759         * config/s390/s390.c (s390_output_indirect_thunk_function): Check
 760         also for flag_dwarf2_cfi_asm.
 761
 762 2018-04-12  Jakub Jelinek  <jakub@redhat.com>
 763
 764         PR rtl-optimization/85342
 765         * regcprop.c (copyprop_hardreg_forward_1): Remove replaced array, use
 766         a bool scalar var inside of the loop instead.  Don't try to update
 767         recog_data.operand after failed apply_change_group.
 768
 769 2018-04-12  Tom de Vries  <tom@codesourcery.com>
 770
 771         PR target/85296
 772         * config/nvptx/nvptx.c (flexible_array_member_type_p): New function.
 773         (nvptx_assemble_decl_begin): Add undefined param.  Declare undefined
 774         array with flexible array member as array without given dimension.
 775         (nvptx_assemble_undefined_decl): Set nvptx_assemble_decl_begin call
 776         argument for undefined param to true.
 777
 778 2018-04-11  Aaron Sawdey  <acsawdey@linux.ibm.com>
 779
 780         PR target/85321
 781         * doc/invoke.texi (RS/6000 and PowerPC Options): Document options
 782         -mcall- and -mtraceback=. Remove options -mabi=spe and -mabi=no-spe
 783         from PowerPC section.
 784         * config/rs6000/sysv4.opt (mcall-): Improve help text.
 785         * config/rs6000/rs6000.opt (mblock-compare-inline-limit=): Trim
 786         help text that is too long.
 787         * config/rs6000/rs6000.opt (mblock-compare-inline-loop-limit=): Trim
 788         help text that is too long.
 789         * config/rs6000/rs6000.opt (mstring-compare-inline-limit=): Trim
 790         help text that is too long.
 791
 792 2018-04-11  Uros Bizjak  <ubizjak@gmail.com>
 793
 794         * config/alpha/alpha.md (stack_probe_internal): Rename
 795         from "probe_stack".  Update all callers.
 796
 797 2018-04-11  Alexander Monakov  <amonakov@ispras.ru>
 798
 799         PR rtl-optimization/84566
 800         * sched-deps.c (sched_analyze_insn): Check deps->readonly when invoking
 801         sched_macro_fuse_insns.
 802
 803 2018-04-11  Alexander Monakov  <amonakov@ispras.ru>
 804
 805         PR target/84301
 806         * sched-rgn.c (add_branch_dependences): Move sel_sched_p check here...
 807         (compute_block_dependences): ... from here.
 808
 809 2018-04-11  Jakub Jelinek  <jakub@redhat.com>
 810
 811         PR tree-optimization/85331
 812         * vec-perm-indices.h (vec_perm_indices::clamp): Change input type
 813         from int to HOST_WIDE_INT.
 814
 815 2018-04-11  Martin Jambor  <mjambor@suse.cz>
 816
 817         PR ipa/84149
 818         * ipa-cp.c (propagate_vals_across_pass_through): Expand comment.
 819         (cgraph_edge_brings_value_p): New parameter dest_val, check if it is
 820         not the same as the source val.
 821         (cgraph_edge_brings_value_p): New parameter.
 822         (gather_edges_for_value): Pass destination value to
 823         cgraph_edge_brings_value_p.
 824         (perhaps_add_new_callers): Likewise.
 825         (get_info_about_necessary_edges): Likewise and exclude values brought
 826         only by self-recursive edges.
 827         (create_specialized_node): Redirect only clones of self-calling edges.
 828         (+self_recursive_pass_through_p): New function.
 829         (find_more_scalar_values_for_callers_subset): Use it.
 830         (find_aggregate_values_for_callers_subset): Likewise.
 831         (known_aggs_to_agg_replacement_list): Removed.
 832         (decide_whether_version_node): Re-calculate known constants for all
 833         remaining context clones.
 834
 835 2018-04-11  Richard Biener  <rguenther@suse.de>
 836
 837         PR lto/85339
 838         * dwarf2out.c (dwarf2out_finish): Remove DW_AT_stmt_list attribute
 839         from early DWARF output.
 840         (dwarf2out_early_finish): Output line info unconditionally into
 841         early DWARF and add reference to it.
 842
 843 2018-04-11  Jakub Jelinek  <jakub@redhat.com>
 844
 845         PR target/85281
 846         * config/i386/sse.md (iptr): Add V16SFmode and V8DFmode cases.
 847         (<avx512>_vec_dup<mode><mask_name>): Use a single pattern for modes
 848         other than V2DFmode using iptr mode attribute.
 849         (<avx512>_vec_dup<mode><mask_name>): Use iptr mode attribute.
 850
 851 2018-04-11  Alexander Monakov  <amonakov@ispras.ru>
 852
 853         PR rtl-optimization/84659
 854         * sel-sched-ir.c (sel_init_pipelining): Invoke cleanup_cfg.
 855
 856 2018-04-11  Jakub Jelinek  <jakub@redhat.com>
 857
 858         PR debug/85302
 859         * dwarf2out.c (skip_loc_list_entry): Don't call size_of_locs if
 860         SIZEP is NULL.
 861         (output_loc_list): Pass address of a dummy size variable even in the
 862         locview handling loop.
 863         (index_location_lists): Add comment on why skip_loc_list_entry can't
 864         call size_of_locs.
 865
 866 2018-04-11  Thomas Preud'homme  <thomas.preudhomme@arm.com>
 867
 868         PR target/85261
 869         * config/arm/arm-builtins.c (arm_expand_builtin): Force input operand
 870         into register.
 871
 872 2018-04-10  Aaron Sawdey  <acsawdey@linux.ibm.com>
 873
 874         PR target/85321
 875         * doc/invoke.texi (RS/6000 and PowerPC Options): Document options
 876         -mblock-compare-inline-limit, -mblock-compare-inline-loop-limit,
 877         and -mstring-compare-inline-limit.
 878
 879 2018-04-10  Segher Boessenkool  <segher@kernel.crashing.org>
 880
 881         PR target/85287
 882         * gcc/config/rs6000/rs6000.md (allocate_stack): Put the residual size
 883         for stack clash protection in a register whenever we need it to be in
 884         a register.
 885
 886 2018-04-10  Segher Boessenkool  <segher@kernel.crashing.org>
 887
 888         * common/config/rs6000/rs6000-common.c (rs6000_option_init_struct):
 889         Enable -fasynchronous-unwind-tables by default if OBJECT_FORMAT_ELF.
 890
 891 2018-04-10  Segher Boessenkool  <segher@kernel.crashing.org>
 892
 893         PR target/85321
 894         * config/rs6000/rs6000.opt (mtraceback=): Show the allowed values in
 895         the help text.
 896         (mlong-double-): Ditto.
 897         * config/rs6000/sysv4.opt (msdata=): Ditto.
 898         (mtls-size=): Ditto.
 899
 900 2018-04-10  Kelvin Nilsen  <kelvin@gcc.gnu.org>
 901
 902         * config/rs6000/rs6000-c.c (altivec_overloaded_builtins): Remove
 903         erroneous entries for
 904         "vector int vec_ldl (int, long int *)", and
 905         "vector unsigned int vec_ldl (int, unsigned long int *)".
 906         Add comments and entries for
 907         "vector bool char vec_ldl (int, bool char *)",
 908         "vector bool short vec_ldl (int, bool short *)",
 909         "vector bool int vec_ldl (int, bool int *)",
 910         "vector bool long long vec_ldl (int, bool long long *)",
 911         "vector pixel vec_ldl (int, pixel *)",
 912         "vector long long vec_ldl (int, long long *)",
 913         "vector unsigned long long vec_ldl (int, unsigned long long *)".
 914         * config/rs6000/rs6000.c (rs6000_init_builtins): Initialize new
 915         type tree bool_long_long_type_node and correct definition of
 916         bool_V2DI_type_node to make reference to this new type tree.
 917         (rs6000_mangle_type): Replace erroneous reference to
 918         bool_long_type_node with bool_long_long_type_node.
 919         * config/rs6000/rs6000.h (enum rs6000_builtin_type_index): Add
 920         comments to emphasize sign distinctions for char and int types and
 921         replace RS6000_BTI_bool_long constant with
 922         RS6000_BTI_bool_long_long constant.  Also add comment to restrict
 923         use of RS6000_BTI_pixel.
 924         (bool_long_type_node): Remove this macro definition.
 925         (bool_long_long_type_node): New macro definition
 926
 927 2018-04-10  Jakub Jelinek  <jakub@redhat.com>
 928
 929         PR rtl-optimization/85300
 930         * combine.c (subst): Handle subst of CONST_SCALAR_INT_P new_rtx also
 931         into FLOAT and UNSIGNED_FLOAT like ZERO_EXTEND, return a CLOBBER if
 932         simplify_unary_operation fails.
 933
 934 2018-04-10  Martin Liska  <mliska@suse.cz>
 935
 936         * gdbhooks.py: Add pretty-printers for varpool_node, symtab_node,
 937         cgraph_edge and ipa_ref.
 938
 939 2018-04-10  Jakub Jelinek  <jakub@redhat.com>
 940
 941         PR target/85177
 942         PR target/85255
 943         * config/i386/sse.md
 944         (<extract_type>_vinsert<shuffletype><extract_suf>_mask): Fix
 945         computation of the VEC_MERGE selector from mask.
 946         (<extract_type>_vinsert<shuffletype><extract_suf>_1<mask_name>):
 947         Fix decoding of the VEC_MERGE selector into mask.
 948
 949 2018-04-10  Richard Sandiford  <richard.sandiford@linaro.org>
 950
 951         PR tree-optimization/85286
 952         * tree-vect-data-refs.c (vect_get_smallest_scalar_type):
 953
 954 2018-04-10  Richard Sandiford  <richard.sandiford@linaro.org>
 955
 956         * final.c (final_1): Set insn_last_address as well as
 957         insn_current_address.
 958
 959 2018-04-10  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
 960
 961         PR target/85173
 962         * explow.c (emit_stack_probe): Call validize_mem on memory location
 963         before passing it to gen_probe_stack.  Create address operand and
 964         legitimize it for the probe_stack_address case.
 965
 966 2018-04-09  Jan Hubicka  <jh@suse.cz>
 967
 968         PR lto/85078
 969         * ipa-devirt.c (rebuild_type_inheritance-hash): New.
 970         * ipa-utils.h (rebuild_type_inheritance-hash): Declare.
 971         * tree.c (free_lang_data_in_type): Fix handling of binfos;
 972         walk basetypes.
 973         (free_lang_data): Rebuild type inheritance graph.
 974
 975 2018-04-09  Martin Sebor  <msebor@redhat.com>
 976
 977         * invoke.texi (-finline-small-functions): Mention other optimization
 978         options.
 979         (-findirect-inlining, -fpartial-inlining): Same.
 980         (-finline-functions-called-once): Same.
 981         (-freorder-blocks-and-partition): Same.
 982
 983 2018-04-09  Jan Hubicka  <jh@suse.cz>
 984
 985         PR rtl/84058
 986         * cfgcleanup.c (try_forward_edges): Do not give up on crossing
 987         jumps; choose last target that matches the criteria (i.e.
 988         no partition changes for non-crossing jumps).
 989         * cfgrtl.c (cfg_layout_redirect_edge_and_branch): Add basic
 990         support for redirecting crossing jumps to non-crossing.
 991
 992 2018-04-09  Alexey Brodkin  <abrodkin@synopsys.com>
 993
 994         * config/arc/arc.c (arc_expand_prologue): Set stack usage info
 995         also for naked functions.
 996
 997 2018-04-09  Claudiu Zissulescu  <claziss@synopsys.com>
 998
 999         * config/arc/arc.md (add_shift): New pattern.
1000         (add_shift2): Likewise.
1001         (sub_shift): Likewise.
1002         (sub_shift_cmp0_noout): Likewise.
1003         (compare_si_ashiftsi): Likewise.
1004         (xbfu_cmp0_noout): New combine pattern.
1005         (xbfu_cmp0"): Likewise.
1006         (movsi_set_cc_insn): Place the predicable variant first.
1007         (commutative_binary_cmp0_noout): Remove clobber.
1008         (commutative_binary_cmp0): New pattern.
1009         (noncommutative_binary_cmp0): Likewise.
1010         (noncommutative_binary_cmp0_noout): Likewise.
1011         (noncommutative_binary_comparison_result_used): Removed.
1012         (rsub_cmp0): New pattern.
1013         (rsub_cmp0_noout): Likewise.
1014         (extzvsi): Changed, keep only meaningful variants.
1015         (SQH, SEZ): New iterators.
1016         (SQH_postfix): New mode attribute.
1017         (SEZ_prefix): New code attribute.
1018         (<SEZ_prefix>xt<SQH_postfix>_cmp0_noout): New instruction pattern.
1019         (<SEZ_prefix>xt<SQH_postfix>_cmp0): Likewise.
1020         * config/arc/predicates.md (cc_set_register): Use CC_REG instead
1021         of numerical value.
1022         (noncommutative_operator): Check the availability of barrel
1023         shifter option.
1024
1025 2018-04-09  Richard Biener  <rguenther@suse.de>
1026
1027         PR tree-optimization/85284
1028         * tree-ssa-loop-niter.c (number_of_iterations_exit_assumptions):
1029         Only use the niter constraining form of simple_iv when the exit
1030         is always executed.
1031
1032 2018-04-09  Tom de Vries  <tom@codesourcery.com>
1033
1034         PR target/84041
1035         * config/nvptx/nvptx.md (define_c_enum "unspecv"): Add UNSPECV_MEMBAR.
1036         (define_expand "*memory_barrier"): New define_expand.
1037         (define_insn "memory_barrier"): New insn.
1038
1039 2018-04-09  Andrey Belevantsev  <abel@ispras.ru>
1040
1041         PR rtl-optimization/80463
1042         PR rtl-optimization/83972
1043         PR rtl-optimization/83480
1044
1045         * sel-sched-ir.c (has_dependence_note_mem_dep): Take into account the
1046         correct producer for the insn.
1047         (tidy_control_flow): Fixup seqnos in case of debug insns.
1048
1049 2018-04-09  Andrey Belevantsev  <abel@ispras.ru>
1050
1051         PR rtl-optimization/83913
1052
1053         * sel-sched-ir.c (merge_expr_data): Choose the middle between two
1054         different sched-times when merging exprs.
1055
1056 2018-04-09  Andrey Belevantsev  <abel@ispras.ru>
1057
1058         PR rtl-optimization/83962
1059
1060         * sel-sched-ir.c (tidy_control_flow): Correct the order in which we call
1061         tidy_fallthru_edge and tidy_control_flow.
1062
1063 2018-04-09  Andrey Belevantsev  <abel@ispras.ru>
1064
1065         PR rtl-optimization/83530
1066
1067         * sel-sched.c (force_next_insn): New global variable.
1068         (remove_insn_for_debug): When force_next_insn is true, also leave only
1069         next insn in the ready list.
1070         (sel_sched_region): When the region wasn't scheduled, make another pass
1071         over it with force_next_insn set to 1.
1072
1073 2018-04-08  Monk Chiang  <sh.chiang04@gmail.com>
1074
1075         * config.gcc (nds32le-*-*, nds32be-*-*): Add nds32/nds32_intrinsic.h
1076         into tm_file.
1077         * config/nds32/constants.md (unspec_volatile_element): Add enum values
1078         for interrupt control.
1079         * config/nds32/nds32-intrinsic.c: Implementation of intrinsic
1080         functions for interrupt control.
1081         * config/nds32/nds32-intrinsic.md: Likewise.
1082         * config/nds32/nds32_intrinsic.h: Likewise.
1083         * config/nds32/nds32.h (nds32_builtins): Likewise.
1084
1085 2018-04-08  Chung-Ju Wu  <jasonwucj@gmail.com>
1086
1087         * config/nds32/nds32.c (nds32_init_machine_status,
1088         nds32_legitimate_index_p, nds32_legitimate_address_p): Consider
1089         strict_aligned_p field.
1090         (nds32_expand_to_rtl_hook): New function.
1091         (TARGET_EXPAND_TO_RTL_HOOK): Define.
1092         * config/nds32/nds32.h (machine_function): Add strict_aligned_p field.
1093
1094 2018-04-08  Kito Cheng  <kito.cheng@gmail.com>
1095             Chung-Ju Wu  <jasonwucj@gmail.com>
1096
1097         * config.gcc (nds32*-*-*): Check that n7 is valid to --with-cpu.
1098         * config/nds32/nds32-n7.md: New file.
1099         * config/nds32/nds32-opts.h (nds32_cpu_type): Add CPU_N7.
1100         * config/nds32/nds32-pipelines-auxiliary.c: Implementation for n7
1101         pipeline.
1102         * config/nds32/nds32-protos.h: More declarations for n7 pipeline.
1103         * config/nds32/nds32.md (pipeline_model): Add n7.
1104         * config/nds32/nds32.opt (mcpu): Support n7 pipeline cpus.
1105         * config/nds32/pipelines.md: Include n7 settings.
1106
1107 2018-04-08  Kito Cheng  <kito.cheng@gmail.com>
1108             Chung-Ju Wu  <jasonwucj@gmail.com>
1109
1110         * config.gcc (nds32*-*-*): Check that e8 is valid to --with-cpu.
1111         * config/nds32/nds32-e8.md: New file.
1112         * config/nds32/nds32-opts.h (nds32-cpu_type): Add CPU_E8.
1113         * config/nds32/nds32-pipelines-auxiliary.c: Implementation for e8
1114         pipeline.
1115         * config/nds32/nds32-protos.h: More declarations for e8 pipeline.
1116         * config/nds32/nds32.md (pipeline_model): Add e8.
1117         * config/nds32/nds32.opt (mcpu): Support e8 pipeline cpus.
1118         * config/nds32/pipelines.md: Include e8 settings.
1119
1120 2018-04-08  Kito Cheng  <kito.cheng@gmail.com>
1121             Chung-Ju Wu  <jasonwucj@gmail.com>
1122
1123         * config.gcc (nds32*-*-*): Check that n6/n8/s8 are valid to --with-cpu.
1124         * config/nds32/nds32-n8.md: New file.
1125         * config/nds32/nds32-opts.h (nds32_cpu_type): Add CPU_N6 and CPU_N8.
1126         * config/nds32/nds32-pipelines-auxiliary.c: Implementation for n8
1127         pipeline.
1128         * config/nds32/nds32-protos.h: More declarations for n8 pipeline.
1129         * config/nds32/nds32-utils.c: More implementations for n8 pipeline.
1130         * config/nds32/nds32.md (pipeline_model): Add n8.
1131         * config/nds32/nds32.opt (mcpu): Support n8 pipeline cpus.
1132         * config/nds32/pipelines.md: Include n8 settings.
1133
1134 2018-04-08  Kito Cheng  <kito.cheng@gmail.com>
1135             Chung-Ju Wu  <jasonwucj@gmail.com>
1136
1137         * config.gcc (nds32*): Add nds32-utils.o into extra_objs.
1138         * config/nds32/nds32-n9-2r1w.md: New file.
1139         * config/nds32/nds32-n9-3r2w.md: New file.
1140         * config/nds32/nds32-opts.h (nds32_cpu_type, nds32_mul_type,
1141         nds32_register_ports): New or modify for cpu n9.
1142         * config/nds32/nds32-pipelines-auxiliary.c: Implementation for n9
1143         pipeline.
1144         * config/nds32/nds32-protos.h: More declarations for n9 pipeline.
1145         * config/nds32/nds32-utils.c: New file.
1146         * config/nds32/nds32.h (TARGET_PIPELINE_N9, TARGET_PIPELINE_SIMPLE,
1147         TARGET_MUL_SLOW): Define.
1148         * config/nds32/nds32.md (pipeline_model): New attribute.
1149         * config/nds32/nds32.opt (mcpu, mconfig-mul, mconfig-register-ports):
1150         New options that support cpu n9.
1151         * config/nds32/pipelines.md: Include n9 settings.
1152         * config/nds32/t-nds32 (nds32-utils.o): Add dependency.
1153
1154 2018-04-08  Chung-Ju Wu  <jasonwucj@gmail.com>
1155
1156         * config/nds32/nds32-md-auxiliary.c (output_cond_branch): Output align
1157         information if necessary.
1158         (output_cond_branch_compare_zero): Likewise.
1159         * config/nds32/nds32.c (nds32_adjust_insn_length): Consider align case.
1160         (nds32_target_alignment): Refine for alignment.
1161         * config/nds32/nds32.h (NDS32_ALIGN_P): Define.
1162         (FUNCTION_BOUNDARY): Modify.
1163         * config/nds32/nds32.md (call_internal, call_value_internal): Consider
1164         align case.
1165         * config/nds32/nds32.opt (malways-align, malign-functions): New.
1166
1167 2018-04-08  Monk Chiang  <sh.chiang04@gmail.com>
1168
1169         * config/nds32/constants.md (unspec_volatile_element): Add values for
1170         TLB operation and data prefetch.
1171         * config/nds32/nds32-intrinsic.c: Implementation of intrinsic
1172         functions for TLB operation and data prefetch.
1173         * config/nds32/nds32-intrinsic.md: Likewise.
1174         * config/nds32/nds32_intrinsic.h: Likewise.
1175         * config/nds32/nds32.c (nds32_dpref_names): Likewise.
1176         (nds32_print_operand): Likewise.
1177         * config/nds32/nds32.h (nds32_builtins): Likewise.
1178
1179 2018-04-07  Thomas Koenig  <tkoenig@gcc.gnu.org>
1180         Andrew Pinski <pinsika@gcc.gnu.org>
1181
1182         PR middle-end/82976
1183         * match.pd: Use constant_boolean_node of correct type instead of
1184         boolean_true_node or boolean_false_node for simplifying
1185         pointer comparisons to zero.
1186
1187 2018-04-07  Jakub Jelinek  <jakub@redhat.com>
1188
1189         PR tree-optimization/80021
1190         * tree.c (verify_type_variant): Make error call in verify_variant_match
1191         translatable and remove final full stop.
1192
1193 2018-04-07  Chung-Ju Wu  <jasonwucj@gmail.com>
1194
1195         * config/nds32/constants.md (unspec_volatile_element): Add
1196         UNSPEC_VOLATILE_EH_RETURN.
1197         * config/nds32/nds32-md-auxiliary.c (nds32_output_stack_push,
1198         nds32_output_stack_pop): Support dwarf exception handling process.
1199         * config/nds32/nds32-protos.h (nds32_dynamic_chain_address): Declare.
1200         * config/nds32/nds32.c (nds32_init_machine_status): Support dwarf
1201         exception handling process.
1202         (nds32_compute_stack_frame): Likewise.
1203         (nds32_return_addr_rtx): Likewise.
1204         (nds32_initial_elimination_offset): Likewise.
1205         (nds32_expand_prologue): Likewise.
1206         (nds32_expand_epilogue): Likewise.
1207         (nds32_dynamic_chain_address): New function.
1208         * config/nds32/nds32.h (machine_function): Add fields for dwarf
1209         exception handling.
1210         (DYNAMIC_CHAIN_ADDRESS): Define.
1211         (EH_RETURN_DATA_REGNO): Define.
1212         (EH_RETURN_STACKADJ_RTX): Define.
1213         * config/nds32/nds32.md (eh_return, nds32_eh_return): Implement
1214         patterns for dwarf exception handling.
1215
1216 2018-04-07  Chung-Ju Wu  <jasonwucj@gmail.com>
1217
1218         * config/nds32/nds32.h: Clean up obsolete macros.
1219
1220 2018-04-07  Monk Chiang  <sh.chiang04@gmail.com>
1221
1222         * config/nds32/constants.md (unspec_element, unspec_volatile_element):
1223         Add enum values for particular instructions.
1224         * config/nds32/nds32-intrinsic.c: Implementation of expanding
1225         particular intrinsic functions.
1226         * config/nds32/nds32-intrinsic.md: Likewise.
1227         * config/nds32/nds32_intrinsic.h: Likewise.
1228         * config/nds32/nds32.h (nds32_builtins): Likewise.
1229         * config/nds32/nds32.md (type): Add pbsad and pbsada.
1230         (btst, ave): New patterns for particular instructions.
1231
1232 2018-04-07  Monk Chiang  <sh.chiang04@gmail.com>
1233
1234         * config/nds32/constants.md (unspec_element, unspec_volatile_element):
1235         Add enum values for atomic load/store and memory sync.
1236         * config/nds32/nds32-intrinsic.c: Implementation for atomic load/store
1237         and memory sync.
1238         * config/nds32/nds32-intrinsic.md: Likewise.
1239         * config/nds32/nds32_intrinsic.h: Likewise.
1240         * config/nds32/nds32.h (nds32_builtins): Likewise.
1241
1242 2018-04-07  Jakub Jelinek  <jakub@redhat.com>
1243
1244         PR tree-optimization/85257
1245         * fold-const.c (native_encode_vector): If not all elts could fit
1246         and off is -1, return 0 rather than offset.
1247         * tree-ssa-sccvn.c (vn_reference_lookup_3): Pass
1248         (offseti - offset2) / BITS_PER_UNIT as 4th argument to
1249         native_encode_expr.  Verify len * BITS_PER_UNIT >= maxsizei.  Don't
1250         adjust buffer in native_interpret_expr call.
1251
1252 2018-04-07  Monk Chiang  <sh.chiang04@gmail.com>
1253
1254         * config/nds32/constants.md (unspec_volatile_element): Add cache
1255         control enum values.
1256         * config/nds32/nds32-intrinsic.c: Add cache control expand functions.
1257         * config/nds32/nds32-intrinsic.md: Add cache control patterns.
1258         * config/nds32/nds32.c (nds32_cctl_names): New.
1259         (nds32_print_operand): Handle cache control register names.
1260         * config/nds32/nds32.h (nds32_builtins): New enum values.
1261         * config/nds32/nds32_intrinsic.h: Add cache control enum types and
1262         macros.
1263         * config/nds32/nds32.md (type): Add mmu.
1264         * config/nds32/pipelines.md (simple_insn): Add mmu.
1265
1266 2018-04-07  Chung-Ju Wu  <jasonwucj@gmail.com>
1267
1268         * config/nds32/nds32.md (type): Remove call.
1269         * config/nds32/pipelines.md (simple_insn): Likewise.
1270
1271 2018-04-07  Monk Chiang  <sh.chiang04@gmail.com>
1272
1273         * config/nds32/constants.md (unspec_volatile_element): Add
1274         UNSPEC_VOLATILE_FMFCSR, UNSPEC_VOLATILE_FMTCSR and
1275         UNSPEC_VOLATILE_FMFCFG.
1276         * config/nds32/nds32-intrinsic.c (bdesc_noarg): New builtin
1277         description for fmfcfg and fmfcsr.
1278         (bdesc_1arg): Add fmtcsr.
1279         (bdesc_2arg): Add fcpynss, fcpyss, fcpynsd and fcpysd.
1280         (nds32_expand_builtin_impl): Deal with FPU intrinsic functions.
1281         * config/nds32/nds32-intrinsic.md (unspec_fcpynsd, unspec_fcpysd,
1282         unspec_fcpynss, unspec_fcpysd, unspec_fcpyss, unspec_fmfcsr,
1283         unspec_fmfcfg): New patterns.
1284         * config/nds32/nds32.h (nds32_builtins): Add NDS32_BUILTIN_FMFCFG,
1285         NDS32_BUILTIN_FMFCSR, NDS32_BUILTIN_FMTCSR, NDS32_BUILTIN_FCPYNSS,
1286         NDS32_BUILTIN_FCPYSS,NDS32_BUILTIN_FCPYNSD and NDS32_BUILTIN_FCPYSD.
1287         * config/nds32/nds32_intrinsic.h (__nds32__fcpynsd, __nds32__fcpynss,
1288         __nds32__fcpysd, __nds32__fcpyss, __nds32__fmfcsr, __nds32__fmtcsr,
1289         __nds32__fmfcfg): Define.
1290
1291 2018-04-07  Monk Chiang  <sh.chiang04@gmail.com>
1292
1293         * config/nds32/nds32.c (nds32_intrinsic_register_names): Add more
1294         intrinsic register names.
1295         * config/nds32/nds32_intrinsic.h (nds32_intrinsic_registers): Add more
1296         intrinsic register enum values and macros.
1297
1298 2018-04-07  Chung-Ju Wu  <jasonwucj@gmail.com>
1299
1300         * config/nds32/nds32.c (nds32_legitimate_index_p): Modify condition
1301         for load/store addressing form.
1302         (nds32_print_operand_address): Likewise.
1303
1304 2018-04-06  Eric Botcazou  <ebotcazou@adacore.com>
1305
1306         PR target/85196
1307         * config/sparc/sparc.c (sparc_expand_move): Deal with symbolic operands
1308         based on LABEL_REF.  Remove useless assertion.
1309         (pic_address_needs_scratch): Fix formatting.
1310         (sparc_legitimize_pic_address): Minor tweaks.
1311         (sparc_delegitimize_address): Adjust assertion accordingly.
1312         * config/sparc/sparc.md (movsi_pic_label_ref): Change label_ref_operand
1313         into symbolic_operand.
1314         (movsi_high_pic_label_ref): Likewise.
1315         (movsi_lo_sum_pic_label_ref): Likewise.
1316         (movdi_pic_label_ref): Likewise.
1317         (movdi_high_pic_label_ref): Likewise.
1318         (movdi_lo_sum_pic_label_ref): Likewise.
1319
1320 2018-04-06  Amaan Cheval  <amaan.cheval@gmail.com>
1321
1322         * config.gcc (x86_64-*-rtems*): Add rtems.h to tm_file for
1323         custom LIB_SPEC setup.
1324
1325 2018-04-06  Ruslan Bukin  <br@bsdpad.com>
1326             Kito Cheng  <kito.cheng@gmail.com>
1327
1328         * config.gcc (riscv*-*-freebsd*): Add RISC-V FreeBSD support.
1329         * config/riscv/freebsd.h: New.
1330
1331 2018-04-06  Chung-Ju Wu  <jasonwucj@gmail.com>
1332
1333         * config/nds32/nds32.c (nds32_adjust_insn_length): Refine.
1334         * config/nds32/nds32.h (ADJUST_INSN_LENGTH): Change the location in
1335         file.
1336
1337 2018-04-06  Chung-Ju Wu  <jasonwucj@gmail.com>
1338             Kito Cheng  <kito.cheng@gmail.com>
1339
1340         * config/nds32/nds32-md-auxiliary.c (nds32_output_return,
1341         nds32_output_call, nds32_symbol_binds_local_p): New functions.
1342         * config/nds32/nds32-protos.h (nds32_output_call,
1343         nds32_output_return): Declare.
1344         * config/nds32/nds32.md: Refine all the call and return patterns.
1345
1346 2018-04-06  Jakub Jelinek  <jakub@redhat.com>
1347
1348         PR debug/85252
1349         * dwarf2out.c (rtl_for_decl_init): For STRING_CST initializer only
1350         build CONST_STRING if TYPE_MAX_VALUE is non-NULL and is INTEGER_CST.
1351
1352         PR rtl-optimization/84872
1353         * cfgloopmanip.c (create_preheader): Use make_forwarder_block even if
1354         nentry == 1 when CP_FALLTHRU_PREHEADERS and single_entry is
1355         EDGE_CROSSING edge.
1356
1357 2018-04-06  Tamar Christina  <tamar.christina@arm.com>
1358
1359         * expr.c (copy_blkmode_to_reg): Revert 254862.
1360         * doc/sourcebuild.texi (word_mode_no_slow_unalign): Likewise.
1361
1362 2018-04-06  Richard Biener  <rguenther@suse.de>
1363
1364         PR middle-end/85244
1365         * tree-dfa.c (get_ref_base_and_extent): Reset seen_variable_array_ref
1366         after seeing a component reference with an adjacent field.  Treat
1367         refs to arrays at struct end of external decls similar to
1368         refs to unconstrained commons.
1369
1370 2018-04-06  Jakub Jelinek  <jakub@redhat.com>
1371
1372         PR sanitizer/85213
1373         * fold-const.c (twoval_comparison_p): Remove SAVE_P argument and don't
1374         look through SAVE_EXPRs with non-side-effects argument.  Adjust
1375         recursive calls.
1376         (fold_comparison): Adjust twoval_comparison_p caller, don't handle
1377         save_p here.
1378
1379 2018-04-06  Richard Biener  <rguenther@suse.de>
1380
1381         PR middle-end/85180
1382         * alias.c (find_base_term): New wrapper around find_base_term
1383         unwinding CSELIB_VAL_PTR changes.
1384         (find_base_term): Do not restore CSELIB_VAL_PTR during the
1385         recursion.
1386
1387 2018-04-06  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>
1388
1389         * config/s390/s390.c (s390_z10_optimize_cmp): Expand dedicated NOP
1390         instructions.
1391         * config/s390/s390.md (UNSPECV_NOP_LR_0, UNSPECV_NOP_LR_1): New
1392         constant definitions.
1393         ("nop"): lr 0,0 -> nopr r0
1394         ("nop_lr0", "nop_lr1"): New insn definitions.
1395
1396 2018-04-06  Chung-Ju Wu  <jasonwucj@gmail.com>
1397
1398         * config/nds32/nds32.md (*stack_push, *stack_pop): Use
1399         NDS32_V3PUSH_AVAILABLE_P macro.
1400
1401 2018-04-06  Monk Chiang  <sh.chiang04@gmail.com>
1402             Chung-Ju Wu  <jasonwucj@gmail.com>
1403
1404         * config.gcc (nds32*-*-*): Add v2j v3f v3s checking.
1405         (nds32*-*-*): Add float and fpu_config into supported_defaults.
1406         * common/config/nds32/nds32-common.c (TARGET_DEFAULT_TARGET_FLAGS):
1407         Include TARGET_DEFAULT_FPU_ISA and TARGET_DEFAULT_FPU_FMA.
1408         * config/nds32/constants.md (unspec_element): Add UNSPEC_COPYSIGN,
1409         UNSPEC_FCPYNSD, UNSPEC_FCPYNSS, UNSPEC_FCPYSD and UNSPEC_FCPYSS.
1410         * config/nds32/constraints.md: New constraints and checking for hard
1411         float configuration.
1412         * config/nds32/iterators.md: New mode iterator and attribute for hard
1413         float configuration.
1414         * config/nds32/nds32-doubleword.md: Use hard float alternatives and
1415         patterns.
1416         * config/nds32/nds32-fpu.md: New file.
1417         * config/nds32/nds32-md-auxiliary.c: New functions and checkings to
1418         deal with hard float code generation.
1419         * config/nds32/nds32-opts.h (nds32_arch_type): Add ARCH_V3F and
1420         ARCH_V3S.
1421         (abi_type, float_reg_number): New enum type.
1422         * config/nds32/nds32-predicates.c: New predicates for hard float.
1423         * config/nds32/nds32-protos.h: Declare functions for hard float.
1424         * config/nds32/nds32.c: Implementation for hard float configuration.
1425         * config/nds32/nds32.h: Definitions for hard float configuration.
1426         * config/nds32/nds32.md: Include hard float machine description and
1427         modify patterns for hard float configuration.
1428         * config/nds32/nds32.opt: New options for hard float configuration.
1429         * config/nds32/predicates.md: New predicates for hard float
1430         configuration.
1431
1432 2018-04-06  Kuan-Lin Chen  <kuanlinchentw@gmail.com>
1433
1434         * common/config/nds32/nds32-common.c
1435         (nds32_option_optimization_table): Enable -mreleax-hint by default.
1436
1437 2018-04-05  Jakub Jelinek  <jakub@redhat.com>
1438
1439         PR middle-end/85195
1440         * match.pd (BIT_FIELD_REF CONSTRUCTOR@0 @1 @2): Use view_convert around
1441         CONSTRUCTOR_ELT (ctor, ...)->value.
1442
1443 2018-04-05  Uros Bizjak  <ubizjak@gmail.com>
1444
1445         PR target/85193
1446         * config/i386/i386.md (define_attr "memory"): Handle rotate1 type.
1447
1448 2018-04-05  Tom de Vries  <tom@codesourcery.com>
1449
1450         PR target/85204
1451         * config/nvptx/nvptx.c (nvptx_single): Fix neutering of bb with only
1452         cond jump.
1453
1454 2018-04-05  Shiva Chen  <shiva0217@gmail.com>
1455             Kito Cheng  <kito.cheng@gmail.com>
1456
1457         * config/nds32/constraints.md (U33): Fine-tune checking condition.
1458         * config/nds32/nds32-md-auxiliary.c (nds32_mem_format): Ditto.
1459         * config/nds32/nds32.h (nds32_16bit_address_type): Add
1460         ADDRESS_POST_MODIFY_LO_REG_IMM3U.
1461
1462 2018-04-05  Shiva Chen  <shiva0217@gmail.com>
1463             Kito Cheng  <kito.cheng@gmail.com>
1464
1465         * config/nds32/constraints.md (Ufe): New memory constraint.
1466         * config/nds32/nds32-md-auxiliary.c (nds32_mem_format,
1467         nds32_output_16bit_load): Consider r8 register for lwi45.fe format.
1468         * config/nds32/nds32.c (nds32_print_operand): Output lwi45.fe
1469         operands.
1470         * config/nds32/nds32.h (nds32_16bit_address_type): Add ADDRESS_R8_IMM7U.
1471         * config/nds32/nds32.md (*mov<mode>): Adjust pattern.
1472
1473 2018-04-05  Chung-Ju Wu  <jasonwucj@gmail.com>
1474
1475         * config/nds32/nds32.md: Use optimize_size in the condition for
1476         alu-shift instructions.
1477
1478 2018-04-05  Chung-Ju Wu  <jasonwucj@gmail.com>
1479
1480         * config/nds32/nds32.md (divsi4, udivsi4): New patterns.
1481
1482 2018-04-05  Chung-Ju Wu  <jasonwucj@gmail.com>
1483
1484         * config/nds32/nds32.md (negsi2): Refine pattern.
1485
1486 2018-04-05  Kito Cheng  <kito.cheng@gmail.com>
1487             Chung-Ju Wu  <jasonwucj@gmail.com>
1488
1489         * config/nds32/iterators.md (shift_rotate): New code iterator.
1490         (shift): New code attribute.
1491         * config/nds32/nds32-md-auxiliary.c (nds32_expand_constant): New.
1492         * config/nds32/nds32-protos.h (nds32_expand_constant): Declare.
1493         * config/nds32/nds32.c (nds32_print_operand): Deal with more cases.
1494         * config/nds32/nds32.md (addsi3, *add_srli): Refine implementation for
1495         bit-wise operations.
1496         (andsi3, *andsi3): Ditto.
1497         (iorsi3, *iorsi3, *or_slli, *or_srli): Ditto.
1498         (xorsi3, *xorsi3, *xor_slli, *xor_srli): Ditto.
1499         (<shift>si3, *ashlsi3, *ashrsi3, *lshrsi3, *rotrsi3): Ditto.
1500         * config/nds32/predicates.md (nds32_rimm5u_operand, nds32_and_operand,
1501         nds32_ior_operand, nds32_xor_operand): New predicates.
1502
1503 2018-04-05  Chung-Ju Wu  <jasonwucj@gmail.com>
1504
1505         * config/nds32/nds32.md (add<mode>3, sub<mode>3): Rename to ...
1506         (addsi3, subsi3): ... this.
1507
1508 2018-04-05  Chung-Ju Wu  <jasonwucj@gmail.com>
1509
1510         * config/nds32/nds32.md (*sub_srli, *and_slli): Fine-tune predicator.
1511
1512 2018-04-05  Chung-Ju Wu  <jasonwucj@gmail.com>
1513
1514         * config/nds32/nds32.md: Adjust indention.
1515
1516 2018-04-05  Kito Cheng  <kito.cheng@gmail.com>
1517
1518         * config/nds32/nds32.md (feature): New attribute.
1519
1520 2018-04-05  Chung-Ju Wu  <jasonwucj@gmail.com>
1521
1522         * config/nds32/nds32.md (subtype): New attribute.
1523
1524 2018-04-04  Thomas Preud'homme  <thomas.preudhomme@arm.com>
1525
1526         PR target/85203
1527         * config/arm/arm-builtins.c (arm_expand_builtin): Change
1528         expansion to perform a bitwise AND of the argument followed by a
1529         boolean negation of the result.
1530
1531 2018-04-04  Peter Bergner  <bergner@vnet.ibm.com>
1532
1533         PR rtl-optimization/84878
1534         * ddg.c (add_cross_iteration_register_deps): Use DF_REF_BB to determine
1535         the basic block.  Assert the use reference is not artificial and that
1536         it has an associated insn.
1537
1538 2018-04-04  Michael Matz  <matz@suse.de>
1539
1540         * builtins.c (compute_objsize): Pass correct operand
1541         to array_at_struct_end_p.
1542
1543 2018-04-04  Richard Biener  <rguenther@suse.de>
1544
1545         PR lto/85176
1546         * dwarf2out.c (dwarf2out_register_external_die): Peel namespaces
1547         from contexts for DINFO_LEVEL_TERSE and below.
1548
1549 2018-04-04  Kito Cheng  <kito.cheng@gmail.com>
1550
1551         * config/nds32/nds32-doubleword.md (move_<mode>): Require
1552         resiter_operand condition.
1553         * config/nds32/nds32.md (*move<mode>): Ditto.
1554
1555 2018-04-04  Kito Cheng  <kito.cheng@gmail.com>
1556             Monk Chiang  <sh.chiang04@gmail.com>
1557
1558         * config/nds32/nds32.md (movmisalign<mode>): New pattern.
1559
1560 2018-04-04  Chung-Ju Wu  <jasonwucj@gmail.com>
1561
1562         * config/nds32/nds32.md (movqi, movhi): Merge into mov<mode>.
1563
1564 2018-04-04  Chung-Ju Wu  <jasonwucj@gmail.com>
1565             Kito Cheng  <kito.cheng@gmail.com>
1566
1567         * config/nds32/nds32-md-auxiliary.c (nds32_inverse_cond_code,
1568         nds32_cond_code_str, output_cond_branch,
1569         output_cond_branch_compare_zero, nds32_expand_cbranch,
1570         nds32_expand_cstore, nds32_expand_movcc,
1571         nds32_output_cbranchsi4_equality_zero,
1572         nds32_output_cbranchsi4_equality_reg,
1573         nds32_output_cbranchsi4_equality_reg_or_const_int,
1574         nds32_output_cbranchsi4_greater_less_zero: New functions.
1575         * config/nds32/nds32-protos.h (nds32_expand_cbranch,
1576         nds32_expand_cstore, nds32_expand_movcc,
1577         nds32_output_cbranchsi4_equality_zero,
1578         nds32_output_cbranchsi4_equality_reg,
1579         nds32_output_cbranchsi4_equality_reg_or_const_int,
1580         nds32_output_cbranchsi4_greater_less_zero): Declare.
1581         * config/nds32/predicates.md (nds32_movecc_comparison_operator,
1582         nds32_rimm11s_operand): New predicates.
1583         * config/nds32/nds32.h (nds32_expand_result_type): New enum type.
1584         * config/nds32/nds32.md: Rewrite all the branch and conditional move
1585         patterns.
1586
1587 2018-04-04  Kito Cheng  <kito.cheng@gmail.com>
1588
1589         * config/nds32/nds32-doubleword.md: Refine all the instruction type.
1590         * config/nds32/nds32.md: Ditto.
1591         * config/nds32/pipelines.md: Ditto.
1592
1593 2018-04-04  Richard Biener  <rguenther@suse.de>
1594
1595         PR tree-optimization/85168
1596         * tree-ssa-sccvn.c (vn_reference_maybe_forwprop_address): Avoid
1597         propagating abnormals.
1598
1599 2018-04-04  Chung-Ju Wu  <jasonwucj@gmail.com>
1600
1601         * config/nds32/nds32.md (enabled): Use yes/no for this attribute.
1602
1603 2018-04-04  Chung-Ju Wu  <jasonwucj@gmail.com>
1604             Kito Cheng  <kito.cheng@gmail.com>
1605
1606         * config/nds32/nds32-md-auxiliary.c (nds32_long_call_p): New function.
1607         * config/nds32/nds32-protos.h (nds32_long_call_p): Declare.
1608         * config/nds32/nds32.c (nds32_function_ok_for_sibcall): New function.
1609         (TARGET_FUNCTION_OK_FOR_SIBCALL): Define.
1610         * config/nds32/nds32.md (sibcall_internal): New.
1611         (sibcall_register): Remove.
1612         (sibcall_immediate): Remove.
1613         (sibcall_value_internal): New.
1614         (sibcall_value_register): Remove.
1615         (sibcall_value_immediate): Remove.
1616         * config/nds32/predicates.md (nds32_general_register_operand): New.
1617         (nds32_call_address_operand): New.
1618
1619 2018-04-03  Jakub Jelinek  <jakub@redhat.com>
1620
1621         PR rtl-optimization/85167
1622         * shrink-wrap.c (move_insn_for_shrink_wrap): Don't set bb_uses and
1623         bb_defs if *split_p, instead preinitialize it to NULL.
1624
1625         PR tree-optimization/85156
1626         * builtins.c (fold_builtin_expect): Use save_expr on arg1 to avoid
1627         evaluating the argument multiple times.
1628
1629 2018-04-03  Bill Schmidt  <wschmidt@linux.ibm.com>
1630
1631         * config/rs6000/emmintrin.h (_mm_cvtpd_epi32): Use __vector rather
1632         than vector.
1633         (_mm_cvtpd_ps): Likewise.
1634         (_mm_cvttpd_epi32): Likewise.
1635         * config/rs6000/mmintrin.h (_mm_unpacklo_pi8): Likewise.
1636         * config/rs6000/xmmintrin.h: For strict-ANSI C++ or C11, undefine
1637         vector, pixel, and bool following altivec.h include.
1638
1639 2018-04-03  Martin Sebor  <msebor@redhat.com>
1640
1641         * doc/extend.texi (Common Function Attributes): Clarify.
1642         (const attribute): Likewise.
1643         (pure attribute): Likewise.
1644
1645 2018-04-03  Jakub Jelinek  <jakub@redhat.com>
1646
1647         PR target/85169
1648         * config/i386/i386.c (ix86_expand_vector_set): Use
1649         HOST_WIDE_INT_1U << elt instead of 1 << elt.  Formatting fix.
1650
1651 2018-04-03  Uros Bizjak  <ubizjak@gmail.com>
1652
1653         * config/i386/i386.c (emit_i387_cw_initialization): Always use logic
1654         instructions when changing rounding bits to preserve precision bits
1655         in the x87 control word.
1656
1657 2018-04-03  Martin Liska  <mliska@suse.cz>
1658
1659         PR tree-optimization/82491
1660         * rtl.h (strip_offset_and_add): Replace += suboffset with
1661         poly_uint64 () + suboffset.
1662
1663 2018-03-29  Martin Liska  <mliska@suse.cz>
1664             Martin Jambor  <mjambor@suse.cz>
1665
1666         PR ipa/84947
1667         * ipa-cp.c (propagate_bits_across_jump_function): Bail out if
1668         param_type is not an integral or pointer type.
1669
1670 2018-04-03  Richard Biener  <rguenther@suse.de>
1671
1672         * sese.h (recompute_all_dominators): Remove.
1673
1674 2018-04-02  Martin Sebor  <msebor@redhat.com>
1675
1676         * doc/invoke.texi (-Wrestrict): Fix typos.
1677
1678 2018-04-02  Jim Wilson  <jimw@sifive.com>
1679
1680         * config/riscv/riscv.h (SHIFT_COUNT_TRUNCATED): Set to zero.
1681         * config/riscv/riscv.md (<optab>si3): Use QImode shift count.
1682         (<optab>di3, <optab>si3_extend): Likewise.
1683         (<optab>si3_mask, <optab>si3_mask_1): New.
1684         (<optab>di3_mask, <optab>di3_mask_1): New.
1685         (<optab>si3_extend_mask, <optab>si3_extend_mask_1): New.
1686         (lshrsi3_zero_extend_1): Use VOIDmode shift count.
1687         * config/riscv/sync.md (atomic_test_and_set): Emit QImode shift count.
1688
1689 2018-04-02  Gerald Pfeifer  <gerald@pfeifer.com>
1690
1691         * doc/cpp.texi (Variadic Macros): Fix line continuation in an
1692         example.
1693
1694 2018-04-02  Chung-Ju Wu  <jasonwucj@gmail.com>
1695
1696         * config/nds32/nds32.c (TARGET_CANONICALIZE_COMPARISON): Define.
1697         (nds32_canonicalize_comparison): New function.
1698
1699 2018-04-01  Chung-Ju Wu  <jasonwucj@gmail.com>
1700             Kito Cheng  <kito.cheng@gmail.com>
1701             Kuan-Lin Chen  <kuanlinchentw@gmail.com>
1702
1703         * config.gcc (nds32): Add nds32-relax-opt.o into extra_objs.
1704         * config/nds32/constants.md (unspec_volatile_element): Add
1705         UNSPEC_VOLATILE_RELAX_GROUP.
1706         * config/nds32/nds32-relax-opt.c: New file.
1707         * config/nds32/nds32-predicates.c
1708         (nds32_symbol_load_store_p): New function.
1709         * config/nds32/nds32-protos.h
1710         (nds32_symbol_load_store_p): Declare function.
1711         (make_pass_nds32_relax_opt): Declare new rtl pass function.
1712         * config/nds32/nds32.c
1713         (nds32_register_pass): New function to register pass.
1714         (nds32_register_passes): New function to register passes.
1715         * config/nds32/nds32.md (relax_group): New pattern.
1716         * config/nds32/nds32.opt (mrelax-hint): New option.
1717         * config/nds32/t-nds32 (nds32-relax-opt.o): New dependency.
1718
1719 2018-04-01  Kito Cheng  <kito.cheng@gmail.com>
1720
1721         * config/nds32/t-nds32: Modify files dependency.
1722
1723 2018-04-01  Chung-Ju Wu  <jasonwucj@gmail.com>
1724
1725         * config/nds32/nds32.h (FUNCTION_PROFILER): Output newline character.
1726         (PROFILE_HOOK): Define its implementation.
1727
1728 2018-04-01  Chung-Ju Wu  <jasonwucj@gmail.com>
1729
1730         * config/nds32/nds32.h (WCHAR_TYPE, WCHAR_TYPE_SIZE): Use unsigned int
1731         type and 32-bit size.
1732
1733 2018-04-01  Jakub Jelinek  <jakub@redhat.com>
1734
1735         PR middle-end/85090
1736         * config/i386/sse.md (V): Add V64QI and V32HI for TARGET_AVX512F.
1737         (V_128_256): New mode iterator.
1738         (*avx512dq_vextract<shuffletype>64x2_1 splitter): New define_split.
1739         (*avx512f_vextract<shuffletype>32x4_1 splitter): Likewise.
1740         (xop_pcmov_<mode><avxsizesuffix>): Use V_128_256 mode iterator instead
1741         of V.
1742         * config/i386/i386.c (ix86_expand_vector_set): Improve V32HImode and
1743         V64QImode expansion for !TARGET_AVX512BW && TARGET_AVX512F.
1744
1745 2018-03-31  Segher Boessenkool  <segher@kernel.crashing.org>
1746
1747         PR target/83315
1748         * config/rs6000/xmmintrin.h (_mm_set_ps, _mm_max_ps): Handle (quiet)
1749         NaN inputs correctly.
1750
1751 2018-03-30  Peter Bergner  <bergner@vnet.ibm.com>
1752
1753         PR target/80546
1754         * config/rs6000/vsx.md (??r): New mode attribute.
1755         (*vsx_mov<mode>_64bit): Use it.
1756         (*vsx_mov<mode>_32bit): Likewise.
1757
1758 2018-03-30  Martin Sebor  <msebor@redhat.com>
1759
1760         PR tree-optimization/84818
1761         * builtins.c (check_access): Use warning_n.
1762
1763 2018-03-30  Aaron Sawdey  <acsawdey@linux.vnet.ibm.com>
1764
1765         PR target/83822
1766         * config/rs6000/rs6000-string.c (expand_compare_loop): Fix redundant
1767         condition.
1768         * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): Fix redundant
1769         condition.
1770
1771 2018-03-30  Julia Koval  <julia.koval@intel.com>
1772
1773         PR target/84413
1774         * x86-tune.def (movx, partial_reg_dependency): Enable for
1775         m_SKYLAKE_AVX512.
1776
1777 2018-03-29  Vladimir Makarov  <vmakarov@redhat.com>
1778
1779         PR inline-asm/84985
1780         * lra-constraints.c (process_alt_operands): Move setting
1781         this_alternative_matches below.
1782
1783 2018-03-29  Martin Liska  <mliska@suse.cz>
1784
1785         PR lto/84995.
1786         * doc/invoke.texi: Document how LTO works with debug info.
1787         Describe auto-load support of binutils.  Mention 'x86-64'
1788         as valid option value of -march option.
1789
1790 2018-03-29  Jakub Jelinek  <jakub@redhat.com>
1791
1792         * config/i386/sse.md (<avx512>_blendm<mode>): Use <sseintprefix>.
1793
1794         PR c/85094
1795         * fold-const.c (operand_equal_p): Handle DEBUG_BEGIN_STMT.
1796         For STATEMENT_LIST, pass down OEP_LEXICOGRAPHIC and maybe
1797         OEP_NO_HASH_CHECK for recursive call, to avoid exponential
1798         checking.
1799
1800 2018-03-28  Peter Bergner  <bergner@vnet.ibm.com>
1801
1802         PR target/84912
1803         * config/rs6000/rs6000.h (RS6000_BTM_POWERPC64): New define.
1804         (RS6000_BTM_COMMON): Add RS6000_BTM_POWERPC64.
1805         * config/rs6000/rs6000.c (rs6000_builtin_mask_calculate): Add support
1806         for RS6000_BTM_POWERPC64.
1807         (rs6000_invalid_builtin): Add handling for RS6000_BTM_POWERPC64
1808         (rs6000_builtin_mask_names): Add RS6000_BTM_POWERPC64.
1809         * config/rs6000/rs6000-builtin.def (BU_P7_POWERPC64_MISC_2): New macro
1810         definition.
1811         (DIVDE): Use it.
1812         (DIVDEU): Likewise.
1813
1814 2018-03-28 Carl Love  <cel@us.ibm.com>
1815
1816         Revert
1817         2017-09-27  Carl Love  <cel@us.ibm.com>
1818
1819         * config/rs6000/rs6000-builtin.def (BU_FP_1MISC_1): Add define macro.
1820         (FCTID, FCTIW): Add BU_FP_MISC_1 macro expansion for builtins.
1821         * config/rs6000/rs6000.md (lrintsfsi2): Add define_insn for the
1822         fctiw instruction.
1823
1824 2018-03-28  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
1825
1826         * config/rs6000/xmmintrin.h (_mm_max_pi16): Use __vector __bool
1827         instead of __vector bool.
1828         (_mm_max_pu8): Likewise.
1829         (_mm_min_pi16): Likewise.
1830
1831 2018-03-28  Peter Bergner  <bergner@vnet.ibm.com>
1832
1833         PR target/84912
1834         * config/rs6000/rs6000-builtin.def (DIVWEO): Delete macro expansion.
1835         (DIVWEUO): Likewise.
1836         (DIVDEO): Likewise.
1837         (DIVDEUO): Likewise.
1838         * config/rs6000/rs6000.c (builtin_function_type): Remove support for
1839         DIVWEUO and DIVDEUO.
1840         * config/rs6000/rs6000.md (UNSPEC_DIVEO, UNSPEC_DIVEUO): Delete unspecs.
1841         (UNSPEC_DIV_EXTEND): Remove deleted unspecs.
1842         (div_extend): Likewise.
1843         * doc/extend.texi (__builtin_divweo): Remove documentation for deleted
1844         builtin function.
1845         (__builtin_divweuo): Likewise.
1846         (__builtin_divdeo): Likewise.
1847         (__builtin_divdeuo): Likewise.
1848
1849 2018-03-28  Jakub Jelinek  <jakub@redhat.com>
1850
1851         PR target/85095
1852         * config/i386/i386.md (*add<mode>3_carry_0, *addsi3_carry_zext_0,
1853         *sub<mode>3_carry_0, *subsi3_carry_zext_0): New patterns.
1854
1855         PR tree-optimization/82004
1856         * gimple-match-head.c (optimize_pow_to_exp): New function.
1857         * match.pd (pow(C,x) -> exp(log(C)*x)): Wrap with #if GIMPLE.
1858         Don't fold to exp if optimize_pow_to_exp is false.
1859
1860 2018-03-28  Martin Liska  <mliska@suse.cz>
1861
1862         PR other/84819
1863         * calls.c (initialize_argument_information): Fix trailing space.
1864         * common.opt: Fix typo and provide better explanation for
1865         -fsanitize-coverage option.
1866         * config/i386/i386.opt: Fix typo.
1867
1868 2018-03-28  Jakub Jelinek  <jakub@redhat.com>
1869             Martin Liska  <mliska@suse.cz>
1870
1871         PR sanitizer/85081
1872         * gimplify.c (asan_poison_variable): Don't do the check for
1873         gimplify_omp_ctxp here.
1874         (gimplify_decl_expr): Do it here.
1875         (gimplify_target_expr): Likewise.
1876
1877 2018-03-28  Martin Liska  <mliska@suse.cz>
1878
1879         PR target/84988
1880         * config/i386/i386.c (ix86_function_arg_advance): Do not call
1881         chkp_type_bounds_count if MPX is not enabled.
1882
1883 2018-03-27  Chung-Ju Wu  <jasonwucj@gmail.com>
1884
1885         * config/nds32/nds32.h (BRANCH_COST): Adjust cost.
1886
1887 2018-03-27  Michael Meissner  <meissner@linux.vnet.ibm.com>
1888
1889         PR target/84914
1890         * config/rs6000/rs6000.c (create_complex_muldiv): New helper
1891         function to create the function decl for complex long double
1892         multiply and divide for -mabi=ieeelongdouble.
1893         (init_float128_ieee): Call it.
1894
1895 2018-03-27  H.J. Lu  <hongjiu.lu@intel.com>
1896
1897         PR target/85044
1898         * config/i386/i386.c (ix86_trampoline_init): Insert ENDBR for
1899         -fcf-protection=branch -mibt.
1900         * config/i386/i386.h (TRAMPOLINE_SIZE): Increased by 4 bytes.
1901
1902 2018-03-27  Ramana Radhakrishnan  <ramana.radhakrishnan@arm.com>
1903
1904         PR target/81863
1905         * config/arm/arm.c (arm_valid_symbolic_address): Handle arm_word_relocations
1906
1907 2018-03-27  Cesar Philippidis  <cesar@codesourcery.com>
1908
1909         PR target/85056
1910         * config/nvptx/nvptx.c (nvptx_assemble_decl_begin): Add '[]' to
1911         extern array declarations.
1912
1913 2018-03-27  Richard Biener  <rguenther@suse.de>
1914
1915         PR middle-end/84067
1916         * match.pd ((A * C) +- (B * C) -> (A+-B) * C): Guard with
1917         explicit single_use checks.
1918
1919 2018-03-27  Richard Biener  <rguenther@suse.de>
1920
1921         PR tree-optimization/85082
1922         * tree-ssa-sccvn.c (vn_reference_lookup_or_insert_for_pieces):
1923         Valueize the VUSE.
1924
1925 2018-03-27  Ramana Radhakrishnan  <ramana.radhakrishnan@arm.com>
1926
1927         * config.gcc (aarch64*-*-linux*): New TARGET_DEFAULT_ASYNC_UNWIND_TABLES
1928         * common/config/aarch64/aarch64-common.c (aarch64_optimization_table[]):
1929         Turn on fasynchronous-unwind-tables and funwind-tables.
1930
1931 2018-03-26  Uros Bizjak  <ubizjak@gmail.com>
1932
1933         PR target/85073
1934         * config/i386/i386.md (*bmi_blsr_<mode>_cmp): New insn pattern.
1935         (*bmi_blsr_<mode>_ccz): Ditto.
1936
1937 2018-03-26  Tom de Vries  <tom@codesourcery.com>
1938
1939         PR tree-optimization/85063
1940         * omp-general.c (offloading_function_p): New function.  Factor out
1941         of ...
1942         * omp-offload.c (pass_omp_target_link::gate): ... here.
1943         * omp-general.h (offloading_function_p): Declare.
1944         * tree-switch-conversion.c (build_one_array): Mark CSWTCH.x variable
1945         with attribute omp declare target for offloading functions.
1946
1947 2018-03-24  Richard Sandiford  <richard.sandiford@linaro.org>
1948
1949         PR tree-optimization/84005
1950         * tree-data-ref.h (get_base_for_alignment): Declare.
1951         * tree-data-ref.c (get_base_for_alignment_1): New function.
1952         (get_base_for_alignment): Likewise.
1953         * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Use
1954         get_base_for_alignment to find a suitable base object, instead
1955         of always using drb->base_address.
1956
1957 2018-03-23  Jakub Jelinek  <jakub@redhat.com>
1958
1959         PR inline-asm/85022
1960         * emit-rtl.c (init_emit_regs): Indicate that VOIDmode MEMs don't have
1961         known size by default.
1962
1963 2018-03-23  Vladimir Makarov  <vmakarov@redhat.com>
1964
1965         PR inline-asm/85030
1966         * lra-constraints.c (process_alt_operands): Don't match BLKmode
1967         and non BLKmode operands.
1968
1969 2018-03-23  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
1970
1971         PR target/85026
1972         * config/arm/arm.md (unaligned_loadhis): Remove first alternative.
1973         Clean up attributes.
1974
1975 2018-03-23  Richard Biener  <rguenther@suse.de>
1976
1977         PR debug/85020
1978         * dwarf2out.c (rtl_for_decl_location): Do not generate RTL early when
1979         we are going to emit early debug for LTO.
1980
1981 2018-03-23  Jakub Jelinek  <jakub@redhat.com>
1982
1983         PR inline-asm/85034
1984         * function.c (match_asm_constraints_1): Don't optimize if input
1985         doesn't satisfy general_operand predicate for output's mode.
1986
1987         PR inline-asm/85022
1988         * alias.c (write_dependence_p): Don't require for x_canonicalized
1989         non-VOIDmode if x has VOIDmode.
1990
1991         PR sanitizer/85029
1992         * sanopt.c (maybe_optimize_ubsan_ptr_ifn): If DECL_REGISTER (base),
1993         just don't try to optimize it rather than assert it never happens.
1994
1995 2018-03-22  Kelvin Nilsen  <kelvin@gcc.gnu.org>
1996
1997         * config/rs6000/rs6000-builtin.def: Remove various BU_ALTIVEC_X
1998         macro expansions for definition of ST_INTERNAL_<mode> and
1999         LD_INTERNAL_<mode> builtins.
2000         * config/rs6000/rs6000-protos.h (rs6000_address_for_altivec):
2001         Remove prototype.
2002         * config/rs6000/rs6000.c (altivec_expand_ld_builtin): Delete this
2003         function.
2004         (altivec_expand_st_builtin): Likewise.
2005         (altivec_expand_builtin): Remove calls to deleted functions.
2006         (rs6000_address_for_altivec): Delete this function.
2007         * config/rs6000/vector.md: Remove expands for
2008         vector_altivec_load_<mode> and vector_altivec_store_<mode>.
2009
2010 2018-03-22  Sudakshina Das  <sudi.das@arm.com>
2011
2012         PR target/84826
2013         * config/arm/arm.h (machine_function): Add static_chain_stack_bytes.
2014         * config/arm/arm.c (arm_compute_static_chain_stack_bytes): Avoid
2015         re-computing once computed.
2016         (arm_expand_prologue): Compute machine->static_chain_stack_bytes.
2017         (arm_init_machine_status): Initialize
2018         machine->static_chain_stack_bytes.
2019
2020 2018-03-22  Kelvin Nilsen  <kelvin@gcc.gnu.org>
2021
2022         PR target/84760
2023         * doc/extend.texi: Add four new prototypes for vec_ld.
2024         * config/rs6000/rs6000-builtin.def (LVX_V1TI): Reorder symbol
2025         definitions for more logical presentation.
2026         * config/rs6000/rs6000-c.c: (altivec_overloaded_builtins): Add
2027         entries for V1TI variants of __builtin_altivec_ld builtin.
2028         * config/rs6000/rs6000.c: (altivec_expand_lv_builtin): Add test and
2029         handling of V1TI variant of LVX icode pattern.
2030         (altivec_expand_builtin): Add case for ALTIVEC_BUILTIN_LVX_V1TI.
2031         (rs6000_gimple_fold_builtin): Likewise.
2032         (altivec_init_builtins): Add code to define
2033         __builtin_altivec_lvx_v1ti function.
2034
2035 2018-03-22  Jakub Jelinek  <jakub@redhat.com>
2036
2037         PR inline-asm/84941
2038         * function.c (match_asm_constraints_1): Don't do the optimization
2039         if input isn't a REG, SUBREG, MEM or constant.
2040
2041 2018-03-22  Tom de Vries  <tom@codesourcery.com>
2042
2043         PR tree-optimization/84956
2044         * tree-ssa-tail-merge.c (find_clusters_1): Skip bbs with
2045         bb_has_abnormal_pred.
2046
2047 2018-03-22  Jakub Jelinek  <jakub@redhat.com>
2048
2049         PR sanitizer/85018
2050         * dwarf2asm.c (dw2_output_indirect_constant_1): Set
2051         DECL_INITIAL (decl) to decl at the end.
2052         * varasm.c (use_blocks_for_decl_p): Revert the 2018-03-20 change,
2053         adjust the comment.
2054
2055 2018-03-21  Joseph Myers  <joseph@codesourcery.com>
2056
2057         * doc/extend.texi (__builtin_tgmath): Document when complex
2058         integer types are treated as _Complex _Float64.
2059
2060 2018-03-21  Tom de Vries  <tom@codesourcery.com>
2061
2062         * doc/extend.texi (__builtin_extend_pointer): Remove pasto.
2063
2064 2018-03-21  Jakub Jelinek  <jakub@redhat.com>
2065
2066         PR tree-optimization/84960
2067         * tree-cfg.c (remove_bb): Don't move forced labels into bb->prev_bb
2068         if it is ENTRY block, move them into single succ of ENTRY in that case.
2069
2070 2018-03-21  Richard Sandiford  <richard.sandiford@linaro.org>
2071
2072         PR tree-optimization/84811
2073         * poly-int.h (poly_span_traits): Remove the T3 parameter and
2074         promote HOST_WIDE_INT T2 - T1 results to unsigned HOST_WIDE_INT.
2075         (maybe_in_range_p, known_in_range_p, ranges_known_overlap_p):
2076         (known_subrange_p): Update accordingly.  Cast each value involved
2077         in the size comparison, rather than casting the result of the
2078         subtraction.
2079
2080 2018-03-21  Jakub Jelinek  <jakub@redhat.com>
2081
2082         PR tree-optimization/84982
2083         * gimple-ssa-store-merging.c (invert_op): Handle boolean inversion
2084         by flipping the least significant bit rather than all bits from
2085         bitpos to bitpos + bitsize - 1.
2086
2087 2018-03-21  Nathan Sidwell  <nathan@acm.org>
2088
2089         * doc/extend.texi (Deprecated Features): Remove mention of
2090         long-deleted deprecations.
2091
2092 2018-03-21  Rainer Orth  <ro@CeBiTec.Uni-Bielefeld.DE>
2093
2094         PR jit/84288
2095         * configure.ac (gcc_cv_ld_soname) <*-*-solaris2*>: Set.
2096         * configure: Regenerate.
2097
2098 2018-03-21  Tom de Vries  <tom@codesourcery.com>
2099
2100         PR tree-optimization/83126
2101         * tree-parloops.c (num_phis): New function.
2102         (gen_parallel_loop): Detect and handle canonicalize_loop_ivs failure.
2103
2104 2018-03-21  Nathan Sidwell  <nathan@acm.org>
2105
2106         * doc/extend.texi (Deprecated Features): Update deprecated flags,
2107         mention anon-struct/union members and trailing attributes.
2108
2109 2018-03-21  Bin Cheng  <bin.cheng@arm.com>
2110
2111         PR tree-optimization/84969
2112         * tree-loop-distribution.c (fuse_memset_builtins): Don't reorder
2113         builtin memset partitions if they set different rhs values.
2114
2115 2018-03-21  Jakub Jelinek  <jakub@redhat.com>
2116
2117         PR rtl-optimization/84989
2118         * simplify-rtx.c (simplify_unary_operation_1): Don't try to simplify
2119         VEC_DUPLICATE with scalar result mode.
2120
2121 2018-03-21  Martin Liska  <mliska@suse.cz>
2122
2123         PR ipa/84963
2124         * ipa-icf.c (sem_item_optimizer::fixup_points_to_sets): Remove
2125         not intended return statement.
2126
2127 2018-03-21  Martin Liska  <mliska@suse.cz>
2128
2129         PR target/84988
2130         * tree-chkp.c (CHKP_ARRAY_MAX_CHECK_STEPS): Define a new macro.
2131         (chkp_find_bound_slots_1): Limit number of iterations.
2132
2133 2018-03-20  David H. Gutteridge  <dhgutteridge@sympatico.ca>
2134
2135         PR target/84838
2136         * Minor grammar fixes for x86 options.
2137
2138 2018-03-20  Jakub Jelinek  <jakub@redhat.com>
2139
2140         PR debug/84875
2141         * dce.c (delete_unmarked_insns): Don't remove frame related noop moves
2142         holding REG_CFA_RESTORE notes, instead turn them into a USE.
2143
2144 2018-03-20  Peter Bergner  <bergner@vnet.ibm.com>
2145
2146         PR target/83789
2147         * config/rs6000/altivec.md (altivec_lvx_<mode>_2op): Delete define_insn.
2148         (altivec_lvx_<mode>_1op): Likewise.
2149         (altivec_stvx_<mode>_2op): Likewise.
2150         (altivec_stvx_<mode>_1op): Likewise.
2151         (altivec_lvx_<VM2:mode>): New define_expand.
2152         (altivec_stvx_<VM2:mode>): Likewise.
2153         (altivec_lvx_<VM2:mode>_2op_<P:mptrsize>): New define_insn.
2154         (altivec_lvx_<VM2:mode>_1op_<P:mptrsize>): Likewise.
2155         (altivec_stvx_<VM2:mode>_2op_<P:mptrsize>): Likewise.
2156         (altivec_stvx_<VM2:mode>_1op_<P:mptrsize>): Likewise.
2157         * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Use new expanders.
2158         (rs6000_gen_lvx): Likewise.
2159         * config/rs6000/rs6000.c (altivec_expand_lv_builtin): Likewise.
2160         (altivec_expand_stv_builtin): Likewise.
2161         (altivec_expand_builtin): Likewise.
2162         * config/rs6000/vector.md: Likewise.
2163
2164 2018-03-20  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
2165
2166         PR target/82518
2167         * config/arm/arm.c (arm_array_mode_supported_p): Return false for
2168         BYTES_BIG_ENDIAN.
2169
2170 2018-03-20  Richard Biener  <rguenther@suse.de>
2171
2172         PR target/84986
2173         * config/i386/i386.c (ix86_add_stmt_cost): Only cost
2174         sign-conversions as zero, fall back to standard scalar_stmt
2175         cost for the rest.
2176
2177 2018-03-20  Martin Liska  <mliska@suse.cz>
2178
2179         PR ipa/84825
2180         * predict.c (rebuild_frequencies): Handle case when we have
2181         PROFILE_ABSENT, but flag_guess_branch_prob is false.
2182
2183 2018-03-20  Jakub Jelinek  <jakub@redhat.com>
2184
2185         PR target/84990
2186         * dwarf2asm.c (dw2_output_indirect_constant_1): Temporarily turn off
2187         flag_section_anchors.
2188         * varasm.c (use_blocks_for_decl_p): Remove hack for
2189         dw2_force_const_mem.
2190
2191         PR target/84845
2192         * config/aarch64/aarch64.md (*aarch64_reg_<mode>3_neg_mask2): Rename
2193         to ...
2194         (*aarch64_<optab>_reg_<mode>3_neg_mask2): ... this.  If pseudos can't
2195         be created, use lowpart_subreg of operands[0] rather than operands[0]
2196         itself.
2197         (*aarch64_reg_<mode>3_minus_mask): Rename to ...
2198         (*aarch64_ashl_reg_<mode>3_minus_mask): ... this.
2199         (*aarch64_<optab>_reg_di3_mask2): Use const_int_operand predicate
2200         and n constraint instead of aarch64_shift_imm_di and Usd.
2201         (*aarch64_reg_<optab>_minus<mode>3): Rename to ...
2202         (*aarch64_<optab>_reg_minus<mode>3): ... this.
2203
2204 2018-03-20  Sudakshina Das  <sudi.das@arm.com>
2205
2206         PR target/82989
2207         * config/arm/neon.md (ashldi3_neon): Update ?s for constraints
2208         to favor GPR over NEON registers.
2209         (<shift>di3_neon): Likewise.
2210
2211 2018-03-20  Tom de Vries  <tom@codesourcery.com>
2212
2213         PR target/84952
2214         * config/nvptx/nvptx.c (nvptx_single): Don't neuter bar.sync.
2215         (nvptx_process_pars): Emit bar.sync asap and alap.
2216
2217 2018-03-20  Tom de Vries  <tom@codesourcery.com>
2218
2219         PR target/84954
2220         * config/nvptx/nvptx.c (prevent_branch_around_nothing): Also update
2221         seen_label if seen_label is already set.
2222
2223 2018-03-20  Jakub Jelinek  <jakub@redhat.com>
2224
2225         PR target/84945
2226         * config/i386/i386.c (fold_builtin_cpu): For features above 31
2227         use __cpu_features2 variable instead of __cpu_model.__cpu_features[0].
2228         Use 1U instead of 1.  Formatting fixes.
2229
2230         PR c/84953
2231         * builtins.c (fold_builtin_strpbrk): For strpbrk(x, "") use type
2232         instead of TREE_TYPE (s1) for the return value.
2233
2234 2018-03-19  Jakub Jelinek  <jakub@redhat.com>
2235
2236         PR tree-optimization/84946
2237         * gimple-ssa-store-merging.c (mem_valid_for_store_merging): Compute
2238         bitsize + bitsize in poly_uint64 rather than poly_int64.
2239
2240         PR sanitizer/78651
2241         * dwarf2asm.c: Include fold-const.c.
2242         (dw2_output_indirect_constant_1): Set DECL_INITIAL (decl) to ADDR_EXPR
2243         of decl rather than decl itself.
2244
2245         PR rtl-optimization/84643
2246         * memmodel.h (enum memmodel): Add MEMMODEL_MAX enumerator.
2247
2248 2018-03-19  Maxim Ostapenko  <m.ostapenko@samsung.com>
2249
2250         PR sanitizer/78651
2251         * dwarf2asm.c (dw2_output_indirect_constant_1): Disable ASan before
2252         calling assemble_variable.
2253
2254 2018-03-19  Sudakshina Das  <sudi.das@arm.com>
2255
2256         PR target/81647
2257         * config/aarch64/aarch64-simd.md (vec_cmp<mode><v_int_equiv>): Modify
2258         instructions for UNLT, UNLE, UNGT, UNGE, UNEQ, UNORDERED and ORDERED.
2259
2260 2018-03-19  Jim Wilson  <jimw@sifive.com>
2261
2262         PR bootstrap/84856
2263         * config/riscv/riscv.c (riscv_function_arg_boundary): Use
2264         PREFERRED_STACK_BOUNDARY instead of STACK_BOUNDARY.
2265         (riscv_first_stack_step): Likewise.
2266         (riscv_option_override): Use STACK_BOUNDARY instead of
2267         MIN_STACK_BOUNDARY.
2268         * config/riscv/riscv.h (STACK_BOUNDARY): Renamed from
2269         MIN_STACK_BOUNDARY.
2270         (BIGGEST_ALIGNMENT): Set to 128.
2271         (PREFERRED_STACK_BOUNDARY): Renamed from STACK_BOUNDARY.
2272         (RISCV_STACK_ALIGN): Use PREFERRED_STACK_BOUNDARY instead of
2273         STACK_BOUNDARY.
2274
2275 2018-03-19  Richard Biener  <rguenther@suse.de>
2276
2277         PR tree-optimization/84933
2278         * tree-vrp.c (set_and_canonicalize_value_range): Treat out-of-bound
2279         values as -INF/INF when canonicalizing an ANTI_RANGE to a RANGE.
2280
2281 2018-03-19  Richard Biener  <rguenther@suse.de>
2282
2283         PR tree-optimization/84859
2284         * tree-ssa-phiopt.c (single_trailing_store_in_bb): New function.
2285         (cond_if_else_store_replacement): Perform sinking operation on
2286         single-store BBs regardless of MAX_STORES_TO_SINK setting.
2287         Generalize what a BB with a single eligible store is.
2288
2289 2018-03-19  Richard Biener  <rguenther@suse.de>
2290
2291         PR tree-optimization/84929
2292         * tree-data-ref.c (analyze_siv_subscript_cst_affine): Guard
2293         chrec_is_positive against non-chrec arg.
2294
2295 2018-03-19  Tamar Christina  <tamar.christina@arm.com>
2296
2297         PR target/84711
2298         * config/arm/arm.c (arm_can_change_mode_class): revert r258554.
2299
2300 2018-03-18  Martin Liska  <mliska@suse.cz>
2301
2302         PR rtl-optimization/84635
2303         * regrename.c (build_def_use): Use matches_mode only when
2304         matches >= 0.
2305
2306 2018-03-18  Richard Sandiford  <richard.sandiford@linaro.org>
2307
2308         PR tree-optimization/84913
2309         * tree-vect-loop.c (vectorizable_reduction): Don't try to
2310         vectorize chains of COND_EXPRs.
2311
2312 2018-03-18  Chung-Ju Wu  <jasonwucj@gmail.com>
2313
2314         * config/nds32/nds32.h (MAX_REGS_PER_ADDRESS): Fix the value.
2315
2316 2018-03-18  Chung-Ju Wu  <jasonwucj@gmail.com>
2317
2318         * config/nds32/nds32.h (LOGICAL_OP_NON_SHORT_CIRCUIT): Define.
2319
2320 2018-03-18  Chung-Ju Wu  <jasonwucj@gmail.com>
2321
2322         * config/nds32/nds32.h (CLZ_DEFINED_VALUE_AT_ZERO): Define.
2323
2324 2018-03-17  Chung-Ju Wu  <jasonwucj@gmail.com>
2325             Kito Cheng  <kito.cheng@gmail.com>
2326
2327         * config/nds32/nds32-protos.h (nds32_adjust_reg_alloc_order): Declare.
2328         * config/nds32/nds32.c (nds32_reg_alloc_order_for_speed): New array.
2329         (nds32_adjust_reg_alloc_order): New function.
2330         * config/nds32/nds32.h (ADJUST_REG_ALLOC_ORDER): Define.
2331
2332 2018-03-17  Kito Cheng  <kito.cheng@gmail.com>
2333
2334         * config/nds32/nds32.c (nds32_asm_output_mi_thunk,
2335         nds32_print_operand, nds32_print_operand_address): Use
2336         HOST_WIDE_INT_PRINT_DEC instead.
2337
2338 2018-03-17  Chung-Ju Wu  <jasonwucj@gmail.com>
2339
2340         * config/nds32/nds32.c (nds32_register_priority): Modify cost.
2341
2342 2018-03-17  Jakub Jelinek  <jakub@redhat.com>
2343
2344         PR target/84902
2345         * config/i386/i386.c (initial_ix86_tune_features,
2346         initial_ix86_arch_features): Use unsigned HOST_WIDE_INT rather than
2347         unsigned long long.
2348         (set_ix86_tune_features): Change ix86_tune_mask from unsigned int
2349         to unsigned HOST_WIDE_INT, initialize to HOST_WIDE_INT_1U << ix86_tune
2350         rather than 1u << ix86_tune.  Formatting fix.
2351         (ix86_option_override_internal): Change ix86_arch_mask from
2352         unsigned int to unsigned HOST_WIDE_INT, initialize to
2353         HOST_WIDE_INT_1U << ix86_arch rather than 1u << ix86_arch.
2354         (ix86_function_specific_restore): Likewise.
2355
2356 2018-03-16  Jakub Jelinek  <jakub@redhat.com>
2357
2358         PR target/84899
2359         * postreload.c (reload_combine_recognize_pattern): Perform
2360         INTVAL addition in unsigned HOST_WIDE_INT type to avoid UB and
2361         truncate_int_for_mode the result for the destination's mode.
2362
2363         PR c/84909
2364         * hsa-gen.c (mem_type_for_type): Fix comment typo.
2365         * tree-vect-loop-manip.c (vect_create_cond_for_niters_checks):
2366         Likewise.
2367         * gimple-ssa-warn-restrict.c (builtin_memref::set_base_and_offset):
2368         Likewise.
2369
2370 2018-03-16  Vladimir Makarov  <vmakarov@redhat.com>
2371
2372         PR target/84876
2373         * lra-assigns.c (lra_split_hard_reg_for): Don't use
2374         regno_allocno_class_array and sorted_pseudos.
2375         * lra-constraints.c (spill_hard_reg_in_range): Ignore hard regs in
2376         insns where regno is used.
2377
2378 2018-03-16  Martin Liska  <mliska@suse.cz>
2379
2380         PR ipa/84833
2381         * multiple_target.c (create_dispatcher_calls): Redirect
2382         reference in the symbol table.
2383
2384 2018-03-16  Martin Liska  <mliska@suse.cz>
2385
2386         PR ipa/84722
2387         * multiple_target.c (create_dispatcher_calls): Redirect also
2388         an alias.
2389
2390 2018-03-16  Jakub Jelinek  <jakub@redhat.com>
2391
2392         PR c++/79937
2393         PR c++/82410
2394         * tree.h (TARGET_EXPR_NO_ELIDE): Define.
2395         * gimplify.c (gimplify_modify_expr_rhs): Don't elide TARGET_EXPRs with
2396         TARGET_EXPR_NO_ELIDE flag set unless *expr_p is INIT_EXPR.
2397
2398 2018-03-16  Julia Koval  <julia.koval@intel.com>
2399
2400         * doc/invoke.texi (Skylake Server): Add CLWB.
2401         Cannonlake): Remove CLWB.
2402
2403 2018-03-16  Jakub Jelinek  <jakub@redhat.com>
2404
2405         PR tree-optimization/84841
2406         * tree-ssa-reassoc.c (INTEGER_CONST_TYPE): Change to 1 << 4 from
2407         1 << 3.
2408         (FLOAT_ONE_CONST_TYPE): Define.
2409         (constant_type): Return FLOAT_ONE_CONST_TYPE for -1.0 and 1.0.
2410         (sort_by_operand_rank): Put entries with higher constant_type last
2411         rather than first to match comments.
2412
2413 2018-03-15  Sandra Loosemore  <sandra@codesourcery.com>
2414
2415         * config/nios2/nios2.md (movsi_internal): Fix thinko in
2416         split predicate.
2417
2418 2018-03-15  Jakub Jelinek  <jakub@redhat.com>
2419
2420         PR c++/79085
2421         * calls.c (expand_call): For TREE_ADDRESSABLE rettype ignore alignment
2422         check and use address of target always.
2423
2424 2018-03-15  H.J. Lu  <hongjiu.lu@intel.com>
2425
2426         PR target/84574
2427         * config/i386/i386.c (indirect_thunk_needed): Update comments.
2428         (indirect_thunk_bnd_needed): Likewise.
2429         (indirect_thunks_used): Likewise.
2430         (indirect_thunks_bnd_used): Likewise.
2431         (indirect_return_needed): New.
2432         (indirect_return_bnd_needed): Likewise.
2433         (output_indirect_thunk_function): Add a bool argument for
2434         function return.
2435         (output_indirect_thunk_function): Don't generate alias for
2436         function return thunk.
2437         (ix86_code_end): Call output_indirect_thunk_function to generate
2438         function return thunks.
2439         (ix86_output_function_return): Set indirect_return_bnd_needed
2440         and indirect_return_needed instead of indirect_thunk_bnd_needed
2441         and indirect_thunk_needed.
2442
2443 2018-03-15  Olga Makhotina  <olga.makhotina@intel.com>
2444
2445         * config/i386/sgxintrin.h (_enclv_u32): New intrinsic.
2446         (__enclv_bc, __enclv_cd, __enclv_generic): New definitions.
2447         (ERDINFO, ETRACKC, ELDBC, ELDUC): New leaves.
2448
2449 2018-03-15  David Malcolm  <dmalcolm@redhat.com>
2450             Paul Hua <paul.hua.gm@gmail.com>
2451
2452         PR c/84852
2453         * gcc.dg/fixits-pr84852-1.c: Fix filename in dg-regexp.
2454
2455 2018-03-15  Segher Boessenkool  <segher@kernel.crashing.org>
2456
2457         * config/rs6000/rs6000.c (abi_v4_pass_in_fpr): Add back the
2458         TARGET_DOUBLE_FLOAT and TARGET_SINGLE_FLOAT conditions on the DFmode
2459         resp. SFmode cases.
2460
2461 2018-03-15  Tamar Christina  <tamar.christina@arm.com>
2462
2463         PR target/84711
2464         * config/arm/arm.c (arm_can_change_mode_class): Use GET_MODE_UNIT_SIZE
2465         instead of GET_MODE_SIZE when comparing Units.
2466
2467 2018-03-15  Vladimir Mezentsev  <vladimir.mezentsev@oracle.com>
2468
2469         PR target/68256
2470         * varasm.c (hash_section): Return an unchangeble hash value
2471         * config/aarch64/aarch64.c (aarch64_use_blocks_for_constant_p):
2472         Return !aarch64_can_use_per_function_literal_pools_p ().
2473
2474 2018-03-15  Jakub Jelinek  <jakub@redhat.com>
2475
2476         PR target/84860
2477         * optabs.c (emit_conditional_move): Pass address of cmode's copy
2478         rather than address of cmode as last argument to prepare_cmp_insn.
2479
2480 2018-03-15  Julia Koval  <julia.koval@intel.com>
2481
2482         * config/i386/i386.c (F_AVX512VBMI2, F_GFNI, F_VPCLMULQDQ,
2483         F_AVX512VNNI, F_AVX512BITALG): New.
2484
2485 2018-03-14  John David Anglin  <danglin@gcc.gnu.org>
2486
2487         PR target/83451
2488         * config/pa/pa.c (pa_emit_move_sequence):  Always emit secondary reload
2489         insn for floating-point loads and stores.
2490
2491 2018-03-14  Carl Love  <cel@us.ibm.com>
2492
2493         * config/rs6000/rs6000-c.c: Add macro definitions for
2494         ALTIVEC_BUILTIN_VEC_PERMXOR.
2495         * config/rs6000/rs6000.h: Add #define for vec_permxor builtin.
2496         * config/rs6000/rs6000-builtin.def: Add macro expansions for VPERMXOR.
2497         * config/rs6000/altivec.md (altivec_vpermxor): New define expand.
2498         * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Add case
2499         UNSPEC_VPERMXOR.
2500         * config/doc/extend.texi: Add prototypes for vec_permxor.
2501
2502 2018-03-14  David Malcolm  <dmalcolm@redhat.com>
2503
2504         PR c/84852
2505         * diagnostic-show-locus.c (class layout_point): Convert m_line
2506         from int to linenum_type.
2507         (line_span::comparator): Use linenum "compare" function when
2508         comparing line numbers.
2509         (test_line_span): New function.
2510         (layout_range::contains_point): Convert param "row" from int to
2511         linenum_type.
2512         (layout_range::intersects_line_p): Likewise.
2513         (layout::will_show_line_p): Likewise.
2514         (layout::print_source_line): Likewise.
2515         (layout::should_print_annotation_line_p): Likewise.
2516         (layout::print_annotation_line): Likewise.
2517         (layout::print_leading_fixits): Likewise.
2518         (layout::annotation_line_showed_range_p): Likewise.
2519         (struct line_corrections): Likewise for field m_row.
2520         (line_corrections::line_corrections): Likewise for param "row".
2521         (layout::print_trailing_fixits): Likewise.
2522         (layout::get_state_at_point): Likewise.
2523         (layout::get_x_bound_for_row): Likewise.
2524         (layout::print_line): Likewise.
2525         (diagnostic_show_locus): Likewise for locals "last_line" and
2526         "row".
2527         (selftest::diagnostic_show_locus_c_tests): Call test_line_span.
2528         * input.c (selftest::test_linenum_comparisons): New function.
2529         (selftest::input_c_tests): Call it.
2530         * selftest.c (selftest::test_assertions): Test ASSERT_GT,
2531         ASSERT_GT_AT, ASSERT_LT, and ASSERT_LT_AT.
2532         * selftest.h (ASSERT_GT): New macro.
2533         (ASSERT_GT_AT): New macro.
2534         (ASSERT_LT): New macro.
2535         (ASSERT_LT_AT): New macro.
2536
2537 2018-03-14  Segher Boessenkool  <segher@kernel.crashing.org>
2538
2539         PR rtl-optimization/84780
2540         * combine.c (distribute_links): Don't make a link based on pc_rtx.
2541
2542 2018-03-14  Martin Liska  <mliska@suse.cz>
2543
2544         * tree.c (record_node_allocation_statistics): Use
2545         get_stats_node_kind.
2546         (get_stats_node_kind): New function extracted from
2547         record_node_allocation_statistics.
2548         (free_node): Use get_stats_node_kind.
2549
2550 2018-03-14  Richard Biener  <rguenther@suse.de>
2551
2552         * tree-ssa-pre.c (compute_antic_aux): Remove code that asserts
2553         that the value-set of ANTIC_IN doesn't grow.
2554
2555         Revert
2556         * tree-ssa-pre.c (struct bb_bitmap_sets): Add visited_with_visited_succs
2557         member.
2558         (BB_VISITED_WITH_VISITED_SUCCS): New define.
2559         (compute_antic): Initialize BB_VISITED_WITH_VISITED_SUCCS.
2560
2561 2018-03-14  Julia Koval  <julia.koval@intel.com>
2562
2563         * config.gcc (icelake-client, icelake-server): New.
2564         (icelake): Remove.
2565         * config/i386/i386.c (initial_ix86_tune_features): Extend to 64 bit.
2566         (initial_ix86_arch_features): Ditto.
2567         (PTA_SKYLAKE): Add SGX.
2568         (PTA_ICELAKE): Remove.
2569         (PTA_ICELAKE_CLIENT): New.
2570         (PTA_ICELAKE_SERVER): New.
2571         (ix86_option_override_internal): Split up icelake on icelake client and
2572         icelake server.
2573         (get_builtin_code_for_version): Ditto.
2574         (fold_builtin_cpu): Ditto.
2575         * config/i386/driver-i386.c (config/i386/driver-i386.c): Ditto.
2576         * config/i386/i386-c.c (ix86_target_macros_internal): Ditto
2577         * config/i386/i386.h (processor_type): Ditto.
2578         * doc/invoke.texi: Ditto.
2579
2580 2018-03-14  Jakub Jelinek  <jakub@redhat.com>
2581
2582         PR sanitizer/83392
2583         * sanopt.c (maybe_optimize_ubsan_ptr_ifn): Handle also
2584         INTEGER_CST offset, add it together with bitpos / 8 and
2585         sign extend based on POINTER_SIZE.
2586
2587         PR target/84844
2588         Revert
2589         2017-04-20  Uros Bizjak  <ubizjak@gmail.com>
2590
2591         PR target/78090
2592         * config/i386/constraints.md (Yc): New register constraint.
2593         * config/i386/i386.md (*float<SWI48:mode><MODEF:mode>2_mixed):
2594         Use Yc constraint for alternative 2 of operand 0.  Remove
2595         preferred_for_speed attribute.
2596
2597 2018-03-14  Richard Biener  <rguenther@suse.de>
2598
2599         PR tree-optimization/84830
2600         * tree-ssa-pre.c (compute_antic_aux): Intersect the new ANTIC_IN
2601         with the old one to avoid oscillations.
2602
2603 2018-03-13  Vladimir Makarov  <vmakarov@redhat.com>
2604
2605         PR target/83712
2606         * lra-assigns.c (find_all_spills_for): Ignore uninteresting
2607         pseudos.
2608         (assign_by_spills): Return a flag of reload assignment failure.
2609         Do not process the reload assignment failures.  Do not spill other
2610         reload pseudos if they has the same reg class.  Update n if
2611         necessary.
2612         (lra_assign): Add a return arg.  Set up from the result of
2613         assign_by_spills call.
2614         (find_reload_regno_insns, lra_split_hard_reg_for): New functions.
2615         * lra-constraints.c (split_reg): Add a new arg.  Use it instead of
2616         usage_insns if it is not NULL.
2617         (spill_hard_reg_in_range): New function.
2618         (split_if_necessary, inherit_in_ebb): Pass a new arg to split_reg.
2619         * lra-int.h (spill_hard_reg_in_range, lra_split_hard_reg_for): New
2620         function prototypes.
2621         (lra_assign): Change prototype.
2622         * lra.c (lra): Add code to deal with fails by splitting hard reg
2623         live ranges.
2624
2625 2018-03-01  Palmer Dabbelt  <palmer@sifive.com>
2626
2627         * config/riscv/riscv.opt (mrelax): New option.
2628         * config/riscv/riscv.c (riscv_file_start): Emit ".option
2629         "norelax" when riscv_mrelax is disabled.
2630         * doc/invoke.texi (RISC-V): Document "-mrelax" and "-mno-relax".
2631
2632 2018-03-13  Aaron Sawdey  <acsawdey@linux.vnet.ibm.com>
2633
2634         PR target/84743
2635         * config/rs6000/rs6000.c (rs6000_reassociation_width): Disable parallel
2636         reassociation for int modes.
2637
2638 2018-03-13  Richard Sandiford  <richard.sandiford@linaro.org>
2639
2640         * tree-vect-loop-manip.c (vect_maybe_permute_loop_masks):
2641         Reverse the choice between VEC_UNPACK_LO_EXPR and VEC_UNPACK_HI_EXPR
2642         for big-endian.
2643         * config/aarch64/iterators.md (hi_lanes_optab): New int attribute.
2644         * config/aarch64/aarch64-sve.md
2645         (*aarch64_sve_<perm_insn><perm_hilo><mode>): Rename to...
2646         (aarch64_sve_<perm_insn><perm_hilo><mode>): ...this.
2647         (*extend<mode><Vwide>2): Rename to...
2648         (aarch64_sve_extend<mode><Vwide>2): ...this.
2649         (vec_unpack<su>_<perm_hilo>_<mode>): Turn into a define_expand,
2650         renaming the old pattern to...
2651         (aarch64_sve_punpk<perm_hilo>_<mode>): ...this.  Only define
2652         unsigned packs.
2653         (vec_unpack<su>_<perm_hilo>_<SVE_BHSI:mode>): Turn into a
2654         define_expand, renaming the old pattern to...
2655         (aarch64_sve_<su>unpk<perm_hilo>_<SVE_BHSI:mode>): ...this.
2656         (*vec_unpacku_<perm_hilo>_<mode>_no_convert): Delete.
2657         (vec_unpacks_<perm_hilo>_<mode>): Take BYTES_BIG_ENDIAN into
2658         account when deciding which SVE instruction the optab should use.
2659         (vec_unpack<su_optab>_float_<perm_hilo>_vnx4si): Likewise.
2660
2661 2018-03-13  Richard Sandiford  <richard.sandiford@linaro.org>
2662
2663         * config/aarch64/aarch64.md (V4_REGNUM, V8_REGNUM, V12_REGNUM)
2664         (V20_REGNUM, V24_REGNUM, V28_REGNUM, P1_REGNUM, P2_REGNUM, P3_REGNUM)
2665         (P4_REGNUM, P5_REGNUM, P6_REGNUM, P8_REGNUM, P9_REGNUM, P10_REGNUM)
2666         (P11_REGNUM, P12_REGNUM, P13_REGNUM, P14_REGNUM): New define_constants.
2667         (tlsdesc_small_<mode>): Turn a define_expand and use
2668         tlsdesc_small_sve_<mode> for SVE.  Rename original define_insn to...
2669         (tlsdesc_small_advsimd_<mode>): ...this.
2670         (tlsdesc_small_sve_<mode>): New pattern.
2671
2672 2018-03-13  Richard Sandiford  <richard.sandiford@linaro.org>
2673
2674         * config/aarch64/iterators.md (UNSPEC_SMUL_HIGHPART)
2675         (UNSPEC_UMUL_HIGHPART): New constants.
2676         (MUL_HIGHPART): New int iteraor.
2677         (su): Handle UNSPEC_SMUL_HIGHPART and UNSPEC_UMUL_HIGHPART.
2678         * config/aarch64/aarch64-sve.md (<su>mul<mode>3_highpart): New
2679         define_expand.
2680         (*<su>mul<mode>3_highpart): New define_insn.
2681
2682 2018-03-13  Eric Botcazou  <ebotcazou@adacore.com>
2683
2684         PR lto/84805
2685         * ipa-devirt.c (odr_subtypes_equivalent_p): Do not get the ODR type of
2686         incomplete types.
2687
2688 2018-03-13  Martin Liska  <mliska@suse.cz>
2689
2690         PR ipa/84658.
2691         * (sem_item_optimizer::sem_item_optimizer): Initialize new
2692         vector.
2693         (sem_item_optimizer::~sem_item_optimizer): Release it.
2694         (sem_item_optimizer::merge_classes): Register variable aliases.
2695         (sem_item_optimizer::fixup_pt_set): New function.
2696         (sem_item_optimizer::fixup_points_to_sets): Likewise.
2697         * ipa-icf.h: Declare new variables and functions.
2698
2699 2018-03-13  Jakub Jelinek  <jakub@redhat.com>
2700
2701         PR middle-end/84834
2702         * match.pd ((A & C) != 0 ? D : 0): Use INTEGER_CST@2 instead of
2703         integer_pow2p@2 and test integer_pow2p in condition.
2704         (A < 0 ? C : 0): Similarly for @1.
2705
2706         PR middle-end/84831
2707         * stmt.c (parse_output_constraint): If the CONSTRAINT_LEN (*p, p)
2708         characters starting at p contain '\0' character, don't look beyond
2709         that.
2710
2711         PR target/84827
2712         * config/i386/i386.md (round<mode>2): For 387 fancy math, disable
2713         pattern if -ftrapping-math -fno-fp-int-builtin-inexact.
2714
2715         PR target/84828
2716         * reg-stack.c (change_stack): Change update_end var from int to
2717         rtx_insn *, if non-NULL don't update just BB_END (current_block), but
2718         also call set_block_for_insn on the newly added insns and rescan.
2719
2720         PR target/84786
2721         * config/i386/sse.md (sse2_loadhpd): Use Yv constraint rather than v
2722         on the last operand.
2723
2724         PR c++/84704
2725         * tree.c (stabilize_reference_1): Return save_expr (e) for
2726         STATEMENT_LIST even if it doesn't have side-effects.
2727
2728 2018-03-12  Jonathan Wakely  <jwakely@redhat.com>
2729
2730         * doc/invoke.texi (-mclflushopt): Fix spelling of option.
2731
2732 2018-03-12  Renlin Li  <renlin.li@arm.com>
2733
2734         * config/aarch64/aarch64.md (movhf_aarch64): Fix mode argument to
2735         aarch64_output_scalar_simd_mov_immediate.
2736
2737 2018-03-12  Martin Sebor  <msebor@redhat.com>
2738
2739         PR tree-optimization/83456
2740         * gimple-fold.c (gimple_fold_builtin_memory_op): Avoid warning
2741         for perfectly overlapping calls to memcpy.
2742         (gimple_fold_builtin_memory_chk): Same.
2743         (gimple_fold_builtin_strcpy): Handle no-warning.
2744         (gimple_fold_builtin_stxcpy_chk): Same.
2745         * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Handle no-warning.
2746
2747 2018-03-12  Segher Boessenkool  <segher@kernel.crashing.org>
2748
2749         * config/rs6000/rs6000.c (abi_v4_pass_in_fpr): Add bool "named"
2750         parameter.  Use it for SFmode.
2751         (rs6000_function_arg_advance_1): Adjust.
2752         (rs6000_function_arg): Adjust.
2753         (rs6000_gimplify_va_arg): Pass false for that new parameter.
2754
2755 2018-03-12  Segher Boessenkool  <segher@kernel.crashing.org>
2756
2757         PR rtl-optimization/84169
2758         PR rtl-optimization/84780
2759         * combine.c (can_combine_p): Check for a 2-insn combination whether
2760         the destination register is used between the two insns, too.
2761
2762 2018-03-12  Richard Biener  <rguenther@suse.de>
2763
2764         PR tree-optimization/84803
2765         * tree-if-conv.c (ifcvt_memrefs_wont_trap): Don't do anything
2766         for refs DR analysis didn't process.
2767
2768 2018-03-12  Richard Biener  <rguenther@suse.de>
2769
2770         PR tree-optimization/84777
2771         * tree-ssa-loop-ch.c (should_duplicate_loop_header_p): For
2772         force-vectorize loops ignore whether we are optimizing for size.
2773
2774 2018-03-12  Chung-Ju Wu  <jasonwucj@gmail.com>
2775
2776         * config/nds32/nds32.c (nds32_md_asm_adjust): New function.
2777         (TARGET_MD_ASM_ADJUST): Define.
2778
2779 2018-03-12  Monk Chiang  <sh.chiang04@gmail.com>
2780             Kito Cheng  <kito.cheng@gmail.com>
2781             Chung-Ju Wu  <jasonwucj@gmail.com>
2782
2783         * config/nds32/nds32.c (nds32_compute_stack_frame,
2784         nds32_emit_stack_push_multiple, nds32_emit_stack_pop_multiple,
2785         nds32_emit_stack_v3push, nds32_emit_stack_v3pop,
2786         nds32_emit_adjust_frame, nds32_expand_prologue, nds32_expand_epilogue,
2787         nds32_expand_prologue_v3push, nds32_expand_epilogue_v3pop): Refine.
2788         * config/nds32/nds32.h (NDS32_FIRST_CALLEE_SAVE_GPR_REGNUM,
2789         NDS32_LAST_CALLEE_SAVE_GPR_REGNUM, NDS32_V3PUSH_AVAILABLE_P): New.
2790         * config/nds32/nds32.md (prologue, epilogue): Use macro
2791         NDS32_V3PUSH_AVAILABLE_P to do checking.
2792
2793 2018-03-11  Jakub Jelinek  <jakub@redhat.com>
2794
2795         PR debug/58150
2796         * dwarf2out.c (gen_enumeration_type_die): Don't guard adding
2797         DW_AT_declaration for ENUM_IS_OPAQUE on -gdwarf-4 or -gno-strict-dwarf,
2798         but on TYPE_SIZE.  Don't do anything for ENUM_IS_OPAQUE if not creating
2799         a new die.  Don't set TREE_ASM_WRITTEN if ENUM_IS_OPAQUE.  Guard
2800         addition of most attributes on !orig_type_die or the attribute not
2801         being present already.  Assert TYPE_VALUES is NULL for ENUM_IS_OPAQUE.
2802
2803 2018-03-11  Kito Cheng  <kito.cheng@gmail.com>
2804             Chung-Ju Wu  <jasonwucj@gmail.com>
2805
2806         * config/nds32/nds32.c (nds32_cpu_cpp_builtins): Modify to define
2807         __NDS32_VH__ macro.
2808         * config/nds32/nds32.opt (mvh): New option.
2809
2810 2018-03-11  Kito Cheng  <kito.cheng@gmail.com>
2811             Chung-Ju Wu  <jasonwucj@gmail.com>
2812
2813         * config/nds32/nds32-protos.h (nds32_cpu_cpp_builtins): Declare
2814         function.
2815         * config/nds32/nds32.c (nds32_cpu_cpp_builtins): New function.
2816         * config/nds32/nds32.h (TARGET_CPU_CPP_BUILTINS): Modify its
2817         definition.
2818
2819 2018-03-11  Kito Cheng  <kito.cheng@gmail.com>
2820             Chung-Ju Wu  <jasonwucj@gmail.com>
2821
2822         * config/nds32/nds32-memory-manipulation.c (nds32_expand_strlen): New
2823         function.
2824         * config/nds32/nds32-multiple.md (strlensi): New pattern.
2825         * config/nds32/nds32-protos.h (nds32_expand_strlen): Declare function.
2826
2827 2018-03-11  Monk Chiang  <sh.chiang04@gmail.com>
2828             Kito Cheng  <kito.cheng@gmail.com>
2829             Chung-Ju Wu  <jasonwucj@gmail.com>
2830
2831         * config/nds32/constants.md (unspec_element): Add UNSPEC_FFB,
2832         UNSPEC_FFMISM and UNSPEC_FLMISM.
2833         * config/nds32/nds32-intrinsic.c (bdesc_2arg): Add builtin description
2834         for ffb, ffmism and flmism.
2835         * config/nds32/nds32-intrinsic.md (unspec_ffb): Define new pattern.
2836         (unspec_ffmism): Ditto.
2837         (unspec_flmism): Ditto.
2838         (nds32_expand_builtin_impl): Check if string extension is available.
2839         * config/nds32/nds32.h (nds32_builtins): Add NDS32_BUILTIN_FFB,
2840         NDS32_BUILTIN_FFMISM and NDS32_BUILTIN_FLMISM.
2841
2842 2018-03-10  Vladimir Makarov  <vmakarov@redhat.com>
2843
2844         Reverting patch:
2845         2018-03-09  Vladimir Makarov  <vmakarov@redhat.com>
2846
2847         PR target/83712
2848         * lra-assigns.c (assign_by_spills): Return a flag of reload
2849         assignment failure.  Do not process the reload assignment
2850         failures.  Do not spill other reload pseudos if they has the same
2851         reg class.
2852         (lra_assign): Add a return arg.  Set up from the result of
2853         assign_by_spills call.
2854         (find_reload_regno_insns, lra_split_hard_reg_for): New functions.
2855         * lra-constraints.c (split_reg): Add a new arg.  Use it instead of
2856         usage_insns if it is not NULL.
2857         (spill_hard_reg_in_range): New function.
2858         (split_if_necessary, inherit_in_ebb): Pass a new arg to split_reg.
2859         * lra-int.h (spill_hard_reg_in_range, lra_split_hard_reg_for): New
2860         function prototypes.
2861         (lra_assign): Change prototype.
2862         * lra.c (lra): Add code to deal with fails by splitting hard reg
2863         live ranges.
2864
2865 2018-03-10  H.J. Lu  <hongjiu.lu@intel.com>
2866
2867         PR target/84807
2868         * config/i386/i386.opt: Replace Enforcment with Enforcement.
2869
2870 2018-03-10  Alexandre Oliva  <aoliva@redhat.com>
2871
2872         PR debug/84620
2873         * dwarf2out.h (dw_val_class): Add dw_val_class_symview.
2874         (dw_val_node): Add val_symbolic_view.
2875         * dwarf2out.c (dw_line_info_table): Add symviews_since_reset.
2876         (symview_upper_bound): New.
2877         (new_line_info_table): Initialize symviews_since_reset.
2878         (dwarf2out_source_line): Count symviews_since_reset and set
2879         symview_upper_bound.
2880         (dw_val_equal_p): Handle symview.
2881         (add_AT_symview): New.
2882         (print_dw_val): Handle symview.
2883         (attr_checksum, attr_checksum_ordered): Likewise.
2884         (same_dw_val_p, size_of_die): Likewise.
2885         (value_format, output_die): Likewise.
2886         (add_high_low_attributes): Use add_AT_symview for entry_view.
2887         (dwarf2out_finish): Reset symview_upper_bound, clear
2888         zero_view_p.
2889
2890 2018-03-09  Peter Bergner  <bergner@vnet.ibm.com>
2891
2892         PR target/83969
2893         * config/rs6000/rs6000.c (rs6000_offsettable_memref_p): New prototype.
2894         Add strict argument and use it.
2895         (rs6000_split_multireg_move): Update for new strict argument.
2896         (mem_operand_gpr): Disallow all non-offsettable addresses.
2897         * config/rs6000/rs6000.md (*movdi_internal64): Use YZ constraint.
2898
2899 2018-03-09  Jakub Jelinek  <jakub@redhat.com>
2900
2901         PR target/84772
2902         * config/rs6000/rs6000.c (rs6000_gimplify_va_arg): Mark va_arg_tmp
2903         temporary TREE_ADDRESSABLE before gimplification of BUILT_IN_MEMCPY.
2904         * config/powerpcspe/powerpcspe.c (rs6000_gimplify_va_arg): Likewise.
2905
2906         PR c++/84767
2907         * tree-inline.c (copy_tree_body_r): For INDIRECT_REF of a remapped
2908         decl, use remap_type if we want to use the type.
2909
2910 2018-03-09  Martin Sebor  <msebor@redhat.com>
2911
2912         PR tree-optimization/84526
2913         * gimple-ssa-warn-restrict.c (builtin_memref::set_base_and_offset):
2914         Remove dead code.
2915         (builtin_access::generic_overlap): Be prepared to handle non-array
2916         base objects.
2917
2918 2018-03-09  Alexandre Oliva  <aoliva@redhat.com>
2919
2920         PR rtl-optimization/84682
2921         * lra-constraints.c (process_address_1): Check is_address flag
2922         for address constraints.
2923         (process_alt_operands): Likewise.
2924         * lra.c (lra_set_insn_recog_data): Pass asm operand locs to
2925         preprocess_constraints.
2926         * recog.h (preprocess_constraints): Add oploc parameter.
2927         Adjust callers.
2928         * recog.c (preprocess_constraints): Test address_operand for
2929         CT_ADDRESS constraints.
2930
2931 2018-03-09  Vladimir Makarov  <vmakarov@redhat.com>
2932
2933         PR target/83712
2934         * lra-assigns.c (assign_by_spills): Return a flag of reload
2935         assignment failure.  Do not process the reload assignment
2936         failures.  Do not spill other reload pseudos if they has the same
2937         reg class.
2938         (lra_assign): Add a return arg.  Set up from the result of
2939         assign_by_spills call.
2940         (find_reload_regno_insns, lra_split_hard_reg_for): New functions.
2941         * lra-constraints.c (split_reg): Add a new arg.  Use it instead of
2942         usage_insns if it is not NULL.
2943         (spill_hard_reg_in_range): New function.
2944         (split_if_necessary, inherit_in_ebb): Pass a new arg to split_reg.
2945         * lra-int.h (spill_hard_reg_in_range, lra_split_hard_reg_for): New
2946         function prototypes.
2947         (lra_assign): Change prototype.
2948         * lra.c (lra): Add code to deal with fails by splitting hard reg
2949         live ranges.
2950
2951 2018-03-09  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
2952
2953         PR target/83193
2954         * common/config/arm/arm-common.c (arm_parse_arch_option_name):
2955         Accept complain bool parameter.  Only emit errors if it is true.
2956         (arm_parse_cpu_option_name): Likewise.
2957         (arm_target_thumb_only): Adjust callers of the above.
2958         * config/arm/arm-protos.h (arm_parse_cpu_option_name): Adjust
2959         prototype to take a default true bool parameter.
2960         (arm_parse_arch_option_name): Likewise.
2961
2962 2018-03-09  David Malcolm  <dmalcolm@redhat.com>
2963             Francois-Xavier Coudert  <fxcoudert@gcc.gnu.org>
2964
2965         PR jit/64089
2966         PR jit/84288
2967         * Makefile.in (LD_VERSION_SCRIPT_OPTION, LD_SONAME_OPTION): New.
2968         * configure: Regenerate.
2969         * configure.ac ("linker --version-script option"): New.
2970         ("linker soname option"): New.
2971
2972 2018-03-09  Richard Biener  <rguenther@suse.de>
2973
2974         PR tree-optimization/84775
2975         * tree-if-conv.c (add_bb_predicate_gimplified_stmts): Delink
2976         immediate uses of predicate stmts and mark them modified.
2977
2978         Revert
2979         PR tree-optimization/84178
2980         * tree-if-conv.c (combine_blocks): Move insert_gimplified_predicates
2981         to caller.
2982         (version_loop_for_if_conversion): Delay update_ssa call.
2983         (tree_if_conversion): Delay update_ssa until after predicate
2984         insertion.
2985
2986 2018-03-09  Eric Botcazou  <ebotcazou@adacore.com>
2987
2988         PR target/84763
2989         * config/i386/winnt.c (i386_pe_seh_cold_init): Use small pre-allocation
2990         when the function accesses prior frames.
2991
2992 2018-03-08  Jakub Jelinek  <jakub@redhat.com>
2993
2994         PR debug/84456
2995         * dwarf2out.c (dw_loc_list): If list && loc_list->first->next, call
2996         gen_llsym, otherwise call maybe_gen_llsym.
2997
2998         PR inline-asm/84742
2999         * recog.c (asm_operand_ok): Return 0 if multi-character constraint
3000         has ',' character inside of it.
3001
3002 2018-03-08  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
3003
3004         PR target/84748
3005         * config/aarch64/aarch64.md (*compare_cstore<mode>_insn): Mark pattern
3006         as clobbering CC_REGNUM.
3007
3008 2018-03-08  Richard Biener  <rguenther@suse.de>
3009
3010         PR middle-end/84552
3011         * tree-scalar-evolution.c: Include tree-into-ssa.h.
3012         (follow_copies_to_constant): Do not follow SSA names registered
3013         for update.
3014
3015 2018-03-08  Richard Biener  <rguenther@suse.de>
3016
3017         PR tree-optimization/84178
3018         * tree-if-conv.c (combine_blocks): Move insert_gimplified_predicates
3019         to caller.
3020         (version_loop_for_if_conversion): Delay update_ssa call.
3021         (tree_if_conversion): Delay update_ssa until after predicate
3022         insertion.
3023
3024 2018-03-08  David Malcolm  <dmalcolm@redhat.com>
3025
3026         PR tree-optimization/84178
3027         * tree-if-conv.c (release_bb_predicate): Remove the
3028         the assertion that the stmts have NULL use_ops.
3029         Discard the statements, asserting that they haven't
3030         yet been added to a BB.
3031
3032 2018-03-08  Richard Biener  <rguenther@suse.de>
3033
3034         PR tree-optimization/84746
3035         * tree-ssa-pre.c (find_leader_in_sets): Deal with SET1 being NULL.
3036         (phi_translate): Pass in destination ANTIC_OUT set.
3037         (phi_translate_1): Likewise.  For a simplified result lookup
3038         a leader in ANTIC_OUT and AVAIL_OUT, not the ANTIC_IN sets.
3039         (phi_translate_set): Adjust.
3040         (do_pre_regular_insertion): Likewise.
3041         (do_pre_partial_partial_insertion): Likewise.
3042
3043 2018-03-08  Martin Liska  <mliska@suse.cz>
3044
3045         PR gcov-profile/84735
3046         * doc/gcov.texi: Document usage of profile files.
3047         * gcov-io.h: Document changes in the format.
3048
3049 2018-03-08  Alexandre Oliva  <aoliva@redhat.com>
3050
3051         PR debug/84404
3052         PR debug/84408
3053         * dwarf2out.c (struct dw_line_info_table): Update comments for
3054         view == -1.
3055         (FORCE_RESET_NEXT_VIEW): New.
3056         (FORCE_RESETTING_VIEW_P): New.
3057         (RESETTING_VIEW_P): Check for -1 too.
3058         (ZERO_VIEW_P): Likewise.
3059         (new_line_info_table): Force-reset next view.
3060         (dwarf2out_begin_function): Likewise.
3061         (dwarf2out_source_line): Simplify zero_view_p initialization.
3062         Test FORCE_RESETTING_VIEW_P and RESETTING_VIEW_P instead of
3063         view directly.  Omit view when omitting .loc at line 0.
3064
3065 2018-03-08  Jakub Jelinek  <jakub@redhat.com>
3066
3067         PR tree-optimization/84740
3068         * tree-switch-conversion.c (process_switch): Call build_constructors
3069         only if info.phi_count is non-zero.
3070
3071         PR tree-optimization/84739
3072         * tree-tailcall.c (find_tail_calls): Check call arguments against
3073         DECL_ARGUMENTS (current_function_decl) rather than
3074         DECL_ARGUMENTS (func) when checking for tail recursion.
3075
3076 2018-03-07  Jakub Jelinek  <jakub@redhat.com>
3077
3078         * doc/contrib.texi: Add entries for Martin Liska, David Malcolm,
3079         Marek Polacek, extend Vladimir Makarov's, Jonathan Wakely's and
3080         Volker Reichelt's entry and add entries for people that perform
3081         GCC fuzzy testing and report numerous bugs.
3082
3083 2018-03-07  Segher Boessenkool  <segher@kernel.crashing.org>
3084
3085         PR target/82411
3086         * config/rs6000/rs6000.c (rs6000_elf_in_small_data_p): Don't put
3087         readonly data in sdata, if that is disabled.
3088         * config/rs6000/sysv4.opt (mreadonly-in-sdata): New option.
3089         * doc/invoke.texi (RS/6000 and PowerPC Options): Document
3090         -mreadonly-in-sdata option.
3091
3092 2018-03-07  Martin Sebor  <msebor@redhat.com>
3093
3094         PR tree-optimization/84468
3095         * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Consider successor
3096         basic block when looking for nul assignment.
3097
3098 2018-03-07  Eric Botcazou  <ebotcazou@adacore.com>
3099
3100         PR target/84277
3101         * except.h (output_function_exception_table): Adjust prototype.
3102         * except.c (output_function_exception_table): Remove FNNAME parameter
3103         and add SECTION parameter.  Ouput one part of the table at a time.
3104         * final.c (final_scan_insn_1) <NOTE_INSN_SWITCH_TEXT_SECTIONS>: Output
3105         the first part of the exception table and emit unwind directives.
3106         * config/i386/i386-protos.h (i386_pe_end_cold_function): Declare.
3107         (i386_pe_seh_cold_init): Likewise.
3108         * config/i386/cygming.h (ASM_DECLARE_COLD_FUNCTION_NAME): New macro.
3109         (ASM_DECLARE_COLD_FUNCTION_SIZE): Likewise.
3110         * config/i386/i386.c (x86_expand_epilogue): Fix wording in comment.
3111         (ix86_output_call_insn): Emit a nop in one more case for SEH.
3112         * config/i386/winnt.c: Include except.h.
3113         (struct seh_frame_state): Add reg_offset, after_prologue and
3114         in_cold_section fields.
3115         (i386_pe_seh_end_prologue): Set seh->after_prologue.
3116         (i386_pe_seh_cold_init): New function.
3117         (i386_pe_seh_fini): Add COLD parameter and bail out if it is not equal
3118         to seh->in_cold_section.
3119         (seh_emit_push): Record the offset of the push.
3120         (seh_emit_save): Record the offet of the save.
3121         (i386_pe_seh_unwind_emit): Deal with NOTE_INSN_SWITCH_TEXT_SECTIONS.
3122         Test seh->after_prologue to disregard the epilogue.
3123         (i386_pe_end_function): Pass FALSE to i386_pe_seh_fini.
3124         (i386_pe_end_cold_function): New function.
3125
3126 2018-03-07  Jakub Jelinek  <jakub@redhat.com>
3127
3128         PR fortran/84565
3129         * config/aarch64/predicates.md (aarch64_simd_reg_or_zero): Use
3130         aarch64_simd_or_scalar_imm_zero rather than aarch64_simd_imm_zero.
3131
3132         PR c++/84704
3133         * gimple-expr.c (create_tmp_var_raw): Set DECL_NAMELESS flag
3134         on tmp_var.
3135         * tree-pretty-print.c (dump_decl_name): For TDF_COMPARE_DEBUG,
3136         don't print names of DECL_NAMELESS DECL_IGNORED_P decls.
3137
3138         PR middle-end/84723
3139         * multiple_target.c: Include tree-inline.h and intl.h.
3140         (expand_target_clones): Diagnose and fail if node->definition and
3141         !tree_versionable_function_p (node->decl).
3142
3143 2018-03-06  John David Anglin  <danglin@gcc.gnu.org>
3144
3145         * config/pa/pa.h (ASM_GENERATE_INTERNAL_LABEL): Revise to use
3146         sprint_ul.
3147         (ASM_OUTPUT_ADDR_VEC_ELT): Revise for above change.
3148         (ASM_OUTPUT_ADDR_DIFF_ELT): Likewise.
3149         * config/pa/pa64-hpux.h (ASM_GENERATE_INTERNAL_LABEL): Revise as above.
3150
3151 2018-03-06  Jakub Jelinek  <jakub@redhat.com>
3152
3153         PR target/84710
3154         * combine.c (try_combine): Use reg_or_subregno instead of handling
3155         just paradoxical SUBREGs and REGs.
3156
3157 2018-03-06  Claudiu Zissulescu  <claziss@synopsys.com>
3158
3159          * config/arc/arc.c (arc_finalize_pic): Remove function.
3160          (arc_must_save_register): We use single base PIC register, remove
3161          checks to save/restore the PIC register.
3162          (arc_expand_prologue): Likewise.
3163          * config/arc/arc-protos.h (arc_set_default_type_attributes):
3164          Remove.
3165          (arc_verify_short): Likewise.
3166          (arc_attr_type): Likewise.
3167          * config/arc/arc.c (arc_set_default_type_attributes): Remove.
3168          (walk_stores): Likewise.
3169          (arc_address_cost): Make it static.
3170          (arc_verify_short): Likewise.
3171          (branch_dest): Likewise.
3172          (arc_attr_type): Likewise.
3173          * config/arc/arc.c (TARGET_ADJUST_INSN_LENGTH): Remove.
3174          (TARGET_INSN_LENGTH_PARAMETERS): Likewise.
3175          (arc_final_prescan_insn): Remove inserting the nops due to
3176          hardware hazards.  It is done in reorg step.
3177          (insn_length_variant_t): Remove.
3178          (insn_length_parameters_t): Likewise.
3179          (arc_insn_length_parameters): Likewise.
3180          (arc_get_insn_variants): Likewise.
3181          * config/arc/arc.h (TARGET_UPSIZE_DBR): Remove.
3182
3183 2018-03-06  Jakub Jelinek  <jakub@redhat.com>
3184
3185         PR inline-asm/84683
3186         * reg-stack.c (move_for_stack_reg): If any_malformed_asm, avoid
3187         assertion failure.
3188
3189         PR tree-optimization/84687
3190         * omp-simd-clone.c (simd_clone_create): Clear DECL_BUILT_IN_CLASS
3191         on new_node->decl.
3192         * match.pd (pow(C,x)*expN(y) -> expN(logN(C)*x+y)): New optimization.
3193
3194 2018-03-05  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
3195
3196         * config/rs6000/rs6000-builtin.def (rs6000_speculation_barrier):
3197         Rename to ppc_speculation_barrier.
3198         * config/rs6000/rs6000.c (rs6000_init_builtins): Rename builtin to
3199         __builtin_ppc_speculation_barrier.
3200
3201 2018-03-05  Jakub Jelinek  <jakub@redhat.com>
3202
3203         PR target/84700
3204         * combine.c (combine_simplify_rtx): Don't try to simplify if
3205         if_then_else_cond returned non-NULL, but either true_rtx or false_rtx
3206         are equal to x.
3207
3208 2018-03-05  Segher Boessenkool  <segher@kernel.crashing.org>
3209
3210         * config/rs6000/rs6000.c (rs6000_loop_align): Don't align tiny loops
3211         to 32 bytes when compiling for POWER9.
3212
3213 2018-03-05  Jakub Jelinek  <jakub@redhat.com>
3214
3215         PR target/84564
3216         * config/i386/i386.c (ix86_function_ok_for_sibcall): Check for
3217         regparm >= 3 with no arg reg available also for calls with
3218         flag_force_indirect_call.  Pass decl to ix86_function_regparm.
3219
3220         PR target/84524
3221         * config/i386/sse.md (*<code><mode>3): Replace <mask_prefix3> with
3222         orig,vex.
3223         (*<plusminus_insn><mode>3): Likewise.  Remove <mask_operand3> uses.
3224
3225 2018-03-05  Peter Bergner  <bergner@vnet.ibm.com>
3226
3227         PR target/84264
3228         * config/rs6000/vector.md (mov<mode>): Disallow altivec memory operands.
3229
3230 2018-03-05  Richard Biener  <rguenther@suse.de>
3231
3232         PR tree-optimization/84486
3233         * tree-ssa-pre.c (create_expression_by_pieces): Remove dead code.
3234         When inserting a __builtin_assume_aligned call set the LHS
3235         SSA name alignment info accordingly.
3236
3237 2018-03-05  Wilco Dijkstra  <wdijkstr@arm.com>
3238
3239         PR tree-optimization/84114
3240         * config/aarch64/aarch64.c (aarch64_reassociation_width)
3241         Avoid reassociation of FLOAT_MODE addition.
3242
3243 2018-03-05  Olga Makhotina  <olga.makhotina@intel.com>
3244
3245         * common/config/i386/i386-common.c (OPTION_MASK_ISA_PCONFIG_SET,
3246         OPTION_MASK_ISA_PCONFIG_UNSET, OPTION_MASK_ISA_WBNOINVD_SET,
3247         OPTION_MASK_ISA_WBNOINVD_UNSET): New definitions.
3248         (ix86_handle_option): Handle -mpconfig and -mwbnoinvd.
3249         * config.gcc (pconfigintrin.h, wbnoinvdintrin.h) : Add headers.
3250         * config/i386/cpuid.h (bit_PCONFIG, bit_WBNOINVD): New.
3251         * config/i386/driver-i386.c (host_detect_local_cpu): Detect -mpconfig
3252         and -mwbnoinvd.
3253         * config/i386/i386-builtin.def (__builtin_ia32_wbnoinvd,
3254         __builtin_ia32_wbinvd): New builtins.
3255         (SPECIAL_ARGS2): New.
3256         * config/i386/i386-c.c (__WBNOINVD__, __PCONFIG__): New.
3257         (SPECIAL_ARGS2): New.
3258         * config/i386/i386.c (ix86_target_string): Add -mpconfig and -mwbnoinvd.
3259         (ix86_valid_target_attribute_inner_p): Ditto.
3260         (ix86_init_mmx_sse_builtins): Add special_args2.
3261         * config/i386/i386.h (TARGET_PCONFIG, TARGET_PCONFIG_P, TARGET_WBNOINVD,
3262         TARGET_WBNOINVD_P): New.
3263         * config/i386/i386.md (UNSPECV_WBINVD, UNSPECV_WBNOINVD): New.
3264         (define_insn "wbinvd", define_insn "wbnoinvd"): New.
3265         * config/i386/i386.opt: Add -mpconfig and -mwbnoinvd.
3266         * config/i386/immintrin.h (_wbinvd): New intrinsic.
3267         * config/i386/pconfigintrin.h: New file.
3268         * config/i386/wbnoinvdintrin.h: Ditto.
3269         * config/i386/x86intrin.h: Add headers pconfigintrin.h and wbnoinvdintrin.h.
3270         * doc/invoke.texi (-mpconfig, -mwbnoinvd): New.
3271
3272 2018-03-05  Richard Biener  <rguenther@suse.de>
3273
3274         PR tree-optimization/84670
3275         * tree-ssa-pre.c (struct bb_bitmap_sets): Add visited_with_visited_succs
3276         member.
3277         (BB_VISITED_WITH_VISITED_SUCCS): New define.
3278         (compute_antic): Initialize BB_VISITED_WITH_VISITED_SUCCS.
3279         (compute_antic_aux): Only assert the number of values in ANTIC_IN
3280         doesn't grow if all successors (recursively) were visited at least
3281         once.
3282
3283 2018-03-05  Richard Biener  <rguenther@suse.de>
3284
3285         PR tree-optimization/84650
3286         * tree-ssa-loop-im.c (pass_lim::execute): Reset the SCEV cache
3287         if executed in the loop pipeline.
3288
3289 2018-03-05  Sandra Loosemore  <sandra@codesourcery.com>
3290
3291         * doc/configfiles.texi (Configuration Files): Move info about
3292         conditionalizing $target-protos.h to...
3293         * doc/sourcebuild.texi (Back End): Here.  Explain how $target.h
3294         differs from $target-protos.h.
3295
3296 2018-03-05  Kito Cheng  <kito.cheng@gmail.com>
3297             Chung-Ju Wu  <jasonwucj@gmail.com>
3298
3299         * config/nds32/nds32-protos.h (nds32_expand_setmem): Declare.
3300         * config/nds32/nds32-multiple.md (setmemsi): Define.
3301         * config/nds32/nds32-memory-manipulation.c
3302         (nds32_gen_dup_4_byte_to_word_value): New.
3303         (emit_setmem_word_loop): New.
3304         (emit_setmem_byte_loop): New.
3305         (nds32_expand_setmem_loop): New.
3306         (nds32_expand_setmem_loop_v3m): New.
3307         (nds32_expand_setmem_unroll): New.
3308         (nds32_expand_setmem): New.
3309
3310 2018-03-04  Kito Cheng  <kito.cheng@gmail.com>
3311             Chung-Ju Wu  <jasonwucj@gmail.com>
3312
3313         * config/nds32/nds32-memory-manipulation.c
3314         (nds32_emit_load_store): New.
3315         (nds32_emit_post_inc_load_store): New.
3316         (nds32_emit_mem_move): New.
3317         (nds32_emit_mem_move_block): New.
3318         (nds32_expand_movmemsi_loop_unknown_size): New.
3319         (nds32_expand_movmemsi_loop_known_size): New.
3320         (nds32_expand_movmemsi_loop): New.
3321         (nds32_expand_movmemsi_unroll): New.
3322         (nds32_expand_movmemqi): Rename ...
3323         (nds32_expand_movmemsi): ... to this.
3324         * config/nds32/nds32-multiple.md (movmemqi): Rename ...
3325         (movmemsi): ... to this.
3326         * config/nds32/nds32-protos.h (nds32_expand_movmemqi): Rename ...
3327         (nds32_expand_movmemsi): ... to this.
3328
3329 2018-03-04  Kito Cheng  <kito.cheng@gmail.com>
3330             Monk Chiang  <sh.chiang04@gmail.com>
3331             Chung-Ju Wu  <jasonwucj@gmail.com>
3332
3333         * config/nds32/nds32-protos.h
3334         (nds32_expand_load_multiple): New arguments.
3335         (nds32_expand_store_multiple): Ditto.
3336         (nds32_valid_multiple_load_store): Rename ...
3337         (nds32_valid_multiple_load_store_p): ... to this.
3338         * config/nds32/nds32-memory-manipulation.c
3339         (nds32_expand_load_multiple): Refine implementation.
3340         (nds32_expand_store_multiple): Ditto.
3341         * config/nds32/nds32-multiple.md
3342         (load_multiple): Update nds32_expand_load_multiple interface.
3343         (store_multiple): Update nds32_expand_store_multiple interface.
3344         * config/nds32/nds32-predicates.c
3345         (nds32_valid_multiple_load_store): Rename ...
3346         (nds32_valid_multiple_load_store_p): ... to this and refine
3347         implementation.
3348         * config/nds32/predicates.md
3349         (nds32_load_multiple_and_update_address_operation): New predicate.
3350         (nds32_store_multiple_and_update_address_operation): New predicate.
3351
3352 2018-03-04  Kito Cheng  <kito.cheng@gmail.com>
3353             Chung-Ju Wu  <jasonwucj@gmail.com>
3354
3355         * config/nds32/nds32.md (type): Add load_multiple and store_multiple.
3356         (combo): New attribute.
3357         * config/nds32/nds32-multiple.md: Refine patterns with new attributes.
3358
3359 2018-03-03  Chung-Ju Wu  <jasonwucj@gmail.com>
3360
3361         * config/nds32/nds32.opt: Change -mcmodel= default value.
3362
3363 2018-03-03  Kito Cheng  <kito.cheng@gmail.com>
3364             Monk Chiang  <sh.chiang04@gmail.com>
3365             Chung-Ju Wu  <jasonwucj@gmail.com>
3366
3367         * config/nds32/constants.md (unspec_element): New enum.
3368         * config/nds32/constraints.md (Umw): New constraint.
3369         * config/nds32/nds32-intrinsic.c: Add more builtin functions.
3370         * config/nds32/nds32-intrinsic.md: Likewise.
3371         * config/nds32/nds32-md-auxiliary.c (nds32_regno_to_enable4): New.
3372         (nds32_valid_smw_lwm_base_p): New.
3373         (nds32_output_smw_single_word): New.
3374         (nds32_output_lmw_single_word): New.
3375         (nds32_expand_unaligned_load): New.
3376         (nds32_expand_unaligned_store): New.
3377         * config/nds32/nds32-protos.h (nds32_valid_smw_lwm_base_p): Declare.
3378         (nds32_output_smw_single_word): Declare.
3379         (nds32_output_lmw_single_word): Declare.
3380         (nds32_expand_unaligned_load): Declare.
3381         (nds32_expand_unaligned_store): Declare.
3382         * config/nds32/nds32.h (nds32_builtins): Add NDS32_BUILTIN_UALOAD_HW,
3383         NDS32_BUILTIN_UALOAD_W, NDS32_BUILTIN_UALOAD_DW,
3384         NDS32_BUILTIN_UASTORE_HW, NDS32_BUILTIN_UASTORE_W,
3385         NDS32_BUILTIN_UASTORE_DW.
3386         * config/nds32/predicates.md (nds32_lmw_smw_base_operand): New
3387         predicate.
3388
3389 2018-03-03  Monk Chiang  <sh.chiang04@gmail.com>
3390             Kito Cheng  <kito.cheng@gmail.com>
3391             Chung-Ju Wu  <jasonwucj@gmail.com>
3392
3393         * config/nds32/nds32-intrinsic.c
3394         (nds32_expand_builtin_null_ftype_reg): Delete.
3395         (nds32_expand_builtin_reg_ftype_imm): Ditto.
3396         (nds32_expand_builtin_null_ftype_reg_imm): Ditto.
3397         (nds32_read_argument): New.
3398         (nds32_legitimize_target): Ditto.
3399         (nds32_legitimize_argument): Ditto.
3400         (nds32_check_constant_argument): Ditto.
3401         (nds32_expand_unop_builtin): Ditto.
3402         (nds32_expand_unopimm_builtin): Ditto.
3403         (nds32_expand_binop_builtin): Ditto.
3404         (nds32_builtin_decl_impl): Ditto.
3405         (builtin_description): Ditto.
3406         (nds32_expand_builtin_impl): Rewrite with new infrastructure.
3407         (nds32_init_builtins_impl): Ditto.
3408         * config/nds32/nds32.c (TARGET_BUILTIN_DECL): Define.
3409         (nds32_builtin_decl): New.
3410         * config/nds32/nds32.h (nds32_builtins): Add NDS32_BUILTIN_COUNT.
3411         * config/nds32/nds32-protos.h (nds32_builtin_decl_impl): Declare.
3412
3413 2018-03-02  Jeff Law  <law@redhat.com>
3414
3415         * reorg.c (stop_search_p): Handle DEBUG_INSN.
3416         (redundant_insn, fill_simple_delay_slots): Likewise.
3417         (fill_slots_from_thread): Likewise.
3418         * resource.c (mark_referenced_resources): Likewise.
3419         (mark_set_resources, find_dead_or_set_registers): Likewise.
3420
3421 2018-03-02  Jakub Jelinek  <jakub@redhat.com>
3422
3423         * substring-locations.h (format_warning_va): Formatting fix for
3424         ATTRIBUTE_GCC_DIAG.
3425         (format_warning_at_substring): Fix up ATTRIBUTE_GCC_DIAG second
3426         argument.
3427         (format_warning_n_va, format_warning_at_substring_n): New prototypes.
3428         * substring-locations.c: Include intl.h.
3429         (format_warning_va): Turned into small wrapper around
3430         format_warning_n_va, renamed to ...
3431         (format_warning_n_va): ... this, add N and PLURAL_GMSGID arguments,
3432         rename GMSGID to SINGULAR_GMSGID, if SINGULAR_GMSGID != PLURAL_GMSGID,
3433         use ngettext.
3434         (format_warning_at_substring_n): New function.
3435         * gimple-ssa-sprintf.c: Remove GCC diagnostic ignored pragma.
3436         (fmtwarn): Add ATTRIBUTE_GCC_DIAG.  Turn into a copy of
3437         format_warning_at_substring with just a shorter name instead of
3438         const function pointer.
3439         (fmtwarn_n): New function.
3440         (maybe_warn, format_directive, parse_directive): Use fmtwarn_n where
3441         appropriate, get rid of all the fmtstr temporaries, move conditionals
3442         with G_() wrapped string literals directly into fmtwarn arguments,
3443         cast dir.len to (int), formatting fixes.
3444
3445 2018-03-02  Thomas Schwinge  <thomas@codesourcery.com>
3446
3447         * doc/invoke.texi: Remove "Cilk Plus" references.
3448
3449 2018-03-02  Jakub Jelinek  <jakub@redhat.com>
3450             Richard Biener  <rguenther@suse.de>
3451
3452         PR ipa/84628
3453         * expr.c (expand_expr_real_1) <case CALL_EXPR>: Don't emit diagnostics
3454         for error or warning attributes if CALL_FROM_THUNK_P is set.
3455         Formatting fixes.
3456
3457 2018-03-02  Jakub Jelinek  <jakub@redhat.com>
3458
3459         PR target/56540
3460         * config/pa/pa.h (TARGET_CPU_CPP_BUILTINS): Predefine
3461         __SIZEOF_128__ macro if HPUX_LONG_DOUBLE_LIBRARY.
3462
3463         PR target/56540
3464         * config/ia64/ia64.h (TARGET_CPU_CPP_BUILTINS): Predefine
3465         __SIZEOF_{FPREG,FLOAT{80,128}}__ macros.
3466
3467         * predict.c (test_prediction_value_range): Use PROB_UNINITIALIZED
3468         instead of -1U in last predictors element's probability member.
3469
3470 2018-03-02  Eric Botcazou  <ebotcazou@adacore.com>
3471
3472         PR ipa/83983
3473         * ipa-devirt.c (odr_subtypes_equivalent_p): Get the ODR type of both
3474         arguments if they are comparable.
3475
3476 2018-03-02  Richard Sandiford  <richard.sandiford@linaro.org>
3477
3478         PR tree-optimization/84634
3479         * tree-vect-stmts.c (vectorizable_store, vectorizable_load): Replace
3480         masks and masked_loop_p with a single loop_masks, making sure it's
3481         null for bb vectorization.
3482
3483 2018-03-02  Richard Sandiford  <richard.sandiford@linaro.org>
3484
3485         * tree-vect-data-refs.c (vect_analyze_data_ref_dependence)
3486         (vect_analyze_data_ref_access): Use loop->safe_len rather than
3487         loop->force_vectorize to check whether there is no alias.
3488
3489 2018-03-02  Jakub Jelinek  <jakub@redhat.com>
3490
3491         PR target/84614
3492         * rtl.h (prev_real_nondebug_insn, next_real_nondebug_insn): New
3493         prototypes.
3494         * emit-rtl.c (next_real_insn, prev_real_insn): Fix up function
3495         comments.
3496         (next_real_nondebug_insn, prev_real_nondebug_insn): New functions.
3497         * cfgcleanup.c (try_head_merge_bb): Use prev_real_nondebug_insn
3498         instead of a loop around prev_real_insn.
3499         * combine.c (move_deaths): Use prev_real_nondebug_insn instead of
3500         prev_real_insn.
3501
3502         PR inline-asm/84625
3503         * config/i386/i386.c (ix86_print_operand): Use conditional
3504         output_operand_lossage instead of gcc_assert if CONST_VECTOR is not
3505         zero vector.
3506
3507 2018-03-02  Richard Biener  <rguenther@suse.de>
3508
3509         PR tree-optimization/84427
3510         * tree-ssa-pre.c (bitmap_remove_expr_from_set): Remove.
3511         (bitmap_set_subtract_values): Rewrite to handle multiple
3512         exprs per value.
3513         (clean): Likewise.
3514         (prune_clobbered_mems): Likewise.
3515         (phi_translate): Take edge instead of pred/phiblock.
3516         (phi_translate_1): Likewise.
3517         (phi_translate_set): Likewise.  Insert all translated
3518         exprs for a value into the set, keeping possibly multiple
3519         expressions per value.
3520         (compute_antic_aux): Adjust for phi_translate changes.
3521         When intersecting union the expressions and prune those
3522         not in the final value set, keeping possibly multiple
3523         expressions per value.  Do not use value-insertion
3524         for unioning ANTIC_OUT U EXP_GEN - TMP_GEN but merge
3525         all expressions.  Add verification that the value-sets
3526         only shrink during iteration.
3527         (compute_partial_antic_aux): Adjust for the phi_translate changes.
3528         (do_pre_regular_insertion): Likewise.
3529         (do_pre_partial_partial_insertion): Likewise.
3530
3531 2018-03-02  Richard Biener  <rguenther@suse.de>
3532
3533         PR target/82005
3534         * config/darwin.c (saved_debug_info_level): New static global.
3535         (darwin_asm_lto_start): Disable debug info generation for LTO out.
3536         (darwin_asm_lto_end): Restore debug info generation settings.
3537
3538 2018-03-01  Martin Liska  <mliska@suse.cz>
3539
3540         PR sanitizer/82484
3541         * sanopt.c (sanitize_rewrite_addressable_params): Do not handle
3542         volatile arguments.
3543
3544 2018-03-01  Richard Biener  <rguenther@suse.de>
3545
3546         PR debug/84645
3547         * dwarf2out.c (gen_variable_die): Properly handle late VLA
3548         type annotation with LTO when debug was disabled at compile-time.
3549
3550 2018-03-01  Matthew Fortune  <mfortune@gmail.com>
3551
3552         * config/mips/mips.c (mips_final_prescan_insn): Fix incorrect
3553         XINT with INTVAL.
3554         (mips_final_postscan_insn): Likewise.
3555
3556 2018-03-01  Richard Sandiford  <richard.sandiford@linaro.org>
3557
3558         PR rtl-optimization/84528
3559         * alias.c (init_alias_target): Add commentary.
3560         (init_alias_analysis): Only give HARD_FRAME_POINTER_REGNUM
3561         a unique base value if the frame pointer is not eliminated
3562         to the stack pointer.
3563
3564 2018-03-01  Tom de Vries  <tom@codesourcery.com>
3565
3566         PR rtl-optimization/83327
3567         * lra-int.h (hard_regs_spilled_into): Declare.
3568         * lra.c (hard_regs_spilled_into): Define.
3569         (init_reg_info): Init hard_regs_spilled_into.
3570         * lra-spills.c (assign_spill_hard_regs): Update hard_regs_spilled_into.
3571         * lra-lives.c (make_hard_regno_born, make_hard_regno_dead)
3572         (process_bb_lives): Handle hard_regs_spilled_into.
3573         (lra_create_live_ranges_1): Before doing liveness propagation, clear
3574         regs in all_hard_regs_bitmap if set in hard_regs_spilled_into.
3575
3576 2018-02-28  David Edelsohn  <dje.gcc@gmail.com>
3577
3578         * config.gcc (powerpc-ibm-aix7.1.*): New stanza.
3579         (powerpc-ibm-aix[789]*): Default to AIX 7.2.
3580         * config/rs6000/aix71.h (TARGET_DEFAULT): Revert to Power4 ISA.
3581         * config/rs6000/aix72.h: New file.
3582
3583 2018-02-28  Jakub Jelinek  <jakub@redhat.com>
3584
3585         * gimple-ssa-warn-restrict.c (maybe_diag_overlap): Use warning_n
3586         instead of warning_at with conditional singular and plural messages
3587         where possible.
3588
3589         PR target/52991
3590         * stor-layout.c (update_alignment_for_field): For
3591         targetm.ms_bitfield_layout_p (rli->t), if !is_bitfield
3592         && !DECL_PACKED (field), do the alignment update, just use
3593         only desired_align instead of MAX (type_align, desired_align)
3594         as the alignment.
3595         (place_field): Don't do known_align < desired_align handling
3596         early if targetm.ms_bitfield_layout_p (rli->t) and rli->prev_field
3597         is non-NULL, instead do it after rli->prev_field handling and
3598         only if not within a bitfield word.  For DECL_PACKED (field)
3599         use type_align of BITS_PER_UNIT.
3600
3601 2018-02-28  Eric Botcazou  <ebotcazou@adacore.com>
3602
3603         * config/aarch64/aarch64.c (aarch64_emit_probe_stack_range): Remove
3604         superfluous parentheses and trailing spaces.
3605
3606 2018-02-28  Richard Biener  <rguenther@suse.de>
3607
3608         PR tree-optimization/84584
3609         * graphite-scop-detection.c (scop_detection::add_scop): Discard
3610         SCoPs with fake exit edge.
3611
3612 2018-02-28  Martin Liska  <mliska@suse.cz>
3613
3614         PR testsuite/84597
3615         * timevar.c (timer::print): Fix format to properly print 100%
3616         values.
3617
3618 2018-02-28  Richard Biener  <rguenther@suse.de>
3619
3620         PR middle-end/84607
3621         * genmatch.c (capture_info::walk_match): Do not mark
3622         captured expressions without operands as expr_p given
3623         they act more like predicates and should be subject to
3624         "lost tail" side-effect preserving.
3625
3626 2018-02-28  Alexandre Oliva  <aoliva@redhat.com>
3627
3628         PR rtl-optimization/81611
3629         * auto-inc-dec.c (attempt_change): Move dead note from
3630         mem_insn if it's the next use of regno
3631         (find_address): Take address use of reg holding
3632         non-incremented value.  Add parm to limit search to the named
3633         reg only.
3634         (merge_in_block): Attempt to use a mem insn that is the next
3635         use of the original regno.
3636
3637 2018-02-27  Martin Sebor  <msebor@redhat.com>
3638
3639         PR c++/83871
3640         * gcc/doc/invoke.texi (-Wmissing-attributes): New option.
3641         * gcc/print-tree.c (print_node): Handle DECL_UNINLINABLE.
3642
3643 2018-02-27  Martin Sebor  <msebor@redhat.com>
3644
3645         PR translation/84207
3646         * diagnostic-core.h (warning_n, error_n, inform_n): Change
3647         n argument to unsigned HOST_WIDE_INT.
3648         * diagnostic.c (warning_n, error_n, inform_n): Ditto.
3649         (diagnostic_n_impl): Ditto.  Handle arguments in excess of LONG_MAX.
3650         * gimple-ssa-sprintf.c (format_directive): Simplify inform_n call.
3651         * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Use warning_n.
3652
3653 2018-02-27  Richard Biener  <rguenther@suse.de>
3654
3655         PR tree-optimization/84512
3656         * tree-vect-loop.c (vect_compute_single_scalar_iteration_cost):
3657         Do not use the estimate returned from record_stmt_cost for
3658         the scalar iteration cost but sum properly using add_stmt_cost.
3659
3660 2018-02-27  Richard Biener  <rguenther@suse.de>
3661
3662         PR tree-optimization/84466
3663         * graphite-scop-detection.c (scop_detection::stmt_simple_for_scop_p):
3664         Adjust last change to less strictly validate use operands.
3665
3666 2018-02-27  Martin Liska  <mliska@suse.cz>
3667
3668         PR gcov-profile/84548
3669         * gcov.c (process_file): Allow partial overlap and consider it
3670         also as group functions.
3671         (output_lines): Properly calculate range of lines for a group.
3672
3673 2018-02-27  Martin Liska  <mliska@suse.cz>
3674
3675         * timevar.c (timer::print_row): Remove 'usr', 'sys', 'wall' and
3676         'ggc' suffixes.  Change first column width.
3677         (timer::print): Fix formatting of the column.
3678
3679 2018-02-27  Alexandre Oliva  <aoliva@redhat.com>
3680
3681         * tree-ssa-live.c (remove_unused_scope_block_p): Do not
3682         preserve inline entry blocks for the sake of debug inline
3683         entry point markers alone.
3684         (remove_unused_locals): Suggest in comments a better place to
3685         force the preservation of inline entry blocks that are
3686         otherwise unused, but do not preserve them.
3687
3688 2018-02-26  H.J. Lu  <hongjiu.lu@intel.com>
3689
3690         * config/i386/i386.c (ix86_output_indirect_jmp): Update comments.
3691
3692 2018-02-26  H.J. Lu  <hongjiu.lu@intel.com>
3693
3694         PR target/84039
3695         * config/i386/constraints.md (Bs): Replace
3696         ix86_indirect_branch_register with
3697         TARGET_INDIRECT_BRANCH_REGISTER.
3698         (Bw): Likewise.
3699         * config/i386/i386.md (indirect_jump): Likewise.
3700         (tablejump): Likewise.
3701         (*sibcall_memory): Likewise.
3702         (*sibcall_value_memory): Likewise.
3703         Peepholes of indirect call and jump via memory: Likewise.
3704         (*sibcall_GOT_32): Disallowed for TARGET_INDIRECT_BRANCH_REGISTER.
3705         (*sibcall_value_GOT_32): Likewise.
3706         * config/i386/predicates.md (indirect_branch_operand): Likewise.
3707         (GOT_memory_operand): Likewise.
3708         (call_insn_operand): Likewise.
3709         (sibcall_insn_operand): Likewise.
3710         (GOT32_symbol_operand): Likewise.
3711         * config/i386/i386.h (TARGET_INDIRECT_BRANCH_REGISTER): New.
3712
3713 2018-02-26  Eric Botcazou  <ebotcazou@adacore.com>
3714
3715         PR rtl-optimization/83496
3716         * reorg.c (steal_delay_list_from_target): Change REDUNDANT array from
3717         booleans to RTXes.  Call fix_reg_dead_note on every non-null element.
3718         (steal_delay_list_from_fallthrough): Call fix_reg_dead_note on a
3719         redundant insn, if any.
3720         (relax_delay_slots): Likewise.
3721         (update_reg_unused_notes): Rename REDUNDANT_INSN to OTHER_INSN.
3722
3723 2018-02-26  Richard Sandiford  <richard.sandiford@linaro.org>
3724
3725         PR tree-optimization/83965
3726         * tree-vect-patterns.c (vect_reassociating_reduction_p): Assume
3727         that grouped statements are part of a reduction chain.  Return
3728         true if the statement is not marked as a reduction itself but
3729         is part of a group.
3730         (vect_recog_dot_prod_pattern): Don't check whether the statement
3731         is part of a group here.
3732         (vect_recog_sad_pattern): Likewise.
3733         (vect_recog_widen_sum_pattern): Likewise.
3734
3735 2018-02-26  Eric Botcazou  <ebotcazou@adacore.com>
3736
3737         PR debug/84545
3738         * final.c (rest_of_clean_state): Also look for calls inside sequences.
3739
3740 2018-02-26  H.J. Lu  <hongjiu.lu@intel.com>
3741
3742         PR target/84530
3743         * config/i386/i386-protos.h (ix86_output_indirect_jmp): Remove
3744         the bool argument.
3745         (ix86_output_indirect_function_return): New prototype.
3746         (ix86_split_simple_return_pop_internal): Likewise.
3747         * config/i386/i386.c (indirect_return_via_cx): New.
3748         (indirect_return_via_cx_bnd): Likewise.
3749         (indirect_thunk_name): Handle return va CX_REG.
3750         (output_indirect_thunk_function): Create alias for
3751         __x86_return_thunk_[re]cx and __x86_return_thunk_[re]cx_bnd.
3752         (ix86_output_indirect_jmp): Remove the bool argument.
3753         (ix86_output_indirect_function_return): New function.
3754         (ix86_split_simple_return_pop_internal): Likewise.
3755         * config/i386/i386.md (*indirect_jump): Don't pass false
3756         to ix86_output_indirect_jmp.
3757         (*tablejump_1): Likewise.
3758         (simple_return_pop_internal): Change it to define_insn_and_split.
3759         Call ix86_split_simple_return_pop_internal to split it for
3760         -mfunction-return=.
3761         (simple_return_indirect_internal): Call
3762         ix86_output_indirect_function_return instead of
3763         ix86_output_indirect_jmp.
3764
3765 2018-02-26  Jakub Jelinek  <jakub@redhat.com>
3766
3767         PR bootstrap/84405
3768         * vec.h (vec_default_construct): For BROKEN_VALUE_INITIALIZATION use
3769         memset and value initialization afterwards.
3770
3771 2018-02-26  Christophe Lyon  <christophe.lyon@linaro.org>
3772
3773         * Makefile.in (lto-wrapper): Use ALL_LINKERFLAGS.
3774
3775 2018-02-26  Ramana Radhakrishnan  <ramana.radhakrishnan@arm.com>
3776
3777         PR target/84521
3778         * common/config/aarch64/aarch64-common.c
3779         (aarch_option_optimization_table[]): Switch
3780         off fomit-frame-pointer
3781
3782 2018-02-26  Kito Cheng  <kito.cheng@gmail.com>
3783             Chung-Ju Wu  <jasonwucj@gmail.com>
3784
3785         * config/nds32/nds32-multiple.md (load_multiple): Disallow
3786         volatile memory.
3787         (store_multiple): Ditto.
3788
3789 2018-02-26  Kito Cheng  <kito.cheng@gmail.com>
3790
3791         * config.gcc: Add --with-cpu support for nds32 target.
3792         * config/nds32/nds32-opts.h (nds32_cpu_type): New.
3793         * config/nds32/nds32.opt: Add -mcpu= option.
3794
3795 2018-02-25  Segher Boessenkool  <segher@kernel.crashing.org>
3796
3797         * config/rs6000/rs6000.opt (mvrsave=no, mvrsave=yes, isel=no,
3798         isel=yes): Warn for these deprecated options.
3799
3800 2018-02-23  David Edelsohn  <dje.gcc@gmail.com>
3801
3802         * config/rs6000/aix71.h (TARGET_DEFAULT): Change to
3803         ISA_2_5_MASKS_EMBEDDED.
3804
3805 2018-02-23  Jakub Jelinek  <jakub@redhat.com>
3806
3807         * ipa-prop.c (ipa_vr_ggc_hash_traits::hash): Hash p->min and
3808         p->max as pointers rather than using iterative_hash_expr.
3809
3810 2018-02-23  Carl Love  <cel@us.ibm.com>
3811
3812         * config/rs6000/rs6000-builtin.def: Change VSIGNED2 and VUNSIGNED2
3813         macro expansions from BU_VSX_2 to BU_P8V_VSX_2 and BU_VSX_OVERLOAD_2 to
3814         BU_P8V_OVERLOAD_2.
3815         * config/rs6000/rs6000-c.c: Change VSX_BUILTIN_VEC_VSIGNED2 to
3816         P8V_BUILTIN_VEC_VSIGNED2.  Change VSX_BUILTIN_VEC_VUNSIGNED2 to
3817         P8V_BUILTIN_VEC_VUNSIGNED2.
3818
3819 2018-02-22  Vladimir Makarov  <vmakarov@redhat.com>
3820
3821         PR target/81572
3822         * lra-int.h (LRA_UNKNOWN_ALT, LRA_NON_CLOBBERED_ALT): New macros.
3823         * lra.c (lra_set_insn_recog_data, lra_update_insn_recog_data): Use
3824         LRA_UNKNOWN_ALT.
3825         * lra-constraints.c (curr_insn_transform): Set up
3826         LRA_NON_CLOBBERED_ALT for moves processed on the fast path.  Use
3827         LRA_UNKNOWN_ALT.
3828         (remove_inheritance_pseudos): Use LRA_UNKNOWN_ALT.
3829         * lra-eliminations.c (spill_pseudos): Ditto.
3830         (process_insn_for_elimination): Ditto.
3831         * lra-lives.c (reg_early_clobber_p): Use the new macros.
3832         * lra-spills.c (spill_pseudos): Use LRA_UNKNOWN_ALT and
3833         LRA_NON_CLOBBERED_ALT.
3834
3835 2018-02-22  Martin Sebor  <msebor@redhat.com>
3836
3837         PR tree-optimization/84480
3838         * gimple-fold.c (gimple_fold_builtin_strcpy): Move warnings
3839         to maybe_diag_stxncpy_trunc.  Call it.
3840         * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Integrate warnings
3841         from gimple_fold_builtin_strcpy.  Print inlining stack.
3842         (handle_builtin_stxncpy): Print inlining stack.
3843         * tree-ssa-strlen.h (maybe_diag_stxncpy_trunc): Declare.
3844
3845 2018-02-22  H.J. Lu  <hongjiu.lu@intel.com>
3846
3847         PR target/84176
3848         * config/i386/i386.c (ix86_set_indirect_branch_type): Issue an
3849         error when -mindirect-branch=thunk-extern, -fcf-protection=branch
3850         and -fcheck-pointer-bounds are used together.
3851         (indirect_thunk_prefix): New enum.
3852         (indirect_thunk_need_prefix): New function.
3853         (indirect_thunk_name): Replace need_bnd_p with need_prefix.  Use
3854         "_nt" instead of "_bnd" for NOTRACK prefix.
3855         (output_indirect_thunk): Replace need_bnd_p with need_prefix.
3856         (output_indirect_thunk_function): Likewise.
3857         (): Likewise.
3858         (ix86_code_end): Update output_indirect_thunk_function calls.
3859         (ix86_output_indirect_branch_via_reg): Replace
3860         ix86_bnd_prefixed_insn_p with indirect_thunk_need_prefix.
3861         (ix86_output_indirect_branch_via_push): Likewise.
3862         (ix86_output_function_return): Likewise.
3863         * doc/invoke.texi: Document -mindirect-branch=thunk-extern is
3864         incompatible with -fcf-protection=branch and
3865         -fcheck-pointer-bounds.
3866
3867 2018-02-22  Steve Ellcey  <sellcey@cavium.com>
3868
3869         PR target/83335
3870         * config/aarch64/aarch64.c (aarch64_print_address_internal):
3871         Change gcc_assert call to output_operand_lossage.
3872
3873 2018-02-22  Steve Ellcey  <sellcey@cavium.com>
3874
3875         * doc/extend.texi (__builtin_extend_pointer): Document builtin.
3876
3877 2018-02-22  DJ Delorie  <dj@redhat.com>
3878             Sebastian Perta  <sebastian.perta@renesas.com>
3879             Oleg Endo  <olegendo@gcc.gnu.org>
3880
3881         * config/rx/rx.c (rx_rtx_costs): New function.
3882         (TARGET_RTX_COSTS): Override to use rx_rtx_costs.
3883
3884 2018-02-22  Thomas Preud'homme  <thomas.preudhomme@arm.com>
3885
3886         * config/arm/t-multilib: Map Armv8-R to Armv7 multilibs.
3887
3888 2018-02-22  Martin Liska  <mliska@suse.cz>
3889
3890         PR driver/83193
3891         * common/config/arm/arm-common.c (arm_print_hint_for_cpu_option):
3892         Add "native" as a possible value.
3893
3894 2018-02-22  Martin Liska  <mliska@suse.cz>
3895
3896         PR driver/83193
3897         * config/i386/i386.c (ix86_option_override_internal):
3898         Add "native" as a possible value for -march and -mtune.
3899
3900 2018-02-22  Jakub Jelinek  <jakub@redhat.com>
3901
3902         PR target/84502
3903         * stor-layout.c (finalize_type_size): Propagate TYPE_EMPTY_P flag
3904         to all type variants.
3905
3906         PR tree-optimization/84503
3907         * gimple-ssa-store-merging.c (merged_store_group::merge_into): Compute
3908         width as info->bitpos + info->bitsize - start.
3909         (merged_store_group::merge_overlapping): Simplify width computation.
3910         (check_no_overlap): New function.
3911         (imm_store_chain_info::try_coalesce_bswap): Compute expected
3912         start + width and last_order of the group, fail if check_no_overlap
3913         fails.
3914         (imm_store_chain_info::coalesce_immediate_stores): Don't merge info
3915         to group if check_no_overlap fails.
3916
3917 2018-02-21  Segher Boessenkool  <segher@kernel.crashing.org>
3918
3919         * config/rs6000/altivec.md: Delete contraint arguments to
3920         define_expand, define_split, and define_peephole2, and in
3921         define_insn_and_split if always unused.
3922         * config/rs6000/darwin.md: Ditto.
3923         * config/rs6000/dfp.md: Ditto.
3924         * config/rs6000/rs6000.md: Ditto.
3925         * config/rs6000/sync.md: Ditto.
3926         * config/rs6000/vector.md: Ditto.
3927         * config/rs6000/vsx.md: Ditto.
3928
3929 2018-02-21  Segher Boessenkool  <segher@kernel.crashing.org>
3930
3931         * config/rs6000/altivec.md: Write output control strings as braced
3932         blocks instead of double-quoted strings.
3933         * config/rs6000/darwin.md: Ditto.
3934         * config/rs6000/rs6000.md: Ditto.
3935         * config/rs6000/vector.md: Ditto.
3936         * config/rs6000/vsx.md: Ditto.
3937
3938 2018-02-21  Jason Merrill  <jason@redhat.com>
3939
3940         PR c++/84314 - ICE with templates and fastcall attribute.
3941         * attribs.c (build_type_attribute_qual_variant): Remove assert.
3942
3943 2018-02-21  Jan Hubicka  <hubicka@ucw.cz>
3944
3945         * ipa-cp.c (determine_versionability): Fix comment typos.
3946
3947 2018-02-21  Jan Hubicka  <hubicka@ucw.cz>
3948
3949         PR c/84229
3950         * ipa-cp.c (determine_versionability): Do not version functions caling
3951         va_arg_pack.
3952
3953 2018-02-21  Martin Liska  <mliska@suse.cz>
3954
3955         PR driver/83193
3956         * config/aarch64/aarch64.c (aarch64_print_hint_for_core_or_arch):
3957         Add "native" as a possible value.
3958         * config/aarch64/aarch64.h (HAVE_LOCAL_CPU_DETECT):  Define
3959         the macro when native cpu detection is available.
3960
3961 2018-02-21  Martin Liska  <mliska@suse.cz>
3962
3963         PR driver/83193
3964         * common/config/arm/arm-common.c (arm_print_hint_for_arch_option):
3965         Add "native" as a possible value.
3966         * config/arm/arm.h (HAVE_LOCAL_CPU_DETECT): Define the macro
3967         when native cpu detection is available.
3968
3969 2018-02-21  Jakub Jelinek  <jakub@redhat.com>
3970             Martin Sebor  <msebor@redhat.com>
3971
3972         PR tree-optimization/84478
3973         * gimple-fold.h (get_range_strlen): Add a bool argument defaulted to
3974         false.
3975         * gimple-fold.c (get_range_strlen): Make minlen const and assume it
3976         can't be NULL.  Change FUZZY from bool to int, for 1 add PHI/COND_EXPR
3977         support which is conservatively correct, for 2 only stay conservative
3978         for maxlen.  Formatting and comment capitalization fixes.  Add STRICT
3979         argument to the 2 argument get_range_strlen, adjust 6 arg
3980         get_range_strlen caller and clear minmaxlen[0] and [1] if it returned
3981         false.
3982         (get_maxval_strlen): Adjust 6 arg get_range_strlen caller.
3983         (gimple_fold_builtin_strlen): Pass true as last argument to
3984         get_range_strlen.
3985
3986 2018-02-20  Martin Sebor  <msebor@redhat.com>
3987
3988         PR middle-end/84095
3989         * gimple-ssa-warn-restrict.c (builtin_memref::extend_offset_range): New.
3990         (builtin_memref::set_base_and_offset): Same.  Handle inner references.
3991         (builtin_memref::builtin_memref): Factor out parts into
3992         set_base_and_offset and call it.
3993
3994 2018-02-20  Richard Sandiford  <richard.sandiford@linaro.org>
3995
3996         PR middle-end/84406
3997         * optabs-query.c (find_widening_optab_handler_and_mode): If from_mode
3998         is a scalar_int_mode, assert that to_mode is a scalar_int_mode with
3999         greater precision.  If to_mode is a MODE_PARTIAL_INT, stop the
4000         search at the associated MODE_INT.
4001
4002 2018-02-20  Jeff Law  <law@redhat.com>
4003
4004         PR middle-end/82123
4005         PR tree-optimization/81592
4006         PR middle-end/79257
4007         * gimple-ssa-sprintf.c (format_integer): Query EVRP range analyzer
4008         for range data rather than using global data.
4009         * gimple-ssa-sprintf.c (get_int_range): Query EVRP range analyzer for
4010         range data rather than using global data.
4011         * gimple-ssa-sprintf.c (get_int_range): Accept vr_values parameter
4012         pass it to children as needed.
4013         (struct directive::fmtresult): Similarly.
4014         (struct directive::set_width): Similarly.
4015         (struct directive::set_precision): Similarly.
4016         (format_integer, format_directive, parse_directive): Similarly.
4017         (format_none): Accept unnamed vr_values parameter.
4018         (format_percent, format_floating, format_character): Similarly.
4019         (format_string, format_plain): Similarly.
4020         * gimple-ssa-sprintf.c (sprintf_dom_walker::handle_gimple_call): Query
4021         the EVRP range analyzer for range data rather than using global data.
4022         * gimple-ssa-sprintf.c: Include alloc-pool.h, vr-values.h and
4023         gimple-ssa-evrp-analyze.h
4024         (class sprintf_dom_walker): Add after_dom_children member function.
4025         Add evrp_range_analyzer member.
4026         (sprintf_dom_walker::before_dom_children): Call into the EVRP
4027         range analyzer as needed.
4028         (sprintf_dom_walker::after_dom_children): New member function.
4029         * gimple-ssa-evrp-analyze.c (evrp_range_analyzer::enter): Do nothing
4030         if not optimizing.
4031         (evrp_range_analyzer::record_ranges_from_stmt): Likewise.
4032         (evrp_range_analyzer::pop_to_marker): Likewise.
4033
4034 2018-02-20  Richard Sandiford  <richard.sandiford@linaro.org>
4035
4036         PR tree-optimization/84419
4037         * internal-fn.c (expand_call_mem_ref): Create a TARGET_MEM_REF
4038         with the required type if its current type is compatible but
4039         different.
4040
4041 2018-02-20  Jakub Jelinek  <jakub@redhat.com>
4042
4043         PR middle-end/82004
4044         * match.pd (pow(C,x) -> exp(log(C)*x)): Delay all folding until
4045         after vectorization.
4046
4047 2018-02-20  Martin Liska  <mliska@suse.cz>
4048
4049         PR driver/83193
4050         * config/aarch64/aarch64.c (aarch64_print_hint_for_core_or_arch): Print
4051         possible values if we don't have a hint.
4052
4053 2018-02-20  Martin Liska  <mliska@suse.cz>
4054
4055         PR c/84310
4056         PR target/79747
4057         * final.c (shorten_branches): Build align_tab array with one
4058         more element.
4059         * opts.c (finish_options): Add alignment option limit check.
4060         (MAX_CODE_ALIGN): Likewise.
4061         (MAX_CODE_ALIGN_VALUE): Likewise.
4062         * doc/invoke.texi: Document maximum allowed option value for
4063         all -falign-* options.
4064
4065 2018-02-19  Jakub Jelinek  <jakub@redhat.com>
4066
4067         PR target/84146
4068         * reg-notes.def (REG_CALL_ARG_LOCATION): New reg note.
4069         * insn-notes.def (NOTE_INSN_CALL_ARG_LOCATION): Remove.
4070         * var-tracking.c (emit_note_insn_var_location): Remove all references
4071         to NOTE_INSN_CALL_ARG_LOCATION.
4072         (emit_notes_in_bb): Emit arguments as REG_CALL_ARG_LOCATION note on
4073         the CALL_INSN rather than separate NOTE_INSN_CALL_ARG_LOCATION note.
4074         Use copy_rtx_if_shared.
4075         * dwarf2out.c (gen_subprogram_die): Use XEXP with 0 instead of
4076         NOTE_VAR_LOCATION on ca_loc->call_arg_loc_note.
4077         (dwarf2out_var_location): Remove handling of
4078         NOTE_INSN_CALL_ARG_LOCATION, instead handle REG_CALL_ARG_LOCATION note
4079         on call_insn.
4080         * final.c (final_scan_insn): Remove all references to
4081         NOTE_INSN_CALL_ARG_LOCATION.
4082         (rest_of_clean_state): Likewise.  Remove REG_CALL_ARG_LOCATION notes
4083         before dumping final insns.
4084         * except.c (emit_note_eh_region_end): Remove all references to
4085         NOTE_INSN_CALL_ARG_LOCATION.
4086         * config/alpha/alpha.c (alpha_pad_function_end): Likewise.
4087         * config/c6x/c6x.c (c6x_gen_bundles): Likewise.
4088         * config/arc/arc.c (hwloop_optimize): Likewise.
4089         * config/arm/arm.c (create_fix_barrier): Likewise.
4090         * config/s390/s390.c (s390_chunkify_start): Likewise.
4091         * config/sh/sh.c (find_barrier): Likewise.
4092         * config/i386/i386.c (rest_of_insert_endbranch,
4093         ix86_seh_fixup_eh_fallthru): Likewise.
4094         * config/xtensa/xtensa.c (hwloop_optimize): Likewise.
4095         * config/iq2000/iq2000.c (final_prescan_insn): Likewise.
4096         * config/frv/frv.c (frv_function_prologue): Likewise.
4097         * emit-rtl.c (try_split): Likewise.  Copy over REG_CALL_ARG_LOCATION
4098         reg note.
4099         (note_outside_basic_block_p): Remove all references to
4100         NOTE_INSN_CALL_ARG_LOCATION.
4101         * gengtype.c (adjust_field_rtx_def): Likewise.
4102         * print-rtl.c (rtx_writer::print_rtx_operand_code_0, print_insn):
4103         Likewise.
4104         * jump.c (cleanup_barriers, delete_related_insns): Likewise.
4105         * cfgrtl.c (force_nonfallthru_and_redirect): Likewise.
4106
4107         PR c++/84444
4108         * builtins.c (builtin_mathfn_code): Don't check if CALL_EXPR_FN (t)
4109         is ADDR_EXPR.
4110
4111         PR tree-optimization/84452
4112         * tree-vect-patterns.c (vect_recog_pow_pattern): Don't call
4113         expand_simd_clones if targetm.simd_clone.compute_vecsize_and_simdlen
4114         is NULL.
4115
4116 2018-02-19  Martin Liska  <mliska@suse.cz>
4117
4118         PR sanitizer/82183
4119         * passes.def: Put pass_sancov_O0 before pass_lower_switch with -O0.
4120
4121 2018-02-19  Martin Liska  <mliska@suse.cz>
4122             Richard Sandiford  <richard.sandiford@linaro.org>
4123
4124         PR tree-optimization/82491
4125         * gimple-fold.c (get_base_constructor): Make earlier bail out
4126         to prevent ubsan.
4127
4128 2018-02-19  Carl Love  <cel@us.ibm.com>
4129
4130         * config/rs6000/rs6000-builtin.def: Change NEG macro expansions from
4131         BU_ALTIVEC_A to BU_P8V_AV_1 and BU_ALTIVEC_OVERLOAD_1 to
4132         BU_P8V_OVERLOAD_1.
4133         * config/rs6000/rs6000-c.c: Change ALTIVEC_BUILTIN_VEC_NEG to
4134         P8V_BUILTIN_VEC_NEG.
4135
4136 2018-02-19  Sebastian Perta  <sebastian.perta@renesas.com>
4137
4138         * config/rl78/rl78.md (movdf): New define expand.
4139
4140 2018-02-19  Martin Liska  <mliska@suse.cz>
4141
4142         PR other/80589
4143         * doc/invoke.texi: Fix typo.
4144         * params.def (PARAM_MAX_LOOP_HEADER_INSNS): Likewise.
4145
4146 2018-02-18  Segher Boessenkool  <segher@kernel.crashing.org>
4147
4148         * config/rs6000/rs6000.c (rs6000_option_override_internal): Don't
4149         handle rs6000_single_float and rs6000_double_float specially for
4150         e500 family CPUs.
4151
4152 2018-02-16  Jeff Law  <law@redhat.com>
4153
4154         * config/rx/rx.c (add_pop_cfi_notes): New function.;
4155         (pop_regs): Use it.
4156
4157 2018-02-16  Jakub Jelinek  <jakub@redhat.com>
4158
4159         PR ipa/84425
4160         * ipa-inline.c (inline_small_functions): Fix a typo.
4161
4162 2018-02-16  Nathan Sidwell  <nathan@acm.org>
4163
4164         * doc/extend.texi (Backwards Compatibility): Americanize 'behaviour'.
4165
4166 2018-02-16  Carl Love  <cel@us.ibm.com>
4167
4168         * config/rs6000/rs6000-builtin.def: Add BU_P8V_VSX_2 macro definition.
4169         Change FLOAT2 expansions from BU_VSX_2 to BU_P8V_VSX_2 and
4170         from BU_VSX_OVERLOAD_2 to BU_P8V_OVERLOAD_2.
4171         * config/rs6000/rs6000-c.c: Changed macro VSX_BUILTIN_VEC_FLOAT2
4172         expansion to P8V_BUILTIN_VEC_FLOAT2.
4173
4174 2018-02-16  Vladimir Makarov  <vmakarov@redhat.com>
4175
4176         PR rtl-optimization/70023
4177         * lra-constraints.c (inherit_in_ebb): Take hard reg mode of
4178         src_regno into account.
4179
4180 2018-02-16  Carl Love  <cel@us.ibm.com>
4181
4182         * config/rs6000/altivec.h: Remove vec_vextract4b and vec_vinsert4b.
4183         * config/rs6000/rs6000-builtin.def: Remove macro expansion for
4184         VEXTRACT4B, VINSERT4B, VINSERT4B_DI and VEXTRACT4B.
4185         * config/rs6000/rs6000.c: Remove case statements for
4186         P9V_BUILTIN_VEXTRACT4B, P9V_BUILTIN_VEC_VEXTRACT4B,
4187         P9V_BUILTIN_VINSERT4B, P9V_BUILTIN_VINSERT4B_DI,
4188         and P9V_BUILTIN_VEC_VINSERT4B.
4189         * config/rs6000/rs6000-c.c (altivec_expand_builtin): Remove entries for
4190         P9V_BUILTIN_VEC_VEXTRACT4B and P9V_BUILTIN_VEC_VINSERT4B.
4191         * config/rs6000/vsx.md:
4192         * doc/extend.texi: Remove vec_vextract4b, non ABI definitions for
4193         vec_insert4b.
4194
4195 2018-02-16  Carl Love  <cel@us.ibm.com>
4196
4197         * config/rs6000/altivec.h: Add builtin names vec_extract4b
4198         vec_insert4b.
4199         * config/rs6000/rs6000-builtin.def: Add INSERT4B and EXTRACT4B
4200         definitions.
4201         * config/rs6000/rs6000-c.c: Add the definitions for
4202         P9V_BUILTIN_VEC_EXTRACT4B and P9V_BUILTIN_VEC_INSERT4B.
4203         * config/rs6000/rs6000.c (altivec_expand_builtin): Add
4204         P9V_BUILTIN_EXTRACT4B and P9V_BUILTIN_INSERT4B case statements.
4205         * config/rs6000/vsx.md: Add define_insn extract4b.  Add define_expand
4206         definition for insert4b and define insn *insert3b_internal.
4207         * doc/extend.texi: Add documentation for vec_extract4b.
4208
4209 2018-02-16  Nathan Sidwell  <nathan@acm.org>
4210
4211         * doc/extend.texi (Backwards Compatibility): Mention friend
4212         injection.  Note for-scope is deprecated.
4213         * doc/invoke.texi (-ffriend-injection): Deprecate.
4214
4215 2018-02-16  Segher Boessenkool  <segher@kernel.crashing.org>
4216
4217         * combine.c (try_combine): When adjusting LOG_LINKS for the destination
4218         that moved to I2, also allow destinations that are a paradoxical
4219         subreg (instead of a normal reg).
4220
4221 2018-02-16  Oleg Endo  <olegendo@gcc.gnu.org>
4222
4223         PR target/83831
4224         * config/rx/rx.c (rx_fuse_in_memory_bitop): Convert shift operand
4225         to QImode.
4226
4227 2018-02-16  Richard Biener  <rguenther@suse.de>
4228
4229         PR tree-optimization/84037
4230         PR tree-optimization/84016
4231         PR target/82862
4232         * config/i386/i386.c (ix86_builtin_vectorization_cost):
4233         Adjust vec_construct for the fact we need additional higher latency
4234         128bit inserts for AVX256 and AVX512 vector builds.
4235         (ix86_add_stmt_cost): Scale vector construction cost for
4236         elementwise loads.
4237
4238 2018-02-16  Richard Biener  <rguenther@suse.de>
4239
4240         PR tree-optimization/84417
4241         * tree-ssa.c (non_rewritable_mem_ref_base): Properly constrain
4242         the MEM_REF offset when conversion to BIT_FIELD_REF is desired.
4243         (non_rewritable_lvalue_p): Likewise, use poly-ints.
4244
4245 2018-02-16  Martin Liska  <mliska@suse.cz>
4246
4247         PR sanitizer/84307
4248         * internal-fn.def (ASAN_CHECK): Set proper flags.
4249         (ASAN_MARK): Likewise.
4250
4251 2018-02-16  Julia Koval  <julia.koval@intel.com>
4252
4253         * config/i386/i386.c (ix86_option_override_internal): Remove PTA_CLWB
4254         from PTA_CANNONLAKE.
4255
4256 2018-02-16  Jakub Jelinek  <jakub@redhat.com>
4257
4258         PR target/84272
4259         * config/aarch64/cortex-a57-fma-steering.c (fma_forest::merge_forest):
4260         Use ++iter rather than iter++ for std::list iterators.
4261         (func_fma_steering::dfs): Likewise.  Don't delete nodes right away,
4262         defer deleting them until all nodes in the forest are processed.  Do
4263         free even leaf nodes.  Change to_process into auto_vec.
4264
4265         PR bootstrap/84405
4266         * system.h (BROKEN_VALUE_INITIALIZATION): Define for GCC < 4.3.
4267         * vec.h (vec_default_construct): Use memset instead of placement new
4268         if BROKEN_VALUE_INITIALIZATION is defined.
4269         * hash-table.h (hash_table<Descriptor, Allocator>::empty_slow): Use
4270         memset instead of value initialization if BROKEN_VALUE_INITIALIZATION
4271         is defined.
4272
4273         PR rtl-optimization/83723
4274         * lra-int.h (lra_substitute_pseudo): Add DEBUG_P argument.
4275         * lra.c (lra_substitute_pseudo): Likewise.  If true, use
4276         gen_rtx_raw_SUBREG instead of gen_rtx_SUBREG.  Pass DEBUG_P to
4277         recursive calls.
4278         (lra_substitute_pseudo_within_insn): Adjust lra_substitute_pseudo
4279         callers.
4280         * lra-constraints.c (inherit_reload_reg, split_reg): Likewise.
4281
4282 2018-02-16  Eric Botcazou  <ebotcazou@adacore.com>
4283
4284         PR rtl-optimization/81443
4285         * rtlanal.c (num_sign_bit_copies1) <SUBREG>: Do not propagate results
4286         from inner REGs to paradoxical SUBREGs.
4287
4288 2018-02-16  Richard Biener  <rguenther@suse.de>
4289
4290         PR tree-optimization/84399
4291         * graphite-scop-detection.c (scop_detection::stmt_simple_for_scop_p):
4292         For operands we can analyze at their definition make sure we can
4293         analyze them at each use as well.
4294
4295 2018-02-16  Richard Biener  <rguenther@suse.de>
4296
4297         PR tree-optimization/84190
4298         * tree-ssa.c (non_rewritable_mem_ref_base): Do not touch
4299         volatile accesses if the decl isn't volatile.
4300
4301 2018-02-15  Jason Merrill  <jason@redhat.com>
4302
4303         PR c++/84314 - ICE with templates and fastcall attribute.
4304         * attribs.c (build_type_attribute_qual_variant): Don't clobber
4305         TYPE_CANONICAL on an existing type.
4306
4307 2018-02-15  Jakub Jelinek  <jakub@redhat.com>
4308
4309         PR tree-optimization/84383
4310         * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Don't look at
4311         dstoff nor call operand_equal_p if dstbase is NULL.
4312
4313         PR tree-optimization/84334
4314         * match.pd ((A +- CST1) +- CST2 -> A + CST3): If A is
4315         also a CONSTANT_CLASS_P, punt.
4316
4317 2018-02-14  Jim Wilson  <jimw@sifive.com>
4318
4319         * config/riscv/riscv.c (riscv_first_stack_step): Move locals after
4320         first SMALL_OPERAND check.  New local min_second_step.  Move assert
4321         to where locals are set.  Add TARGET_RVC support.
4322         * config/riscv/riscv.h (C_SxSP_BITS, SWSP_REACH, SDSP_REACH): New.
4323
4324 2018-02-14  Indu Bhagat  <indu.bhagat@oracle.com>
4325
4326         * doc/invoke.texi: Correct -Wformat-overflow code sample.
4327
4328 2018-02-14  Martin Sebor  <msebor@redhat.com>
4329
4330         PR tree-optimization/83698
4331         * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For
4332         arrays constrain the offset range to their bounds.
4333         (builtin_access::strcat_overlap): Adjust the bounds of overlap offset.
4334         (builtin_access::overlap): Avoid setting the size of overlap if it's
4335         already been set.
4336         (maybe_diag_overlap): Also consider arrays when deciding what values
4337         of offsets to include in diagnostics.
4338
4339 2018-02-14  Martin Sebor  <msebor@redhat.com>
4340
4341         PR c/84108
4342         * attribs.c (diag_attr_exclusions): Consider the exclusion(s)
4343         that correspond to the kind of a declaration.
4344
4345 2018-02-14  John David Anglin  <danglin@gcc.gnu.org>
4346
4347         PR target/83984
4348         * config/pa/pa.md: Load address of PIC label using the linkage table
4349         if the label is nonlocal.
4350
4351 2018-02-14  Kelvin Nilsen  <kelvin@gcc.gnu.org>
4352
4353         * config/rs6000/rs6000.c (rs6000_option_override_internal): Issue
4354         warning message if user requests -maltivec=be.
4355         * doc/invoke.texi: Document deprecation of -maltivec=be.
4356
4357 2018-02-14  Will Schmidt  <will_schmidt@vnet.ibm.com>
4358
4359         PR target/84220
4360         * config/rs6000/rs6000-c.c: Update definitions for
4361         ALTIVEC_BUILTIN_VEC_SLD, ALTIVEC_BUILTIN_VEC_SLDW,
4362         VEC_XXSLDWI and ALTIVEC_BUILTIN_VEC_XXPERMDI builtins.
4363
4364 2018-02-14  Igor Tsimbalist  <igor.v.tsimbalist@intel.com>
4365
4366         PR target/84239
4367         * config/i386/cetintrin.h: Remove _rdssp[d|q] and
4368         add _get_ssp intrinsics. Remove argument from
4369         __builtin_ia32_rdssp[d|q].
4370         * config/i386/i386-builtin-types.def: Add UINT_FTYPE_VOID.
4371         * config/i386/i386-builtin.def: Remove argument from
4372         __builtin_ia32_rdssp[d|q].
4373         * config/i386/i386.c: Use UINT_FTYPE_VOID. Use
4374         ix86_expand_special_args_builtin for _rdssp[d|q].
4375         * config/i386/i386.md: Remove argument from rdssp[si|di] insn.
4376         Clear register before usage.
4377         * doc/extend.texi: Remove argument from __builtin_ia32_rdssp[d|q].
4378         Add documentation for new _get_ssp and _inc_ssp intrinsics.
4379
4380 2018-02-14  Richard Sandiford  <richard.sandiford@linaro.org>
4381
4382         PR tree-optimization/84357
4383         * tree-data-ref.c (object_address_invariant_in_loop_p): Check
4384         operand 1 of an ARRAY_REF too.
4385
4386 2018-02-14  Oleg Endo  <olegendo@gcc.gnu.org>
4387
4388         PR target/83831
4389         * config/rx/rx-protos.h (rx_reg_dead_or_unused_after_insn,
4390         rx_copy_reg_dead_or_unused_notes, rx_fuse_in_memory_bitop): New
4391         declarations.
4392         (set_of_reg): New struct.
4393         (rx_find_set_of_reg, rx_find_use_of_reg): New functions.
4394         * config/rx/rx.c (rx_reg_dead_or_unused_after_insn,
4395         rx_copy_reg_dead_or_unused_notes, rx_fuse_in_memory_bitop): New
4396         functions.
4397         * config/rx/rx.md (andsi3, iorsi3, xorsi3): Convert to insn_and_split.
4398         Split into bitclr, bitset, bitinvert patterns if appropriate.
4399         (*bitset, *bitinvert, *bitclr): Convert to named insn_and_split and
4400         use rx_fuse_in_memory_bitop.
4401         (*bitset_in_memory, *bitinvert_in_memory, *bitclr_in_memory): Convert
4402         to named insn, correct maximum insn length.
4403
4404 2018-02-14  Jozef Lawrynowicz  <jozefl.gcc@gmail.com>
4405
4406         PR target/79242
4407         * machmode.def: Define a complex mode for PARTIAL_INT.
4408         * genmodes.c (complex_class): Return MODE_COMPLEX_INT for
4409         MODE_PARTIAL_INT.
4410         * doc/rtl.texi: Document CSPImode.
4411         * config/msp430/msp430.c (msp430_hard_regno_nregs): Add CPSImode
4412         handling.
4413         (msp430_hard_regno_nregs_with_padding): Likewise.
4414
4415 2018-02-13  Peter Bergner  <bergner@vnet.ibm.com>
4416
4417         PR target/84279
4418         * config/rs6000/rs6000.c (mem_operand_gpr): Disallow altivec addresses.
4419
4420 2018-02-13  Segher Boessenkool  <segher@kernel.crashing.org>
4421
4422         PR rtl-optimization/84169
4423         * combine.c (try_combine): New variable split_i2i3.  Set it to true if
4424         we generated a parallel as new i3 and we split that to new i2 and i3
4425         instructions.  Handle split_i2i3 similar to swap_i2i3: scan the
4426         LOG_LINKs of i3 to see which of those need to link to i2 now.  Link
4427         those to i2, not i1.  Partially rewrite this scan code.
4428
4429 2018-02-13  Jakub Jelinek  <jakub@redhat.com>
4430
4431         PR c/82210
4432         * stor-layout.c (place_field): For variable length fields, adjust
4433         offset_align afterwards not just based on the field's alignment,
4434         but also on the size.
4435
4436         PR middle-end/84309
4437         * match.pd (pow(C,x) -> exp(log(C)*x)): Use exp2s and log2s instead
4438         of exps and logs in the use_exp2 case.
4439
4440 2018-02-13  Jeff Law  <law@redhat.com>
4441
4442         * config/rl/rl78.c (rl78_attribute_table): Fix terminator and
4443         entry for "vector".
4444
4445         * config/rl78/rl78.c (rl78_handle_func_attribute): Mark
4446         ARGS as unused.
4447
4448 2018-02-13  Alexandre Oliva  <aoliva@redhat.com>
4449
4450         PR debug/84342
4451         PR debug/84319
4452         * common.opt (gas-loc-support, gas-locview-support): New.
4453         (ginline-points, ginternal-reset-location-views): New.
4454         * doc/invoke.texi: Document them.  Use @itemx where intended.
4455         (gvariable-location-views): Adjust.
4456         * target.def (reset_location_view): New.
4457         * doc/tm.texi.in (DWARF2_ASM_VIEW_DEBUG_INFO): New.
4458         (TARGET_RESET_LOCATION_VIEW): New.
4459         * doc/tm.texi: Rebuilt.
4460         * dwarf2out.c (dwarf2out_default_as_loc_support): New.
4461         (dwarf2out_default_as_locview_support): New.
4462         (output_asm_line_debug_info): Use option variables.
4463         (dwarf2out_maybe_output_loclist_view_pair): Likewise.
4464         (output_loc_list): Likewise.
4465         (add_high_low_attributes): Check option variables.
4466         Don't output entry view attribute in strict mode.
4467         (gen_inlined_subroutine_die): Check option variables.
4468         (dwarf2out_inline_entry): Likewise.
4469         (init_sections_and_labels): Likewise.
4470         (dwarf2out_early_finish): Likewise.
4471         (maybe_reset_location_view): New, from...
4472         (dwarf2out_var_location): ... here.  Call it.
4473         * debug.h (dwarf2out_default_as_loc_support): Declare.
4474         (dwarf2out_default_as_locview_support): Declare.
4475         * hooks.c (hook_int_rtx_insn_0): New.
4476         * hooks.h (hook_int_rtx_insn_0): Declare.
4477         * toplev.c (process_options): Take -gas-loc-support and
4478         -gas-locview-support from dwarf2out.  Enable
4479         -gvariable-location-views by default only with locview
4480         assembler support.  Enable -ginternal-reset-location-views by
4481         default only if the target defines the corresponding hook.
4482         Enable -ginline-points by default if location views are
4483         enabled; force it disabled if statement frontiers are
4484         disabled.
4485         * tree-inline.c (expand_call_inline): Check option variables.
4486         * tree-ssa-live.c (remove_unused_scope_block_p): Likewise.
4487
4488 2018-02-13  Richard Sandiford  <richard.sandiford@linaro.org>
4489
4490         PR tree-optimization/84321
4491         * tree-vrp.c (intersect_range_with_nonzero_bits): Fix VR_ANTI_RANGE
4492         handling.  Also check whether the anti-range contains any values
4493         that satisfy the mask; switch to a VR_RANGE if not.
4494
4495 2018-02-13  Paolo Bonzini  <bonzini@gnu.org>
4496
4497         PR sanitizer/84340
4498         * internal-fn.def (ASAN_CHECK, ASAN_MARK): Revert changes to fnspec.
4499
4500 2018-02-13  Martin Jambor  <mjambor@suse.cz>
4501
4502         PR c++/83990
4503         * ipa-param-manipulation.c (ipa_modify_call_arguments): Use location
4504         of call statements, also set location of a load to a temporary.
4505
4506 2018-02-13  Sebastian Perta  <sebastian.perta@renesas.com>
4507
4508         * config/rl78/rl78.c (add_vector_labels): New function.
4509         * config/rl78/rl78.c (rl78_handle_vector_attribute): New function.
4510         * config/rl78/rl78.c (rl78_start_function): Call add_vector_labels.
4511         * config/rl78/rl78.c (rl78_handle_func_attribute): Removed the assert
4512         which checks that no arguments are passed.
4513         * config/rl78/rl78.c (rl78_attribute_table): Add "vector" attribute.
4514         * doc/extend.texi: Documentation for the new attribute.
4515
4516 2018-02-13  Andreas Schwab  <schwab@suse.de>
4517
4518         * config/riscv/linux.h (CPP_SPEC): Define.
4519
4520 2018-02-13  Jakub Jelinek  <jakub@redhat.com>
4521
4522         PR target/84335
4523         * config/i386/i386.c (ix86_init_mmx_sse_builtins): Pass
4524         OPTION_MASK_ISA_AES | OPTION_MASK_ISA_SSE2 instead of
4525         OPTION_MASK_ISA_AES as first argument to def_builtin_const
4526         for AES builtins.  Pass OPTION_MASK_ISA_PCLMUL | OPTION_MASK_ISA_SSE2
4527         instead of OPTION_MASK_ISA_PCLMUL as first argument to
4528         def_builtin_const for __builtin_ia32_pclmulqdq128 builtin.
4529         * config/i386/wmmintrin.h: If __SSE2__ is not defined, enable it
4530         temporarily for AES and PCLMUL builtins.
4531
4532         PR tree-optimization/84339
4533         * gimple-fold.c (get_range_strlen): Set *FLEXP to true when handling
4534         ARRAY_REF where first operand is array_at_struct_end_p COMPONENT_REF.
4535         Formatting fixes.
4536
4537         PR middle-end/84309
4538         * match.pd (pow(C,x) -> exp(log(C)*x)): Optimize instead into
4539         exp2(log2(C)*x) if C is a power of 2 and c99 runtime is available.
4540         * generic-match-head.c (canonicalize_math_after_vectorization_p): New
4541         inline function.
4542         * gimple-match-head.c (canonicalize_math_after_vectorization_p): New
4543         inline function.
4544         * omp-simd-clone.h: New file.
4545         * omp-simd-clone.c: Include omp-simd-clone.h.
4546         (expand_simd_clones): No longer static.
4547         * tree-vect-patterns.c: Include fold-const-call.h, attribs.h,
4548         cgraph.h and omp-simd-clone.h.
4549         (vect_recog_pow_pattern): Optimize pow(C,x) to exp(log(C)*x).
4550         (vect_recog_widen_shift_pattern): Formatting fix.
4551         (vect_pattern_recog_1): Don't check optab for calls.
4552
4553         PR target/84336
4554         * config/i386/sse.md (<avx512>_vpermi2var<mode>3_mask): Force
4555         operands[2] into a REG before using gen_lowpart on it.
4556
4557 2018-02-12  Jeff Law  <law@redhat.com>
4558
4559         PR target/83760
4560         * config/sh/sh.c (find_barrier): Consider a sibling call
4561         a barrier as well.
4562
4563         * cse.c (try_back_substitute_reg): Move any REG_ARGS_SIZE note when
4564         successfully back substituting a reg.
4565
4566 2018-02-12  Richard Biener  <rguenther@suse.de>
4567
4568         PR tree-optimization/84037
4569         * tree-vect-slp.c (vect_analyze_slp_cost): Add visited
4570         parameter, move visited init to caller.
4571         (vect_slp_analyze_operations): Separate cost from validity
4572         check, initialize visited once for all instances.
4573         (vect_schedule_slp): Analyze map to CSE vectorized nodes once
4574         for all instances.
4575         * tree-vect-stmts.c (vect_model_simple_cost): Make early
4576         out an assert.
4577         (vect_model_promotion_demotion_cost): Likewise.
4578         (vectorizable_bswap): Guard cost modeling with !slp_node
4579         instead of !PURE_SLP_STMT to avoid double-counting on hybrid
4580         SLP stmts.
4581         (vectorizable_call): Likewise.
4582         (vectorizable_conversion): Likewise.
4583         (vectorizable_assignment): Likewise.
4584         (vectorizable_shift): Likewise.
4585         (vectorizable_operation): Likewise.
4586         (vectorizable_store): Likewise.
4587         (vectorizable_load): Likewise.
4588         (vectorizable_condition): Likewise.
4589         (vectorizable_comparison): Likewise.
4590
4591 2018-02-12  Paolo Bonzini  <bonzini@gnu.org>
4592
4593         PR sanitizer/84307
4594         * internal-fn.def (ASAN_CHECK): Fix fnspec to account for return value.
4595         (ASAN_MARK): Fix fnspec to account for return value, change pointer
4596         argument from 'R' to 'W' so that the pointed-to datum is clobbered.
4597
4598 2018-02-08  Jan Hubicka  <hubicka@ucw.cz>
4599
4600         PR middle-end/83665
4601         * params.def (inline-min-speedup): Increase from 8 to 15.
4602         (max-inline-insns-auto): Decrease from 40 to 30.
4603         * ipa-split.c (consider_split): Add some buffer for function to
4604         be considered inlining candidate.
4605         * invoke.texi (max-inline-insns-auto, inline-min-speedup): UPdate
4606         default values.
4607
4608 2018-02-12  Richard Biener  <rguenther@suse.de>
4609
4610         PR tree-optimization/84037
4611         * tree-vect-slp.c (vect_build_slp_tree_2): Try swapping the
4612         matched stmts if we cannot swap the non-matched ones.
4613
4614 2018-02-12  Olga Makhotina  <olga.makhotina@intel.com>
4615
4616         * config/i386/avx512fintrin.h (_mm_mask_scalef_round_sd,
4617         _mm_maskz_scalef_round_sd, _mm_mask_scalef_round_ss,
4618         _mm_maskz_scalef_round_ss): New intrinsics.
4619         (__builtin_ia32_scalefsd_round, __builtin_ia32_scalefss_round): Fix.
4620         * config/i386/i386-builtin.def (__builtin_ia32_scalefsd_round,
4621         __builtin_ia32_scalefss_round): Remove.
4622         (__builtin_ia32_scalefsd_mask_round,
4623         __builtin_ia32_scalefss_mask_round): New intrinsics.
4624         * config/i386/sse.md (vmscalef<mode><round_name>): Renamed to ...
4625         (vmscalef<mode><mask_scalar_name><round_scalar_name>): ... this.
4626         ((match_operand:VF_128 2 "<round_nimm_predicate>"
4627         "<round_constraint>")): Changed to ...
4628         ((match_operand:VF_128 2 "<round_scalar_nimm_predicate>"
4629         "<round_scalar_constraint>")): ... this.
4630         ("vscalef<ssescalarmodesuffix>\t{<round_op3>%2, %1, %0|
4631         %0, %1, %2<round_op3>}"): Changed to ...
4632         ("vscalef<ssescalarmodesuffix>\t{<round_scalar_mask_op3>%2, %1,
4633         %0<mask_scalar_operand3>|%0<mask_scalar_operand3>, %1,
4634         %2<round_scalar_mask_op3>}"): ... this.
4635         * config/i386/subst.md (round_scalar_nimm_predicate): New.
4636
4637 2018-02-12  Olga Makhotina  <olga.makhotina@intel.com>
4638
4639         * config/i386/avx512fintrin.h (_mm_mask_sqrt_round_sd)
4640         (_mm_maskz_sqrt_round_sd, _mm_mask_sqrt_round_ss)
4641         (_mm_maskz_sqrt_round_ss): New intrinsics.
4642         (__builtin_ia32_sqrtsd_round, __builtin_ia32_sqrtss_round): Remove.
4643         (__builtin_ia32_sqrtsd_mask_round)
4644         (__builtin_ia32_sqrtss_mask_round): New builtins.
4645         * config/i386/i386-builtin.def (__builtin_ia32_sqrtsd_round)
4646         (__builtin_ia32_sqrtss_round): Remove.
4647         (__builtin_ia32_sqrtsd_mask_round)
4648         (__builtin_ia32_sqrtss_mask_round): New builtins.
4649         * config/i386/sse.md (vmsqrt<mode>2<round_name>): Renamed to ...
4650         (vmsqrt<mode>2<mask_scalar_name><round_scalar_name>): ... this.
4651         ((match_operand:VF_128 1 "vector_operand"
4652         "xBm,<round_constraint>")): Changed to ...
4653         ((match_operand:VF_128 1 "vector_operand"
4654         "xBm,<round_scalar_constraint>")): ... this.
4655         (vsqrt<ssescalarmodesuffix>\t{<round_op3>%1, %2, %0|
4656         %0, %2, %<iptr>1<round_op3>}): Changed to ...
4657         (vsqrt<ssescalarmodesuffix>\t{<round_scalar_mask_op3>%1, %2,
4658         %0<mask_scalar_operand3>|%0<mask_scalar_operand3>, %2,
4659         %<iptr>1<round_scalar_mask_op3>}): ... this.
4660         ((set_attr "prefix" "<round_prefix>")): Changed to ...
4661         ((set_attr "prefix" "<round_scalar_prefix>")): ... this.
4662
4663 2018-02-11  Steven Munroe  <munroesj@gcc.gnu.org>
4664
4665         PR target/84266
4666         * config/rs6000/mmintrin.h (_mm_cmpeq_pi32 [_ARCH_PWR9]):
4667         Cast vec_cmpeq result to correct type.
4668         * config/rs6000/mmintrin.h (_mm_cmpgt_pi32 [_ARCH_PWR9]):
4669         Cast vec_cmpgt result to correct type.
4670
4671 2018-02-11  Alexandre Oliva  <aoliva@redhat.com>
4672
4673         * final.c (final_scan_insn_1): Renamed from...
4674         (final_scan_insn): ... this.  New wrapper, to recover
4675         seen from the outermost call in recursive ones.
4676         * config/sparc/sparc.c (output_return): Drop seen from call.
4677         (output_sibcall): Likewise.
4678         * config/visium/visium.c (output_branch): Likewise.
4679
4680 2018-02-10  John David Anglin  <danglin@gcc.gnu.org>
4681
4682         * config/pa/pa.c (hppa_profile_hook): Mark SYMBOL_REF for _mcount as
4683         function label.
4684
4685 2018-02-10  Alan Modra  <amodra@gmail.com>
4686
4687         PR target/84300
4688         * config/rs6000/rs6000.md (split_stack_return): Remove (use ..).
4689         Specify LR as an input.
4690
4691 2018-02-10  Jakub Jelinek  <jakub@redhat.com>
4692
4693         PR sanitizer/83987
4694         * omp-low.c (maybe_remove_omp_member_access_dummy_vars,
4695         remove_member_access_dummy_vars): New functions.
4696         (lower_omp_for, lower_omp_taskreg, lower_omp_target,
4697         lower_omp_1, execute_lower_omp): Use them.
4698
4699         PR rtl-optimization/84308
4700         * shrink-wrap.c (spread_components): Release todo vector.
4701
4702 2018-02-09  Vladimir Makarov  <vmakarov@redhat.com>
4703
4704         PR rtl-optimization/57193
4705         * ira-color.c (struct allocno_color_data): Add member
4706         conflict_allocno_hard_prefs.
4707         (update_conflict_allocno_hard_prefs): New.
4708         (bucket_allocno_compare_func): Add a preference based on
4709         conflict_allocno_hard_prefs.
4710         (push_allocno_to_stack): Update conflict_allocno_hard_prefs.
4711         (color_allocnos): Remove a dead code.  Initiate
4712         conflict_allocno_hard_prefs.  Call update_costs_from_prefs.
4713
4714 2018-02-09  Jakub Jelinek  <jakub@redhat.com>
4715
4716         PR target/84226
4717         * config/rs6000/vsx.md (p9_xxbrq_v16qi): Change input operand
4718         constraint from =wa to wa.  Avoid a subreg on the output operand,
4719         instead use a pseudo and subreg it in a move.
4720         (p9_xxbrd_<mode>): Changed to ...
4721         (p9_xxbrd_v2di): ... this insn, without VSX_D iterator.
4722         (p9_xxbrd_v2df): New expander.
4723         (p9_xxbrw_<mode>): Changed to ...
4724         (p9_xxbrw_v4si): ... this insn, without VSX_W iterator.
4725         (p9_xxbrw_v4sf): New expander.
4726
4727 2018-02-09  Sebastian Perta  <sebastian.perta@renesas.com>
4728
4729         * config/rx/rx.md (movsicc): Update expander to be matched by GCC.
4730
4731 2018-02-09  Peter Bergner  <bergner@vnet.ibm.com>
4732
4733         PR target/83926
4734         * config/rs6000/vsx.md (vsx_mul_v2di): Handle generating a 64-bit
4735         multiply in 32-bit mode.
4736         (vsx_div_v2di): Handle generating a 64-bit signed divide in 32-bit mode.
4737         (vsx_udiv_v2di): Handle generating a 64-bit unsigned divide in 32-bit
4738         mode.
4739
4740 2018-02-09  Sebastian Perta  <sebastian.perta@renesas.com>
4741
4742         * config/rx/constraints.md (CALL_OP_SYMBOL_REF): Added new constraint
4743         to allow or block "symbol_ref" depending on the value of TARGET_JSR.
4744         * config/rx/rx.md (call_internal): Use CALL_OP_SYMBOL_REF.
4745         * config/rx/rx.md (call_value_internal): Use CALL_OP_SYMBOL_REF.
4746
4747 2018-02-09  Pierre-Marie de Rodat  <derodat@adacore.com>
4748
4749         PR lto/84213
4750         * dwarf2out.c (is_trivial_indirect_ref): New function.
4751         (dwarf2out_late_global_decl): Do not generate a location
4752         attribute for variables that have a non-trivial DECL_VALUE_EXPR
4753         and that are not defined in the current unit.
4754
4755 2018-02-09  Eric Botcazou  <ebotcazou@adacore.com>
4756
4757         * optabs.c (prepare_cmp_insn): Try harder to emit a direct comparison
4758         instead of a libcall for UNORDERED.
4759
4760 2018-02-09  Tamar Christina  <tamar.christina@arm.com>
4761
4762         PR target/82641
4763         * config/arm/arm-c.c (arm_cpu_builtins): Un-define __ARM_FEATURE_LDREX,
4764         __ARM_ARCH_PROFILE, __ARM_ARCH_ISA_THUMB, __ARM_FP and __ARM_NEON_FP.
4765
4766 2018-02-09  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>
4767
4768         PR target/PR84295
4769         * config/s390/s390.c (s390_set_current_function): Invoke
4770         s390_indirect_branch_settings also if fndecl didn't change.
4771
4772 2018-02-09  Alexandre Oliva  <aoliva@redhat.com>
4773
4774         * config/rs6000/rs6000.md (blockage): Set length to zero.
4775
4776 2018-02-09  Eric Botcazou  <ebotcazou@adacore.com>
4777
4778         * expr.c (optimize_bitfield_assignment_op): Remove obsolete assertion.
4779
4780 2018-02-09  Jakub Jelinek  <jakub@redhat.com>
4781
4782         PR sanitizer/84285
4783         * gcc.c (STATIC_LIBASAN_LIBS, STATIC_LIBTSAN_LIBS,
4784         STATIC_LIBLSAN_LIBS, STATIC_LIBUBSAN_LIBS): Handle -static like
4785         -static-lib*san.
4786
4787         PR debug/84252
4788         * var-tracking.c (vt_add_function_parameter): Punt for non-onepart
4789         PARALLEL incoming that failed vt_get_decl_and_offset check.
4790
4791         PR middle-end/84237
4792         * output.h (bss_initializer_p): Add NAMED argument, defaulted to false.
4793         * varasm.c (bss_initializer_p): Add NAMED argument, if true, ignore
4794         TREE_READONLY bit.
4795         (get_variable_section): For decls in named .bss* sections pass true as
4796         second argument to bss_initializer_p.
4797
4798 2018-02-09  Marek Polacek  <polacek@redhat.com>
4799             Jakub Jelinek  <jakub@redhat.com>
4800
4801         PR c++/83659
4802         * fold-const.c (fold_indirect_ref_1): Use VECTOR_TYPE_P macro.
4803         Formatting fixes.  Verify first that tree_fits_poly_int64_p (op01).
4804         Sync some changes from cxx_fold_indirect_ref.
4805
4806 2018-02-09  Alexandre Oliva  <aoliva@redhat.com>
4807
4808         * cfgexpand.c (expand_gimple_basic_block): Handle inline entry
4809         markers.
4810         * dwarf2out.c (dwarf2_debug_hooks): Enable inline_entry hook.
4811         (BLOCK_INLINE_ENTRY_LABEL): New.
4812         (dwarf2out_var_location): Disregard inline entry markers.
4813         (inline_entry_data): New struct.
4814         (inline_entry_data_hasher): New hashtable type.
4815         (inline_entry_data_hasher::hash): New.
4816         (inline_entry_data_hasher::equal): New.
4817         (inline_entry_data_table): New variable.
4818         (add_high_low_attributes): Add DW_AT_entry_pc and
4819         DW_AT_GNU_entry_view attributes if a pending entry is found
4820         in inline_entry_data_table.  Add old entry_pc attribute only
4821         if debug nonbinding markers are disabled.
4822         (gen_inlined_subroutine_die): Set BLOCK_DIE if nonbinding
4823         markers are enabled.
4824         (block_within_block_p, dwarf2out_inline_entry): New.
4825         (dwarf2out_finish): Check that no entries remained in
4826         inline_entry_data_table.
4827         * final.c (reemit_insn_block_notes): Handle inline entry notes.
4828         (final_scan_insn, notice_source_line): Likewise.
4829         (rest_of_clean_state): Skip inline entry markers.
4830         * gimple-pretty-print.c (dump_gimple_debug): Handle inline entry
4831         markers.
4832         * gimple.c (gimple_build_debug_inline_entry): New.
4833         * gimple.h (enum gimple_debug_subcode): Add
4834         GIMPLE_DEBUG_INLINE_ENTRY.
4835         (gimple_build_debug_inline_entry): Declare.
4836         (gimple_debug_inline_entry_p): New.
4837         (gimple_debug_nonbind_marker_p): Adjust.
4838         * insn-notes.def (INLINE_ENTRY): New.
4839         * print-rtl.c (rtx_writer::print_rtx_operand_code_0): Handle
4840         inline entry marker notes.
4841         (print_insn): Likewise.
4842         * rtl.h (NOTE_MARKER_P): Add INLINE_ENTRY support.
4843         (INSN_DEBUG_MARKER_KIND): Likewise.
4844         (GEN_RTX_DEBUG_MARKER_INLINE_ENTRY_PAT): New.
4845         * tree-inline.c (expand_call_inline): Build and insert
4846         debug_inline_entry stmt.
4847         * tree-ssa-live.c (remove_unused_scope_block_p): Preserve
4848         inline entry blocks early, if nonbind markers are enabled.
4849         (dump_scope_block): Dump fragment info.
4850         * var-tracking.c (reemit_marker_as_note): Handle inline entry note.
4851         * doc/gimple.texi (gimple_debug_inline_entry_p): New.
4852         (gimple_build_debug_inline_entry): New.
4853         * doc/invoke.texi (gstatement-frontiers, gno-statement-frontiers):
4854         Enable/disable inline entry points too.
4855         * doc/rtl.texi (NOTE_INSN_INLINE_ENTRY): New.
4856         (DEBUG_INSN): Describe inline entry markers.
4857
4858         * common.opt (gvariable-location-views): New.
4859         (gvariable-location-views=incompat5): New.
4860         * config.in: Rebuilt.
4861         * configure: Rebuilt.
4862         * configure.ac: Test assembler for view support.
4863         * dwarf2asm.c (dw2_asm_output_symname_uleb128): New.
4864         * dwarf2asm.h (dw2_asm_output_symname_uleb128): Declare.
4865         * dwarf2out.c (var_loc_view): New typedef.
4866         (struct dw_loc_list_struct): Add vl_symbol, vbegin, vend.
4867         (dwarf2out_locviews_in_attribute): New.
4868         (dwarf2out_locviews_in_loclist): New.
4869         (dw_val_equal_p): Compare val_view_list of dw_val_class_view_lists.
4870         (enum dw_line_info_opcode): Add LI_adv_address.
4871         (struct dw_line_info_table): Add view.
4872         (RESET_NEXT_VIEW, RESETTING_VIEW_P): New macros.
4873         (DWARF2_ASM_VIEW_DEBUG_INFO): Define default.
4874         (zero_view_p): New variable.
4875         (ZERO_VIEW_P): New macro.
4876         (output_asm_line_debug_info): New.
4877         (struct var_loc_node): Add view.
4878         (add_AT_view_list, AT_loc_list): New.
4879         (add_var_loc_to_decl): Add view param.  Test it against last.
4880         (new_loc_list): Add view params.  Record them.
4881         (AT_loc_list_ptr): Handle loc and view lists.
4882         (view_list_to_loc_list_val_node): New.
4883         (print_dw_val): Handle dw_val_class_view_list.
4884         (size_of_die): Likewise.
4885         (value_format): Likewise.
4886         (loc_list_has_views): New.
4887         (gen_llsym): Set vl_symbol too.
4888         (maybe_gen_llsym, skip_loc_list_entry): New.
4889         (dwarf2out_maybe_output_loclist_view_pair): New.
4890         (output_loc_list): Output view list or entries too.
4891         (output_view_list_offset): New.
4892         (output_die): Handle dw_val_class_view_list.
4893         (output_dwarf_version): New.
4894         (output_compilation_unit_header): Use it.
4895         (output_skeleton_debug_sections): Likewise.
4896         (output_rnglists, output_line_info): Likewise.
4897         (output_pubnames, output_aranges): Update version comments.
4898         (output_one_line_info_table): Output view numbers in asm comments.
4899         (dw_loc_list): Determine current endview, pass it to new_loc_list.
4900         Call maybe_gen_llsym.
4901         (loc_list_from_tree_1): Adjust.
4902         (add_AT_location_description): Create view list attribute if
4903         needed, check it's absent otherwise.
4904         (convert_cfa_to_fb_loc_list): Adjust.
4905         (maybe_emit_file): Call output_asm_line_debug_info for test.
4906         (dwarf2out_var_location): Reset views as needed.  Precompute
4907         add_var_loc_to_decl args.  Call get_attr_min_length only if we have the
4908         attribute.  Set view.
4909         (new_line_info_table): Reset next view.
4910         (set_cur_line_info_table): Call output_asm_line_debug_info for test.
4911         (dwarf2out_source_line): Likewise.  Output view resets and labels to
4912         the assembler, or select appropriate line info opcodes.
4913         (prune_unused_types_walk_attribs): Handle dw_val_class_view_list.
4914         (optimize_string_length): Catch it.  Adjust.
4915         (resolve_addr): Copy vl_symbol along with ll_symbol.  Handle
4916         dw_val_class_view_list, and remove it if no longer needed.
4917         (hash_loc_list): Hash view numbers.
4918         (loc_list_hasher::equal): Compare them.
4919         (optimize_location_lists): Check whether a view list symbol is
4920         needed, and whether the locview attribute is present, and
4921         whether they match.  Remove the locview attribute if no longer
4922         needed.
4923         (index_location_lists): Call skip_loc_list_entry for test.
4924         (dwarf2out_finish): Call output_asm_line_debug_info for test.
4925         Use output_dwarf_version.
4926         * dwarf2out.h (enum dw_val_class): Add dw_val_class_view_list.
4927         (struct dw_val_node): Add val_view_list.
4928         * final.c (SEEN_NEXT_VIEW): New.
4929         (set_next_view_needed): New.
4930         (clear_next_view_needed): New.
4931         (maybe_output_next_view): New.
4932         (final_start_function): Rename to...
4933         (final_start_function_1): ... this.  Take pointer to FIRST,
4934         add SEEN parameter.  Emit param bindings in the initial view.
4935         (final_start_function): Reintroduce SEEN-less interface.
4936         (final): Rename to...
4937         (final_1): ... this.  Take SEEN parameter.  Output final pending
4938         next view at the end.
4939         (final): Reintroduce seen-less interface.
4940         (final_scan_insn): Output pending next view before switching
4941         sections or ending a block.  Mark the next view as needed when
4942         outputting variable locations.  Notify debug backend of section
4943         changes, and of location view changes.
4944         (rest_of_handle_final): Adjust.
4945         * toplev.c (process_options): Autodetect value for debug variable
4946         location views option.  Warn on incompat5 without -gdwarf-5.
4947         * doc/invoke.texi (gvariable-location-views): New.
4948         (gvariable-location-views=incompat5): New.
4949         (gno-variable-location-views): New.
4950
4951 2018-02-08  David Malcolm  <dmalcolm@redhat.com>
4952
4953         PR tree-optimization/84136
4954         * tree-cfg.c (find_taken_edge_computed_goto): Remove assertion
4955         that the result of find_edge is non-NULL.
4956
4957 2018-02-08  Sergey Shalnov  <sergey.shalnov@intel.com>
4958
4959         PR target/83008
4960         * config/i386/x86-tune-costs.h (skylake_cost): Fix cost of
4961         storing integer register in SImode.  Fix cost of 256 and 512
4962         byte aligned SSE register store.
4963
4964 2018-02-08  Sergey Shalnov  <sergey.shalnov@intel.com>
4965
4966         * config/i386/i386.c (ix86_multiplication_cost): Fix
4967         multiplication cost for TARGET_AVX512DQ.
4968
4969 2018-02-08  Marek Polacek  <polacek@redhat.com>
4970
4971         PR tree-optimization/84238
4972         * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Verify the result of
4973         get_range_strlen.
4974
4975 2018-02-08  Richard Sandiford  <richard.sandiford@linaro.org>
4976
4977         PR tree-optimization/84265
4978         * tree-vect-stmts.c (vectorizable_store): Don't treat
4979         VMAT_CONTIGUOUS accesses as grouped.
4980         (vectorizable_load): Likewise.
4981
4982 2018-02-08  Richard Sandiford  <richard.sandiford@linaro.org>
4983
4984         PR tree-optimization/81635
4985         * wide-int.h (wi::round_down_for_mask, wi::round_up_for_mask): Declare.
4986         * wide-int.cc (wi::round_down_for_mask, wi::round_up_for_mask)
4987         (test_round_for_mask): New functions.
4988         (wide_int_cc_tests): Call test_round_for_mask.
4989         * tree-vrp.h (intersect_range_with_nonzero_bits): Declare.
4990         * tree-vrp.c (intersect_range_with_nonzero_bits): New function.
4991         * tree-data-ref.c (split_constant_offset_1): Use it to refine the
4992         range returned by get_range_info.
4993
4994 2018-02-08  Jan Hubicka  <hubicka@ucw.cz>
4995
4996         PR ipa/81360
4997         * cgraph.h (symtab_node::output_to_lto_symbol_table_p): Declare
4998         * symtab.c: Include builtins.h
4999         (symtab_node::output_to_lto_symbol_table_p): Move here
5000         from lto-streamer-out.c:output_symbol_p.
5001         * lto-streamer-out.c (write_symbol): Turn early exit to assert.
5002         (output_symbol_p): Move all logic to symtab.c
5003         (produce_symtab): Update.
5004
5005 2018-02-08  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>
5006
5007         * config/s390/s390-opts.h (enum indirect_branch): Define.
5008         * config/s390/s390-protos.h (s390_return_addr_from_memory)
5009         (s390_indirect_branch_via_thunk)
5010         (s390_indirect_branch_via_inline_thunk): Add function prototypes.
5011         (enum s390_indirect_branch_type): Define.
5012         * config/s390/s390.c (struct s390_frame_layout, struct
5013         machine_function): Remove.
5014         (indirect_branch_prez10thunk_mask, indirect_branch_z10thunk_mask)
5015         (indirect_branch_table_label_no, indirect_branch_table_name):
5016         Define variables.
5017         (INDIRECT_BRANCH_NUM_OPTIONS): Define macro.
5018         (enum s390_indirect_branch_option): Define.
5019         (s390_return_addr_from_memory): New function.
5020         (s390_handle_string_attribute): New function.
5021         (s390_attribute_table): Add new attribute handler.
5022         (s390_execute_label): Handle UNSPEC_EXECUTE_JUMP patterns.
5023         (s390_indirect_branch_via_thunk): New function.
5024         (s390_indirect_branch_via_inline_thunk): New function.
5025         (s390_function_ok_for_sibcall): When jumping via thunk disallow
5026         sibling call optimization for non z10 compiles.
5027         (s390_emit_call): Force indirect branch target to be a single
5028         register.  Add r1 clobber for non-z10 compiles.
5029         (s390_emit_epilogue): Emit return jump via return_use expander.
5030         (s390_reorg): Handle JUMP_INSNs as execute targets.
5031         (s390_option_override_internal): Perform validity checks for the
5032         new command line options.
5033         (s390_indirect_branch_attrvalue): New function.
5034         (s390_indirect_branch_settings): New function.
5035         (s390_set_current_function): Invoke s390_indirect_branch_settings.
5036         (s390_output_indirect_thunk_function):  New function.
5037         (s390_code_end): Implement target hook.
5038         (s390_case_values_threshold): Implement target hook.
5039         (TARGET_ASM_CODE_END, TARGET_CASE_VALUES_THRESHOLD): Define target
5040         macros.
5041         * config/s390/s390.h (struct s390_frame_layout)
5042         (struct machine_function): Move here from s390.c.
5043         (TARGET_INDIRECT_BRANCH_NOBP_RET)
5044         (TARGET_INDIRECT_BRANCH_NOBP_JUMP)
5045         (TARGET_INDIRECT_BRANCH_NOBP_JUMP_THUNK)
5046         (TARGET_INDIRECT_BRANCH_NOBP_JUMP_INLINE_THUNK)
5047         (TARGET_INDIRECT_BRANCH_NOBP_CALL)
5048         (TARGET_DEFAULT_INDIRECT_BRANCH_TABLE)
5049         (TARGET_INDIRECT_BRANCH_THUNK_NAME_EXRL)
5050         (TARGET_INDIRECT_BRANCH_THUNK_NAME_EX)
5051         (TARGET_INDIRECT_BRANCH_TABLE): Define macros.
5052         * config/s390/s390.md (UNSPEC_EXECUTE_JUMP)
5053         (INDIRECT_BRANCH_THUNK_REGNUM): Define constants.
5054         (mnemonic attribute): Add values which aren't recognized
5055         automatically.
5056         ("*cjump_long", "*icjump_long", "*basr", "*basr_r"): Disable
5057         pattern for branch conversion.  Fix mnemonic attribute.
5058         ("*c<code>", "*sibcall_br", "*sibcall_value_br", "*return"): Emit
5059         indirect branch via thunk if requested.
5060         ("indirect_jump", "<code>"): Expand patterns for branch conversion.
5061         ("*indirect_jump"): Disable for branch conversion using out of
5062         line thunks.
5063         ("indirect_jump_via_thunk<mode>_z10")
5064         ("indirect_jump_via_thunk<mode>")
5065         ("indirect_jump_via_inlinethunk<mode>_z10")
5066         ("indirect_jump_via_inlinethunk<mode>", "*casesi_jump")
5067         ("casesi_jump_via_thunk<mode>_z10", "casesi_jump_via_thunk<mode>")
5068         ("casesi_jump_via_inlinethunk<mode>_z10")
5069         ("casesi_jump_via_inlinethunk<mode>", "*basr_via_thunk<mode>_z10")
5070         ("*basr_via_thunk<mode>", "*basr_r_via_thunk_z10")
5071         ("*basr_r_via_thunk", "return<mode>_prez10"): New pattern.
5072         ("*indirect2_jump"): Disable for branch conversion.
5073         ("casesi_jump"): Turn into expander and expand patterns for branch
5074         conversion.
5075         ("return_use"): New expander.
5076         ("*return"): Emit return via thunk and rename it to ...
5077         ("*return<mode>"): ... this one.
5078         * config/s390/s390.opt: Add new options and and enum for the
5079         option values.
5080
5081 2018-02-08  Richard Sandiford  <richard.sandiford@linaro.org>
5082
5083         * lra-constraints.c (match_reload): Unconditionally use
5084         gen_lowpart_SUBREG, rather than selecting between that
5085         and equivalent gen_rtx_SUBREG code.
5086
5087 2018-02-08  Richard Biener  <rguenther@suse.de>
5088
5089         PR tree-optimization/84233
5090         * tree-ssa-phiprop.c (propagate_with_phi): Use separate
5091         changed flag instead of boguously re-using phi_inserted.
5092
5093 2018-02-08  Martin Jambor  <mjambor@suse.cz>
5094
5095         * hsa-gen.c (get_symbol_for_decl): Set program allocation for
5096         static local variables.
5097
5098 2018-02-08  Richard Biener  <rguenther@suse.de>
5099
5100         PR tree-optimization/84278
5101         * tree-vect-stmts.c (vectorizable_store): When looking for
5102         smaller vector types to perform grouped strided loads/stores
5103         make sure the mode is supported by the target.
5104         (vectorizable_load): Likewise.
5105
5106 2018-02-08  Wilco Dijkstra  <wdijkstr@arm.com>
5107
5108         * config/aarch64/aarch64.c (aarch64_components_for_bb):
5109         Increase LDP/STP opportunities by adding adjacent callee-saves.
5110
5111 2018-02-08  Wilco Dijkstra  <wdijkstr@arm.com>
5112
5113         PR rtl-optimization/84068
5114         PR rtl-optimization/83459
5115         * haifa-sched.c (rank_for_schedule): Fix SCHED_PRESSURE_MODEL sorting.
5116
5117 2018-02-08  Aldy Hernandez  <aldyh@redhat.com>
5118
5119         PR tree-optimization/84224
5120         * gimple-ssa-warn-alloca.c (pass_walloca::execute): Remove assert.
5121         * calls.c (gimple_alloca_call_p): Only return TRUE when we have
5122         non-zero arguments.
5123
5124 2018-02-07  Iain Sandoe  <iain@codesourcery.com>
5125
5126         PR target/84113
5127         * config/rs6000/altivec.md (*restore_world): Remove LR use.
5128         * config/rs6000/predicates.md (restore_world_operation): Adjust op
5129         count, remove one USE.
5130
5131 2018-02-07  Michael Meissner  <meissner@linux.vnet.ibm.com>
5132
5133         * doc/install.texi (Configuration): Document the
5134         --with-long-double-format={ibm,ieee} PowerPC configuration
5135         options.
5136
5137         PR target/84154
5138         * config/rs6000/rs6000.md (fix_trunc<SFDF:mode><QHI:mode>2):
5139         Convert from define_expand to be define_insn_and_split.  Rework
5140         float/double/_Float128 conversions to QI/HI/SImode to work with
5141         both ISA 2.07 (power8) or ISA 3.0 (power9).  Fix regression where
5142         conversions to QI/HImode types did a store and then a load to
5143         truncate the value.  For conversions to VSX registers, don't split
5144         the insn, instead emit the code directly.  Use the code iterator
5145         any_fix to combine signed and unsigned conversions.
5146         (fix<uns>_trunc<SFDF:mode>si2_p8): Likewise.
5147         (fixuns_trunc<SFDF:mode><QHI:mode>2): Likewise.
5148         (fix_trunc<IEEE128:mode><QHI:mode>2): Likewise.
5149         (fix<uns>_trunc<SFDF:mode><QHI:mode>2): Likewise.
5150         (fix_<mode>di2_hw): Likewise.
5151         (fixuns_<mode>di2_hw): Likewise.
5152         (fix_<mode>si2_hw): Likewise.
5153         (fixuns_<mode>si2_hw): Likewise.
5154         (fix<uns>_<IEEE128:mode><SDI:mode>2_hw): Likewise.
5155         (fix<uns>_trunc<IEEE128:mode><QHI:mode>2): Likewise.
5156         (fctiw<u>z_<mode>_smallint): Rename fctiw<u>z_<mode>_smallint to
5157         fix<uns>_trunc<SFDF:mode>si2_p8.
5158         (fix_trunc<SFDF:mode><QHI:mode>2_internal): Delete, no longer
5159         used.
5160         (fixuns_trunc<SFDF:mode><QHI:mode>2_internal): Likewise.
5161         (fix<uns>_<mode>_mem): Likewise.
5162         (fctiw<u>z_<mode>_mem): Likewise.
5163         (fix<uns>_<mode>_mem): Likewise.
5164         (fix<uns>_trunc<SFDF:mode><QHSI:mode>2_mem): On ISA 3.0, prevent
5165         the register allocator from doing a direct move to the GPRs to do
5166         a store, and instead use the ISA 3.0 store byte/half-word from
5167         vector register instruction.  For IEEE 128-bit floating point,
5168         also optimize stores of 32-bit ints.
5169         (fix<uns>_trunc<IEEE128:mode><QHSI:mode>2_mem): Likewise.
5170
5171 2018-02-07  Alan Hayward  <alan.hayward@arm.com>
5172
5173         * genextract.c (push_pathstr_operand): New function to support
5174         [a-zA-Z].
5175         (walk_rtx): Call push_pathstr_operand.
5176         (print_path): Support [a-zA-Z].
5177
5178 2018-02-07  Richard Biener  <rguenther@suse.de>
5179
5180         PR tree-optimization/84037
5181         * tree-vectorizer.h (struct _loop_vec_info): Add ivexpr_map member.
5182         (cse_and_gimplify_to_preheader): Declare.
5183         (vect_get_place_in_interleaving_chain): Likewise.
5184         * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
5185         ivexpr_map.
5186         (_loop_vec_info::~_loop_vec_info): Delete it.
5187         (cse_and_gimplify_to_preheader): New function.
5188         * tree-vect-slp.c (vect_get_place_in_interleaving_chain): Export.
5189         * tree-vect-stmts.c (vectorizable_store): CSE base and steps.
5190         (vectorizable_load): Likewise.  For grouped stores always base
5191         the IV on the first element.
5192         * tree-vect-loop-manip.c (vect_loop_versioning): Unshare versioning
5193         condition before gimplifying.
5194
5195 2018-02-07  Jakub Jelinek  <jakub@redhat.com>
5196
5197         * tree-eh.c (operation_could_trap_helper_p): Ignore honor_trapv for
5198         *DIV_EXPR and *MOD_EXPR.
5199
5200 2018-02-07  H.J. Lu  <hongjiu.lu@intel.com>
5201
5202         PR target/84248
5203         * config/i386/i386.c (ix86_option_override_internal): Mask out
5204         the CF_SET bit when checking -fcf-protection.
5205
5206 2018-02-07  Tom de Vries  <tom@codesourcery.com>
5207
5208         PR libgomp/84217
5209         * omp-expand.c (expand_oacc_collapse_init): Ensure diff_type is large
5210         enough.
5211
5212 2018-02-07  Richard Biener  <rguenther@suse.de>
5213
5214         PR tree-optimization/84204
5215         * tree-chrec.c (chrec_fold_plus_1): Remove size limiting in
5216         this place.
5217
5218         PR tree-optimization/84205
5219         * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Also
5220         special-case isl_ast_op_zdiv_r.
5221
5222         PR tree-optimization/84223
5223         * graphite-scop-detection.c (gather_bbs::before_dom_children):
5224         Only add conditions from within the region.
5225         (gather_bbs::after_dom_children): Adjust.
5226
5227 2018-02-07  Georg-Johann Lay  <avr@gjlay.de>
5228
5229         PR target/84209
5230         * config/avr/avr.h (GENERAL_REGNO_P, GENERAL_REG_P): New macros.
5231         * config/avr/avr.md: Only post-reload split REG-REG moves if
5232         either register is GENERAL_REG_P.
5233
5234 2018-02-07  Jakub Jelinek  <jakub@redhat.com>
5235
5236         PR tree-optimization/84235
5237         * tree-ssa-scopedtables.c
5238         (avail_exprs_stack::simplify_binary_operation): Fir MINUS_EXPR, punt
5239         if the subtraction is performed in floating point type where NaNs are
5240         honored.  For *DIV_EXPR, punt for ALL_FRACT_MODE_Ps where we can't
5241         build 1.  Formatting fix.
5242
5243 2018-02-06  Jakub Jelinek  <jakub@redhat.com>
5244
5245         PR target/84146
5246         * config/i386/i386.c (rest_of_insert_endbranch): Only skip
5247         NOTE_INSN_CALL_ARG_LOCATION after a call, not anything else,
5248         and skip it regardless of bb boundaries.  Use CALL_P macro,
5249         don't test INSN_P (insn) together with CALL_P or JUMP_P check
5250         unnecessarily, formatting fix.
5251
5252 2018-02-06  Michael Collison  <michael.collison@arm.com>
5253
5254         * config/arm/thumb2.md:
5255         (*thumb2_mov_negscc): Split only if TARGET_THUMB2 && !arm_restrict_it.
5256         (*thumb_mov_notscc): Ditto.
5257
5258 2018-02-06  Michael Meissner  <meissner@linux.vnet.ibm.com>
5259
5260         PR target/84154
5261         * config/rs6000/rs6000.md (su code attribute): Use "u" for
5262         unsigned_fix, not "s".
5263
5264 2018-02-06  Rainer Orth  <ro@CeBiTec.Uni-Bielefeld.DE>
5265
5266         * configure.ac (gcc_fn_eh_frame_ro): New function.
5267         (gcc_cv_as_cfi_directive): Check both 32 and 64-bit assembler for
5268         correct .eh_frame permissions.
5269         * configure: Regenerate.
5270
5271 2018-02-06  Andrew Jenner  <andrew@codeourcery.com>
5272
5273         * doc/invoke.texi: Add section for the PowerPC SPE backend. Remove
5274         irrelevant options.
5275
5276 2018-02-06  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
5277
5278         * config/rs6000/rs6000.c (rs6000_option_override_internal):
5279         Display warning message for -mno-speculate-indirect-jumps.
5280
5281 2018-02-06  Andrew Jenner  <andrew@codesourcery.com>
5282
5283         * config/powerpcspe/powerpcspe.opt: (msimple-fpu, mfpu) Add
5284         Undocumented.
5285         * config/powerpcspe/sysv4.opt (mbit-align): Likewise.
5286
5287 2018-02-06  Aldy Hernandez  <aldyh@redhat.com>
5288
5289         PR tree-optimization/84225
5290         * tree-eh.c (find_trapping_overflow): Only call
5291         operation_no_trapping_overflow when ANY_INTEGRAL_TYPE_P.
5292
5293 2018-02-06  Igor Tsimbalist  <igor.v.tsimbalist@intel.com>
5294
5295         PR target/84145
5296         * config/i386/i386.c: Reimplement the check of possible options
5297         -mibt/-mshstk conbination. Change error messages.
5298         * doc/invoke.texi: Fix a typo: remove extra '='.
5299
5300 2018-02-06  Marek Polacek  <polacek@redhat.com>
5301
5302         PR tree-optimization/84228
5303         * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Skip debug statements.
5304
5305 2018-02-06  Tamar Christina  <tamar.christina@arm.com>
5306
5307         PR target/82641
5308         * config/arm/arm.c (arm_print_asm_arch_directives): Record already
5309         emitted arch directives.
5310         * config/arm/arm-c.c (arm_cpu_builtins): Undefine __ARM_ARCH and
5311         __ARM_FEATURE_COPROC before changing architectures.
5312
5313 2018-02-06  Richard Biener  <rguenther@suse.de>
5314
5315         * config/i386/i386.c (print_reg): Fix typo.
5316         (ix86_loop_unroll_adjust): Do not unroll beyond the original nunroll.
5317
5318 2018-02-06  Eric Botcazou  <ebotcazou@adacore.com>
5319
5320         * configure: Regenerate.
5321
5322 2018-02-05  Martin Sebor  <msebor@redhat.com>
5323
5324         PR tree-optimization/83369
5325         * tree-ssa-ccp.c (pass_post_ipa_warn::execute): Use %G to print
5326         inlining context.
5327
5328 2018-02-05  Martin Liska  <mliska@suse.cz>
5329
5330         * doc/invoke.texi: Cherry-pick upstream r323995.
5331
5332 2018-02-05  Richard Sandiford  <richard.sandiford@linaro.org>
5333
5334         * ira.c (ira_init_register_move_cost): Adjust comment.
5335
5336 2018-02-05  Martin Liska  <mliska@suse.cz>
5337
5338         PR gcov-profile/84137
5339         * doc/gcov.texi: Fix typo in documentation.
5340
5341 2018-02-05  Martin Liska  <mliska@suse.cz>
5342
5343         PR gcov-profile/83879
5344         * doc/gcov.texi: Document necessity of --dynamic-list-data when
5345         using dlopen functionality.
5346
5347 2018-02-05  Olga Makhotina  <olga.makhotina@intel.com>
5348
5349         * config/i386/avx512dqintrin.h (_mm_mask_range_sd, _mm_maskz_range_sd,
5350         _mm_mask_range_round_sd, _mm_maskz_range_round_sd, _mm_mask_range_ss,
5351         _mm_maskz_range_ss, _mm_mask_range_round_ss,
5352         _mm_maskz_range_round_ss): New intrinsics.
5353         (__builtin_ia32_rangesd128_round)
5354         (__builtin_ia32_rangess128_round): Remove.
5355         (__builtin_ia32_rangesd128_mask_round,
5356         __builtin_ia32_rangess128_mask_round): New builtins.
5357         * config/i386/i386-builtin.def (__builtin_ia32_rangesd128_round,
5358         __builtin_ia32_rangess128_round): Remove.
5359         (__builtin_ia32_rangesd128_mask_round,
5360         __builtin_ia32_rangess128_mask_round): New builtins.
5361         * config/i386/sse.md (ranges<mode><round_saeonly_name>): Renamed to ...
5362         (ranges<mode><mask_scalar_name><round_saeonly_scalar_name>): ... this.
5363         ((match_operand:VF_128 2 "<round_saeonly_nimm_predicate>"
5364         "<round_saeonly_constraint>")): Changed to ...
5365         ((match_operand:VF_128 2 "<round_saeonly_scalar_nimm_predicate>"
5366         "<round_saeonly_scalar_constraint>")): ... this.
5367         ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_op4>%2, %1, %0|
5368         %0, %1, %2<round_saeonly_op4>, %3}"): Changed to ...
5369         ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_scalar_mask_op4>%2,
5370         %1, %0<mask_scalar_operand4>|%0<mask_scalar_operand4>, %1,
5371         %2<round_saeonly_scalar_mask_op4>, %3}"): ... this.
5372
5373 2018-02-02  Andrew Jenner  <andrew@codesourcery.com>
5374
5375         * config/powerpcspe/powerpcspe.opt: Add Undocumented to irrelevant
5376         options.
5377         * config/powerpcspe/powerpcspe-tables.opt (rs6000_cpu_opt_value):
5378         Remove all values except native, 8540 and 8548.
5379
5380 2018-02-02  H.J. Lu  <hongjiu.lu@intel.com>
5381
5382         * config/i386/i386.c (ix86_output_function_return): Pass
5383         INVALID_REGNUM, instead of -1, as invalid register number to
5384         indirect_thunk_name and output_indirect_thunk.
5385
5386 2018-02-02  Julia Koval  <julia.koval@intel.com>
5387
5388         * config.gcc: Add -march=icelake.
5389         * config/i386/driver-i386.c (host_detect_local_cpu): Detect icelake.
5390         * config/i386/i386-c.c (ix86_target_macros_internal): Handle icelake.
5391         * config/i386/i386.c (processor_costs): Add m_ICELAKE.
5392         (PTA_ICELAKE, PTA_AVX512VNNI, PTA_GFNI, PTA_VAES, PTA_AVX512VBMI2,
5393         PTA_VPCLMULQDQ, PTA_RDPID, PTA_AVX512BITALG): New.
5394         (processor_target_table): Add icelake.
5395         (ix86_option_override_internal): Handle new PTAs.
5396         (get_builtin_code_for_version): Handle icelake.
5397         (M_INTEL_COREI7_ICELAKE): New.
5398         (fold_builtin_cpu): Handle icelake.
5399         * config/i386/i386.h (TARGET_ICELAKE, PROCESSOR_ICELAKE): New.
5400         * doc/invoke.texi: Add -march=icelake.
5401
5402 2018-02-02  Julia Koval  <julia.koval@intel.com>
5403
5404         * config/i386/i386.c (ix86_option_override_internal): Change flags type
5405         to wide_int_bitmask.
5406         * wide-int-bitmask.h: New.
5407
5408 2018-02-02  Igor Tsimbalist  <igor.v.tsimbalist@intel.com>
5409
5410         PR target/84066
5411         * config/i386/i386.md: Replace Pmode with word_mode in
5412         builtin_setjmp_setup and builtin_longjmp to support x32.
5413
5414 2018-02-01  Peter Bergner  <bergner@vnet.ibm.com>
5415
5416         PR target/56010
5417         PR target/83743
5418         * config/rs6000/driver-rs6000.c: #include "diagnostic.h".
5419         #include "opts.h".
5420         (rs6000_supported_cpu_names): New static variable.
5421         (linux_cpu_translation_table): Likewise.
5422         (elf_platform) <cpu>: Define new static variable and use it.
5423         Translate kernel AT_PLATFORM name to canonical name if needed.
5424         Error if platform name is unknown.
5425
5426 2018-02-01  Aldy Hernandez  <aldyh@redhat.com>
5427
5428         PR target/84089
5429         * config/pa/predicates.md (base14_operand): Handle E_VOIDmode.
5430
5431 2018-02-01  Jeff Law  <law@redhat.com>
5432
5433         PR target/84128
5434         * config/i386/i386.c (release_scratch_register_on_entry): Add new
5435         OFFSET and RELEASE_VIA_POP arguments.  Use SP+OFFSET to restore
5436         the scratch if RELEASE_VIA_POP is false.
5437         (ix86_adjust_stack_and_probe_stack_clash): Un-constify SIZE.
5438         If we have to save a temporary register, decrement SIZE appropriately.
5439         Pass new arguments to release_scratch_register_on_entry.
5440         (ix86_adjust_stack_and_probe): Likewise.
5441         (ix86_emit_probe_stack_range): Pass new arguments to
5442         release_scratch_register_on_entry.
5443
5444 2018-02-01  Uros Bizjak  <ubizjak@gmail.com>
5445
5446         PR rtl-optimization/84157
5447         * combine.c (change_zero_ext): Use REG_P predicate in
5448         front of HARD_REGISTER_P predicate.
5449
5450 2018-02-01  Georg-Johann Lay  <avr@gjlay.de>
5451
5452         * config/avr/avr.c (avr_option_override): Move disabling of
5453         -fdelete-null-pointer-checks to...
5454         * common/config/avr/avr-common.c (avr_option_optimization_table):
5455         ...here.
5456
5457 2018-02-01  Richard Sandiford  <richard.sandiford@linaro.org>
5458
5459         PR tree-optimization/81635
5460         * tree-data-ref.c (split_constant_offset_1): For types that
5461         wrap on overflow, try to use range info to prove that wrapping
5462         cannot occur.
5463
5464 2018-02-01  Renlin Li  <renlin.li@arm.com>
5465
5466         PR target/83370
5467         * config/aarch64/aarch64.c (aarch64_class_max_nregs): Handle
5468         TAILCALL_ADDR_REGS.
5469         (aarch64_register_move_cost): Likewise.
5470         * config/aarch64/aarch64.h (reg_class): Rename CALLER_SAVE_REGS to
5471         TAILCALL_ADDR_REGS.
5472         (REG_CLASS_NAMES): Likewise.
5473         (REG_CLASS_CONTENTS): Rename CALLER_SAVE_REGS to
5474         TAILCALL_ADDR_REGS. Remove IP registers.
5475         * config/aarch64/aarch64.md (Ucs): Update register constraint.
5476
5477 2018-02-01  Richard Biener  <rguenther@suse.de>
5478
5479         * domwalk.h (dom_walker::dom_walker): Add additional constructor
5480         for specifying RPO order and allow NULL for that.
5481         * domwalk.c (dom_walker::dom_walker): Likewise.
5482         (dom_walker::walk): Handle NULL RPO order.
5483         * tree-into-ssa.c (rewrite_dom_walker): Do not walk dom children
5484         in RPO order.
5485         (rewrite_update_dom_walker): Likewise.
5486         (mark_def_dom_walker): Likewise.
5487
5488 2018-02-01  Richard Sandiford  <richard.sandiford@linaro.org>
5489
5490         * config/aarch64/aarch64-protos.h (aarch64_split_sve_subreg_move)
5491         (aarch64_maybe_expand_sve_subreg_move): Declare.
5492         * config/aarch64/aarch64.md (UNSPEC_REV_SUBREG): New unspec.
5493         * config/aarch64/predicates.md (aarch64_any_register_operand): New
5494         predicate.
5495         * config/aarch64/aarch64-sve.md (mov<mode>): Optimize subreg moves
5496         that are semantically a reverse operation.
5497         (*aarch64_sve_mov<mode>_subreg_be): New pattern.
5498         * config/aarch64/aarch64.c (aarch64_maybe_expand_sve_subreg_move):
5499         (aarch64_replace_reg_mode, aarch64_split_sve_subreg_move): New
5500         functions.
5501         (aarch64_can_change_mode_class): For big-endian, forbid changes
5502         between two SVE modes if they have different element sizes.
5503
5504 2018-02-01  Richard Sandiford  <richard.sandiford@linaro.org>
5505
5506         * config/aarch64/aarch64.c (aarch64_expand_sve_const_vector): Prefer
5507         the TImode handling for big-endian targets.
5508
5509 2018-02-01  Richard Sandiford  <richard.sandiford@linaro.org>
5510
5511         * config/aarch64/aarch64-sve.md (sve_ld1rq): Replace with...
5512         (*sve_ld1rq<Vesize>): ... this new pattern.  Handle all element sizes,
5513         not just bytes.
5514         * config/aarch64/aarch64.c (aarch64_expand_sve_widened_duplicate):
5515         Remove BSWAP handing for big-endian targets and use the form of
5516         LD1RQ appropariate for the mode.
5517
5518 2018-02-01  Richard Sandiford  <richard.sandiford@linaro.org>
5519
5520         * config/aarch64/aarch64.c (aarch64_simd_valid_immediate): Handle
5521         all CONST_VECTOR_DUPLICATE_P vectors, not just those with a single
5522         duplicated element.
5523
5524 2018-02-01  Richard Sandiford  <richard.sandiford@linaro.org>
5525
5526         PR tearget/83845
5527         * config/aarch64/aarch64.c (aarch64_secondary_reload): Tighten
5528         check for operands that need to go through aarch64_sve_reload_be.
5529
5530 2018-02-01  Jakub Jelinek  <jakub@redhat.com>
5531
5532         PR tree-optimization/81661
5533         PR tree-optimization/84117
5534         * tree-eh.h (rewrite_to_non_trapping_overflow): Declare.
5535         * tree-eh.c: Include gimplify.h.
5536         (find_trapping_overflow, replace_trapping_overflow,
5537         rewrite_to_non_trapping_overflow): New functions.
5538         * tree-vect-loop.c: Include tree-eh.h.
5539         (vect_get_loop_niters): Use rewrite_to_non_trapping_overflow.
5540         * tree-data-ref.c: Include tree-eh.h.
5541         (get_segment_min_max): Use rewrite_to_non_trapping_overflow.
5542
5543 2018-01-31  Uros Bizjak  <ubizjak@gmail.com>
5544
5545         PR rtl-optimization/84123
5546         * combine.c (change_zero_ext): Check if hard register satisfies
5547         can_change_dest_mode before calling gen_lowpart_SUBREG.
5548
5549 2018-01-31  Vladimir Makarov  <vmakarov@redhat.com>
5550
5551         PR target/82444
5552         * ira.c (ira_init_register_move_cost): Remove assert.
5553
5554 2018-01-31  Eric Botcazou  <ebotcazou@adacore.com>
5555
5556         PR rtl-optimization/84071
5557         * doc/tm.texi.in (WORD_REGISTER_OPERATIONS): Add explicit case.
5558         * doc/tm.texi: Regenerate.
5559
5560 2018-01-31  Richard Biener  <rguenther@suse.de>
5561
5562         PR tree-optimization/84132
5563         * tree-data-ref.c (analyze_miv_subscript): Properly
5564         check whether evolution_function_is_affine_multivariate_p
5565         before calling gcd_of_steps_may_divide_p.
5566
5567 2018-01-31  Julia Koval  <julia.koval@intel.com>
5568
5569         PR target/83618
5570         * config/i386/i386.c (ix86_expand_builtin): Handle IX86_BUILTIN_RDPID.
5571         * config/i386/i386.md (rdpid_rex64) New.
5572         (rdpid): Make 32bit only.
5573
5574 2018-01-29  Aldy Hernandez  <aldyh@redhat.com>
5575
5576         PR lto/84105
5577         * tree-pretty-print.c (dump_generic_node): Handle a TYPE_NAME with
5578         an IDENTIFIER_NODE for FUNCTION_TYPE's.
5579
5580 2018-01-31  Eric Botcazou  <ebotcazou@adacore.com>
5581
5582         Revert
5583         2018-01-12  Eric Botcazou  <ebotcazou@adacore.com>
5584
5585         * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
5586
5587 2018-01-31  Eric Botcazou  <ebotcazou@adacore.com>
5588
5589         PR rtl-optimization/84071
5590         * combine.c (record_dead_and_set_regs_1): Record the source unmodified
5591         for a paradoxical SUBREG on a WORD_REGISTER_OPERATIONS target.
5592
5593 2018-01-31  Claudiu Zissulescu  <claziss@synopsys.com>
5594
5595         * config/arc/arc.c (arc_handle_aux_attribute): New function.
5596         (arc_attribute_table): Add 'aux' attribute.
5597         (arc_in_small_data_p): Consider aux like variables.
5598         (arc_is_aux_reg_p): New function.
5599         (arc_asm_output_aligned_decl_local): Ignore 'aux' like variables.
5600         (arc_get_aux_arg): New function.
5601         (prepare_move_operands): Handle aux-register access.
5602         (arc_handle_aux_attribute): New function.
5603         * doc/extend.texi (ARC Variable attributes): Add subsection.
5604
5605 2018-01-31  Claudiu Zissulescu  <claziss@synopsys.com>
5606
5607         * config/arc/arc-protos.h (arc_is_uncached_mem_p): Function proto.
5608         * config/arc/arc.c (arc_handle_uncached_attribute): New function.
5609         (arc_attribute_table): Add 'uncached' attribute.
5610         (arc_print_operand): Print '.di' flag for uncached memory
5611         accesses.
5612         (arc_in_small_data_p): Do not consider for small data the uncached
5613         types.
5614         (arc_is_uncached_mem_p): New function.
5615         * config/arc/predicates.md (compact_store_memory_operand): Check
5616         for uncached memory accesses.
5617         (nonvol_nonimm_operand): Likewise.
5618         * gcc/doc/extend.texi (ARC Type Attribute): New subsection.
5619
5620 2018-01-31  Jakub Jelinek  <jakub@redhat.com>
5621
5622         PR c/84100
5623         * common.opt (falign-functions=, falign-jumps=, falign-labels=,
5624         falign-loops=): Add Optimization flag.
5625
5626 2018-01-30  Jeff Law  <law@redhat.com>
5627
5628         PR target/84064
5629         * i386.c (ix86_adjust_stack_and_probe_stack_clash): New argument
5630         INT_REGISTERS_SAVED.  Check it prior to calling
5631         get_scratch_register_on_entry.
5632         (ix86_adjust_stack_and_probe): Similarly.
5633         (ix86_emit_probe_stack_range): Similarly.
5634         (ix86_expand_prologue): Corresponding changes.
5635
5636 2018-01-30  Rainer Orth  <ro@CeBiTec.Uni-Bielefeld.DE>
5637
5638         PR target/40411
5639         * config/sol2.h (STARTFILE_ARCH_SPEC): Use -std=c*,
5640         -std=iso9899:199409 instead of -pedantic to select values-Xc.o.
5641
5642 2018-01-30  Vladimir Makarov  <vmakarov@redhat.com>
5643
5644         PR target/84112
5645         * lra-constraints.c (curr_insn_transform): Process AND in the
5646         address.
5647
5648 2018-01-30  Jakub Jelinek  <jakub@redhat.com>
5649
5650         PR rtl-optimization/83986
5651         * sched-deps.c (sched_analyze_insn): For frame related insns, add anti
5652         dependence against last_pending_memory_flush in addition to
5653         pending_jump_insns.
5654
5655 2018-01-30  Alexandre Oliva  <aoliva@redhat.com>
5656
5657         PR tree-optimization/81611
5658         * tree-ssa-dom.c (simple_iv_increment_p): Skip intervening
5659         copies.
5660
5661 2018-01-30  Aaron Sawdey  <acsawdey@linux.vnet.ibm.com>
5662
5663         PR target/83758
5664         * config/rs6000/rs6000.c (rs6000_internal_arg_pointer): Only return
5665         a reg rtx.
5666
5667 2018-01-30  Richard Biener  <rguenther@suse.de>
5668             Jakub Jelinek  <jakub@redhat.com>
5669
5670         PR tree-optimization/84111
5671         * tree-ssa-loop-ivcanon.c (tree_unroll_loops_completely_1): Skip
5672         inner loops added during recursion, as they don't have up-to-date
5673         SSA form.
5674
5675 2018-01-30  Jan Hubicka  <hubicka@ucw.cz>
5676
5677         PR ipa/81360
5678         * ipa-inline.c (can_inline_edge_p): Break out late tests to...
5679         (can_inline_edge_by_limits_p): ... here.
5680         (can_early_inline_edge_p, check_callers,
5681         update_caller_keys, update_callee_keys, recursive_inlining,
5682         add_new_edges_to_heap, speculation_useful_p,
5683         inline_small_functions,
5684         inline_small_functions, flatten_function,
5685         inline_to_all_callers_1): Update.
5686
5687 2018-01-30  Jan Hubicka  <hubicka@ucw.cz>
5688
5689         * profile-count.c (profile_count::combine_with_ipa_count): Handle
5690         zeros correctly.
5691
5692 2018-01-30  Richard Biener  <rguenther@suse.de>
5693
5694         PR tree-optimization/83008
5695         * tree-vect-slp.c (vect_analyze_slp_cost_1): Properly cost
5696         invariant and constant vector uses in stmts when they need
5697         more than one stmt.
5698
5699 2018-01-30  Rainer Orth  <ro@CeBiTec.Uni-Bielefeld.DE>
5700
5701         PR bootstrap/84017
5702         * configure.ac (gcc_cv_as_shf_merge): Disable on Solaris 10/x86.
5703         * configure: Regenerate.
5704
5705 2018-01-30  Richard Sandiford  <richard.sandiford@linaro.org>
5706
5707         * config/aarch64/aarch64-sve.md (*vec_extract<mode><Vel>_0): New
5708         pattern.
5709         (*vec_extract<mode><Vel>_v128): Require a nonzero lane number.
5710         Use gen_rtx_REG rather than gen_lowpart.
5711
5712 2018-01-30  Richard Sandiford  <richard.sandiford@linaro.org>
5713
5714         * lra-constraints.c (match_reload): Use subreg_lowpart_offset
5715         rather than 0 when creating partial subregs.
5716
5717 2018-01-30  Richard Sandiford  <richard.sandiford@linaro.org>
5718
5719         * vec-perm-indices.c (vec_perm_indices::series_p): Give examples
5720         of usage.
5721
5722 2018-01-29  Michael Meissner  <meissner@linux.vnet.ibm.com>
5723
5724         PR target/81550
5725         * config/rs6000/rs6000.c (rs6000_setup_reg_addr_masks): If DFmode
5726         and SFmode can go in Altivec registers (-mcpu=power7 for DFmode,
5727         -mcpu=power8 for SFmode) don't set the PRE_INCDEC or PRE_MODIFY
5728         flags.  This restores the settings used before the 2017-07-24.
5729         Turning off pre increment/decrement/modify allows IVOPTS to
5730         optimize DF/SF loops where the index is an int.
5731
5732 2018-01-29  Richard Biener  <rguenther@suse.de>
5733             Kelvin Nilsen  <kelvin@gcc.gnu.org>
5734
5735         PR bootstrap/80867
5736         * tree-vect-stmts.c (vectorizable_call): Don't call
5737         targetm.vectorize_builtin_md_vectorized_function if callee is
5738         NULL.
5739
5740 2018-01-22  Carl Love  <cel@us.ibm.com>
5741
5742         * doc/extend.tex: Fix typo in second arg in
5743         __builtin_bcdadd_{lt|eq|gt|ov} and __builtin_bcdsub_{lt|eq|gt|ov}.
5744
5745 2018-01-29  Richard Biener  <rguenther@suse.de>
5746
5747         PR tree-optimization/84086
5748         * tree-ssanames.c: Include cfgloop.h and tree-scalar-evolution.h.
5749         (flush_ssaname_freelist): When SSA names were released reset
5750         the SCEV hash table.
5751
5752 2018-01-29  Richard Biener  <rguenther@suse.de>
5753
5754         PR tree-optimization/84057
5755         * tree-ssa-loop-ivcanon.c (unloop_loops): Deal with already
5756         removed paths when removing edges.
5757
5758 2018-01-27  H.J. Lu  <hongjiu.lu@intel.com>
5759
5760         * doc/invoke.texi: Replace -mfunction-return==@var{choice} with
5761         -mfunction-return=@var{choice}.
5762
5763 2018-01-27  Bernd Edlinger  <bernd.edlinger@hotmail.de>
5764
5765         PR diagnostic/84034
5766         * diagnostic-show-locus.c (get_line_width_without_trailing_whitespace):
5767         Handle CR like TAB.
5768         (layout::print_source_line): Likewise.
5769         (test_get_line_width_without_trailing_whitespace): Add test cases.
5770
5771 2018-01-27  Jakub Jelinek  <jakub@redhat.com>
5772
5773         PR middle-end/84040
5774         * sched-deps.c (sched_macro_fuse_insns): Return immediately for
5775         debug insns.
5776
5777 2018-01-26  Jim Wilson  <jimw@sifive.com>
5778
5779         * config/riscv/riscv.h (MAX_FIXED_MODE_SIZE): New.
5780
5781         * config/riscv/elf.h (LIB_SPEC): Don't include -lgloss when nosys.specs
5782         specified.
5783
5784 2018-01-26  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
5785
5786         * config/aarch64/aarch64.md: Add peepholes for CMP + SUB -> SUBS
5787         and CMP + SUB-immediate -> SUBS.
5788
5789 2018-01-26  Martin Sebor  <msebor@redhat.com>
5790
5791         PR tree-optimization/83896
5792         * tree-ssa-strlen.c (get_string_len): Rename...
5793         (get_string_cst_length): ...to this.  Return HOST_WIDE_INT.
5794         Avoid assuming length is constant.
5795         (handle_char_store): Use HOST_WIDE_INT for string length.
5796
5797 2018-01-26  Uros Bizjak  <ubizjak@gmail.com>
5798
5799         PR target/81763
5800         * config/i386/i386.md (*andndi3_doubleword): Add earlyclobber
5801         to (=&r,r,rm) alternative. Add (=r,0,rm) and (=r,r,0) alternatives.
5802
5803 2018-01-26  Richard Biener  <rguenther@suse.de>
5804
5805         PR rtl-optimization/84003
5806         * dse.c (record_store): Only record redundant stores when
5807         the earlier store aliases at least all accesses the later one does.
5808
5809 2018-01-26  Jakub Jelinek  <jakub@redhat.com>
5810
5811         PR rtl-optimization/83985
5812         * dce.c (deletable_insn_p): Return false for separate shrink wrapping
5813         REG_CFA_RESTORE insns.
5814         (delete_unmarked_insns): Don't ignore separate shrink wrapping
5815         REG_CFA_RESTORE insns here.
5816
5817         PR c/83989
5818         * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Don't
5819         use SSA_NAME_VAR as base for SSA_NAMEs with non-NULL SSA_NAME_VAR.
5820
5821 2018-01-26  Claudiu Zissulescu  <claziss@synopsys.com>
5822
5823         * config/arc/arc-arch.h (arc_tune_attr): Add ARC_TUNE_CORE_3.
5824         * config/arc/arc.c (arc_sched_issue_rate): Use ARC_TUNE_... .
5825         (arc_init): Likewise.
5826         (arc_override_options): Likewise.
5827         (arc_file_start): Choose Tag_ARC_CPU_variation based on arc_tune
5828         value.
5829         (hwloop_fail): Use TARGET_DBNZ when we want to check for dbnz insn
5830         support.
5831         * config/arc/arc.h (TARGET_DBNZ): Define.
5832         * config/arc/arc.md (attr tune): Add core_3, use ARC_TUNE_... to
5833         properly set the tune attribute.
5834         (dbnz): Use TARGET_DBNZ guard.
5835         * config/arc/arc.opt (mtune): Add core3 option.
5836
5837 2018-01-26  Claudiu Zissulescu  <claziss@synopsys.com>
5838
5839         * config/arc/arc.c (arc_delegitimize_address_0): Refactored to
5840         recognize new pic like addresses.
5841         (arc_delegitimize_address): Clean up.
5842
5843 2018-01-26  Claudiu Zissulescu  <claziss@synopsys.com>
5844
5845         * config/arc/arc-arches.def: Option mrf16 valid for all
5846         architectures.
5847         * config/arc/arc-c.def (__ARC_RF16__): New predefined macro.
5848         * config/arc/arc-cpus.def (em_mini): New cpu with rf16 on.
5849         * config/arc/arc-options.def (FL_RF16): Add mrf16 option.
5850         * config/arc/arc-tables.opt: Regenerate.
5851         * config/arc/arc.c (arc_conditional_register_usage): Handle
5852         reduced register file case.
5853         (arc_file_start): Set must have build attributes.
5854         * config/arc/arc.h (MAX_ARC_PARM_REGS): Conditional define using
5855         mrf16 option value.
5856         * config/arc/arc.opt (mrf16): Add new option.
5857         * config/arc/elf.h (ATTRIBUTE_PCS): Define.
5858         * config/arc/genmultilib.awk: Handle new mrf16 option.
5859         * config/arc/linux.h (ATTRIBUTE_PCS): Define.
5860         * config/arc/t-multilib: Regenerate.
5861         * doc/invoke.texi (ARC Options): Document mrf16 option.
5862
5863 2018-01-26  Claudiu Zissulescu  <claziss@synopsys.com>
5864
5865         * config/arc/arc-protos.h: Add arc_is_secure_call_p proto.
5866         * config/arc/arc.c (arc_handle_secure_attribute): New function.
5867         (arc_attribute_table): Add 'secure_call' attribute.
5868         (arc_print_operand): Print secure call operand.
5869         (arc_function_ok_for_sibcall): Don't optimize tail calls when
5870         secure.
5871         (arc_is_secure_call_p): New function.  * config/arc/arc.md
5872         (call_i): Add support for sjli instruction.
5873         (call_value_i): Likewise.
5874         * config/arc/constraints.md (Csc): New constraint.
5875
5876 2018-01-26  Claudiu Zissulescu  <claziss@synopsys.com>
5877             John Eric Martin  <John.Martin@emmicro-us.com>
5878
5879         * config/arc/arc-protos.h: Add arc_is_jli_call_p proto.
5880         * config/arc/arc.c (_arc_jli_section): New struct.
5881         (arc_jli_section): New type.
5882         (rc_jli_sections): New static variable.
5883         (arc_handle_jli_attribute): New function.
5884         (arc_attribute_table): Add jli_always and jli_fixed attribute.
5885         (arc_file_end): New function.
5886         (TARGET_ASM_FILE_END): Define.
5887         (arc_print_operand): Reuse 'S' letter for JLI output instruction.
5888         (arc_add_jli_section): New function.
5889         (jli_call_scan): Likewise.
5890         (arc_reorg): Call jli_call_scan.
5891         (arc_output_addsi): Remove 'S' from printing asm operand.
5892         (arc_is_jli_call_p): New function.
5893         * config/arc/arc.md (movqi_insn): Remove 'S' from printing asm
5894         operand.
5895         (movhi_insn): Likewise.
5896         (movsi_insn): Likewise.
5897         (movsi_set_cc_insn): Likewise.
5898         (loadqi_update): Likewise.
5899         (load_zeroextendqisi_update): Likewise.
5900         (load_signextendqisi_update): Likewise.
5901         (loadhi_update): Likewise.
5902         (load_zeroextendhisi_update): Likewise.
5903         (load_signextendhisi_update): Likewise.
5904         (loadsi_update): Likewise.
5905         (loadsf_update): Likewise.
5906         (movsicc_insn): Likewise.
5907         (bset_insn): Likewise.
5908         (bxor_insn): Likewise.
5909         (bclr_insn): Likewise.
5910         (bmsk_insn): Likewise.
5911         (bicsi3_insn): Likewise.
5912         (cmpsi_cc_c_insn): Likewise.
5913         (movsi_ne): Likewise.
5914         (movsi_cond_exec): Likewise.
5915         (clrsbsi2): Likewise.
5916         (norm_f): Likewise.
5917         (normw): Likewise.
5918         (swap): Likewise.
5919         (divaw): Likewise.
5920         (flag): Likewise.
5921         (sr): Likewise.
5922         (kflag): Likewise.
5923         (ffs): Likewise.
5924         (ffs_f): Likewise.
5925         (fls): Likewise.
5926         (call_i): Remove 'S' asm letter, add jli instruction.
5927         (call_value_i): Likewise.
5928         * config/arc/arc.op (mjli-always): New option.
5929         * config/arc/constraints.md (Cji): New constraint.
5930         * config/arc/fpx.md (addsf3_fpx): Remove 'S' from printing asm
5931         operand.
5932         (subsf3_fpx): Likewise.
5933         (mulsf3_fpx): Likewise.
5934         * config/arc/simdext.md (vendrec_insn): Remove 'S' from printing
5935         asm operand.
5936         * doc/extend.texi (ARC): Document 'jli-always' and 'jli-fixed'
5937         function attrbutes.
5938         * doc/invoke.texi (ARC): Document mjli-always option.
5939
5940 2018-01-26  Sebastian Perta  <sebastian.perta@renesas.com>
5941
5942         * config/rl78/rl78.c (rl78_addsi3_internal): If operand 2 is const
5943         avoid addition with 0 and use incw and decw where possible.
5944
5945 2018-01-26  Richard Biener  <rguenther@suse.de>
5946
5947         PR tree-optimization/81082
5948         * fold-const.c (fold_plusminus_mult_expr): Do not perform the
5949         association if it requires casting to unsigned.
5950         * match.pd ((A * C) +- (B * C) -> (A+-B)): New patterns derived
5951         from fold_plusminus_mult_expr to catch important cases late when
5952         range info is available.
5953
5954 2018-01-26  Rainer Orth  <ro@CeBiTec.Uni-Bielefeld.DE>
5955
5956         * config/i386/sol2.h (USE_HIDDEN_LINKONCE): Remove.
5957         * configure.ac (hidden_linkonce): New test.
5958         * configure: Regenerate.
5959         * config.in: Regenerate.
5960
5961 2018-01-26  Julia Koval  <julia.koval@intel.com>
5962
5963         * config/i386/avx512bitalgintrin.h (_mm512_bitshuffle_epi64_mask,
5964         _mm512_mask_bitshuffle_epi64_mask, _mm256_bitshuffle_epi64_mask,
5965         _mm256_mask_bitshuffle_epi64_mask, _mm_bitshuffle_epi64_mask,
5966         _mm_mask_bitshuffle_epi64_mask): Fix type.
5967         * config/i386/i386-builtin-types.def (UHI_FTYPE_V2DI_V2DI_UHI,
5968         USI_FTYPE_V4DI_V4DI_USI): Remove.
5969         * config/i386/i386-builtin.def (__builtin_ia32_vpshufbitqmb512_mask,
5970         __builtin_ia32_vpshufbitqmb256_mask,
5971         __builtin_ia32_vpshufbitqmb128_mask): Fix types.
5972         * config/i386/i386.c (ix86_expand_args_builtin): Remove old types.
5973         * config/i386/sse.md (VI1_AVX512VLBW): Change types.
5974
5975 2018-01-26  Alan Modra  <amodra@gmail.com>
5976
5977         PR target/84033
5978         * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Exclude
5979         UNSPEC_VBPERMQ.  Sort other unspecs.
5980
5981 2018-01-25  David Edelsohn  <dje.gcc@gmail.com>
5982
5983         * doc/invoke.texi (PowerPC Options): Document 'native' cpu type.
5984
5985 2018-01-25  Jan Hubicka  <hubicka@ucw.cz>
5986
5987         PR middle-end/83055
5988         * predict.c (drop_profile): Do not push/pop cfun; update also
5989         node->count.
5990         (handle_missing_profiles): Fix logic looking for zero profiles.
5991
5992 2018-01-25  Jakub Jelinek  <jakub@redhat.com>
5993
5994         PR middle-end/83977
5995         * ipa-fnsummary.c (compute_fn_summary): Clear can_change_signature
5996         on functions with #pragma omp declare simd or functions with simd
5997         attribute.
5998         * omp-simd-clone.c (expand_simd_clones): Revert 2018-01-24 change.
5999         * config/i386/i386.c (ix86_simd_clone_compute_vecsize_and_simdlen):
6000         Remove trailing \n from warning_at calls.
6001
6002 2018-01-25  Tom de Vries  <tom@codesourcery.com>
6003
6004         PR target/84028
6005         * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
6006         for neutered workers.
6007
6008 2018-01-24  Joseph Myers  <joseph@codesourcery.com>
6009
6010         PR target/68467
6011         * config/m68k/m68k.c (m68k_promote_function_mode): New function.
6012         (TARGET_PROMOTE_FUNCTION_MODE): New macro.
6013
6014 2018-01-24  Jeff Law  <law@redhat.com>
6015
6016         PR target/83994
6017         * i386.c (get_probe_interval): Move to earlier point.
6018         (ix86_compute_frame_layout): If -fstack-clash-protection and
6019         the frame is larger than the probe interval, then use pushes
6020         to save registers rather than reg->mem moves.
6021         (ix86_expand_prologue): Remove conditional for int_registers_saved
6022         assertion.
6023
6024 2018-01-24  Vladimir Makarov  <vmakarov@redhat.com>
6025
6026         PR target/84014
6027         * ira-build.c (setup_min_max_allocno_live_range_point): Set up
6028         min/max for never referenced object.
6029
6030 2018-01-24  Jakub Jelinek  <jakub@redhat.com>
6031
6032         PR middle-end/83977
6033         * tree.c (free_lang_data_in_decl): Don't clear DECL_ABSTRACT_ORIGIN
6034         here.
6035         * omp-low.c (create_omp_child_function): Remove "omp declare simd"
6036         attributes from DECL_ATTRIBUTES (decl) without affecting
6037         DECL_ATTRIBUTES (current_function_decl).
6038         * omp-simd-clone.c (expand_simd_clones): Ignore DECL_ARTIFICIAL
6039         functions with non-NULL DECL_ABSTRACT_ORIGIN.
6040
6041 2018-01-24  Richard Sandiford  <richard.sandiford@linaro.org>
6042
6043         PR tree-optimization/83979
6044         * fold-const.c (fold_comparison): Use constant_boolean_node
6045         instead of boolean_{true,false}_node.
6046
6047 2018-01-24  Jan Hubicka  <hubicka@ucw.cz>
6048
6049         * ipa-profile.c (ipa_propagate_frequency_1): Fix logic skipping calls
6050         with zero counts.
6051
6052 2018-01-24  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
6053
6054         * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
6055         Simplify the clause that sets the length attribute.
6056         (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
6057         (*sibcall_nonlocal_sysv<mode>): Clean up code block; simplify the
6058         clause that sets the length attribute.
6059         (*sibcall_value_nonlocal_sysv<mode>): Likewise.
6060
6061 2018-01-24  Tom de Vries  <tom@codesourcery.com>
6062
6063         PR target/83589
6064         * config/nvptx/nvptx.c (WORKAROUND_PTXJIT_BUG_2): Define to 1.
6065         (nvptx_pc_set, nvptx_condjump_label): New function. Copy from jump.c.
6066         Add strict parameter.
6067         (prevent_branch_around_nothing): Insert dummy insn between branch to
6068         label and label with no ptx insn inbetween.
6069         * config/nvptx/nvptx.md (define_insn "fake_nop"): New insn.
6070
6071 2018-01-24  Tom de Vries  <tom@codesourcery.com>
6072
6073         PR target/81352
6074         * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
6075         for neutered threads in warp.
6076         * config/nvptx/nvptx.md (define_insn "exit"): New insn.
6077
6078 2018-01-24  Richard Biener  <rguenther@suse.de>
6079
6080         PR tree-optimization/83176
6081         * tree-chrec.c (chrec_fold_plus_1): Handle (signed T){(T) .. }
6082         operands.
6083
6084 2018-01-24  Richard Biener  <rguenther@suse.de>
6085
6086         PR tree-optimization/82819
6087         * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Avoid
6088         code generating pluses that are no-ops in the target precision.
6089
6090 2018-01-24  Richard Biener  <rguenther@suse.de>
6091
6092         PR middle-end/84000
6093         * tree-cfg.c (replace_loop_annotate): Handle annot_expr_parallel_kind.
6094
6095 2018-01-23  Jan Hubicka  <hubicka@ucw.cz>
6096
6097         * cfgcleanup.c (try_crossjump_to_edge): Use combine_with_count
6098         to merge probabilities.
6099         * predict.c (probably_never_executed): Also mark as cold functions
6100         with global 0 profile and guessed local profile.
6101         * profile-count.c (profile_probability::combine_with_count): New
6102         member function.
6103         * profile-count.h (profile_probability::operator*,
6104         profile_probability::operator*=, profile_probability::operator/,
6105         profile_probability::operator/=): Reduce precision to adjusted
6106         and set value to guessed on contradictory divisions.
6107         (profile_probability::combine_with_freq): Remove.
6108         (profile_probability::combine_wiht_count): Declare.
6109         (profile_count::force_nonzero):: Set to adjusted.
6110         (profile_count::probability_in):: Set quality to adjusted.
6111         * tree-ssa-tail-merge.c (replace_block_by): Use
6112         combine_with_count.
6113
6114 2018-01-23  Andrew Waterman  <andrew@sifive.com>
6115             Jim Wilson  <jimw@sifive.com>
6116
6117         * config/riscv/riscv.c (riscv_stack_boundary): New.
6118         (riscv_option_override): Set riscv_stack_boundary.  Handle
6119         riscv_preferred_stack_boundary_arg.
6120         * config/riscv/riscv.h (MIN_STACK_BOUNDARY, ABI_STACK_BOUNDARY): New.
6121         (BIGGEST_ALIGNMENT): Set to STACK_BOUNDARY.
6122         (STACK_BOUNDARY): Set to riscv_stack_boundary.
6123         (RISCV_STACK_ALIGN): Use STACK_BOUNDARY.
6124         * config/riscv/riscv.opt (mpreferred-stack-boundary): New.
6125         * doc/invoke.tex (RISC-V Options): Add -mpreferred-stack-boundary.
6126
6127 2018-01-23  H.J. Lu  <hongjiu.lu@intel.com>
6128
6129         PR target/83905
6130         * config/i386/i386.c (ix86_expand_prologue): Use cost reference
6131         of struct ix86_frame.
6132         (ix86_expand_epilogue): Likewise.  Add a local variable for
6133         the reg_save_offset field in struct ix86_frame.
6134
6135 2018-01-23  Bin Cheng  <bin.cheng@arm.com>
6136
6137         PR tree-optimization/82604
6138         * tree-loop-distribution.c (enum partition_kind): New enum item
6139         PKIND_PARTIAL_MEMSET.
6140         (partition_builtin_p): Support above new enum item.
6141         (generate_code_for_partition): Ditto.
6142         (compute_access_range): Differentiate cases that equality can be
6143         proven at all loops, the innermost loops or no loops.
6144         (classify_builtin_st, classify_builtin_ldst): Adjust call to above
6145         function.  Set PKIND_PARTIAL_MEMSET for partition appropriately.
6146         (finalize_partitions, distribute_loop): Don't fuse partition of
6147         PKIND_PARTIAL_MEMSET kind when distributing 3-level loop nest.
6148         (prepare_perfect_loop_nest): Distribute 3-level loop nest only if
6149         parloop is enabled.
6150
6151 2018-01-23  Martin Liska  <mliska@suse.cz>
6152
6153         * predict.def (PRED_INDIR_CALL): Set probability to PROB_EVEN in
6154         order to ignore the predictor.
6155         (PRED_POLYMORPHIC_CALL): Likewise.
6156         (PRED_RECURSIVE_CALL): Likewise.
6157
6158 2018-01-23  Martin Liska  <mliska@suse.cz>
6159
6160         * tree-profile.c (tree_profiling): Print function header to
6161         aware reader which function we are working on.
6162         * value-prof.c (gimple_find_values_to_profile): Do not print
6163         not interesting value histograms.
6164
6165 2018-01-23  Martin Liska  <mliska@suse.cz>
6166
6167         * profile-count.h (enum profile_quality): Add
6168         profile_uninitialized as the first value. Do not number values
6169         as they are zero based.
6170         (profile_count::verify): Update sanity check.
6171         (profile_probability::verify): Likewise.
6172
6173 2018-01-23  Nathan Sidwell  <nathan@acm.org>
6174
6175         * doc/invoke.texi (ffor-scope): Deprecate.
6176
6177 2018-01-23  David Malcolm  <dmalcolm@redhat.com>
6178
6179         PR tree-optimization/83510
6180         * domwalk.c (set_all_edges_as_executable): New function.
6181         (dom_walker::dom_walker): Convert bool param
6182         "skip_unreachable_blocks" to enum reachability.  Move setup of
6183         edge flags to set_all_edges_as_executable and only do it when
6184         reachability is REACHABLE_BLOCKS.
6185         * domwalk.h (enum dom_walker::reachability): New enum.
6186         (dom_walker::dom_walker): Convert bool param
6187         "skip_unreachable_blocks" to enum reachability.
6188         (set_all_edges_as_executable): New decl.
6189         * graphite-scop-detection.c  (gather_bbs::gather_bbs): Convert
6190         from false for "skip_unreachable_blocks" to ALL_BLOCKS for
6191         "reachability".
6192         * tree-ssa-dom.c (dom_opt_dom_walker::dom_opt_dom_walker): Likewise,
6193         but converting true to REACHABLE_BLOCKS.
6194         * tree-ssa-sccvn.c (sccvn_dom_walker::sccvn_dom_walker): Likewise.
6195         * tree-vrp.c
6196         (check_array_bounds_dom_walker::check_array_bounds_dom_walker):
6197         Likewise, but converting it to REACHABLE_BLOCKS_PRESERVING_FLAGS.
6198         (vrp_dom_walker::vrp_dom_walker): Likewise, but converting it to
6199         REACHABLE_BLOCKS.
6200         (vrp_prop::vrp_finalize): Call set_all_edges_as_executable
6201         if check_all_array_refs will be called.
6202
6203 2018-01-23  David Malcolm  <dmalcolm@redhat.com>
6204
6205         * tree.c (selftest::test_location_wrappers): Add more test
6206         coverage.
6207
6208 2018-01-23  David Malcolm  <dmalcolm@redhat.com>
6209
6210         * sbitmap.c (selftest::test_set_range): Fix memory leaks.
6211         (selftest::test_bit_in_range): Likewise.
6212
6213 2018-01-23  Richard Sandiford  <richard.sandiford@linaro.org>
6214
6215         PR testsuite/83888
6216         * doc/sourcebuild.texi (vect_float): Say that the selector
6217         only describes the situation when -funsafe-math-optimizations is on.
6218         (vect_float_strict): Document.
6219
6220 2018-01-23  Richard Sandiford  <richard.sandiford@linaro.org>
6221
6222         PR tree-optimization/83965
6223         * tree-vect-patterns.c (vect_reassociating_reduction_p): New function.
6224         (vect_recog_dot_prod_pattern, vect_recog_sad_pattern): Use it
6225         instead of checking only for a reduction.
6226         (vect_recog_widen_sum_pattern): Likewise.
6227
6228 2018-01-23  Jan Hubicka  <hubicka@ucw.cz>
6229
6230         * predict.c (probably_never_executed): Only use precise profile info.
6231         (compute_function_frequency): Skip after inlining hack since we now
6232         have quality checking.
6233
6234 2018-01-23  Jan Hubicka  <hubicka@ucw.cz>
6235
6236         * profile-count.h (profile_probability::very_unlikely,
6237         profile_probability::unlikely, profile_probability::even): Set
6238         precision to guessed.
6239
6240 2018-01-23  Richard Biener  <rguenther@suse.de>
6241
6242         PR tree-optimization/83963
6243         * graphite-scop-detection.c (scop_detection::harmful_loop_in_region):
6244         Properly terminate dominator walk when crossing the exit edge not
6245         when visiting its source block.
6246
6247 2018-01-23  Jakub Jelinek  <jakub@redhat.com>
6248
6249         PR c++/83918
6250         * tree.c (maybe_wrap_with_location): Use NON_LVALUE_EXPR rather than
6251         VIEW_CONVERT_EXPR to wrap CONST_DECLs.
6252
6253 2018-01-22  Jakub Jelinek  <jakub@redhat.com>
6254
6255         PR tree-optimization/83957
6256         * omp-expand.c (expand_omp_for_generic): Ignore virtual PHIs.  Remove
6257         semicolon after for body surrounded by braces.
6258
6259         PR tree-optimization/83081
6260         * profile-count.h (profile_probability::split): New method.
6261         * dojump.c (do_jump_1) <case TRUTH_ANDIF_EXPR, case TRUTH_ORIF_EXPR>:
6262         Use profile_probability::split.
6263         (do_compare_rtx_and_jump): Fix adjustment of probabilities
6264         when splitting a single conditional jump into 2.
6265
6266 2018-01-22  David Malcolm  <dmalcolm@redhat.com>
6267
6268         PR tree-optimization/69452
6269         * tree-ssa-loop-im.c (class move_computations_dom_walker): Remove
6270         decl.
6271
6272 2018-01-22  Sebastian Perta  <sebastian.perta@renesas.com>
6273
6274         * config/rl78/rl78-expand.md (bswaphi2): New define_expand.
6275         * config/rl78/rl78-virt.md (*bswaphi2_virt): New define_insn.
6276         * config/rl78/rl78-real.md (*bswaphi2_real): New define_insn.
6277
6278 2018-01-22  Sebastian Perta  <sebastian.perta@renesas.com>
6279
6280         * config/rl78/rl78-protos.h (rl78_split_movdi): New function declaration.
6281         * config/rl78/rl78.md (movdi): New define_expand.
6282         * config/rl78/rl78.c (rl78_split_movdi): New function.
6283
6284 2018-01-22  Michael Meissner  <meissner@linux.vnet.ibm.com>
6285
6286         PR target/83862
6287         * config/rs6000/rs6000-protos.h (rs6000_split_signbit): Delete,
6288         no longer used.
6289         * config/rs6000/rs6000.c (rs6000_split_signbit): Likewise.
6290         * config/rs6000/rs6000.md (signbit<mode>2): Change code for IEEE
6291         128-bit to produce an UNSPEC move to get the double word with the
6292         signbit and then a shift directly to do signbit.
6293         (signbit<mode>2_dm): Replace old IEEE 128-bit signbit
6294         implementation with a new version that just does either a direct
6295         move or a regular move.  Move memory interface to separate insns.
6296         Move insns so they are next to the expander.
6297         (signbit<mode>2_dm_mem_be): New combiner insns to combine load
6298         with signbit move.  Split big and little endian case.
6299         (signbit<mode>2_dm_mem_le): Likewise.
6300         (signbit<mode>2_dm_<su>ext): Delete, no longer used.
6301         (signbit<mode>2_dm2): Likewise.
6302
6303 2018-01-22  Sebastian Perta  <sebastian.perta@renesas.com>
6304
6305         * config/rl78/rl78.md (anddi3): New define_expand.
6306
6307 2018-01-22  Sebastian Perta  <sebastian.perta@renesas.com>
6308
6309         * config/rl78/rl78.md (umindi3): New define_expand.
6310
6311 2018-01-22  Sebastian Perta  <sebastian.perta@renesas.com>
6312
6313         * config/rl78/rl78.md (smindi3): New define_expand.
6314
6315 2018-01-22  Sebastian Perta  <sebastian.perta@renesas.com>
6316
6317         * config/rl78/rl78.md (smaxdi3): New define_expand.
6318
6319 2018-01-22  Carl Love  <cel@us.ibm.com>
6320
6321         * config/rs6000/rs6000-builtin.def (ST_ELEMREV_V1TI, LD_ELEMREV_V1TI,
6322         LVX_V1TI): Add macro expansion.
6323         * config/rs6000/rs6000-c.c (altivec_builtin_types): Add argument
6324         definitions for VSX_BUILTIN_VEC_XST_BE, VSX_BUILTIN_VEC_ST,
6325         VSX_BUILTIN_VEC_XL, LD_ELEMREV_V1TI builtins.
6326         * config/rs6000/rs6000-p8swap.c (insn_is_swappable_p);
6327         Change check to determine if the instruction is a byte reversing
6328         entry.  Fix typo in comment.
6329         * config/rs6000/rs6000.c (altivec_expand_builtin): Add case entry
6330         for VSX_BUILTIN_ST_ELEMREV_V1TI and VSX_BUILTIN_LD_ELEMREV_V1TI.
6331         Add def_builtin calls for new builtins.
6332         * config/rs6000/vsx.md (vsx_st_elemrev_v1ti, vsx_ld_elemrev_v1ti):
6333         Add define_insn expansion.
6334
6335 2018-01-22  Sebastian Perta  <sebastian.perta@renesas.com>
6336
6337         * config/rl78/rl78.md (umaxdi3): New define_expand.
6338
6339 2018-01-22  Sebastian Perta  <sebastian.perta@renesas.com>
6340
6341         * config/rl78/rl78.c (rl78_note_reg_set): Fixed dead reg check
6342         for non-QImode registers.
6343
6344 2018-01-22  Richard Biener  <rguenther@suse.de>
6345
6346         PR tree-optimization/83963
6347         * graphite-scop-detection.c (scop_detection::get_sese): Delay
6348         including the loop exit block.
6349         (scop_detection::merge_sese): Likewise.
6350         (scop_detection::add_scop): Do it here instead.
6351
6352 2018-01-22  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
6353
6354         * doc/sourcebuild.texi (arm_softfloat): Document.
6355
6356 2018-01-21  John David Anglin  <danglin@gcc.gnu.org>
6357
6358         PR gcc/77734
6359         * config/pa/pa.c (pa_function_ok_for_sibcall): Use
6360         targetm.binds_local_p instead of TREE_PUBLIC to check local binding.
6361         Move TARGET_PORTABLE_RUNTIME check after TARGET_64BIT check.
6362
6363 2018-01-21  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
6364             David Edelsohn  <dje.gcc@gmail.com>
6365
6366         PR target/83946
6367         * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
6368         Change "crset eq" to "crset 2".
6369         (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
6370         (*call_indirect_aix<mode>_nospec): Likewise.
6371         (*call_value_indirect_aix<mode>_nospec): Likewise.
6372         (*call_indirect_elfv2<mode>_nospec): Likewise.
6373         (*call_value_indirect_elfv2<mode>_nospec): Likewise.
6374         (*sibcall_nonlocal_sysv<mode>): Change "crset eq" to "crset 2";
6375         change assembly output from . to $.
6376         (*sibcall_value_nonlocal_sysv<mode>): Likewise.
6377         (indirect_jump<mode>_nospec): Change assembly output from . to $.
6378         (*tablejump<mode>_internal1_nospec): Likewise.
6379
6380 2018-01-21  Oleg Endo  <olegendo@gcc.gnu.org>
6381
6382         PR target/80870
6383         * config/sh/sh_optimize_sett_clrt.cc:
6384         Use INCLUDE_ALGORITHM and INCLUDE_VECTOR instead of direct includes.
6385
6386 2018-01-20  Richard Sandiford  <richard.sandiford@linaro.org>
6387
6388         PR tree-optimization/83940
6389         * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): Set
6390         offset_dt to vect_constant_def rather than vect_unknown_def_type.
6391         (vect_check_load_store_mask): Add a mask_dt_out parameter and
6392         use it to pass back the definition type.
6393         (vect_check_store_rhs): Likewise rhs_dt_out.
6394         (vect_build_gather_load_calls): Add a mask_dt argument and use
6395         it instead of a call to vect_is_simple_use.
6396         (vectorizable_store): Update calls to vect_check_load_store_mask
6397         and vect_check_store_rhs.  Use the dt returned by the latter instead
6398         of scatter_src_dt.  Use the cached mask_dt and gs_info.offset_dt
6399         instead of calls to vect_is_simple_use.  Pass the scalar rather
6400         than the vector operand to vect_is_simple_use when handling
6401         second and subsequent copies of an rhs value.
6402         (vectorizable_load): Update calls to vect_check_load_store_mask
6403         and vect_build_gather_load_calls.  Use the cached mask_dt and
6404         gs_info.offset_dt instead of calls to vect_is_simple_use.
6405
6406 2018-01-20  Jakub Jelinek  <jakub@redhat.com>
6407
6408         PR middle-end/83945
6409         * tree-emutls.c: Include gimplify.h.
6410         (lower_emutls_2): New function.
6411         (lower_emutls_1): If ADDR_EXPR is a gimple invariant and walk_tree
6412         with lower_emutls_2 callback finds some TLS decl in it, unshare_expr
6413         it before further processing.
6414
6415         PR target/83930
6416         * simplify-rtx.c (simplify_binary_operation_1) <case UMOD>: Use
6417         UINTVAL (trueop1) instead of INTVAL (op1).
6418
6419 2018-01-19  Jakub Jelinek  <jakub@redhat.com>
6420
6421         PR debug/81570
6422         PR debug/83728
6423         * dwarf2cfi.c (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define to
6424         INCOMING_FRAME_SP_OFFSET if not defined.
6425         (scan_trace): Add ENTRY argument.  If true and
6426         DEFAULT_INCOMING_FRAME_SP_OFFSET != INCOMING_FRAME_SP_OFFSET,
6427         emit a note to adjust the CFA offset.
6428         (create_cfi_notes): Adjust scan_trace callers.
6429         (create_cie_data): Use DEFAULT_INCOMING_FRAME_SP_OFFSET rather than
6430         INCOMING_FRAME_SP_OFFSET in the CIE.
6431         * config/i386/i386.h (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define.
6432         * config/stormy16/stormy16.h (DEFAULT_INCOMING_FRAME_SP_OFFSET):
6433         Likewise.
6434         * doc/tm.texi.in (DEFAULT_INCOMING_FRAME_SP_OFFSET): Document.
6435         * doc/tm.texi: Regenerated.
6436
6437 2018-01-19  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>
6438
6439         PR rtl-optimization/83147
6440         * lra-constraints.c (remove_inheritance_pseudos): Use
6441         lra_substitute_pseudo_within_insn.
6442
6443 2018-01-19  Tom de Vries  <tom@codesourcery.com>
6444             Cesar Philippidis  <cesar@codesourcery.com>
6445
6446         PR target/83920
6447         * config/nvptx/nvptx.c (nvptx_single): Fix jit workaround.
6448
6449 2018-01-19  Cesar Philippidis  <cesar@codesourcery.com>
6450
6451         PR target/83790
6452         * config/nvptx/nvptx.c (output_init_frag): Don't use generic address
6453         spaces for function labels.
6454
6455 2018-01-19  Martin Liska  <mliska@suse.cz>
6456
6457         * predict.def (PRED_LOOP_EXIT): Change from 85 to 89.
6458         (PRED_LOOP_EXIT_WITH_RECURSION): Change from 72 to 78.
6459         (PRED_LOOP_EXTRA_EXIT): Change from 83 to 67.
6460         (PRED_OPCODE_POSITIVE): Change from 64 to 59.
6461         (PRED_TREE_OPCODE_POSITIVE): Change from 64 to 59.
6462         (PRED_CONST_RETURN): Change from 69 to 65.
6463         (PRED_NULL_RETURN): Change from 91 to 71.
6464         (PRED_LOOP_IV_COMPARE_GUESS): Change from 98 to 64.
6465         (PRED_LOOP_GUARD): Change from 66 to 73.
6466
6467 2018-01-19  Martin Liska  <mliska@suse.cz>
6468
6469         * predict.c (predict_insn_def): Add new assert.
6470         (struct branch_predictor): Change type to signed integer.
6471         (test_prediction_value_range): Amend test to cover
6472         PROB_UNINITIALIZED.
6473         * predict.def (PRED_LOOP_ITERATIONS): Use the new constant.
6474         (PRED_LOOP_ITERATIONS_GUESSED): Likewise.
6475         (PRED_LOOP_ITERATIONS_MAX): Likewise.
6476         (PRED_LOOP_IV_COMPARE): Likewise.
6477         * predict.h (PROB_UNINITIALIZED): Define new constant.
6478
6479 2018-01-19  Martin Liska  <mliska@suse.cz>
6480
6481         * predict.c (dump_prediction): Add new format for
6482         analyze_brprob.py script which is enabled with -details
6483         suboption.
6484         * profile-count.h (precise_p): New function.
6485
6486 2018-01-19  Richard Sandiford  <richard.sandiford@linaro.org>
6487
6488         PR tree-optimization/83922
6489         * tree-vect-loop.c (vect_verify_full_masking): Return false if
6490         there are no statements that need masking.
6491         (vect_active_double_reduction_p): New function.
6492         (vect_analyze_loop_operations): Use it when handling phis that
6493         are not in the loop header.
6494
6495 2018-01-19  Richard Sandiford  <richard.sandiford@linaro.org>
6496
6497         PR tree-optimization/83914
6498         * tree-vect-loop.c (vectorizable_induction): Don't convert
6499         init_expr or apply the peeling adjustment for inductions
6500         that are nested within the vectorized loop.
6501
6502 2018-01-19  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
6503
6504         * config/arm/thumb2.md (*thumb2_negsi2_short): Use RSB mnemonic
6505         instead of NEG.
6506
6507 2018-01-18  Jakub Jelinek  <jakub@redhat.com>
6508
6509         PR sanitizer/81715
6510         PR testsuite/83882
6511         * function.h (gimplify_parameters): Add gimple_seq * argument.
6512         * function.c: Include gimple.h and options.h.
6513         (gimplify_parameters): Add cleanup argument, add CLOBBER stmts
6514         for the added local temporaries if needed.
6515         * gimplify.c (gimplify_body): Adjust gimplify_parameters caller,
6516         if there are any parameter cleanups, wrap whole body into a
6517         try/finally with the cleanups.
6518
6519 2018-01-18  Wilco Dijkstra  <wdijkstr@arm.com>
6520
6521         PR target/82964
6522         * config/aarch64/aarch64.c (aarch64_legitimate_constant_p):
6523         Use GET_MODE_CLASS for scalar floating point.
6524
6525 2018-01-18  Jan Hubicka  <hubicka@ucw.cz>
6526
6527         PR ipa/82256
6528         patch by PaX Team
6529         * cgraphclones.c (cgraph_node::create_version_clone_with_body):
6530         Fix call of call_cgraph_insertion_hooks.
6531
6532 2018-01-18  Martin Sebor  <msebor@redhat.com>
6533
6534         * doc/invoke.texi (-Wclass-memaccess): Tweak text.
6535
6536 2018-01-18  Jan Hubicka  <hubicka@ucw.cz>
6537
6538         PR ipa/83619
6539         * cgraph.c (cgraph_edge::redirect_call_stmt_to_callee): Update edge
6540         frequencies.
6541
6542 2018-01-18  Boris Kolpackov  <boris@codesynthesis.com>
6543
6544         PR other/70268
6545         * common.opt: (-ffile-prefix-map): New option.
6546         * opts.c (common_handle_option): Defer it.
6547         * opts-global.c (handle_common_deferred_options): Handle it.
6548         * debug.h (remap_debug_filename, add_debug_prefix_map): Move to...
6549         * file-prefix-map.h: New file.
6550         (remap_debug_filename, add_debug_prefix_map): ...here.
6551         (add_macro_prefix_map, add_file_prefix_map, remap_macro_filename): New.
6552         * final.c (debug_prefix_map, add_debug_prefix_map
6553         remap_debug_filename): Move to...
6554         * file-prefix-map.c: New file.
6555         (file_prefix_map, add_prefix_map, remap_filename) ...here and rename,
6556         generalize, get rid of alloca(), use strrchr() instead of strchr().
6557         (add_macro_prefix_map, add_debug_prefix_map, add_file_prefix_map):
6558         Implement in terms of add_prefix_map().
6559         (remap_macro_filename, remap_debug_filename): Implement in term of
6560         remap_filename().
6561         * Makefile.in (OBJS, PLUGIN_HEADERS): Add new files.
6562         * builtins.c (fold_builtin_FILE): Call remap_macro_filename().
6563         * dbxout.c: Include file-prefix-map.h.
6564         * varasm.c: Likewise.
6565         * vmsdbgout.c: Likewise.
6566         * xcoffout.c: Likewise.
6567         * dwarf2out.c: Likewise plus omit new options from DW_AT_producer.
6568         * doc/cppopts.texi (-fmacro-prefix-map): Document.
6569         * doc/invoke.texi (-ffile-prefix-map): Document.
6570         (-fdebug-prefix-map): Update description.
6571
6572 2018-01-18  Martin Liska  <mliska@suse.cz>
6573
6574         * config/i386/i386.c (indirect_thunk_name): Document that also
6575         lfence is emitted.
6576         (output_indirect_thunk): Document why both instructions
6577         (pause and lfence) are generated.
6578
6579 2018-01-18  Richard Biener  <rguenther@suse.de>
6580
6581         PR tree-optimization/83887
6582         * graphite-scop-detection.c
6583         (scop_detection::get_nearest_dom_with_single_entry): Remove.
6584         (scop_detection::get_nearest_pdom_with_single_exit): Likewise.
6585         (scop_detection::merge_sese): Re-implement with a flood-fill
6586         algorithm that properly finds a SESE region if it exists.
6587
6588 2018-01-18  Jakub Jelinek  <jakub@redhat.com>
6589
6590         PR c/61240
6591         * match.pd ((P + A) - P, P - (P + A), (P + A) - (P + B)): For
6592         pointer_diff optimizations use view_convert instead of convert.
6593
6594 2018-01-17  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
6595
6596         * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
6597         Generate different code for -mno-speculate-indirect-jumps.
6598         (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
6599         (*call_indirect_aix<mode>): Disable for
6600         -mno-speculate-indirect-jumps.
6601         (*call_indirect_aix<mode>_nospec): New define_insn.
6602         (*call_value_indirect_aix<mode>): Disable for
6603         -mno-speculate-indirect-jumps.
6604         (*call_value_indirect_aix<mode>_nospec): New define_insn.
6605         (*sibcall_nonlocal_sysv<mode>): Generate different code for
6606         -mno-speculate-indirect-jumps.
6607         (*sibcall_value_nonlocal_sysv<mode>): Likewise.
6608
6609 2018-01-17  Michael Meissner  <meissner@linux.vnet.ibm.com>
6610
6611         * config/rs6000/rs6000.c (rs6000_emit_move): If we load or store a
6612         long double type, set the flags for noting the default long double
6613         type, even if we don't pass or return a long double type.
6614
6615 2018-01-17  Jan Hubicka  <hubicka@ucw.cz>
6616
6617         PR ipa/83051
6618         * ipa-inline.c (flatten_function): Do not overwrite final inlining
6619         failure.
6620
6621 2018-01-17  Will Schmidt  <will_schmidt@vnet.ibm.com>
6622
6623         * config/rs6000/rs6000.c (rs6000_gimple_builtin): Add gimple folding
6624         support for merge[hl].
6625         (fold_mergehl_helper): New helper function.
6626         (tree-vector-builder.h): New #include for tree_vector_builder usage.
6627         * config/rs6000/altivec.md (altivec_vmrghw_direct): Add xxmrghw insn.
6628         (altivec_vmrglw_direct): Add xxmrglw insn.
6629
6630 2018-01-17  Andrew Waterman  <andrew@sifive.com>
6631
6632         * config/riscv/riscv.c (riscv_conditional_register_usage): If
6633         UNITS_PER_FP_ARG is 0, set call_used_regs to 1 for all FP regs.
6634
6635 2018-01-17  David Malcolm  <dmalcolm@redhat.com>
6636
6637         PR lto/83121
6638         * ipa-devirt.c (add_type_duplicate): When comparing memory layout,
6639         call the lto_location_cache before reading the
6640         DECL_SOURCE_LOCATION of the types.
6641
6642 2018-01-17  Wilco Dijkstra  <wdijkstr@arm.com>
6643             Richard Sandiford  <richard.sandiford@linaro.org>
6644
6645         * config/aarch64/aarch64.md (movti_aarch64): Use Uti constraint.
6646         * config/aarch64/aarch64.c (aarch64_mov128_immediate): New function.
6647         (aarch64_legitimate_constant_p): Just support CONST_DOUBLE
6648         SF/DF/TF mode to avoid creating illegal CONST_WIDE_INT immediates.
6649         * config/aarch64/aarch64-protos.h (aarch64_mov128_immediate):
6650         Add declaration.
6651         * config/aarch64/constraints.md (aarch64_movti_operand):
6652         Limit immediates.
6653         * config/aarch64/predicates.md (Uti): Add new constraint.
6654
6655 2018-01-17 Carl Love  <cel@us.ibm.com>
6656
6657         * config/rs6000/vsx.md (define_expand xl_len_r,
6658         define_expand stxvl, define_expand *stxvl): Add match_dup argument.
6659         (define_insn): Add, match_dup 1 argument to define_insn stxvll and
6660         lxvll.
6661         (define_expand, define_insn): Move the shift left from  the
6662         define_insn to the define_expand for lxvl and stxvl instructions.
6663         * config/rs6000/rs6000-builtin.def (BU_P9V_64BIT_VSX_2): Change LXVL
6664         and XL_LEN_R definitions to PURE.
6665
6666 2018-01-17  Uros Bizjak  <ubizjak@gmail.com>
6667
6668         * config/i386/i386.c (indirect_thunk_name): Declare regno
6669         as unsigned int.  Compare regno with INVALID_REGNUM.
6670         (output_indirect_thunk): Ditto.
6671         (output_indirect_thunk_function): Ditto.
6672         (ix86_code_end): Declare regno as unsigned int.  Use INVALID_REGNUM
6673         in the call to output_indirect_thunk_function.
6674
6675 2018-01-17  Richard Sandiford  <richard.sandiford@linaro.org>
6676
6677         PR middle-end/83884
6678         * expr.c (expand_expr_real_1): Use the size of GET_MODE (op0)
6679         rather than the size of inner_type to determine the stack slot size
6680         when handling VIEW_CONVERT_EXPRs on strict-alignment targets.
6681
6682 2018-01-16  Sebastian Peryt  <sebastian.peryt@intel.com>
6683
6684         PR target/83546
6685         * config/i386/i386.c (ix86_option_override_internal): Add PTA_RDRND
6686         to PTA_SILVERMONT.
6687
6688 2018-01-16  Michael Meissner  <meissner@linux.vnet.ibm.com>
6689
6690         * config.gcc (powerpc*-linux*-*): Add support for 64-bit little
6691         endian Linux systems to optionally enable multilibs for selecting
6692         the long double type if the user configured an explicit type.
6693         * config/rs6000/rs6000.h (TARGET_IEEEQUAD_MULTILIB): Indicate we
6694         have no long double multilibs if not defined.
6695         * config/rs6000/rs6000.c (rs6000_option_override_internal): Do not
6696         warn if the user used -mabi={ieee,ibm}longdouble and we built
6697         multilibs for long double.
6698         * config/rs6000/linux64.h (MULTILIB_DEFAULTS_IEEE): Define as the
6699         appropriate multilib option.
6700         (MULTILIB_DEFAULTS): Add MULTILIB_DEFAULTS_IEEE to the default
6701         multilib options.
6702         * config/rs6000/t-ldouble-linux64le-ibm: New configuration files
6703         for building long double multilibs.
6704         * config/rs6000/t-ldouble-linux64le-ieee: Likewise.
6705
6706 2018-01-16  John David Anglin  <danglin@gcc.gnu.org>
6707
6708         * config.gcc (hppa*-*-linux*): Change callee copies ABI to caller
6709         copies.
6710
6711         * config/pa.h (MALLOC_ABI_ALIGNMENT): Set 32-bit alignment default to
6712         64 bits.
6713         * config/pa/pa32-linux.h (MALLOC_ABI_ALIGNMENT): Set alignment to
6714         128 bits.
6715
6716         * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Cleanup type and mode
6717         variables.
6718
6719         * config/pa/pa.c (pa_function_arg_size): Apply CEIL to GET_MODE_SIZE
6720         return value.
6721
6722 2018-01-16  Eric Botcazou  <ebotcazou@adacore.com>
6723
6724         * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For an
6725         ADDR_EXPR, do not count the offset of a COMPONENT_REF twice.
6726
6727 2018-01-16  Kelvin Nilsen  <kelvin@gcc.gnu.org>
6728
6729         * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Generate
6730         different rtl trees depending on TARGET_64BIT.
6731         (rs6000_gen_lvx): Likewise.
6732
6733 2018-01-16  Eric Botcazou  <ebotcazou@adacore.com>
6734
6735         * config/visium/visium.md (nop): Tweak comment.
6736         (hazard_nop): Likewise.
6737
6738 2018-01-16  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
6739
6740         * config/rs6000/rs6000.c (rs6000_opt_vars): Add entry for
6741         -mspeculate-indirect-jumps.
6742         * config/rs6000/rs6000.md (*call_indirect_elfv2<mode>): Disable
6743         for -mno-speculate-indirect-jumps.
6744         (*call_indirect_elfv2<mode>_nospec): New define_insn.
6745         (*call_value_indirect_elfv2<mode>): Disable for
6746         -mno-speculate-indirect-jumps.
6747         (*call_value_indirect_elfv2<mode>_nospec): New define_insn.
6748         (indirect_jump): Emit different RTL for
6749         -mno-speculate-indirect-jumps.
6750         (*indirect_jump<mode>): Disable for
6751         -mno-speculate-indirect-jumps.
6752         (*indirect_jump<mode>_nospec): New define_insn.
6753         (tablejump): Emit different RTL for
6754         -mno-speculate-indirect-jumps.
6755         (tablejumpsi): Disable for -mno-speculate-indirect-jumps.
6756         (tablejumpsi_nospec): New define_expand.
6757         (tablejumpdi): Disable for -mno-speculate-indirect-jumps.
6758         (tablejumpdi_nospec): New define_expand.
6759         (*tablejump<mode>_internal1): Disable for
6760         -mno-speculate-indirect-jumps.
6761         (*tablejump<mode>_internal1_nospec): New define_insn.
6762         * config/rs6000/rs6000.opt (mspeculate-indirect-jumps): New
6763         option.
6764
6765 2018-01-16  Artyom Skrobov tyomitch@gmail.com
6766
6767         * caller-save.c (insert_save): Drop unnecessary parameter.  All
6768         callers updated.
6769
6770 2018-01-16  Jakub Jelinek  <jakub@redhat.com>
6771             Richard Biener  <rguenth@suse.de>
6772
6773         PR libgomp/83590
6774         * gimplify.c (gimplify_one_sizepos): For is_gimple_constant (expr)
6775         return early, inline manually is_gimple_sizepos.  Make sure if we
6776         call gimplify_expr we don't end up with a gimple constant.
6777         * tree.c (variably_modified_type_p): Don't return true for
6778         is_gimple_constant (_t).  Inline manually is_gimple_sizepos.
6779         * gimplify.h (is_gimple_sizepos): Remove.
6780
6781 2018-01-16  Richard Sandiford  <richard.sandiford@linaro.org>
6782
6783         PR tree-optimization/83857
6784         * tree-vect-loop.c (vect_analyze_loop_operations): Don't call
6785         vectorizable_live_operation for pure SLP statements.
6786         (vectorizable_live_operation): Handle PHIs.
6787
6788 2018-01-16  Richard Biener  <rguenther@suse.de>
6789
6790         PR tree-optimization/83867
6791         * tree-vect-stmts.c (vect_transform_stmt): Precompute
6792         nested_in_vect_loop_p since the scalar stmt may get invalidated.
6793
6794 2018-01-16  Jakub Jelinek  <jakub@redhat.com>
6795
6796         PR c/83844
6797         * stor-layout.c (handle_warn_if_not_align): Use byte_position and
6798         multiple_of_p instead of unchecked tree_to_uhwi and UHWI check.
6799         If off is not INTEGER_CST, issue a may not be aligned warning
6800         rather than isn't aligned.  Use isn%'t rather than isn't.
6801         * fold-const.c (multiple_of_p) <case BIT_AND_EXPR>: Don't fall through
6802         into MULT_EXPR.
6803         <case MULT_EXPR>: Improve the case when bottom and one of the
6804         MULT_EXPR operands are INTEGER_CSTs and bottom is multiple of that
6805         operand, in that case check if the other operand is multiple of
6806         bottom divided by the INTEGER_CST operand.
6807
6808 2018-01-16  Richard Sandiford  <richard.sandiford@linaro.org>
6809
6810         PR target/83858
6811         * config/pa/pa.h (FUNCTION_ARG_SIZE): Delete.
6812         * config/pa/pa-protos.h (pa_function_arg_size): Declare.
6813         * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Use
6814         pa_function_arg_size instead of FUNCTION_ARG_SIZE.
6815         * config/pa/pa.c (pa_function_arg_advance): Likewise.
6816         (pa_function_arg, pa_arg_partial_bytes): Likewise.
6817         (pa_function_arg_size): New function.
6818
6819 2018-01-16  Richard Sandiford  <richard.sandiford@linaro.org>
6820
6821         * fold-const.c (fold_ternary_loc): Construct the vec_perm_indices
6822         in a separate statement.
6823
6824 2018-01-16  Richard Sandiford  <richard.sandiford@linaro.org>
6825
6826         PR tree-optimization/83847
6827         * tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Don't
6828         group gathers and scatters.
6829
6830 2018-01-16  Jakub Jelinek  <jakub@redhat.com>
6831
6832         PR rtl-optimization/86620
6833         * params.def (max-sched-ready-insns): Bump minimum value to 1.
6834
6835         PR rtl-optimization/83213
6836         * recog.c (peep2_attempt): Copy over CROSSING_JUMP_P from peepinsn
6837         to last if both are JUMP_INSNs.
6838
6839         PR tree-optimization/83843
6840         * gimple-ssa-store-merging.c
6841         (imm_store_chain_info::output_merged_store): Handle bit_not_p on
6842         store_immediate_info for bswap/nop orig_stores.
6843
6844 2018-01-15  Andrew Waterman  <andrew@sifive.com>
6845
6846         * config/riscv/riscv.c (riscv_rtx_costs) <MULT>: Increase cost if
6847         !TARGET_MUL.
6848         <UDIV>: Increase cost if !TARGET_DIV.
6849
6850 2018-01-15  Segher Boessenkool  <segher@kernel.crashing.org>
6851
6852         * config/rs6000/rs6000.md (define_attr "type"): Remove delayed_cr.
6853         (define_attr "cr_logical_3op"): New.
6854         (cceq_ior_compare): Adjust.
6855         (cceq_ior_compare_complement): Adjust.
6856         (*cceq_rev_compare): Adjust.
6857         * config/rs6000/rs6000.c (rs6000_adjust_cost): Adjust.
6858         (is_cracked_insn): Adjust.
6859         (insn_must_be_first_in_group): Adjust.
6860         * config/rs6000/40x.md: Adjust.
6861         * config/rs6000/440.md: Adjust.
6862         * config/rs6000/476.md: Adjust.
6863         * config/rs6000/601.md: Adjust.
6864         * config/rs6000/603.md: Adjust.
6865         * config/rs6000/6xx.md: Adjust.
6866         * config/rs6000/7450.md: Adjust.
6867         * config/rs6000/7xx.md: Adjust.
6868         * config/rs6000/8540.md: Adjust.
6869         * config/rs6000/cell.md: Adjust.
6870         * config/rs6000/e300c2c3.md: Adjust.
6871         * config/rs6000/e500mc.md: Adjust.
6872         * config/rs6000/e500mc64.md: Adjust.
6873         * config/rs6000/e5500.md: Adjust.
6874         * config/rs6000/e6500.md: Adjust.
6875         * config/rs6000/mpc.md: Adjust.
6876         * config/rs6000/power4.md: Adjust.
6877         * config/rs6000/power5.md: Adjust.
6878         * config/rs6000/power6.md: Adjust.
6879         * config/rs6000/power7.md: Adjust.
6880         * config/rs6000/power8.md: Adjust.
6881         * config/rs6000/power9.md: Adjust.
6882         * config/rs6000/rs64.md: Adjust.
6883         * config/rs6000/titan.md: Adjust.
6884
6885 2018-01-15  H.J. Lu  <hongjiu.lu@intel.com>
6886
6887         * config/i386/predicates.md (indirect_branch_operand): Rewrite
6888         ix86_indirect_branch_register logic.
6889
6890 2018-01-15  H.J. Lu  <hongjiu.lu@intel.com>
6891
6892         * config/i386/constraints.md (Bs): Update
6893         ix86_indirect_branch_register check.  Don't check
6894         ix86_indirect_branch_register with GOT_memory_operand.
6895         (Bw): Likewise.
6896         * config/i386/predicates.md (GOT_memory_operand): Don't check
6897         ix86_indirect_branch_register here.
6898         (GOT32_symbol_operand): Likewise.
6899
6900 2018-01-15  H.J. Lu  <hongjiu.lu@intel.com>
6901
6902         * config/i386/predicates.md (constant_call_address_operand):
6903         Rewrite ix86_indirect_branch_register logic.
6904         (sibcall_insn_operand): Likewise.
6905
6906 2018-01-15  H.J. Lu  <hongjiu.lu@intel.com>
6907
6908         * config/i386/constraints.md (Bs): Replace
6909         ix86_indirect_branch_thunk_register with
6910         ix86_indirect_branch_register.
6911         (Bw): Likewise.
6912         * config/i386/i386.md (indirect_jump): Likewise.
6913         (tablejump): Likewise.
6914         (*sibcall_memory): Likewise.
6915         (*sibcall_value_memory): Likewise.
6916         Peepholes of indirect call and jump via memory: Likewise.
6917         * config/i386/i386.opt: Likewise.
6918         * config/i386/predicates.md (indirect_branch_operand): Likewise.
6919         (GOT_memory_operand): Likewise.
6920         (call_insn_operand): Likewise.
6921         (sibcall_insn_operand): Likewise.
6922         (GOT32_symbol_operand): Likewise.
6923
6924 2018-01-15  Jakub Jelinek  <jakub@redhat.com>
6925
6926         PR middle-end/83837
6927         * omp-expand.c (expand_omp_atomic_pipeline): Use loaded_val
6928         type rather than type addr's type points to.
6929         (expand_omp_atomic_mutex): Likewise.
6930         (expand_omp_atomic): Likewise.
6931
6932 2018-01-15  H.J. Lu  <hongjiu.lu@intel.com>
6933
6934         PR target/83839
6935         * config/i386/i386.c (output_indirect_thunk_function): Use
6936         ASM_OUTPUT_LABEL, instead of ASM_OUTPUT_DEF, for TARGET_MACHO
6937         for  __x86_return_thunk.
6938
6939 2018-01-15  Richard Biener  <rguenther@suse.de>
6940
6941         PR middle-end/83850
6942         * expmed.c (extract_bit_field_1): Fix typo.
6943
6944 2018-01-15  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
6945
6946         PR target/83687
6947         * config/arm/iterators.md (VF): New mode iterator.
6948         * config/arm/neon.md (neon_vabd<mode>_2): Use the above.
6949         Remove integer-related logic from pattern.
6950         (neon_vabd<mode>_3): Likewise.
6951
6952 2018-01-15  Jakub Jelinek  <jakub@redhat.com>
6953
6954         PR middle-end/82694
6955         * common.opt (fstrict-overflow): No longer an alias.
6956         (fwrapv-pointer): New option.
6957         * tree.h (TYPE_OVERFLOW_WRAPS, TYPE_OVERFLOW_UNDEFINED): Define
6958         also for pointer types based on flag_wrapv_pointer.
6959         * opts.c (common_handle_option) <case OPT_fstrict_overflow>: Set
6960         opts->x_flag_wrap[pv] to !value, clear opts->x_flag_trapv if
6961         opts->x_flag_wrapv got set.
6962         * fold-const.c (fold_comparison, fold_binary_loc): Revert 2017-08-01
6963         changes, just use TYPE_OVERFLOW_UNDEFINED on pointer type instead of
6964         POINTER_TYPE_OVERFLOW_UNDEFINED.
6965         * match.pd: Likewise in address comparison pattern.
6966         * doc/invoke.texi: Document -fwrapv and -fstrict-overflow.
6967
6968 2018-01-15  Richard Biener  <rguenther@suse.de>
6969
6970         PR lto/83804
6971         * tree.c (free_lang_data_in_type): Always unlink TYPE_DECLs
6972         from TYPE_FIELDS.  Free TYPE_BINFO if not used by devirtualization.
6973         Reset type names to their identifier if their TYPE_DECL doesn't
6974         have linkage (and thus is used for ODR and devirt).
6975         (save_debug_info_for_decl): Remove.
6976         (save_debug_info_for_type): Likewise.
6977         (add_tree_to_fld_list): Adjust.
6978         * tree-pretty-print.c (dump_generic_node): Make dumping of
6979         type names more robust.
6980
6981 2018-01-15  Richard Biener  <rguenther@suse.de>
6982
6983         * BASE-VER: Bump to 8.0.1.
6984
6985 2018-01-14  Martin Sebor  <msebor@redhat.com>
6986
6987         PR other/83508
6988         * builtins.c (check_access): Avoid warning when the no-warning bit
6989         is set.
6990
6991 2018-01-14  Cory Fields  <cory-nospam-@coryfields.com>
6992
6993         * tree-ssa-loop-im.c (sort_bbs_in_loop_postorder_cmp): Stabilize sort.
6994         * ira-color (allocno_hard_regs_compare): Likewise.
6995
6996 2018-01-14  Nathan Rossi  <nathan@nathanrossi.com>
6997
6998         PR target/83013
6999         * config/microblaze/microblaze.c (microblaze_asm_output_ident):
7000         Use .pushsection/.popsection.
7001
7002 2018-01-14  Martin Sebor  <msebor@redhat.com>
7003
7004         PR c++/81327
7005         * doc/invoke.texi (-Wlass-memaccess): Document suppression by casting.
7006
7007 2018-01-14  Jakub Jelinek  <jakub@redhat.com>
7008
7009         * config.gcc (i[34567]86-*-*): Remove one duplicate gfniintrin.h
7010         entry from extra_headers.
7011         (x86_64-*-*): Remove two duplicate gfniintrin.h entries from
7012         extra_headers, make the list bitwise identical to the i?86-*-* one.
7013
7014 2018-01-14  H.J. Lu  <hongjiu.lu@intel.com>
7015
7016         * config/i386/i386.c (ix86_set_indirect_branch_type): Disallow
7017         -mcmodel=large with -mindirect-branch=thunk,
7018         -mindirect-branch=thunk-extern, -mfunction-return=thunk and
7019         -mfunction-return=thunk-extern.
7020         * doc/invoke.texi: Document -mcmodel=large is incompatible with
7021         -mindirect-branch=thunk, -mindirect-branch=thunk-extern,
7022         -mfunction-return=thunk and -mfunction-return=thunk-extern.
7023
7024 2018-01-14  H.J. Lu  <hongjiu.lu@intel.com>
7025
7026         * config/i386/i386.c (print_reg): Print the name of the full
7027         integer register without '%'.
7028         (ix86_print_operand): Handle 'V'.
7029          * doc/extend.texi: Document 'V' modifier.
7030
7031 2018-01-14  H.J. Lu  <hongjiu.lu@intel.com>
7032
7033         * config/i386/constraints.md (Bs): Disallow memory operand for
7034         -mindirect-branch-register.
7035         (Bw): Likewise.
7036         * config/i386/predicates.md (indirect_branch_operand): Likewise.
7037         (GOT_memory_operand): Likewise.
7038         (call_insn_operand): Likewise.
7039         (sibcall_insn_operand): Likewise.
7040         (GOT32_symbol_operand): Likewise.
7041         * config/i386/i386.md (indirect_jump): Call convert_memory_address
7042         for -mindirect-branch-register.
7043         (tablejump): Likewise.
7044         (*sibcall_memory): Likewise.
7045         (*sibcall_value_memory): Likewise.
7046         Disallow peepholes of indirect call and jump via memory for
7047         -mindirect-branch-register.
7048         (*call_pop): Replace m with Bw.
7049         (*call_value_pop): Likewise.
7050         (*sibcall_pop_memory): Replace m with Bs.
7051         * config/i386/i386.opt (mindirect-branch-register): New option.
7052         * doc/invoke.texi: Document -mindirect-branch-register option.
7053
7054 2018-01-14  H.J. Lu  <hongjiu.lu@intel.com>
7055
7056         * config/i386/i386-protos.h (ix86_output_function_return): New.
7057         * config/i386/i386.c (ix86_set_indirect_branch_type): Also
7058         set function_return_type.
7059         (indirect_thunk_name): Add ret_p to indicate thunk for function
7060         return.
7061         (output_indirect_thunk_function): Pass false to
7062         indirect_thunk_name.
7063         (ix86_output_indirect_branch_via_reg): Likewise.
7064         (ix86_output_indirect_branch_via_push): Likewise.
7065         (output_indirect_thunk_function): Create alias for function
7066         return thunk if regno < 0.
7067         (ix86_output_function_return): New function.
7068         (ix86_handle_fndecl_attribute): Handle function_return.
7069         (ix86_attribute_table): Add function_return.
7070         * config/i386/i386.h (machine_function): Add
7071         function_return_type.
7072         * config/i386/i386.md (simple_return_internal): Use
7073         ix86_output_function_return.
7074         (simple_return_internal_long): Likewise.
7075         * config/i386/i386.opt (mfunction-return=): New option.
7076         (indirect_branch): Mention -mfunction-return=.
7077         * doc/extend.texi: Document function_return function attribute.
7078         * doc/invoke.texi: Document -mfunction-return= option.
7079
7080 2018-01-14  H.J. Lu  <hongjiu.lu@intel.com>
7081
7082         * config/i386/i386-opts.h (indirect_branch): New.
7083         * config/i386/i386-protos.h (ix86_output_indirect_jmp): Likewise.
7084         * config/i386/i386.c (ix86_using_red_zone): Disallow red-zone
7085         with local indirect jump when converting indirect call and jump.
7086         (ix86_set_indirect_branch_type): New.
7087         (ix86_set_current_function): Call ix86_set_indirect_branch_type.
7088         (indirectlabelno): New.
7089         (indirect_thunk_needed): Likewise.
7090         (indirect_thunk_bnd_needed): Likewise.
7091         (indirect_thunks_used): Likewise.
7092         (indirect_thunks_bnd_used): Likewise.
7093         (INDIRECT_LABEL): Likewise.
7094         (indirect_thunk_name): Likewise.
7095         (output_indirect_thunk): Likewise.
7096         (output_indirect_thunk_function): Likewise.
7097         (ix86_output_indirect_branch_via_reg): Likewise.
7098         (ix86_output_indirect_branch_via_push): Likewise.
7099         (ix86_output_indirect_branch): Likewise.
7100         (ix86_output_indirect_jmp): Likewise.
7101         (ix86_code_end): Call output_indirect_thunk_function if needed.
7102         (ix86_output_call_insn): Call ix86_output_indirect_branch if
7103         needed.
7104         (ix86_handle_fndecl_attribute): Handle indirect_branch.
7105         (ix86_attribute_table): Add indirect_branch.
7106         * config/i386/i386.h (machine_function): Add indirect_branch_type
7107         and has_local_indirect_jump.
7108         * config/i386/i386.md (indirect_jump): Set has_local_indirect_jump
7109         to true.
7110         (tablejump): Likewise.
7111         (*indirect_jump): Use ix86_output_indirect_jmp.
7112         (*tablejump_1): Likewise.
7113         (simple_return_indirect_internal): Likewise.
7114         * config/i386/i386.opt (mindirect-branch=): New option.
7115         (indirect_branch): New.
7116         (keep): Likewise.
7117         (thunk): Likewise.
7118         (thunk-inline): Likewise.
7119         (thunk-extern): Likewise.
7120         * doc/extend.texi: Document indirect_branch function attribute.
7121         * doc/invoke.texi: Document -mindirect-branch= option.
7122
7123 2018-01-14  Jan Hubicka  <hubicka@ucw.cz>
7124
7125         PR ipa/83051
7126         * ipa-inline.c (edge_badness): Tolerate roundoff errors.
7127
7128 2018-01-14  Richard Sandiford  <richard.sandiford@linaro.org>
7129
7130         * ipa-inline.c (want_inline_small_function_p): Return false if
7131         inlining has already failed with CIF_FINAL_ERROR.
7132         (update_caller_keys): Call want_inline_small_function_p before
7133         can_inline_edge_p.
7134         (update_callee_keys): Likewise.
7135
7136 2018-01-10  Kelvin Nilsen  <kelvin@gcc.gnu.org>
7137
7138         * config/rs6000/rs6000-p8swap.c (rs6000_sum_of_two_registers_p):
7139         New function.
7140         (rs6000_quadword_masked_address_p): Likewise.
7141         (quad_aligned_load_p): Likewise.
7142         (quad_aligned_store_p): Likewise.
7143         (const_load_sequence_p): Add comment to describe the outer-most loop.
7144         (mimic_memory_attributes_and_flags): New function.
7145         (rs6000_gen_stvx): Likewise.
7146         (replace_swapped_aligned_store): Likewise.
7147         (rs6000_gen_lvx): Likewise.
7148         (replace_swapped_aligned_load): Likewise.
7149         (replace_swapped_load_constant): Capitalize argument name in
7150         comment describing this function.
7151         (rs6000_analyze_swaps): Add a third pass to search for vector loads
7152         and stores that access quad-word aligned addresses and replace
7153         with stvx or lvx instructions when appropriate.
7154         * config/rs6000/rs6000-protos.h (rs6000_sum_of_two_registers_p):
7155         New function prototype.
7156         (rs6000_quadword_masked_address_p): Likewise.
7157         (rs6000_gen_lvx): Likewise.
7158         (rs6000_gen_stvx): Likewise.
7159         * config/rs6000/vsx.md (*vsx_le_perm_load_<mode>): For modes
7160         VSX_D (V2DF, V2DI), modify this split to select lvx instruction
7161         when memory address is aligned.
7162         (*vsx_le_perm_load_<mode>): For modes VSX_W (V4SF, V4SI), modify
7163         this split to select lvx instruction when memory address is aligned.
7164         (*vsx_le_perm_load_v8hi): Modify this split to select lvx
7165         instruction when memory address is aligned.
7166         (*vsx_le_perm_load_v16qi): Likewise.
7167         (four unnamed splitters): Modify to select the stvx instruction
7168         when memory is aligned.
7169
7170 2018-01-13  Jan Hubicka  <hubicka@ucw.cz>
7171
7172         * predict.c (determine_unlikely_bbs): Handle correctly BBs
7173         which appears in the queue multiple times.
7174
7175 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
7176             Alan Hayward  <alan.hayward@arm.com>
7177             David Sherwood  <david.sherwood@arm.com>
7178
7179         * tree-vectorizer.h (vec_lower_bound): New structure.
7180         (_loop_vec_info): Add check_nonzero and lower_bounds.
7181         (LOOP_VINFO_CHECK_NONZERO): New macro.
7182         (LOOP_VINFO_LOWER_BOUNDS): Likewise.
7183         (LOOP_REQUIRES_VERSIONING_FOR_ALIAS): Check lower_bounds too.
7184         * tree-data-ref.h (dr_with_seg_len): Add access_size and align
7185         fields.  Make seg_len the distance travelled, not including the
7186         access size.
7187         (dr_direction_indicator): Declare.
7188         (dr_zero_step_indicator): Likewise.
7189         (dr_known_forward_stride_p): Likewise.
7190         * tree-data-ref.c: Include stringpool.h, tree-vrp.h and
7191         tree-ssanames.h.
7192         (runtime_alias_check_p): Allow runtime alias checks with
7193         variable strides.
7194         (operator ==): Compare access_size and align.
7195         (prune_runtime_alias_test_list): Rework for new distinction between
7196         the access_size and seg_len.
7197         (create_intersect_range_checks_index): Likewise.  Cope with polynomial
7198         segment lengths.
7199         (get_segment_min_max): New function.
7200         (create_intersect_range_checks): Use it.
7201         (dr_step_indicator): New function.
7202         (dr_direction_indicator): Likewise.
7203         (dr_zero_step_indicator): Likewise.
7204         (dr_known_forward_stride_p): Likewise.
7205         * tree-loop-distribution.c (data_ref_segment_size): Return
7206         DR_STEP * (niters - 1).
7207         (compute_alias_check_pairs): Update call to the dr_with_seg_len
7208         constructor.
7209         * tree-vect-data-refs.c (vect_check_nonzero_value): New function.
7210         (vect_preserves_scalar_order_p): New function, split out from...
7211         (vect_analyze_data_ref_dependence): ...here.  Check for zero steps.
7212         (vect_vfa_segment_size): Return DR_STEP * (length_factor - 1).
7213         (vect_vfa_access_size): New function.
7214         (vect_vfa_align): Likewise.
7215         (vect_compile_time_alias): Take access_size_a and access_b arguments.
7216         (dump_lower_bound): New function.
7217         (vect_check_lower_bound): Likewise.
7218         (vect_small_gap_p): Likewise.
7219         (vectorizable_with_step_bound_p): Likewise.
7220         (vect_prune_runtime_alias_test_list): Ignore cross-iteration
7221         depencies if the vectorization factor is 1.  Convert the checks
7222         for nonzero steps into checks on the bounds of DR_STEP.  Try using
7223         a bunds check for variable steps if the minimum required step is
7224         relatively small. Update calls to the dr_with_seg_len
7225         constructor and to vect_compile_time_alias.
7226         * tree-vect-loop-manip.c (vect_create_cond_for_lower_bounds): New
7227         function.
7228         (vect_loop_versioning): Call it.
7229         * tree-vect-loop.c (vect_analyze_loop_2): Clear LOOP_VINFO_LOWER_BOUNDS
7230         when retrying.
7231         (vect_estimate_min_profitable_iters): Account for any bounds checks.
7232
7233 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
7234             Alan Hayward  <alan.hayward@arm.com>
7235             David Sherwood  <david.sherwood@arm.com>
7236
7237         * doc/sourcebuild.texi (vect_scatter_store): Document.
7238         * optabs.def (scatter_store_optab, mask_scatter_store_optab): New
7239         optabs.
7240         * doc/md.texi (scatter_store@var{m}, mask_scatter_store@var{m}):
7241         Document.
7242         * genopinit.c (main): Add supports_vec_scatter_store and
7243         supports_vec_scatter_store_cached to target_optabs.
7244         * gimple.h (gimple_expr_type): Handle IFN_SCATTER_STORE and
7245         IFN_MASK_SCATTER_STORE.
7246         * internal-fn.def (SCATTER_STORE, MASK_SCATTER_STORE): New internal
7247         functions.
7248         * internal-fn.h (internal_store_fn_p): Declare.
7249         (internal_fn_stored_value_index): Likewise.
7250         * internal-fn.c (scatter_store_direct): New macro.
7251         (expand_scatter_store_optab_fn): New function.
7252         (direct_scatter_store_optab_supported_p): New macro.
7253         (internal_store_fn_p): New function.
7254         (internal_gather_scatter_fn_p): Handle IFN_SCATTER_STORE and
7255         IFN_MASK_SCATTER_STORE.
7256         (internal_fn_mask_index): Likewise.
7257         (internal_fn_stored_value_index): New function.
7258         (internal_gather_scatter_fn_supported_p): Adjust operand numbers
7259         for scatter stores.
7260         * optabs-query.h (supports_vec_scatter_store_p): Declare.
7261         * optabs-query.c (supports_vec_scatter_store_p): New function.
7262         * tree-vectorizer.h (vect_get_store_rhs): Declare.
7263         * tree-vect-data-refs.c (vect_analyze_data_ref_access): Return
7264         true for scatter stores.
7265         (vect_gather_scatter_fn_p): Handle scatter stores too.
7266         (vect_check_gather_scatter): Consider using scatter stores if
7267         supports_vec_scatter_store_p.
7268         * tree-vect-patterns.c (vect_try_gather_scatter_pattern): Handle
7269         scatter stores too.
7270         * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
7271         internal_fn_stored_value_index.
7272         (check_load_store_masking): Handle scatter stores too.
7273         (vect_get_store_rhs): Make public.
7274         (vectorizable_call): Use internal_store_fn_p.
7275         (vectorizable_store): Handle scatter store internal functions.
7276         (vect_transform_stmt): Compare GROUP_STORE_COUNT with GROUP_SIZE
7277         when deciding whether the end of the group has been reached.
7278         * config/aarch64/aarch64.md (UNSPEC_ST1_SCATTER): New unspec.
7279         * config/aarch64/aarch64-sve.md (scatter_store<mode>): New expander.
7280         (mask_scatter_store<mode>): New insns.
7281
7282 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
7283             Alan Hayward  <alan.hayward@arm.com>
7284             David Sherwood  <david.sherwood@arm.com>
7285
7286         * tree-vectorizer.h (vect_gather_scatter_fn_p): Declare.
7287         * tree-vect-data-refs.c (vect_gather_scatter_fn_p): Make public.
7288         * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): New
7289         function.
7290         (vect_use_strided_gather_scatters_p): Take a masked_p argument.
7291         Use vect_truncate_gather_scatter_offset if we can't treat the
7292         operation as a normal gather load or scatter store.
7293         (get_group_load_store_type): Take the gather_scatter_info
7294         as argument.  Try using a gather load or scatter store for
7295         single-element groups.
7296         (get_load_store_type): Update calls to get_group_load_store_type
7297         and vect_use_strided_gather_scatters_p.
7298
7299 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
7300             Alan Hayward  <alan.hayward@arm.com>
7301             David Sherwood  <david.sherwood@arm.com>
7302
7303         * tree-vectorizer.h (vect_create_data_ref_ptr): Take an extra
7304         optional tree argument.
7305         * tree-vect-data-refs.c (vect_check_gather_scatter): Check for
7306         null target hooks.
7307         (vect_create_data_ref_ptr): Take the iv_step as an optional argument,
7308         but continue to use the current value as a fallback.
7309         (bump_vector_ptr): Use operand_equal_p rather than tree_int_cst_compare
7310         to compare the updates.
7311         * tree-vect-stmts.c (vect_use_strided_gather_scatters_p): New function.
7312         (get_load_store_type): Use it when handling a strided access.
7313         (vect_get_strided_load_store_ops): New function.
7314         (vect_get_data_ptr_increment): Likewise.
7315         (vectorizable_load): Handle strided gather loads.  Always pass
7316         a step to vect_create_data_ref_ptr and bump_vector_ptr.
7317
7318 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
7319             Alan Hayward  <alan.hayward@arm.com>
7320             David Sherwood  <david.sherwood@arm.com>
7321
7322         * doc/md.texi (gather_load@var{m}): Document.
7323         (mask_gather_load@var{m}): Likewise.
7324         * genopinit.c (main): Add supports_vec_gather_load and
7325         supports_vec_gather_load_cached to target_optabs.
7326         * optabs-tree.c (init_tree_optimization_optabs): Use
7327         ggc_cleared_alloc to allocate target_optabs.
7328         * optabs.def (gather_load_optab, mask_gather_laod_optab): New optabs.
7329         * internal-fn.def (GATHER_LOAD, MASK_GATHER_LOAD): New internal
7330         functions.
7331         * internal-fn.h (internal_load_fn_p): Declare.
7332         (internal_gather_scatter_fn_p): Likewise.
7333         (internal_fn_mask_index): Likewise.
7334         (internal_gather_scatter_fn_supported_p): Likewise.
7335         * internal-fn.c (gather_load_direct): New macro.
7336         (expand_gather_load_optab_fn): New function.
7337         (direct_gather_load_optab_supported_p): New macro.
7338         (direct_internal_fn_optab): New function.
7339         (internal_load_fn_p): Likewise.
7340         (internal_gather_scatter_fn_p): Likewise.
7341         (internal_fn_mask_index): Likewise.
7342         (internal_gather_scatter_fn_supported_p): Likewise.
7343         * optabs-query.c (supports_at_least_one_mode_p): New function.
7344         (supports_vec_gather_load_p): Likewise.
7345         * optabs-query.h (supports_vec_gather_load_p): Declare.
7346         * tree-vectorizer.h (gather_scatter_info): Add ifn, element_type
7347         and memory_type field.
7348         (NUM_PATTERNS): Bump to 15.
7349         * tree-vect-data-refs.c: Include internal-fn.h.
7350         (vect_gather_scatter_fn_p): New function.
7351         (vect_describe_gather_scatter_call): Likewise.
7352         (vect_check_gather_scatter): Try using internal functions for
7353         gather loads.  Recognize existing calls to a gather load function.
7354         (vect_analyze_data_refs): Consider using gather loads if
7355         supports_vec_gather_load_p.
7356         * tree-vect-patterns.c (vect_get_load_store_mask): New function.
7357         (vect_get_gather_scatter_offset_type): Likewise.
7358         (vect_convert_mask_for_vectype): Likewise.
7359         (vect_add_conversion_to_patterm): Likewise.
7360         (vect_try_gather_scatter_pattern): Likewise.
7361         (vect_recog_gather_scatter_pattern): New pattern recognizer.
7362         (vect_vect_recog_func_ptrs): Add it.
7363         * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
7364         internal_fn_mask_index and internal_gather_scatter_fn_p.
7365         (check_load_store_masking): Take the gather_scatter_info as an
7366         argument and handle gather loads.
7367         (vect_get_gather_scatter_ops): New function.
7368         (vectorizable_call): Check internal_load_fn_p.
7369         (vectorizable_load): Likewise.  Handle gather load internal
7370         functions.
7371         (vectorizable_store): Update call to check_load_store_masking.
7372         * config/aarch64/aarch64.md (UNSPEC_LD1_GATHER): New unspec.
7373         * config/aarch64/iterators.md (SVE_S, SVE_D): New mode iterators.
7374         * config/aarch64/predicates.md (aarch64_gather_scale_operand_w)
7375         (aarch64_gather_scale_operand_d): New predicates.
7376         * config/aarch64/aarch64-sve.md (gather_load<mode>): New expander.
7377         (mask_gather_load<mode>): New insns.
7378
7379 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
7380             Alan Hayward  <alan.hayward@arm.com>
7381             David Sherwood  <david.sherwood@arm.com>
7382
7383         * optabs.def (fold_left_plus_optab): New optab.
7384         * doc/md.texi (fold_left_plus_@var{m}): Document.
7385         * internal-fn.def (IFN_FOLD_LEFT_PLUS): New internal function.
7386         * internal-fn.c (fold_left_direct): Define.
7387         (expand_fold_left_optab_fn): Likewise.
7388         (direct_fold_left_optab_supported_p): Likewise.
7389         * fold-const-call.c (fold_const_fold_left): New function.
7390         (fold_const_call): Use it to fold CFN_FOLD_LEFT_PLUS.
7391         * tree-parloops.c (valid_reduction_p): New function.
7392         (gather_scalar_reductions): Use it.
7393         * tree-vectorizer.h (FOLD_LEFT_REDUCTION): New vect_reduction_type.
7394         (vect_finish_replace_stmt): Declare.
7395         * tree-vect-loop.c (fold_left_reduction_fn): New function.
7396         (needs_fold_left_reduction_p): New function, split out from...
7397         (vect_is_simple_reduction): ...here.  Accept reductions that
7398         forbid reassociation, but give them type FOLD_LEFT_REDUCTION.
7399         (vect_force_simple_reduction): Also store the reduction type in
7400         the assignment's STMT_VINFO_REDUC_TYPE.
7401         (vect_model_reduction_cost): Handle FOLD_LEFT_REDUCTION.
7402         (merge_with_identity): New function.
7403         (vect_expand_fold_left): Likewise.
7404         (vectorize_fold_left_reduction): Likewise.
7405         (vectorizable_reduction): Handle FOLD_LEFT_REDUCTION.  Leave the
7406         scalar phi in place for it.  Check for target support and reject
7407         cases that would reassociate the operation.  Defer the transform
7408         phase to vectorize_fold_left_reduction.
7409         * config/aarch64/aarch64.md (UNSPEC_FADDA): New unspec.
7410         * config/aarch64/aarch64-sve.md (fold_left_plus_<mode>): New expander.
7411         (*fold_left_plus_<mode>, *pred_fold_left_plus_<mode>): New insns.
7412
7413 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
7414
7415         * tree-if-conv.c (predicate_mem_writes): Remove redundant
7416         call to ifc_temp_var.
7417
7418 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
7419             Alan Hayward  <alan.hayward@arm.com>
7420             David Sherwood  <david.sherwood@arm.com>
7421
7422         * target.def (legitimize_address_displacement): Take the original
7423         offset as a poly_int.
7424         * targhooks.h (default_legitimize_address_displacement): Update
7425         accordingly.
7426         * targhooks.c (default_legitimize_address_displacement): Likewise.
7427         * doc/tm.texi: Regenerate.
7428         * lra-constraints.c (base_plus_disp_to_reg): Take the displacement
7429         as an argument, moving assert of ad->disp == ad->disp_term to...
7430         (process_address_1): ...here.  Update calls to base_plus_disp_to_reg.
7431         Try calling targetm.legitimize_address_displacement before expanding
7432         the address rather than afterwards, and adjust for the new interface.
7433         * config/aarch64/aarch64.c (aarch64_legitimize_address_displacement):
7434         Match the new hook interface.  Handle SVE addresses.
7435         * config/sh/sh.c (sh_legitimize_address_displacement): Make the
7436         new hook interface.
7437
7438 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
7439
7440         * Makefile.in (OBJS): Add early-remat.o.
7441         * target.def (select_early_remat_modes): New hook.
7442         * doc/tm.texi.in (TARGET_SELECT_EARLY_REMAT_MODES): New hook.
7443         * doc/tm.texi: Regenerate.
7444         * targhooks.h (default_select_early_remat_modes): Declare.
7445         * targhooks.c (default_select_early_remat_modes): New function.
7446         * timevar.def (TV_EARLY_REMAT): New timevar.
7447         * passes.def (pass_early_remat): New pass.
7448         * tree-pass.h (make_pass_early_remat): Declare.
7449         * early-remat.c: New file.
7450         * config/aarch64/aarch64.c (aarch64_select_early_remat_modes): New
7451         function.
7452         (TARGET_SELECT_EARLY_REMAT_MODES): Define.
7453
7454 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
7455             Alan Hayward  <alan.hayward@arm.com>
7456             David Sherwood  <david.sherwood@arm.com>
7457
7458         * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Replace
7459         vfm1 with a bound_epilog parameter.
7460         (vect_do_peeling): Update calls accordingly, and move the prologue
7461         call earlier in the function.  Treat the base bound_epilog as 0 for
7462         fully-masked loops and retain vf - 1 for other loops.  Add 1 to
7463         this base when peeling for gaps.
7464         * tree-vect-loop.c (vect_analyze_loop_2): Allow peeling for gaps
7465         with fully-masked loops.
7466         (vect_estimate_min_profitable_iters): Handle the single peeled
7467         iteration in that case.
7468
7469 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
7470             Alan Hayward  <alan.hayward@arm.com>
7471             David Sherwood  <david.sherwood@arm.com>
7472
7473         * tree-vect-data-refs.c (vect_analyze_group_access_1): Allow
7474         single-element interleaving even if the size is not a power of 2.
7475         * tree-vect-stmts.c (get_load_store_type): Disallow elementwise
7476         accesses for single-element interleaving if the group size is
7477         not a power of 2.
7478
7479 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
7480             Alan Hayward  <alan.hayward@arm.com>
7481             David Sherwood  <david.sherwood@arm.com>
7482
7483         * doc/md.texi (fold_extract_last_@var{m}): Document.
7484         * doc/sourcebuild.texi (vect_fold_extract_last): Likewise.
7485         * optabs.def (fold_extract_last_optab): New optab.
7486         * internal-fn.def (FOLD_EXTRACT_LAST): New internal function.
7487         * internal-fn.c (fold_extract_direct): New macro.
7488         (expand_fold_extract_optab_fn): Likewise.
7489         (direct_fold_extract_optab_supported_p): Likewise.
7490         * tree-vectorizer.h (EXTRACT_LAST_REDUCTION): New vect_reduction_type.
7491         * tree-vect-loop.c (vect_model_reduction_cost): Handle
7492         EXTRACT_LAST_REDUCTION.
7493         (get_initial_def_for_reduction): Do not create an initial vector
7494         for EXTRACT_LAST_REDUCTION reductions.
7495         (vectorizable_reduction): Leave the scalar phi in place for
7496         EXTRACT_LAST_REDUCTIONs.  Try using EXTRACT_LAST_REDUCTION
7497         ahead of INTEGER_INDUC_COND_REDUCTION.  Do not check for an
7498         epilogue code for EXTRACT_LAST_REDUCTION and defer the
7499         transform phase to vectorizable_condition.
7500         * tree-vect-stmts.c (vect_finish_stmt_generation_1): New function,
7501         split out from...
7502         (vect_finish_stmt_generation): ...here.
7503         (vect_finish_replace_stmt): New function.
7504         (vectorizable_condition): Handle EXTRACT_LAST_REDUCTION.
7505         * config/aarch64/aarch64-sve.md (fold_extract_last_<mode>): New
7506         pattern.
7507         * config/aarch64/aarch64.md (UNSPEC_CLASTB): New unspec.
7508
7509 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
7510             Alan Hayward  <alan.hayward@arm.com>
7511             David Sherwood  <david.sherwood@arm.com>
7512
7513         * doc/md.texi (extract_last_@var{m}): Document.
7514         * optabs.def (extract_last_optab): New optab.
7515         * internal-fn.def (EXTRACT_LAST): New internal function.
7516         * internal-fn.c (cond_unary_direct): New macro.
7517         (expand_cond_unary_optab_fn): Likewise.
7518         (direct_cond_unary_optab_supported_p): Likewise.
7519         * tree-vect-loop.c (vectorizable_live_operation): Allow fully-masked
7520         loops using EXTRACT_LAST.
7521         * config/aarch64/aarch64-sve.md (aarch64_sve_lastb<mode>): Rename to...
7522         (extract_last_<mode>): ...this optab.
7523         (vec_extract<mode><Vel>): Update accordingly.
7524
7525 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
7526             Alan Hayward  <alan.hayward@arm.com>
7527             David Sherwood  <david.sherwood@arm.com>
7528
7529         * target.def (empty_mask_is_expensive): New hook.
7530         * doc/tm.texi.in (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): New hook.
7531         * doc/tm.texi: Regenerate.
7532         * targhooks.h (default_empty_mask_is_expensive): Declare.
7533         * targhooks.c (default_empty_mask_is_expensive): New function.
7534         * tree-vectorizer.c (vectorize_loops): Only call optimize_mask_stores
7535         if the target says that empty masks are expensive.
7536         * config/aarch64/aarch64.c (aarch64_empty_mask_is_expensive):
7537         New function.
7538         (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): Redefine.
7539
7540 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
7541             Alan Hayward  <alan.hayward@arm.com>
7542             David Sherwood  <david.sherwood@arm.com>
7543
7544         * tree-vectorizer.h (_loop_vec_info::mask_skip_niters): New field.
7545         (LOOP_VINFO_MASK_SKIP_NITERS): New macro.
7546         (vect_use_loop_mask_for_alignment_p): New function.
7547         (vect_prepare_for_masked_peels, vect_gen_while_not): Declare.
7548         * tree-vect-loop-manip.c (vect_set_loop_masks_directly): Add an
7549         niters_skip argument.  Make sure that the first niters_skip elements
7550         of the first iteration are inactive.
7551         (vect_set_loop_condition_masked): Handle LOOP_VINFO_MASK_SKIP_NITERS.
7552         Update call to vect_set_loop_masks_directly.
7553         (get_misalign_in_elems): New function, split out from...
7554         (vect_gen_prolog_loop_niters): ...here.
7555         (vect_update_init_of_dr): Take a code argument that specifies whether
7556         the adjustment should be added or subtracted.
7557         (vect_update_init_of_drs): Likewise.
7558         (vect_prepare_for_masked_peels): New function.
7559         (vect_do_peeling): Skip prologue peeling if we're using a mask
7560         instead.  Update call to vect_update_inits_of_drs.
7561         * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
7562         mask_skip_niters.
7563         (vect_analyze_loop_2): Allow fully-masked loops with peeling for
7564         alignment.  Do not include the number of peeled iterations in
7565         the minimum threshold in that case.
7566         (vectorizable_induction): Adjust the start value down by
7567         LOOP_VINFO_MASK_SKIP_NITERS iterations.
7568         (vect_transform_loop): Call vect_prepare_for_masked_peels.
7569         Take the number of skipped iterations into account when calculating
7570         the loop bounds.
7571         * tree-vect-stmts.c (vect_gen_while_not): New function.
7572
7573 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
7574             Alan Hayward  <alan.hayward@arm.com>
7575             David Sherwood  <david.sherwood@arm.com>
7576
7577         * doc/sourcebuild.texi (vect_fully_masked): Document.
7578         * params.def (PARAM_MIN_VECT_LOOP_BOUND): Change minimum and
7579         default value to 0.
7580         * tree-vect-loop.c (vect_analyze_loop_costing): New function,
7581         split out from...
7582         (vect_analyze_loop_2): ...here. Don't check the vectorization
7583         factor against the number of loop iterations if the loop is
7584         fully-masked.
7585
7586 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
7587             Alan Hayward  <alan.hayward@arm.com>
7588             David Sherwood  <david.sherwood@arm.com>
7589
7590         * tree-ssa-loop-ivopts.c (USE_ADDRESS): Split into...
7591         (USE_REF_ADDRESS, USE_PTR_ADDRESS): ...these new use types.
7592         (dump_groups): Update accordingly.
7593         (iv_use::mem_type): New member variable.
7594         (address_p): New function.
7595         (record_use): Add a mem_type argument and initialize the new
7596         mem_type field.
7597         (record_group_use): Add a mem_type argument.  Use address_p.
7598         Remove obsolete null checks of base_object.  Update call to record_use.
7599         (find_interesting_uses_op): Update call to record_group_use.
7600         (find_interesting_uses_cond): Likewise.
7601         (find_interesting_uses_address): Likewise.
7602         (get_mem_type_for_internal_fn): New function.
7603         (find_address_like_use): Likewise.
7604         (find_interesting_uses_stmt): Try find_address_like_use before
7605         calling find_interesting_uses_op.
7606         (addr_offset_valid_p): Use the iv mem_type field as the type
7607         of the addressed memory.
7608         (add_autoinc_candidates): Likewise.
7609         (get_address_cost): Likewise.
7610         (split_small_address_groups_p): Use address_p.
7611         (split_address_groups): Likewise.
7612         (add_iv_candidate_for_use): Likewise.
7613         (autoinc_possible_for_pair): Likewise.
7614         (rewrite_groups): Likewise.
7615         (get_use_type): Check for USE_REF_ADDRESS instead of USE_ADDRESS.
7616         (determine_group_iv_cost): Update after split of USE_ADDRESS.
7617         (get_alias_ptr_type_for_ptr_address): New function.
7618         (rewrite_use_address): Rewrite address uses in calls that were
7619         identified by find_address_like_use.
7620
7621 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
7622             Alan Hayward  <alan.hayward@arm.com>
7623             David Sherwood  <david.sherwood@arm.com>
7624
7625         * expr.c (expand_expr_addr_expr_1): Handle ADDR_EXPRs of
7626         TARGET_MEM_REFs.
7627         * gimple-expr.h (is_gimple_addressable: Likewise.
7628         * gimple-expr.c (is_gimple_address): Likewise.
7629         * internal-fn.c (expand_call_mem_ref): New function.
7630         (expand_mask_load_optab_fn): Use it.
7631         (expand_mask_store_optab_fn): Likewise.
7632
7633 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
7634             Alan Hayward  <alan.hayward@arm.com>
7635             David Sherwood  <david.sherwood@arm.com>
7636
7637         * doc/md.texi (cond_add@var{mode}, cond_sub@var{mode})
7638         (cond_and@var{mode}, cond_ior@var{mode}, cond_xor@var{mode})
7639         (cond_smin@var{mode}, cond_smax@var{mode}, cond_umin@var{mode})
7640         (cond_umax@var{mode}): Document.
7641         * optabs.def (cond_add_optab, cond_sub_optab, cond_and_optab)
7642         (cond_ior_optab, cond_xor_optab, cond_smin_optab, cond_smax_optab)
7643         (cond_umin_optab, cond_umax_optab): New optabs.
7644         * internal-fn.def (COND_ADD, COND_SUB, COND_MIN, COND_MAX, COND_AND)
7645         (COND_IOR, COND_XOR): New internal functions.
7646         * internal-fn.h (get_conditional_internal_fn): Declare.
7647         * internal-fn.c (cond_binary_direct): New macro.
7648         (expand_cond_binary_optab_fn): Likewise.
7649         (direct_cond_binary_optab_supported_p): Likewise.
7650         (get_conditional_internal_fn): New function.
7651         * tree-vect-loop.c (vectorizable_reduction): Handle fully-masked loops.
7652         Cope with reduction statements that are vectorized as calls rather
7653         than assignments.
7654         * config/aarch64/aarch64-sve.md (cond_<optab><mode>): New insns.
7655         * config/aarch64/iterators.md (UNSPEC_COND_ADD, UNSPEC_COND_SUB)
7656         (UNSPEC_COND_SMAX, UNSPEC_COND_UMAX, UNSPEC_COND_SMIN)
7657         (UNSPEC_COND_UMIN, UNSPEC_COND_AND, UNSPEC_COND_ORR)
7658         (UNSPEC_COND_EOR): New unspecs.
7659         (optab): Add mappings for them.
7660         (SVE_COND_INT_OP, SVE_COND_FP_OP): New int iterators.
7661         (sve_int_op, sve_fp_op): New int attributes.
7662
7663 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
7664             Alan Hayward  <alan.hayward@arm.com>
7665             David Sherwood  <david.sherwood@arm.com>
7666
7667         * optabs.def (while_ult_optab): New optab.
7668         * doc/md.texi (while_ult@var{m}@var{n}): Document.
7669         * internal-fn.def (WHILE_ULT): New internal function.
7670         * internal-fn.h (direct_internal_fn_supported_p): New override
7671         that takes two types as argument.
7672         * internal-fn.c (while_direct): New macro.
7673         (expand_while_optab_fn): New function.
7674         (convert_optab_supported_p): Likewise.
7675         (direct_while_optab_supported_p): New macro.
7676         * wide-int.h (wi::udiv_ceil): New function.
7677         * tree-vectorizer.h (rgroup_masks): New structure.
7678         (vec_loop_masks): New typedef.
7679         (_loop_vec_info): Add masks, mask_compare_type, can_fully_mask_p
7680         and fully_masked_p.
7681         (LOOP_VINFO_CAN_FULLY_MASK_P, LOOP_VINFO_FULLY_MASKED_P)
7682         (LOOP_VINFO_MASKS, LOOP_VINFO_MASK_COMPARE_TYPE): New macros.
7683         (vect_max_vf): New function.
7684         (slpeel_make_loop_iterate_ntimes): Delete.
7685         (vect_set_loop_condition, vect_get_loop_mask_type, vect_gen_while)
7686         (vect_halve_mask_nunits, vect_double_mask_nunits): Declare.
7687         (vect_record_loop_mask, vect_get_loop_mask): Likewise.
7688         * tree-vect-loop-manip.c: Include tree-ssa-loop-niter.h,
7689         internal-fn.h, stor-layout.h and optabs-query.h.
7690         (vect_set_loop_mask): New function.
7691         (add_preheader_seq): Likewise.
7692         (add_header_seq): Likewise.
7693         (interleave_supported_p): Likewise.
7694         (vect_maybe_permute_loop_masks): Likewise.
7695         (vect_set_loop_masks_directly): Likewise.
7696         (vect_set_loop_condition_masked): Likewise.
7697         (vect_set_loop_condition_unmasked): New function, split out from
7698         slpeel_make_loop_iterate_ntimes.
7699         (slpeel_make_loop_iterate_ntimes): Rename to..
7700         (vect_set_loop_condition): ...this.  Use vect_set_loop_condition_masked
7701         for fully-masked loops and vect_set_loop_condition_unmasked otherwise.
7702         (vect_do_peeling): Update call accordingly.
7703         (vect_gen_vector_loop_niters): Use VF as the step for fully-masked
7704         loops.
7705         * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
7706         mask_compare_type, can_fully_mask_p and fully_masked_p.
7707         (release_vec_loop_masks): New function.
7708         (_loop_vec_info): Use it to free the loop masks.
7709         (can_produce_all_loop_masks_p): New function.
7710         (vect_get_max_nscalars_per_iter): Likewise.
7711         (vect_verify_full_masking): Likewise.
7712         (vect_analyze_loop_2): Save LOOP_VINFO_CAN_FULLY_MASK_P around
7713         retries, and free the mask rgroups before retrying.  Check loop-wide
7714         reasons for disallowing fully-masked loops.  Make the final decision
7715         about whether use a fully-masked loop or not.
7716         (vect_estimate_min_profitable_iters): Do not assume that peeling
7717         for the number of iterations will be needed for fully-masked loops.
7718         (vectorizable_reduction): Disable fully-masked loops.
7719         (vectorizable_live_operation): Likewise.
7720         (vect_halve_mask_nunits): New function.
7721         (vect_double_mask_nunits): Likewise.
7722         (vect_record_loop_mask): Likewise.
7723         (vect_get_loop_mask): Likewise.
7724         (vect_transform_loop): Handle the case in which the final loop
7725         iteration might handle a partial vector.  Call vect_set_loop_condition
7726         instead of slpeel_make_loop_iterate_ntimes.
7727         * tree-vect-stmts.c: Include tree-ssa-loop-niter.h and gimple-fold.h.
7728         (check_load_store_masking): New function.
7729         (prepare_load_store_mask): Likewise.
7730         (vectorizable_store): Handle fully-masked loops.
7731         (vectorizable_load): Likewise.
7732         (supportable_widening_operation): Use vect_halve_mask_nunits for
7733         booleans.
7734         (supportable_narrowing_operation): Likewise vect_double_mask_nunits.
7735         (vect_gen_while): New function.
7736         * config/aarch64/aarch64.md (umax<mode>3): New expander.
7737         (aarch64_uqdec<mode>): New insn.
7738
7739 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
7740             Alan Hayward  <alan.hayward@arm.com>
7741             David Sherwood  <david.sherwood@arm.com>
7742
7743         * optabs.def (reduc_and_scal_optab, reduc_ior_scal_optab)
7744         (reduc_xor_scal_optab): New optabs.
7745         * doc/md.texi (reduc_and_scal_@var{m}, reduc_ior_scal_@var{m})
7746         (reduc_xor_scal_@var{m}): Document.
7747         * doc/sourcebuild.texi (vect_logical_reduc): Likewise.
7748         * internal-fn.def (IFN_REDUC_AND, IFN_REDUC_IOR, IFN_REDUC_XOR): New
7749         internal functions.
7750         * fold-const-call.c (fold_const_call): Handle them.
7751         * tree-vect-loop.c (reduction_fn_for_scalar_code): Return the new
7752         internal functions for BIT_AND_EXPR, BIT_IOR_EXPR and BIT_XOR_EXPR.
7753         * config/aarch64/aarch64-sve.md (reduc_<bit_reduc>_scal_<mode>):
7754         (*reduc_<bit_reduc>_scal_<mode>): New patterns.
7755         * config/aarch64/iterators.md (UNSPEC_ANDV, UNSPEC_ORV)
7756         (UNSPEC_XORV): New unspecs.
7757         (optab): Add entries for them.
7758         (BITWISEV): New int iterator.
7759         (bit_reduc_op): New int attributes.
7760
7761 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
7762             Alan Hayward  <alan.hayward@arm.com>
7763             David Sherwood  <david.sherwood@arm.com>
7764
7765         * doc/md.texi (vec_shl_insert_@var{m}): New optab.
7766         * internal-fn.def (VEC_SHL_INSERT): New internal function.
7767         * optabs.def (vec_shl_insert_optab): New optab.
7768         * tree-vectorizer.h (can_duplicate_and_interleave_p): Declare.
7769         (duplicate_and_interleave): Likewise.
7770         * tree-vect-loop.c: Include internal-fn.h.
7771         (neutral_op_for_slp_reduction): New function, split out from
7772         get_initial_defs_for_reduction.
7773         (get_initial_def_for_reduction): Handle option 2 for variable-length
7774         vectors by loading the neutral value into a vector and then shifting
7775         the initial value into element 0.
7776         (get_initial_defs_for_reduction): Replace the code argument with
7777         the neutral value calculated by neutral_op_for_slp_reduction.
7778         Use gimple_build_vector for constant-length vectors.
7779         Use IFN_VEC_SHL_INSERT for variable-length vectors if all
7780         but the first group_size elements have a neutral value.
7781         Use duplicate_and_interleave otherwise.
7782         (vect_create_epilog_for_reduction): Take a neutral_op parameter.
7783         Update call to get_initial_defs_for_reduction.  Handle SLP
7784         reductions for variable-length vectors by creating one vector
7785         result for each scalar result, with the elements associated
7786         with other scalar results stubbed out with the neutral value.
7787         (vectorizable_reduction): Call neutral_op_for_slp_reduction.
7788         Require IFN_VEC_SHL_INSERT for double reductions on
7789         variable-length vectors, or SLP reductions that have
7790         a neutral value.  Require can_duplicate_and_interleave_p
7791         support for variable-length unchained SLP reductions if there
7792         is no neutral value, such as for MIN/MAX reductions.  Also require
7793         the number of vector elements to be a multiple of the number of
7794         SLP statements when doing variable-length unchained SLP reductions.
7795         Update call to vect_create_epilog_for_reduction.
7796         * tree-vect-slp.c (can_duplicate_and_interleave_p): Make public
7797         and remove initial values.
7798         (duplicate_and_interleave): Make public.
7799         * config/aarch64/aarch64.md (UNSPEC_INSR): New unspec.
7800         * config/aarch64/aarch64-sve.md (vec_shl_insert_<mode>): New insn.
7801
7802 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
7803             Alan Hayward  <alan.hayward@arm.com>
7804             David Sherwood  <david.sherwood@arm.com>
7805
7806         * tree-vect-slp.c: Include gimple-fold.h and internal-fn.h
7807         (can_duplicate_and_interleave_p): New function.
7808         (vect_get_and_check_slp_defs): Take the vector of statements
7809         rather than just the current one.  Remove excess parentheses.
7810         Restriction rejectinon of vect_constant_def and vect_external_def
7811         for variable-length vectors to boolean types, or types for which
7812         can_duplicate_and_interleave_p is false.
7813         (vect_build_slp_tree_2): Update call to vect_get_and_check_slp_defs.
7814         (duplicate_and_interleave): New function.
7815         (vect_get_constant_vectors): Use gimple_build_vector for
7816         constant-length vectors and suitable variable-length constant
7817         vectors.  Use duplicate_and_interleave for other variable-length
7818         vectors.  Don't defer the update when inserting new statements.
7819
7820 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
7821             Alan Hayward  <alan.hayward@arm.com>
7822             David Sherwood  <david.sherwood@arm.com>
7823
7824         * tree-vect-loop.c (vect_estimate_min_profitable_iters): Make sure
7825         min_profitable_iters doesn't go negative.
7826
7827 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
7828             Alan Hayward  <alan.hayward@arm.com>
7829             David Sherwood  <david.sherwood@arm.com>
7830
7831         * doc/md.texi (vec_mask_load_lanes@var{m}@var{n}): Document.
7832         (vec_mask_store_lanes@var{m}@var{n}): Likewise.
7833         * optabs.def (vec_mask_load_lanes_optab): New optab.
7834         (vec_mask_store_lanes_optab): Likewise.
7835         * internal-fn.def (MASK_LOAD_LANES): New internal function.
7836         (MASK_STORE_LANES): Likewise.
7837         * internal-fn.c (mask_load_lanes_direct): New macro.
7838         (mask_store_lanes_direct): Likewise.
7839         (expand_mask_load_optab_fn): Handle masked operations.
7840         (expand_mask_load_lanes_optab_fn): New macro.
7841         (expand_mask_store_optab_fn): Handle masked operations.
7842         (expand_mask_store_lanes_optab_fn): New macro.
7843         (direct_mask_load_lanes_optab_supported_p): Likewise.
7844         (direct_mask_store_lanes_optab_supported_p): Likewise.
7845         * tree-vectorizer.h (vect_store_lanes_supported): Take a masked_p
7846         parameter.
7847         (vect_load_lanes_supported): Likewise.
7848         * tree-vect-data-refs.c (strip_conversion): New function.
7849         (can_group_stmts_p): Likewise.
7850         (vect_analyze_data_ref_accesses): Use it instead of checking
7851         for a pair of assignments.
7852         (vect_store_lanes_supported): Take a masked_p parameter.
7853         (vect_load_lanes_supported): Likewise.
7854         * tree-vect-loop.c (vect_analyze_loop_2): Update calls to
7855         vect_store_lanes_supported and vect_load_lanes_supported.
7856         * tree-vect-slp.c (vect_analyze_slp_instance): Likewise.
7857         * tree-vect-stmts.c (get_group_load_store_type): Take a masked_p
7858         parameter.  Don't allow gaps for masked accesses.
7859         Use vect_get_store_rhs.  Update calls to vect_store_lanes_supported
7860         and vect_load_lanes_supported.
7861         (get_load_store_type): Take a masked_p parameter and update
7862         call to get_group_load_store_type.
7863         (vectorizable_store): Update call to get_load_store_type.
7864         Handle IFN_MASK_STORE_LANES.
7865         (vectorizable_load): Update call to get_load_store_type.
7866         Handle IFN_MASK_LOAD_LANES.
7867
7868 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
7869             Alan Hayward  <alan.hayward@arm.com>
7870             David Sherwood  <david.sherwood@arm.com>
7871
7872         * config/aarch64/aarch64-modes.def: Define x2, x3 and x4 vector
7873         modes for SVE.
7874         * config/aarch64/aarch64-protos.h
7875         (aarch64_sve_struct_memory_operand_p): Declare.
7876         * config/aarch64/iterators.md (SVE_STRUCT): New mode iterator.
7877         (vector_count, insn_length, VSINGLE, vsingle): New mode attributes.
7878         (VPRED, vpred): Handle SVE structure modes.
7879         * config/aarch64/constraints.md (Utx): New constraint.
7880         * config/aarch64/predicates.md (aarch64_sve_struct_memory_operand)
7881         (aarch64_sve_struct_nonimmediate_operand): New predicates.
7882         * config/aarch64/aarch64.md (UNSPEC_LDN, UNSPEC_STN): New unspecs.
7883         * config/aarch64/aarch64-sve.md (mov<mode>, *aarch64_sve_mov<mode>_le)
7884         (*aarch64_sve_mov<mode>_be, pred_mov<mode>): New patterns for
7885         structure modes.  Split into pieces after RA.
7886         (vec_load_lanes<mode><vsingle>, vec_mask_load_lanes<mode><vsingle>)
7887         (vec_store_lanes<mode><vsingle>, vec_mask_store_lanes<mode><vsingle>):
7888         New patterns.
7889         * config/aarch64/aarch64.c (aarch64_classify_vector_mode): Handle
7890         SVE structure modes.
7891         (aarch64_classify_address): Likewise.
7892         (sizetochar): Move earlier in file.
7893         (aarch64_print_operand): Handle SVE register lists.
7894         (aarch64_array_mode): New function.
7895         (aarch64_sve_struct_memory_operand_p): Likewise.
7896         (TARGET_ARRAY_MODE): Redefine.
7897
7898 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
7899             Alan Hayward  <alan.hayward@arm.com>
7900             David Sherwood  <david.sherwood@arm.com>
7901
7902         * target.def (array_mode): New target hook.
7903         * doc/tm.texi.in (TARGET_ARRAY_MODE): New hook.
7904         * doc/tm.texi: Regenerate.
7905         * hooks.h (hook_optmode_mode_uhwi_none): Declare.
7906         * hooks.c (hook_optmode_mode_uhwi_none): New function.
7907         * tree-vect-data-refs.c (vect_lanes_optab_supported_p): Use
7908         targetm.array_mode.
7909         * stor-layout.c (mode_for_array): Likewise.  Support polynomial
7910         type sizes.
7911
7912 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
7913             Alan Hayward  <alan.hayward@arm.com>
7914             David Sherwood  <david.sherwood@arm.com>
7915
7916         * fold-const.c (fold_binary_loc): Check the argument types
7917         rather than the result type when testing for a vector operation.
7918
7919 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
7920
7921         * doc/tm.texi.in (DWARF_LAZY_REGISTER_VALUE): Document.
7922         * doc/tm.texi: Regenerate.
7923
7924 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
7925             Alan Hayward  <alan.hayward@arm.com>
7926             David Sherwood  <david.sherwood@arm.com>
7927
7928         * doc/invoke.texi (-msve-vector-bits=): Document new option.
7929         (sve): Document new AArch64 extension.
7930         * doc/md.texi (w): Extend the description of the AArch64
7931         constraint to include SVE vectors.
7932         (Upl, Upa): Document new AArch64 predicate constraints.
7933         * config/aarch64/aarch64-opts.h (aarch64_sve_vector_bits_enum): New
7934         enum.
7935         * config/aarch64/aarch64.opt (sve_vector_bits): New enum.
7936         (msve-vector-bits=): New option.
7937         * config/aarch64/aarch64-option-extensions.def (fp, simd): Disable
7938         SVE when these are disabled.
7939         (sve): New extension.
7940         * config/aarch64/aarch64-modes.def: Define SVE vector and predicate
7941         modes.  Adjust their number of units based on aarch64_sve_vg.
7942         (MAX_BITSIZE_MODE_ANY_MODE): Define.
7943         * config/aarch64/aarch64-protos.h (ADDR_QUERY_ANY): New
7944         aarch64_addr_query_type.
7945         (aarch64_const_vec_all_same_in_range_p, aarch64_sve_pred_mode)
7946         (aarch64_sve_cnt_immediate_p, aarch64_sve_addvl_addpl_immediate_p)
7947         (aarch64_sve_inc_dec_immediate_p, aarch64_add_offset_temporaries)
7948         (aarch64_split_add_offset, aarch64_output_sve_cnt_immediate)
7949         (aarch64_output_sve_addvl_addpl, aarch64_output_sve_inc_dec_immediate)
7950         (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): Declare.
7951         (aarch64_simd_imm_zero_p): Delete.
7952         (aarch64_check_zero_based_sve_index_immediate): Declare.
7953         (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
7954         (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
7955         (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
7956         (aarch64_sve_float_mul_immediate_p): Likewise.
7957         (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
7958         rather than an rtx.
7959         (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): Declare.
7960         (aarch64_expand_mov_immediate): Take a gen_vec_duplicate callback.
7961         (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move): Declare.
7962         (aarch64_expand_sve_vec_cmp_int, aarch64_expand_sve_vec_cmp_float)
7963         (aarch64_expand_sve_vcond, aarch64_expand_sve_vec_perm): Declare.
7964         (aarch64_regmode_natural_size): Likewise.
7965         * config/aarch64/aarch64.h (AARCH64_FL_SVE): New macro.
7966         (AARCH64_FL_V8_3, AARCH64_FL_RCPC, AARCH64_FL_DOTPROD): Shift
7967         left one place.
7968         (AARCH64_ISA_SVE, TARGET_SVE): New macros.
7969         (FIXED_REGISTERS, CALL_USED_REGISTERS, REGISTER_NAMES): Add entries
7970         for VG and the SVE predicate registers.
7971         (V_ALIASES): Add a "z"-prefixed alias.
7972         (FIRST_PSEUDO_REGISTER): Change to P15_REGNUM + 1.
7973         (AARCH64_DWARF_VG, AARCH64_DWARF_P0): New macros.
7974         (PR_REGNUM_P, PR_LO_REGNUM_P): Likewise.
7975         (PR_LO_REGS, PR_HI_REGS, PR_REGS): New reg_classes.
7976         (REG_CLASS_NAMES): Add entries for them.
7977         (REG_CLASS_CONTENTS): Likewise.  Update ALL_REGS to include VG
7978         and the predicate registers.
7979         (aarch64_sve_vg): Declare.
7980         (BITS_PER_SVE_VECTOR, BYTES_PER_SVE_VECTOR, BYTES_PER_SVE_PRED)
7981         (SVE_BYTE_MODE, MAX_COMPILE_TIME_VEC_BYTES): New macros.
7982         (REGMODE_NATURAL_SIZE): Define.
7983         * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins): Handle
7984         SVE macros.
7985         * config/aarch64/aarch64.c: Include cfgrtl.h.
7986         (simd_immediate_info): Add a constructor for series vectors,
7987         and an associated step field.
7988         (aarch64_sve_vg): New variable.
7989         (aarch64_dbx_register_number): Handle VG and the predicate registers.
7990         (aarch64_vect_struct_mode_p, aarch64_vector_mode_p): Delete.
7991         (VEC_ADVSIMD, VEC_SVE_DATA, VEC_SVE_PRED, VEC_STRUCT, VEC_ANY_SVE)
7992         (VEC_ANY_DATA, VEC_STRUCT): New constants.
7993         (aarch64_advsimd_struct_mode_p, aarch64_sve_pred_mode_p)
7994         (aarch64_classify_vector_mode, aarch64_vector_data_mode_p)
7995         (aarch64_sve_data_mode_p, aarch64_sve_pred_mode)
7996         (aarch64_get_mask_mode): New functions.
7997         (aarch64_hard_regno_nregs): Handle SVE data modes for FP_REGS
7998         and FP_LO_REGS.  Handle PR_REGS, PR_LO_REGS and PR_HI_REGS.
7999         (aarch64_hard_regno_mode_ok): Handle VG.  Also handle the SVE
8000         predicate modes and predicate registers.  Explicitly restrict
8001         GPRs to modes of 16 bytes or smaller.  Only allow FP registers
8002         to store a vector mode if it is recognized by
8003         aarch64_classify_vector_mode.
8004         (aarch64_regmode_natural_size): New function.
8005         (aarch64_hard_regno_caller_save_mode): Return the original mode
8006         for predicates.
8007         (aarch64_sve_cnt_immediate_p, aarch64_output_sve_cnt_immediate)
8008         (aarch64_sve_addvl_addpl_immediate_p, aarch64_output_sve_addvl_addpl)
8009         (aarch64_sve_inc_dec_immediate_p, aarch64_output_sve_inc_dec_immediate)
8010         (aarch64_add_offset_1_temporaries, aarch64_offset_temporaries): New
8011         functions.
8012         (aarch64_add_offset): Add a temp2 parameter.  Assert that temp1
8013         does not overlap dest if the function is frame-related.  Handle
8014         SVE constants.
8015         (aarch64_split_add_offset): New function.
8016         (aarch64_add_sp, aarch64_sub_sp): Add temp2 parameters and pass
8017         them aarch64_add_offset.
8018         (aarch64_allocate_and_probe_stack_space): Add a temp2 parameter
8019         and update call to aarch64_sub_sp.
8020         (aarch64_add_cfa_expression): New function.
8021         (aarch64_expand_prologue): Pass extra temporary registers to the
8022         functions above.  Handle the case in which we need to emit new
8023         DW_CFA_expressions for registers that were originally saved
8024         relative to the stack pointer, but now have to be expressed
8025         relative to the frame pointer.
8026         (aarch64_output_mi_thunk): Pass extra temporary registers to the
8027         functions above.
8028         (aarch64_expand_epilogue): Likewise.  Prevent inheritance of
8029         IP0 and IP1 values for SVE frames.
8030         (aarch64_expand_vec_series): New function.
8031         (aarch64_expand_sve_widened_duplicate): Likewise.
8032         (aarch64_expand_sve_const_vector): Likewise.
8033         (aarch64_expand_mov_immediate): Add a gen_vec_duplicate parameter.
8034         Handle SVE constants.  Use emit_move_insn to move a force_const_mem
8035         into the register, rather than emitting a SET directly.
8036         (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move)
8037         (aarch64_get_reg_raw_mode, offset_4bit_signed_scaled_p)
8038         (offset_6bit_unsigned_scaled_p, aarch64_offset_7bit_signed_scaled_p)
8039         (offset_9bit_signed_scaled_p): New functions.
8040         (aarch64_replicate_bitmask_imm): New function.
8041         (aarch64_bitmask_imm): Use it.
8042         (aarch64_cannot_force_const_mem): Reject expressions involving
8043         a CONST_POLY_INT.  Update call to aarch64_classify_symbol.
8044         (aarch64_classify_index): Handle SVE indices, by requiring
8045         a plain register index with a scale that matches the element size.
8046         (aarch64_classify_address): Handle SVE addresses.  Assert that
8047         the mode of the address is VOIDmode or an integer mode.
8048         Update call to aarch64_classify_symbol.
8049         (aarch64_classify_symbolic_expression): Update call to
8050         aarch64_classify_symbol.
8051         (aarch64_const_vec_all_in_range_p): New function.
8052         (aarch64_print_vector_float_operand): Likewise.
8053         (aarch64_print_operand): Handle 'N' and 'C'.  Use "zN" rather than
8054         "vN" for FP registers with SVE modes.  Handle (const ...) vectors
8055         and the FP immediates 1.0 and 0.5.
8056         (aarch64_print_address_internal): Handle SVE addresses.
8057         (aarch64_print_operand_address): Use ADDR_QUERY_ANY.
8058         (aarch64_regno_regclass): Handle predicate registers.
8059         (aarch64_secondary_reload): Handle big-endian reloads of SVE
8060         data modes.
8061         (aarch64_class_max_nregs): Handle SVE modes and predicate registers.
8062         (aarch64_rtx_costs): Check for ADDVL and ADDPL instructions.
8063         (aarch64_convert_sve_vector_bits): New function.
8064         (aarch64_override_options): Use it to handle -msve-vector-bits=.
8065         (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
8066         rather than an rtx.
8067         (aarch64_legitimate_constant_p): Use aarch64_classify_vector_mode.
8068         Handle SVE vector and predicate modes.  Accept VL-based constants
8069         that need only one temporary register, and VL offsets that require
8070         no temporary registers.
8071         (aarch64_conditional_register_usage): Mark the predicate registers
8072         as fixed if SVE isn't available.
8073         (aarch64_vector_mode_supported_p): Use aarch64_classify_vector_mode.
8074         Return true for SVE vector and predicate modes.
8075         (aarch64_simd_container_mode): Take the number of bits as a poly_int64
8076         rather than an unsigned int.  Handle SVE modes.
8077         (aarch64_preferred_simd_mode): Update call accordingly.  Handle
8078         SVE modes.
8079         (aarch64_autovectorize_vector_sizes): Add BYTES_PER_SVE_VECTOR
8080         if SVE is enabled.
8081         (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
8082         (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
8083         (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
8084         (aarch64_sve_float_mul_immediate_p): New functions.
8085         (aarch64_sve_valid_immediate): New function.
8086         (aarch64_simd_valid_immediate): Use it as the fallback for SVE vectors.
8087         Explicitly reject structure modes.  Check for INDEX constants.
8088         Handle PTRUE and PFALSE constants.
8089         (aarch64_check_zero_based_sve_index_immediate): New function.
8090         (aarch64_simd_imm_zero_p): Delete.
8091         (aarch64_mov_operand_p): Use aarch64_simd_valid_immediate for
8092         vector modes.  Accept constants in the range of CNT[BHWD].
8093         (aarch64_simd_scalar_immediate_valid_for_move): Explicitly
8094         ask for an Advanced SIMD mode.
8095         (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): New functions.
8096         (aarch64_simd_vector_alignment): Handle SVE predicates.
8097         (aarch64_vectorize_preferred_vector_alignment): New function.
8098         (aarch64_simd_vector_alignment_reachable): Use it instead of
8099         the vector size.
8100         (aarch64_shift_truncation_mask): Use aarch64_vector_data_mode_p.
8101         (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): New
8102         functions.
8103         (MAX_VECT_LEN): Delete.
8104         (expand_vec_perm_d): Add a vec_flags field.
8105         (emit_unspec2, aarch64_expand_sve_vec_perm): New functions.
8106         (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_zip)
8107         (aarch64_evpc_ext): Don't apply a big-endian lane correction
8108         for SVE modes.
8109         (aarch64_evpc_rev): Rename to...
8110         (aarch64_evpc_rev_local): ...this.  Use a predicated operation for SVE.
8111         (aarch64_evpc_rev_global): New function.
8112         (aarch64_evpc_dup): Enforce a 64-byte range for SVE DUP.
8113         (aarch64_evpc_tbl): Use MAX_COMPILE_TIME_VEC_BYTES instead of
8114         MAX_VECT_LEN.
8115         (aarch64_evpc_sve_tbl): New function.
8116         (aarch64_expand_vec_perm_const_1): Update after rename of
8117         aarch64_evpc_rev.  Handle SVE permutes too, trying
8118         aarch64_evpc_rev_global and using aarch64_evpc_sve_tbl rather
8119         than aarch64_evpc_tbl.
8120         (aarch64_vectorize_vec_perm_const): Initialize vec_flags.
8121         (aarch64_sve_cmp_operand_p, aarch64_unspec_cond_code)
8122         (aarch64_gen_unspec_cond, aarch64_expand_sve_vec_cmp_int)
8123         (aarch64_emit_unspec_cond, aarch64_emit_unspec_cond_or)
8124         (aarch64_emit_inverted_unspec_cond, aarch64_expand_sve_vec_cmp_float)
8125         (aarch64_expand_sve_vcond): New functions.
8126         (aarch64_modes_tieable_p): Use aarch64_vector_data_mode_p instead
8127         of aarch64_vector_mode_p.
8128         (aarch64_dwarf_poly_indeterminate_value): New function.
8129         (aarch64_compute_pressure_classes): Likewise.
8130         (aarch64_can_change_mode_class): Likewise.
8131         (TARGET_GET_RAW_RESULT_MODE, TARGET_GET_RAW_ARG_MODE): Redefine.
8132         (TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT): Likewise.
8133         (TARGET_VECTORIZE_GET_MASK_MODE): Likewise.
8134         (TARGET_DWARF_POLY_INDETERMINATE_VALUE): Likewise.
8135         (TARGET_COMPUTE_PRESSURE_CLASSES): Likewise.
8136         (TARGET_CAN_CHANGE_MODE_CLASS): Likewise.
8137         * config/aarch64/constraints.md (Upa, Upl, Uav, Uat, Usv, Usi, Utr)
8138         (Uty, Dm, vsa, vsc, vsd, vsi, vsn, vsl, vsm, vsA, vsM, vsN): New
8139         constraints.
8140         (Dn, Dl, Dr): Accept const as well as const_vector.
8141         (Dz): Likewise.  Compare against CONST0_RTX.
8142         * config/aarch64/iterators.md: Refer to "Advanced SIMD" instead
8143         of "vector" where appropriate.
8144         (SVE_ALL, SVE_BH, SVE_BHS, SVE_BHSI, SVE_HSDI, SVE_HSF, SVE_SD)
8145         (SVE_SDI, SVE_I, SVE_F, PRED_ALL, PRED_BHS): New mode iterators.
8146         (UNSPEC_SEL, UNSPEC_ANDF, UNSPEC_IORF, UNSPEC_XORF, UNSPEC_COND_LT)
8147         (UNSPEC_COND_LE, UNSPEC_COND_EQ, UNSPEC_COND_NE, UNSPEC_COND_GE)
8148         (UNSPEC_COND_GT, UNSPEC_COND_LO, UNSPEC_COND_LS, UNSPEC_COND_HS)
8149         (UNSPEC_COND_HI, UNSPEC_COND_UO): New unspecs.
8150         (Vetype, VEL, Vel, VWIDE, Vwide, vw, vwcore, V_INT_EQUIV)
8151         (v_int_equiv): Extend to SVE modes.
8152         (Vesize, V128, v128, Vewtype, V_FP_EQUIV, v_fp_equiv, VPRED): New
8153         mode attributes.
8154         (LOGICAL_OR, SVE_INT_UNARY, SVE_FP_UNARY): New code iterators.
8155         (optab): Handle popcount, smin, smax, umin, umax, abs and sqrt.
8156         (logical_nn, lr, sve_int_op, sve_fp_op): New code attributs.
8157         (LOGICALF, OPTAB_PERMUTE, UNPACK, UNPACK_UNSIGNED, SVE_COND_INT_CMP)
8158         (SVE_COND_FP_CMP): New int iterators.
8159         (perm_hilo): Handle the new unpack unspecs.
8160         (optab, logicalf_op, su, perm_optab, cmp_op, imm_con): New int
8161         attributes.
8162         * config/aarch64/predicates.md (aarch64_sve_cnt_immediate)
8163         (aarch64_sve_addvl_addpl_immediate, aarch64_split_add_offset_immediate)
8164         (aarch64_pluslong_or_poly_operand, aarch64_nonmemory_operand)
8165         (aarch64_equality_operator, aarch64_constant_vector_operand)
8166         (aarch64_sve_ld1r_operand, aarch64_sve_ldr_operand): New predicates.
8167         (aarch64_sve_nonimmediate_operand): Likewise.
8168         (aarch64_sve_general_operand): Likewise.
8169         (aarch64_sve_dup_operand, aarch64_sve_arith_immediate): Likewise.
8170         (aarch64_sve_sub_arith_immediate, aarch64_sve_inc_dec_immediate)
8171         (aarch64_sve_logical_immediate, aarch64_sve_mul_immediate): Likewise.
8172         (aarch64_sve_dup_immediate, aarch64_sve_cmp_vsc_immediate): Likewise.
8173         (aarch64_sve_cmp_vsd_immediate, aarch64_sve_index_immediate): Likewise.
8174         (aarch64_sve_float_arith_immediate): Likewise.
8175         (aarch64_sve_float_arith_with_sub_immediate): Likewise.
8176         (aarch64_sve_float_mul_immediate, aarch64_sve_arith_operand): Likewise.
8177         (aarch64_sve_add_operand, aarch64_sve_logical_operand): Likewise.
8178         (aarch64_sve_lshift_operand, aarch64_sve_rshift_operand): Likewise.
8179         (aarch64_sve_mul_operand, aarch64_sve_cmp_vsc_operand): Likewise.
8180         (aarch64_sve_cmp_vsd_operand, aarch64_sve_index_operand): Likewise.
8181         (aarch64_sve_float_arith_operand): Likewise.
8182         (aarch64_sve_float_arith_with_sub_operand): Likewise.
8183         (aarch64_sve_float_mul_operand): Likewise.
8184         (aarch64_sve_vec_perm_operand): Likewise.
8185         (aarch64_pluslong_operand): Include aarch64_sve_addvl_addpl_immediate.
8186         (aarch64_mov_operand): Accept const_poly_int and const_vector.
8187         (aarch64_simd_lshift_imm, aarch64_simd_rshift_imm): Accept const
8188         as well as const_vector.
8189         (aarch64_simd_imm_zero, aarch64_simd_imm_minus_one): Move earlier
8190         in file.  Use CONST0_RTX and CONSTM1_RTX.
8191         (aarch64_simd_or_scalar_imm_zero): Likewise.  Add match_codes.
8192         (aarch64_simd_reg_or_zero): Accept const as well as const_vector.
8193         Use aarch64_simd_imm_zero.
8194         * config/aarch64/aarch64-sve.md: New file.
8195         * config/aarch64/aarch64.md: Include it.
8196         (VG_REGNUM, P0_REGNUM, P7_REGNUM, P15_REGNUM): New register numbers.
8197         (UNSPEC_REV, UNSPEC_LD1_SVE, UNSPEC_ST1_SVE, UNSPEC_MERGE_PTRUE)
8198         (UNSPEC_PTEST_PTRUE, UNSPEC_UNPACKSHI, UNSPEC_UNPACKUHI)
8199         (UNSPEC_UNPACKSLO, UNSPEC_UNPACKULO, UNSPEC_PACK)
8200         (UNSPEC_FLOAT_CONVERT, UNSPEC_WHILE_LO): New unspec constants.
8201         (sve): New attribute.
8202         (enabled): Disable instructions with the sve attribute unless
8203         TARGET_SVE.
8204         (movqi, movhi): Pass CONST_POLY_INT operaneds through
8205         aarch64_expand_mov_immediate.
8206         (*mov<mode>_aarch64, *movsi_aarch64, *movdi_aarch64): Handle
8207         CNT[BHSD] immediates.
8208         (movti): Split CONST_POLY_INT moves into two halves.
8209         (add<mode>3): Accept aarch64_pluslong_or_poly_operand.
8210         Split additions that need a temporary here if the destination
8211         is the stack pointer.
8212         (*add<mode>3_aarch64): Handle ADDVL and ADDPL immediates.
8213         (*add<mode>3_poly_1): New instruction.
8214         (set_clobber_cc): New expander.
8215
8216 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
8217
8218         * simplify-rtx.c (simplify_immed_subreg): Add an inner_bytes
8219         parameter and use it instead of GET_MODE_SIZE (innermode).  Use
8220         inner_bytes * BITS_PER_UNIT instead of GET_MODE_BITSIZE (innermode).
8221         Use CEIL (inner_bytes, GET_MODE_UNIT_SIZE (innermode)) instead of
8222         GET_MODE_NUNITS (innermode).  Also add a first_elem parameter.
8223         Change innermode from fixed_mode_size to machine_mode.
8224         (simplify_subreg): Update call accordingly.  Handle a constant-sized
8225         subreg of a variable-length CONST_VECTOR.
8226
8227 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
8228             Alan Hayward  <alan.hayward@arm.com>
8229             David Sherwood  <david.sherwood@arm.com>
8230
8231         * tree-ssa-address.c (mem_ref_valid_without_offset_p): New function.
8232         (add_offset_to_base): New function, split out from...
8233         (create_mem_ref): ...here.  When handling a scale other than 1,
8234         check first whether the address is valid without the offset.
8235         Add it into the base if so, leaving the index and scale as-is.
8236
8237 2018-01-12  Jakub Jelinek  <jakub@redhat.com>
8238
8239         PR c++/83778
8240         * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Call
8241         fold_for_warn before checking if arg2 is INTEGER_CST.
8242
8243 2018-01-12  Segher Boessenkool  <segher@kernel.crashing.org>
8244
8245         * config/rs6000/predicates.md (load_multiple_operation): Delete.
8246         (store_multiple_operation): Delete.
8247         * config/rs6000/rs6000-cpus.def (601): Remove MASK_STRING.
8248         * config/rs6000/rs6000-protos.h (rs6000_output_load_multiple): Delete.
8249         * config/rs6000/rs6000-string.c (expand_block_move): Delete everything
8250         guarded by TARGET_STRING.
8251         (rs6000_output_load_multiple): Delete.
8252         * config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
8253         OPTION_MASK_STRING / TARGET_STRING handling.
8254         (print_operand) <'N', 'O'>: Add comment that these are unused now.
8255         (const rs6000_opt_masks) <"string">: Change mask to 0.
8256         * config/rs6000/rs6000.h (TARGET_DEFAULT): Remove MASK_STRING.
8257         (MASK_STRING): Delete.
8258         * config/rs6000/rs6000.md (*mov<mode>_string): Delete TARGET_STRING
8259         parts.  Simplify.
8260         (load_multiple): Delete.
8261         (*ldmsi8): Delete.
8262         (*ldmsi7): Delete.
8263         (*ldmsi6): Delete.
8264         (*ldmsi5): Delete.
8265         (*ldmsi4): Delete.
8266         (*ldmsi3): Delete.
8267         (store_multiple): Delete.
8268         (*stmsi8): Delete.
8269         (*stmsi7): Delete.
8270         (*stmsi6): Delete.
8271         (*stmsi5): Delete.
8272         (*stmsi4): Delete.
8273         (*stmsi3): Delete.
8274         (movmemsi_8reg): Delete.
8275         (corresponding unnamed define_insn): Delete.
8276         (movmemsi_6reg): Delete.
8277         (corresponding unnamed define_insn): Delete.
8278         (movmemsi_4reg): Delete.
8279         (corresponding unnamed define_insn): Delete.
8280         (movmemsi_2reg): Delete.
8281         (corresponding unnamed define_insn): Delete.
8282         (movmemsi_1reg): Delete.
8283         (corresponding unnamed define_insn): Delete.
8284         * config/rs6000/rs6000.opt (mno-string): New.
8285         (mstring): Replace by deprecation warning stub.
8286         * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mstring.
8287
8288 2018-01-12  Jakub Jelinek  <jakub@redhat.com>
8289
8290         * regrename.c (regrename_do_replace): If replacing the same
8291         reg multiple times, try to reuse last created gen_raw_REG.
8292
8293         PR debug/81155
8294         * bb-reorder.c (pass_partition_blocks::gate): In lto don't partition
8295         main to workaround a bug in GDB.
8296
8297 2018-01-12  Tom de Vries  <tom@codesourcery.com>
8298
8299         PR target/83737
8300         * config.gcc (nvptx*-*-*): Set use_gcc_stdint=wrap.
8301
8302 2018-01-12  Vladimir Makarov  <vmakarov@redhat.com>
8303
8304         PR rtl-optimization/80481
8305         * ira-color.c (get_cap_member): New function.
8306         (allocnos_conflict_by_live_ranges_p): Use it.
8307         (slot_coalesced_allocno_live_ranges_intersect_p): Add assert.
8308         (setup_slot_coalesced_allocno_live_ranges): Ditto.
8309
8310 2018-01-12  Uros Bizjak  <ubizjak@gmail.com>
8311
8312         PR target/83628
8313         * config/alpha/alpha.md (*saddsi_1): New insn_ans_split pattern.
8314         (*saddl_se_1): Ditto.
8315         (*ssubsi_1): Ditto.
8316         (*ssubl_se_1): Ditto.
8317
8318 2018-01-12  Richard Sandiford  <richard.sandiford@linaro.org>
8319
8320         * tree-predcom.c (aff_combination_dr_offset): Use wi::to_poly_widest
8321         rather than wi::to_widest for DR_INITs.
8322         * tree-vect-data-refs.c (vect_find_same_alignment_drs): Use
8323         wi::to_poly_offset rather than wi::to_offset for DR_INIT.
8324         (vect_analyze_data_ref_accesses): Require both DR_INITs to be
8325         INTEGER_CSTs.
8326         (vect_analyze_group_access_1): Note that here.
8327
8328 2018-01-12  Richard Sandiford  <richard.sandiford@linaro.org>
8329
8330         * tree-vectorizer.c (get_vec_alignment_for_array_type): Handle
8331         polynomial type sizes.
8332
8333 2018-01-12  Richard Sandiford  <richard.sandiford@linaro.org>
8334
8335         * gimplify.c (gimple_add_tmp_var_fn): Allow variables to have a
8336         poly_uint64 size, rather than requiring an unsigned HOST_WIDE_INT size.
8337         (gimple_add_tmp_var): Likewise.
8338
8339 2018-01-12  Martin Liska  <mliska@suse.cz>
8340
8341         * gimple.c (gimple_alloc_counts): Use uint64_t instead of int.
8342         (gimple_alloc_sizes): Likewise.
8343         (dump_gimple_statistics): Use PRIu64 in printf format.
8344         * gimple.h: Change uint64_t to int.
8345
8346 2018-01-12  Martin Liska  <mliska@suse.cz>
8347
8348         * tree-core.h: Use uint64_t instead of int.
8349         * tree.c (tree_node_counts): Likewise.
8350         (tree_node_sizes): Likewise.
8351         (dump_tree_statistics): Use PRIu64 in printf format.
8352
8353 2018-01-12  Martin Liska  <mliska@suse.cz>
8354
8355         * Makefile.in: As qsort_chk is implemented in vec.c, add
8356         vec.o to linkage of gencfn-macros.
8357         * tree.c (build_new_poly_int_cst): Add CXX_MEM_STAT_INFO as it's
8358         passing the info to record_node_allocation_statistics.
8359         (test_vector_cst_patterns): Add CXX_MEM_STAT_INFO to declaration
8360         and pass the info.
8361         * ggc-common.c (struct ggc_usage): Add operator== and use
8362         it in operator< and compare function.
8363         * mem-stats.h (struct mem_usage): Likewise.
8364         * vec.c (struct vec_usage): Remove operator< and compare
8365         function. Can be simply inherited.
8366
8367 2018-01-12  Martin Jambor  <mjambor@suse.cz>
8368
8369         PR target/81616
8370         * params.def: New parameter PARAM_AVOID_FMA_MAX_BITS.
8371         * tree-ssa-math-opts.c: Include domwalk.h.
8372         (convert_mult_to_fma_1): New function.
8373         (fma_transformation_info): New type.
8374         (fma_deferring_state): Likewise.
8375         (cancel_fma_deferring): New function.
8376         (result_of_phi): Likewise.
8377         (last_fma_candidate_feeds_initial_phi): Likewise.
8378         (convert_mult_to_fma): Added deferring logic, split actual
8379         transformation to convert_mult_to_fma_1.
8380         (math_opts_dom_walker): New type.
8381         (math_opts_dom_walker::after_dom_children): New method, body moved
8382         here from pass_optimize_widening_mul::execute, added deferring logic
8383         bits.
8384         (pass_optimize_widening_mul::execute): Moved most of code to
8385         math_opts_dom_walker::after_dom_children.
8386         * config/i386/x86-tune.def (X86_TUNE_AVOID_128FMA_CHAINS): New.
8387         * config/i386/i386.c (ix86_option_override_internal): Added
8388         maybe_setting of PARAM_AVOID_FMA_MAX_BITS.
8389
8390 2018-01-12  Richard Biener  <rguenther@suse.de>
8391
8392         PR debug/83157
8393         * dwarf2out.c (gen_variable_die): Do not reset old_die for
8394         inline instance vars.
8395
8396 2018-01-12  Oleg Endo  <olegendo@gcc.gnu.org>
8397
8398         PR target/81819
8399         * config/rx/rx.c (rx_is_restricted_memory_address):
8400         Handle SUBREG case.
8401
8402 2018-01-12  Richard Biener  <rguenther@suse.de>
8403
8404         PR tree-optimization/80846
8405         * target.def (split_reduction): New target hook.
8406         * targhooks.c (default_split_reduction): New function.
8407         * targhooks.h (default_split_reduction): Declare.
8408         * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
8409         target requests first reduce vectors by combining low and high
8410         parts.
8411         * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
8412         (get_vectype_for_scalar_type_and_size): Export.
8413         * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
8414         * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
8415         * doc/tm.texi: Regenerate.
8416         * config/i386/i386.c (ix86_split_reduction): Implement
8417         TARGET_VECTORIZE_SPLIT_REDUCTION.
8418
8419 2018-01-12  Eric Botcazou  <ebotcazou@adacore.com>
8420
8421         PR target/83368
8422         * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
8423         in PIC mode except for TARGET_VXWORKS_RTP.
8424         * config/sparc/sparc.c: Include cfgrtl.h.
8425         (TARGET_INIT_PIC_REG): Define.
8426         (TARGET_USE_PSEUDO_PIC_REG): Likewise.
8427         (sparc_pic_register_p): New predicate.
8428         (sparc_legitimate_address_p): Use it.
8429         (sparc_legitimize_pic_address): Likewise.
8430         (sparc_delegitimize_address): Likewise.
8431         (sparc_mode_dependent_address_p): Likewise.
8432         (gen_load_pcrel_sym): Remove 4th parameter.
8433         (load_got_register): Adjust call to above.  Remove obsolete stuff.
8434         (sparc_expand_prologue): Do not call load_got_register here.
8435         (sparc_flat_expand_prologue): Likewise.
8436         (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
8437         (sparc_use_pseudo_pic_reg): New function.
8438         (sparc_init_pic_reg): Likewise.
8439         * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
8440         (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
8441
8442 2018-01-12  Christophe Lyon  <christophe.lyon@linaro.org>
8443
8444         * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
8445         Add item for branch_cost.
8446
8447 2018-01-12  Eric Botcazou  <ebotcazou@adacore.com>
8448
8449         PR rtl-optimization/83565
8450         * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
8451         not extend the result to a larger mode for rotate operations.
8452         (num_sign_bit_copies1): Likewise.
8453
8454 2018-01-12  Rainer Orth  <ro@CeBiTec.Uni-Bielefeld.DE>
8455
8456         PR target/40411
8457         * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
8458         -symbolic.
8459         Use values-Xc.o for -pedantic.
8460         Link with values-xpg4.o for C90, values-xpg6.o otherwise.
8461
8462 2018-01-12  Martin Liska  <mliska@suse.cz>
8463
8464         PR ipa/83054
8465         * ipa-devirt.c (final_warning_record::grow_type_warnings):
8466         New function.
8467         (possible_polymorphic_call_targets): Use it.
8468         (ipa_devirt): Likewise.
8469
8470 2018-01-12  Martin Liska  <mliska@suse.cz>
8471
8472         * profile-count.h (enum profile_quality): Use 0 as invalid
8473         enum value of profile_quality.
8474
8475 2018-01-12  Chung-Ju Wu  <jasonwucj@gmail.com>
8476
8477         * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
8478         -mext-string options.
8479
8480 2018-01-12  Richard Biener  <rguenther@suse.de>
8481
8482         * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
8483         DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
8484         * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
8485         Likewise.
8486         * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
8487
8488 2018-01-11  Michael Meissner  <meissner@linux.vnet.ibm.com>
8489
8490         * configure.ac (--with-long-double-format): Add support for the
8491         configuration option to change the default long double format on
8492         PowerPC systems.
8493         * config.gcc (powerpc*-linux*-*): Likewise.
8494         * configure: Regenerate.
8495         * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
8496         double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
8497         used without modification.
8498
8499 2018-01-11  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
8500
8501         * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
8502         (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
8503         * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
8504         MISC_BUILTIN_SPEC_BARRIER.
8505         (rs6000_init_builtins): Likewise.
8506         * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
8507         enum value.
8508         (speculation_barrier): New define_insn.
8509         * doc/extend.texi: Document __builtin_speculation_barrier.
8510
8511 2018-01-11  Jakub Jelinek  <jakub@redhat.com>
8512
8513         PR target/83203
8514         * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
8515         is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
8516         * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
8517         iterators.
8518         (ssescalarmodesuffix): Add 512-bit vectors.  Use "d" or "q" for
8519         integral modes instead of "ss" and "sd".
8520         (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
8521         vectors with 32-bit and 64-bit elements.
8522         (vecdupssescalarmodesuffix): New mode attribute.
8523         (vec_dup<mode>): Use it.
8524
8525 2018-01-11  H.J. Lu  <hongjiu.lu@intel.com>
8526
8527         PR target/83330
8528         * config/i386/i386.c (ix86_compute_frame_layout): Align stack
8529         frame if argument is passed on stack.
8530
8531 2018-01-11  Jakub Jelinek  <jakub@redhat.com>
8532
8533         PR target/82682
8534         * ree.c (combine_reaching_defs): Optimize also
8535         reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
8536         reg2=any_extend(exp); reg1=reg2;, formatting fix.
8537
8538 2018-01-11  Jan Hubicka  <hubicka@ucw.cz>
8539
8540         PR middle-end/83189
8541         * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
8542
8543 2018-01-11  Jan Hubicka  <hubicka@ucw.cz>
8544
8545         PR middle-end/83718
8546         * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
8547         after they are computed.
8548
8549 2018-01-11  Bin Cheng  <bin.cheng@arm.com>
8550
8551         PR tree-optimization/83695
8552         * gimple-loop-linterchange.cc
8553         (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
8554         reset cached scev information after interchange.
8555         (pass_linterchange::execute): Remove call to scev_reset_htab.
8556
8557 2018-01-11  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
8558
8559         * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
8560         vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
8561         vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
8562         vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
8563         vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
8564         vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
8565         * config/arm/arm_neon_builtins.def (vfmal_lane_low,
8566         vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
8567         vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
8568         vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
8569         vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
8570         * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
8571         (V_lane_reg): Likewise.
8572         * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
8573         New define_expand.
8574         (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
8575         (vfmal_lane_low<mode>_intrinsic,
8576         vfmal_lane_low<vfmlsel2><mode>_intrinsic,
8577         vfmal_lane_high<vfmlsel2><mode>_intrinsic,
8578         vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
8579         vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
8580         vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
8581         vfmsl_lane_high<mode>_intrinsic): New define_insns.
8582
8583 2018-01-11  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
8584
8585         * config/arm/arm-cpus.in (fp16fml): New feature.
8586         (ALL_SIMD): Add fp16fml.
8587         (armv8.2-a): Add fp16fml as an option.
8588         (armv8.3-a): Likewise.
8589         (armv8.4-a): Add fp16fml as part of fp16.
8590         * config/arm/arm.h (TARGET_FP16FML): Define.
8591         * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
8592         when appropriate.
8593         * config/arm/arm-modes.def (V2HF): Define.
8594         * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
8595         vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
8596         vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
8597         * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
8598         vfmsl_low, vfmsl_high): New set of builtins.
8599         * config/arm/iterators.md (PLUSMINUS): New code iterator.
8600         (vfml_op): New code attribute.
8601         (VFMLHALVES): New int iterator.
8602         (VFML, VFMLSEL): New mode attributes.
8603         (V_reg): Define mapping for V2HF.
8604         (V_hi, V_lo): New mode attributes.
8605         (VF_constraint): Likewise.
8606         (vfml_half, vfml_half_selector): New int attributes.
8607         * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
8608         define_expand.
8609         (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
8610         vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
8611         New define_insn.
8612         * config/arm/t-arm-elf (v8_fps): Add fp16fml.
8613         * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
8614         * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
8615         * doc/invoke.texi (ARM Options): Document fp16fml.  Update armv8.4-a
8616         documentation.
8617         * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
8618         Document new effective target and option set.
8619
8620 2018-01-11  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
8621
8622         * config/arm/arm-cpus.in (armv8_4): New feature.
8623         (ARMv8_4a): New fgroup.
8624         (armv8.4-a): New arch.
8625         * config/arm/arm-tables.opt: Regenerate.
8626         * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
8627         * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
8628         * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
8629         Add matching rules for -march=armv8.4-a and extensions.
8630         * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
8631
8632 2018-01-11  Oleg Endo  <olegendo@gcc.gnu.org>
8633
8634         PR target/81821
8635         * config/rx/rx.md (BW): New mode attribute.
8636         (sync_lock_test_and_setsi): Add mode suffix to insn output.
8637
8638 2018-01-11  Richard Biener  <rguenther@suse.de>
8639
8640         PR tree-optimization/83435
8641         * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
8642         * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
8643         * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
8644
8645 2018-01-11  Richard Sandiford  <richard.sandiford@linaro.org>
8646             Alan Hayward  <alan.hayward@arm.com>
8647             David Sherwood  <david.sherwood@arm.com>
8648
8649         * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
8650         field.
8651         (aarch64_classify_address): Initialize it.  Track polynomial offsets.
8652         (aarch64_print_address_internal): Use it to check for a zero offset.
8653
8654 2018-01-11  Richard Sandiford  <richard.sandiford@linaro.org>
8655             Alan Hayward  <alan.hayward@arm.com>
8656             David Sherwood  <david.sherwood@arm.com>
8657
8658         * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
8659         * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
8660         Return a poly_int64 rather than a HOST_WIDE_INT.
8661         (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
8662         rather than a HOST_WIDE_INT.
8663         * config/aarch64/aarch64.h (aarch64_frame): Protect with
8664         HAVE_POLY_INT_H rather than HOST_WIDE_INT.  Change locals_offset,
8665         hard_fp_offset, frame_size, initial_adjust, callee_offset and
8666         final_offset from HOST_WIDE_INT to poly_int64.
8667         * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
8668         to_constant when getting the number of units in an Advanced SIMD
8669         mode.
8670         (aarch64_builtin_vectorized_function): Check for a constant number
8671         of units.
8672         * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
8673         GET_MODE_SIZE.
8674         (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
8675         attribute instead of GET_MODE_NUNITS.
8676         * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
8677         (aarch64_class_max_nregs): Use the constant_lowest_bound of the
8678         GET_MODE_SIZE for fixed-size registers.
8679         (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
8680         (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
8681         (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
8682         (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
8683         (aarch64_print_operand, aarch64_print_address_internal)
8684         (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
8685         (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
8686         (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
8687         Handle polynomial GET_MODE_SIZE.
8688         (aarch64_hard_regno_caller_save_mode): Likewise.  Return modes
8689         wider than SImode without modification.
8690         (tls_symbolic_operand_type): Use strip_offset instead of split_const.
8691         (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
8692         (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
8693         passing and returning SVE modes.
8694         (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
8695         rather than GEN_INT.
8696         (aarch64_emit_probe_stack_range): Take the size as a poly_int64
8697         rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
8698         (aarch64_allocate_and_probe_stack_space): Likewise.
8699         (aarch64_layout_frame): Cope with polynomial offsets.
8700         (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
8701         start_offset as a poly_int64 rather than a HOST_WIDE_INT.  Track
8702         polynomial offsets.
8703         (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
8704         (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
8705         poly_int64 rather than a HOST_WIDE_INT.
8706         (aarch64_get_separate_components, aarch64_process_components)
8707         (aarch64_expand_prologue, aarch64_expand_epilogue)
8708         (aarch64_use_return_insn_p): Handle polynomial frame offsets.
8709         (aarch64_anchor_offset): New function, split out from...
8710         (aarch64_legitimize_address): ...here.
8711         (aarch64_builtin_vectorization_cost): Handle polynomial
8712         TYPE_VECTOR_SUBPARTS.
8713         (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
8714         GET_MODE_NUNITS.
8715         (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
8716         number of elements from the PARALLEL rather than the mode.
8717         (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
8718         rather than GET_MODE_BITSIZE.
8719         (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
8720         (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
8721         (aarch64_expand_vec_perm_const_1): Handle polynomial
8722         d->perm.length () and d->perm elements.
8723         (aarch64_evpc_tbl): Likewise.  Use nelt rather than GET_MODE_NUNITS.
8724         Apply to_constant to d->perm elements.
8725         (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
8726         polynomial CONST_VECTOR_NUNITS.
8727         (aarch64_move_pointer): Take amount as a poly_int64 rather
8728         than an int.
8729         (aarch64_progress_pointer): Avoid temporary variable.
8730         * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
8731         the mode attribute instead of GET_MODE.
8732
8733 2018-01-11  Richard Sandiford  <richard.sandiford@linaro.org>
8734             Alan Hayward  <alan.hayward@arm.com>
8735             David Sherwood  <david.sherwood@arm.com>
8736
8737         * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
8738         x exists before using it.
8739         (aarch64_add_constant_internal): Rename to...
8740         (aarch64_add_offset_1): ...this.  Replace regnum with separate
8741         src and dest rtxes.  Handle the case in which they're different,
8742         including when the offset is zero.  Replace scratchreg with an rtx.
8743         Use 2 additions if there is no spare register into which we can
8744         move a 16-bit constant.
8745         (aarch64_add_constant): Delete.
8746         (aarch64_add_offset): Replace reg with separate src and dest
8747         rtxes.  Take a poly_int64 offset instead of a HOST_WIDE_INT.
8748         Use aarch64_add_offset_1.
8749         (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
8750         an rtx rather than an int.  Take the delta as a poly_int64
8751         rather than a HOST_WIDE_INT.  Use aarch64_add_offset.
8752         (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
8753         (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
8754         aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
8755         (aarch64_expand_epilogue): Update calls to aarch64_add_offset
8756         and aarch64_add_sp.
8757         (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
8758         aarch64_add_constant.
8759
8760 2018-01-11  Richard Sandiford  <richard.sandiford@linaro.org>
8761
8762         * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
8763         Use scalar_float_mode.
8764
8765 2018-01-11  Richard Sandiford  <richard.sandiford@linaro.org>
8766
8767         * config/aarch64/aarch64-simd.md
8768         (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
8769         (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
8770         (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
8771         (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
8772         (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
8773         (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
8774         (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
8775         (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
8776         (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
8777         (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
8778
8779 2018-01-11  Prathamesh Kulkarni  <prathamesh.kulkarni@linaro.org>
8780
8781         PR target/83514
8782         * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
8783         targ_options->x_arm_arch_string is non NULL.
8784
8785 2018-01-11  Tamar Christina  <tamar.christina@arm.com>
8786
8787         * config/aarch64/aarch64.h
8788         (AARCH64_FL_FOR_ARCH8_4): Add  AARCH64_FL_DOTPROD.
8789
8790 2018-01-11  Sudakshina Das  <sudi.das@arm.com>
8791
8792         PR target/82096
8793         * expmed.c (emit_store_flag_force): Swap if const op0
8794         and change VOIDmode to mode of op0.
8795
8796 2018-01-11  Richard Sandiford  <richard.sandiford@linaro.org>
8797
8798         PR rtl-optimization/83761
8799         * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
8800         than bytes to mode_for_size.
8801
8802 2018-01-10  Jan Hubicka  <hubicka@ucw.cz>
8803
8804         PR middle-end/83189
8805         * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
8806         * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
8807         profile.
8808
8809 2018-01-10  Jan Hubicka  <hubicka@ucw.cz>
8810
8811         PR middle-end/83575
8812         * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
8813         when in layout mode.
8814         (cfg_layout_finalize): Do not verify cfg before we are out of layout.
8815         * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
8816         partition fixup.
8817
8818 2018-01-10  Michael Collison  <michael.collison@arm.com>
8819
8820         * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
8821         * config/aarch64/aarch64-option-extension.def: Add
8822         AARCH64_OPT_EXTENSION of 'fp16fml'.
8823         * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
8824         (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
8825         * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
8826         * config/aarch64/constraints.md (Ui7): New constraint.
8827         * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
8828         (VFMLA_SEL_W): Ditto.
8829         (f16quad): Ditto.
8830         (f16mac1): Ditto.
8831         (VFMLA16_LOW): New int iterator.
8832         (VFMLA16_HIGH): Ditto.
8833         (UNSPEC_FMLAL): New unspec.
8834         (UNSPEC_FMLSL): Ditto.
8835         (UNSPEC_FMLAL2): Ditto.
8836         (UNSPEC_FMLSL2): Ditto.
8837         (f16mac): New code attribute.
8838         * config/aarch64/aarch64-simd-builtins.def
8839         (aarch64_fmlal_lowv2sf): Ditto.
8840         (aarch64_fmlsl_lowv2sf): Ditto.
8841         (aarch64_fmlalq_lowv4sf): Ditto.
8842         (aarch64_fmlslq_lowv4sf): Ditto.
8843         (aarch64_fmlal_highv2sf): Ditto.
8844         (aarch64_fmlsl_highv2sf): Ditto.
8845         (aarch64_fmlalq_highv4sf): Ditto.
8846         (aarch64_fmlslq_highv4sf): Ditto.
8847         (aarch64_fmlal_lane_lowv2sf): Ditto.
8848         (aarch64_fmlsl_lane_lowv2sf): Ditto.
8849         (aarch64_fmlal_laneq_lowv2sf): Ditto.
8850         (aarch64_fmlsl_laneq_lowv2sf): Ditto.
8851         (aarch64_fmlalq_lane_lowv4sf): Ditto.
8852         (aarch64_fmlsl_lane_lowv4sf): Ditto.
8853         (aarch64_fmlalq_laneq_lowv4sf): Ditto.
8854         (aarch64_fmlsl_laneq_lowv4sf): Ditto.
8855         (aarch64_fmlal_lane_highv2sf): Ditto.
8856         (aarch64_fmlsl_lane_highv2sf): Ditto.
8857         (aarch64_fmlal_laneq_highv2sf): Ditto.
8858         (aarch64_fmlsl_laneq_highv2sf): Ditto.
8859         (aarch64_fmlalq_lane_highv4sf): Ditto.
8860         (aarch64_fmlsl_lane_highv4sf): Ditto.
8861         (aarch64_fmlalq_laneq_highv4sf): Ditto.
8862         (aarch64_fmlsl_laneq_highv4sf): Ditto.
8863         * config/aarch64/aarch64-simd.md:
8864         (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
8865         (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
8866         (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
8867         (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
8868         (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
8869         (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
8870         (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
8871         (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
8872         (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
8873         (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
8874         (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
8875         (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
8876         (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
8877         (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
8878         (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
8879         (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
8880         (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
8881         (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
8882         (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
8883         (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
8884         * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
8885         (vfmlsl_low_u32): Ditto.
8886         (vfmlalq_low_u32): Ditto.
8887         (vfmlslq_low_u32): Ditto.
8888         (vfmlal_high_u32): Ditto.
8889         (vfmlsl_high_u32): Ditto.
8890         (vfmlalq_high_u32): Ditto.
8891         (vfmlslq_high_u32): Ditto.
8892         (vfmlal_lane_low_u32): Ditto.
8893         (vfmlsl_lane_low_u32): Ditto.
8894         (vfmlal_laneq_low_u32): Ditto.
8895         (vfmlsl_laneq_low_u32): Ditto.
8896         (vfmlalq_lane_low_u32): Ditto.
8897         (vfmlslq_lane_low_u32): Ditto.
8898         (vfmlalq_laneq_low_u32): Ditto.
8899         (vfmlslq_laneq_low_u32): Ditto.
8900         (vfmlal_lane_high_u32): Ditto.
8901         (vfmlsl_lane_high_u32): Ditto.
8902         (vfmlal_laneq_high_u32): Ditto.
8903         (vfmlsl_laneq_high_u32): Ditto.
8904         (vfmlalq_lane_high_u32): Ditto.
8905         (vfmlslq_lane_high_u32): Ditto.
8906         (vfmlalq_laneq_high_u32): Ditto.
8907         (vfmlslq_laneq_high_u32): Ditto.
8908         * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
8909         (AARCH64_FL_FOR_ARCH8_4): New.
8910         (AARCH64_ISA_F16FML): New ISA flag.
8911         (TARGET_F16FML): New feature flag for fp16fml.
8912         (doc/invoke.texi): Document new fp16fml option.
8913
8914 2018-01-10  Michael Collison  <michael.collison@arm.com>
8915
8916         * config/aarch64/aarch64-builtins.c:
8917         (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
8918         * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
8919         (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
8920         * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
8921         (AARCH64_ISA_SHA3): New ISA flag.
8922         (TARGET_SHA3): New feature flag for sha3.
8923         * config/aarch64/iterators.md (sha512_op): New int attribute.
8924         (CRYPTO_SHA512): New int iterator.
8925         (UNSPEC_SHA512H): New unspec.
8926         (UNSPEC_SHA512H2): Ditto.
8927         (UNSPEC_SHA512SU0): Ditto.
8928         (UNSPEC_SHA512SU1): Ditto.
8929         * config/aarch64/aarch64-simd-builtins.def
8930         (aarch64_crypto_sha512hqv2di): New builtin.
8931         (aarch64_crypto_sha512h2qv2di): Ditto.
8932         (aarch64_crypto_sha512su0qv2di): Ditto.
8933         (aarch64_crypto_sha512su1qv2di): Ditto.
8934         (aarch64_eor3qv8hi): Ditto.
8935         (aarch64_rax1qv2di): Ditto.
8936         (aarch64_xarqv2di): Ditto.
8937         (aarch64_bcaxqv8hi): Ditto.
8938         * config/aarch64/aarch64-simd.md:
8939         (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
8940         (aarch64_crypto_sha512su0qv2di): Ditto.
8941         (aarch64_crypto_sha512su1qv2di): Ditto.
8942         (aarch64_eor3qv8hi): Ditto.
8943         (aarch64_rax1qv2di): Ditto.
8944         (aarch64_xarqv2di): Ditto.
8945         (aarch64_bcaxqv8hi): Ditto.
8946         * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
8947         (vsha512h2q_u64): Ditto.
8948         (vsha512su0q_u64): Ditto.
8949         (vsha512su1q_u64): Ditto.
8950         (veor3q_u16): Ditto.
8951         (vrax1q_u64): Ditto.
8952         (vxarq_u64): Ditto.
8953         (vbcaxq_u16): Ditto.
8954         * config/arm/types.md (crypto_sha512): New type attribute.
8955         (crypto_sha3): Ditto.
8956         (doc/invoke.texi): Document new sha3 option.
8957
8958 2018-01-10  Michael Collison  <michael.collison@arm.com>
8959
8960         * config/aarch64/aarch64-builtins.c:
8961         (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
8962         * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
8963         (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
8964         (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
8965         * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
8966         (AARCH64_ISA_SM4): New ISA flag.
8967         (TARGET_SM4): New feature flag for sm4.
8968         * config/aarch64/aarch64-simd-builtins.def
8969         (aarch64_sm3ss1qv4si): Ditto.
8970         (aarch64_sm3tt1aq4si): Ditto.
8971         (aarch64_sm3tt1bq4si): Ditto.
8972         (aarch64_sm3tt2aq4si): Ditto.
8973         (aarch64_sm3tt2bq4si): Ditto.
8974         (aarch64_sm3partw1qv4si): Ditto.
8975         (aarch64_sm3partw2qv4si): Ditto.
8976         (aarch64_sm4eqv4si): Ditto.
8977         (aarch64_sm4ekeyqv4si): Ditto.
8978         * config/aarch64/aarch64-simd.md:
8979         (aarch64_sm3ss1qv4si): Ditto.
8980         (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
8981         (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
8982         (aarch64_sm4eqv4si): Ditto.
8983         (aarch64_sm4ekeyqv4si): Ditto.
8984         * config/aarch64/iterators.md (sm3tt_op): New int iterator.
8985         (sm3part_op): Ditto.
8986         (CRYPTO_SM3TT): Ditto.
8987         (CRYPTO_SM3PART): Ditto.
8988         (UNSPEC_SM3SS1): New unspec.
8989         (UNSPEC_SM3TT1A): Ditto.
8990         (UNSPEC_SM3TT1B): Ditto.
8991         (UNSPEC_SM3TT2A): Ditto.
8992         (UNSPEC_SM3TT2B): Ditto.
8993         (UNSPEC_SM3PARTW1): Ditto.
8994         (UNSPEC_SM3PARTW2): Ditto.
8995         (UNSPEC_SM4E): Ditto.
8996         (UNSPEC_SM4EKEY): Ditto.
8997         * config/aarch64/constraints.md (Ui2): New constraint.
8998         * config/aarch64/predicates.md (aarch64_imm2): New predicate.
8999         * config/arm/types.md (crypto_sm3): New type attribute.
9000         (crypto_sm4): Ditto.
9001         * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
9002         (vsm3tt1aq_u32): Ditto.
9003         (vsm3tt1bq_u32): Ditto.
9004         (vsm3tt2aq_u32): Ditto.
9005         (vsm3tt2bq_u32): Ditto.
9006         (vsm3partw1q_u32): Ditto.
9007         (vsm3partw2q_u32): Ditto.
9008         (vsm4eq_u32): Ditto.
9009         (vsm4ekeyq_u32): Ditto.
9010         (doc/invoke.texi): Document new sm4 option.
9011
9012 2018-01-10  Michael Collison  <michael.collison@arm.com>
9013
9014         * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
9015         * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
9016         (AARCH64_FL_FOR_ARCH8_4): New.
9017         (AARCH64_FL_V8_4): New flag.
9018         (doc/invoke.texi): Document new armv8.4-a option.
9019
9020 2018-01-10  Michael Collison  <michael.collison@arm.com>
9021
9022         * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
9023         (__ARM_FEATURE_AES): Define if TARGET_AES is true.
9024         (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
9025         * config/aarch64/aarch64-option-extension.def: Add
9026         AARCH64_OPT_EXTENSION of 'sha2'.
9027         (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
9028         (crypto): Disable sha2 and aes if crypto disabled.
9029         (crypto): Enable aes and sha2 if enabled.
9030         (simd): Disable sha2 and aes if simd disabled.
9031         * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
9032         New flags.
9033         (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
9034         (TARGET_SHA2): New feature flag for sha2.
9035         (TARGET_AES): New feature flag for aes.
9036         * config/aarch64/aarch64-simd.md:
9037         (aarch64_crypto_aes<aes_op>v16qi): Make pattern
9038         conditional on TARGET_AES.
9039         (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
9040         (aarch64_crypto_sha1hsi): Make pattern conditional
9041         on TARGET_SHA2.
9042         (aarch64_crypto_sha1hv4si): Ditto.
9043         (aarch64_be_crypto_sha1hv4si): Ditto.
9044         (aarch64_crypto_sha1su1v4si): Ditto.
9045         (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
9046         (aarch64_crypto_sha1su0v4si): Ditto.
9047         (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
9048         (aarch64_crypto_sha256su0v4si): Ditto.
9049         (aarch64_crypto_sha256su1v4si): Ditto.
9050         (doc/invoke.texi): Document new aes and sha2 options.
9051
9052 2018-01-10  Martin Sebor  <msebor@redhat.com>
9053
9054         PR tree-optimization/83781
9055         * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
9056         as string arrays.
9057
9058 2018-01-11  Martin Sebor  <msebor@gmail.com>
9059             Prathamesh Kulkarni  <prathamesh.kulkarni@linaro.org>
9060
9061         PR tree-optimization/83501
9062         PR tree-optimization/81703
9063
9064         * tree-ssa-strlen.c (get_string_cst): Rename...
9065         (get_string_len): ...to this.  Handle global constants.
9066         (handle_char_store): Adjust.
9067
9068 2018-01-10  Kito Cheng  <kito.cheng@gmail.com>
9069             Jim Wilson  <jimw@sifive.com>
9070
9071         * config/riscv/riscv-protos.h (riscv_output_return): New.
9072         * config/riscv/riscv.c (struct machine_function): New naked_p field.
9073         (riscv_attribute_table, riscv_output_return),
9074         (riscv_handle_fndecl_attribute, riscv_naked_function_p),
9075         (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
9076         (riscv_compute_frame_info): Only compute frame->mask if not a naked
9077         function.
9078         (riscv_expand_prologue): Add early return for naked function.
9079         (riscv_expand_epilogue): Likewise.
9080         (riscv_function_ok_for_sibcall): Return false for naked function.
9081         (riscv_set_current_function): New.
9082         (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
9083         (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
9084         * config/riscv/riscv.md (simple_return): Call riscv_output_return.
9085         * doc/extend.texi (RISC-V Function Attributes): New.
9086
9087 2018-01-10  Michael Meissner  <meissner@linux.vnet.ibm.com>
9088
9089         * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
9090         check for 128-bit long double before checking TCmode.
9091         * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
9092         128-bit long doubles before checking TFmode or TCmode.
9093         (FLOAT128_IBM_P): Likewise.
9094
9095 2018-01-10  Martin Sebor  <msebor@redhat.com>
9096
9097         PR tree-optimization/83671
9098         * builtins.c (c_strlen): Unconditionally return zero for the empty
9099         string.
9100         Use -Warray-bounds for warnings.
9101         * gimple-fold.c (get_range_strlen): Handle non-constant lengths
9102         for non-constant array indices with COMPONENT_REF, arrays of
9103         arrays, and pointers to arrays.
9104         (gimple_fold_builtin_strlen): Determine and set length range for
9105         non-constant character arrays.
9106
9107 2018-01-10  Aldy Hernandez  <aldyh@redhat.com>
9108
9109         PR middle-end/81897
9110         * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
9111         empty blocks.
9112
9113 2018-01-10  Eric Botcazou  <ebotcazou@adacore.com>
9114
9115         * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
9116
9117 2018-01-10  Peter Bergner  <bergner@vnet.ibm.com>
9118
9119         PR target/83399
9120         * config/rs6000/rs6000.c (print_operand) <'y'>: Use
9121         VECTOR_MEM_ALTIVEC_OR_VSX_P.
9122         * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
9123         indexed_or_indirect_operand predicate.
9124         (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
9125         (*vsx_le_perm_load_v8hi): Likewise.
9126         (*vsx_le_perm_load_v16qi): Likewise.
9127         (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
9128         (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
9129         (*vsx_le_perm_store_v8hi): Likewise.
9130         (*vsx_le_perm_store_v16qi): Likewise.
9131         (eight unnamed splitters): Likewise.
9132
9133 2018-01-10  Peter Bergner  <bergner@vnet.ibm.com>
9134
9135         * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
9136         * config/rs6000/emmintrin.h: Likewise.
9137         * config/rs6000/mmintrin.h: Likewise.
9138         * config/rs6000/xmmintrin.h: Likewise.
9139
9140 2018-01-10  David Malcolm  <dmalcolm@redhat.com>
9141
9142         PR c++/43486
9143         * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
9144         "public_flag".
9145         * tree.c (tree_nop_conversion): Return true for location wrapper
9146         nodes.
9147         (maybe_wrap_with_location): New function.
9148         (selftest::check_strip_nops): New function.
9149         (selftest::test_location_wrappers): New function.
9150         (selftest::tree_c_tests): Call it.
9151         * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
9152         (maybe_wrap_with_location): New decl.
9153         (EXPR_LOCATION_WRAPPER_P): New macro.
9154         (location_wrapper_p): New inline function.
9155         (tree_strip_any_location_wrapper): New inline function.
9156
9157 2018-01-10  H.J. Lu  <hongjiu.lu@intel.com>
9158
9159         PR target/83735
9160         * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
9161         stack_realign_offset for the largest alignment of stack slot
9162         actually used.
9163         (ix86_find_max_used_stack_alignment): New function.
9164         (ix86_finalize_stack_frame_flags): Use it.  Set
9165         max_used_stack_alignment if we don't realign stack.
9166         * config/i386/i386.h (machine_function): Add
9167         max_used_stack_alignment.
9168
9169 2018-01-10  Christophe Lyon  <christophe.lyon@linaro.org>
9170
9171         * config/arm/arm.opt (-mbranch-cost): New option.
9172         * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
9173         account.
9174
9175 2018-01-10  Segher Boessenkool  <segher@kernel.crashing.org>
9176
9177         PR target/83629
9178         * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
9179         load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
9180
9181 2018-01-10  Richard Biener  <rguenther@suse.de>
9182
9183         PR debug/83765
9184         * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
9185         early out so it also covers the case where we have a non-NULL
9186         origin.
9187
9188 2018-01-10  Richard Sandiford  <richard.sandiford@linaro.org>
9189
9190         PR tree-optimization/83753
9191         * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
9192         for non-strided grouped accesses if the number of elements is 1.
9193
9194 2018-01-10  Jan Hubicka  <hubicka@ucw.cz>
9195
9196         PR target/81616
9197         * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
9198         * i386.h (TARGET_USE_GATHER): Define.
9199         * x86-tune.def (X86_TUNE_USE_GATHER): New.
9200
9201 2018-01-10  Martin Liska  <mliska@suse.cz>
9202
9203         PR bootstrap/82831
9204         * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
9205         * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
9206         partitioning.
9207         * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
9208         CLEANUP_NO_PARTITIONING is not set.
9209
9210 2018-01-10  Richard Sandiford  <richard.sandiford@linaro.org>
9211
9212         * doc/rtl.texi: Remove documentation of (const ...) wrappers
9213         for vectors, as a partial revert of r254296.
9214         * rtl.h (const_vec_p): Delete.
9215         (const_vec_duplicate_p): Don't test for vector CONSTs.
9216         (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
9217         * expmed.c (make_tree): Likewise.
9218
9219         Revert:
9220         * common.md (E, F): Use CONSTANT_P instead of checking for
9221         CONST_VECTOR.
9222         * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
9223         checking for CONST_VECTOR.
9224
9225 2018-01-09  Jan Hubicka  <hubicka@ucw.cz>
9226
9227         PR middle-end/83575
9228         * predict.c (force_edge_cold): Handle in more sane way edges
9229         with no prediction.
9230
9231 2018-01-09  Carl Love  <cel@us.ibm.com>
9232
9233         * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
9234         V4SI, V4SF types.
9235         (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
9236         * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
9237         VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
9238         VMRGOW_V2DI, VMRGOW_V2DF.  Remove definition for VMRGOW.
9239         * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
9240         P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW):  Add definitions.
9241         * config/rs6000/rs6000-protos.h: Add extern defition for
9242         rs6000_generate_float2_double_code.
9243         * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
9244         function.
9245         * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
9246         (float2_v2df): Add define_expand.
9247
9248 2018-01-09  Uros Bizjak  <ubizjak@gmail.com>
9249
9250         PR target/83628
9251         * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
9252         op_mode in the force_to_mode call.
9253
9254 2018-01-09  Richard Sandiford  <richard.sandiford@linaro.org>
9255
9256         * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
9257         instead of checking each element individually.
9258         (aarch64_evpc_uzp): Likewise.
9259         (aarch64_evpc_zip): Likewise.
9260         (aarch64_evpc_ext): Likewise.
9261         (aarch64_evpc_rev): Likewise.
9262         (aarch64_evpc_dup): Test the encoding for a single duplicated element,
9263         instead of checking each element individually.  Return true without
9264         generating rtl if
9265         (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
9266         whether all selected elements come from the same input, instead of
9267         checking each element individually.  Remove calls to gen_rtx_REG,
9268         start_sequence and end_sequence and instead assert that no rtl is
9269         generated.
9270
9271 2018-01-09  Richard Sandiford  <richard.sandiford@linaro.org>
9272
9273         * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
9274         order of HIGH and CONST checks.
9275
9276 2018-01-09  Richard Sandiford  <richard.sandiford@linaro.org>
9277
9278         * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
9279         if the destination isn't an SSA_NAME.
9280
9281 2018-01-09  Richard Biener  <rguenther@suse.de>
9282
9283         PR tree-optimization/83668
9284         * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
9285         move prologue...
9286         (canonicalize_loop_form): ... here, renamed from ...
9287         (canonicalize_loop_closed_ssa_form): ... this and amended to
9288         swap successor edges for loop exit blocks to make us use
9289         the RPO order we need for initial schedule generation.
9290
9291 2018-01-09  Joseph Myers  <joseph@codesourcery.com>
9292
9293         PR tree-optimization/64811
9294         * match.pd: When optimizing comparisons with Inf, avoid
9295         introducing or losing exceptions from comparisons with NaN.
9296
9297 2018-01-09  Martin Liska  <mliska@suse.cz>
9298
9299         PR sanitizer/82517
9300         * asan.c (shadow_mem_size): Add gcc_assert.
9301
9302 2018-01-09  Georg-Johann Lay  <avr@gjlay.de>
9303
9304         Don't save registers in main().
9305
9306         PR target/83738
9307         * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
9308         * config/avr/avr.opt (-mmain-is-OS_task): New target option.
9309         * config/avr/avr.c (avr_set_current_function): Don't error if
9310         naked, OS_task or OS_main are specified at the same time.
9311         (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
9312         OS_main.
9313         (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
9314         attribute.
9315         * common/config/avr/avr-common.c (avr_option_optimization_table):
9316         Switch on -mmain-is-OS_task for optimizing compilations.
9317
9318 2018-01-09  Richard Biener  <rguenther@suse.de>
9319
9320         PR tree-optimization/83572
9321         * graphite.c: Include cfganal.h.
9322         (graphite_transform_loops): Connect infinite loops to exit
9323         and remove fake edges at the end.
9324
9325 2018-01-09  Jan Hubicka  <hubicka@ucw.cz>
9326
9327         * ipa-inline.c (edge_badness): Revert accidental checkin.
9328
9329 2018-01-09  Jan Hubicka  <hubicka@ucw.cz>
9330
9331         PR ipa/80763
9332         * ipa-comdats.c (set_comdat_group): Only set comdat group of real
9333         symbols; not inline clones.
9334
9335 2018-01-09  Jakub Jelinek  <jakub@redhat.com>
9336
9337         PR target/83507
9338         * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
9339         hard registers.  Formatting fixes.
9340
9341         PR preprocessor/83722
9342         * gcc.c (try_generate_repro): Pass
9343         &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
9344         &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
9345         do_report_bug.
9346
9347 2018-01-08  Monk Chiang  <sh.chiang04@gmail.com>
9348             Kito Cheng  <kito.cheng@gmail.com>
9349
9350         * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
9351         (riscv_leaf_function_p): Delete.
9352         (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
9353
9354 2018-01-08  Aaron Sawdey  <acsawdey@linux.vnet.ibm.com>
9355
9356         * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
9357         function.
9358         (do_ifelse): New function.
9359         (do_isel): New function.
9360         (do_sub3): New function.
9361         (do_add3): New function.
9362         (do_load_mask_compare): New function.
9363         (do_overlap_load_compare): New function.
9364         (expand_compare_loop): New function.
9365         (expand_block_compare): Call expand_compare_loop() when appropriate.
9366         * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
9367         option description.
9368         (-mblock-compare-inline-loop-limit): New option.
9369
9370 2018-01-08  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
9371
9372         PR target/83677
9373         * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
9374         Reverse order of second and third operands in first alternative.
9375         * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
9376         of first and second elements in UNSPEC_VPERMR vector.
9377         (altivec_expand_vec_perm_le): Likewise.
9378
9379 2018-01-08  Jeff Law  <law@redhat.com>
9380
9381         PR rtl-optimizatin/81308
9382         * tree-switch-conversion.c (cfg_altered): New file scoped static.
9383         (process_switch): If group_case_labels makes a change, then set
9384         cfg_altered.
9385         (pass_convert_switch::execute): If a switch is converted, then
9386         set cfg_altered.  Return TODO_cfg_cleanup if cfg_altered is true.
9387
9388         PR rtl-optimization/81308
9389         * recog.c (split_all_insns): Conditionally cleanup the CFG after
9390         splitting insns.
9391
9392 2018-01-08  Vidya Praveen  <vidyapraveen@arm.com>
9393
9394         PR target/83663 - Revert r255946
9395         * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
9396         generation for cases where splatting a value is not useful.
9397         * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
9398         across a vec_duplicate and a paradoxical subreg forming a vector
9399         mode to a vec_concat.
9400
9401 2018-01-08  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
9402
9403         * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
9404         -march=armv8.3-a variants.
9405         * config/arm/t-multilib: Likewise.
9406         * config/arm/t-arm-elf: Likewise.  Handle dotprod extension.
9407
9408 2018-01-08  Aaron Sawdey  <acsawdey@linux.vnet.ibm.com>
9409
9410         * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
9411         to generate rtl.
9412         (cceq_ior_compare_complement): Give it a name so I can use it, and
9413         change boolean_or_operator predicate to boolean_operator so it can
9414         be used to generate a crand.
9415         (eqne): New code iterator.
9416         (bd/bd_neg): New code_attrs.
9417         (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
9418         a single define_insn.
9419         (<bd>tf_<mode>): A new insn pattern for the conditional form branch
9420         decrement (bdnzt/bdnzf/bdzt/bdzf).
9421         * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
9422         with the new names of the branch decrement patterns, and added the
9423         names of the branch decrement conditional patterns.
9424
9425 2018-01-08  Richard Biener  <rguenther@suse.de>
9426
9427         PR tree-optimization/83563
9428         * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
9429         cache.
9430
9431 2018-01-08  Richard Biener  <rguenther@suse.de>
9432
9433         PR middle-end/83713
9434         * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
9435
9436 2018-01-08  Richard Biener  <rguenther@suse.de>
9437
9438         PR tree-optimization/83685
9439         * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
9440         references to abnormals.
9441
9442 2018-01-08  Richard Biener  <rguenther@suse.de>
9443
9444         PR lto/83719
9445         * dwarf2out.c (output_indirect_strings): Handle empty
9446         skeleton_debug_str_hash.
9447         (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
9448
9449 2018-01-08  Claudiu Zissulescu  <claziss@synopsys.com>
9450
9451         * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
9452         (emit_store_direct): Likewise.
9453         (arc_trampoline_adjust_address): Likewise.
9454         (arc_asm_trampoline_template): New function.
9455         (arc_initialize_trampoline): Use asm_trampoline_template.
9456         (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
9457         * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
9458         * config/arc/arc.md (flush_icache): Delete pattern.
9459
9460 2018-01-08  Claudiu Zissulescu  <claziss@synopsys.com>
9461
9462         * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
9463         * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
9464         munaligned-access.
9465
9466 2018-01-08  Sebastian Huber  <sebastian.huber@embedded-brains.de>
9467
9468         PR target/83681
9469         * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
9470         by not USED_FOR_TARGET.
9471         (make_pass_resolve_sw_modes): Likewise.
9472
9473 2018-01-08  Sebastian Huber  <sebastian.huber@embedded-brains.de>
9474
9475         * config/nios2/nios2.h (nios2_section_threshold): Guard by not
9476         USED_FOR_TARGET.
9477
9478 2018-01-08  Richard Biener  <rguenther@suse.de>
9479
9480         PR middle-end/83580
9481         * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
9482
9483 2018-01-08  Richard Biener  <rguenther@suse.de>
9484
9485         PR middle-end/83517
9486         * match.pd ((t * 2) / 2) -> t): Add missing :c.
9487
9488 2018-01-06  Aldy Hernandez  <aldyh@redhat.com>
9489
9490         PR middle-end/81897
9491         * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
9492         basic blocks with a small number of successors.
9493         (convert_control_dep_chain_into_preds): Improve handling of
9494         forwarder blocks.
9495         (dump_predicates): Split apart into...
9496         (dump_pred_chain): ...here...
9497         (dump_pred_info): ...and here.
9498         (can_one_predicate_be_invalidated_p): Add debugging printfs.
9499         (can_chain_union_be_invalidated_p): Improve check for invalidation
9500         of paths.
9501         (uninit_uses_cannot_happen): Avoid unnecessary if
9502         convert_control_dep_chain_into_preds yielded nothing.
9503
9504 2018-01-06  Martin Sebor  <msebor@redhat.com>
9505
9506         PR tree-optimization/83640
9507         * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
9508         subtracting negative offset from size.
9509         (builtin_access::overlap): Adjust offset bounds of the access to fall
9510         within the size of the object if possible.
9511
9512 2018-01-06  Richard Sandiford  <richard.sandiford@linaro.org>
9513
9514         PR rtl-optimization/83699
9515         * expmed.c (extract_bit_field_1): Restrict the vector usage of
9516         extract_bit_field_as_subreg to cases in which the extracted
9517         value is also a vector.
9518
9519         * lra-constraints.c (process_alt_operands): Test for the equivalence
9520         substitutions when detecting a possible reload cycle.
9521
9522 2018-01-06  Jakub Jelinek  <jakub@redhat.com>
9523
9524         PR debug/83480
9525         * toplev.c (process_options): Don't enable debug_nonbind_markers_p
9526         by default if flag_selective_schedling{,2}.  Formatting fixes.
9527
9528         PR rtl-optimization/83682
9529         * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
9530         if it has non-VECTOR_MODE element mode.
9531         (vec_duplicate_p): Likewise.
9532
9533         PR middle-end/83694
9534         * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
9535         and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
9536
9537 2018-01-05  Jakub Jelinek  <jakub@redhat.com>
9538
9539         PR target/83604
9540         * config/i386/i386-builtin.def
9541         (__builtin_ia32_vgf2p8affineinvqb_v64qi,
9542         __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
9543         Require also OPTION_MASK_ISA_AVX512F in addition to
9544         OPTION_MASK_ISA_GFNI.
9545         (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
9546         __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
9547         OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
9548         to OPTION_MASK_ISA_GFNI.
9549         (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
9550         OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
9551         OPTION_MASK_ISA_AVX512BW.
9552         (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
9553         OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
9554         addition to OPTION_MASK_ISA_GFNI.
9555         (__builtin_ia32_vgf2p8affineinvqb_v16qi,
9556         __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
9557         Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
9558         to OPTION_MASK_ISA_GFNI.
9559         * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
9560         a requirement for all ISAs rather than any of them with a few
9561         exceptions.
9562         (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
9563         processing.
9564         (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
9565         bitmasks to be enabled with 3 exceptions, instead of requiring any
9566         enabled ISA with lots of exceptions.
9567         * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
9568         vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
9569         Change avx512bw in isa attribute to avx512f.
9570         * config/i386/sgxintrin.h: Add license boilerplate.
9571         * config/i386/vaesintrin.h: Likewise.  Fix macro spelling __AVX512F
9572         to __AVX512F__ and __AVX512VL to __AVX512VL__.
9573         (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
9574         _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
9575         defined.
9576         * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
9577         _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
9578         temporarily sse2 rather than sse if not enabled already.
9579
9580         PR target/83604
9581         * config/i386/sse.md (VI248_VLBW): Rename to ...
9582         (VI248_AVX512VL): ... this.  Don't guard V32HI with TARGET_AVX512BW.
9583         (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
9584         vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
9585         vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
9586         vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
9587         mode iterator instead of VI248_VLBW.
9588
9589 2018-01-05  Jan Hubicka  <hubicka@ucw.cz>
9590
9591         * ipa-fnsummary.c (record_modified_bb_info): Add OP.
9592         (record_modified): Skip clobbers; add debug output.
9593         (param_change_prob): Use sreal frequencies.
9594
9595 2018-01-05  Richard Sandiford  <richard.sandiford@linaro.org>
9596
9597         * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
9598         punt for user-aligned variables.
9599
9600 2018-01-05  Richard Sandiford  <richard.sandiford@linaro.org>
9601
9602         * tree-chrec.c (chrec_contains_symbols): Return true for
9603         POLY_INT_CST.
9604
9605 2018-01-05  Sudakshina Das  <sudi.das@arm.com>
9606
9607         PR target/82439
9608         * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
9609         of (x|y) == x for BICS pattern.
9610
9611 2018-01-05  Jakub Jelinek  <jakub@redhat.com>
9612
9613         PR tree-optimization/83605
9614         * gimple-ssa-strength-reduction.c: Include tree-eh.h.
9615         (find_candidates_dom_walker::before_dom_children): Ignore stmts that
9616         can throw.
9617
9618 2018-01-05  Sebastian Huber  <sebastian.huber@embedded-brains.de>
9619
9620         * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
9621         * config/epiphany/rtems.h: New file.
9622
9623 2018-01-04  Jakub Jelinek  <jakub@redhat.com>
9624             Uros Bizjak  <ubizjak@gmail.com>
9625
9626         PR target/83554
9627         * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
9628         QIreg_operand instead of register_operand predicate.
9629         * config/i386/i386.c (ix86_rop_should_change_byte_p,
9630         set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
9631         comments instead of -fmitigate[-_]rop.
9632
9633 2018-01-04  Rainer Orth  <ro@CeBiTec.Uni-Bielefeld.DE>
9634
9635         PR bootstrap/81926
9636         * cgraphunit.c (symbol_table::compile): Switch to text_section
9637         before calling assembly_start debug hook.
9638         * run-rtl-passes.c (run_rtl_passes): Likewise.
9639         Include output.h.
9640
9641 2018-01-04  Richard Sandiford  <richard.sandiford@linaro.org>
9642
9643         * tree-vrp.c (extract_range_from_binary_expr_1): Check
9644         range_int_cst_p rather than !symbolic_range_p before calling
9645         extract_range_from_multiplicative_op_1.
9646
9647 2018-01-04  Jeff Law  <law@redhat.com>
9648
9649         * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
9650         redundant test in assertion.
9651
9652 2018-01-04  Richard Sandiford  <richard.sandiford@linaro.org>
9653
9654         * doc/rtl.texi: Document machine_mode wrapper classes.
9655
9656 2018-01-04  Richard Sandiford  <richard.sandiford@linaro.org>
9657
9658         * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
9659         using tree_to_uhwi.
9660
9661 2018-01-04  Richard Sandiford  <richard.sandiford@linaro.org>
9662
9663         * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
9664         the VEC_PERM_EXPR fold to fail.
9665
9666 2018-01-04  Jakub Jelinek  <jakub@redhat.com>
9667
9668         PR debug/83585
9669         * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
9670         to switched_sections.
9671
9672 2018-01-04  Richard Sandiford  <richard.sandiford@linaro.org>
9673
9674         PR target/83680
9675         * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
9676         test for d.testing.
9677
9678 2018-01-04  Peter Bergner  <bergner@vnet.ibm.com>
9679
9680         PR target/83387
9681         * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
9682         allow arguments in FP registers if TARGET_HARD_FLOAT is false.
9683
9684 2018-01-04  Jakub Jelinek  <jakub@redhat.com>
9685
9686         PR debug/83666
9687         * cfgexpand.c (expand_debug_expr) <case BIT_FIELD_REF>: Punt if mode
9688         is BLKmode and bitpos not zero or mode change is needed.
9689
9690 2018-01-04  Richard Sandiford  <richard.sandiford@linaro.org>
9691
9692         PR target/83675
9693         * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
9694         TARGET_VIS2.
9695
9696 2018-01-04  Uros Bizjak  <ubizjak@gmail.com>
9697
9698         PR target/83628
9699         * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
9700         instead of MULT rtx.  Update all corresponding splitters.
9701         (*saddl_se): Ditto.
9702         (*ssub<modesuffix>): Ditto.
9703         (*ssubl_se): Ditto.
9704         (*cmp_sadd_di): Update split patterns.
9705         (*cmp_sadd_si): Ditto.
9706         (*cmp_sadd_sidi): Ditto.
9707         (*cmp_ssub_di): Ditto.
9708         (*cmp_ssub_si): Ditto.
9709         (*cmp_ssub_sidi): Ditto.
9710         * config/alpha/predicates.md (const23_operand): New predicate.
9711         * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
9712         Look for ASHIFT, not MULT inner operand.
9713         (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
9714
9715 2018-01-04  Martin Liska  <mliska@suse.cz>
9716
9717         PR gcov-profile/83669
9718         * gcov.c (output_intermediate_file): Add version to intermediate
9719         gcov file.
9720         * doc/gcov.texi: Document new field 'version' in intermediate
9721         file format. Fix location of '-k' option of gcov command.
9722
9723 2018-01-04  Martin Liska  <mliska@suse.cz>
9724
9725         PR ipa/82352
9726         * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
9727
9728 2018-01-04  Jakub Jelinek  <jakub@redhat.com>
9729
9730         * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
9731
9732 2018-01-03  Martin Sebor  <msebor@redhat.com>
9733
9734         PR tree-optimization/83655
9735         * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
9736         checking calls with invalid arguments.
9737
9738 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
9739
9740         * tree-vect-stmts.c (vect_get_store_rhs): New function.
9741         (vectorizable_mask_load_store): Delete.
9742         (vectorizable_call): Return false for masked loads and stores.
9743         (vectorizable_store): Handle IFN_MASK_STORE.  Use vect_get_store_rhs
9744         instead of gimple_assign_rhs1.
9745         (vectorizable_load): Handle IFN_MASK_LOAD.
9746         (vect_transform_stmt): Don't set is_store for call_vec_info_type.
9747
9748 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
9749
9750         * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
9751         split out from..,
9752         (vectorizable_mask_load_store): ...here.
9753         (vectorizable_load): ...and here.
9754
9755 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
9756
9757         * tree-vect-stmts.c (vect_build_all_ones_mask)
9758         (vect_build_zero_merge_argument): New functions, split out from...
9759         (vectorizable_load): ...here.
9760
9761 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
9762
9763         * tree-vect-stmts.c (vect_check_store_rhs): New function,
9764         split out from...
9765         (vectorizable_mask_load_store): ...here.
9766         (vectorizable_store): ...and here.
9767
9768 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
9769
9770         * tree-vect-stmts.c (vect_check_load_store_mask): New function,
9771         split out from...
9772         (vectorizable_mask_load_store): ...here.
9773
9774 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
9775
9776         * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
9777         (vect_model_store_cost): Take a vec_load_store_type instead of a
9778         vect_def_type.
9779         * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
9780         (vect_model_store_cost): Take a vec_load_store_type instead of a
9781         vect_def_type.
9782         (vectorizable_mask_load_store): Update accordingly.
9783         (vectorizable_store): Likewise.
9784         * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
9785
9786 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
9787
9788         * tree-vect-loop.c (vect_transform_loop): Stub out scalar
9789         IFN_MASK_LOAD calls here rather than...
9790         * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
9791
9792 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
9793             Alan Hayward  <alan.hayward@arm.com>
9794             David Sherwood  <david.sherwood@arm.com>
9795
9796         * expmed.c (extract_bit_field_1): For vector extracts,
9797         fall back to extract_bit_field_as_subreg if vec_extract
9798         isn't available.
9799
9800 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
9801             Alan Hayward  <alan.hayward@arm.com>
9802             David Sherwood  <david.sherwood@arm.com>
9803
9804         * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
9805         they are variable or constant sized.
9806         (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
9807         slots for constant-sized data.
9808
9809 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
9810             Alan Hayward  <alan.hayward@arm.com>
9811             David Sherwood  <david.sherwood@arm.com>
9812
9813         * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
9814         handling COND_EXPRs with boolean comparisons, try to find a better
9815         basis for the mask type than the boolean itself.
9816
9817 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
9818
9819         * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
9820         is calculated and how it can be overridden.
9821         * genmodes.c (max_bitsize_mode_any_mode): New variable.
9822         (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
9823         if defined.
9824         (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
9825         if nonzero.
9826
9827 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
9828             Alan Hayward  <alan.hayward@arm.com>
9829             David Sherwood  <david.sherwood@arm.com>
9830
9831         * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
9832         Remove the mode argument.
9833         (aarch64_simd_valid_immediate): Remove the mode and inverse
9834         arguments.
9835         * config/aarch64/iterators.md (bitsize): New iterator.
9836         * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
9837         (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
9838         * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
9839         aarch64_simd_valid_immediate.
9840         * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
9841         (aarch64_reg_or_bic_imm): Likewise.
9842         * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
9843         with an insn_type enum and msl with a modifier_type enum.
9844         Replace element_width with a scalar_mode.  Change the shift
9845         to unsigned int.  Add constructors for scalar_float_mode and
9846         scalar_int_mode elements.
9847         (aarch64_vect_float_const_representable_p): Delete.
9848         (aarch64_can_const_movi_rtx_p)
9849         (aarch64_simd_scalar_immediate_valid_for_move)
9850         (aarch64_simd_make_constant): Update call to
9851         aarch64_simd_valid_immediate.
9852         (aarch64_advsimd_valid_immediate_hs): New function.
9853         (aarch64_advsimd_valid_immediate): Likewise.
9854         (aarch64_simd_valid_immediate): Remove mode and inverse
9855         arguments.  Rewrite to use the above.  Use const_vec_duplicate_p
9856         to detect duplicated constants and use aarch64_float_const_zero_rtx_p
9857         and aarch64_float_const_representable_p on the result.
9858         (aarch64_output_simd_mov_immediate): Remove mode argument.
9859         Update call to aarch64_simd_valid_immediate and use of
9860         simd_immediate_info.
9861         (aarch64_output_scalar_simd_mov_immediate): Update call
9862         accordingly.
9863
9864 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
9865             Alan Hayward  <alan.hayward@arm.com>
9866             David Sherwood  <david.sherwood@arm.com>
9867
9868         * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
9869         (mode_nunits): Likewise CONST_MODE_NUNITS.
9870         * machmode.def (ADJUST_NUNITS): Document.
9871         * genmodes.c (mode_data::need_nunits_adj): New field.
9872         (blank_mode): Update accordingly.
9873         (adj_nunits): New variable.
9874         (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
9875         parameter.
9876         (emit_mode_size_inline): Set need_bytesize_adj for all modes
9877         listed in adj_nunits.
9878         (emit_mode_nunits_inline): Set need_nunits_adj for all modes
9879         listed in adj_nunits.  Don't emit case statements for such modes.
9880         (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
9881         and CONST_MODE_PRECISION.  Make CONST_MODE_SIZE expand to
9882         nothing if adj_nunits is nonnull.
9883         (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
9884         (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
9885         (emit_mode_fbit): Update use of print_maybe_const_decl.
9886         (emit_move_size): Likewise.  Treat the array as non-const
9887         if adj_nunits.
9888         (emit_mode_adjustments): Handle adj_nunits.
9889
9890 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
9891
9892         * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
9893         * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
9894         (VECTOR_MODES): Use it.
9895         (make_vector_modes): Take the prefix as an argument.
9896
9897 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
9898             Alan Hayward  <alan.hayward@arm.com>
9899             David Sherwood  <david.sherwood@arm.com>
9900
9901         * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
9902         * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
9903         for MODE_VECTOR_BOOL.
9904         * machmode.def (VECTOR_BOOL_MODE): Document.
9905         * genmodes.c (VECTOR_BOOL_MODE): New macro.
9906         (make_vector_bool_mode): New function.
9907         (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
9908         MODE_VECTOR_BOOL.
9909         * lto-streamer-in.c (lto_input_mode_table): Likewise.
9910         * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
9911         Likewise.
9912         * stor-layout.c (int_mode_for_mode): Likewise.
9913         * tree.c (build_vector_type_for_mode): Likewise.
9914         * varasm.c (output_constant_pool_2): Likewise.
9915         * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
9916         CONSTM1_RTX (BImode) are the same thing.  Initialize const_tiny_rtx
9917         for MODE_VECTOR_BOOL.
9918         * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
9919         of mode class checks.
9920         * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
9921         instead of a list of mode class checks.
9922         (expand_vector_scalar_condition): Likewise.
9923         (type_for_widest_vector_mode): Handle BImode as an inner mode.
9924
9925 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
9926             Alan Hayward  <alan.hayward@arm.com>
9927             David Sherwood  <david.sherwood@arm.com>
9928
9929         * machmode.h (mode_size): Change from unsigned short to
9930         poly_uint16_pod.
9931         (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
9932         (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
9933         or if measurement_type is not polynomial.
9934         (fixed_size_mode::includes_p): Check for constant-sized modes.
9935         * genmodes.c (emit_mode_size_inline): Make mode_size_inline
9936         return a poly_uint16 rather than an unsigned short.
9937         (emit_mode_size): Change the type of mode_size from unsigned short
9938         to poly_uint16_pod.  Use ZERO_COEFFS for the initializer.
9939         (emit_mode_adjustments): Cope with polynomial vector sizes.
9940         * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
9941         for GET_MODE_SIZE.
9942         * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
9943         for GET_MODE_SIZE.
9944         * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
9945         * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
9946         * caller-save.c (setup_save_areas): Likewise.
9947         (replace_reg_with_saved_mem): Likewise.
9948         * calls.c (emit_library_call_value_1): Likewise.
9949         * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
9950         * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
9951         (gen_lowpart_for_combine): Likewise.
9952         * convert.c (convert_to_integer_1): Likewise.
9953         * cse.c (equiv_constant, cse_insn): Likewise.
9954         * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
9955         (cselib_subst_to_values): Likewise.
9956         * dce.c (word_dce_process_block): Likewise.
9957         * df-problems.c (df_word_lr_mark_ref): Likewise.
9958         * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
9959         * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
9960         (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
9961         (rtl_for_decl_location): Likewise.
9962         * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
9963         * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
9964         * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
9965         (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
9966         (expand_expr_real_1): Likewise.
9967         * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
9968         (pad_below): Likewise.
9969         * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
9970         * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
9971         * ira.c (get_subreg_tracking_sizes): Likewise.
9972         * ira-build.c (ira_create_allocno_objects): Likewise.
9973         * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
9974         (ira_sort_regnos_for_alter_reg): Likewise.
9975         * ira-costs.c (record_operand_costs): Likewise.
9976         * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
9977         (resolve_simple_move): Likewise.
9978         * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
9979         (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
9980         (lra_constraints): Likewise.
9981         (CONST_POOL_OK_P): Reject variable-sized modes.
9982         * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
9983         (add_pseudo_to_slot, lra_spill): Likewise.
9984         * omp-low.c (omp_clause_aligned_alignment): Likewise.
9985         * optabs-query.c (get_best_extraction_insn): Likewise.
9986         * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
9987         * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
9988         (expand_mult_highpart, valid_multiword_target_p): Likewise.
9989         * recog.c (offsettable_address_addr_space_p): Likewise.
9990         * regcprop.c (maybe_mode_change): Likewise.
9991         * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
9992         * regrename.c (build_def_use): Likewise.
9993         * regstat.c (dump_reg_info): Likewise.
9994         * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
9995         (find_reloads, find_reloads_subreg_address): Likewise.
9996         * reload1.c (eliminate_regs_1): Likewise.
9997         * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
9998         * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
9999         (simplify_binary_operation_1, simplify_subreg): Likewise.
10000         * targhooks.c (default_function_arg_padding): Likewise.
10001         (default_hard_regno_nregs, default_class_max_nregs): Likewise.
10002         * tree-cfg.c (verify_gimple_assign_binary): Likewise.
10003         (verify_gimple_assign_ternary): Likewise.
10004         * tree-inline.c (estimate_move_cost): Likewise.
10005         * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
10006         * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
10007         (get_address_cost_ainc): Likewise.
10008         * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
10009         (vect_supportable_dr_alignment): Likewise.
10010         * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
10011         (vectorizable_reduction): Likewise.
10012         * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
10013         (vectorizable_operation, vectorizable_load): Likewise.
10014         * tree.c (build_same_sized_truth_vector_type): Likewise.
10015         * valtrack.c (cleanup_auto_inc_dec): Likewise.
10016         * var-tracking.c (emit_note_insn_var_location): Likewise.
10017         * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
10018         (ADDR_VEC_ALIGN): Likewise.
10019
10020 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10021             Alan Hayward  <alan.hayward@arm.com>
10022             David Sherwood  <david.sherwood@arm.com>
10023
10024         * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
10025         unsigned short.
10026         (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
10027         or if measurement_type is polynomial.
10028         * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
10029         * combine.c (make_extraction): Likewise.
10030         * dse.c (find_shift_sequence): Likewise.
10031         * dwarf2out.c (mem_loc_descriptor): Likewise.
10032         * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
10033         (extract_bit_field, extract_low_bits): Likewise.
10034         * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
10035         (optimize_bitfield_assignment_op, expand_assignment): Likewise.
10036         (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
10037         * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
10038         * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
10039         * reload.c (find_reloads): Likewise.
10040         * reload1.c (alter_reg): Likewise.
10041         * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
10042         * targhooks.c (default_secondary_memory_needed_mode): Likewise.
10043         * tree-if-conv.c (predicate_mem_writes): Likewise.
10044         * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
10045         * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
10046         * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
10047         * valtrack.c (dead_debug_insert_temp): Likewise.
10048         * varasm.c (mergeable_constant_section): Likewise.
10049         * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
10050
10051 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10052             Alan Hayward  <alan.hayward@arm.com>
10053             David Sherwood  <david.sherwood@arm.com>
10054
10055         * expr.c (expand_assignment): Cope with polynomial mode sizes
10056         when assigning to a CONCAT.
10057
10058 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10059             Alan Hayward  <alan.hayward@arm.com>
10060             David Sherwood  <david.sherwood@arm.com>
10061
10062         * machmode.h (mode_precision): Change from unsigned short to
10063         poly_uint16_pod.
10064         (mode_to_precision): Return a poly_uint16 rather than an unsigned
10065         short.
10066         (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
10067         or if measurement_type is not polynomial.
10068         (HWI_COMPUTABLE_MODE_P): Turn into a function.  Optimize the case
10069         in which the mode is already known to be a scalar_int_mode.
10070         * genmodes.c (emit_mode_precision): Change the type of mode_precision
10071         from unsigned short to poly_uint16_pod.  Use ZERO_COEFFS for the
10072         initializer.
10073         * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
10074         for GET_MODE_PRECISION.
10075         * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
10076         for GET_MODE_PRECISION.
10077         * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
10078         as polynomial.
10079         (try_combine, find_split_point, combine_simplify_rtx): Likewise.
10080         (expand_field_assignment, make_extraction): Likewise.
10081         (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
10082         (get_last_value): Likewise.
10083         * convert.c (convert_to_integer_1): Likewise.
10084         * cse.c (cse_insn): Likewise.
10085         * expr.c (expand_expr_real_1): Likewise.
10086         * lra-constraints.c (simplify_operand_subreg): Likewise.
10087         * optabs-query.c (can_atomic_load_p): Likewise.
10088         * optabs.c (expand_atomic_load): Likewise.
10089         (expand_atomic_store): Likewise.
10090         * ree.c (combine_reaching_defs): Likewise.
10091         * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
10092         * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
10093         * tree.h (type_has_mode_precision_p): Likewise.
10094         * ubsan.c (instrument_si_overflow): Likewise.
10095
10096 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10097             Alan Hayward  <alan.hayward@arm.com>
10098             David Sherwood  <david.sherwood@arm.com>
10099
10100         * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
10101         polynomial numbers of units.
10102         (SET_TYPE_VECTOR_SUBPARTS): Likewise.
10103         (valid_vector_subparts_p): New function.
10104         (build_vector_type): Remove temporary shim and take the number
10105         of units as a poly_uint64 rather than an int.
10106         (build_opaque_vector_type): Take the number of units as a
10107         poly_uint64 rather than an int.
10108         * tree.c (build_vector_from_ctor): Handle polynomial
10109         TYPE_VECTOR_SUBPARTS.
10110         (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
10111         (uniform_vector_p, vector_type_mode, build_vector): Likewise.
10112         (build_vector_from_val): If the number of units is variable,
10113         use build_vec_duplicate_cst for constant operands and
10114         VEC_DUPLICATE_EXPR otherwise.
10115         (make_vector_type): Remove temporary is_constant ().
10116         (build_vector_type, build_opaque_vector_type): Take the number of
10117         units as a poly_uint64 rather than an int.
10118         (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
10119         VECTOR_CST_NELTS.
10120         * cfgexpand.c (expand_debug_expr): Likewise.
10121         * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
10122         (store_constructor, expand_expr_real_1): Likewise.
10123         (const_scalar_mask_from_tree): Likewise.
10124         * fold-const-call.c (fold_const_reduction): Likewise.
10125         * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
10126         (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
10127         (native_encode_vector, vec_cst_ctor_to_array): Likewise.
10128         (fold_relational_const): Likewise.
10129         (native_interpret_vector): Likewise.  Change the size from an
10130         int to an unsigned int.
10131         * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
10132         TYPE_VECTOR_SUBPARTS.
10133         (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
10134         (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
10135         duplicating a non-constant operand into a variable-length vector.
10136         * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
10137         TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
10138         * ipa-icf.c (sem_variable::equals): Likewise.
10139         * match.pd: Likewise.
10140         * omp-simd-clone.c (simd_clone_subparts): Likewise.
10141         * print-tree.c (print_node): Likewise.
10142         * stor-layout.c (layout_type): Likewise.
10143         * targhooks.c (default_builtin_vectorization_cost): Likewise.
10144         * tree-cfg.c (verify_gimple_comparison): Likewise.
10145         (verify_gimple_assign_binary): Likewise.
10146         (verify_gimple_assign_ternary): Likewise.
10147         (verify_gimple_assign_single): Likewise.
10148         * tree-pretty-print.c (dump_generic_node): Likewise.
10149         * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
10150         (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
10151         * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
10152         (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
10153         (vect_shift_permute_load_chain): Likewise.
10154         * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
10155         (expand_vector_condition, optimize_vector_constructor): Likewise.
10156         (lower_vec_perm, get_compute_type): Likewise.
10157         * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
10158         (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
10159         * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
10160         (vect_recog_mask_conversion_pattern): Likewise.
10161         * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
10162         (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
10163         * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
10164         (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
10165         (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
10166         (vectorizable_shift, vectorizable_operation, vectorizable_store)
10167         (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
10168         (supportable_widening_operation): Likewise.
10169         (supportable_narrowing_operation): Likewise.
10170         * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
10171         Likewise.
10172         * varasm.c (output_constant): Likewise.
10173
10174 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10175             Alan Hayward  <alan.hayward@arm.com>
10176             David Sherwood  <david.sherwood@arm.com>
10177
10178         * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
10179         so that both the length == 3 and length != 3 cases set up their
10180         own permute vectors.  Add comments explaining why we know the
10181         number of elements is constant.
10182         (vect_permute_load_chain): Likewise.
10183
10184 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10185             Alan Hayward  <alan.hayward@arm.com>
10186             David Sherwood  <david.sherwood@arm.com>
10187
10188         * machmode.h (mode_nunits): Change from unsigned char to
10189         poly_uint16_pod.
10190         (ONLY_FIXED_SIZE_MODES): New macro.
10191         (pod_mode::measurement_type, scalar_int_mode::measurement_type)
10192         (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
10193         (complex_mode::measurement_type, fixed_size_mode::measurement_type):
10194         New typedefs.
10195         (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
10196         (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
10197         or if measurement_type is not polynomial.
10198         * genmodes.c (ZERO_COEFFS): New macro.
10199         (emit_mode_nunits_inline): Make mode_nunits_inline return a
10200         poly_uint16.
10201         (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
10202         Use ZERO_COEFFS when emitting initializers.
10203         * data-streamer.h (bp_pack_poly_value): New function.
10204         (bp_unpack_poly_value): Likewise.
10205         * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
10206         for GET_MODE_NUNITS.
10207         * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
10208         for GET_MODE_NUNITS.
10209         * tree.c (make_vector_type): Remove temporary shim and make
10210         the real function take the number of units as a poly_uint64
10211         rather than an int.
10212         (build_vector_type_for_mode): Handle polynomial nunits.
10213         * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
10214         * emit-rtl.c (const_vec_series_p_1): Likewise.
10215         (gen_rtx_CONST_VECTOR): Likewise.
10216         * fold-const.c (test_vec_duplicate_folding): Likewise.
10217         * genrecog.c (validate_pattern): Likewise.
10218         * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
10219         * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
10220         * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
10221         (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
10222         (expand_vec_cond_expr, expand_mult_highpart): Likewise.
10223         * rtlanal.c (subreg_get_info): Likewise.
10224         * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
10225         (vect_grouped_load_supported): Likewise.
10226         * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
10227         * tree-vect-loop.c (have_whole_vector_shift): Likewise.
10228         * simplify-rtx.c (simplify_unary_operation_1): Likewise.
10229         (simplify_const_unary_operation, simplify_binary_operation_1)
10230         (simplify_const_binary_operation, simplify_ternary_operation)
10231         (test_vector_ops_duplicate, test_vector_ops): Likewise.
10232         (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
10233         instead of CONST_VECTOR_NUNITS.
10234         * varasm.c (output_constant_pool_2): Likewise.
10235         * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
10236         explicit-encoded elements in the XVEC for variable-length vectors.
10237
10238 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10239
10240         * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
10241
10242 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10243             Alan Hayward  <alan.hayward@arm.com>
10244             David Sherwood  <david.sherwood@arm.com>
10245
10246         * coretypes.h (fixed_size_mode): Declare.
10247         (fixed_size_mode_pod): New typedef.
10248         * builtins.h (target_builtins::x_apply_args_mode)
10249         (target_builtins::x_apply_result_mode): Change type to
10250         fixed_size_mode_pod.
10251         * builtins.c (apply_args_size, apply_result_size, result_vector)
10252         (expand_builtin_apply_args_1, expand_builtin_apply)
10253         (expand_builtin_return): Update accordingly.
10254
10255 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10256
10257         * cse.c (hash_rtx_cb): Hash only the encoded elements.
10258         * cselib.c (cselib_hash_rtx): Likewise.
10259         * expmed.c (make_tree): Build VECTOR_CSTs directly from the
10260         CONST_VECTOR encoding.
10261
10262 2018-01-03  Jakub Jelinek  <jakub@redhat.com>
10263             Jeff Law  <law@redhat.com>
10264
10265         PR target/83641
10266         * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
10267         noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
10268         only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
10269         and add REG_CFA_ADJUST_CFA notes in that case to both insns.
10270
10271         PR target/83641
10272         * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
10273         explicitly probe *sp in a noreturn function if there were any callee
10274         register saves or frame pointer is needed.
10275
10276 2018-01-03  Jakub Jelinek  <jakub@redhat.com>
10277
10278         PR debug/83621
10279         * cfgexpand.c (expand_debug_expr): Return NULL if mode is
10280         BLKmode for ternary, binary or unary expressions.
10281
10282         PR debug/83645
10283         * var-tracking.c (delete_vta_debug_insn): New inline function.
10284         (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
10285         insns from get_insns () to NULL instead of each bb separately.
10286         Use delete_vta_debug_insn.  No longer static.
10287         (vt_debug_insns_local, variable_tracking_main_1): Adjust
10288         delete_vta_debug_insns callers.
10289         * rtl.h (delete_vta_debug_insns): Declare.
10290         * final.c (rest_of_handle_final): Call delete_vta_debug_insns
10291         instead of variable_tracking_main.
10292
10293 2018-01-03  Martin Sebor  <msebor@redhat.com>
10294
10295         PR tree-optimization/83603
10296         * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
10297         arguments past the endof the argument list in functions declared
10298         without a prototype.
10299         * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
10300         Avoid checking when arguments are null.
10301
10302 2018-01-03  Martin Sebor  <msebor@redhat.com>
10303
10304         PR c/83559
10305         * doc/extend.texi (attribute const): Fix a typo.
10306         * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
10307         issuing -Wsuggest-attribute for void functions.
10308
10309 2018-01-03  Martin Sebor  <msebor@redhat.com>
10310
10311         * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
10312         offset_int::from instead of wide_int::to_shwi.
10313         (maybe_diag_overlap): Remove assertion.
10314         Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
10315         * gimple-ssa-sprintf.c (format_directive): Same.
10316         (parse_directive): Same.
10317         (sprintf_dom_walker::compute_format_length): Same.
10318         (try_substitute_return_value): Same.
10319
10320 2018-01-03  Jeff Law  <law@redhat.com>
10321
10322         PR middle-end/83654
10323         * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
10324         non-constant residual for zero at runtime and avoid probing in
10325         that case.  Reorganize code for trailing problem to mirror handling
10326         of the residual.
10327
10328 2018-01-03  Prathamesh Kulkarni  <prathamesh.kulkarni@linaro.org>
10329
10330         PR tree-optimization/83501
10331         * tree-ssa-strlen.c (get_string_cst): New.
10332         (handle_char_store): Call get_string_cst.
10333
10334 2018-01-03  Martin Liska  <mliska@suse.cz>
10335
10336         PR tree-optimization/83593
10337         * tree-ssa-strlen.c: Include tree-cfg.h.
10338         (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
10339         (strlen_dom_walker): Add new member variable m_cleanup_cfg.
10340         (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
10341         to false.
10342         (strlen_dom_walker::before_dom_children): Call
10343         gimple_purge_dead_eh_edges. Dump tranformation with details
10344         dump flags.
10345         (strlen_dom_walker::before_dom_children): Update call by adding
10346         new argument cleanup_eh.
10347         (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
10348
10349 2018-01-03  Martin Liska  <mliska@suse.cz>
10350
10351         PR ipa/83549
10352         * cif-code.def (VARIADIC_THUNK): New enum value.
10353         * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
10354         thunks.
10355
10356 2018-01-03  Jan Beulich  <jbeulich@suse.com>
10357
10358         * sse.md (mov<mode>_internal): Tighten condition for when to use
10359         vmovdqu<ssescalarsize> for TI and OI modes.
10360
10361 2018-01-03  Jakub Jelinek  <jakub@redhat.com>
10362
10363         Update copyright years.
10364
10365 2018-01-03  Martin Liska  <mliska@suse.cz>
10366
10367         PR ipa/83594
10368         * ipa-visibility.c (function_and_variable_visibility): Skip
10369         functions with noipa attribure.
10370
10371 2018-01-03  Jakub Jelinek  <jakub@redhat.com>
10372
10373         * gcc.c (process_command): Update copyright notice dates.
10374         * gcov-dump.c (print_version): Ditto.
10375         * gcov.c (print_version): Ditto.
10376         * gcov-tool.c (print_version): Ditto.
10377         * gengtype.c (create_file): Ditto.
10378         * doc/cpp.texi: Bump @copying's copyright year.
10379         * doc/cppinternals.texi: Ditto.
10380         * doc/gcc.texi: Ditto.
10381         * doc/gccint.texi: Ditto.
10382         * doc/gcov.texi: Ditto.
10383         * doc/install.texi: Ditto.
10384         * doc/invoke.texi: Ditto.
10385
10386 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10387
10388         * vector-builder.h (vector_builder::m_full_nelts): Change from
10389         unsigned int to poly_uint64.
10390         (vector_builder::full_nelts): Update prototype accordingly.
10391         (vector_builder::new_vector): Likewise.
10392         (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
10393         (vector_builder::operator ==): Likewise.
10394         (vector_builder::finalize): Likewise.
10395         * int-vector-builder.h (int_vector_builder::int_vector_builder):
10396         Take the number of elements as a poly_uint64 rather than an
10397         unsigned int.
10398         * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
10399         from unsigned int to poly_uint64.
10400         (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
10401         (vec_perm_indices::new_vector): Likewise.
10402         (vec_perm_indices::length): Likewise.
10403         (vec_perm_indices::nelts_per_input): Likewise.
10404         (vec_perm_indices::input_nelts): Likewise.
10405         * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
10406         number of elements per input as a poly_uint64 rather than an
10407         unsigned int.  Use the original encoding for variable-length
10408         vectors, rather than clamping each individual element.
10409         For the second and subsequent elements in each pattern,
10410         clamp the step and base before clamping their sum.
10411         (vec_perm_indices::series_p): Handle polynomial element counts.
10412         (vec_perm_indices::all_in_range_p): Likewise.
10413         (vec_perm_indices_to_tree): Likewise.
10414         (vec_perm_indices_to_rtx): Likewise.
10415         * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
10416         * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
10417         (tree_vector_builder::new_binary_operation): Handle polynomial
10418         element counts.  Return false if we need to know the number
10419         of elements at compile time.
10420         * fold-const.c (fold_vec_perm): Punt if the number of elements
10421         isn't known at compile time.
10422
10423 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10424
10425         * vec-perm-indices.h (vec_perm_builder): Change element type
10426         from HOST_WIDE_INT to poly_int64.
10427         (vec_perm_indices::element_type): Update accordingly.
10428         (vec_perm_indices::clamp): Handle polynomial element_types.
10429         * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
10430         (vec_perm_indices::all_in_range_p): Likewise.
10431         (tree_to_vec_perm_builder): Check for poly_int64 trees rather
10432         than shwi trees.
10433         * vector-builder.h (vector_builder::stepped_sequence_p): Handle
10434         polynomial vec_perm_indices element types.
10435         * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
10436         * fold-const.c (fold_vec_perm): Likewise.
10437         * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
10438         * tree-vect-generic.c (lower_vec_perm): Likewise.
10439         * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
10440         * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
10441         element type to HOST_WIDE_INT.
10442
10443 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10444             Alan Hayward  <alan.hayward@arm.com>
10445             David Sherwood  <david.sherwood@arm.com>
10446
10447         * alias.c (addr_side_effect_eval): Take the size as a poly_int64
10448         rather than an int.  Use plus_constant.
10449         (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
10450         Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
10451
10452 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10453             Alan Hayward  <alan.hayward@arm.com>
10454             David Sherwood  <david.sherwood@arm.com>
10455
10456         * calls.c (emit_call_1, expand_call): Change struct_value_size from
10457         a HOST_WIDE_INT to a poly_int64.
10458
10459 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10460             Alan Hayward  <alan.hayward@arm.com>
10461             David Sherwood  <david.sherwood@arm.com>
10462
10463         * calls.c (load_register_parameters): Cope with polynomial
10464         mode sizes.  Require a constant size for BLKmode parameters
10465         that aren't described by a PARALLEL.  If BLOCK_REG_PADDING
10466         forces a parameter to be padded at the lsb end in order to
10467         fill a complete number of words, require the parameter size
10468         to be ordered wrt UNITS_PER_WORD.
10469
10470 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10471             Alan Hayward  <alan.hayward@arm.com>
10472             David Sherwood  <david.sherwood@arm.com>
10473
10474         * reload1.c (spill_stack_slot_width): Change element type
10475         from unsigned int to poly_uint64_pod.
10476         (alter_reg): Treat mode sizes as polynomial.
10477
10478 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10479             Alan Hayward  <alan.hayward@arm.com>
10480             David Sherwood  <david.sherwood@arm.com>
10481
10482         * reload.c (complex_word_subreg_p): New function.
10483         (reload_inner_reg_of_subreg, push_reload): Use it.
10484
10485 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10486             Alan Hayward  <alan.hayward@arm.com>
10487             David Sherwood  <david.sherwood@arm.com>
10488
10489         * lra-constraints.c (process_alt_operands): Reject matched
10490         operands whose sizes aren't ordered.
10491         (match_reload): Refer to this check here.
10492
10493 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10494             Alan Hayward  <alan.hayward@arm.com>
10495             David Sherwood  <david.sherwood@arm.com>
10496
10497         * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
10498         that the mode size is in the set {1, 2, 4, 8, 16}.
10499
10500 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10501             Alan Hayward  <alan.hayward@arm.com>
10502             David Sherwood  <david.sherwood@arm.com>
10503
10504         * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
10505         Use plus_constant instead of gen_rtx_PLUS.
10506
10507 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10508             Alan Hayward  <alan.hayward@arm.com>
10509             David Sherwood  <david.sherwood@arm.com>
10510
10511         * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
10512         * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
10513         * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
10514         * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
10515         * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
10516         * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
10517         * config/i386/i386-protos.h (ix86_push_rounding): Declare.
10518         * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
10519         * config/i386/i386.c (ix86_push_rounding): ...this new function.
10520         * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
10521         a poly_int64.
10522         * config/m32c/m32c.c (m32c_push_rounding): Likewise.
10523         * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
10524         * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
10525         * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
10526         * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
10527         * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
10528         * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
10529         * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
10530         * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
10531         * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
10532         function.
10533         * expr.c (emit_move_resolve_push): Treat the input and result
10534         of PUSH_ROUNDING as a poly_int64.
10535         (emit_move_complex_push, emit_single_push_insn_1): Likewise.
10536         (emit_push_insn): Likewise.
10537         * lra-eliminations.c (mark_not_eliminable): Likewise.
10538         * recog.c (push_operand): Likewise.
10539         * reload1.c (elimination_effects): Likewise.
10540         * rtlanal.c (nonzero_bits1): Likewise.
10541         * calls.c (store_one_arg): Likewise.  Require the padding to be
10542         known at compile time.
10543
10544 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10545             Alan Hayward  <alan.hayward@arm.com>
10546             David Sherwood  <david.sherwood@arm.com>
10547
10548         * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
10549         Use plus_constant instead of gen_rtx_PLUS.
10550
10551 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10552             Alan Hayward  <alan.hayward@arm.com>
10553             David Sherwood  <david.sherwood@arm.com>
10554
10555         * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
10556         rather than an int.
10557
10558 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10559             Alan Hayward  <alan.hayward@arm.com>
10560             David Sherwood  <david.sherwood@arm.com>
10561
10562         * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
10563         instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
10564         via stack temporaries.  Treat the mode size as polynomial too.
10565
10566 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10567             Alan Hayward  <alan.hayward@arm.com>
10568             David Sherwood  <david.sherwood@arm.com>
10569
10570         * expr.c (expand_expr_real_2): When handling conversions involving
10571         unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
10572         multiplying int_size_in_bytes by BITS_PER_UNIT.  Treat GET_MODE_BISIZE
10573         as a poly_uint64 too.
10574
10575 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10576             Alan Hayward  <alan.hayward@arm.com>
10577             David Sherwood  <david.sherwood@arm.com>
10578
10579         * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
10580
10581 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10582             Alan Hayward  <alan.hayward@arm.com>
10583             David Sherwood  <david.sherwood@arm.com>
10584
10585         * combine.c (can_change_dest_mode): Handle polynomial
10586         REGMODE_NATURAL_SIZE.
10587         * expmed.c (store_bit_field_1): Likewise.
10588         * expr.c (store_constructor): Likewise.
10589         * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
10590         and polynomial REGMODE_NATURAL_SIZE.
10591         (gen_lowpart_common): Likewise.
10592         * reginfo.c (record_subregs_of_mode): Likewise.
10593         * rtlanal.c (read_modify_subreg_p): Likewise.
10594
10595 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10596             Alan Hayward  <alan.hayward@arm.com>
10597             David Sherwood  <david.sherwood@arm.com>
10598
10599         * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
10600         numbers of elements.
10601
10602 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10603             Alan Hayward  <alan.hayward@arm.com>
10604             David Sherwood  <david.sherwood@arm.com>
10605
10606         * match.pd: Cope with polynomial numbers of vector elements.
10607
10608 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10609             Alan Hayward  <alan.hayward@arm.com>
10610             David Sherwood  <david.sherwood@arm.com>
10611
10612         * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
10613         in a POINTER_PLUS_EXPR.
10614
10615 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10616             Alan Hayward  <alan.hayward@arm.com>
10617             David Sherwood  <david.sherwood@arm.com>
10618
10619         * omp-simd-clone.c (simd_clone_subparts): New function.
10620         (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
10621         (ipa_simd_modify_function_body): Likewise.
10622
10623 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10624             Alan Hayward  <alan.hayward@arm.com>
10625             David Sherwood  <david.sherwood@arm.com>
10626
10627         * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
10628         (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
10629         (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
10630         (expand_vector_condition, vector_element): Likewise.
10631         (subparts_gt): New function.
10632         (get_compute_type): Use subparts_gt.
10633         (count_type_subparts): Delete.
10634         (expand_vector_operations_1): Use subparts_gt instead of
10635         count_type_subparts.
10636
10637 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10638             Alan Hayward  <alan.hayward@arm.com>
10639             David Sherwood  <david.sherwood@arm.com>
10640
10641         * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
10642         (vect_compile_time_alias): ...this new function.  Do the calculation
10643         on poly_ints rather than trees.
10644         (vect_prune_runtime_alias_test_list): Update call accordingly.
10645
10646 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10647             Alan Hayward  <alan.hayward@arm.com>
10648             David Sherwood  <david.sherwood@arm.com>
10649
10650         * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
10651         numbers of units.
10652         (vect_schedule_slp_instance): Likewise.
10653
10654 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10655             Alan Hayward  <alan.hayward@arm.com>
10656             David Sherwood  <david.sherwood@arm.com>
10657
10658         * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
10659         constant and extern definitions for variable-length vectors.
10660         (vect_get_constant_vectors): Note that the number of units
10661         is known to be constant.
10662
10663 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10664             Alan Hayward  <alan.hayward@arm.com>
10665             David Sherwood  <david.sherwood@arm.com>
10666
10667         * tree-vect-stmts.c (vectorizable_conversion): Treat the number
10668         of units as polynomial.  Choose between WIDE and NARROW based
10669         on multiple_p.
10670
10671 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10672             Alan Hayward  <alan.hayward@arm.com>
10673             David Sherwood  <david.sherwood@arm.com>
10674
10675         * tree-vect-stmts.c (simd_clone_subparts): New function.
10676         (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
10677
10678 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10679             Alan Hayward  <alan.hayward@arm.com>
10680             David Sherwood  <david.sherwood@arm.com>
10681
10682         * tree-vect-stmts.c (vectorizable_call): Treat the number of
10683         vectors as polynomial.  Use build_index_vector for
10684         IFN_GOMP_SIMD_LANE.
10685
10686 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10687             Alan Hayward  <alan.hayward@arm.com>
10688             David Sherwood  <david.sherwood@arm.com>
10689
10690         * tree-vect-stmts.c (get_load_store_type): Treat the number of
10691         units as polynomial.  Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
10692         for variable-length vectors.
10693         (vectorizable_mask_load_store): Treat the number of units as
10694         polynomial, asserting that it is constant if the condition has
10695         already been enforced.
10696         (vectorizable_store, vectorizable_load): Likewise.
10697
10698 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10699             Alan Hayward  <alan.hayward@arm.com>
10700             David Sherwood  <david.sherwood@arm.com>
10701
10702         * tree-vect-loop.c (vectorizable_live_operation): Treat the number
10703         of units as polynomial.  Punt if we can't tell at compile time
10704         which vector contains the final result.
10705
10706 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10707             Alan Hayward  <alan.hayward@arm.com>
10708             David Sherwood  <david.sherwood@arm.com>
10709
10710         * tree-vect-loop.c (vectorizable_induction): Treat the number
10711         of units as polynomial.  Punt on SLP inductions.  Use an integer
10712         VEC_SERIES_EXPR for variable-length integer reductions.  Use a
10713         cast of such a series for variable-length floating-point
10714         reductions.
10715
10716 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10717             Alan Hayward  <alan.hayward@arm.com>
10718             David Sherwood  <david.sherwood@arm.com>
10719
10720         * tree.h (build_index_vector): Declare.
10721         * tree.c (build_index_vector): New function.
10722         * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
10723         of units as polynomial, forcibly converting it to a constant if
10724         vectorizable_reduction has already enforced the condition.
10725         (vect_create_epilog_for_reduction): Likewise.  Use build_index_vector
10726         to create a {1,2,3,...} vector.
10727         (vectorizable_reduction): Treat the number of units as polynomial.
10728         Choose vectype_in based on the largest scalar element size rather
10729         than the smallest number of units.  Enforce the restrictions
10730         relied on above.
10731
10732 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10733             Alan Hayward  <alan.hayward@arm.com>
10734             David Sherwood  <david.sherwood@arm.com>
10735
10736         * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
10737         number of units as polynomial.
10738
10739 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10740             Alan Hayward  <alan.hayward@arm.com>
10741             David Sherwood  <david.sherwood@arm.com>
10742
10743         * target.h (vector_sizes, auto_vector_sizes): New typedefs.
10744         * target.def (autovectorize_vector_sizes): Return the vector sizes
10745         by pointer, using vector_sizes rather than a bitmask.
10746         * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
10747         * targhooks.c (default_autovectorize_vector_sizes): Likewise.
10748         * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
10749         Likewise.
10750         * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
10751         * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
10752         * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
10753         * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
10754         * omp-general.c (omp_max_vf): Likewise.
10755         * omp-low.c (omp_clause_aligned_alignment): Likewise.
10756         * optabs-query.c (can_vec_mask_load_store_p): Likewise.
10757         * tree-vect-loop.c (vect_analyze_loop): Likewise.
10758         * tree-vect-slp.c (vect_slp_bb): Likewise.
10759         * doc/tm.texi: Regenerate.
10760         * tree-vectorizer.h (current_vector_size): Change from an unsigned int
10761         to a poly_uint64.
10762         * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
10763         the vector size as a poly_uint64 rather than an unsigned int.
10764         (current_vector_size): Change from an unsigned int to a poly_uint64.
10765         (get_vectype_for_scalar_type): Update accordingly.
10766         * tree.h (build_truth_vector_type): Take the size and number of
10767         units as a poly_uint64 rather than an unsigned int.
10768         (build_vector_type): Add a temporary overload that takes
10769         the number of units as a poly_uint64 rather than an unsigned int.
10770         * tree.c (make_vector_type): Likewise.
10771         (build_truth_vector_type): Take the number of units as a poly_uint64
10772         rather than an unsigned int.
10773
10774 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10775             Alan Hayward  <alan.hayward@arm.com>
10776             David Sherwood  <david.sherwood@arm.com>
10777
10778         * target.def (get_mask_mode): Take the number of units and length
10779         as poly_uint64s rather than unsigned ints.
10780         * targhooks.h (default_get_mask_mode): Update accordingly.
10781         * targhooks.c (default_get_mask_mode): Likewise.
10782         * config/i386/i386.c (ix86_get_mask_mode): Likewise.
10783         * doc/tm.texi: Regenerate.
10784
10785 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10786             Alan Hayward  <alan.hayward@arm.com>
10787             David Sherwood  <david.sherwood@arm.com>
10788
10789         * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
10790         * omp-general.c (omp_max_vf): Likewise.
10791         * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
10792         (expand_omp_simd): Handle polynomial safelen.
10793         * omp-low.c (omplow_simd_context): Add a default constructor.
10794         (omplow_simd_context::max_vf): Change from int to poly_uint64.
10795         (lower_rec_simd_input_clauses): Update accordingly.
10796         (lower_rec_input_clauses): Likewise.
10797
10798 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10799             Alan Hayward  <alan.hayward@arm.com>
10800             David Sherwood  <david.sherwood@arm.com>
10801
10802         * tree-vectorizer.h (vect_nunits_for_cost): New function.
10803         * tree-vect-loop.c (vect_model_reduction_cost): Use it.
10804         * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
10805         (vect_analyze_slp_cost): Likewise.
10806         * tree-vect-stmts.c (vect_model_store_cost): Likewise.
10807         (vect_model_load_cost): Likewise.
10808
10809 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10810             Alan Hayward  <alan.hayward@arm.com>
10811             David Sherwood  <david.sherwood@arm.com>
10812
10813         * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
10814         (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
10815         from an unsigned int * to a poly_uint64_pod *.
10816         (calculate_unrolling_factor): New function.
10817         (vect_analyze_slp_instance): Use it.  Track polynomial max_nunits.
10818
10819 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10820             Alan Hayward  <alan.hayward@arm.com>
10821             David Sherwood  <david.sherwood@arm.com>
10822
10823         * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
10824         from an unsigned int to a poly_uint64.
10825         (_loop_vec_info::slp_unrolling_factor): Likewise.
10826         (_loop_vec_info::vectorization_factor): Change from an int
10827         to a poly_uint64.
10828         (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
10829         (vect_get_num_vectors): New function.
10830         (vect_update_max_nunits, vect_vf_for_cost): Likewise.
10831         (vect_get_num_copies): Use vect_get_num_vectors.
10832         (vect_analyze_data_ref_dependences): Change max_vf from an int *
10833         to an unsigned int *.
10834         (vect_analyze_data_refs): Change min_vf from an int * to a
10835         poly_uint64 *.
10836         (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
10837         than an unsigned HOST_WIDE_INT.
10838         * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
10839         (vect_analyze_data_ref_dependence): Change max_vf from an int *
10840         to an unsigned int *.
10841         (vect_analyze_data_ref_dependences): Likewise.
10842         (vect_compute_data_ref_alignment): Handle polynomial vf.
10843         (vect_enhance_data_refs_alignment): Likewise.
10844         (vect_prune_runtime_alias_test_list): Likewise.
10845         (vect_shift_permute_load_chain): Likewise.
10846         (vect_supportable_dr_alignment): Likewise.
10847         (dependence_distance_ge_vf): Take the vectorization factor as a
10848         poly_uint64 rather than an unsigned HOST_WIDE_INT.
10849         (vect_analyze_data_refs): Change min_vf from an int * to a
10850         poly_uint64 *.
10851         * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
10852         vfm1 as a poly_uint64 rather than an int.  Make the same change
10853         for the returned bound_scalar.
10854         (vect_gen_vector_loop_niters): Handle polynomial vf.
10855         (vect_do_peeling): Likewise.  Update call to
10856         vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
10857         (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
10858         be constant.
10859         * tree-vect-loop.c (vect_determine_vectorization_factor)
10860         (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
10861         (vect_get_known_peeling_cost): Likewise.
10862         (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
10863         (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
10864         (vect_transform_loop): Likewise.  Use the lowest possible VF when
10865         updating the upper bounds of the loop.
10866         (vect_min_worthwhile_factor): Make static.  Return an unsigned int
10867         rather than an int.
10868         * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
10869         polynomial unroll factors.
10870         (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
10871         (vect_make_slp_decision): Likewise.
10872         (vect_supported_load_permutation_p): Likewise, and polynomial
10873         vf too.
10874         (vect_analyze_slp_cost): Handle polynomial vf.
10875         (vect_slp_analyze_node_operations): Likewise.
10876         (vect_slp_analyze_bb_1): Likewise.
10877         (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
10878         than an unsigned HOST_WIDE_INT.
10879         * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
10880         (vectorizable_load): Handle polynomial vf.
10881         * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
10882         a poly_uint64.
10883         (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
10884
10885 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10886             Alan Hayward  <alan.hayward@arm.com>
10887             David Sherwood  <david.sherwood@arm.com>
10888
10889         * match.pd: Handle bit operations involving three constants
10890         and try to fold one pair.
10891
10892 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
10893
10894         * tree-vect-loop-manip.c: Include gimple-fold.h.
10895         (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
10896         niters_maybe_zero parameters.  Handle other cases besides a step of 1.
10897         (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
10898         Add a path that uses a step of VF instead of 1, but disable it
10899         for now.
10900         (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
10901         and niters_no_overflow parameters.  Update calls to
10902         slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
10903         Create a new SSA name if the latter choses to use a ste other
10904         than zero, and return it via niters_vector_mult_vf_var.
10905         * tree-vect-loop.c (vect_transform_loop): Update calls to
10906         vect_do_peeling, vect_gen_vector_loop_niters and
10907         slpeel_make_loop_iterate_ntimes.
10908         * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
10909         (vect_gen_vector_loop_niters): Update declarations after above changes.
10910
10911 2018-01-02  Michael Meissner  <meissner@linux.vnet.ibm.com>
10912
10913         * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
10914         128-bit round to integer instructions.
10915         (ceil<mode>2): Likewise.
10916         (btrunc<mode>2): Likewise.
10917         (round<mode>2): Likewise.
10918
10919 2018-01-02  Aaron Sawdey  <acsawdey@linux.vnet.ibm.com>
10920
10921         * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
10922         unaligned VSX load/store on P8/P9.
10923         (expand_block_clear): Allow the use of unaligned VSX
10924         load/store on P8/P9.
10925
10926 2018-01-02  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
10927
10928         * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
10929         New function.
10930         (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
10931         swap associated with both a load and a store.
10932
10933 2018-01-02  Andrew Waterman  <andrew@sifive.com>
10934
10935         * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
10936         * config/riscv/riscv.md (clear_cache): Use it.
10937
10938 2018-01-02  Artyom Skrobov  <tyomitch@gmail.com>
10939
10940         * web.c: Remove out-of-date comment.
10941
10942 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
10943
10944         * expr.c (fixup_args_size_notes): Check that any existing
10945         REG_ARGS_SIZE notes are correct, and don't try to re-add them.
10946         (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
10947         (emit_single_push_insn): ...here.
10948
10949 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
10950
10951         * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
10952         (const_vector_encoded_nelts): New function.
10953         (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
10954         (const_vector_int_elt, const_vector_elt): Declare.
10955         * emit-rtl.c (const_vector_int_elt_1): New function.
10956         (const_vector_elt): Likewise.
10957         * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
10958         of CONST_VECTOR_ELT.
10959
10960 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
10961
10962         * expr.c: Include rtx-vector-builder.h.
10963         (const_vector_mask_from_tree): Use rtx_vector_builder and operate
10964         directly on the tree encoding.
10965         (const_vector_from_tree): Likewise.
10966         * optabs.c: Include rtx-vector-builder.h.
10967         (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
10968         sequence of "u" values.
10969         * vec-perm-indices.c: Include rtx-vector-builder.h.
10970         (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
10971         directly on the vec_perm_indices encoding.
10972
10973 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
10974
10975         * doc/rtl.texi (const_vector): Describe new encoding scheme.
10976         * Makefile.in (OBJS): Add rtx-vector-builder.o.
10977         * rtx-vector-builder.h: New file.
10978         * rtx-vector-builder.c: Likewise.
10979         * rtl.h (rtx_def::u2): Add a const_vector field.
10980         (CONST_VECTOR_NPATTERNS): New macro.
10981         (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
10982         (CONST_VECTOR_DUPLICATE_P): Likewise.
10983         (CONST_VECTOR_STEPPED_P): Likewise.
10984         (CONST_VECTOR_ENCODED_ELT): Likewise.
10985         (const_vec_duplicate_p): Check for a duplicated vector encoding.
10986         (unwrap_const_vec_duplicate): Likewise.
10987         (const_vec_series_p): Check for a non-duplicated vector encoding.
10988         Say that the function only returns true for integer vectors.
10989         * emit-rtl.c: Include rtx-vector-builder.h.
10990         (gen_const_vec_duplicate_1): Delete.
10991         (gen_const_vector): Call gen_const_vec_duplicate instead of
10992         gen_const_vec_duplicate_1.
10993         (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
10994         (gen_const_vec_duplicate): Use rtx_vector_builder.
10995         (gen_const_vec_series): Likewise.
10996         (gen_rtx_CONST_VECTOR): Likewise.
10997         * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
10998         (swap_const_vector_halves): Take an rtx pointer rather than rtx.
10999         Build a new vector rather than modifying a CONST_VECTOR in-place.
11000         (handle_special_swappables): Update call accordingly.
11001         * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
11002         (swap_const_vector_halves): Take an rtx pointer rather than rtx.
11003         Build a new vector rather than modifying a CONST_VECTOR in-place.
11004         (handle_special_swappables): Update call accordingly.
11005
11006 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
11007
11008         * simplify-rtx.c (simplify_const_binary_operation): Use
11009         CONST_VECTOR_ELT instead of XVECEXP.
11010
11011 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
11012
11013         * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
11014         the selector elements to be different from the data elements
11015         if the selector is a VECTOR_CST.
11016         * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
11017         ssizetype for the selector.
11018
11019 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
11020
11021         * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
11022         before testing each element individually.
11023         * tree-vect-generic.c (lower_vec_perm): Likewise.
11024
11025 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
11026
11027         * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
11028         * selftest-run-tests.c (selftest::run_tests): Call it.
11029         * vector-builder.h (vector_builder::operator ==): New function.
11030         (vector_builder::operator !=): Likewise.
11031         * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
11032         (vec_perm_indices::all_from_input_p): New function.
11033         * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
11034         (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
11035         * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
11036         instead of reading the VECTOR_CST directly.  Detect whether both
11037         vector inputs are the same before constructing the vec_perm_indices,
11038         and update the number of inputs argument accordingly.  Use the
11039         utility functions added above.  Only construct sel2 if we need to.
11040
11041 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
11042
11043         * optabs.c (expand_vec_perm_var): Use an explicit encoding for
11044         the broadcast of the low byte.
11045         (expand_mult_highpart): Use an explicit encoding for the permutes.
11046         * optabs-query.c (can_mult_highpart_p): Likewise.
11047         * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
11048         * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
11049         (vectorizable_bswap): Likewise.
11050         * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
11051         explicit encoding for the power-of-2 permutes.
11052         (vect_permute_store_chain): Likewise.
11053         (vect_grouped_load_supported): Likewise.
11054         (vect_permute_load_chain): Likewise.
11055
11056 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
11057
11058         * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
11059         * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
11060         * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
11061         * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
11062         * tree-vect-stmts.c (vectorizable_bswap): Likewise.
11063         (vect_gen_perm_mask_any): Likewise.
11064
11065 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
11066
11067         * int-vector-builder.h: New file.
11068         * vec-perm-indices.h: Include int-vector-builder.h.
11069         (vec_perm_indices): Redefine as an int_vector_builder.
11070         (auto_vec_perm_indices): Delete.
11071         (vec_perm_builder): Redefine as a stand-alone class.
11072         (vec_perm_indices::vec_perm_indices): New function.
11073         (vec_perm_indices::clamp): Likewise.
11074         * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
11075         (vec_perm_indices::new_vector): New function.
11076         (vec_perm_indices::new_expanded_vector): Update for new
11077         vec_perm_indices class.
11078         (vec_perm_indices::rotate_inputs): New function.
11079         (vec_perm_indices::all_in_range_p): Operate directly on the
11080         encoded form, without computing elided elements.
11081         (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
11082         encoding.  Update for new vec_perm_indices class.
11083         * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
11084         the given vec_perm_builder.
11085         (expand_vec_perm_var): Update vec_perm_builder constructor.
11086         (expand_mult_highpart): Use vec_perm_builder instead of
11087         auto_vec_perm_indices.
11088         * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
11089         vec_perm_indices instead of auto_vec_perm_indices.  Use a single
11090         or double series encoding as appropriate.
11091         * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
11092         vec_perm_indices instead of auto_vec_perm_indices.
11093         * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
11094         * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
11095         (vect_permute_store_chain): Likewise.
11096         (vect_grouped_load_supported): Likewise.
11097         (vect_permute_load_chain): Likewise.
11098         (vect_shift_permute_load_chain): Likewise.
11099         * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
11100         (vect_transform_slp_perm_load): Likewise.
11101         (vect_schedule_slp_instance): Likewise.
11102         * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
11103         (vectorizable_mask_load_store): Likewise.
11104         (vectorizable_bswap): Likewise.
11105         (vectorizable_store): Likewise.
11106         (vectorizable_load): Likewise.
11107         * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
11108         vec_perm_indices instead of auto_vec_perm_indices.  Use
11109         tree_to_vec_perm_builder to read the vector from a tree.
11110         * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
11111         vec_perm_builder instead of a vec_perm_indices.
11112         (have_whole_vector_shift): Use vec_perm_builder and
11113         vec_perm_indices instead of auto_vec_perm_indices.  Leave the
11114         truncation to calc_vec_perm_mask_for_shift.
11115         (vect_create_epilog_for_reduction): Likewise.
11116         * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
11117         from auto_vec_perm_indices to vec_perm_indices.
11118         (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
11119         instead of changing individual elements.
11120         (aarch64_vectorize_vec_perm_const): Use new_vector to install
11121         the vector in d.perm.
11122         * config/arm/arm.c (expand_vec_perm_d::perm): Change
11123         from auto_vec_perm_indices to vec_perm_indices.
11124         (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
11125         instead of changing individual elements.
11126         (arm_vectorize_vec_perm_const): Use new_vector to install
11127         the vector in d.perm.
11128         * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
11129         Update vec_perm_builder constructor.
11130         (rs6000_expand_interleave): Likewise.
11131         * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
11132         (rs6000_expand_interleave): Likewise.
11133
11134 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
11135
11136         * optabs-query.c (can_vec_perm_var_p): Check whether lowering
11137         to qimode could truncate the indices.
11138         * optabs.c (expand_vec_perm_var): Likewise.
11139
11140 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
11141
11142         * Makefile.in (OBJS): Add vec-perm-indices.o.
11143         * vec-perm-indices.h: New file.
11144         * vec-perm-indices.c: Likewise.
11145         * target.h (vec_perm_indices): Replace with a forward class
11146         declaration.
11147         (auto_vec_perm_indices): Move to vec-perm-indices.h.
11148         * optabs.h: Include vec-perm-indices.h.
11149         (expand_vec_perm): Delete.
11150         (selector_fits_mode_p, expand_vec_perm_var): Declare.
11151         (expand_vec_perm_const): Declare.
11152         * target.def (vec_perm_const_ok): Replace with...
11153         (vec_perm_const): ...this new hook.
11154         * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
11155         (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
11156         * doc/tm.texi: Regenerate.
11157         * optabs.def (vec_perm_const): Delete.
11158         * doc/md.texi (vec_perm_const): Likewise.
11159         (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
11160         * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
11161         expand_vec_perm for constant permutation vectors.  Assert that
11162         the mode of variable permutation vectors is the integer equivalent
11163         of the mode that is being permuted.
11164         * optabs-query.h (selector_fits_mode_p): Declare.
11165         * optabs-query.c: Include vec-perm-indices.h.
11166         (selector_fits_mode_p): New function.
11167         (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
11168         is defined, instead of checking whether the vec_perm_const_optab
11169         exists.  Use targetm.vectorize.vec_perm_const instead of
11170         targetm.vectorize.vec_perm_const_ok.  Check whether the indices
11171         fit in the vector mode before using a variable permute.
11172         * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
11173         vec_perm_indices instead of an rtx.
11174         (expand_vec_perm): Replace with...
11175         (expand_vec_perm_const): ...this new function.  Take the selector
11176         as a vec_perm_indices rather than an rtx.  Also take the mode of
11177         the selector.  Update call to shift_amt_for_vec_perm_mask.
11178         Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
11179         Use vec_perm_indices::new_expanded_vector to expand the original
11180         selector into bytes.  Check whether the indices fit in the vector
11181         mode before using a variable permute.
11182         (expand_vec_perm_var): Make global.
11183         (expand_mult_highpart): Use expand_vec_perm_const.
11184         * fold-const.c: Includes vec-perm-indices.h.
11185         * tree-ssa-forwprop.c: Likewise.
11186         * tree-vect-data-refs.c: Likewise.
11187         * tree-vect-generic.c: Likewise.
11188         * tree-vect-loop.c: Likewise.
11189         * tree-vect-slp.c: Likewise.
11190         * tree-vect-stmts.c: Likewise.
11191         * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
11192         Delete.
11193         * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
11194         * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
11195         (aarch64_vectorize_vec_perm_const_ok): Fuse into...
11196         (aarch64_vectorize_vec_perm_const): ...this new function.
11197         (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
11198         (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
11199         * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
11200         * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
11201         * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
11202         (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
11203         (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
11204         into...
11205         (arm_vectorize_vec_perm_const): ...this new function.  Explicitly
11206         check for NEON modes.
11207         * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
11208         * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
11209         * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
11210         (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
11211         into...
11212         (ix86_vectorize_vec_perm_const): ...this new function.  Incorporate
11213         the old VEC_PERM_CONST conditions.
11214         * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
11215         * config/ia64/vect.md (vec_perm_const<mode>): Delete.
11216         * config/ia64/ia64.c (ia64_expand_vec_perm_const)
11217         (ia64_vectorize_vec_perm_const_ok): Merge into...
11218         (ia64_vectorize_vec_perm_const): ...this new function.
11219         * config/mips/loongson.md (vec_perm_const<mode>): Delete.
11220         * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
11221         * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
11222         * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
11223         * config/mips/mips.c (mips_expand_vec_perm_const)
11224         (mips_vectorize_vec_perm_const_ok): Merge into...
11225         (mips_vectorize_vec_perm_const): ...this new function.
11226         * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
11227         * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
11228         * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
11229         * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
11230         * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
11231         (rs6000_expand_vec_perm_const): Delete.
11232         * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
11233         Delete.
11234         (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
11235         (altivec_expand_vec_perm_const_le): Take each operand individually.
11236         Operate on constant selectors rather than rtxes.
11237         (altivec_expand_vec_perm_const): Likewise.  Update call to
11238         altivec_expand_vec_perm_const_le.
11239         (rs6000_expand_vec_perm_const): Delete.
11240         (rs6000_vectorize_vec_perm_const_ok): Delete.
11241         (rs6000_vectorize_vec_perm_const): New function.
11242         (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
11243         an element count and rtx array.
11244         (rs6000_expand_extract_even): Update call accordingly.
11245         (rs6000_expand_interleave): Likewise.
11246         * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
11247         * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
11248         * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
11249         * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
11250         (rs6000_expand_vec_perm_const): Delete.
11251         * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
11252         (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
11253         (altivec_expand_vec_perm_const_le): Take each operand individually.
11254         Operate on constant selectors rather than rtxes.
11255         (altivec_expand_vec_perm_const): Likewise.  Update call to
11256         altivec_expand_vec_perm_const_le.
11257         (rs6000_expand_vec_perm_const): Delete.
11258         (rs6000_vectorize_vec_perm_const_ok): Delete.
11259         (rs6000_vectorize_vec_perm_const): New function.  Remove stray
11260         reference to the SPE evmerge intructions.
11261         (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
11262         an element count and rtx array.
11263         (rs6000_expand_extract_even): Update call accordingly.
11264         (rs6000_expand_interleave): Likewise.
11265         * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
11266         * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
11267         new function.
11268         (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
11269
11270 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
11271
11272         * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
11273         vector mode and that that mode matches the mode of the data
11274         being permuted.
11275         (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
11276         out into expand_vec_perm_var.  Do all CONST_VECTOR handling here,
11277         directly using expand_vec_perm_1 when forcing selectors into
11278         registers.
11279         (expand_vec_perm_var): New function, split out from expand_vec_perm.
11280
11281 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
11282
11283         * optabs-query.h (can_vec_perm_p): Delete.
11284         (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
11285         * optabs-query.c (can_vec_perm_p): Split into...
11286         (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
11287         (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
11288         particular selector is valid.
11289         * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
11290         * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
11291         (vect_grouped_load_supported): Likewise.
11292         (vect_shift_permute_load_chain): Likewise.
11293         * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
11294         (vect_transform_slp_perm_load): Likewise.
11295         * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
11296         (vectorizable_bswap): Likewise.
11297         (vect_gen_perm_mask_checked): Likewise.
11298         * fold-const.c (fold_ternary_loc): Likewise.  Don't take
11299         implementations of variable permutation vectors into account
11300         when deciding which selector to use.
11301         * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
11302         vec_perm_const_optab is supported; instead use can_vec_perm_const_p
11303         with a false third argument.
11304         * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
11305         to test whether the constant selector is valid and can_vec_perm_var_p
11306         to test whether a variable selector is valid.
11307
11308 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
11309
11310         * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
11311         * optabs-query.c (can_vec_perm_p): Likewise.
11312         * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
11313         instead of vec_perm_indices.
11314         * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
11315         (vect_gen_perm_mask_checked): Likewise,
11316         * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
11317         (vect_gen_perm_mask_checked): Likewise,
11318
11319 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
11320
11321         * optabs-query.h (qimode_for_vec_perm): Declare.
11322         * optabs-query.c (can_vec_perm_p): Split out qimode search to...
11323         (qimode_for_vec_perm): ...this new function.
11324         * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
11325
11326 2018-01-02  Aaron Sawdey  <acsawdey@linux.vnet.ibm.com>
11327
11328         * rtlanal.c (canonicalize_condition): Return 0 if final rtx
11329         does not have a conditional at the top.
11330
11331 2018-01-02  Richard Biener  <rguenther@suse.de>
11332
11333         * ipa-inline.c (big_speedup_p): Fix expression.
11334
11335 2018-01-02  Jan Hubicka  <hubicka@ucw.cz>
11336
11337         PR target/81616
11338         * config/i386/x86-tune-costs.h: Increase cost of integer load costs
11339         for generic 4->6.
11340
11341 2018-01-02  Jan Hubicka  <hubicka@ucw.cz>
11342
11343         PR target/81616
11344         Generic tuning.
11345         * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
11346         cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
11347         and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
11348         cond_taken_branch_cost 3->4.
11349
11350 2018-01-01  Jakub Jelinek  <jakub@redhat.com>
11351
11352         PR tree-optimization/83581
11353         * tree-loop-distribution.c (pass_loop_distribution::execute): Return
11354         TODO_cleanup_cfg if any changes have been made.
11355
11356         PR middle-end/83608
11357         * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
11358         convert_modes if target mode has the right side, but different mode
11359         class.
11360
11361         PR middle-end/83609
11362         * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
11363         last argument when extracting from CONCAT.  If either from_real or
11364         from_imag is NULL, use expansion through memory.  If result is not
11365         a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
11366         the parts directly to inner mode, if even that fails, use expansion
11367         through memory.
11368
11369         PR middle-end/83623
11370         * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
11371         check for bswap in mode rather than HImode and use that in expand_unop
11372         too.
11373 \f
11374 Copyright (C) 2018 Free Software Foundation, Inc.
11375
11376 Copying and distribution of this file, with or without modification,
11377 are permitted in any medium without royalty provided the copyright
11378 notice and this notice are preserved.