gcc/ChangeLog

   1 2018-05-16  Richard Biener  <rguenther@suse.de>
   2
   3         * tree-cfg.c (verify_gimple_assign_ternary): Properly
   4         verify the [VEC_]COND_EXPR embedded comparison.
   5
   6 2018-05-15  Martin Sebor  <msebor@redhat.com>
   7
   8         PR tree-optimization/85753
   9         * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Handle
  10         RECORD_TYPE in addition to ARRAY_TYPE.
  11
  12 2018-05-15  Martin Sebor  <msebor@redhat.com>
  13
  14         PR middle-end/85643
  15         * calls.c (get_attr_nonstring_decl): Handle MEM_REF.
  16
  17 2018-05-15  Richard Biener  <rguenther@suse.de>
  18
  19         * tree-ssa-dse.c (dse_classify_store): Remove use_stmt parameter,
  20         add by_clobber_p one.  Change algorithm to collect all defs
  21         representing uses we need to walk and try reducing them to
  22         a single one before failing.
  23         (dse_dom_walker::dse_optimize_stmt): Adjust.
  24
  25 2018-05-13  Mark Wielaard  <mark@klomp.org>
  26
  27         * dwarf2out.c (dwarf_OP): Handle DW_OP_addrx and DW_OP_constx.
  28         (size_of_loc_descr): Likewise.
  29         (output_loc_operands): Likewise.
  30         (output_loc_operands_raw): Likewise.
  31         (dw_addr_op): Use dwarf_OP () for DW_OP_constx and DW_OP_addrx.
  32         (resolve_addr_in_expr): Handle DW_OP_addrx and DW_OP_constx.
  33         (hash_loc_operands): Likewise.
  34         (compare_loc_operands): Likewise.
  35
  36 2018-05-14  Mark Wielaard  <mark@klomp.org>
  37
  38         * dwarf2out.c (count_index_addrs): New function.
  39         (dwarf2out_finish): Use count_index_addrs to calculate addrs_length.
  40
  41 2018-05-15  Prathamesh Kulkarni  <prathamesh.kulkarni@linaro.org>
  42
  43         PR tree-optimization/83648
  44         * ipa-pure-const.c (malloc_candidate_p): Allow function with NULL
  45         return value as malloc candidate.
  46
  47 2018-05-15  Prathamesh Kulkarni  <prathamesh.kulkarni@linaro.org>
  48
  49         PR ipa/85734
  50         * ipa-pure-const.c (warn_function_malloc): Pass value of known_finite param
  51         as true in call to suggest_attribute.
  52
  53 2018-05-14  Segher Boessenkool  <segher@kernel.crashing.org>
  54
  55         * doc/invoke.texi (RS/6000 and PowerPC Options): Delete @itemx for
  56         -mreadonly-in-sdata.
  57
  58 2018-05-14  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
  59
  60         * config/aarch64/aarch64-simd.md (*aarch64_crypto_aese_fused):
  61         New pattern.
  62         (aarch64_crypto_aesd_fused): Likewise.
  63
  64 2018-05-14  Wilco Dijkstra  <wdijkstr@arm.com>
  65
  66         * config/aarch64/aarch64.md (mov<mode>): Remove '*' in alternatives.
  67         (movsi_aarch64): Likewise.
  68         (load_pairsi): Likewise.
  69         (load_pairdi): Likewise.
  70         (store_pairsi): Likewise.
  71         (store_pairdi): Likewise.
  72         (load_pairsf): Likewise.
  73         (load_pairdf): Likewise.
  74         (store_pairsf): Likewise.
  75         (store_pairdf): Likewise.
  76         (zero_extend): Likewise.
  77         (trunc): Swap alternatives.
  78         (fcvt_target): Add '?' to prefer w over r.
  79
  80 2018-05-14  Jakub Jelinek  <jakub@redhat.com>
  81
  82         PR target/85756
  83         * config/i386/i386.md: Disallow non-commutative arithmetics in
  84         last twpeephole for mem {+,-,&,|,^}= x; mem != 0 after cmpelim
  85         optimization.  Use COMMUTATIVE_ARITH_P test rather than != MINUS
  86         in the peephole2 before it.
  87
  88 2018-05-14  Sebastian Peryt  <sebastian.peryt@intel.com>
  89
  90         * common/config/i386/i386-common.c (OPTION_MASK_ISA_CLDEMOTE_SET,
  91         OPTION_MASK_ISA_CLDEMOTE_UNSET): New defines.
  92         (ix86_handle_option): Handle -mcldemote.
  93         * config.gcc: New header.
  94         * config/i386/cldemoteintrin.h: New file.
  95         * config/i386/cpuid.h (bit_CLDEMOTE): New bit.
  96         * config/i386/driver-i386.c (host_detect_local_cpu): Detect
  97         -mcldemote.
  98         * config/i386/i386-c.c (ix86_target_macros_internal): Handle
  99         OPTION_MASK_ISA_CLDEMOTE.
 100         * config/i386/i386.c (ix86_target_string): Add -mcldemote.
 101         (ix86_valid_target_attribute_inner_p): Ditto.
 102         (enum ix86_builtins): Add IX86_BUILTIN_CLDEMOTE.
 103         (ix86_init_mmx_sse_builtins): Define __builtin_ia32_cldemote.
 104         (ix86_expand_builtin): Expand IX86_BUILTIN_CLDEMOTE.
 105         * config/i386/i386.h (TARGET_CLDEMOTE, TARGET_CLDEMOTE_P): New.
 106         * config/i386/i386.md (UNSPECV_CLDEMOTE): New.
 107         (cldemote): New.
 108         * config/i386/i386.opt: Add -mcldemote.
 109         * config/i386/x86intrin.h: New header.
 110         * doc/invoke.texi: Add -mcldemote.
 111
 112 2018-05-14  Richard Biener  <rguenther@suse.de>
 113
 114         * doc/match-and-simplify.texi: Adjust :s documentation.
 115
 116 2018-05-14  Alexander Monakov  <amonakov@ispras.ru>
 117
 118         * sort.cc (REORDER_23): Pass the type for the temporaries instead of
 119         intended memcpy size.
 120         (REORDER_45): Likewise.
 121
 122 2018-05-13  Alexander Monakov  <amonakov@ispras.ru>
 123
 124         * sort.cc: New file.
 125         * system.h [!CHECKING_P] (qsort): Redirect to gcc_qsort.
 126         * vec.c (qsort_chk): Use gcc_qsort.
 127         * Makefile.in (OBJS-libcommon): Add sort.o.
 128         (build/sort.o): New target.  Use it...
 129         (BUILD_RTL): ... here, and...
 130         (build/gencfn-macros): ... here, and...
 131         (build/genmatch): ... here.
 132
 133 2018-05-13  Kito Cheng  <kito.cheng@gmail.com>
 134             Chung-Ju Wu  <jasonwucj@gmail.com>
 135
 136         * config.gcc (nds32*-*-*): Check that n15 is valid to --with-cpu.
 137         * config/nds32/nds32-graywolf.md: New file.
 138         * config/nds32/nds32-opts.h (nds32_cpu_type): Add CPU_GRAYWOLF.
 139         * config/nds32/nds32-pipelines-auxiliary.c: Implementation for n15
 140         pipeline.
 141         * config/nds32/nds32-protos.h: More declarations for n15 pipeline.
 142         * config/nds32/nds32-utils.c: More implementations for n15 pipeline.
 143         * config/nds32/nds32.md (pipeline_model): Add graywolf.
 144         * config/nds32/nds32.opt (mcpu): Support n15 pipeline cpus.
 145         * config/nds32/pipelines.md: Include n15 settings.
 146
 147 2018-05-13  Kito Cheng  <kito.cheng@gmail.com>
 148             Chung-Ju Wu  <jasonwucj@gmail.com>
 149
 150         * config.gcc (nds32*-*-*): Check that n12/n13 are valid to --with-cpu.
 151         * config/nds32/nds32-n13.md: New file.
 152         * config/nds32/nds32-opts.h (nds32_cpu_type): Add CPU_N12 and CPU_N13.
 153         * config/nds32/nds32-pipelines-auxiliary.c: Implementation for n12/n13
 154         pipeline.
 155         * config/nds32/nds32-protos.h: More declarations for n12/n13 pipeline.
 156         * config/nds32/nds32.md (pipeline_model): Add n13.
 157         * config/nds32/nds32.opt (mcpu): Support n12/n13 pipeline cpus.
 158         * config/nds32/pipelines.md: Include n13 settings.
 159
 160 2018-05-13  Kito Cheng  <kito.cheng@gmail.com>
 161             Chung-Ju Wu  <jasonwucj@gmail.com>
 162
 163         * config.gcc (nds32*-*-*): Check that n10/d10 are valid to --with-cpu.
 164         * config/nds32/nds32-n10.md: New file.
 165         * config/nds32/nds32-opts.h (nds32_cpu_type): Add CPU_N10.
 166         * config/nds32/nds32-pipelines-auxiliary.c: Implementation for n10
 167         pipeline.
 168         * config/nds32/nds32-protos.h: More declarations for n10 pipeline.
 169         * config/nds32/nds32-utils.c: More implementations for n10 pipeline.
 170         * config/nds32/nds32.md (pipeline_model): Add n10.
 171         * config/nds32/nds32.opt (mcpu): Support n10 pipeline cpus.
 172         * config/nds32/pipelines.md: Include n10 settings.
 173
 174 2018-05-13  Monk Chiang  <sh.chiang04@gmail.com>
 175             Kito Cheng  <kito.cheng@gmail.com>
 176             Chung-Ju Wu  <jasonwucj@gmail.com>
 177
 178         * config.gcc (nds32be-*-*): Handle --with-ext-dsp.
 179         * config/nds32/constants.md (unspec_element, unspec_volatile_element):
 180         Add enum values for DSP extension instructions.
 181         * config/nds32/constraints.md (Iu06, IU06, CVp5, CVs5, CVs2, CVhi):
 182         New constraints.
 183         * config/nds32/iterators.md (shifts, shiftrt, sat_plus, all_plus,
 184         sat_minus, all_minus, plus_minus, extend, sumax, sumin, sumin_max):
 185         New code iterators.
 186         (su, zs, uk, opcode, add_rsub, add_sub): New code attributes.
 187         * config/nds32/nds32-dspext.md: New file for DSP implementation.
 188         * config/nds32/nds32-intrinsic.c: Implementation of DSP extension.
 189         * config/nds32/nds32-intrinsic.md: Likewise.
 190         * config/nds32/nds32_intrinsic.h: Likewise.
 191         * config/nds32/nds32-md-auxiliary.c: Likewise.
 192         * config/nds32/nds32-memory-manipulation.c: Consider DSP extension.
 193         * config/nds32/nds32-predicates.c (const_vector_to_hwint): New.
 194         (nds32_valid_CVp5_p, nds32_valid_CVs5_p): New.
 195         (nds32_valid_CVs2_p, nds32_valid_CVhi_p): New.
 196         * config/nds32/nds32-protos.h: New declarations for DSP extension.
 197         * config/nds32/nds32-utils.c (extract_mac_non_acc_rtx): New case
 198         TYPE_DMAC in switch statement.
 199         * config/nds32/nds32.c: New checking and implementation for DSP
 200         extension instructions.
 201         * config/nds32/nds32.h: Likewise.
 202         * config/nds32/nds32.md: Likewise.
 203         * config/nds32/nds32.opt (mhw-abs, mext-dsp): New options.
 204         * config/nds32/predicates.md: Implement new predicates for DSP
 205         extension.
 206
 207 2018-05-11  Michael Meissner  <meissner@linux.vnet.ibm.com>
 208
 209         * config/rs6000/rs6000.md (mov<mode>_softfloat, FMOVE32):
 210         Reformat alternatives and attributes so it is easier to identify
 211         which constraints/attributes go with which instruction.
 212         (mov<mode>_hardfloat32, FMOVE64): Likewise.
 213         (mov<mode>_softfloat32, FMOVE64): Likewise.
 214         (mov<mode>_hardfloat64, FMOVE64): Likewise.
 215         (mov<mode>_softfloat64, FMOVE64): Likewise.
 216
 217 2018-05-11  Kelvin Nilsen  <kelvin@gcc.gnu.org>
 218
 219         * doc/extend.texi (PowerPC Built-in Functions): Rename this
 220         subsection.
 221         (Basic PowerPC Built-in Functions): The new name of the
 222         subsection previously known as "PowerPC Built-in Functions".
 223         (Basic PowerPC Built-in Functions Available on all Configurations):
 224         New subsubsection.
 225         (Basic PowerPC Built-in Functions Available on ISA 2.05): Likewise.
 226         (Basic PowerPC Built-in Functions Available on ISA 2.06): Likewise.
 227         (Basic PowerPC Built-in Functions Available on ISA 2.07): Likewise.
 228         (Basic PowerPC Built-in Functions Available on ISA 3.0): Likewise.
 229
 230 2018-05-11  Martin Jambor  <mjambor@suse.cz>
 231
 232         PR ipa/85655
 233         * ipa-cp.c (intersect_with_plats): Check that the lattice contains
 234         single const.
 235
 236 2018-05-11  Richard Earnshaw  <rearnsha@arm.com>
 237
 238         PR target/85733
 239         * config/arm/arm-cpus.in (fgroup ARMv6m): Add be8 feature.
 240
 241 2018-05-11  Sebastian Peryt  <sebastian.peryt@intel.com>
 242
 243         * common/config/i386/i386-common.c (OPTION_MASK_ISA_WAITPKG_SET,
 244         OPTION_MASK_ISA_WAITPKG_UNSET): New defines.
 245         (ix86_handle_option): Handle -mwaitpkg.
 246         * config.gcc: New header.
 247         * config/i386/cpuid.h (bit_WAITPKG): New bit.
 248         * config/i386/driver-i386.c (host_detect_local_cpu): Detect -mwaitpkg.
 249         * config/i386/i386-builtin-types.def ((UINT8, UNSIGNED, UINT64)): New
 250         function type.
 251         * config/i386/i386-c.c (ix86_target_macros_internal): Handle
 252         OPTION_MASK_ISA_WAITPKG.
 253         * config/i386/i386.c (ix86_target_string): Add -mwaitpkg.
 254         (ix86_option_override_internal): Add PTA_WAITPKG.
 255         (ix86_valid_target_attribute_inner_p): Add -mwaitpkg.
 256         (enum ix86_builtins): Add IX86_BUILTIN_UMONITOR, IX86_BUILTIN_UMWAIT,
 257         IX86_BUILTIN_TPAUSE.
 258         (ix86_init_mmx_sse_builtins): Define __builtin_ia32_umonitor,
 259         __builtin_ia32_umwait and __builtin_ia32_tpause.
 260         (ix86_expand_builtin): Expand IX86_BUILTIN_UMONITOR,
 261         IX86_BUILTIN_UMWAIT, IX86_BUILTIN_TPAUSE.
 262         * config/i386/i386.h (TARGET_WAITPKG, TARGET_WAITPKG_P): New.
 263         * config/i386/i386.md (UNSPECV_UMWAIT, UNSPECV_UMONITOR,
 264         UNSPECV_TPAUSE): New.
 265         (umwait, umwait_rex64, umonitor_<mode>, tpause, tpause_rex64): New.
 266         * config/i386/i386.opt: Add -mwaitpkg.
 267         * config/i386/waitpkgintrin.h: New file.
 268         * config/i386/x86intrin.h: New header.
 269         * doc/invoke.texi: Add -mwaitpkg.
 270
 271 2018-05-11  Richard Earnshaw  <rearnsha@arm.com>
 272
 273         PR target/85606
 274         * config/arm/arm-cpus.in: Add comment that ARMv6-m and ARMv6S-m are now
 275         equivalent.
 276         (cortex-m0): Use armv6s-m isa.
 277         (cortex-m0plus): Likewise.
 278         (cortex-m1): Likewise.
 279         (cortex-m0.small-multiply): Likewise.
 280         (cortex-m0plus.small-multiply): Likewise.
 281         (cortex-m1.small-multiply): Likewise.
 282
 283 2018-05-11  Allan Sandfeld Jensen  <allan.jensen@qt.io>
 284             Jakub Jelinek  <jakub@redhat.com>
 285
 286         PR tree-optimization/85692
 287         * tree-ssa-forwprop.c (simplify_vector_constructor): Try two
 288         source permute as well.
 289
 290 2018-05-11  Martin Liska  <mliska@suse.cz>
 291
 292         PR sanitizer/85556
 293         * doc/extend.texi: Document LLVM style format for no_sanitize
 294         attribute.
 295
 296 2018-05-10  Michael Meissner  <meissner@linux.vnet.ibm.com>
 297
 298         * config/rs6000/rs6000.c (mode_supports_dq_form): Rename
 299         mode_supports_vsx_dform_quad to mode_supports_dq_form.
 300         (mode_supports_vsx_dform_quad): Likewise.
 301         (mode_supports_vmx_dform): Move these functions to be next to the
 302         other mode_supports functions.
 303         (mode_supports_dq_form): Likewise.
 304         (quad_address_p): Change calls of mode_supports_vsx_dform_quad to
 305         mode_supports_dq_form.
 306         (reg_offset_addressing_ok_p): Likewise.
 307         (offsettable_ok_by_alignment): Likewise.
 308         (rs6000_legitimate_offset_address_p): Likewise.
 309         (legitimate_lo_sum_address_p): Likewise.
 310         (rs6000_legitimize_address): Likewise.
 311         (rs6000_legitimize_reload_address): Likewise.
 312         (rs6000_secondary_reload_inner): Likewise.
 313         (rs6000_preferred_reload_class): Likewise.
 314         (rs6000_output_move_128bit): Likewise.
 315
 316 2018-05-10  Uros Bizjak  <ubizjak@gmail.com>
 317
 318         * config/i386/i386.c (ix86_expand_builtin) <case IX86_BUILTIN_RDPID>:
 319         Generate SImode target register for null target.
 320         <case IX86_BUILTIN_XGETBV>: Ditto.
 321         <case IX86_BUILTIN_XSETBV>: Optimize LSHIFTRT generation.
 322         * config/i386/xsaveintrin.h (_xgetbv): Add missing return.
 323
 324 2018-05-10  Carl Love  <cel@us.ibm.com>
 325
 326         * config/rs6000/rs6000.md (prefetch): Generate ISA 2.06 instructions
 327         dcbtt and dcbtstt if operands[2] is 0.
 328
 329 2018-05-10  Uros Bizjak  <ubizjak@gmail.com>
 330
 331         PR target/85693
 332         * config/i386/sse.md (usadv64qi): New expander.
 333
 334 2018-05-10  Segher Boessenkool  <segher@kernel.crashing.org>
 335
 336         * config/rs6000/altivec.md (altivec_vmrghb, altivec_vmrghh,
 337         altivec_vmrghw, altivec_vmrglb, altivec_vmrglh, altivec_vmrglw): Remove
 338         -maltivec=be support.
 339         (vec_widen_umult_even_v16qi, vec_widen_smult_even_v16qi,
 340         vec_widen_umult_even_v8hi, vec_widen_smult_even_v8hi,
 341         vec_widen_umult_even_v4si, vec_widen_smult_even_v4si,
 342         vec_widen_umult_odd_v16qi, vec_widen_smult_odd_v16qi,
 343         vec_widen_umult_odd_v8hi, vec_widen_smult_odd_v8hi,
 344         vec_widen_umult_odd_v4si, vec_widen_smult_odd_v4si, altivec_vpkpx,
 345         altivec_vpks<VI_char>ss, altivec_vpks<VI_char>us,
 346         altivec_vpku<VI_char>us, altivec_vpku<VI_char>um, altivec_vsum2sws,
 347         altivec_vsumsws): Adjust.
 348         (altivec_vspltb *altivec_vspltb_internal, altivec_vsplth,
 349         *altivec_vsplth_internal, altivec_vspltw, *altivec_vspltw_internal,
 350         altivec_vspltsf, *altivec_vspltsf_internal): Remove -maltivec=be
 351         support.
 352         (altivec_vperm_<mode>, altivec_vperm_<mode>_uns,
 353         altivec_vupkhs<VU_char>, altivec_vupkls<VU_char>, altivec_vupkhpx,
 354         altivec_vupklpx, altivec_lvsl, altivec_lvsr): Adjust.
 355         (altivec_lve<VI_char>x): Delete expand.
 356         (*altivec_lve<VI_char>x_internal): Rename to...
 357         (altivec_lve<VI_char>x): ... this.
 358         (altivec_lvxl_<mode>): Delete expand.
 359         (*altivec_lvxl_<mode>_internal): Rename to ...
 360         (altivec_lvxl_<mode>): ... this.
 361         (altivec_stvxl_<mode>): Delete expand.
 362         (*altivec_stvxl_<mode>_internal): Rename to ...
 363         (altivec_stvxl_<mode>): ... this.
 364         (altivec_stve<VI_char>x): Delete expand.
 365         (*altivec_stve<VI_char>x_internal): Rename to ...
 366         (altivec_stve<VI_char>x): ... this.
 367         (doublee<mode>2, unsdoubleev4si2, doubleo<mode>2, unsdoubleov4si2,
 368         doubleh<mode>2, unsdoublehv4si2, doublel<mode>2, unsdoublelv4si2,
 369         reduc_plus_scal_<mode>): Adjust.
 370         * config/rs6000/rs6000-c.c (rs6000_target_modify_macros): Adjust
 371         comment.
 372         (rs6000_cpu_cpp_builtins): Adjust.
 373         (altivec_resolve_overloaded_builtin): Remove -maltivec=be support.
 374         * config/rs6000/rs6000-protos.h (altivec_expand_lvx_be,
 375         altivec_expand_stvx_be, altivec_expand_stvex_be): Delete.
 376         * config/rs6000/rs6000.c (rs6000_option_override_internal): Remove
 377         -maltivec=be support.
 378         (rs6000_split_vec_extract_var): Adjust.
 379         (rs6000_split_v4si_init): Adjust.
 380         (swap_selector_for_mode): Delete.
 381         (altivec_expand_lvx_be, altivec_expand_stvx_be,
 382         altivec_expand_stvex_be): Delete.
 383         (altivec_expand_lv_builtin, altivec_expand_stv_builtin): Remove
 384         -maltivec=be support.
 385         (rs6000_gimple_fold_builtin): Ditto.
 386         (rs6000_generate_float2_double_code, rs6000_generate_float2_code):
 387         Adjust.
 388         * config/rs6000/rs6000.h (VECTOR_ELT_ORDER_BIG): Delete.
 389         (TARGET_DIRECT_MOVE_64BIT): Adjust.
 390         * config/rs6000/rs6000.md (split for extendsidi2 for vectors): Adjust.
 391         * config/rs6000/rs6000.opt (maltivec=le, maltivec=be): Delete.
 392         * config/rs6000/vsx.md (floate<mode>, unsfloatev2di, floato<mode>,
 393         unsfloatov2di, vsignedo_v2df, vsignede_v2df, vunsignedo_v2df,
 394         vunsignede_v2df, vsx_extract_<mode>_p9, *vsx_extract_si,
 395         *vsx_extract_<mode>_p8, *vsx_extract_si_<uns>float_df,
 396         *vsx_extract_si_<uns>float_<mode>, vsx_set_<mode>_p9, vsx_set_v4sf_p9,
 397         *vsx_insert_extract_v4sf_p9, *vsx_insert_extract_v4sf_p9_2, and an
 398         anonymous split): Adjust.
 399         (vsx_mergel_<mode>, vsx_mergeh_<mode>): Remove -maltivec=be support.
 400         (vsx_xxspltd_<mode>, extract4b, insert4b): Adjust.
 401
 402 2018-05-10  Eric Botcazou  <ebotcazou@adacore.com>
 403
 404         * configure.ac (gcc_gxx_include_dir_add_sysroot): Set it to 1 only
 405         when --with-gxx-include-dir is also specified.
 406         * configure: Regenerate.
 407
 408 2018-05-09  Jim Wilson  <jimw@sifive.com>
 409
 410         PR target/84797
 411         * config.gcc (riscv*-*-*): Handle --with-multilib-list.
 412         * config/riscv/t-withmultilib: New.
 413         * config/riscv/withmultilib.h: New.
 414         * doc/install.texi: Document RISC-V --with-multilib-list support.
 415
 416 2018-05-09  Richard Biener  <rguenther@suse.de>
 417
 418         * tree-vect-slp.c (vect_bb_slp_scalar_cost): Fill a cost
 419         vector.
 420         (vect_bb_vectorization_profitable_p): Adjust.  Compute
 421         actual scalar cost using the cost vector and the add_stmt_cost
 422         machinery.
 423
 424 2018-05-09  Segher Boessenkool  <segher@kernel.crashing.org>
 425
 426         PR rtl-optimization/85645
 427         * config/rs6000/rs6000.c (rs6000_emit_prologue_components): Put a SET
 428         in the REG_CFA_REGISTER note for LR, don't leave it empty.
 429
 430 2018-05-09  Segher Boessenkool  <segher@kernel.crashing.org>
 431
 432         PR rtl-optimization/85645
 433         * shrink-wrap.c (spread_components): Return a boolean saying if
 434         anything was changed.
 435         (try_shrink_wrapping_separate): Iterate spread_components until
 436         nothing changes anymore.
 437
 438 2018-05-09  Segher Boessenkool  <segher@kernel.crashing.org>
 439
 440         PR rtl-optimization/85645
 441         * regrename.c (build_def_use): Also kill the chains that include the
 442         destination of a REG_CFA_REGISTER note.
 443
 444 2018-05-09  Segher Boessenkool  <segher@kernel.crashing.org>
 445
 446         PR rtl-optimization/85645
 447         *  regcprop.c (copyprop_hardreg_forward_1): Don't propagate into an
 448         insn that has a REG_CFA_REGISTER note.
 449
 450 2018-05-09  Richard Sandiford  <richard.sandiford@linaro.org>
 451
 452         * cfgexpand.c (expand_clobber): New function.
 453         (expand_gimple_stmt_1): Use it.
 454         * tree-vect-stmts.c (vect_clobber_variable): New function,
 455         split out from...
 456         (vectorizable_simd_clone_call): ...here.
 457         (vectorizable_store): Emit a clobber either side of an
 458         IFN_STORE_LANES sequence.
 459         (vectorizable_load): Emit a clobber after an IFN_LOAD_LANES sequence.
 460
 461 2018-05-09  Tom de Vries  <tom@codesourcery.com>
 462
 463         PR target/85626
 464         * config/nvptx/nvptx.md (define_insn "trap", define_insn "trap_if_true")
 465         (define_insn "trap_if_false"): Add exit after trap.
 466
 467 2018-05-09  Eric Botcazou  <ebotcazou@adacore.com>
 468
 469         PR rtl-optimization/85638
 470         * bb-reorder.c: Include common/common-target.h.
 471         (create_forwarder_block): New function extracted from...
 472         (fix_up_crossing_landing_pad): ...here.  Rename into...
 473         (dw2_fix_up_crossing_landing_pad): ...this.
 474         (sjlj_fix_up_crossing_landing_pad): New function.
 475         (find_rarely_executed_basic_blocks_and_crossing_edges): In SJLJ mode,
 476         call sjlj_fix_up_crossing_landing_pad if there are incoming EH edges
 477         from both partitions and exit the loop after one iteration.
 478
 479 2018-05-08  Kelvin Nilsen  <kelvin@gcc.gnu.org>
 480
 481         Revert:
 482         * doc/extend.texi (PowerPC Built-in Functions): Rename this
 483         subsection.
 484         (Basic PowerPC Built-in Functions): The new name of the
 485         subsection previously known as "PowerPC Built-in Functions".
 486         (Basic PowerPC Built-in Functions Available on all Configurations):
 487         New subsubsection.
 488         (Basic PowerPC Built-in Functions Available on ISA 2.05): New
 489         subsubsection.
 490         (Basic PowerPC Built-in Functions Available on ISA 2.06): New
 491         subsubsection.
 492         (Basic PowerPC Built-in Functions Available on ISA 2.07): New
 493         subsubsection.
 494         (Basic PowerPC Built-in Functions Available on ISA 3.0): New
 495         subsubsection.
 496
 497 2018-05-08  Jim Wilson  <jimw@sifive.com>
 498
 499         * config/riscv/linux.h (MUSL_ABI_SUFFIX): Delete unnecessary backslash.
 500         (LD_EMUL_SUFFIX): New.
 501         (LINK_SPEC): Use it.
 502
 503 2018-05-08  Kelvin Nilsen  <kelvin@gcc.gnu.org>
 504
 505         * doc/extend.texi (PowerPC Built-in Functions): Rename this
 506         subsection.
 507         (Basic PowerPC Built-in Functions): The new name of the
 508         subsection previously known as "PowerPC Built-in Functions".
 509         (Basic PowerPC Built-in Functions Available on all Configurations):
 510         New subsubsection.
 511         (Basic PowerPC Built-in Functions Available on ISA 2.05): New
 512         subsubsection.
 513         (Basic PowerPC Built-in Functions Available on ISA 2.06): New
 514         subsubsection.
 515         (Basic PowerPC Built-in Functions Available on ISA 2.07): New
 516         subsubsection.
 517         (Basic PowerPC Built-in Functions Available on ISA 3.0): New
 518         subsubsection.
 519
 520 2018-05-08  Jakub Jelinek  <jakub@redhat.com>
 521
 522         PR target/85683
 523         * config/i386/i386.md: Add peepholes for mem {+,-,&,|,^}= x; mem != 0
 524         after cmpelim optimization.
 525
 526 2018-05-08  Olga Makhotina  <olga.makhotina@intel.com>
 527
 528         * config.gcc: Support "goldmont".
 529         * config/i386/driver-i386.c (host_detect_local_cpu): Detect "goldmont".
 530         * config/i386/i386-c.c (ix86_target_macros_internal): Handle
 531         PROCESSOR_GOLDMONT.
 532         * config/i386/i386.c (m_GOLDMONT): Define.
 533         (processor_target_table): Add "goldmont".
 534         (PTA_GOLDMONT): Define.
 535         (ix86_lea_outperforms): Add TARGET_GOLDMONT.
 536         (get_builtin_code_for_version): Handle PROCESSOR_GOLDMONT.
 537         (fold_builtin_cpu): Add M_INTEL_GOLDMONT.
 538         (fold_builtin_cpu): Add "goldmont".
 539         (ix86_add_stmt_cost): Add TARGET_GOLDMONT.
 540         (ix86_option_override_internal): Add "goldmont".
 541         * config/i386/i386.h (processor_costs): Define TARGET_GOLDMONT.
 542         (processor_type): Add PROCESSOR_GOLDMONT.
 543         * config/i386/i386.md: Add CPU "glm".
 544         * config/i386/glm.md: New file.
 545         * config/i386/x86-tune.def: Add m_GOLDMONT.
 546         * doc/invoke.texi: Add goldmont as x86 -march=/-mtune= CPU type.
 547
 548 2018-05-08  Jakub Jelinek  <jakub@redhat.com>
 549
 550         PR target/85572
 551         * config/i386/i386.c (ix86_expand_sse2_abs): Handle E_V2DImode and
 552         E_V4DImode.
 553         * config/i386/sse.md (abs<mode>2): Use VI_AVX2 iterator instead of
 554         VI1248_AVX512VL_AVX512BW.  Handle V2DImode and V4DImode if not
 555         TARGET_AVX512VL using ix86_expand_sse2_abs.  Formatting fixes.
 556
 557         PR target/85317
 558         * config/i386/i386.c (ix86_fold_builtin): Handle
 559         IX86_BUILTIN_{,P}MOVMSK{PS,PD,B}{,128,256}.
 560
 561         PR target/85480
 562         * config/i386/sse.md (ssequaterinsnmode): New mode attribute.
 563         (*<extract_type>_vinsert<shuffletype><extract_suf>_0): New pattern.
 564
 565 2018-05-08  Richard Earnshaw  <rearnsha@arm.com>
 566
 567         PR target/85658
 568         * config/arm/parsecpu.awk (check_cpu): Fix operator precedence.
 569         (check_arch): Likewise.
 570         (check_fpu): Return the result rather than printing it.
 571         (end arch): Fix operator precedence.
 572         (end cpu): Likewise.
 573         (END): Print the result from check_fpu.
 574
 575 2018-05-08  Richard Sandiford  <richard.sandiford@linaro.org>
 576             Alan Hayward  <alan.hayward@arm.com>
 577             David Sherwood  <david.sherwood@arm.com>
 578
 579         * config/aarch64/aarch64-sve.md (*pred_cmp<cmp_op><mode>_combine)
 580         (*pred_cmp<cmp_op><mode>, *fcm<cmp_op><mode>_and_combine)
 581         (*fcmuo<mode>_and_combine, *fcm<cmp_op><mode>_and)
 582         (*fcmuo<mode>_and): New patterns.
 583
 584 2018-05-08  Richard Sandiford  <richard.sandiford@linaro.org>
 585
 586         * config/aarch64/iterators.md (UNSPEC_COND_LO, UNSPEC_COND_LS)
 587         (UNSPEC_COND_HI, UNSPEC_COND_HS, UNSPEC_COND_UO): Delete.
 588         (SVE_INT_CMP, SVE_FP_CMP): New code iterators.
 589         (cmp_op, sve_imm_con): New code attributes.
 590         (SVE_COND_INT_CMP, imm_con): Delete.
 591         (cmp_op): Remove above unspecs from int attribute.
 592         * config/aarch64/aarch64-sve.md (*vec_cmp<cmp_op>_<mode>): Rename
 593         to...
 594         (*cmp<cmp_op><mode>): ...this.  Use UNSPEC_MERGE_PTRUE instead of
 595         comparison-specific unspecs.
 596         (*vec_cmp<cmp_op>_<mode>_ptest): Rename to...
 597         (*cmp<cmp_op><mode>_ptest): ...this and adjust likewise.
 598         (*vec_cmp<cmp_op>_<mode>_cc): Rename to...
 599         (*cmp<cmp_op><mode>_cc): ...this and adjust likewise.
 600         (*vec_fcm<cmp_op><mode>): Rename to...
 601         (*fcm<cmp_op><mode>): ...this and adjust likewise.
 602         (*vec_fcmuo<mode>): Rename to...
 603         (*fcmuo<mode>): ...this and adjust likewise.
 604         (*pred_fcm<cmp_op><mode>): New pattern.
 605         * config/aarch64/aarch64.c (aarch64_emit_unop, aarch64_emit_binop)
 606         (aarch64_emit_sve_ptrue_op, aarch64_emit_sve_ptrue_op_cc): New
 607         functions.
 608         (aarch64_unspec_cond_code): Remove handling of LTU, GTU, LEU, GEU
 609         and UNORDERED.
 610         (aarch64_gen_unspec_cond, aarch64_emit_unspec_cond): Delete.
 611         (aarch64_emit_sve_predicated_cond): New function.
 612         (aarch64_expand_sve_vec_cmp_int): Use aarch64_emit_sve_ptrue_op_cc.
 613         (aarch64_emit_unspec_cond_or): Replace with...
 614         (aarch64_emit_sve_or_conds): ...this new function.  Use
 615         aarch64_emit_sve_ptrue_op for the individual comparisons and
 616         aarch64_emit_binop to OR them together.
 617         (aarch64_emit_inverted_unspec_cond): Replace with...
 618         (aarch64_emit_sve_inverted_cond): ...this new function.  Use
 619         aarch64_emit_sve_ptrue_op for the comparison and
 620         aarch64_emit_unop to invert the result.
 621         (aarch64_expand_sve_vec_cmp_float): Update after the above
 622         changes.  Use aarch64_emit_sve_ptrue_op for native comparisons.
 623
 624 2018-05-07  Nathan Sidwell  <nathan@acm.org>
 625
 626         * doc/invoke.texi (C++ Dialect Options): Remove -ffor-scope.
 627         * doc/extend.texi (Deprecated Features): Remove -fno-for-scope
 628         (Backwards Compatibility): Likewise.
 629
 630 2018-05-07  Luis Machado  <luis.machado@linaro.org>
 631
 632         PR bootstrap/85681
 633         Revert:
 634         2018-05-07  Luis Machado  <luis.machado@linaro.org>
 635
 636         * config/aarch64/aarch64-protos.h (cpu_prefetch_tune)
 637         <prefetch_dynamic_strides>: New const bool field.
 638         * config/aarch64/aarch64.c (generic_prefetch_tune): Update to include
 639         prefetch_dynamic_strides.
 640         (exynosm1_prefetch_tune): Likewise.
 641         (thunderxt88_prefetch_tune): Likewise.
 642         (thunderx_prefetch_tune): Likewise.
 643         (thunderx2t99_prefetch_tune): Likewise.
 644         (qdf24xx_prefetch_tune): Likewise. Set prefetch_dynamic_strides to false.
 645         (aarch64_override_options_internal): Update to set
 646         PARAM_PREFETCH_DYNAMIC_STRIDES.
 647         * doc/invoke.texi (prefetch-dynamic-strides): Document new option.
 648         * params.def (PARAM_PREFETCH_DYNAMIC_STRIDES): New.
 649         * params.h (PARAM_PREFETCH_DYNAMIC_STRIDES): Define.
 650         * tree-ssa-loop-prefetch.c (should_issue_prefetch_p): Account for
 651         prefetch-dynamic-strides setting.
 652
 653         2018-05-07  Luis Machado  <luis.machado@linaro.org>
 654
 655         * config/aarch64/aarch64-protos.h (cpu_prefetch_tune)
 656         <minimum_stride>: New const int field.
 657         * config/aarch64/aarch64.c (generic_prefetch_tune): Update to include
 658         minimum_stride field.
 659         (exynosm1_prefetch_tune): Likewise.
 660         (thunderxt88_prefetch_tune): Likewise.
 661         (thunderx_prefetch_tune): Likewise.
 662         (thunderx2t99_prefetch_tune): Likewise.
 663         (qdf24xx_prefetch_tune): Likewise. Set minimum_stride to 2048.
 664         (aarch64_override_options_internal): Update to set
 665         PARAM_PREFETCH_MINIMUM_STRIDE.
 666         * doc/invoke.texi (prefetch-minimum-stride): Document new option.
 667         * params.def (PARAM_PREFETCH_MINIMUM_STRIDE): New.
 668         * params.h (PARAM_PREFETCH_MINIMUM_STRIDE): Define.
 669         * tree-ssa-loop-prefetch.c (should_issue_prefetch_p): Return false if
 670         stride is constant and is below the minimum stride threshold.
 671
 672 2018-05-07  Luis Machado  <luis.machado@linaro.org>
 673
 674         * config/aarch64/aarch64.c (qdf24xx_prefetch_tune) <l2_cache_size>: Set
 675         to 512.
 676
 677 2018-05-07  Luis Machado  <luis.machado@linaro.org>
 678
 679         * config/aarch64/aarch64-protos.h (cpu_prefetch_tune)
 680         <prefetch_dynamic_strides>: New const bool field.
 681         * config/aarch64/aarch64.c (generic_prefetch_tune): Update to include
 682         prefetch_dynamic_strides.
 683         (exynosm1_prefetch_tune): Likewise.
 684         (thunderxt88_prefetch_tune): Likewise.
 685         (thunderx_prefetch_tune): Likewise.
 686         (thunderx2t99_prefetch_tune): Likewise.
 687         (qdf24xx_prefetch_tune): Likewise. Set prefetch_dynamic_strides to false.
 688         (aarch64_override_options_internal): Update to set
 689         PARAM_PREFETCH_DYNAMIC_STRIDES.
 690         * doc/invoke.texi (prefetch-dynamic-strides): Document new option.
 691         * params.def (PARAM_PREFETCH_DYNAMIC_STRIDES): New.
 692         * params.h (PARAM_PREFETCH_DYNAMIC_STRIDES): Define.
 693         * tree-ssa-loop-prefetch.c (should_issue_prefetch_p): Account for
 694         prefetch-dynamic-strides setting.
 695
 696 2018-05-07  Luis Machado  <luis.machado@linaro.org>
 697
 698         * config/aarch64/aarch64-protos.h (cpu_prefetch_tune)
 699         <minimum_stride>: New const int field.
 700         * config/aarch64/aarch64.c (generic_prefetch_tune): Update to include
 701         minimum_stride field.
 702         (exynosm1_prefetch_tune): Likewise.
 703         (thunderxt88_prefetch_tune): Likewise.
 704         (thunderx_prefetch_tune): Likewise.
 705         (thunderx2t99_prefetch_tune): Likewise.
 706         (qdf24xx_prefetch_tune): Likewise. Set minimum_stride to 2048.
 707         (aarch64_override_options_internal): Update to set
 708         PARAM_PREFETCH_MINIMUM_STRIDE.
 709         * doc/invoke.texi (prefetch-minimum-stride): Document new option.
 710         * params.def (PARAM_PREFETCH_MINIMUM_STRIDE): New.
 711         * params.h (PARAM_PREFETCH_MINIMUM_STRIDE): Define.
 712         * tree-ssa-loop-prefetch.c (should_issue_prefetch_p): Return false if
 713         stride is constant and is below the minimum stride threshold.
 714
 715 2018-05-06  Jakub Jelinek  <jakub@redhat.com>
 716
 717         PR c++/85659
 718         * cfgexpand.c (expand_asm_stmt): Don't create a temporary if
 719         the type is addressable.  Don't force op into register if it has
 720         BLKmode.
 721
 722 2018-05-05  Roland McGrath  <mcgrathr@google.com>
 723
 724         PR other/77609
 725         * varasm.c (default_section_type_flags): Set SECTION_NOTYPE for
 726         any section for which we don't know a specific type it should have,
 727         regardless of name.  Previously this was done only for the exact
 728         names ".init_array", ".fini_array", and ".preinit_array".
 729         (default_elf_asm_named_section): Add comment about
 730         relationship with default_section_type_flags and SECTION_NOTYPE.
 731         (get_section): Don't consider it a type conflict if one side has
 732         SECTION_NOTYPE and the other doesn't, as long as neither has the
 733         SECTION_BSS et al used in the default_section_type_flags logic.
 734
 735 2018-05-05  Tom de Vries  <tom@codesourcery.com>
 736
 737         PR target/85653
 738         * config/nvptx/nvptx.c (WORKAROUND_PTXJIT_BUG_3): Define.
 739         (workaround_barsyncs): New function.
 740         (nvptx_reorg): Use workaround_barsyncs.
 741         * config/nvptx/nvptx.md (define_c_enum "unspecv"): Add UNSPECV_MEMBAR.
 742         (define_expand "nvptx_membar_cta"): New define_expand.
 743         (define_insn "*nvptx_membar_cta"): New insn.
 744
 745 2018-05-04  Pekka Jääskeläinen  <pekka.jaaskelainen@parmance.com>
 746
 747         * brig-builtins.def: Add consts to ptrs etc. in BRIG builtin defs.
 748         To improve optimization opportunities.
 749         * builtin-types.def: The new needed builtin types for the above.
 750
 751 2018-05-04  Richard Biener  <rguenther@suse.de>
 752
 753         * bb-reorder.c (sanitize_hot_paths): Release hot_bbs_to_check.
 754         * gimple-ssa-store-merging.c
 755         (imm_store_chain_info::output_merged_store): Remove redundant create,
 756         release split_store vector contents on failure.
 757         * tree-vect-slp.c (vect_schedule_slp_instance): Avoid leaking
 758         scalar stmt vector on cache hit.
 759
 760 2018-05-04  Segher Boessenkool  <segher@kernel.crashing.org>
 761
 762         * common/config/rs6000/rs6000-common.c (rs6000_handle_option): Remove
 763         Xilinx FP support.
 764         * config.gcc (powerpc-xilinx-eabi*): Remove.
 765         * config/rs6000/predicates.md (easy_fp_constant): Remove Xilinx FP
 766         support.
 767         (fusion_addis_mem_combo_load): Ditto.
 768         * config/rs6000/rs6000-c.c (rs6000_target_modify_macros): Remove Xilinx
 769         FP support.
 770         (rs6000_cpu_cpp_builtins): Ditto.
 771         * config/rs6000/rs6000-linux.c
 772         (rs6000_linux_float_exceptions_rounding_supported_p): Ditto.
 773         * config/rs6000/rs6000-opts.h (enum fpu_type_t): Delete.
 774         * config/rs6000/rs6000.c (rs6000_debug_reg_global): Remove Xilinx FP
 775         support.
 776         (rs6000_setup_reg_addr_masks): Ditto.
 777         (rs6000_init_hard_regno_mode_ok): Ditto.
 778         (rs6000_option_override_internal): Ditto.
 779         (legitimate_lo_sum_address_p): Ditto.
 780         (rs6000_legitimize_address): Ditto.
 781         (rs6000_legitimize_reload_address): Ditto.
 782         (rs6000_legitimate_address_p): Ditto.
 783         (abi_v4_pass_in_fpr): Ditto.
 784         (setup_incoming_varargs): Ditto.
 785         (rs6000_gimplify_va_arg): Ditto.
 786         (rs6000_split_multireg_move): Ditto.
 787         (rs6000_savres_strategy): Ditto.
 788         (rs6000_emit_prologue_components): Ditto.
 789         (rs6000_emit_epilogue_components): Ditto.
 790         (rs6000_emit_prologue): Ditto.
 791         (rs6000_emit_epilogue): Ditto.
 792         (rs6000_elf_file_end): Ditto.
 793         (rs6000_function_value): Ditto.
 794         (rs6000_libcall_value): Ditto.
 795         * config/rs6000/rs6000.h: Ditto.
 796         (TARGET_MINMAX_SF, TARGET_MINMAX_DF): Delete, merge to ...
 797         (TARGET_MINMAX): ... this.  New.
 798         (TARGET_SF_FPR, TARGET_DF_FPR, TARGET_SF_INSN, TARGET_DF_INSN): Delete.
 799         * config/rs6000/rs6000.md: Remove Xilinx FP support.
 800         (*movsi_internal1_single): Delete.
 801         * config/rs6000/rs6000.opt (msingle-float, mdouble-float, msimple-fpu,
 802         mfpu=, mxilinx-fpu): Delete.
 803         * config/rs6000/singlefp.h: Delete.
 804         * config/rs6000/sysv4.h: Remove Xilinx FP support.
 805         * config/rs6000/t-rs6000: Ditto.
 806         * config/rs6000/t-xilinx: Delete.
 807         * gcc/config/rs6000/titan.md: Adjust for fp_type removal.
 808         * gcc/config/rs6000/vsx.md: Remove Xilinx FP support.
 809         (VStype_simple): Delete.
 810         (VSfptype_simple, VSfptype_mul, VSfptype_div, VSfptype_sqrt): Delete.
 811         * config/rs6000/xfpu.h: Delete.
 812         * config/rs6000/xfpu.md: Delete.
 813         * config/rs6000/xilinx.h: Delete.
 814         * config/rs6000/xilinx.opt: Delete.
 815         * gcc/doc/invoke.texi (RS/6000 and PowerPC Options): Remove
 816         -msingle-float, -mdouble-float, -msimple-fpu, -mfpu=, and -mxilinx-fpu.
 817
 818 2018-05-04  Tom de Vries  <tom@codesourcery.com>
 819
 820         PR libgomp/85639
 821         * builtins.c (expand_builtin_goacc_parlevel_id_size): Handle null target
 822         if ignore == 0.
 823
 824 2018-05-04  Richard Biener  <rguenther@suse.de>
 825
 826         PR middle-end/85627
 827         * tree-complex.c (update_complex_assignment): We are always in SSA form.
 828         (expand_complex_div_wide): Likewise.
 829         (expand_complex_operations_1): Likewise.
 830         (expand_complex_libcall): Preserve EH info of the original stmt.
 831         (tree_lower_complex): Handle removed blocks.
 832         * tree.c (build_common_builtin_nodes): Do not set ECF_NOTRHOW
 833         on complex multiplication and division libcall builtins.
 834
 835 2018-05-04  Richard Biener  <rguenther@suse.de>
 836
 837         PR middle-end/85574
 838         * fold-const.c (negate_expr_p): Restrict negation of operand
 839         zero of a division to when we know that can happen without
 840         overflow.
 841         (fold_negate_expr_1): Likewise.
 842
 843 2018-05-04  Jakub Jelinek  <jakub@redhat.com>
 844
 845         PR libstdc++/85466
 846         * real.h (real_nextafter): Declare.
 847         * real.c (real_nextafter): New function.
 848         * fold-const-call.c (fold_const_nextafter): New function.
 849         (fold_const_call_sss): Call it for CASE_CFN_NEXTAFTER and
 850         CASE_CFN_NEXTTOWARD.
 851         (fold_const_call_1): For CASE_CFN_NEXTTOWARD call fold_const_call_sss
 852         even when arg1_mode is different from arg0_mode.
 853
 854 2018-05-03  Nathan Sidwell  <nathan@acm.org>
 855
 856         * doc/extend.texi (Deprecated Features): Remove
 857         -ffriend-injection.
 858         (Backwards Compatibility): Likewise.
 859         * doc/invoke.texi (C++ Language Options): Likewise.
 860         (C++ Dialect Options): Likewise.
 861
 862 2018-05-03  Jakub Jelinek  <jakub@redhat.com>
 863
 864         PR target/85530
 865         * config/i386/avx512fintrin.h (_mm512_mullox_epi64,
 866         _mm512_mask_mullox_epi64): New intrinsics.
 867
 868 2018-05-03  Tom de Vries  <tom@codesourcery.com>
 869
 870         PR testsuite/85106
 871         * doc/sourcebuild.texi (Commands for use in dg-final, Scan optimization
 872         dump files): Add offload-tree.
 873
 874 2018-05-03  Richard Biener  <rguenther@suse.de>
 875
 876         PR tree-optimization/85615
 877         * tree-ssa-threadupdate.c (thread_block_1): Only allow exits
 878         to loops not nested in BBs loop father to avoid creating multi-entry
 879         loops.
 880
 881 2018-05-03  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
 882
 883         PR tree-optimization/70291
 884         * tree-complex.c (expand_complex_libcall): Add type, inplace_p
 885         arguments.  Change return type to tree.  Emit libcall as a new
 886         statement rather than replacing existing one when inplace_p is true.
 887         (expand_complex_multiplication_components): New function.
 888         (expand_complex_multiplication): Expand floating-point complex
 889         multiplication using the above.
 890         (expand_complex_division): Rename inner_type parameter to type.
 891         Update expand_complex_libcall call-site.
 892         (expand_complex_operations_1): Update expand_complex_multiplication
 893         and expand_complex_division call-sites.
 894
 895 2018-05-02  Jakub Jelinek  <jakub@redhat.com>
 896
 897         PR target/85582
 898         * config/i386/i386.md (*ashl<dwi>3_doubleword_mask,
 899         *ashl<dwi>3_doubleword_mask_1, *<shift_insn><dwi>3_doubleword_mask,
 900         *<shift_insn><dwi>3_doubleword_mask_1): In condition require that
 901         the highest significant bit of the shift count mask is clear.  In
 902         check whether and[sq]i3 is needed verify that all significant bits
 903         of the shift count other than the highest are set.
 904
 905 2018-05-02  Tom de Vries  <tom@codesourcery.com>
 906
 907         PR libgomp/82428
 908         * builtins.def (DEF_GOACC_BUILTIN_ONLY): Define.
 909         * omp-builtins.def (BUILT_IN_GOACC_PARLEVEL_ID)
 910         (BUILT_IN_GOACC_PARLEVEL_SIZE): New builtin.
 911         * builtins.c (expand_builtin_goacc_parlevel_id_size): New function.
 912         (expand_builtin): Call expand_builtin_goacc_parlevel_id_size.
 913         * doc/extend.texi (Other Builtins): Add __builtin_goacc_parlevel_id and
 914         __builtin_goacc_parlevel_size.
 915
 916 2018-05-02  Richard Biener  <rguenther@suse.de>
 917
 918         PR tree-optimization/85597
 919         * tree-vect-stmts.c (vectorizable_operation): For ternary SLP
 920         do not use split vect_get_vec_defs call but call vect_get_slp_defs
 921         directly.
 922
 923 2018-05-02  Tom de Vries  <tom@codesourcery.com>
 924
 925         PR testsuite/85106
 926         * doc/sourcebuild.texi (Commands for use in dg-final, Scan optimization
 927         dump files): Add ltrans-tree.
 928
 929 2018-05-02  Tom de Vries  <tom@codesourcery.com>
 930
 931         PR testsuite/85106
 932         * doc/sourcebuild.texi (Commands for use in dg-final, Scan optimization
 933         dump files): Add wpa-ipa.
 934
 935 2018-05-02  Segher Boessenkool  <segher@kernel.crashing.org>
 936
 937         * config.gcc (powerpc*-*-*): Remove paired.h.  Unsupport the
 938         powerpc*-*-linux*paired* target.
 939         * config/rs6000/750cl.h: Delete.
 940         * config/rs6000/paired.h: Delete.
 941         * config/rs6000/paired.md: Delete.
 942         * config/rs6000/predicates.md (easy_vector_constant): Remove paired
 943         float support.
 944         * config/rs6000/rs6000-builtin.def: Remove paired float support.
 945         * config/rs6000/rs6000-c.c (rs6000_target_modify_macros): Update
 946         comment.  Remove paired float support.
 947         * config/rs6000/rs6000-modes.def: Remove V2SF and V2SI.
 948         * config/rs6000/rs6000-opts.h (enum rs6000_vector): Delete
 949         VECTOR_PAIRED.
 950         * config/rs6000/rs6000-protos.h (paired_expand_vector_init,
 951         paired_emit_vector_cond_expr, paired_expand_vector_move): Delete
 952         declarations.
 953         * config/rs6000/rs6000.c: Remove paired float support.
 954         (paired_expand_vector_init, paired_expand_vector_move,
 955         paired_emit_vector_compare, paired_emit_vector_cond_expr,
 956         (paired_expand_lv_builtin, paired_expand_stv_builtin,
 957         paired_expand_builtin, paired_expand_predicate_builtin,
 958         paired_init_builtins): Delete.
 959         * config/rs6000/rs6000.h: Remove paired float support.
 960         * config/rs6000/rs6000.md: Remove paired float support.
 961         (move_from_CR_ov_bit): Delete.
 962         * config/rs6000/rs6000.opt (mpaired): Delete.
 963         * config/rs6000/t-rs6000: Remove paired.md from MD_INCLUDES.
 964         * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mpaired.
 965
 966 2018-05-02  Richard Biener  <rguenther@suse.de>
 967
 968         PR middle-end/85567
 969         * gimplify.c (gimplify_save_expr): When in SSA form allow
 970         SAVE_EXPRs to compute to SSA vars.
 971
 972 2018-05-02  Jakub Jelinek  <jakub@redhat.com>
 973
 974         PR target/85582
 975         * config/i386/i386.md (*ashl<dwi>3_doubleword_mask,
 976         *ashl<dwi>3_doubleword_mask_1, *<shift_insn><dwi>3_doubleword_mask,
 977         *<shift_insn><dwi>3_doubleword_mask_1): If and[sq]i3 is needed, don't
 978         clobber operands[2], instead use a new pseudo.  Formatting fixes.
 979
 980 2018-05-02  Richard Sandiford  <richard.sandiford@linaro.org>
 981
 982         PR tree-optimization/85586
 983         * tree-vect-data-refs.c (vect_analyze_data_ref_dependence): Only
 984         exit early for statements in the same group if the accesses are
 985         not strided.
 986
 987 2018-05-02  Tom de Vries  <tom@codesourcery.com>
 988
 989         PR lto/85451
 990         * lto-wrapper.c (compile_offload_image): Add "could not find mkoffload"
 991         error message.
 992
 993 2018-05-01  Marc Glisse  <marc.glisse@inria.fr>
 994
 995         PR tree-optimization/85143
 996         * match.pd (A<B&A<C): Extend to BIT_IOR_EXPR.
 997
 998 2018-05-01  Tom de Vries  <tom@codesourcery.com>
 999
1000         PR lto/85451
1001         * config/nvptx/mkoffload.c (main): Suggest using -B in "offload compiler
1002         not found" error message.
1003
1004 2018-05-01  Tom de Vries  <tom@codesourcery.com>
1005
1006         PR other/83786
1007         * vec.h (VEC_ORDERED_REMOVE_IF, VEC_ORDERED_REMOVE_IF_FROM_TO): Define.
1008         * vec.c (test_ordered_remove_if): New function.
1009         (vec_c_tests): Call test_ordered_remove_if.
1010         * dwarf2cfi.c (connect_traces): Use VEC_ORDERED_REMOVE_IF_FROM_TO.
1011         * lto-streamer-out.c (prune_offload_funcs): Use VEC_ORDERED_REMOVE_IF.
1012         * tree-vect-patterns.c (vect_pattern_recog_1): Use
1013         VEC_ORDERED_REMOVE_IF.
1014
1015 2018-05-01  Prathamesh Kulkarni  <prathamesh.kulkarni@linaro.org>
1016
1017         PR tree-optimization/82665
1018         * vr-values.c (vr_values::extract_range_from_binary_expr): Handle
1019         pointer subtraction where arguments come from a memchr call.
1020
1021 2018-05-01  Jakub Jelinek  <jakub@redhat.com>
1022
1023         * configure.ac (LD_AS_NEEDED_OPTION, LD_NO_AS_NEEDED_OPTION): Use
1024         --push-state --as-needed and --pop-state instead of --as-needed and
1025         --no-as-needed if ld supports it.
1026         * configure: Regenerated.
1027
1028         PR web/85578
1029         * doc/install.texi2html: Replace _002d with - and _002a with * in
1030         generated html files using sed.
1031
1032 2018-04-30  David Malcolm  <dmalcolm@redhat.com>
1033
1034         PR c++/85523
1035         * gcc-rich-location.c (blank_line_before_p): New function.
1036         (use_new_line): New function.
1037         (gcc_rich_location::add_fixit_insert_formatted): New function.
1038         * gcc-rich-location.h
1039         (gcc_rich_location::add_fixit_insert_formatted): New function.
1040
1041 2018-04-30  David Malcolm  <dmalcolm@redhat.com>
1042
1043         * selftest.c (assert_streq): Rename "expected" and "actual" to
1044         "val1" and "val2".  Extend NULL-handling to cover both inputs
1045         symmetrically, while still requiring both to be non-NULL for a pass.
1046         * selftest.h (assert_streq): Rename "expected" and "actual" to
1047         "val1" and "val2".
1048         (ASSERT_EQ): Likewise.
1049         (ASSERT_EQ_AT): Likewise.
1050         (ASSERT_KNOWN_EQ): Likewise.
1051         (ASSERT_KNOWN_EQ_AT): Likewise.
1052         (ASSERT_NE): Likewise.
1053         (ASSERT_MAYBE_NE): Likewise.
1054         (ASSERT_MAYBE_NE_AT): Likewise.
1055         (ASSERT_STREQ): Likewise.  Clarify that both must be non-NULL for
1056         the assertion to pass.
1057         (ASSERT_STREQ_AT): Likewise.
1058
1059 2018-04-30  Jonathan Wakely  <jwakely@redhat.com>
1060
1061         * doc/invoke.texi (-fpie, -fPIE): Fix grammar and clarify
1062         interaction with -pie.
1063
1064 2018-04-30  David Malcolm  <dmalcolm@redhat.com>
1065
1066         * selftest.h: Fix alphabetization of per-source-file selftest
1067         declarations.
1068
1069 2018-04-30  Jason Merrill  <jason@redhat.com>
1070
1071         PR c++/61982 - dead stores to destroyed objects.
1072         * gimplify.c (gimplify_modify_expr): Simplify complex lvalue on LHS
1073         of clobber.
1074
1075 2018-04-30  Jason Merrill  <jason@redhat.com>
1076
1077         * tree.c (build_clobber): New.
1078         * tree.h: Declare it.
1079         * gimplify.c (gimplify_bind_expr, gimplify_target_expr): Use it.
1080
1081 2018-04-30  David Malcolm  <dmalcolm@redhat.com>
1082
1083         * diagnostic-show-locus.c (layout::layout): Update for
1084         location_get_source_line returning a char_span.
1085         (struct char_span): Move to input.h.
1086         (struct correction): Update for fields in char_span becoming
1087         private.
1088         (struct source_line): Update for location_get_source_line
1089         returning a char_span.
1090         (layout::print_line): Likewise.
1091         * edit-context.c (edited_file::print_content): Likewise.
1092         (edited_file::print_diff_hunk): Likewise.
1093         (edited_file::print_run_of_changed_lines): Likewise.
1094         (edited_file::get_num_lines): Likewise.
1095         (edited_line::edited_line): Likewise.
1096         * final.c (asm_show_source): Likewise.
1097         * input.c (location_get_source_line): Convert return type
1098         from const char * to char_span, losing the final "line_len"
1099         param.
1100         (dump_location_info): Update for the above.
1101         (get_substring_ranges_for_loc): Likewise.  Use a char_span
1102         when handling the literal within the line.
1103         (test_reading_source_line): Update for location_get_source_line
1104         returning a char_span.
1105         * input.h (class char_span): Move here from
1106         diagnostic-show-locus.c, converting from a struct to a class.
1107         Make data members private.
1108         (char_span::operator bool): New.
1109         (char_span::length): New.
1110         (char_span::get_buffer): New.
1111         (char_span::operator[]): New.
1112         (char_span::subspan): Make const.
1113         (char_span::xstrdup): New.
1114         (location_get_source_line): Convert return type from const char *
1115         to char_span, losing the final "line_size" param.
1116
1117 2018-04-30  Jan Hubicka  <jh@suse.cz>
1118
1119         * lto-wrapper.c (ltrans_priorities): New static var.
1120         (cmp_priority): New.
1121         (run_gcc): Read priorities and if doing parallel build order
1122         the Makefile by them.
1123
1124 2018-04-30  David Malcolm  <dmalcolm@redhat.com>
1125
1126         * input.h (builtins_location_check): Convert to a STATIC_ASSERT.
1127
1128 2018-04-30  Richard Biener  <rguenther@suse.de>
1129
1130         * tree-cfg.c (verify_address): Remove base argument, add
1131         flag whether to check TREE_ADDRESSABLE and do that.
1132         (verify_expr): Remove.
1133         (verify_types_in_gimple_reference): Add pieces from verify_expr.
1134         (verify_gimple_assign_single): Likewise.
1135         (verify_gimple_switch): Likewise.
1136         (verify_expr_location_1): Dereference tp once.  Add (disabled)
1137         piece from verify_expr.
1138         (verify_gimple_in_cfg): Do not call verify_expr on all ops.
1139
1140 2018-04-30  Claudiu Zissulescu  <claziss@synopsys.com>
1141
1142         * config/arc/linux.h (CLEAR_INSN_CACHE): Define.
1143
1144 2018-04-30  Claudiu Zissulescu  <claziss@synopsys.com>
1145
1146         * config/arc/arc-protos.h (prepare_extend_operands): Remove.
1147         (small_data_pattern): Likewise.
1148         (arc_rewrite_small_data): Likewise.
1149         * config/arc/arc.c (LEGITIMATE_SMALL_DATA_OFFSET_P): Remove.
1150         (LEGITIMATE_SMALL_DATA_ADDRESS_P): Likewise.
1151         (get_symbol_alignment): New function.
1152         (legitimate_small_data_address_p): Likewise.
1153         (legitimate_scaled_address): Update, call
1154         legitimate_small_data_address_p.
1155         (output_sdata): New static variable.
1156         (arc_print_operand): Update how we handle small data operands.
1157         (arc_print_operand_address): Likewise.
1158         (arc_legitimate_address_p): Update, use
1159         legitimate_small_data_address_p.
1160         (arc_rewrite_small_data_p): Remove.
1161         (arc_rewrite_small_data_1): Likewise.
1162         (arc_rewrite_small_data): Likewise.
1163         (small_data_pattern): Likewise.
1164         (compact_sda_memory_operand): Update to use
1165         legitimate_small_data_address_p and get_symbol_alignment.
1166         (prepare_move_operands): Don't rewite sdata pattern.
1167         (prepare_extend_operands): Remove.
1168         * config/arc/arc.md (zero_extendqihi2): Don't rewrite sdata
1169         pattern.
1170         (zero_extendqisi2): Likewise.
1171         (zero_extendhisi2): Likewise.
1172         (extendqihi2): Likewise.
1173         (extendqisi2): Likewise.
1174         (extendhisi2): Likewise.
1175         (addsi3): Likewise.
1176         (subsi3): Likewise.
1177         (andsi3): Likewise.
1178         * config/arc/constraints.md (Usd): Change it to memory constraint.
1179
1180 2018-04-30  Claudiu Zissulescu  <claziss@synopsys.com>
1181
1182         * config/arc/arc.c (arc_split_move): Allow signed 6-bit constants
1183         as source of std instructions.
1184         * config/arc/arc.md (movsi_insn): Update pattern predicate to
1185         allow 6-bit constants as source for store instructions.
1186         (movdi_insn): Update instruction pattern to allow 6-bit constants
1187         as source for store instructions.
1188
1189 2018-04-30  Jonathan Wakely  <jwakely@redhat.com>
1190
1191         * doc/invoke.texi (-fdebug-types-section): Fix grammar.
1192
1193 2018-04-30  Nathan Sidwell  <nathan@acm.org>
1194             Sandra Loosemore <sandra@codesourcery.com>
1195
1196         * dumpfile.c (dump_open): Allow '-' for stdout.
1197         * doc/invoke.texi (Developer Options): Document dump filename
1198         determination early.  Document stdin/stdout selection.
1199
1200 2018-04-30 Andrew Sadek  <andrew.sadek.se@gmail.com>
1201
1202         Microblaze Target: PIC data text relative
1203
1204         * config/microblaze/microblaze.opt: add new option -mpic-data-text-rel.
1205         * config/microblaze/microblaze-protos.h (microblaze_constant_address_p):
1206         Add declaration.
1207         * gcc/config/microblaze/microblaze.h (microblaze_constant_address_p):
1208         CONSTANT_ADDRESS_P definition to microblaze_constant_address_p.
1209         * config/microblaze/microblaze.c (TARGET_PIC_DATA_TEXT_REL):
1210         New addressing mode for data-text relative position indepenedent code.
1211         (microblaze_classify_unspec): add 'UNSPEC_TEXT' case ->
1212         'ADDRESS_SYMBOLIC_TXT_REL'.
1213         (microblaze_classify_address): Add handling for UNSPEC + CONST_INT.
1214         (microblaze_legitimate_pic_operand): Exclude function calls from
1215         pic operands in case of TARGET_PIC_DATA_TEXT_REL option.
1216         (microblaze_legitimize_address): Generate 'UNSPEC_TEXT' for all possible
1217         addresses cases.
1218         (microblaze_address_insns): Add 'ADDRESS_SYMBOLIC_TXT_REL' case.
1219         (print_operand): Add 'ADDRESS_SYMBOLIC_TXT_REL' case.
1220         (print_operand_address): Add 'ADDRESS_SYMBOLIC_TXT_REL' case + handling
1221         for 'address + offset'.
1222         (microblaze_expand_prologue): Add new function prologue call for
1223         'r20' assignation.
1224         (microblaze_asm_generate_pic_addr_dif_vec): Override new target hook
1225         'TARGET_ASM_GENERATE_PIC_ADDR_DIFF_VEC' to disable address diff vector
1226         table in case of TARGET_PIC_DATA_TEXT_REL.
1227         (expand_pic_symbol_ref): Add handling for 'UNSPEC_TEXT'.
1228         * gcc/config/microblaze/microblaze.md (TARGET_PIC_DATA_TEXT_REL):
1229         Add new macros 'UNSPEC_TEXT',
1230         'UNSPEC_SET_TEXT' + add rule for setting r20 in function prologue
1231         + exclude function calls from 'UNSPEC_PLT' in case of data text
1232         relative mode.
1233         * doc/tm.texi.in (TARGET_ASM_GENERATE_PIC_ADDR_DIFF_VEC): Add
1234         new target hook for generating address diff vector tables in case of
1235         flag_pic.
1236         * doc/tm.texi : Regenerate.
1237         * stmt.c (TARGET_ASM_GENERATE_PIC_ADDR_DIFF_VEC): Append new condition
1238         'targetm.asm_out.generate_pic_addr_diff_vec' to flag_pic in case
1239         of addr diff vector generation.
1240         * target.def (TARGET_ASM_GENERATE_PIC_ADDR_DIFF_VEC): Add
1241         target hook definition.
1242         * targhooks.h, gcc/targhooks.c (TARGET_ASM_GENERATE_PIC_ADDR_DIFF_VEC):
1243         Add default function for generate_pic_addr_diff_vec -> flag_pic.
1244         * doc/invoke.texi (Add new pic option): Add new microblaze pic
1245         option for data text relative.
1246
1247 2018-04-30  Richard Biener  <rguenther@suse.de>
1248
1249         * tree-chrec.h (evolution_function_is_constant_p): Remove
1250         redundant check.
1251         * tree-cfg.c (tree_node_can_be_shared): Re-order checks.
1252
1253 2018-04-30  Richard Biener  <rguenther@suse.de>
1254
1255         PR bootstrap/85571
1256         * dwarf2out.c (gen_producer_string): Ignore -fchecking[=].
1257
1258 2018-04-30  Richard Biener  <rguenther@suse.de>
1259
1260         PR tree-optimization/28364
1261         PR tree-optimization/85275
1262         * tree-ssa-loop-ch.c (ch_base::copy_headers): Stop after
1263         copying first exit test.
1264
1265 2018-04-28  Mark Wielaard  <mark@klomp.org>
1266
1267         * dwarf2out.c (dwarf2out_finish): Add .debug_addr table header for
1268         dwarf_version >= 5.
1269         (dwarf_AT): Handle DW_AT_addr_base.
1270         (add_top_level_skeleton_die_attrs): Use dwarf_AT for DW_AT_addr_base.
1271
1272 2018-04-28  Uros Bizjak  <ubizjak@gmail.com>
1273
1274         PR target/84431
1275         * config/i386/i386.md (*ashl<dwi>3_doubleword_mask): New pattern.
1276         (*ashl<dwi>3_doubleword_mask_1): Ditto.
1277         (*<shift_insn><dwi>3_doubleword_mask): Ditto.
1278         (*<shift_insn><dwi>3_doubleword_mask_1): Ditto.
1279
1280 2018-04-28  Richard Biener  <rguenther@suse.de>
1281
1282         * tree-cfg.c (verify_gimple_phi): Take a gphi * argument.
1283         (verify_gimple_in_cfg): Rename visited_stmts to visited_throwing_stmts
1284         to reflect use.  Only add interesting stmts.
1285
1286 2018-04-27  Martin Jambor  <mjambor@suse.cz>
1287
1288         PR ipa/85549
1289         * ipa-cp.c (find_aggregate_values_for_callers_subset): Make sure
1290         the jump function allows for passing through aggregate values.
1291
1292 2018-04-27  David Malcolm  <dmalcolm@redhat.com>
1293
1294         * input.h (in_system_header_at): Convert from macro to inline
1295         function.
1296         (from_macro_expansion_at): Likewise.
1297         (from_macro_definition_at): Likewise.
1298
1299 2018-04-27  Jeff Law  <law@redhat.com>
1300
1301         * config.gcc: Mark tile* targets as deprecated/obsolete.
1302
1303 2018-04-27  Richard Biener  <rguenther@suse.de>
1304
1305         * config/aarch64/aarch64.c: Simplify ap.__stack advance and
1306         fix for ILP32.
1307
1308 2018-04-27  Richard Biener  <rguenther@suse.de>
1309
1310         * tree-cfg.c (verify_expr): Make dead code hit gcc_unreachable.
1311
1312 2018-04-27  Uros Bizjak  <ubizjak@gmail.com>
1313
1314         * config/i386/i386.md (*movti_internal): Substitute Ye constraint
1315         with Yd constraint. Set "preferred_for_speed" attribute from
1316         TARGET_INTER_UNIT_MOVES_{FROM,TO}_VEC for alternatives
1317         with Yd constraint.
1318         (*movdi_internal): Ditto.
1319         (movti_interunit splitters): Remove
1320         TARGET_INTER_UNIT_MOVES_{FROM,TO}_VEC from insn condition.
1321         (movdi_interunit splitters): Ditto.
1322         * config/i386/constraints.md (Ye): Remove.
1323         (Yd): Do not depend on TARGET_INTER_UNIT_MOVES_{FROM,TO}_VEC.
1324
1325 2018-04-27  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
1326
1327         PR target/85512
1328         * config/aarch64/constraints.md (Usg): Limit to 31.
1329         (Usj): Limit to 63.
1330
1331 2018-04-27  Jakub Jelinek  <jakub@redhat.com>
1332
1333         PR tree-optimization/85529
1334         * tree-ssa-reassoc.c (optimize_range_tests_var_bound): Add FIRST_BB
1335         argument.  Don't call get_nonzero_bits if opcode is ERROR_MARK_NODE,
1336         rhs2 def stmt's bb is dominated by first_bb and it isn't an obvious
1337         zero extension or masking of the MSB bit.
1338         (optimize_range_tests): Add FIRST_BB argument, pass it through
1339         to optimize_range_tests_var_bound.
1340         (maybe_optimize_range_tests, reassociate_bb): Adjust
1341         optimize_range_tests callers.
1342
1343 2018-04-26  Richard Biener  <rguenther@suse.de>
1344             Jakub Jelinek  <jakub@redhat.com>
1345
1346         * cgraph.h (symbol_table): Just declare debug method here.
1347         * symtab.c (symbol_table::debug): Define.
1348
1349 2018-04-26  Eric Botcazou  <ebotcazou@adacore.com>
1350
1351         * loop-invariant.c (may_assign_reg_p): Return false for frame pointer.
1352
1353 2018-04-26  Uros Bizjak  <ubizjak@gmail.com>
1354
1355         * config/i386/i386.md ("isa" attribute): Add x64_sse2.
1356         ("enabled" attribute): Handle x64_sse2 "isa" attribute.
1357         (*movdi_internal): Substitute Yi and Yj constraint with x
1358         and Ym and Yn constraint with y constraint.  Update "isa"
1359         attribute and set "preferred_for_speed" attribute from
1360         TARGET_INTER_UNIT_MOVES_{FROM,TO}_VEC for updated alternatives.
1361         (*movsi_internal): Ditto.
1362         (*movdf_internal): Ditto.
1363         (*movsf_internal): Ditto.
1364         (*zero_extendsidi2): Ditto.
1365         * config/i386/sse.md (vec_set<mode>_0): Ditto.
1366         (sse2_loadld): Ditto.
1367         (*vec_extract<ssevecmodelower>_0): Ditto.
1368         (*vec_extractv4si_0_zext_sse4): Ditto.
1369         (vec_concatv2di): Ditto.
1370         (*vec_dup<mode>): Ditto.
1371         * config/i386/mmx.md (*mov<mode>_internal): Ditto.
1372         * config/i386/constraints.md (Yi): Remove.
1373         (Yj): Remove.
1374         (Ym): Remove.
1375         (Yn): Remove.
1376
1377 2018-04-26  Nathan Sidwell  <nathan@acm.org>
1378
1379         * dumpfile.c (dump_open): New.
1380         (dump_open_alternate_stream, dump_start, dump_begin): Call it.
1381         (dump_finish): Detect stdio/stderr by value not name.
1382
1383 2018-04-26  Jonathan Wakely  <jwakely@redhat.com>
1384
1385         * doc/invoke.texi (-Wreturn-type): Document default status for C++.
1386
1387 2018-04-26  Tom de Vries  <tom@codesourcery.com>
1388
1389         PR target/84952
1390         * config/nvptx/nvptx.c (verify_neutering_jumps)
1391         (verify_neutering_labels): New function
1392         (nvptx_single): Use verify_neutering_jumps and verify_neutering_labels.
1393
1394 2018-04-26  Tom de Vries  <tom@codesourcery.com>
1395
1396         PR target/84025
1397         * config/nvptx/nvptx.c (needs_neutering_p): New function.
1398         (nvptx_single): Use needs_neutering_p to skip over insns that do not
1399         need neutering.
1400
1401 2018-04-26  Richard Biener <rguenther@suse.de>
1402             Tom de Vries  <tom@codesourcery.com>
1403
1404         PR lto/85422
1405         * lto-streamer-out.c (output_function): Fixup loops if required to match
1406         discovery done in the reader.
1407
1408 2018-04-26  Richard Biener  <rguenther@suse.de>
1409
1410         PR tree-optimization/85116
1411         * tree-ssa-loop-ch.c (do_while_loop_p): A do-while loop should
1412         have a loop exit from the single latch predecessor.  Remove
1413         case of header with just condition.
1414         (ch_base::copy_headers): Exclude infinite loops from any
1415         processing.
1416         (pass_ch::execute): Record exits.
1417
1418 2018-04-26  Richard Biener  <rguenther@suse.de>
1419
1420         * tree-vect-data-refs.c (vect_get_data_access_cost): Get
1421         prologue cost vector and pass it to vect_get_load_cost.
1422         (vect_get_peeling_costs_all_drs): Likewise.
1423         (vect_peeling_hash_get_lowest_cost): Likewise.
1424         (vect_enhance_data_refs_alignment): Likewise.
1425
1426 2018-04-26  Richard Biener  <rguenther@suse.de>
1427
1428         PR middle-end/85450
1429         * tree-cfg.c (verify_gimple_assign_unary): Restore proper
1430         checking of integer<->pointer conversions.
1431         * omp-expand.c (expand_omp_for_static_nochunk): Avoid
1432         sign-/zero-extending pointer types.
1433         (expand_omp_for_static_chunk): Likewise.
1434
1435 2018-03-22  Hans-Peter Nilsson  <hp@axis.com>
1436             Jean Lee  <xiaoyur347@gmail.com>
1437
1438         * config/mips/mips.c (mips_asan_shadow_offset): New function.
1439         (TARGET_ASAN_SHADOW_OFFSET): Define.
1440         * config/mips/mips.h (FRAME_GROWS_DOWNWARD): Augment to also be
1441         true for -fsanitize=address.
1442
1443 2018-04-25  Mark Wielaard  <mark@klomp.org>
1444
1445         * dwarf2out.c (file_info_cmp): Sort longer dir prefixes before
1446         shorter ones.
1447
1448 2018-04-25  Jakub Jelinek  <jakub@redhat.com>
1449
1450         * config/i386/i386.md (*x86_mov<mode>cc_0_m1): Use type "alu1" rather
1451         than "alu", remove explicit "memory" and "imm_disp" attributes.
1452         (*x86_mov<mode>cc_0_m1_se, *x86_mov<mode>cc_0_m1_neg): Likewise.
1453
1454         PR middle-end/85414
1455         * simplify-rtx.c (simplify_unary_operation_1) <case SIGN_EXTEND,
1456         case ZERO_EXTEND>: Pass SUBREG_REG (op) rather than op to
1457         gen_lowpart_no_emit.
1458
1459 2018-04-25  Sebastian Peryt  <sebastian.peryt@intel.com>
1460
1461         PR target/85473
1462         * config/i386/i386.c (ix86_expand_builtin): Change memory
1463         operand to XI, extend p0 to Pmode.
1464         * config/i386/i386.md: Change unspec volatile and operand
1465         1 mode to XI, change operand 0 mode to P.
1466
1467 2018-04-25  Chung-Ju Wu  <jasonwucj@gmail.com>
1468
1469         * config/nds32/nds32-predicates.c (nds32_can_use_bclr_p): Mask with
1470         GET_MODE_MASK before any checking.
1471         (nds32_can_use_bset_p): Likewise.
1472         (nds32_can_use_btgl_p): Likewise.
1473
1474 2018-04-25  Chung-Ju Wu  <jasonwucj@gmail.com>
1475
1476         * config/nds32/nds32-doubleword.md: New define_split pattern for
1477         illegal register number.
1478
1479 2018-04-25  Chung-Ju Wu  <jasonwucj@gmail.com>
1480
1481         * config/nds32/nds32.c (nds32_print_operand): Set op_value ealier.
1482
1483 2018-04-25  Chung-Ju Wu  <jasonwucj@gmail.com>
1484
1485         * config/nds32/nds32.h (ASM_APP_ON): Add missing newline character.
1486
1487 2018-04-25  Richard Biener  <rguenther@suse.de>
1488
1489         * lto-streamer.h (LTO_major_version): Bump to 8.
1490
1491 2018-04-25  Jakub Jelinek  <jakub@redhat.com>
1492
1493         * BASE-VER: Set to 9.0.0.
1494
1495 2018-04-24  Segher Boessenkool  <segher@kernel.crashing.org>
1496
1497         * config/rs6000/rs6000.c (init_float128_ieee): Fix spelling mistakes
1498         in __abskf2 and __powikf2.
1499
1500 2018-04-24  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
1501
1502         PR target/85512
1503         * config/aarch64/constraints.md (Usg, Usj): New constraints.
1504         * config/aarch64/iterators.md (cmode_simd): New mode attribute.
1505         * config/aarch64/aarch64.md (*aarch64_ashr_sisd_or_int_<mode>3):
1506         Use the above on operand 2.  Reindent.
1507         (*aarch64_lshr_sisd_or_int_<mode>3): Likewise.
1508
1509 2018-04-24  H.J. Lu  <hongjiu.lu@intel.com>
1510
1511         PR target/85485
1512         * common/config/i386/i386-common.c (ix86_handle_option): Don't
1513         handle OPT_mcet.
1514         * config/i386/i386.opt (mcet): Removed.
1515         * doc/install.texi: Remove -mcet documentation.
1516         * doc/invoke.texi: Likewise.
1517
1518 2018-04-24  H.J. Lu  <hongjiu.lu@intel.com>
1519
1520         PR target/85485
1521         * doc/install.texi: Remove -mcet from bootstrap-cet.
1522
1523 2018-04-24  Jakub Jelinek  <jakub@redhat.com>
1524
1525         PR target/85511
1526         * config/i386/i386.c (ix86_init_mmx_sse_builtins): Don't define
1527         __builtin_ia32_readeflags_u32 and __builtin_ia32_writeeflags_u32
1528         if TARGET_64BIT.
1529
1530         PR target/85503
1531         * config/rs6000/rs6000-p8swap.c (const_load_sequence_p): Punt if
1532         const_vector is not CONST_VECTOR or SYMBOL_REF for a constant pool
1533         containing a CONST_VECTOR.
1534
1535 2018-04-24  Cesar Philippidis  <cesar@codesourcery.com>
1536
1537         * doc/install.texi: Update newlib dependency for nvptx.
1538
1539 2018-04-24  Jakub Jelinek  <jakub@redhat.com>
1540
1541         PR target/85508
1542         * config/i386/i386.c (ix86_expand_vector_init_one_var): Use UINTVAL
1543         instead of INTVAL when shifting x left.
1544
1545 2018-04-24  Andreas Krebbel  <krebbel@linux.ibm.com>
1546
1547         PR tree-optimization/85478
1548         * tree-vect-loop.c (vect_analyze_loop_2): Do not call
1549         vect_grouped_store_supported for single element vectors.
1550
1551 2018-04-24  Richard Biener  <rguenther@suse.de>
1552
1553         PR target/85491
1554         * config/i386/i386.c (ix86_add_stmt_cost): Restrict strided
1555         load cost increase to the case of non-constant step.
1556
1557 2018-04-24  Jakub Jelinek  <jakub@redhat.com>
1558
1559         PR target/84828
1560         * reg-stack.c (move_for_stack_reg): Don't fail assertion about dead
1561         destination if any_malformed_asm.
1562
1563 2018-04-23  Eric Botcazou  <ebotcazou@adacore.com>
1564
1565         PR middle-end/85496
1566         * expr.c (store_field): In the bitfield case, if the value comes from
1567         a function call and is returned in registers by means of a PARALLEL,
1568         do not change the mode of the temporary unless BLKmode and VOIDmode.
1569
1570 2018-04-23  Andrey Belevantsev  <abel@ispras.ru>
1571
1572         PR rtl-optimization/85423
1573         * sel-sched-ir.c (has_dependence_note_mem_dep): Only discard
1574         dependencies to debug insns when the previous insn is non-debug.
1575
1576 2018-04-23  Claudiu Zissulescu  <claziss@synopsys.com>
1577
1578         * config/arc/arc.md ("vunspec"): Delete it, unify all the unspec
1579         enums into a single definition.
1580         (fls): Fix predicates and printing.
1581         (seti): Likewise.
1582
1583 2018-04-23  Claudiu Zissulescu  <claziss@synopsys.com>
1584
1585         * config/arc/arc-protos.h (check_if_valid_sleep_operand): Remove.
1586         * config/arc/arc.c (arc_expand_builtin): Sleep accepts registers
1587         and short u6 immediate.
1588         (check_if_valid_sleep_operand): Remove.
1589         * config/arc/arc.md (Sleep): Accepts registers and u6 immediates.
1590
1591 2018-04-22  Chung-Ju Wu  <jasonwucj@gmail.com>
1592
1593         * config/nds32/nds32.c (nds32_compute_stack_frame): Consider
1594         flag_always_save_lp condition.
1595         * config/nds32/nds32.opt (malways-save-lp): New option.
1596
1597 2018-04-22  Shiva Chen  <shiva0217@gmail.com>
1598
1599         * config/nds32/nds32-protos.h (nds32_use_load_post_increment): Declare.
1600         * config/nds32/nds32.c (nds32_use_load_post_increment): New.
1601         * config/nds32/nds32.h
1602         (USE_LOAD_POST_INCREMENT, USE_LOAD_POST_DECREMENT): Define.
1603         (USE_STORE_POST_INCREMENT, USE_STORE_POST_DECREMENT): Define.
1604
1605 2018-04-22  Shiva Chen  <shiva0217@gmail.com>
1606
1607         * config/nds32/nds32-protos.h (nds32_ls_333_p): Remove.
1608         * config/nds32/nds32.c (nds32_ls_333_p): Remove.
1609
1610 2018-04-22  Shiva Chen  <shiva0217@gmail.com>
1611             Chung-Ju Wu  <jasonwucj@gmail.com>
1612
1613         * config/nds32/nds32-protos.h (nds32_case_vector_shorten_mode):
1614         Declare.
1615         * config/nds32/nds32.c (nds32_case_vector_shorten_mode): New function.
1616         * config/nds32/nds32.h (CASE_VECTOR_SHORTEN_MODE): Modify.
1617
1618 2018-04-22  Chung-Ju Wu  <jasonwucj@gmail.com>
1619
1620         * config/nds32/nds32.c (nds32_compute_stack_frame): Fix wrong value.
1621
1622 2018-04-22  Chung-Ju Wu  <jasonwucj@gmail.com>
1623
1624         * config/nds32/nds32-protos.h (nds32_data_alignment,
1625         nds32_local_alignment): Declare.
1626         * config/nds32/nds32.c (nds32_data_alignment, nds32_constant_alignment,
1627         nds32_local_alignment): New functions.
1628         (TARGET_CONSTANT_ALIGNMENT): Define.
1629         * config/nds32/nds32.h (DATA_ALIGNMENT, LOCAL_ALIGNMENT): Define.
1630
1631 2018-04-22  Chung-Ju Wu  <jasonwucj@gmail.com>
1632
1633         * config/nds32/nds32.c
1634         (TARGET_HARD_REGNO_MODE_OK): Move to the bottom of file.
1635         (TARGET_MODES_TIEABLE_P): Likewise.
1636
1637 2018-04-22  Chung-Ju Wu  <jasonwucj@gmail.com>
1638
1639         * config/nds32/nds32.c (nds32_asm_file_start): Display optimization
1640         level Ofast and Og.
1641
1642 2018-04-22  Monk Chiang  <sh.chiang04@gmail.com>
1643             Chung-Ju Wu  <jasonwucj@gmail.com>
1644
1645         * config/nds32/constants.md (unspec_volatile_element): Add enum values
1646         for unaligned access.
1647         * config/nds32/nds32-intrinsic.c: Implementation of expanding
1648         unaligned access.
1649         * config/nds32/nds32-intrinsic.md: Likewise.
1650         * config/nds32/nds32_intrinsic.h: Likewise.
1651         * config/nds32/nds32.h (nds32_builtins): Likewise.
1652         * config/nds32/nds32.opt (munaligned-access): New option.
1653         * config/nds32/nds32.c (nds32_asm_file_start): Display
1654         flag_unaligned_access status.
1655
1656 2018-04-20  Kito Cheng  <kito.cheng@gmail.com>
1657
1658         * config/riscv/elf.h (LINK_SPEC): Pass --no-relax if
1659         -mno-relax is present.
1660         * config/riscv/linux.h (LINK_SPEC): Ditto.
1661
1662 2018-04-20  Martin Sebor  <msebor@redhat.com>
1663
1664         PR c/85365
1665         * gimple-fold.c (gimple_fold_builtin_strcpy): Suppress -Wrestrict
1666         for null pointers.
1667         (gimple_fold_builtin_stxcpy_chk): Same.
1668         * gimple-ssa-warn-restrict.c (check_bounds_or_overlap): Same.
1669
1670 2018-04-20  Michael Meissner  <meissner@linux.ibm.com>
1671
1672         PR target/85456
1673         * config/rs6000/rs6000.c (init_float128_ieee): Add support to call
1674         __powikf2 when long double is IEEE 128-bit.
1675
1676 2018-04-20  Kito Cheng  <kito.cheng@gmail.com>
1677
1678         * config/riscv/riscv.c (riscv_first_stack_step): Round up min
1679         step to make sure stack always aligned.
1680
1681 2018-04-20  Carl Love  <cel@us.ibm.com>
1682
1683         PR target/83402
1684         * config/rs6000/rs6000-c.c (rs6000_gimple_fold_builtin): Add
1685         size check for arg0.
1686
1687 2018-04-20  Nathan Sidwell  <nathan@codesourcery.com>
1688             Tom de Vries  <tom@codesourcery.com>
1689
1690         PR target/85445
1691         * config/nvptx/nvptx.c (nvptx_emit_forking, nvptx_emit_joining):
1692         Emit insns for calls too.
1693         (nvptx_find_par): Always look for worker-level predecessor insn.
1694         (nvptx_propagate): Add is_call parm, return bool.  Copy frame for
1695         calls.
1696         (nvptx_vpropagate, nvptx_wpropagate): Adjust.
1697         (nvptx_process_pars): Propagate frames for calls.
1698
1699 2018-04-20  H.J. Lu  <hongjiu.lu@intel.com>
1700
1701         PR target/85469
1702         * common/config/i386/i386-common.c (OPTION_MASK_ISA_IBT_SET):
1703         Removed.
1704         (OPTION_MASK_ISA_IBT_UNSET): Likewise.
1705         (ix86_handle_option): Don't handle OPT_mibt.
1706         * config/i386/cet.h: Check __CET__ instead of __IBT__ and
1707         __SHSTK__.
1708         * config/i386/driver-i386.c (host_detect_local_cpu): Remove
1709         has_ibt and ibt.
1710         * config/i386/i386-c.c (ix86_target_macros_internal): Don't
1711         check OPTION_MASK_ISA_IBT nor flag_cf_protection.
1712         (ix86_target_macros): Define __CET__ with flag_cf_protection
1713         for -fcf-protection.
1714         * config/i386/i386.c (isa2_opts): Remove -mibt.
1715         * config/i386/i386.h (TARGET_IBT): Removed.
1716         (TARGET_IBT_P): Likewise.
1717         (ix86_valid_target_attribute_inner_p): Don't check OPT_mibt.
1718         * config/i386/i386.md (nop_endbr): Don't check TARGET_IBT.
1719         * config/i386/i386.opt (mcet): Update help message.
1720         (mshstk): Likewise.
1721         (mibt): Removed.
1722         * doc/invoke.texi: Remove -mibt.  Document __CET__.  Document
1723         -mcet as an alias for -mshstk.
1724
1725 2018-04-20  Richard Biener <rguenther@suse.de>
1726
1727         PR middle-end/85475
1728         * match.pd ((X * CST) * Y -> (X * Y) * CST): Avoid exponential
1729         complexity by forcing a single use of the multiply operand.
1730
1731 2018-04-20  Martin Jambor  <mjambor@suse.cz>
1732
1733         ipa/85449
1734         * ipa-cp.c (cgraph_edge_brings_value_p): Move check for self-feeding
1735         recursion dependency to only apply to non-clones.
1736
1737 2018-04-20  Martin Jambor  <mjambor@suse.cz>
1738
1739         ipa/85447
1740         * ipa-cp.c (create_specialized_node): Check that clones of
1741         self-recursive edges exist during IPA-CP.
1742
1743 2018-04-19  Toon Moene  <toon@moene.org>
1744
1745         * doc/invoke.texi: Add -floop-unroll-and-jam to options enabled
1746         by -O3.
1747
1748 2018-04-19  Jakub Jelinek  <jakub@redhat.com>
1749
1750         PR tree-optimization/85467
1751         * fold-const.c (fold_ternary_loc) <case BIT_FIELD_REF>: Use
1752         VECTOR_TYPE_P macro.  If type is vector type, VIEW_CONVERT_EXPR the
1753         VECTOR_CST element to type.
1754
1755 2018-04-19  H.J. Lu  <hongjiu.lu@intel.com>
1756
1757         PR target/85397
1758         * config/i386/i386.h (STACK_SAVEAREA_MODE): New.
1759         * config/i386/i386.md (builtin_setjmp_setup): Removed.
1760         (builtin_longjmp): Likewise.
1761         (save_stack_nonlocal): New pattern.
1762         (restore_stack_nonlocal): Likewise.
1763
1764 2018-04-19  H.J. Lu  <hongjiu.lu@intel.com>
1765
1766         PR target/85404
1767         * config/i386/cet.c (file_end_indicate_exec_stack_and_cet):
1768         Replace ASM_OUTPUT_LABEL with fprintf.
1769
1770 2018-04-19  H.J. Lu  <hongjiu.lu@intel.com>
1771
1772         PR target/85417
1773         * config/i386/cet.c (file_end_indicate_exec_stack_and_cet):
1774         Check flag_cf_protection instead of TARGET_IBT and TARGET_SHSTK.
1775         * config/i386/i386-c.c (ix86_target_macros_internal): Also
1776         define __IBT__ and __SHSTK__ for -fcf-protection.
1777         * config/i386/i386.c (pass_insert_endbranch::gate): Don't check
1778         TARGET_IBT.
1779         (ix86_trampoline_init): Likewise.
1780         (x86_output_mi_thunk): Likewise.
1781         (ix86_notrack_prefixed_insn_p): Likewise.
1782         (ix86_option_override_internal): Don't disallow -fcf-protection.
1783         * config/i386/i386.md (rdssp<mode>): Also enable for
1784         -fcf-protection.
1785         (incssp<mode>): Likewise.
1786         (nop_endbr): Likewise.
1787         * config/i386/i386.opt (mcet): Change help message to built-in
1788         functions only.
1789         (mibt): Likewise.
1790         (mshstk): Likewise.
1791         * doc/invoke.texi: Remove -mcet, -mibt and -mshstk condition
1792         on -fcf-protection.  Change -mcet, -mibt and -mshstk to only
1793         enable CET built-in functions.
1794
1795 2018-04-19  Sebastian Peryt  <sebastian.peryt@intel.com>
1796
1797         * common/config/i386/i386-common.c
1798         (OPTION_MASK_ISA_MOVDIRI_SET, OPTION_MASK_ISA_MOVDIR64B_SET,
1799         OPTION_MASK_ISA_MOVDIRI_UNSET,
1800         OPTION_MASK_ISA_MOVDIR64B_UNSET): New defines.
1801         (ix86_handle_option): Handle -mmovdiri and -mmovdir64b.
1802         * config.gcc (movdirintrin.h): New header.
1803         * config/i386/cpuid.h (bit_MOVDIRI,
1804         bit_MOVDIR64B): New bits.
1805         * config/i386/driver-i386.c (host_detect_local_cpu): Detect -mmovdiri
1806         and -mmvodir64b.
1807         * config/i386/i386-builtin-types.def ((VOID, PUNSIGNED, UNSIGNED),
1808         (VOID, PVOID, PCVOID)): New function types.
1809         * config/i386/i386-builtin.def (__builtin_ia32_directstoreu_u32,
1810         __builtin_ia32_directstoreu_u64,
1811          __builtin_ia32_movdir64b): New builtins.
1812         * config/i386/i386-c.c (__MOVDIRI__, __MOVDIR64B__): New.
1813         * config/i386/i386.c (ix86_target_string): Added -mmovdir64b
1814         and -mmovdiri.
1815         (ix86_valid_target_attribute_inner_p): Ditto.
1816         (ix86_expand_special_args_builtin): Added VOID_FTYPE_PUNSIGNED_UNSIGNED
1817         and VOID_FTYPE_PUNSIGNED_UNSIGNED.
1818         (ix86_expand_builtin): Expand IX86_BUILTIN_MOVDIR64B.
1819         * config/i386/i386.h (TARGET_MOVDIRI, TARGET_MOVDIRI_P,
1820         TARGET_MOVDIR64B, TARGET_MOVDIR64B_P): New.
1821         * config/i386/i386.md (UNSPECV_MOVDIRI, UNSPECV_MOVDIR64B): New.
1822         (movdiri<mode>, movdir64b_<mode>): New.
1823         * config/i386/i386.opt: Add -mmovdiri and -mmovdir64b.
1824         * config/i386/immintrin.h: Include movdirintrin.h.
1825         * config/i386/movdirintrin.h: New file.
1826         * doc/invoke.texi: Added -mmovdiri and -mmovdir64b.
1827
1828 2018-04-19  Richard Biener  <rguenther@suse.de>
1829
1830         PR middle-end/85455
1831         * cfg.c (clear_bb_flags): When loop state says we have
1832         marked irreducible regions also preserve BB_IRREDUCIBLE_LOOP.
1833
1834 2018-04-19  Richard Biener  <rguenther@suse.de>
1835
1836         PR tree-optimization/84737
1837         * tree-vect-data-refs.c (vect_copy_ref_info): New function
1838         copying restrict info.
1839         (vect_setup_realignment): Use it.
1840         * tree-vectorizer.h (vect_copy_ref_info): Declare.
1841         * tree-vect-stmts.c (vectorizable_store): Copy ref info from
1842         the first DR to all generated stores.
1843         (vectorizable_load): Likewise for loads.
1844
1845 2018-04-19  Jakub Jelinek  <jakub@redhat.com>
1846
1847         PR tree-optimization/85446
1848         * match.pd ((intptr_t) x eq/ne CST to x eq/ne (typeof x) cst): Require
1849         the integral and pointer types to have the same precision.
1850
1851         * doc/install.texi: Document --disable-cet being the default and
1852         --enable-cet=auto.
1853
1854 2018-04-18  Martin Liska  <mliska@suse.cz>
1855
1856         * ipa-devirt.c (odr_subtypes_equivalent_p): Fix GNU coding
1857         style.
1858
1859 2018-04-18  Martin Liska  <mliska@suse.cz>
1860
1861         Revert
1862         2018-03-02  Eric Botcazou  <ebotcazou@adacore.com>
1863
1864         PR ipa/83983
1865         * ipa-devirt.c (odr_subtypes_equivalent_p): Get the ODR type of both
1866         arguments if they are comparable.
1867
1868 2018-04-18  Martin Liska  <mliska@suse.cz>
1869
1870         Revert
1871         2018-03-13  Eric Botcazou  <ebotcazou@adacore.com>
1872
1873         PR lto/84805
1874         * ipa-devirt.c (odr_subtypes_equivalent_p): Do not get the ODR type of
1875         incomplete types.
1876
1877 2018-04-18  H.J. Lu  <hongjiu.lu@intel.com>
1878
1879         PR target/85388
1880         * config/i386/i386.c (ix86_expand_split_stack_prologue): Insert
1881         ENDBR after calling __morestack.
1882
1883 2018-04-18  David Malcolm  <dmalcolm@redhat.com>
1884
1885         PR jit/85384
1886         * configure.ac (gcc-driver-name.h): Honor --with-gcc-major-version
1887         by using gcc_base_ver to generate a gcc_driver_version, and use
1888         it when generating GCC_DRIVER_NAME.
1889         * configure: Regenerate.
1890
1891 2018-04-18  Jakub Jelinek  <jakub@redhat.com>
1892
1893         PR target/81084
1894         * config.gcc: Obsolete powerpc*-*-*spe*.
1895
1896 2018-04-17  Jakub Jelinek  <jakub@redhat.com>
1897
1898         PR debug/84637
1899         * dbxout.c (dbxout_int): Perform negation in unsigned int type.
1900         (stabstr_D): Change type of unum from unsigned int to
1901         unsigned HOST_WIDE_INT.  Perform negation in unsigned HOST_WIDE_INT
1902         type.
1903
1904 2018-04-17  Jim Wilson  <jimw@sifive.com>
1905
1906         PR 84856
1907         * config/riscv/riscv.c (riscv_compute_frame_info): Add calls to
1908         RISCV_STACK_ALIGN when using outgoing_args_size and pretend_args_size.
1909         Set arg_pointer_offset after using pretend_args_size.
1910
1911 2018-04-17  Jakub Jelinek  <jakub@redhat.com>
1912
1913         PR rtl-optimization/85431
1914         * dse.c (record_store): Ignore zero width stores.
1915
1916         PR sanitizer/85230
1917         * asan.c (handle_builtin_stack_restore): Adjust comment.  Emit
1918         __asan_allocas_unpoison call and last_alloca_addr = new_sp before
1919         __builtin_stack_restore rather than after it.
1920         * builtins.c (expand_asan_emit_allocas_unpoison): Pass
1921         arg1 + (virtual_dynamic_stack_rtx - stack_pointer_rtx) as second
1922         argument instead of virtual_dynamic_stack_rtx.
1923
1924 2018-04-17  Kelvin Nilsen  <kelvin@gcc.gnu.org>
1925
1926         * config/rs6000/rs6000-protos.h (rs6000_builtin_is_supported_p):
1927         New prototype.
1928         * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin):
1929         Add note to error message to explain internal mapping of overloaded
1930         built-in function name to non-overloaded built-in function name.
1931         * config/rs6000/rs6000.c (rs6000_builtin_is_supported_p): New
1932         function.
1933
1934 2018-04-17  Michael Meissner  <meissner@linux.vnet.ibm.com>
1935
1936         PR target/85424
1937         * config/rs6000/rs6000.md (pack<mode>): Do not try handle a pack
1938         where the inputs overlap with the output.
1939
1940 2018-04-17  Jakub Jelinek  <jakub@redhat.com>
1941
1942         * config/i386/sse.md (vec_extract_lo_<mode><mask_name>): Add
1943         (=v, v) alternative and explicit "memory" attribute.
1944         (vec_extract_lo_<mode><mask_name>): Likewise.  Also add
1945         "type", "prefix", "prefix_extra", "length_immediate" and "mode"
1946         attributes.
1947         (vec_extract_lo_<mode><mask_name>): Add (=v, v) alternative and use
1948         "sselog1" type instead of "sselog".
1949         (vec_extract_hi_<mode><mask_name>): Use "sselog1" type instead of
1950         "sselog".  Remove explicit "memory" attribute.
1951         (vec_extract_lo_v32hi): Add (=v, v) alternative and explicit "memory",
1952         "type", "prefix", "prefix_extra", "length_immediate" and "mode"
1953         attributes.
1954         (vec_extract_hi_v32hi): Merge all alternatives into one, use
1955         "sselog1" type instead of "sselog".  Remove explicit "memory"
1956         attribute.
1957         (vec_extract_hi_v16hi): Merge each pair of alternatives into one,
1958         use "sselog1" type instead of "sselog".  Remove explicit "memory"
1959         attribute.
1960         (vec_extract_lo_v64qi): Add (=v, v) alternative and explicit "memory",
1961         "type", "prefix", "prefix_extra", "length_immediate" and "mode"
1962         attributes.
1963         (vec_extract_hi_v64qi): Merge all alternatives into one, use
1964         "sselog1" type instead of "sselog".  Remove explicit "memory"
1965         attribute.
1966         (vec_extract_hi_v32qi): Merge each pair of alternatives into one,
1967         use "sselog1" type instead of "sselog".  Remove explicit "memory"
1968         attribute.
1969
1970         PR target/85430
1971         * config/i386/i386.md (*ashlqi3_1_slp): Use alu1 type instead of alu.
1972
1973         PR middle-end/85414
1974         * rtlhooks.c (gen_lowpart_if_possible): Don't call gen_lowpart_SUBREG
1975         on a SUBREG.
1976
1977 2018-04-17  Martin Jambor  <mjambor@suse.cz>
1978
1979         PR ipa/85421
1980         * ipa-cp.c (create_specialized_node): Call
1981         expand_all_artificial_thunks if necessary.
1982
1983 2018-04-17  Martin Liska  <mliska@suse.cz>
1984
1985         PR lto/85405
1986         * ipa-devirt.c (odr_types_equivalent_p): Remove trailing
1987         in message, remote space in between '_G' and '('.
1988
1989 2018-04-17  Jakub Jelinek  <jakub@redhat.com>
1990
1991         PR target/85281
1992         * config/i386/sse.md (reduces<mode><mask_scalar_name>,
1993         avx512f_vmcmp<mode>3<round_saeonly_name>,
1994         avx512f_vmcmp<mode>3_mask<round_saeonly_name>,
1995         avx512f_sgetexp<mode><mask_scalar_name><round_saeonly_scalar_name>,
1996         avx512f_rndscale<mode><round_saeonly_name>,
1997         avx512dq_ranges<mode><mask_scalar_name><round_saeonly_scalar_name>,
1998         avx512f_vgetmant<mode><mask_scalar_name><round_saeonly_scalar_name>):
1999         Use %<iptr>2 instead of %2 for -masm=intel.
2000         (avx512f_vcvtss2usi<round_name>, avx512f_vcvtss2usiq<round_name>,
2001         avx512f_vcvttss2usi<round_saeonly_name>,
2002         avx512f_vcvttss2usiq<round_saeonly_name>): Use %k1 instead of %1 for
2003         -masm=intel.
2004         (avx512f_vcvtsd2usi<round_name>, avx512f_vcvtsd2usiq<round_name>,
2005         avx512f_vcvttsd2usi<round_saeonly_name>,
2006         avx512f_vcvttsd2usiq<round_saeonly_name>, ufloatv2siv2df2<mask_name>):
2007         Use %q1 instead of %1 for -masm=intel.
2008         (avx512f_sfixupimm<mode><sd_maskz_name><round_saeonly_name>,
2009         avx512f_sfixupimm<mode>_mask<round_saeonly_name>): Use %<iptr>3 instead
2010         of %3 for -masm=intel.
2011         (sse2_shufpd_v2df_mask): Fix a typo, change %{6%} to %{%6%} for
2012         -masm=intel.
2013         (*avx512vl_<code>v2div2qi2_store): Use %w0 instead of %0 for
2014         -masm=intel.
2015         (*avx512vl_<code><mode>v4qi2_store): Use %k0 instead of %0 for
2016         -masm=intel.
2017         (avx512vl_<code><mode>v4qi2_mask_store): Use a single pattern with
2018         %k0 and %1 for -masm=intel rather than two patterns, one with %0 and
2019         %g1.
2020         (*avx512vl_<code><mode>v8qi2_store): Use %q0 instead of %0 for
2021         -masm=intel.
2022         (avx512vl_<code><mode>v8qi2_mask_store): Use a single pattern with
2023         %q0 and %1 for -masm=intel rather than two patterns, one with %0 and
2024         %g1 and one with %0 and %1.
2025         (avx512er_vmrcp28<mode><round_saeonly_name>,
2026         avx512er_vmrsqrt28<mode><round_saeonly_name>): Use %<iptr>1 instead of
2027         %1 for -masm=intel.
2028         (avx5124fmaddps_4fmaddps_mask, avx5124fmaddps_4fmaddss_mask,
2029         avx5124fmaddps_4fnmaddps_mask, avx5124fmaddps_4fnmaddss_mask,
2030         avx5124vnniw_vp4dpwssd_mask, avx5124vnniw_vp4dpwssds_mask): Swap order
2031         of %0 and %{%4%} for -masm=intel.
2032         (avx5124fmaddps_4fmaddps_maskz, avx5124fmaddps_4fmaddss_maskz,
2033         avx5124fmaddps_4fnmaddps_maskz, avx5124fmaddps_4fnmaddss_maskz,
2034         avx5124vnniw_vp4dpwssd_maskz, avx5124vnniw_vp4dpwssds_maskz): Swap
2035         order of %0 and %{%5%}%{z%} for -masm=intel.
2036
2037 2018-04-17  Jan Hubicka  <jh@suse.cz>
2038
2039         PR lto/85405
2040         * ipa-devirt.c (odr_types_equivalent_p): Handle bit fields.
2041
2042 2018-04-17  Martin Liska  <mliska@suse.cz>
2043
2044         PR ipa/85329
2045         * multiple_target.c (create_dispatcher_calls): Set apostrophes
2046         for target_clone error message.  Make default implementation
2047         clone to be a local declaration.
2048         (separate_attrs): Add new argument and check for an empty
2049         string.
2050         (expand_target_clones): Handle it.
2051         (ipa_target_clone): Make redirection just for target_clones
2052         functions.
2053
2054 2018-04-16  Cesar Philippidis  <cesar@codesourcery.com>
2055             Tom de Vries  <tom@codesourcery.com>
2056
2057         PR middle-end/84955
2058         * omp-expand.c (expand_oacc_for): Add dummy false branch for
2059         tiled basic blocks without omp continue statements.
2060
2061 2018-04-16  Aaron Sawdey  <acsawdey@linux.ibm.com>
2062
2063         PR target/83660
2064         * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Mark
2065         vec_extract expression as having side effects to make sure it gets
2066         a cleanup point.
2067
2068 2018-04-16  H.J. Lu  <hongjiu.lu@intel.com>
2069
2070         PR target/85403
2071         * config/i386/i386.c (get_builtin_code_for_version): Check
2072         error_mark_node.
2073
2074 2018-04-16  Olga Makhotina  <olga.makhotina@intel.com>
2075
2076         PR target/84331
2077         * gcc/config.gcc: Support "skylake".
2078         * gcc/config/i386/i386-c.c (ix86_target_macros_internal): Handle
2079         PROCESSOR_SKYLAKE.
2080         * gcc/config/i386/i386.c (m_SKYLAKE): Define.
2081         (processor_target_table): Add "skylake".
2082         (ix86_option_override_internal): Add "skylake".
2083         (get_builtin_code_for_version): Handle PROCESSOR_SKYLAKE,
2084         PROCESSOR_CANNONLAKE.
2085         (get_builtin_code_for_version): Fix priority for
2086         PROCESSOR_ICELAKE_CLIENT, PROCESSOR_ICELAKE_SERVER,
2087         PROCESSOR_SKYLAKE-AVX512.
2088         * gcc/config/i386/i386.h (processor_costs): Define TARGET_SKYLAKE.
2089         (processor_type): Add PROCESSOR_SKYLAKE.
2090
2091 2018-04-16  Paolo Carlini  <paolo.carlini@oracle.com>
2092             Jason Merrill  <jason@redhat.com>
2093
2094         PR c++/85112
2095         * convert.c (convert_to_integer_1): Use direct recursion for
2096         enumeral types and types with a precision less than the number
2097         of bits in their mode.
2098
2099 2018-04-16  Julia Koval  <julia.koval@intel.com>
2100
2101         PR target/84413
2102         * config/i386/x86-tune.def (X86_TUNE_SSE_UNALIGNED_LOAD_OPTIMAL,
2103         X86_TUNE_SSE_UNALIGNED_STORE_OPTIMAL): Add m_SKYLAKE_AVX512
2104
2105 2018-04-14  Segher Boessenkool  <segher@kernel.crashing.org>
2106
2107         PR target/85293
2108         * config/rs6000/rs6000.opt (mdirect-move): Make deprecated.
2109         * doc/invoke.texi (RS/6000 and PowerPC Options): Remove -mdirect-move
2110         and -mno-direct-move.
2111
2112 2018-04-13  Paul A. Clarke  <pc@us.ibm.com>
2113
2114         PR target/83402
2115         * config/rs6000/emmintrin.h (_mm_slli_epi{16,32,64}):
2116         Ensure that vec_splat_s32 is only called with 0 <= shift < 16.
2117         Ensure negative shifts result in {0}.
2118
2119 2018-04-13  Vladimir Makarov  <vmakarov@redhat.com>
2120
2121         PR rtl-optimization/79916
2122         * config/rs6000/rs6000.c (rs6000_emit_move): Use assigned hard
2123         regs (if any) to define how to gnerate SD moves when LRA is in
2124         progress.
2125
2126 2018-04-13  Jakub Jelinek  <jakub@redhat.com>
2127
2128         PR rtl-optimization/85393
2129         * except.h (expand_dw2_landing_pad_for_region): Remove declaration.
2130         * except.c (expand_dw2_landing_pad_for_region): Make static.
2131         * bb-reorder.c (fix_up_crossing_landing_pad): In new_bb emit just
2132         a label and unconditional jump to old_bb, rather than
2133         expand_dw2_landing_pad_for_region insn(s) and jump to single_succ
2134         basic block.
2135
2136         PR rtl-optimization/85376
2137         * simplify-rtx.c (simplify_const_unary_operation): For CLZ and CTZ and
2138         zero op0, if C?Z_DEFINED_VALUE_AT_ZERO is false, return NULL_RTX
2139         instead of a specific value.
2140
2141 2018-04-13  Jan Hubicka  <hubicka@ucw.cz>
2142             Bin Cheng  <bin.cheng@arm.com>
2143
2144         PR tree-optimization/82965
2145         PR tree-optimization/83991
2146         * cfgloopanal.c (expected_loop_iterations_unbounded): Add
2147         by_profile_only parameter.
2148         * cfgloopmanip.c (scale_loop_profile): Further scale loop's profile
2149         information if the loop was predicted to iterate too many times.
2150         * cfgloop.h (expected_loop_iterations_unbounded): Update prototype
2151
2152 2018-04-13  Jan Hubicka  <hubicka@ucw.cz>
2153
2154         PR lto/71991
2155         * config/i386/i386.c (ix86_can_inline_p): Allow safe transitions for
2156         always inline.
2157
2158 2018-04-13  Martin Liska  <mliska@suse.cz>
2159             Jakub Jelinek  <jakub@redhat.com>
2160
2161         PR middle-end/81657
2162         * expr.h (enum block_op_methods): Add BLOCK_OP_NO_LIBCALL_RET.
2163         * expr.c (emit_block_move_hints): Handle BLOCK_OP_NO_LIBCALL_RET.
2164         * builtins.c (expand_builtin_memory_copy_args): Use
2165         BLOCK_OP_NO_LIBCALL_RET method for mempcpy with non-ignored target,
2166         handle dest_addr == pc_rtx.
2167
2168 2018-04-12  Segher Boessenkool  <segher@kernel.crashing.org>
2169
2170         PR target/85291
2171         * config/rs6000/rs6000.md (fix_trunc<mode>si2): Use legacy code if
2172         asked to not generate direct moves.
2173         (fix_trunc<mode>si2_stfiwx): Similar.
2174         (fix_trunc<mode>si2_internal): Similar.
2175
2176 2018-04-12  Jakub Jelinek  <jakub@redhat.com>
2177
2178         PR debug/83157
2179         * var-tracking.c (add_stores): Handle STRICT_LOW_PART SET_DEST.
2180         * cselib.c (cselib_record_sets): For STRICT_LOW_PART dest,
2181         lookup if dest in some wider mode is known to be const0_rtx and
2182         if so, record permanent equivalence for it to be ZERO_EXTEND of
2183         the narrower mode destination.
2184
2185 2018-04-12  Cesar Philippidis  <cesar@codesourcery.com>
2186
2187         * lto-streamer-out.c (output_function): Revert 259346.
2188         * omp-expand.c (expand_oacc_for): Likewise.
2189
2190 2018-04-12  Alexander Monakov  <amonakov@ispras.ru>
2191
2192         PR rtl-optimization/85354
2193         * sel-sched-ir.c (sel_init_pipelining): Move cfg_cleanup call...
2194         * sel-sched.c (sel_global_init): ... here.
2195
2196 2018-04-12  Eric Botcazou  <ebotcazou@adacore.com>
2197
2198         PR target/85238
2199         * lto-wrapper.c (debug_objcopy): Open the files in binary mode.
2200         * dwarf2out.c (dwarf2out_early_finish): Do not generate assembly in LTO
2201         mode for PE-COFF targets.
2202         * config/i386/i386-protos.h (i386_pe_asm_lto_start): Declare.
2203         (i386_pe_asm_lto_end): Likewise.
2204         * config/i386/cygming.h (TARGET_ASM_LTO_START): Define.
2205         (TARGET_ASM_LTO_END): Likewise.
2206         * config/i386/winnt.c (saved_debug_info_level): New static variable.
2207         (i386_pe_asm_lto_start): New function.
2208         (i386_pe_asm_lto_end): Likewise.
2209
2210 2018-04-12  Cesar Philippidis  <cesar@codesourcery.com>
2211             Richard Biener  <rguenther@suse.de>
2212
2213         PR middle-end/84955
2214         * lto-streamer-out.c (output_function): Fix CFG loop state before
2215         streaming out.
2216         * omp-expand.c (expand_oacc_for): Handle calls to internal
2217         functions like regular functions.
2218
2219 2018-04-12  Richard Biener  <rguenther@suse.de>
2220
2221         PR lto/85371
2222         * dwarf2out.c (init_sections_and_labels): Use debug_line_section[_label]
2223         for the early LTO debug to properly generate references to it
2224         during DIE emission.  Do not re-use that for the skeleton for
2225         split-dwarf.
2226         (dwarf2out_early_finish): Likewise.
2227
2228 2018-04-12  Jakub Jelinek  <jakub@redhat.com>
2229
2230         PR target/85328
2231         * config/i386/sse.md
2232         (<mask_codefor>avx512dq_vextract<shuffletype>64x2_1<mask_name> split,
2233         <mask_codefor>avx512f_vextract<shuffletype>32x4_1<mask_name> split,
2234         vec_extract_lo_<mode><mask_name> split, vec_extract_lo_v32hi,
2235         vec_extract_lo_v64qi): For non-AVX512VL if input is xmm16+ reg
2236         and output is a reg, avoid creating invalid lowpart subreg, but
2237         instead split into a 512-bit move.  Don't split if not AVX512VL,
2238         input is xmm16+ reg and output is a mem.
2239         (vec_extract_lo_<mode><mask_name>, vec_extract_lo_v32hi,
2240         vec_extract_lo_v64qi): Don't require split if not AVX512VL, input is
2241         xmm16+ reg and output is a mem.
2242
2243 2018-04-12  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>
2244
2245         * config/s390/s390.c (s390_output_indirect_thunk_function): Check
2246         also for flag_dwarf2_cfi_asm.
2247
2248 2018-04-12  Jakub Jelinek  <jakub@redhat.com>
2249
2250         PR rtl-optimization/85342
2251         * regcprop.c (copyprop_hardreg_forward_1): Remove replaced array, use
2252         a bool scalar var inside of the loop instead.  Don't try to update
2253         recog_data.operand after failed apply_change_group.
2254
2255 2018-04-12  Tom de Vries  <tom@codesourcery.com>
2256
2257         PR target/85296
2258         * config/nvptx/nvptx.c (flexible_array_member_type_p): New function.
2259         (nvptx_assemble_decl_begin): Add undefined param.  Declare undefined
2260         array with flexible array member as array without given dimension.
2261         (nvptx_assemble_undefined_decl): Set nvptx_assemble_decl_begin call
2262         argument for undefined param to true.
2263
2264 2018-04-11  Aaron Sawdey  <acsawdey@linux.ibm.com>
2265
2266         PR target/85321
2267         * doc/invoke.texi (RS/6000 and PowerPC Options): Document options
2268         -mcall- and -mtraceback=. Remove options -mabi=spe and -mabi=no-spe
2269         from PowerPC section.
2270         * config/rs6000/sysv4.opt (mcall-): Improve help text.
2271         * config/rs6000/rs6000.opt (mblock-compare-inline-limit=): Trim
2272         help text that is too long.
2273         * config/rs6000/rs6000.opt (mblock-compare-inline-loop-limit=): Trim
2274         help text that is too long.
2275         * config/rs6000/rs6000.opt (mstring-compare-inline-limit=): Trim
2276         help text that is too long.
2277
2278 2018-04-11  Uros Bizjak  <ubizjak@gmail.com>
2279
2280         * config/alpha/alpha.md (stack_probe_internal): Rename
2281         from "probe_stack".  Update all callers.
2282
2283 2018-04-11  Alexander Monakov  <amonakov@ispras.ru>
2284
2285         PR rtl-optimization/84566
2286         * sched-deps.c (sched_analyze_insn): Check deps->readonly when invoking
2287         sched_macro_fuse_insns.
2288
2289 2018-04-11  Alexander Monakov  <amonakov@ispras.ru>
2290
2291         PR target/84301
2292         * sched-rgn.c (add_branch_dependences): Move sel_sched_p check here...
2293         (compute_block_dependences): ... from here.
2294
2295 2018-04-11  Jakub Jelinek  <jakub@redhat.com>
2296
2297         PR tree-optimization/85331
2298         * vec-perm-indices.h (vec_perm_indices::clamp): Change input type
2299         from int to HOST_WIDE_INT.
2300
2301 2018-04-11  Martin Jambor  <mjambor@suse.cz>
2302
2303         PR ipa/84149
2304         * ipa-cp.c (propagate_vals_across_pass_through): Expand comment.
2305         (cgraph_edge_brings_value_p): New parameter dest_val, check if it is
2306         not the same as the source val.
2307         (cgraph_edge_brings_value_p): New parameter.
2308         (gather_edges_for_value): Pass destination value to
2309         cgraph_edge_brings_value_p.
2310         (perhaps_add_new_callers): Likewise.
2311         (get_info_about_necessary_edges): Likewise and exclude values brought
2312         only by self-recursive edges.
2313         (create_specialized_node): Redirect only clones of self-calling edges.
2314         (+self_recursive_pass_through_p): New function.
2315         (find_more_scalar_values_for_callers_subset): Use it.
2316         (find_aggregate_values_for_callers_subset): Likewise.
2317         (known_aggs_to_agg_replacement_list): Removed.
2318         (decide_whether_version_node): Re-calculate known constants for all
2319         remaining context clones.
2320
2321 2018-04-11  Richard Biener  <rguenther@suse.de>
2322
2323         PR lto/85339
2324         * dwarf2out.c (dwarf2out_finish): Remove DW_AT_stmt_list attribute
2325         from early DWARF output.
2326         (dwarf2out_early_finish): Output line info unconditionally into
2327         early DWARF and add reference to it.
2328
2329 2018-04-11  Jakub Jelinek  <jakub@redhat.com>
2330
2331         PR target/85281
2332         * config/i386/sse.md (iptr): Add V16SFmode and V8DFmode cases.
2333         (<avx512>_vec_dup<mode><mask_name>): Use a single pattern for modes
2334         other than V2DFmode using iptr mode attribute.
2335         (<avx512>_vec_dup<mode><mask_name>): Use iptr mode attribute.
2336
2337 2018-04-11  Alexander Monakov  <amonakov@ispras.ru>
2338
2339         PR rtl-optimization/84659
2340         * sel-sched-ir.c (sel_init_pipelining): Invoke cleanup_cfg.
2341
2342 2018-04-11  Jakub Jelinek  <jakub@redhat.com>
2343
2344         PR debug/85302
2345         * dwarf2out.c (skip_loc_list_entry): Don't call size_of_locs if
2346         SIZEP is NULL.
2347         (output_loc_list): Pass address of a dummy size variable even in the
2348         locview handling loop.
2349         (index_location_lists): Add comment on why skip_loc_list_entry can't
2350         call size_of_locs.
2351
2352 2018-04-11  Thomas Preud'homme  <thomas.preudhomme@arm.com>
2353
2354         PR target/85261
2355         * config/arm/arm-builtins.c (arm_expand_builtin): Force input operand
2356         into register.
2357
2358 2018-04-10  Aaron Sawdey  <acsawdey@linux.ibm.com>
2359
2360         PR target/85321
2361         * doc/invoke.texi (RS/6000 and PowerPC Options): Document options
2362         -mblock-compare-inline-limit, -mblock-compare-inline-loop-limit,
2363         and -mstring-compare-inline-limit.
2364
2365 2018-04-10  Segher Boessenkool  <segher@kernel.crashing.org>
2366
2367         PR target/85287
2368         * gcc/config/rs6000/rs6000.md (allocate_stack): Put the residual size
2369         for stack clash protection in a register whenever we need it to be in
2370         a register.
2371
2372 2018-04-10  Segher Boessenkool  <segher@kernel.crashing.org>
2373
2374         * common/config/rs6000/rs6000-common.c (rs6000_option_init_struct):
2375         Enable -fasynchronous-unwind-tables by default if OBJECT_FORMAT_ELF.
2376
2377 2018-04-10  Segher Boessenkool  <segher@kernel.crashing.org>
2378
2379         PR target/85321
2380         * config/rs6000/rs6000.opt (mtraceback=): Show the allowed values in
2381         the help text.
2382         (mlong-double-): Ditto.
2383         * config/rs6000/sysv4.opt (msdata=): Ditto.
2384         (mtls-size=): Ditto.
2385
2386 2018-04-10  Kelvin Nilsen  <kelvin@gcc.gnu.org>
2387
2388         * config/rs6000/rs6000-c.c (altivec_overloaded_builtins): Remove
2389         erroneous entries for
2390         "vector int vec_ldl (int, long int *)", and
2391         "vector unsigned int vec_ldl (int, unsigned long int *)".
2392         Add comments and entries for
2393         "vector bool char vec_ldl (int, bool char *)",
2394         "vector bool short vec_ldl (int, bool short *)",
2395         "vector bool int vec_ldl (int, bool int *)",
2396         "vector bool long long vec_ldl (int, bool long long *)",
2397         "vector pixel vec_ldl (int, pixel *)",
2398         "vector long long vec_ldl (int, long long *)",
2399         "vector unsigned long long vec_ldl (int, unsigned long long *)".
2400         * config/rs6000/rs6000.c (rs6000_init_builtins): Initialize new
2401         type tree bool_long_long_type_node and correct definition of
2402         bool_V2DI_type_node to make reference to this new type tree.
2403         (rs6000_mangle_type): Replace erroneous reference to
2404         bool_long_type_node with bool_long_long_type_node.
2405         * config/rs6000/rs6000.h (enum rs6000_builtin_type_index): Add
2406         comments to emphasize sign distinctions for char and int types and
2407         replace RS6000_BTI_bool_long constant with
2408         RS6000_BTI_bool_long_long constant.  Also add comment to restrict
2409         use of RS6000_BTI_pixel.
2410         (bool_long_type_node): Remove this macro definition.
2411         (bool_long_long_type_node): New macro definition
2412
2413 2018-04-10  Jakub Jelinek  <jakub@redhat.com>
2414
2415         PR rtl-optimization/85300
2416         * combine.c (subst): Handle subst of CONST_SCALAR_INT_P new_rtx also
2417         into FLOAT and UNSIGNED_FLOAT like ZERO_EXTEND, return a CLOBBER if
2418         simplify_unary_operation fails.
2419
2420 2018-04-10  Martin Liska  <mliska@suse.cz>
2421
2422         * gdbhooks.py: Add pretty-printers for varpool_node, symtab_node,
2423         cgraph_edge and ipa_ref.
2424
2425 2018-04-10  Jakub Jelinek  <jakub@redhat.com>
2426
2427         PR target/85177
2428         PR target/85255
2429         * config/i386/sse.md
2430         (<extract_type>_vinsert<shuffletype><extract_suf>_mask): Fix
2431         computation of the VEC_MERGE selector from mask.
2432         (<extract_type>_vinsert<shuffletype><extract_suf>_1<mask_name>):
2433         Fix decoding of the VEC_MERGE selector into mask.
2434
2435 2018-04-10  Richard Sandiford  <richard.sandiford@linaro.org>
2436
2437         PR tree-optimization/85286
2438         * tree-vect-data-refs.c (vect_get_smallest_scalar_type):
2439
2440 2018-04-10  Richard Sandiford  <richard.sandiford@linaro.org>
2441
2442         * final.c (final_1): Set insn_last_address as well as
2443         insn_current_address.
2444
2445 2018-04-10  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
2446
2447         PR target/85173
2448         * explow.c (emit_stack_probe): Call validize_mem on memory location
2449         before passing it to gen_probe_stack.  Create address operand and
2450         legitimize it for the probe_stack_address case.
2451
2452 2018-04-09  Jan Hubicka  <jh@suse.cz>
2453
2454         PR lto/85078
2455         * ipa-devirt.c (rebuild_type_inheritance-hash): New.
2456         * ipa-utils.h (rebuild_type_inheritance-hash): Declare.
2457         * tree.c (free_lang_data_in_type): Fix handling of binfos;
2458         walk basetypes.
2459         (free_lang_data): Rebuild type inheritance graph.
2460
2461 2018-04-09  Martin Sebor  <msebor@redhat.com>
2462
2463         * invoke.texi (-finline-small-functions): Mention other optimization
2464         options.
2465         (-findirect-inlining, -fpartial-inlining): Same.
2466         (-finline-functions-called-once): Same.
2467         (-freorder-blocks-and-partition): Same.
2468
2469 2018-04-09  Jan Hubicka  <jh@suse.cz>
2470
2471         PR rtl/84058
2472         * cfgcleanup.c (try_forward_edges): Do not give up on crossing
2473         jumps; choose last target that matches the criteria (i.e.
2474         no partition changes for non-crossing jumps).
2475         * cfgrtl.c (cfg_layout_redirect_edge_and_branch): Add basic
2476         support for redirecting crossing jumps to non-crossing.
2477
2478 2018-04-09  Alexey Brodkin  <abrodkin@synopsys.com>
2479
2480         * config/arc/arc.c (arc_expand_prologue): Set stack usage info
2481         also for naked functions.
2482
2483 2018-04-09  Claudiu Zissulescu  <claziss@synopsys.com>
2484
2485         * config/arc/arc.md (add_shift): New pattern.
2486         (add_shift2): Likewise.
2487         (sub_shift): Likewise.
2488         (sub_shift_cmp0_noout): Likewise.
2489         (compare_si_ashiftsi): Likewise.
2490         (xbfu_cmp0_noout): New combine pattern.
2491         (xbfu_cmp0"): Likewise.
2492         (movsi_set_cc_insn): Place the predicable variant first.
2493         (commutative_binary_cmp0_noout): Remove clobber.
2494         (commutative_binary_cmp0): New pattern.
2495         (noncommutative_binary_cmp0): Likewise.
2496         (noncommutative_binary_cmp0_noout): Likewise.
2497         (noncommutative_binary_comparison_result_used): Removed.
2498         (rsub_cmp0): New pattern.
2499         (rsub_cmp0_noout): Likewise.
2500         (extzvsi): Changed, keep only meaningful variants.
2501         (SQH, SEZ): New iterators.
2502         (SQH_postfix): New mode attribute.
2503         (SEZ_prefix): New code attribute.
2504         (<SEZ_prefix>xt<SQH_postfix>_cmp0_noout): New instruction pattern.
2505         (<SEZ_prefix>xt<SQH_postfix>_cmp0): Likewise.
2506         * config/arc/predicates.md (cc_set_register): Use CC_REG instead
2507         of numerical value.
2508         (noncommutative_operator): Check the availability of barrel
2509         shifter option.
2510
2511 2018-04-09  Richard Biener  <rguenther@suse.de>
2512
2513         PR tree-optimization/85284
2514         * tree-ssa-loop-niter.c (number_of_iterations_exit_assumptions):
2515         Only use the niter constraining form of simple_iv when the exit
2516         is always executed.
2517
2518 2018-04-09  Tom de Vries  <tom@codesourcery.com>
2519
2520         PR target/84041
2521         * config/nvptx/nvptx.md (define_c_enum "unspecv"): Add UNSPECV_MEMBAR.
2522         (define_expand "*memory_barrier"): New define_expand.
2523         (define_insn "memory_barrier"): New insn.
2524
2525 2018-04-09  Andrey Belevantsev  <abel@ispras.ru>
2526
2527         PR rtl-optimization/80463
2528         PR rtl-optimization/83972
2529         PR rtl-optimization/83480
2530
2531         * sel-sched-ir.c (has_dependence_note_mem_dep): Take into account the
2532         correct producer for the insn.
2533         (tidy_control_flow): Fixup seqnos in case of debug insns.
2534
2535 2018-04-09  Andrey Belevantsev  <abel@ispras.ru>
2536
2537         PR rtl-optimization/83913
2538
2539         * sel-sched-ir.c (merge_expr_data): Choose the middle between two
2540         different sched-times when merging exprs.
2541
2542 2018-04-09  Andrey Belevantsev  <abel@ispras.ru>
2543
2544         PR rtl-optimization/83962
2545
2546         * sel-sched-ir.c (tidy_control_flow): Correct the order in which we call
2547         tidy_fallthru_edge and tidy_control_flow.
2548
2549 2018-04-09  Andrey Belevantsev  <abel@ispras.ru>
2550
2551         PR rtl-optimization/83530
2552
2553         * sel-sched.c (force_next_insn): New global variable.
2554         (remove_insn_for_debug): When force_next_insn is true, also leave only
2555         next insn in the ready list.
2556         (sel_sched_region): When the region wasn't scheduled, make another pass
2557         over it with force_next_insn set to 1.
2558
2559 2018-04-08  Monk Chiang  <sh.chiang04@gmail.com>
2560
2561         * config.gcc (nds32le-*-*, nds32be-*-*): Add nds32/nds32_intrinsic.h
2562         into tm_file.
2563         * config/nds32/constants.md (unspec_volatile_element): Add enum values
2564         for interrupt control.
2565         * config/nds32/nds32-intrinsic.c: Implementation of intrinsic
2566         functions for interrupt control.
2567         * config/nds32/nds32-intrinsic.md: Likewise.
2568         * config/nds32/nds32_intrinsic.h: Likewise.
2569         * config/nds32/nds32.h (nds32_builtins): Likewise.
2570
2571 2018-04-08  Chung-Ju Wu  <jasonwucj@gmail.com>
2572
2573         * config/nds32/nds32.c (nds32_init_machine_status,
2574         nds32_legitimate_index_p, nds32_legitimate_address_p): Consider
2575         strict_aligned_p field.
2576         (nds32_expand_to_rtl_hook): New function.
2577         (TARGET_EXPAND_TO_RTL_HOOK): Define.
2578         * config/nds32/nds32.h (machine_function): Add strict_aligned_p field.
2579
2580 2018-04-08  Kito Cheng  <kito.cheng@gmail.com>
2581             Chung-Ju Wu  <jasonwucj@gmail.com>
2582
2583         * config.gcc (nds32*-*-*): Check that n7 is valid to --with-cpu.
2584         * config/nds32/nds32-n7.md: New file.
2585         * config/nds32/nds32-opts.h (nds32_cpu_type): Add CPU_N7.
2586         * config/nds32/nds32-pipelines-auxiliary.c: Implementation for n7
2587         pipeline.
2588         * config/nds32/nds32-protos.h: More declarations for n7 pipeline.
2589         * config/nds32/nds32.md (pipeline_model): Add n7.
2590         * config/nds32/nds32.opt (mcpu): Support n7 pipeline cpus.
2591         * config/nds32/pipelines.md: Include n7 settings.
2592
2593 2018-04-08  Kito Cheng  <kito.cheng@gmail.com>
2594             Chung-Ju Wu  <jasonwucj@gmail.com>
2595
2596         * config.gcc (nds32*-*-*): Check that e8 is valid to --with-cpu.
2597         * config/nds32/nds32-e8.md: New file.
2598         * config/nds32/nds32-opts.h (nds32-cpu_type): Add CPU_E8.
2599         * config/nds32/nds32-pipelines-auxiliary.c: Implementation for e8
2600         pipeline.
2601         * config/nds32/nds32-protos.h: More declarations for e8 pipeline.
2602         * config/nds32/nds32.md (pipeline_model): Add e8.
2603         * config/nds32/nds32.opt (mcpu): Support e8 pipeline cpus.
2604         * config/nds32/pipelines.md: Include e8 settings.
2605
2606 2018-04-08  Kito Cheng  <kito.cheng@gmail.com>
2607             Chung-Ju Wu  <jasonwucj@gmail.com>
2608
2609         * config.gcc (nds32*-*-*): Check that n6/n8/s8 are valid to --with-cpu.
2610         * config/nds32/nds32-n8.md: New file.
2611         * config/nds32/nds32-opts.h (nds32_cpu_type): Add CPU_N6 and CPU_N8.
2612         * config/nds32/nds32-pipelines-auxiliary.c: Implementation for n8
2613         pipeline.
2614         * config/nds32/nds32-protos.h: More declarations for n8 pipeline.
2615         * config/nds32/nds32-utils.c: More implementations for n8 pipeline.
2616         * config/nds32/nds32.md (pipeline_model): Add n8.
2617         * config/nds32/nds32.opt (mcpu): Support n8 pipeline cpus.
2618         * config/nds32/pipelines.md: Include n8 settings.
2619
2620 2018-04-08  Kito Cheng  <kito.cheng@gmail.com>
2621             Chung-Ju Wu  <jasonwucj@gmail.com>
2622
2623         * config.gcc (nds32*): Add nds32-utils.o into extra_objs.
2624         * config/nds32/nds32-n9-2r1w.md: New file.
2625         * config/nds32/nds32-n9-3r2w.md: New file.
2626         * config/nds32/nds32-opts.h (nds32_cpu_type, nds32_mul_type,
2627         nds32_register_ports): New or modify for cpu n9.
2628         * config/nds32/nds32-pipelines-auxiliary.c: Implementation for n9
2629         pipeline.
2630         * config/nds32/nds32-protos.h: More declarations for n9 pipeline.
2631         * config/nds32/nds32-utils.c: New file.
2632         * config/nds32/nds32.h (TARGET_PIPELINE_N9, TARGET_PIPELINE_SIMPLE,
2633         TARGET_MUL_SLOW): Define.
2634         * config/nds32/nds32.md (pipeline_model): New attribute.
2635         * config/nds32/nds32.opt (mcpu, mconfig-mul, mconfig-register-ports):
2636         New options that support cpu n9.
2637         * config/nds32/pipelines.md: Include n9 settings.
2638         * config/nds32/t-nds32 (nds32-utils.o): Add dependency.
2639
2640 2018-04-08  Chung-Ju Wu  <jasonwucj@gmail.com>
2641
2642         * config/nds32/nds32-md-auxiliary.c (output_cond_branch): Output align
2643         information if necessary.
2644         (output_cond_branch_compare_zero): Likewise.
2645         * config/nds32/nds32.c (nds32_adjust_insn_length): Consider align case.
2646         (nds32_target_alignment): Refine for alignment.
2647         * config/nds32/nds32.h (NDS32_ALIGN_P): Define.
2648         (FUNCTION_BOUNDARY): Modify.
2649         * config/nds32/nds32.md (call_internal, call_value_internal): Consider
2650         align case.
2651         * config/nds32/nds32.opt (malways-align, malign-functions): New.
2652
2653 2018-04-08  Monk Chiang  <sh.chiang04@gmail.com>
2654
2655         * config/nds32/constants.md (unspec_volatile_element): Add values for
2656         TLB operation and data prefetch.
2657         * config/nds32/nds32-intrinsic.c: Implementation of intrinsic
2658         functions for TLB operation and data prefetch.
2659         * config/nds32/nds32-intrinsic.md: Likewise.
2660         * config/nds32/nds32_intrinsic.h: Likewise.
2661         * config/nds32/nds32.c (nds32_dpref_names): Likewise.
2662         (nds32_print_operand): Likewise.
2663         * config/nds32/nds32.h (nds32_builtins): Likewise.
2664
2665 2018-04-07  Thomas Koenig  <tkoenig@gcc.gnu.org>
2666         Andrew Pinski <pinsika@gcc.gnu.org>
2667
2668         PR middle-end/82976
2669         * match.pd: Use constant_boolean_node of correct type instead of
2670         boolean_true_node or boolean_false_node for simplifying
2671         pointer comparisons to zero.
2672
2673 2018-04-07  Jakub Jelinek  <jakub@redhat.com>
2674
2675         PR tree-optimization/80021
2676         * tree.c (verify_type_variant): Make error call in verify_variant_match
2677         translatable and remove final full stop.
2678
2679 2018-04-07  Chung-Ju Wu  <jasonwucj@gmail.com>
2680
2681         * config/nds32/constants.md (unspec_volatile_element): Add
2682         UNSPEC_VOLATILE_EH_RETURN.
2683         * config/nds32/nds32-md-auxiliary.c (nds32_output_stack_push,
2684         nds32_output_stack_pop): Support dwarf exception handling process.
2685         * config/nds32/nds32-protos.h (nds32_dynamic_chain_address): Declare.
2686         * config/nds32/nds32.c (nds32_init_machine_status): Support dwarf
2687         exception handling process.
2688         (nds32_compute_stack_frame): Likewise.
2689         (nds32_return_addr_rtx): Likewise.
2690         (nds32_initial_elimination_offset): Likewise.
2691         (nds32_expand_prologue): Likewise.
2692         (nds32_expand_epilogue): Likewise.
2693         (nds32_dynamic_chain_address): New function.
2694         * config/nds32/nds32.h (machine_function): Add fields for dwarf
2695         exception handling.
2696         (DYNAMIC_CHAIN_ADDRESS): Define.
2697         (EH_RETURN_DATA_REGNO): Define.
2698         (EH_RETURN_STACKADJ_RTX): Define.
2699         * config/nds32/nds32.md (eh_return, nds32_eh_return): Implement
2700         patterns for dwarf exception handling.
2701
2702 2018-04-07  Chung-Ju Wu  <jasonwucj@gmail.com>
2703
2704         * config/nds32/nds32.h: Clean up obsolete macros.
2705
2706 2018-04-07  Monk Chiang  <sh.chiang04@gmail.com>
2707
2708         * config/nds32/constants.md (unspec_element, unspec_volatile_element):
2709         Add enum values for particular instructions.
2710         * config/nds32/nds32-intrinsic.c: Implementation of expanding
2711         particular intrinsic functions.
2712         * config/nds32/nds32-intrinsic.md: Likewise.
2713         * config/nds32/nds32_intrinsic.h: Likewise.
2714         * config/nds32/nds32.h (nds32_builtins): Likewise.
2715         * config/nds32/nds32.md (type): Add pbsad and pbsada.
2716         (btst, ave): New patterns for particular instructions.
2717
2718 2018-04-07  Monk Chiang  <sh.chiang04@gmail.com>
2719
2720         * config/nds32/constants.md (unspec_element, unspec_volatile_element):
2721         Add enum values for atomic load/store and memory sync.
2722         * config/nds32/nds32-intrinsic.c: Implementation for atomic load/store
2723         and memory sync.
2724         * config/nds32/nds32-intrinsic.md: Likewise.
2725         * config/nds32/nds32_intrinsic.h: Likewise.
2726         * config/nds32/nds32.h (nds32_builtins): Likewise.
2727
2728 2018-04-07  Jakub Jelinek  <jakub@redhat.com>
2729
2730         PR tree-optimization/85257
2731         * fold-const.c (native_encode_vector): If not all elts could fit
2732         and off is -1, return 0 rather than offset.
2733         * tree-ssa-sccvn.c (vn_reference_lookup_3): Pass
2734         (offseti - offset2) / BITS_PER_UNIT as 4th argument to
2735         native_encode_expr.  Verify len * BITS_PER_UNIT >= maxsizei.  Don't
2736         adjust buffer in native_interpret_expr call.
2737
2738 2018-04-07  Monk Chiang  <sh.chiang04@gmail.com>
2739
2740         * config/nds32/constants.md (unspec_volatile_element): Add cache
2741         control enum values.
2742         * config/nds32/nds32-intrinsic.c: Add cache control expand functions.
2743         * config/nds32/nds32-intrinsic.md: Add cache control patterns.
2744         * config/nds32/nds32.c (nds32_cctl_names): New.
2745         (nds32_print_operand): Handle cache control register names.
2746         * config/nds32/nds32.h (nds32_builtins): New enum values.
2747         * config/nds32/nds32_intrinsic.h: Add cache control enum types and
2748         macros.
2749         * config/nds32/nds32.md (type): Add mmu.
2750         * config/nds32/pipelines.md (simple_insn): Add mmu.
2751
2752 2018-04-07  Chung-Ju Wu  <jasonwucj@gmail.com>
2753
2754         * config/nds32/nds32.md (type): Remove call.
2755         * config/nds32/pipelines.md (simple_insn): Likewise.
2756
2757 2018-04-07  Monk Chiang  <sh.chiang04@gmail.com>
2758
2759         * config/nds32/constants.md (unspec_volatile_element): Add
2760         UNSPEC_VOLATILE_FMFCSR, UNSPEC_VOLATILE_FMTCSR and
2761         UNSPEC_VOLATILE_FMFCFG.
2762         * config/nds32/nds32-intrinsic.c (bdesc_noarg): New builtin
2763         description for fmfcfg and fmfcsr.
2764         (bdesc_1arg): Add fmtcsr.
2765         (bdesc_2arg): Add fcpynss, fcpyss, fcpynsd and fcpysd.
2766         (nds32_expand_builtin_impl): Deal with FPU intrinsic functions.
2767         * config/nds32/nds32-intrinsic.md (unspec_fcpynsd, unspec_fcpysd,
2768         unspec_fcpynss, unspec_fcpysd, unspec_fcpyss, unspec_fmfcsr,
2769         unspec_fmfcfg): New patterns.
2770         * config/nds32/nds32.h (nds32_builtins): Add NDS32_BUILTIN_FMFCFG,
2771         NDS32_BUILTIN_FMFCSR, NDS32_BUILTIN_FMTCSR, NDS32_BUILTIN_FCPYNSS,
2772         NDS32_BUILTIN_FCPYSS,NDS32_BUILTIN_FCPYNSD and NDS32_BUILTIN_FCPYSD.
2773         * config/nds32/nds32_intrinsic.h (__nds32__fcpynsd, __nds32__fcpynss,
2774         __nds32__fcpysd, __nds32__fcpyss, __nds32__fmfcsr, __nds32__fmtcsr,
2775         __nds32__fmfcfg): Define.
2776
2777 2018-04-07  Monk Chiang  <sh.chiang04@gmail.com>
2778
2779         * config/nds32/nds32.c (nds32_intrinsic_register_names): Add more
2780         intrinsic register names.
2781         * config/nds32/nds32_intrinsic.h (nds32_intrinsic_registers): Add more
2782         intrinsic register enum values and macros.
2783
2784 2018-04-07  Chung-Ju Wu  <jasonwucj@gmail.com>
2785
2786         * config/nds32/nds32.c (nds32_legitimate_index_p): Modify condition
2787         for load/store addressing form.
2788         (nds32_print_operand_address): Likewise.
2789
2790 2018-04-06  Eric Botcazou  <ebotcazou@adacore.com>
2791
2792         PR target/85196
2793         * config/sparc/sparc.c (sparc_expand_move): Deal with symbolic operands
2794         based on LABEL_REF.  Remove useless assertion.
2795         (pic_address_needs_scratch): Fix formatting.
2796         (sparc_legitimize_pic_address): Minor tweaks.
2797         (sparc_delegitimize_address): Adjust assertion accordingly.
2798         * config/sparc/sparc.md (movsi_pic_label_ref): Change label_ref_operand
2799         into symbolic_operand.
2800         (movsi_high_pic_label_ref): Likewise.
2801         (movsi_lo_sum_pic_label_ref): Likewise.
2802         (movdi_pic_label_ref): Likewise.
2803         (movdi_high_pic_label_ref): Likewise.
2804         (movdi_lo_sum_pic_label_ref): Likewise.
2805
2806 2018-04-06  Amaan Cheval  <amaan.cheval@gmail.com>
2807
2808         * config.gcc (x86_64-*-rtems*): Add rtems.h to tm_file for
2809         custom LIB_SPEC setup.
2810
2811 2018-04-06  Ruslan Bukin  <br@bsdpad.com>
2812             Kito Cheng  <kito.cheng@gmail.com>
2813
2814         * config.gcc (riscv*-*-freebsd*): Add RISC-V FreeBSD support.
2815         * config/riscv/freebsd.h: New.
2816
2817 2018-04-06  Chung-Ju Wu  <jasonwucj@gmail.com>
2818
2819         * config/nds32/nds32.c (nds32_adjust_insn_length): Refine.
2820         * config/nds32/nds32.h (ADJUST_INSN_LENGTH): Change the location in
2821         file.
2822
2823 2018-04-06  Chung-Ju Wu  <jasonwucj@gmail.com>
2824             Kito Cheng  <kito.cheng@gmail.com>
2825
2826         * config/nds32/nds32-md-auxiliary.c (nds32_output_return,
2827         nds32_output_call, nds32_symbol_binds_local_p): New functions.
2828         * config/nds32/nds32-protos.h (nds32_output_call,
2829         nds32_output_return): Declare.
2830         * config/nds32/nds32.md: Refine all the call and return patterns.
2831
2832 2018-04-06  Jakub Jelinek  <jakub@redhat.com>
2833
2834         PR debug/85252
2835         * dwarf2out.c (rtl_for_decl_init): For STRING_CST initializer only
2836         build CONST_STRING if TYPE_MAX_VALUE is non-NULL and is INTEGER_CST.
2837
2838         PR rtl-optimization/84872
2839         * cfgloopmanip.c (create_preheader): Use make_forwarder_block even if
2840         nentry == 1 when CP_FALLTHRU_PREHEADERS and single_entry is
2841         EDGE_CROSSING edge.
2842
2843 2018-04-06  Tamar Christina  <tamar.christina@arm.com>
2844
2845         * expr.c (copy_blkmode_to_reg): Revert 254862.
2846         * doc/sourcebuild.texi (word_mode_no_slow_unalign): Likewise.
2847
2848 2018-04-06  Richard Biener  <rguenther@suse.de>
2849
2850         PR middle-end/85244
2851         * tree-dfa.c (get_ref_base_and_extent): Reset seen_variable_array_ref
2852         after seeing a component reference with an adjacent field.  Treat
2853         refs to arrays at struct end of external decls similar to
2854         refs to unconstrained commons.
2855
2856 2018-04-06  Jakub Jelinek  <jakub@redhat.com>
2857
2858         PR sanitizer/85213
2859         * fold-const.c (twoval_comparison_p): Remove SAVE_P argument and don't
2860         look through SAVE_EXPRs with non-side-effects argument.  Adjust
2861         recursive calls.
2862         (fold_comparison): Adjust twoval_comparison_p caller, don't handle
2863         save_p here.
2864
2865 2018-04-06  Richard Biener  <rguenther@suse.de>
2866
2867         PR middle-end/85180
2868         * alias.c (find_base_term): New wrapper around find_base_term
2869         unwinding CSELIB_VAL_PTR changes.
2870         (find_base_term): Do not restore CSELIB_VAL_PTR during the
2871         recursion.
2872
2873 2018-04-06  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>
2874
2875         * config/s390/s390.c (s390_z10_optimize_cmp): Expand dedicated NOP
2876         instructions.
2877         * config/s390/s390.md (UNSPECV_NOP_LR_0, UNSPECV_NOP_LR_1): New
2878         constant definitions.
2879         ("nop"): lr 0,0 -> nopr r0
2880         ("nop_lr0", "nop_lr1"): New insn definitions.
2881
2882 2018-04-06  Chung-Ju Wu  <jasonwucj@gmail.com>
2883
2884         * config/nds32/nds32.md (*stack_push, *stack_pop): Use
2885         NDS32_V3PUSH_AVAILABLE_P macro.
2886
2887 2018-04-06  Monk Chiang  <sh.chiang04@gmail.com>
2888             Chung-Ju Wu  <jasonwucj@gmail.com>
2889
2890         * config.gcc (nds32*-*-*): Add v2j v3f v3s checking.
2891         (nds32*-*-*): Add float and fpu_config into supported_defaults.
2892         * common/config/nds32/nds32-common.c (TARGET_DEFAULT_TARGET_FLAGS):
2893         Include TARGET_DEFAULT_FPU_ISA and TARGET_DEFAULT_FPU_FMA.
2894         * config/nds32/constants.md (unspec_element): Add UNSPEC_COPYSIGN,
2895         UNSPEC_FCPYNSD, UNSPEC_FCPYNSS, UNSPEC_FCPYSD and UNSPEC_FCPYSS.
2896         * config/nds32/constraints.md: New constraints and checking for hard
2897         float configuration.
2898         * config/nds32/iterators.md: New mode iterator and attribute for hard
2899         float configuration.
2900         * config/nds32/nds32-doubleword.md: Use hard float alternatives and
2901         patterns.
2902         * config/nds32/nds32-fpu.md: New file.
2903         * config/nds32/nds32-md-auxiliary.c: New functions and checkings to
2904         deal with hard float code generation.
2905         * config/nds32/nds32-opts.h (nds32_arch_type): Add ARCH_V3F and
2906         ARCH_V3S.
2907         (abi_type, float_reg_number): New enum type.
2908         * config/nds32/nds32-predicates.c: New predicates for hard float.
2909         * config/nds32/nds32-protos.h: Declare functions for hard float.
2910         * config/nds32/nds32.c: Implementation for hard float configuration.
2911         * config/nds32/nds32.h: Definitions for hard float configuration.
2912         * config/nds32/nds32.md: Include hard float machine description and
2913         modify patterns for hard float configuration.
2914         * config/nds32/nds32.opt: New options for hard float configuration.
2915         * config/nds32/predicates.md: New predicates for hard float
2916         configuration.
2917
2918 2018-04-06  Kuan-Lin Chen  <kuanlinchentw@gmail.com>
2919
2920         * common/config/nds32/nds32-common.c
2921         (nds32_option_optimization_table): Enable -mreleax-hint by default.
2922
2923 2018-04-05  Jakub Jelinek  <jakub@redhat.com>
2924
2925         PR middle-end/85195
2926         * match.pd (BIT_FIELD_REF CONSTRUCTOR@0 @1 @2): Use view_convert around
2927         CONSTRUCTOR_ELT (ctor, ...)->value.
2928
2929 2018-04-05  Uros Bizjak  <ubizjak@gmail.com>
2930
2931         PR target/85193
2932         * config/i386/i386.md (define_attr "memory"): Handle rotate1 type.
2933
2934 2018-04-05  Tom de Vries  <tom@codesourcery.com>
2935
2936         PR target/85204
2937         * config/nvptx/nvptx.c (nvptx_single): Fix neutering of bb with only
2938         cond jump.
2939
2940 2018-04-05  Shiva Chen  <shiva0217@gmail.com>
2941             Kito Cheng  <kito.cheng@gmail.com>
2942
2943         * config/nds32/constraints.md (U33): Fine-tune checking condition.
2944         * config/nds32/nds32-md-auxiliary.c (nds32_mem_format): Ditto.
2945         * config/nds32/nds32.h (nds32_16bit_address_type): Add
2946         ADDRESS_POST_MODIFY_LO_REG_IMM3U.
2947
2948 2018-04-05  Shiva Chen  <shiva0217@gmail.com>
2949             Kito Cheng  <kito.cheng@gmail.com>
2950
2951         * config/nds32/constraints.md (Ufe): New memory constraint.
2952         * config/nds32/nds32-md-auxiliary.c (nds32_mem_format,
2953         nds32_output_16bit_load): Consider r8 register for lwi45.fe format.
2954         * config/nds32/nds32.c (nds32_print_operand): Output lwi45.fe
2955         operands.
2956         * config/nds32/nds32.h (nds32_16bit_address_type): Add ADDRESS_R8_IMM7U.
2957         * config/nds32/nds32.md (*mov<mode>): Adjust pattern.
2958
2959 2018-04-05  Chung-Ju Wu  <jasonwucj@gmail.com>
2960
2961         * config/nds32/nds32.md: Use optimize_size in the condition for
2962         alu-shift instructions.
2963
2964 2018-04-05  Chung-Ju Wu  <jasonwucj@gmail.com>
2965
2966         * config/nds32/nds32.md (divsi4, udivsi4): New patterns.
2967
2968 2018-04-05  Chung-Ju Wu  <jasonwucj@gmail.com>
2969
2970         * config/nds32/nds32.md (negsi2): Refine pattern.
2971
2972 2018-04-05  Kito Cheng  <kito.cheng@gmail.com>
2973             Chung-Ju Wu  <jasonwucj@gmail.com>
2974
2975         * config/nds32/iterators.md (shift_rotate): New code iterator.
2976         (shift): New code attribute.
2977         * config/nds32/nds32-md-auxiliary.c (nds32_expand_constant): New.
2978         * config/nds32/nds32-protos.h (nds32_expand_constant): Declare.
2979         * config/nds32/nds32.c (nds32_print_operand): Deal with more cases.
2980         * config/nds32/nds32.md (addsi3, *add_srli): Refine implementation for
2981         bit-wise operations.
2982         (andsi3, *andsi3): Ditto.
2983         (iorsi3, *iorsi3, *or_slli, *or_srli): Ditto.
2984         (xorsi3, *xorsi3, *xor_slli, *xor_srli): Ditto.
2985         (<shift>si3, *ashlsi3, *ashrsi3, *lshrsi3, *rotrsi3): Ditto.
2986         * config/nds32/predicates.md (nds32_rimm5u_operand, nds32_and_operand,
2987         nds32_ior_operand, nds32_xor_operand): New predicates.
2988
2989 2018-04-05  Chung-Ju Wu  <jasonwucj@gmail.com>
2990
2991         * config/nds32/nds32.md (add<mode>3, sub<mode>3): Rename to ...
2992         (addsi3, subsi3): ... this.
2993
2994 2018-04-05  Chung-Ju Wu  <jasonwucj@gmail.com>
2995
2996         * config/nds32/nds32.md (*sub_srli, *and_slli): Fine-tune predicator.
2997
2998 2018-04-05  Chung-Ju Wu  <jasonwucj@gmail.com>
2999
3000         * config/nds32/nds32.md: Adjust indention.
3001
3002 2018-04-05  Kito Cheng  <kito.cheng@gmail.com>
3003
3004         * config/nds32/nds32.md (feature): New attribute.
3005
3006 2018-04-05  Chung-Ju Wu  <jasonwucj@gmail.com>
3007
3008         * config/nds32/nds32.md (subtype): New attribute.
3009
3010 2018-04-04  Thomas Preud'homme  <thomas.preudhomme@arm.com>
3011
3012         PR target/85203
3013         * config/arm/arm-builtins.c (arm_expand_builtin): Change
3014         expansion to perform a bitwise AND of the argument followed by a
3015         boolean negation of the result.
3016
3017 2018-04-04  Peter Bergner  <bergner@vnet.ibm.com>
3018
3019         PR rtl-optimization/84878
3020         * ddg.c (add_cross_iteration_register_deps): Use DF_REF_BB to determine
3021         the basic block.  Assert the use reference is not artificial and that
3022         it has an associated insn.
3023
3024 2018-04-04  Michael Matz  <matz@suse.de>
3025
3026         * builtins.c (compute_objsize): Pass correct operand
3027         to array_at_struct_end_p.
3028
3029 2018-04-04  Richard Biener  <rguenther@suse.de>
3030
3031         PR lto/85176
3032         * dwarf2out.c (dwarf2out_register_external_die): Peel namespaces
3033         from contexts for DINFO_LEVEL_TERSE and below.
3034
3035 2018-04-04  Kito Cheng  <kito.cheng@gmail.com>
3036
3037         * config/nds32/nds32-doubleword.md (move_<mode>): Require
3038         resiter_operand condition.
3039         * config/nds32/nds32.md (*move<mode>): Ditto.
3040
3041 2018-04-04  Kito Cheng  <kito.cheng@gmail.com>
3042             Monk Chiang  <sh.chiang04@gmail.com>
3043
3044         * config/nds32/nds32.md (movmisalign<mode>): New pattern.
3045
3046 2018-04-04  Chung-Ju Wu  <jasonwucj@gmail.com>
3047
3048         * config/nds32/nds32.md (movqi, movhi): Merge into mov<mode>.
3049
3050 2018-04-04  Chung-Ju Wu  <jasonwucj@gmail.com>
3051             Kito Cheng  <kito.cheng@gmail.com>
3052
3053         * config/nds32/nds32-md-auxiliary.c (nds32_inverse_cond_code,
3054         nds32_cond_code_str, output_cond_branch,
3055         output_cond_branch_compare_zero, nds32_expand_cbranch,
3056         nds32_expand_cstore, nds32_expand_movcc,
3057         nds32_output_cbranchsi4_equality_zero,
3058         nds32_output_cbranchsi4_equality_reg,
3059         nds32_output_cbranchsi4_equality_reg_or_const_int,
3060         nds32_output_cbranchsi4_greater_less_zero: New functions.
3061         * config/nds32/nds32-protos.h (nds32_expand_cbranch,
3062         nds32_expand_cstore, nds32_expand_movcc,
3063         nds32_output_cbranchsi4_equality_zero,
3064         nds32_output_cbranchsi4_equality_reg,
3065         nds32_output_cbranchsi4_equality_reg_or_const_int,
3066         nds32_output_cbranchsi4_greater_less_zero): Declare.
3067         * config/nds32/predicates.md (nds32_movecc_comparison_operator,
3068         nds32_rimm11s_operand): New predicates.
3069         * config/nds32/nds32.h (nds32_expand_result_type): New enum type.
3070         * config/nds32/nds32.md: Rewrite all the branch and conditional move
3071         patterns.
3072
3073 2018-04-04  Kito Cheng  <kito.cheng@gmail.com>
3074
3075         * config/nds32/nds32-doubleword.md: Refine all the instruction type.
3076         * config/nds32/nds32.md: Ditto.
3077         * config/nds32/pipelines.md: Ditto.
3078
3079 2018-04-04  Richard Biener  <rguenther@suse.de>
3080
3081         PR tree-optimization/85168
3082         * tree-ssa-sccvn.c (vn_reference_maybe_forwprop_address): Avoid
3083         propagating abnormals.
3084
3085 2018-04-04  Chung-Ju Wu  <jasonwucj@gmail.com>
3086
3087         * config/nds32/nds32.md (enabled): Use yes/no for this attribute.
3088
3089 2018-04-04  Chung-Ju Wu  <jasonwucj@gmail.com>
3090             Kito Cheng  <kito.cheng@gmail.com>
3091
3092         * config/nds32/nds32-md-auxiliary.c (nds32_long_call_p): New function.
3093         * config/nds32/nds32-protos.h (nds32_long_call_p): Declare.
3094         * config/nds32/nds32.c (nds32_function_ok_for_sibcall): New function.
3095         (TARGET_FUNCTION_OK_FOR_SIBCALL): Define.
3096         * config/nds32/nds32.md (sibcall_internal): New.
3097         (sibcall_register): Remove.
3098         (sibcall_immediate): Remove.
3099         (sibcall_value_internal): New.
3100         (sibcall_value_register): Remove.
3101         (sibcall_value_immediate): Remove.
3102         * config/nds32/predicates.md (nds32_general_register_operand): New.
3103         (nds32_call_address_operand): New.
3104
3105 2018-04-03  Jakub Jelinek  <jakub@redhat.com>
3106
3107         PR rtl-optimization/85167
3108         * shrink-wrap.c (move_insn_for_shrink_wrap): Don't set bb_uses and
3109         bb_defs if *split_p, instead preinitialize it to NULL.
3110
3111         PR tree-optimization/85156
3112         * builtins.c (fold_builtin_expect): Use save_expr on arg1 to avoid
3113         evaluating the argument multiple times.
3114
3115 2018-04-03  Bill Schmidt  <wschmidt@linux.ibm.com>
3116
3117         * config/rs6000/emmintrin.h (_mm_cvtpd_epi32): Use __vector rather
3118         than vector.
3119         (_mm_cvtpd_ps): Likewise.
3120         (_mm_cvttpd_epi32): Likewise.
3121         * config/rs6000/mmintrin.h (_mm_unpacklo_pi8): Likewise.
3122         * config/rs6000/xmmintrin.h: For strict-ANSI C++ or C11, undefine
3123         vector, pixel, and bool following altivec.h include.
3124
3125 2018-04-03  Martin Sebor  <msebor@redhat.com>
3126
3127         * doc/extend.texi (Common Function Attributes): Clarify.
3128         (const attribute): Likewise.
3129         (pure attribute): Likewise.
3130
3131 2018-04-03  Jakub Jelinek  <jakub@redhat.com>
3132
3133         PR target/85169
3134         * config/i386/i386.c (ix86_expand_vector_set): Use
3135         HOST_WIDE_INT_1U << elt instead of 1 << elt.  Formatting fix.
3136
3137 2018-04-03  Uros Bizjak  <ubizjak@gmail.com>
3138
3139         * config/i386/i386.c (emit_i387_cw_initialization): Always use logic
3140         instructions when changing rounding bits to preserve precision bits
3141         in the x87 control word.
3142
3143 2018-04-03  Martin Liska  <mliska@suse.cz>
3144
3145         PR tree-optimization/82491
3146         * rtl.h (strip_offset_and_add): Replace += suboffset with
3147         poly_uint64 () + suboffset.
3148
3149 2018-03-29  Martin Liska  <mliska@suse.cz>
3150             Martin Jambor  <mjambor@suse.cz>
3151
3152         PR ipa/84947
3153         * ipa-cp.c (propagate_bits_across_jump_function): Bail out if
3154         param_type is not an integral or pointer type.
3155
3156 2018-04-03  Richard Biener  <rguenther@suse.de>
3157
3158         * sese.h (recompute_all_dominators): Remove.
3159
3160 2018-04-02  Martin Sebor  <msebor@redhat.com>
3161
3162         * doc/invoke.texi (-Wrestrict): Fix typos.
3163
3164 2018-04-02  Jim Wilson  <jimw@sifive.com>
3165
3166         * config/riscv/riscv.h (SHIFT_COUNT_TRUNCATED): Set to zero.
3167         * config/riscv/riscv.md (<optab>si3): Use QImode shift count.
3168         (<optab>di3, <optab>si3_extend): Likewise.
3169         (<optab>si3_mask, <optab>si3_mask_1): New.
3170         (<optab>di3_mask, <optab>di3_mask_1): New.
3171         (<optab>si3_extend_mask, <optab>si3_extend_mask_1): New.
3172         (lshrsi3_zero_extend_1): Use VOIDmode shift count.
3173         * config/riscv/sync.md (atomic_test_and_set): Emit QImode shift count.
3174
3175 2018-04-02  Gerald Pfeifer  <gerald@pfeifer.com>
3176
3177         * doc/cpp.texi (Variadic Macros): Fix line continuation in an
3178         example.
3179
3180 2018-04-02  Chung-Ju Wu  <jasonwucj@gmail.com>
3181
3182         * config/nds32/nds32.c (TARGET_CANONICALIZE_COMPARISON): Define.
3183         (nds32_canonicalize_comparison): New function.
3184
3185 2018-04-01  Chung-Ju Wu  <jasonwucj@gmail.com>
3186             Kito Cheng  <kito.cheng@gmail.com>
3187             Kuan-Lin Chen  <kuanlinchentw@gmail.com>
3188
3189         * config.gcc (nds32): Add nds32-relax-opt.o into extra_objs.
3190         * config/nds32/constants.md (unspec_volatile_element): Add
3191         UNSPEC_VOLATILE_RELAX_GROUP.
3192         * config/nds32/nds32-relax-opt.c: New file.
3193         * config/nds32/nds32-predicates.c
3194         (nds32_symbol_load_store_p): New function.
3195         * config/nds32/nds32-protos.h
3196         (nds32_symbol_load_store_p): Declare function.
3197         (make_pass_nds32_relax_opt): Declare new rtl pass function.
3198         * config/nds32/nds32.c
3199         (nds32_register_pass): New function to register pass.
3200         (nds32_register_passes): New function to register passes.
3201         * config/nds32/nds32.md (relax_group): New pattern.
3202         * config/nds32/nds32.opt (mrelax-hint): New option.
3203         * config/nds32/t-nds32 (nds32-relax-opt.o): New dependency.
3204
3205 2018-04-01  Kito Cheng  <kito.cheng@gmail.com>
3206
3207         * config/nds32/t-nds32: Modify files dependency.
3208
3209 2018-04-01  Chung-Ju Wu  <jasonwucj@gmail.com>
3210
3211         * config/nds32/nds32.h (FUNCTION_PROFILER): Output newline character.
3212         (PROFILE_HOOK): Define its implementation.
3213
3214 2018-04-01  Chung-Ju Wu  <jasonwucj@gmail.com>
3215
3216         * config/nds32/nds32.h (WCHAR_TYPE, WCHAR_TYPE_SIZE): Use unsigned int
3217         type and 32-bit size.
3218
3219 2018-04-01  Jakub Jelinek  <jakub@redhat.com>
3220
3221         PR middle-end/85090
3222         * config/i386/sse.md (V): Add V64QI and V32HI for TARGET_AVX512F.
3223         (V_128_256): New mode iterator.
3224         (*avx512dq_vextract<shuffletype>64x2_1 splitter): New define_split.
3225         (*avx512f_vextract<shuffletype>32x4_1 splitter): Likewise.
3226         (xop_pcmov_<mode><avxsizesuffix>): Use V_128_256 mode iterator instead
3227         of V.
3228         * config/i386/i386.c (ix86_expand_vector_set): Improve V32HImode and
3229         V64QImode expansion for !TARGET_AVX512BW && TARGET_AVX512F.
3230
3231 2018-03-31  Segher Boessenkool  <segher@kernel.crashing.org>
3232
3233         PR target/83315
3234         * config/rs6000/xmmintrin.h (_mm_set_ps, _mm_max_ps): Handle (quiet)
3235         NaN inputs correctly.
3236
3237 2018-03-30  Peter Bergner  <bergner@vnet.ibm.com>
3238
3239         PR target/80546
3240         * config/rs6000/vsx.md (??r): New mode attribute.
3241         (*vsx_mov<mode>_64bit): Use it.
3242         (*vsx_mov<mode>_32bit): Likewise.
3243
3244 2018-03-30  Martin Sebor  <msebor@redhat.com>
3245
3246         PR tree-optimization/84818
3247         * builtins.c (check_access): Use warning_n.
3248
3249 2018-03-30  Aaron Sawdey  <acsawdey@linux.vnet.ibm.com>
3250
3251         PR target/83822
3252         * config/rs6000/rs6000-string.c (expand_compare_loop): Fix redundant
3253         condition.
3254         * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): Fix redundant
3255         condition.
3256
3257 2018-03-30  Julia Koval  <julia.koval@intel.com>
3258
3259         PR target/84413
3260         * x86-tune.def (movx, partial_reg_dependency): Enable for
3261         m_SKYLAKE_AVX512.
3262
3263 2018-03-29  Vladimir Makarov  <vmakarov@redhat.com>
3264
3265         PR inline-asm/84985
3266         * lra-constraints.c (process_alt_operands): Move setting
3267         this_alternative_matches below.
3268
3269 2018-03-29  Martin Liska  <mliska@suse.cz>
3270
3271         PR lto/84995.
3272         * doc/invoke.texi: Document how LTO works with debug info.
3273         Describe auto-load support of binutils.  Mention 'x86-64'
3274         as valid option value of -march option.
3275
3276 2018-03-29  Jakub Jelinek  <jakub@redhat.com>
3277
3278         * config/i386/sse.md (<avx512>_blendm<mode>): Use <sseintprefix>.
3279
3280         PR c/85094
3281         * fold-const.c (operand_equal_p): Handle DEBUG_BEGIN_STMT.
3282         For STATEMENT_LIST, pass down OEP_LEXICOGRAPHIC and maybe
3283         OEP_NO_HASH_CHECK for recursive call, to avoid exponential
3284         checking.
3285
3286 2018-03-28  Peter Bergner  <bergner@vnet.ibm.com>
3287
3288         PR target/84912
3289         * config/rs6000/rs6000.h (RS6000_BTM_POWERPC64): New define.
3290         (RS6000_BTM_COMMON): Add RS6000_BTM_POWERPC64.
3291         * config/rs6000/rs6000.c (rs6000_builtin_mask_calculate): Add support
3292         for RS6000_BTM_POWERPC64.
3293         (rs6000_invalid_builtin): Add handling for RS6000_BTM_POWERPC64
3294         (rs6000_builtin_mask_names): Add RS6000_BTM_POWERPC64.
3295         * config/rs6000/rs6000-builtin.def (BU_P7_POWERPC64_MISC_2): New macro
3296         definition.
3297         (DIVDE): Use it.
3298         (DIVDEU): Likewise.
3299
3300 2018-03-28 Carl Love  <cel@us.ibm.com>
3301
3302         Revert
3303         2017-09-27  Carl Love  <cel@us.ibm.com>
3304
3305         * config/rs6000/rs6000-builtin.def (BU_FP_1MISC_1): Add define macro.
3306         (FCTID, FCTIW): Add BU_FP_MISC_1 macro expansion for builtins.
3307         * config/rs6000/rs6000.md (lrintsfsi2): Add define_insn for the
3308         fctiw instruction.
3309
3310 2018-03-28  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
3311
3312         * config/rs6000/xmmintrin.h (_mm_max_pi16): Use __vector __bool
3313         instead of __vector bool.
3314         (_mm_max_pu8): Likewise.
3315         (_mm_min_pi16): Likewise.
3316
3317 2018-03-28  Peter Bergner  <bergner@vnet.ibm.com>
3318
3319         PR target/84912
3320         * config/rs6000/rs6000-builtin.def (DIVWEO): Delete macro expansion.
3321         (DIVWEUO): Likewise.
3322         (DIVDEO): Likewise.
3323         (DIVDEUO): Likewise.
3324         * config/rs6000/rs6000.c (builtin_function_type): Remove support for
3325         DIVWEUO and DIVDEUO.
3326         * config/rs6000/rs6000.md (UNSPEC_DIVEO, UNSPEC_DIVEUO): Delete unspecs.
3327         (UNSPEC_DIV_EXTEND): Remove deleted unspecs.
3328         (div_extend): Likewise.
3329         * doc/extend.texi (__builtin_divweo): Remove documentation for deleted
3330         builtin function.
3331         (__builtin_divweuo): Likewise.
3332         (__builtin_divdeo): Likewise.
3333         (__builtin_divdeuo): Likewise.
3334
3335 2018-03-28  Jakub Jelinek  <jakub@redhat.com>
3336
3337         PR target/85095
3338         * config/i386/i386.md (*add<mode>3_carry_0, *addsi3_carry_zext_0,
3339         *sub<mode>3_carry_0, *subsi3_carry_zext_0): New patterns.
3340
3341         PR tree-optimization/82004
3342         * gimple-match-head.c (optimize_pow_to_exp): New function.
3343         * match.pd (pow(C,x) -> exp(log(C)*x)): Wrap with #if GIMPLE.
3344         Don't fold to exp if optimize_pow_to_exp is false.
3345
3346 2018-03-28  Martin Liska  <mliska@suse.cz>
3347
3348         PR other/84819
3349         * calls.c (initialize_argument_information): Fix trailing space.
3350         * common.opt: Fix typo and provide better explanation for
3351         -fsanitize-coverage option.
3352         * config/i386/i386.opt: Fix typo.
3353
3354 2018-03-28  Jakub Jelinek  <jakub@redhat.com>
3355             Martin Liska  <mliska@suse.cz>
3356
3357         PR sanitizer/85081
3358         * gimplify.c (asan_poison_variable): Don't do the check for
3359         gimplify_omp_ctxp here.
3360         (gimplify_decl_expr): Do it here.
3361         (gimplify_target_expr): Likewise.
3362
3363 2018-03-28  Martin Liska  <mliska@suse.cz>
3364
3365         PR target/84988
3366         * config/i386/i386.c (ix86_function_arg_advance): Do not call
3367         chkp_type_bounds_count if MPX is not enabled.
3368
3369 2018-03-27  Chung-Ju Wu  <jasonwucj@gmail.com>
3370
3371         * config/nds32/nds32.h (BRANCH_COST): Adjust cost.
3372
3373 2018-03-27  Michael Meissner  <meissner@linux.vnet.ibm.com>
3374
3375         PR target/84914
3376         * config/rs6000/rs6000.c (create_complex_muldiv): New helper
3377         function to create the function decl for complex long double
3378         multiply and divide for -mabi=ieeelongdouble.
3379         (init_float128_ieee): Call it.
3380
3381 2018-03-27  H.J. Lu  <hongjiu.lu@intel.com>
3382
3383         PR target/85044
3384         * config/i386/i386.c (ix86_trampoline_init): Insert ENDBR for
3385         -fcf-protection=branch -mibt.
3386         * config/i386/i386.h (TRAMPOLINE_SIZE): Increased by 4 bytes.
3387
3388 2018-03-27  Ramana Radhakrishnan  <ramana.radhakrishnan@arm.com>
3389
3390         PR target/81863
3391         * config/arm/arm.c (arm_valid_symbolic_address): Handle arm_word_relocations
3392
3393 2018-03-27  Cesar Philippidis  <cesar@codesourcery.com>
3394
3395         PR target/85056
3396         * config/nvptx/nvptx.c (nvptx_assemble_decl_begin): Add '[]' to
3397         extern array declarations.
3398
3399 2018-03-27  Richard Biener  <rguenther@suse.de>
3400
3401         PR middle-end/84067
3402         * match.pd ((A * C) +- (B * C) -> (A+-B) * C): Guard with
3403         explicit single_use checks.
3404
3405 2018-03-27  Richard Biener  <rguenther@suse.de>
3406
3407         PR tree-optimization/85082
3408         * tree-ssa-sccvn.c (vn_reference_lookup_or_insert_for_pieces):
3409         Valueize the VUSE.
3410
3411 2018-03-27  Ramana Radhakrishnan  <ramana.radhakrishnan@arm.com>
3412
3413         * config.gcc (aarch64*-*-linux*): New TARGET_DEFAULT_ASYNC_UNWIND_TABLES
3414         * common/config/aarch64/aarch64-common.c (aarch64_optimization_table[]):
3415         Turn on fasynchronous-unwind-tables and funwind-tables.
3416
3417 2018-03-26  Uros Bizjak  <ubizjak@gmail.com>
3418
3419         PR target/85073
3420         * config/i386/i386.md (*bmi_blsr_<mode>_cmp): New insn pattern.
3421         (*bmi_blsr_<mode>_ccz): Ditto.
3422
3423 2018-03-26  Tom de Vries  <tom@codesourcery.com>
3424
3425         PR tree-optimization/85063
3426         * omp-general.c (offloading_function_p): New function.  Factor out
3427         of ...
3428         * omp-offload.c (pass_omp_target_link::gate): ... here.
3429         * omp-general.h (offloading_function_p): Declare.
3430         * tree-switch-conversion.c (build_one_array): Mark CSWTCH.x variable
3431         with attribute omp declare target for offloading functions.
3432
3433 2018-03-24  Richard Sandiford  <richard.sandiford@linaro.org>
3434
3435         PR tree-optimization/84005
3436         * tree-data-ref.h (get_base_for_alignment): Declare.
3437         * tree-data-ref.c (get_base_for_alignment_1): New function.
3438         (get_base_for_alignment): Likewise.
3439         * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Use
3440         get_base_for_alignment to find a suitable base object, instead
3441         of always using drb->base_address.
3442
3443 2018-03-23  Jakub Jelinek  <jakub@redhat.com>
3444
3445         PR inline-asm/85022
3446         * emit-rtl.c (init_emit_regs): Indicate that VOIDmode MEMs don't have
3447         known size by default.
3448
3449 2018-03-23  Vladimir Makarov  <vmakarov@redhat.com>
3450
3451         PR inline-asm/85030
3452         * lra-constraints.c (process_alt_operands): Don't match BLKmode
3453         and non BLKmode operands.
3454
3455 2018-03-23  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
3456
3457         PR target/85026
3458         * config/arm/arm.md (unaligned_loadhis): Remove first alternative.
3459         Clean up attributes.
3460
3461 2018-03-23  Richard Biener  <rguenther@suse.de>
3462
3463         PR debug/85020
3464         * dwarf2out.c (rtl_for_decl_location): Do not generate RTL early when
3465         we are going to emit early debug for LTO.
3466
3467 2018-03-23  Jakub Jelinek  <jakub@redhat.com>
3468
3469         PR inline-asm/85034
3470         * function.c (match_asm_constraints_1): Don't optimize if input
3471         doesn't satisfy general_operand predicate for output's mode.
3472
3473         PR inline-asm/85022
3474         * alias.c (write_dependence_p): Don't require for x_canonicalized
3475         non-VOIDmode if x has VOIDmode.
3476
3477         PR sanitizer/85029
3478         * sanopt.c (maybe_optimize_ubsan_ptr_ifn): If DECL_REGISTER (base),
3479         just don't try to optimize it rather than assert it never happens.
3480
3481 2018-03-22  Kelvin Nilsen  <kelvin@gcc.gnu.org>
3482
3483         * config/rs6000/rs6000-builtin.def: Remove various BU_ALTIVEC_X
3484         macro expansions for definition of ST_INTERNAL_<mode> and
3485         LD_INTERNAL_<mode> builtins.
3486         * config/rs6000/rs6000-protos.h (rs6000_address_for_altivec):
3487         Remove prototype.
3488         * config/rs6000/rs6000.c (altivec_expand_ld_builtin): Delete this
3489         function.
3490         (altivec_expand_st_builtin): Likewise.
3491         (altivec_expand_builtin): Remove calls to deleted functions.
3492         (rs6000_address_for_altivec): Delete this function.
3493         * config/rs6000/vector.md: Remove expands for
3494         vector_altivec_load_<mode> and vector_altivec_store_<mode>.
3495
3496 2018-03-22  Sudakshina Das  <sudi.das@arm.com>
3497
3498         PR target/84826
3499         * config/arm/arm.h (machine_function): Add static_chain_stack_bytes.
3500         * config/arm/arm.c (arm_compute_static_chain_stack_bytes): Avoid
3501         re-computing once computed.
3502         (arm_expand_prologue): Compute machine->static_chain_stack_bytes.
3503         (arm_init_machine_status): Initialize
3504         machine->static_chain_stack_bytes.
3505
3506 2018-03-22  Kelvin Nilsen  <kelvin@gcc.gnu.org>
3507
3508         PR target/84760
3509         * doc/extend.texi: Add four new prototypes for vec_ld.
3510         * config/rs6000/rs6000-builtin.def (LVX_V1TI): Reorder symbol
3511         definitions for more logical presentation.
3512         * config/rs6000/rs6000-c.c: (altivec_overloaded_builtins): Add
3513         entries for V1TI variants of __builtin_altivec_ld builtin.
3514         * config/rs6000/rs6000.c: (altivec_expand_lv_builtin): Add test and
3515         handling of V1TI variant of LVX icode pattern.
3516         (altivec_expand_builtin): Add case for ALTIVEC_BUILTIN_LVX_V1TI.
3517         (rs6000_gimple_fold_builtin): Likewise.
3518         (altivec_init_builtins): Add code to define
3519         __builtin_altivec_lvx_v1ti function.
3520
3521 2018-03-22  Jakub Jelinek  <jakub@redhat.com>
3522
3523         PR inline-asm/84941
3524         * function.c (match_asm_constraints_1): Don't do the optimization
3525         if input isn't a REG, SUBREG, MEM or constant.
3526
3527 2018-03-22  Tom de Vries  <tom@codesourcery.com>
3528
3529         PR tree-optimization/84956
3530         * tree-ssa-tail-merge.c (find_clusters_1): Skip bbs with
3531         bb_has_abnormal_pred.
3532
3533 2018-03-22  Jakub Jelinek  <jakub@redhat.com>
3534
3535         PR sanitizer/85018
3536         * dwarf2asm.c (dw2_output_indirect_constant_1): Set
3537         DECL_INITIAL (decl) to decl at the end.
3538         * varasm.c (use_blocks_for_decl_p): Revert the 2018-03-20 change,
3539         adjust the comment.
3540
3541 2018-03-21  Joseph Myers  <joseph@codesourcery.com>
3542
3543         * doc/extend.texi (__builtin_tgmath): Document when complex
3544         integer types are treated as _Complex _Float64.
3545
3546 2018-03-21  Tom de Vries  <tom@codesourcery.com>
3547
3548         * doc/extend.texi (__builtin_extend_pointer): Remove pasto.
3549
3550 2018-03-21  Jakub Jelinek  <jakub@redhat.com>
3551
3552         PR tree-optimization/84960
3553         * tree-cfg.c (remove_bb): Don't move forced labels into bb->prev_bb
3554         if it is ENTRY block, move them into single succ of ENTRY in that case.
3555
3556 2018-03-21  Richard Sandiford  <richard.sandiford@linaro.org>
3557
3558         PR tree-optimization/84811
3559         * poly-int.h (poly_span_traits): Remove the T3 parameter and
3560         promote HOST_WIDE_INT T2 - T1 results to unsigned HOST_WIDE_INT.
3561         (maybe_in_range_p, known_in_range_p, ranges_known_overlap_p):
3562         (known_subrange_p): Update accordingly.  Cast each value involved
3563         in the size comparison, rather than casting the result of the
3564         subtraction.
3565
3566 2018-03-21  Jakub Jelinek  <jakub@redhat.com>
3567
3568         PR tree-optimization/84982
3569         * gimple-ssa-store-merging.c (invert_op): Handle boolean inversion
3570         by flipping the least significant bit rather than all bits from
3571         bitpos to bitpos + bitsize - 1.
3572
3573 2018-03-21  Nathan Sidwell  <nathan@acm.org>
3574
3575         * doc/extend.texi (Deprecated Features): Remove mention of
3576         long-deleted deprecations.
3577
3578 2018-03-21  Rainer Orth  <ro@CeBiTec.Uni-Bielefeld.DE>
3579
3580         PR jit/84288
3581         * configure.ac (gcc_cv_ld_soname) <*-*-solaris2*>: Set.
3582         * configure: Regenerate.
3583
3584 2018-03-21  Tom de Vries  <tom@codesourcery.com>
3585
3586         PR tree-optimization/83126
3587         * tree-parloops.c (num_phis): New function.
3588         (gen_parallel_loop): Detect and handle canonicalize_loop_ivs failure.
3589
3590 2018-03-21  Nathan Sidwell  <nathan@acm.org>
3591
3592         * doc/extend.texi (Deprecated Features): Update deprecated flags,
3593         mention anon-struct/union members and trailing attributes.
3594
3595 2018-03-21  Bin Cheng  <bin.cheng@arm.com>
3596
3597         PR tree-optimization/84969
3598         * tree-loop-distribution.c (fuse_memset_builtins): Don't reorder
3599         builtin memset partitions if they set different rhs values.
3600
3601 2018-03-21  Jakub Jelinek  <jakub@redhat.com>
3602
3603         PR rtl-optimization/84989
3604         * simplify-rtx.c (simplify_unary_operation_1): Don't try to simplify
3605         VEC_DUPLICATE with scalar result mode.
3606
3607 2018-03-21  Martin Liska  <mliska@suse.cz>
3608
3609         PR ipa/84963
3610         * ipa-icf.c (sem_item_optimizer::fixup_points_to_sets): Remove
3611         not intended return statement.
3612
3613 2018-03-21  Martin Liska  <mliska@suse.cz>
3614
3615         PR target/84988
3616         * tree-chkp.c (CHKP_ARRAY_MAX_CHECK_STEPS): Define a new macro.
3617         (chkp_find_bound_slots_1): Limit number of iterations.
3618
3619 2018-03-20  David H. Gutteridge  <dhgutteridge@sympatico.ca>
3620
3621         PR target/84838
3622         * Minor grammar fixes for x86 options.
3623
3624 2018-03-20  Jakub Jelinek  <jakub@redhat.com>
3625
3626         PR debug/84875
3627         * dce.c (delete_unmarked_insns): Don't remove frame related noop moves
3628         holding REG_CFA_RESTORE notes, instead turn them into a USE.
3629
3630 2018-03-20  Peter Bergner  <bergner@vnet.ibm.com>
3631
3632         PR target/83789
3633         * config/rs6000/altivec.md (altivec_lvx_<mode>_2op): Delete define_insn.
3634         (altivec_lvx_<mode>_1op): Likewise.
3635         (altivec_stvx_<mode>_2op): Likewise.
3636         (altivec_stvx_<mode>_1op): Likewise.
3637         (altivec_lvx_<VM2:mode>): New define_expand.
3638         (altivec_stvx_<VM2:mode>): Likewise.
3639         (altivec_lvx_<VM2:mode>_2op_<P:mptrsize>): New define_insn.
3640         (altivec_lvx_<VM2:mode>_1op_<P:mptrsize>): Likewise.
3641         (altivec_stvx_<VM2:mode>_2op_<P:mptrsize>): Likewise.
3642         (altivec_stvx_<VM2:mode>_1op_<P:mptrsize>): Likewise.
3643         * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Use new expanders.
3644         (rs6000_gen_lvx): Likewise.
3645         * config/rs6000/rs6000.c (altivec_expand_lv_builtin): Likewise.
3646         (altivec_expand_stv_builtin): Likewise.
3647         (altivec_expand_builtin): Likewise.
3648         * config/rs6000/vector.md: Likewise.
3649
3650 2018-03-20  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
3651
3652         PR target/82518
3653         * config/arm/arm.c (arm_array_mode_supported_p): Return false for
3654         BYTES_BIG_ENDIAN.
3655
3656 2018-03-20  Richard Biener  <rguenther@suse.de>
3657
3658         PR target/84986
3659         * config/i386/i386.c (ix86_add_stmt_cost): Only cost
3660         sign-conversions as zero, fall back to standard scalar_stmt
3661         cost for the rest.
3662
3663 2018-03-20  Martin Liska  <mliska@suse.cz>
3664
3665         PR ipa/84825
3666         * predict.c (rebuild_frequencies): Handle case when we have
3667         PROFILE_ABSENT, but flag_guess_branch_prob is false.
3668
3669 2018-03-20  Jakub Jelinek  <jakub@redhat.com>
3670
3671         PR target/84990
3672         * dwarf2asm.c (dw2_output_indirect_constant_1): Temporarily turn off
3673         flag_section_anchors.
3674         * varasm.c (use_blocks_for_decl_p): Remove hack for
3675         dw2_force_const_mem.
3676
3677         PR target/84845
3678         * config/aarch64/aarch64.md (*aarch64_reg_<mode>3_neg_mask2): Rename
3679         to ...
3680         (*aarch64_<optab>_reg_<mode>3_neg_mask2): ... this.  If pseudos can't
3681         be created, use lowpart_subreg of operands[0] rather than operands[0]
3682         itself.
3683         (*aarch64_reg_<mode>3_minus_mask): Rename to ...
3684         (*aarch64_ashl_reg_<mode>3_minus_mask): ... this.
3685         (*aarch64_<optab>_reg_di3_mask2): Use const_int_operand predicate
3686         and n constraint instead of aarch64_shift_imm_di and Usd.
3687         (*aarch64_reg_<optab>_minus<mode>3): Rename to ...
3688         (*aarch64_<optab>_reg_minus<mode>3): ... this.
3689
3690 2018-03-20  Sudakshina Das  <sudi.das@arm.com>
3691
3692         PR target/82989
3693         * config/arm/neon.md (ashldi3_neon): Update ?s for constraints
3694         to favor GPR over NEON registers.
3695         (<shift>di3_neon): Likewise.
3696
3697 2018-03-20  Tom de Vries  <tom@codesourcery.com>
3698
3699         PR target/84952
3700         * config/nvptx/nvptx.c (nvptx_single): Don't neuter bar.sync.
3701         (nvptx_process_pars): Emit bar.sync asap and alap.
3702
3703 2018-03-20  Tom de Vries  <tom@codesourcery.com>
3704
3705         PR target/84954
3706         * config/nvptx/nvptx.c (prevent_branch_around_nothing): Also update
3707         seen_label if seen_label is already set.
3708
3709 2018-03-20  Jakub Jelinek  <jakub@redhat.com>
3710
3711         PR target/84945
3712         * config/i386/i386.c (fold_builtin_cpu): For features above 31
3713         use __cpu_features2 variable instead of __cpu_model.__cpu_features[0].
3714         Use 1U instead of 1.  Formatting fixes.
3715
3716         PR c/84953
3717         * builtins.c (fold_builtin_strpbrk): For strpbrk(x, "") use type
3718         instead of TREE_TYPE (s1) for the return value.
3719
3720 2018-03-19  Jakub Jelinek  <jakub@redhat.com>
3721
3722         PR tree-optimization/84946
3723         * gimple-ssa-store-merging.c (mem_valid_for_store_merging): Compute
3724         bitsize + bitsize in poly_uint64 rather than poly_int64.
3725
3726         PR sanitizer/78651
3727         * dwarf2asm.c: Include fold-const.c.
3728         (dw2_output_indirect_constant_1): Set DECL_INITIAL (decl) to ADDR_EXPR
3729         of decl rather than decl itself.
3730
3731         PR rtl-optimization/84643
3732         * memmodel.h (enum memmodel): Add MEMMODEL_MAX enumerator.
3733
3734 2018-03-19  Maxim Ostapenko  <m.ostapenko@samsung.com>
3735
3736         PR sanitizer/78651
3737         * dwarf2asm.c (dw2_output_indirect_constant_1): Disable ASan before
3738         calling assemble_variable.
3739
3740 2018-03-19  Sudakshina Das  <sudi.das@arm.com>
3741
3742         PR target/81647
3743         * config/aarch64/aarch64-simd.md (vec_cmp<mode><v_int_equiv>): Modify
3744         instructions for UNLT, UNLE, UNGT, UNGE, UNEQ, UNORDERED and ORDERED.
3745
3746 2018-03-19  Jim Wilson  <jimw@sifive.com>
3747
3748         PR bootstrap/84856
3749         * config/riscv/riscv.c (riscv_function_arg_boundary): Use
3750         PREFERRED_STACK_BOUNDARY instead of STACK_BOUNDARY.
3751         (riscv_first_stack_step): Likewise.
3752         (riscv_option_override): Use STACK_BOUNDARY instead of
3753         MIN_STACK_BOUNDARY.
3754         * config/riscv/riscv.h (STACK_BOUNDARY): Renamed from
3755         MIN_STACK_BOUNDARY.
3756         (BIGGEST_ALIGNMENT): Set to 128.
3757         (PREFERRED_STACK_BOUNDARY): Renamed from STACK_BOUNDARY.
3758         (RISCV_STACK_ALIGN): Use PREFERRED_STACK_BOUNDARY instead of
3759         STACK_BOUNDARY.
3760
3761 2018-03-19  Richard Biener  <rguenther@suse.de>
3762
3763         PR tree-optimization/84933
3764         * tree-vrp.c (set_and_canonicalize_value_range): Treat out-of-bound
3765         values as -INF/INF when canonicalizing an ANTI_RANGE to a RANGE.
3766
3767 2018-03-19  Richard Biener  <rguenther@suse.de>
3768
3769         PR tree-optimization/84859
3770         * tree-ssa-phiopt.c (single_trailing_store_in_bb): New function.
3771         (cond_if_else_store_replacement): Perform sinking operation on
3772         single-store BBs regardless of MAX_STORES_TO_SINK setting.
3773         Generalize what a BB with a single eligible store is.
3774
3775 2018-03-19  Richard Biener  <rguenther@suse.de>
3776
3777         PR tree-optimization/84929
3778         * tree-data-ref.c (analyze_siv_subscript_cst_affine): Guard
3779         chrec_is_positive against non-chrec arg.
3780
3781 2018-03-19  Tamar Christina  <tamar.christina@arm.com>
3782
3783         PR target/84711
3784         * config/arm/arm.c (arm_can_change_mode_class): revert r258554.
3785
3786 2018-03-18  Martin Liska  <mliska@suse.cz>
3787
3788         PR rtl-optimization/84635
3789         * regrename.c (build_def_use): Use matches_mode only when
3790         matches >= 0.
3791
3792 2018-03-18  Richard Sandiford  <richard.sandiford@linaro.org>
3793
3794         PR tree-optimization/84913
3795         * tree-vect-loop.c (vectorizable_reduction): Don't try to
3796         vectorize chains of COND_EXPRs.
3797
3798 2018-03-18  Chung-Ju Wu  <jasonwucj@gmail.com>
3799
3800         * config/nds32/nds32.h (MAX_REGS_PER_ADDRESS): Fix the value.
3801
3802 2018-03-18  Chung-Ju Wu  <jasonwucj@gmail.com>
3803
3804         * config/nds32/nds32.h (LOGICAL_OP_NON_SHORT_CIRCUIT): Define.
3805
3806 2018-03-18  Chung-Ju Wu  <jasonwucj@gmail.com>
3807
3808         * config/nds32/nds32.h (CLZ_DEFINED_VALUE_AT_ZERO): Define.
3809
3810 2018-03-17  Chung-Ju Wu  <jasonwucj@gmail.com>
3811             Kito Cheng  <kito.cheng@gmail.com>
3812
3813         * config/nds32/nds32-protos.h (nds32_adjust_reg_alloc_order): Declare.
3814         * config/nds32/nds32.c (nds32_reg_alloc_order_for_speed): New array.
3815         (nds32_adjust_reg_alloc_order): New function.
3816         * config/nds32/nds32.h (ADJUST_REG_ALLOC_ORDER): Define.
3817
3818 2018-03-17  Kito Cheng  <kito.cheng@gmail.com>
3819
3820         * config/nds32/nds32.c (nds32_asm_output_mi_thunk,
3821         nds32_print_operand, nds32_print_operand_address): Use
3822         HOST_WIDE_INT_PRINT_DEC instead.
3823
3824 2018-03-17  Chung-Ju Wu  <jasonwucj@gmail.com>
3825
3826         * config/nds32/nds32.c (nds32_register_priority): Modify cost.
3827
3828 2018-03-17  Jakub Jelinek  <jakub@redhat.com>
3829
3830         PR target/84902
3831         * config/i386/i386.c (initial_ix86_tune_features,
3832         initial_ix86_arch_features): Use unsigned HOST_WIDE_INT rather than
3833         unsigned long long.
3834         (set_ix86_tune_features): Change ix86_tune_mask from unsigned int
3835         to unsigned HOST_WIDE_INT, initialize to HOST_WIDE_INT_1U << ix86_tune
3836         rather than 1u << ix86_tune.  Formatting fix.
3837         (ix86_option_override_internal): Change ix86_arch_mask from
3838         unsigned int to unsigned HOST_WIDE_INT, initialize to
3839         HOST_WIDE_INT_1U << ix86_arch rather than 1u << ix86_arch.
3840         (ix86_function_specific_restore): Likewise.
3841
3842 2018-03-16  Jakub Jelinek  <jakub@redhat.com>
3843
3844         PR target/84899
3845         * postreload.c (reload_combine_recognize_pattern): Perform
3846         INTVAL addition in unsigned HOST_WIDE_INT type to avoid UB and
3847         truncate_int_for_mode the result for the destination's mode.
3848
3849         PR c/84909
3850         * hsa-gen.c (mem_type_for_type): Fix comment typo.
3851         * tree-vect-loop-manip.c (vect_create_cond_for_niters_checks):
3852         Likewise.
3853         * gimple-ssa-warn-restrict.c (builtin_memref::set_base_and_offset):
3854         Likewise.
3855
3856 2018-03-16  Vladimir Makarov  <vmakarov@redhat.com>
3857
3858         PR target/84876
3859         * lra-assigns.c (lra_split_hard_reg_for): Don't use
3860         regno_allocno_class_array and sorted_pseudos.
3861         * lra-constraints.c (spill_hard_reg_in_range): Ignore hard regs in
3862         insns where regno is used.
3863
3864 2018-03-16  Martin Liska  <mliska@suse.cz>
3865
3866         PR ipa/84833
3867         * multiple_target.c (create_dispatcher_calls): Redirect
3868         reference in the symbol table.
3869
3870 2018-03-16  Martin Liska  <mliska@suse.cz>
3871
3872         PR ipa/84722
3873         * multiple_target.c (create_dispatcher_calls): Redirect also
3874         an alias.
3875
3876 2018-03-16  Jakub Jelinek  <jakub@redhat.com>
3877
3878         PR c++/79937
3879         PR c++/82410
3880         * tree.h (TARGET_EXPR_NO_ELIDE): Define.
3881         * gimplify.c (gimplify_modify_expr_rhs): Don't elide TARGET_EXPRs with
3882         TARGET_EXPR_NO_ELIDE flag set unless *expr_p is INIT_EXPR.
3883
3884 2018-03-16  Julia Koval  <julia.koval@intel.com>
3885
3886         * doc/invoke.texi (Skylake Server): Add CLWB.
3887         Cannonlake): Remove CLWB.
3888
3889 2018-03-16  Jakub Jelinek  <jakub@redhat.com>
3890
3891         PR tree-optimization/84841
3892         * tree-ssa-reassoc.c (INTEGER_CONST_TYPE): Change to 1 << 4 from
3893         1 << 3.
3894         (FLOAT_ONE_CONST_TYPE): Define.
3895         (constant_type): Return FLOAT_ONE_CONST_TYPE for -1.0 and 1.0.
3896         (sort_by_operand_rank): Put entries with higher constant_type last
3897         rather than first to match comments.
3898
3899 2018-03-15  Sandra Loosemore  <sandra@codesourcery.com>
3900
3901         * config/nios2/nios2.md (movsi_internal): Fix thinko in
3902         split predicate.
3903
3904 2018-03-15  Jakub Jelinek  <jakub@redhat.com>
3905
3906         PR c++/79085
3907         * calls.c (expand_call): For TREE_ADDRESSABLE rettype ignore alignment
3908         check and use address of target always.
3909
3910 2018-03-15  H.J. Lu  <hongjiu.lu@intel.com>
3911
3912         PR target/84574
3913         * config/i386/i386.c (indirect_thunk_needed): Update comments.
3914         (indirect_thunk_bnd_needed): Likewise.
3915         (indirect_thunks_used): Likewise.
3916         (indirect_thunks_bnd_used): Likewise.
3917         (indirect_return_needed): New.
3918         (indirect_return_bnd_needed): Likewise.
3919         (output_indirect_thunk_function): Add a bool argument for
3920         function return.
3921         (output_indirect_thunk_function): Don't generate alias for
3922         function return thunk.
3923         (ix86_code_end): Call output_indirect_thunk_function to generate
3924         function return thunks.
3925         (ix86_output_function_return): Set indirect_return_bnd_needed
3926         and indirect_return_needed instead of indirect_thunk_bnd_needed
3927         and indirect_thunk_needed.
3928
3929 2018-03-15  Olga Makhotina  <olga.makhotina@intel.com>
3930
3931         * config/i386/sgxintrin.h (_enclv_u32): New intrinsic.
3932         (__enclv_bc, __enclv_cd, __enclv_generic): New definitions.
3933         (ERDINFO, ETRACKC, ELDBC, ELDUC): New leaves.
3934
3935 2018-03-15  David Malcolm  <dmalcolm@redhat.com>
3936             Paul Hua <paul.hua.gm@gmail.com>
3937
3938         PR c/84852
3939         * gcc.dg/fixits-pr84852-1.c: Fix filename in dg-regexp.
3940
3941 2018-03-15  Segher Boessenkool  <segher@kernel.crashing.org>
3942
3943         * config/rs6000/rs6000.c (abi_v4_pass_in_fpr): Add back the
3944         TARGET_DOUBLE_FLOAT and TARGET_SINGLE_FLOAT conditions on the DFmode
3945         resp. SFmode cases.
3946
3947 2018-03-15  Tamar Christina  <tamar.christina@arm.com>
3948
3949         PR target/84711
3950         * config/arm/arm.c (arm_can_change_mode_class): Use GET_MODE_UNIT_SIZE
3951         instead of GET_MODE_SIZE when comparing Units.
3952
3953 2018-03-15  Vladimir Mezentsev  <vladimir.mezentsev@oracle.com>
3954
3955         PR target/68256
3956         * varasm.c (hash_section): Return an unchangeble hash value
3957         * config/aarch64/aarch64.c (aarch64_use_blocks_for_constant_p):
3958         Return !aarch64_can_use_per_function_literal_pools_p ().
3959
3960 2018-03-15  Jakub Jelinek  <jakub@redhat.com>
3961
3962         PR target/84860
3963         * optabs.c (emit_conditional_move): Pass address of cmode's copy
3964         rather than address of cmode as last argument to prepare_cmp_insn.
3965
3966 2018-03-15  Julia Koval  <julia.koval@intel.com>
3967
3968         * config/i386/i386.c (F_AVX512VBMI2, F_GFNI, F_VPCLMULQDQ,
3969         F_AVX512VNNI, F_AVX512BITALG): New.
3970
3971 2018-03-14  John David Anglin  <danglin@gcc.gnu.org>
3972
3973         PR target/83451
3974         * config/pa/pa.c (pa_emit_move_sequence):  Always emit secondary reload
3975         insn for floating-point loads and stores.
3976
3977 2018-03-14  Carl Love  <cel@us.ibm.com>
3978
3979         * config/rs6000/rs6000-c.c: Add macro definitions for
3980         ALTIVEC_BUILTIN_VEC_PERMXOR.
3981         * config/rs6000/rs6000.h: Add #define for vec_permxor builtin.
3982         * config/rs6000/rs6000-builtin.def: Add macro expansions for VPERMXOR.
3983         * config/rs6000/altivec.md (altivec_vpermxor): New define expand.
3984         * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Add case
3985         UNSPEC_VPERMXOR.
3986         * config/doc/extend.texi: Add prototypes for vec_permxor.
3987
3988 2018-03-14  David Malcolm  <dmalcolm@redhat.com>
3989
3990         PR c/84852
3991         * diagnostic-show-locus.c (class layout_point): Convert m_line
3992         from int to linenum_type.
3993         (line_span::comparator): Use linenum "compare" function when
3994         comparing line numbers.
3995         (test_line_span): New function.
3996         (layout_range::contains_point): Convert param "row" from int to
3997         linenum_type.
3998         (layout_range::intersects_line_p): Likewise.
3999         (layout::will_show_line_p): Likewise.
4000         (layout::print_source_line): Likewise.
4001         (layout::should_print_annotation_line_p): Likewise.
4002         (layout::print_annotation_line): Likewise.
4003         (layout::print_leading_fixits): Likewise.
4004         (layout::annotation_line_showed_range_p): Likewise.
4005         (struct line_corrections): Likewise for field m_row.
4006         (line_corrections::line_corrections): Likewise for param "row".
4007         (layout::print_trailing_fixits): Likewise.
4008         (layout::get_state_at_point): Likewise.
4009         (layout::get_x_bound_for_row): Likewise.
4010         (layout::print_line): Likewise.
4011         (diagnostic_show_locus): Likewise for locals "last_line" and
4012         "row".
4013         (selftest::diagnostic_show_locus_c_tests): Call test_line_span.
4014         * input.c (selftest::test_linenum_comparisons): New function.
4015         (selftest::input_c_tests): Call it.
4016         * selftest.c (selftest::test_assertions): Test ASSERT_GT,
4017         ASSERT_GT_AT, ASSERT_LT, and ASSERT_LT_AT.
4018         * selftest.h (ASSERT_GT): New macro.
4019         (ASSERT_GT_AT): New macro.
4020         (ASSERT_LT): New macro.
4021         (ASSERT_LT_AT): New macro.
4022
4023 2018-03-14  Segher Boessenkool  <segher@kernel.crashing.org>
4024
4025         PR rtl-optimization/84780
4026         * combine.c (distribute_links): Don't make a link based on pc_rtx.
4027
4028 2018-03-14  Martin Liska  <mliska@suse.cz>
4029
4030         * tree.c (record_node_allocation_statistics): Use
4031         get_stats_node_kind.
4032         (get_stats_node_kind): New function extracted from
4033         record_node_allocation_statistics.
4034         (free_node): Use get_stats_node_kind.
4035
4036 2018-03-14  Richard Biener  <rguenther@suse.de>
4037
4038         * tree-ssa-pre.c (compute_antic_aux): Remove code that asserts
4039         that the value-set of ANTIC_IN doesn't grow.
4040
4041         Revert
4042         * tree-ssa-pre.c (struct bb_bitmap_sets): Add visited_with_visited_succs
4043         member.
4044         (BB_VISITED_WITH_VISITED_SUCCS): New define.
4045         (compute_antic): Initialize BB_VISITED_WITH_VISITED_SUCCS.
4046
4047 2018-03-14  Julia Koval  <julia.koval@intel.com>
4048
4049         * config.gcc (icelake-client, icelake-server): New.
4050         (icelake): Remove.
4051         * config/i386/i386.c (initial_ix86_tune_features): Extend to 64 bit.
4052         (initial_ix86_arch_features): Ditto.
4053         (PTA_SKYLAKE): Add SGX.
4054         (PTA_ICELAKE): Remove.
4055         (PTA_ICELAKE_CLIENT): New.
4056         (PTA_ICELAKE_SERVER): New.
4057         (ix86_option_override_internal): Split up icelake on icelake client and
4058         icelake server.
4059         (get_builtin_code_for_version): Ditto.
4060         (fold_builtin_cpu): Ditto.
4061         * config/i386/driver-i386.c (config/i386/driver-i386.c): Ditto.
4062         * config/i386/i386-c.c (ix86_target_macros_internal): Ditto
4063         * config/i386/i386.h (processor_type): Ditto.
4064         * doc/invoke.texi: Ditto.
4065
4066 2018-03-14  Jakub Jelinek  <jakub@redhat.com>
4067
4068         PR sanitizer/83392
4069         * sanopt.c (maybe_optimize_ubsan_ptr_ifn): Handle also
4070         INTEGER_CST offset, add it together with bitpos / 8 and
4071         sign extend based on POINTER_SIZE.
4072
4073         PR target/84844
4074         Revert
4075         2017-04-20  Uros Bizjak  <ubizjak@gmail.com>
4076
4077         PR target/78090
4078         * config/i386/constraints.md (Yc): New register constraint.
4079         * config/i386/i386.md (*float<SWI48:mode><MODEF:mode>2_mixed):
4080         Use Yc constraint for alternative 2 of operand 0.  Remove
4081         preferred_for_speed attribute.
4082
4083 2018-03-14  Richard Biener  <rguenther@suse.de>
4084
4085         PR tree-optimization/84830
4086         * tree-ssa-pre.c (compute_antic_aux): Intersect the new ANTIC_IN
4087         with the old one to avoid oscillations.
4088
4089 2018-03-13  Vladimir Makarov  <vmakarov@redhat.com>
4090
4091         PR target/83712
4092         * lra-assigns.c (find_all_spills_for): Ignore uninteresting
4093         pseudos.
4094         (assign_by_spills): Return a flag of reload assignment failure.
4095         Do not process the reload assignment failures.  Do not spill other
4096         reload pseudos if they has the same reg class.  Update n if
4097         necessary.
4098         (lra_assign): Add a return arg.  Set up from the result of
4099         assign_by_spills call.
4100         (find_reload_regno_insns, lra_split_hard_reg_for): New functions.
4101         * lra-constraints.c (split_reg): Add a new arg.  Use it instead of
4102         usage_insns if it is not NULL.
4103         (spill_hard_reg_in_range): New function.
4104         (split_if_necessary, inherit_in_ebb): Pass a new arg to split_reg.
4105         * lra-int.h (spill_hard_reg_in_range, lra_split_hard_reg_for): New
4106         function prototypes.
4107         (lra_assign): Change prototype.
4108         * lra.c (lra): Add code to deal with fails by splitting hard reg
4109         live ranges.
4110
4111 2018-03-01  Palmer Dabbelt  <palmer@sifive.com>
4112
4113         * config/riscv/riscv.opt (mrelax): New option.
4114         * config/riscv/riscv.c (riscv_file_start): Emit ".option
4115         "norelax" when riscv_mrelax is disabled.
4116         * doc/invoke.texi (RISC-V): Document "-mrelax" and "-mno-relax".
4117
4118 2018-03-13  Aaron Sawdey  <acsawdey@linux.vnet.ibm.com>
4119
4120         PR target/84743
4121         * config/rs6000/rs6000.c (rs6000_reassociation_width): Disable parallel
4122         reassociation for int modes.
4123
4124 2018-03-13  Richard Sandiford  <richard.sandiford@linaro.org>
4125
4126         * tree-vect-loop-manip.c (vect_maybe_permute_loop_masks):
4127         Reverse the choice between VEC_UNPACK_LO_EXPR and VEC_UNPACK_HI_EXPR
4128         for big-endian.
4129         * config/aarch64/iterators.md (hi_lanes_optab): New int attribute.
4130         * config/aarch64/aarch64-sve.md
4131         (*aarch64_sve_<perm_insn><perm_hilo><mode>): Rename to...
4132         (aarch64_sve_<perm_insn><perm_hilo><mode>): ...this.
4133         (*extend<mode><Vwide>2): Rename to...
4134         (aarch64_sve_extend<mode><Vwide>2): ...this.
4135         (vec_unpack<su>_<perm_hilo>_<mode>): Turn into a define_expand,
4136         renaming the old pattern to...
4137         (aarch64_sve_punpk<perm_hilo>_<mode>): ...this.  Only define
4138         unsigned packs.
4139         (vec_unpack<su>_<perm_hilo>_<SVE_BHSI:mode>): Turn into a
4140         define_expand, renaming the old pattern to...
4141         (aarch64_sve_<su>unpk<perm_hilo>_<SVE_BHSI:mode>): ...this.
4142         (*vec_unpacku_<perm_hilo>_<mode>_no_convert): Delete.
4143         (vec_unpacks_<perm_hilo>_<mode>): Take BYTES_BIG_ENDIAN into
4144         account when deciding which SVE instruction the optab should use.
4145         (vec_unpack<su_optab>_float_<perm_hilo>_vnx4si): Likewise.
4146
4147 2018-03-13  Richard Sandiford  <richard.sandiford@linaro.org>
4148
4149         * config/aarch64/aarch64.md (V4_REGNUM, V8_REGNUM, V12_REGNUM)
4150         (V20_REGNUM, V24_REGNUM, V28_REGNUM, P1_REGNUM, P2_REGNUM, P3_REGNUM)
4151         (P4_REGNUM, P5_REGNUM, P6_REGNUM, P8_REGNUM, P9_REGNUM, P10_REGNUM)
4152         (P11_REGNUM, P12_REGNUM, P13_REGNUM, P14_REGNUM): New define_constants.
4153         (tlsdesc_small_<mode>): Turn a define_expand and use
4154         tlsdesc_small_sve_<mode> for SVE.  Rename original define_insn to...
4155         (tlsdesc_small_advsimd_<mode>): ...this.
4156         (tlsdesc_small_sve_<mode>): New pattern.
4157
4158 2018-03-13  Richard Sandiford  <richard.sandiford@linaro.org>
4159
4160         * config/aarch64/iterators.md (UNSPEC_SMUL_HIGHPART)
4161         (UNSPEC_UMUL_HIGHPART): New constants.
4162         (MUL_HIGHPART): New int iteraor.
4163         (su): Handle UNSPEC_SMUL_HIGHPART and UNSPEC_UMUL_HIGHPART.
4164         * config/aarch64/aarch64-sve.md (<su>mul<mode>3_highpart): New
4165         define_expand.
4166         (*<su>mul<mode>3_highpart): New define_insn.
4167
4168 2018-03-13  Eric Botcazou  <ebotcazou@adacore.com>
4169
4170         PR lto/84805
4171         * ipa-devirt.c (odr_subtypes_equivalent_p): Do not get the ODR type of
4172         incomplete types.
4173
4174 2018-03-13  Martin Liska  <mliska@suse.cz>
4175
4176         PR ipa/84658.
4177         * (sem_item_optimizer::sem_item_optimizer): Initialize new
4178         vector.
4179         (sem_item_optimizer::~sem_item_optimizer): Release it.
4180         (sem_item_optimizer::merge_classes): Register variable aliases.
4181         (sem_item_optimizer::fixup_pt_set): New function.
4182         (sem_item_optimizer::fixup_points_to_sets): Likewise.
4183         * ipa-icf.h: Declare new variables and functions.
4184
4185 2018-03-13  Jakub Jelinek  <jakub@redhat.com>
4186
4187         PR middle-end/84834
4188         * match.pd ((A & C) != 0 ? D : 0): Use INTEGER_CST@2 instead of
4189         integer_pow2p@2 and test integer_pow2p in condition.
4190         (A < 0 ? C : 0): Similarly for @1.
4191
4192         PR middle-end/84831
4193         * stmt.c (parse_output_constraint): If the CONSTRAINT_LEN (*p, p)
4194         characters starting at p contain '\0' character, don't look beyond
4195         that.
4196
4197         PR target/84827
4198         * config/i386/i386.md (round<mode>2): For 387 fancy math, disable
4199         pattern if -ftrapping-math -fno-fp-int-builtin-inexact.
4200
4201         PR target/84828
4202         * reg-stack.c (change_stack): Change update_end var from int to
4203         rtx_insn *, if non-NULL don't update just BB_END (current_block), but
4204         also call set_block_for_insn on the newly added insns and rescan.
4205
4206         PR target/84786
4207         * config/i386/sse.md (sse2_loadhpd): Use Yv constraint rather than v
4208         on the last operand.
4209
4210         PR c++/84704
4211         * tree.c (stabilize_reference_1): Return save_expr (e) for
4212         STATEMENT_LIST even if it doesn't have side-effects.
4213
4214 2018-03-12  Jonathan Wakely  <jwakely@redhat.com>
4215
4216         * doc/invoke.texi (-mclflushopt): Fix spelling of option.
4217
4218 2018-03-12  Renlin Li  <renlin.li@arm.com>
4219
4220         * config/aarch64/aarch64.md (movhf_aarch64): Fix mode argument to
4221         aarch64_output_scalar_simd_mov_immediate.
4222
4223 2018-03-12  Martin Sebor  <msebor@redhat.com>
4224
4225         PR tree-optimization/83456
4226         * gimple-fold.c (gimple_fold_builtin_memory_op): Avoid warning
4227         for perfectly overlapping calls to memcpy.
4228         (gimple_fold_builtin_memory_chk): Same.
4229         (gimple_fold_builtin_strcpy): Handle no-warning.
4230         (gimple_fold_builtin_stxcpy_chk): Same.
4231         * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Handle no-warning.
4232
4233 2018-03-12  Segher Boessenkool  <segher@kernel.crashing.org>
4234
4235         * config/rs6000/rs6000.c (abi_v4_pass_in_fpr): Add bool "named"
4236         parameter.  Use it for SFmode.
4237         (rs6000_function_arg_advance_1): Adjust.
4238         (rs6000_function_arg): Adjust.
4239         (rs6000_gimplify_va_arg): Pass false for that new parameter.
4240
4241 2018-03-12  Segher Boessenkool  <segher@kernel.crashing.org>
4242
4243         PR rtl-optimization/84169
4244         PR rtl-optimization/84780
4245         * combine.c (can_combine_p): Check for a 2-insn combination whether
4246         the destination register is used between the two insns, too.
4247
4248 2018-03-12  Richard Biener  <rguenther@suse.de>
4249
4250         PR tree-optimization/84803
4251         * tree-if-conv.c (ifcvt_memrefs_wont_trap): Don't do anything
4252         for refs DR analysis didn't process.
4253
4254 2018-03-12  Richard Biener  <rguenther@suse.de>
4255
4256         PR tree-optimization/84777
4257         * tree-ssa-loop-ch.c (should_duplicate_loop_header_p): For
4258         force-vectorize loops ignore whether we are optimizing for size.
4259
4260 2018-03-12  Chung-Ju Wu  <jasonwucj@gmail.com>
4261
4262         * config/nds32/nds32.c (nds32_md_asm_adjust): New function.
4263         (TARGET_MD_ASM_ADJUST): Define.
4264
4265 2018-03-12  Monk Chiang  <sh.chiang04@gmail.com>
4266             Kito Cheng  <kito.cheng@gmail.com>
4267             Chung-Ju Wu  <jasonwucj@gmail.com>
4268
4269         * config/nds32/nds32.c (nds32_compute_stack_frame,
4270         nds32_emit_stack_push_multiple, nds32_emit_stack_pop_multiple,
4271         nds32_emit_stack_v3push, nds32_emit_stack_v3pop,
4272         nds32_emit_adjust_frame, nds32_expand_prologue, nds32_expand_epilogue,
4273         nds32_expand_prologue_v3push, nds32_expand_epilogue_v3pop): Refine.
4274         * config/nds32/nds32.h (NDS32_FIRST_CALLEE_SAVE_GPR_REGNUM,
4275         NDS32_LAST_CALLEE_SAVE_GPR_REGNUM, NDS32_V3PUSH_AVAILABLE_P): New.
4276         * config/nds32/nds32.md (prologue, epilogue): Use macro
4277         NDS32_V3PUSH_AVAILABLE_P to do checking.
4278
4279 2018-03-11  Jakub Jelinek  <jakub@redhat.com>
4280
4281         PR debug/58150
4282         * dwarf2out.c (gen_enumeration_type_die): Don't guard adding
4283         DW_AT_declaration for ENUM_IS_OPAQUE on -gdwarf-4 or -gno-strict-dwarf,
4284         but on TYPE_SIZE.  Don't do anything for ENUM_IS_OPAQUE if not creating
4285         a new die.  Don't set TREE_ASM_WRITTEN if ENUM_IS_OPAQUE.  Guard
4286         addition of most attributes on !orig_type_die or the attribute not
4287         being present already.  Assert TYPE_VALUES is NULL for ENUM_IS_OPAQUE.
4288
4289 2018-03-11  Kito Cheng  <kito.cheng@gmail.com>
4290             Chung-Ju Wu  <jasonwucj@gmail.com>
4291
4292         * config/nds32/nds32.c (nds32_cpu_cpp_builtins): Modify to define
4293         __NDS32_VH__ macro.
4294         * config/nds32/nds32.opt (mvh): New option.
4295
4296 2018-03-11  Kito Cheng  <kito.cheng@gmail.com>
4297             Chung-Ju Wu  <jasonwucj@gmail.com>
4298
4299         * config/nds32/nds32-protos.h (nds32_cpu_cpp_builtins): Declare
4300         function.
4301         * config/nds32/nds32.c (nds32_cpu_cpp_builtins): New function.
4302         * config/nds32/nds32.h (TARGET_CPU_CPP_BUILTINS): Modify its
4303         definition.
4304
4305 2018-03-11  Kito Cheng  <kito.cheng@gmail.com>
4306             Chung-Ju Wu  <jasonwucj@gmail.com>
4307
4308         * config/nds32/nds32-memory-manipulation.c (nds32_expand_strlen): New
4309         function.
4310         * config/nds32/nds32-multiple.md (strlensi): New pattern.
4311         * config/nds32/nds32-protos.h (nds32_expand_strlen): Declare function.
4312
4313 2018-03-11  Monk Chiang  <sh.chiang04@gmail.com>
4314             Kito Cheng  <kito.cheng@gmail.com>
4315             Chung-Ju Wu  <jasonwucj@gmail.com>
4316
4317         * config/nds32/constants.md (unspec_element): Add UNSPEC_FFB,
4318         UNSPEC_FFMISM and UNSPEC_FLMISM.
4319         * config/nds32/nds32-intrinsic.c (bdesc_2arg): Add builtin description
4320         for ffb, ffmism and flmism.
4321         * config/nds32/nds32-intrinsic.md (unspec_ffb): Define new pattern.
4322         (unspec_ffmism): Ditto.
4323         (unspec_flmism): Ditto.
4324         (nds32_expand_builtin_impl): Check if string extension is available.
4325         * config/nds32/nds32.h (nds32_builtins): Add NDS32_BUILTIN_FFB,
4326         NDS32_BUILTIN_FFMISM and NDS32_BUILTIN_FLMISM.
4327
4328 2018-03-10  Vladimir Makarov  <vmakarov@redhat.com>
4329
4330         Reverting patch:
4331         2018-03-09  Vladimir Makarov  <vmakarov@redhat.com>
4332
4333         PR target/83712
4334         * lra-assigns.c (assign_by_spills): Return a flag of reload
4335         assignment failure.  Do not process the reload assignment
4336         failures.  Do not spill other reload pseudos if they has the same
4337         reg class.
4338         (lra_assign): Add a return arg.  Set up from the result of
4339         assign_by_spills call.
4340         (find_reload_regno_insns, lra_split_hard_reg_for): New functions.
4341         * lra-constraints.c (split_reg): Add a new arg.  Use it instead of
4342         usage_insns if it is not NULL.
4343         (spill_hard_reg_in_range): New function.
4344         (split_if_necessary, inherit_in_ebb): Pass a new arg to split_reg.
4345         * lra-int.h (spill_hard_reg_in_range, lra_split_hard_reg_for): New
4346         function prototypes.
4347         (lra_assign): Change prototype.
4348         * lra.c (lra): Add code to deal with fails by splitting hard reg
4349         live ranges.
4350
4351 2018-03-10  H.J. Lu  <hongjiu.lu@intel.com>
4352
4353         PR target/84807
4354         * config/i386/i386.opt: Replace Enforcment with Enforcement.
4355
4356 2018-03-10  Alexandre Oliva  <aoliva@redhat.com>
4357
4358         PR debug/84620
4359         * dwarf2out.h (dw_val_class): Add dw_val_class_symview.
4360         (dw_val_node): Add val_symbolic_view.
4361         * dwarf2out.c (dw_line_info_table): Add symviews_since_reset.
4362         (symview_upper_bound): New.
4363         (new_line_info_table): Initialize symviews_since_reset.
4364         (dwarf2out_source_line): Count symviews_since_reset and set
4365         symview_upper_bound.
4366         (dw_val_equal_p): Handle symview.
4367         (add_AT_symview): New.
4368         (print_dw_val): Handle symview.
4369         (attr_checksum, attr_checksum_ordered): Likewise.
4370         (same_dw_val_p, size_of_die): Likewise.
4371         (value_format, output_die): Likewise.
4372         (add_high_low_attributes): Use add_AT_symview for entry_view.
4373         (dwarf2out_finish): Reset symview_upper_bound, clear
4374         zero_view_p.
4375
4376 2018-03-09  Peter Bergner  <bergner@vnet.ibm.com>
4377
4378         PR target/83969
4379         * config/rs6000/rs6000.c (rs6000_offsettable_memref_p): New prototype.
4380         Add strict argument and use it.
4381         (rs6000_split_multireg_move): Update for new strict argument.
4382         (mem_operand_gpr): Disallow all non-offsettable addresses.
4383         * config/rs6000/rs6000.md (*movdi_internal64): Use YZ constraint.
4384
4385 2018-03-09  Jakub Jelinek  <jakub@redhat.com>
4386
4387         PR target/84772
4388         * config/rs6000/rs6000.c (rs6000_gimplify_va_arg): Mark va_arg_tmp
4389         temporary TREE_ADDRESSABLE before gimplification of BUILT_IN_MEMCPY.
4390         * config/powerpcspe/powerpcspe.c (rs6000_gimplify_va_arg): Likewise.
4391
4392         PR c++/84767
4393         * tree-inline.c (copy_tree_body_r): For INDIRECT_REF of a remapped
4394         decl, use remap_type if we want to use the type.
4395
4396 2018-03-09  Martin Sebor  <msebor@redhat.com>
4397
4398         PR tree-optimization/84526
4399         * gimple-ssa-warn-restrict.c (builtin_memref::set_base_and_offset):
4400         Remove dead code.
4401         (builtin_access::generic_overlap): Be prepared to handle non-array
4402         base objects.
4403
4404 2018-03-09  Alexandre Oliva  <aoliva@redhat.com>
4405
4406         PR rtl-optimization/84682
4407         * lra-constraints.c (process_address_1): Check is_address flag
4408         for address constraints.
4409         (process_alt_operands): Likewise.
4410         * lra.c (lra_set_insn_recog_data): Pass asm operand locs to
4411         preprocess_constraints.
4412         * recog.h (preprocess_constraints): Add oploc parameter.
4413         Adjust callers.
4414         * recog.c (preprocess_constraints): Test address_operand for
4415         CT_ADDRESS constraints.
4416
4417 2018-03-09  Vladimir Makarov  <vmakarov@redhat.com>
4418
4419         PR target/83712
4420         * lra-assigns.c (assign_by_spills): Return a flag of reload
4421         assignment failure.  Do not process the reload assignment
4422         failures.  Do not spill other reload pseudos if they has the same
4423         reg class.
4424         (lra_assign): Add a return arg.  Set up from the result of
4425         assign_by_spills call.
4426         (find_reload_regno_insns, lra_split_hard_reg_for): New functions.
4427         * lra-constraints.c (split_reg): Add a new arg.  Use it instead of
4428         usage_insns if it is not NULL.
4429         (spill_hard_reg_in_range): New function.
4430         (split_if_necessary, inherit_in_ebb): Pass a new arg to split_reg.
4431         * lra-int.h (spill_hard_reg_in_range, lra_split_hard_reg_for): New
4432         function prototypes.
4433         (lra_assign): Change prototype.
4434         * lra.c (lra): Add code to deal with fails by splitting hard reg
4435         live ranges.
4436
4437 2018-03-09  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
4438
4439         PR target/83193
4440         * common/config/arm/arm-common.c (arm_parse_arch_option_name):
4441         Accept complain bool parameter.  Only emit errors if it is true.
4442         (arm_parse_cpu_option_name): Likewise.
4443         (arm_target_thumb_only): Adjust callers of the above.
4444         * config/arm/arm-protos.h (arm_parse_cpu_option_name): Adjust
4445         prototype to take a default true bool parameter.
4446         (arm_parse_arch_option_name): Likewise.
4447
4448 2018-03-09  David Malcolm  <dmalcolm@redhat.com>
4449             Francois-Xavier Coudert  <fxcoudert@gcc.gnu.org>
4450
4451         PR jit/64089
4452         PR jit/84288
4453         * Makefile.in (LD_VERSION_SCRIPT_OPTION, LD_SONAME_OPTION): New.
4454         * configure: Regenerate.
4455         * configure.ac ("linker --version-script option"): New.
4456         ("linker soname option"): New.
4457
4458 2018-03-09  Richard Biener  <rguenther@suse.de>
4459
4460         PR tree-optimization/84775
4461         * tree-if-conv.c (add_bb_predicate_gimplified_stmts): Delink
4462         immediate uses of predicate stmts and mark them modified.
4463
4464         Revert
4465         PR tree-optimization/84178
4466         * tree-if-conv.c (combine_blocks): Move insert_gimplified_predicates
4467         to caller.
4468         (version_loop_for_if_conversion): Delay update_ssa call.
4469         (tree_if_conversion): Delay update_ssa until after predicate
4470         insertion.
4471
4472 2018-03-09  Eric Botcazou  <ebotcazou@adacore.com>
4473
4474         PR target/84763
4475         * config/i386/winnt.c (i386_pe_seh_cold_init): Use small pre-allocation
4476         when the function accesses prior frames.
4477
4478 2018-03-08  Jakub Jelinek  <jakub@redhat.com>
4479
4480         PR debug/84456
4481         * dwarf2out.c (dw_loc_list): If list && loc_list->first->next, call
4482         gen_llsym, otherwise call maybe_gen_llsym.
4483
4484         PR inline-asm/84742
4485         * recog.c (asm_operand_ok): Return 0 if multi-character constraint
4486         has ',' character inside of it.
4487
4488 2018-03-08  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
4489
4490         PR target/84748
4491         * config/aarch64/aarch64.md (*compare_cstore<mode>_insn): Mark pattern
4492         as clobbering CC_REGNUM.
4493
4494 2018-03-08  Richard Biener  <rguenther@suse.de>
4495
4496         PR middle-end/84552
4497         * tree-scalar-evolution.c: Include tree-into-ssa.h.
4498         (follow_copies_to_constant): Do not follow SSA names registered
4499         for update.
4500
4501 2018-03-08  Richard Biener  <rguenther@suse.de>
4502
4503         PR tree-optimization/84178
4504         * tree-if-conv.c (combine_blocks): Move insert_gimplified_predicates
4505         to caller.
4506         (version_loop_for_if_conversion): Delay update_ssa call.
4507         (tree_if_conversion): Delay update_ssa until after predicate
4508         insertion.
4509
4510 2018-03-08  David Malcolm  <dmalcolm@redhat.com>
4511
4512         PR tree-optimization/84178
4513         * tree-if-conv.c (release_bb_predicate): Remove the
4514         the assertion that the stmts have NULL use_ops.
4515         Discard the statements, asserting that they haven't
4516         yet been added to a BB.
4517
4518 2018-03-08  Richard Biener  <rguenther@suse.de>
4519
4520         PR tree-optimization/84746
4521         * tree-ssa-pre.c (find_leader_in_sets): Deal with SET1 being NULL.
4522         (phi_translate): Pass in destination ANTIC_OUT set.
4523         (phi_translate_1): Likewise.  For a simplified result lookup
4524         a leader in ANTIC_OUT and AVAIL_OUT, not the ANTIC_IN sets.
4525         (phi_translate_set): Adjust.
4526         (do_pre_regular_insertion): Likewise.
4527         (do_pre_partial_partial_insertion): Likewise.
4528
4529 2018-03-08  Martin Liska  <mliska@suse.cz>
4530
4531         PR gcov-profile/84735
4532         * doc/gcov.texi: Document usage of profile files.
4533         * gcov-io.h: Document changes in the format.
4534
4535 2018-03-08  Alexandre Oliva  <aoliva@redhat.com>
4536
4537         PR debug/84404
4538         PR debug/84408
4539         * dwarf2out.c (struct dw_line_info_table): Update comments for
4540         view == -1.
4541         (FORCE_RESET_NEXT_VIEW): New.
4542         (FORCE_RESETTING_VIEW_P): New.
4543         (RESETTING_VIEW_P): Check for -1 too.
4544         (ZERO_VIEW_P): Likewise.
4545         (new_line_info_table): Force-reset next view.
4546         (dwarf2out_begin_function): Likewise.
4547         (dwarf2out_source_line): Simplify zero_view_p initialization.
4548         Test FORCE_RESETTING_VIEW_P and RESETTING_VIEW_P instead of
4549         view directly.  Omit view when omitting .loc at line 0.
4550
4551 2018-03-08  Jakub Jelinek  <jakub@redhat.com>
4552
4553         PR tree-optimization/84740
4554         * tree-switch-conversion.c (process_switch): Call build_constructors
4555         only if info.phi_count is non-zero.
4556
4557         PR tree-optimization/84739
4558         * tree-tailcall.c (find_tail_calls): Check call arguments against
4559         DECL_ARGUMENTS (current_function_decl) rather than
4560         DECL_ARGUMENTS (func) when checking for tail recursion.
4561
4562 2018-03-07  Jakub Jelinek  <jakub@redhat.com>
4563
4564         * doc/contrib.texi: Add entries for Martin Liska, David Malcolm,
4565         Marek Polacek, extend Vladimir Makarov's, Jonathan Wakely's and
4566         Volker Reichelt's entry and add entries for people that perform
4567         GCC fuzzy testing and report numerous bugs.
4568
4569 2018-03-07  Segher Boessenkool  <segher@kernel.crashing.org>
4570
4571         PR target/82411
4572         * config/rs6000/rs6000.c (rs6000_elf_in_small_data_p): Don't put
4573         readonly data in sdata, if that is disabled.
4574         * config/rs6000/sysv4.opt (mreadonly-in-sdata): New option.
4575         * doc/invoke.texi (RS/6000 and PowerPC Options): Document
4576         -mreadonly-in-sdata option.
4577
4578 2018-03-07  Martin Sebor  <msebor@redhat.com>
4579
4580         PR tree-optimization/84468
4581         * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Consider successor
4582         basic block when looking for nul assignment.
4583
4584 2018-03-07  Eric Botcazou  <ebotcazou@adacore.com>
4585
4586         PR target/84277
4587         * except.h (output_function_exception_table): Adjust prototype.
4588         * except.c (output_function_exception_table): Remove FNNAME parameter
4589         and add SECTION parameter.  Ouput one part of the table at a time.
4590         * final.c (final_scan_insn_1) <NOTE_INSN_SWITCH_TEXT_SECTIONS>: Output
4591         the first part of the exception table and emit unwind directives.
4592         * config/i386/i386-protos.h (i386_pe_end_cold_function): Declare.
4593         (i386_pe_seh_cold_init): Likewise.
4594         * config/i386/cygming.h (ASM_DECLARE_COLD_FUNCTION_NAME): New macro.
4595         (ASM_DECLARE_COLD_FUNCTION_SIZE): Likewise.
4596         * config/i386/i386.c (x86_expand_epilogue): Fix wording in comment.
4597         (ix86_output_call_insn): Emit a nop in one more case for SEH.
4598         * config/i386/winnt.c: Include except.h.
4599         (struct seh_frame_state): Add reg_offset, after_prologue and
4600         in_cold_section fields.
4601         (i386_pe_seh_end_prologue): Set seh->after_prologue.
4602         (i386_pe_seh_cold_init): New function.
4603         (i386_pe_seh_fini): Add COLD parameter and bail out if it is not equal
4604         to seh->in_cold_section.
4605         (seh_emit_push): Record the offset of the push.
4606         (seh_emit_save): Record the offet of the save.
4607         (i386_pe_seh_unwind_emit): Deal with NOTE_INSN_SWITCH_TEXT_SECTIONS.
4608         Test seh->after_prologue to disregard the epilogue.
4609         (i386_pe_end_function): Pass FALSE to i386_pe_seh_fini.
4610         (i386_pe_end_cold_function): New function.
4611
4612 2018-03-07  Jakub Jelinek  <jakub@redhat.com>
4613
4614         PR fortran/84565
4615         * config/aarch64/predicates.md (aarch64_simd_reg_or_zero): Use
4616         aarch64_simd_or_scalar_imm_zero rather than aarch64_simd_imm_zero.
4617
4618         PR c++/84704
4619         * gimple-expr.c (create_tmp_var_raw): Set DECL_NAMELESS flag
4620         on tmp_var.
4621         * tree-pretty-print.c (dump_decl_name): For TDF_COMPARE_DEBUG,
4622         don't print names of DECL_NAMELESS DECL_IGNORED_P decls.
4623
4624         PR middle-end/84723
4625         * multiple_target.c: Include tree-inline.h and intl.h.
4626         (expand_target_clones): Diagnose and fail if node->definition and
4627         !tree_versionable_function_p (node->decl).
4628
4629 2018-03-06  John David Anglin  <danglin@gcc.gnu.org>
4630
4631         * config/pa/pa.h (ASM_GENERATE_INTERNAL_LABEL): Revise to use
4632         sprint_ul.
4633         (ASM_OUTPUT_ADDR_VEC_ELT): Revise for above change.
4634         (ASM_OUTPUT_ADDR_DIFF_ELT): Likewise.
4635         * config/pa/pa64-hpux.h (ASM_GENERATE_INTERNAL_LABEL): Revise as above.
4636
4637 2018-03-06  Jakub Jelinek  <jakub@redhat.com>
4638
4639         PR target/84710
4640         * combine.c (try_combine): Use reg_or_subregno instead of handling
4641         just paradoxical SUBREGs and REGs.
4642
4643 2018-03-06  Claudiu Zissulescu  <claziss@synopsys.com>
4644
4645          * config/arc/arc.c (arc_finalize_pic): Remove function.
4646          (arc_must_save_register): We use single base PIC register, remove
4647          checks to save/restore the PIC register.
4648          (arc_expand_prologue): Likewise.
4649          * config/arc/arc-protos.h (arc_set_default_type_attributes):
4650          Remove.
4651          (arc_verify_short): Likewise.
4652          (arc_attr_type): Likewise.
4653          * config/arc/arc.c (arc_set_default_type_attributes): Remove.
4654          (walk_stores): Likewise.
4655          (arc_address_cost): Make it static.
4656          (arc_verify_short): Likewise.
4657          (branch_dest): Likewise.
4658          (arc_attr_type): Likewise.
4659          * config/arc/arc.c (TARGET_ADJUST_INSN_LENGTH): Remove.
4660          (TARGET_INSN_LENGTH_PARAMETERS): Likewise.
4661          (arc_final_prescan_insn): Remove inserting the nops due to
4662          hardware hazards.  It is done in reorg step.
4663          (insn_length_variant_t): Remove.
4664          (insn_length_parameters_t): Likewise.
4665          (arc_insn_length_parameters): Likewise.
4666          (arc_get_insn_variants): Likewise.
4667          * config/arc/arc.h (TARGET_UPSIZE_DBR): Remove.
4668
4669 2018-03-06  Jakub Jelinek  <jakub@redhat.com>
4670
4671         PR inline-asm/84683
4672         * reg-stack.c (move_for_stack_reg): If any_malformed_asm, avoid
4673         assertion failure.
4674
4675         PR tree-optimization/84687
4676         * omp-simd-clone.c (simd_clone_create): Clear DECL_BUILT_IN_CLASS
4677         on new_node->decl.
4678         * match.pd (pow(C,x)*expN(y) -> expN(logN(C)*x+y)): New optimization.
4679
4680 2018-03-05  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
4681
4682         * config/rs6000/rs6000-builtin.def (rs6000_speculation_barrier):
4683         Rename to ppc_speculation_barrier.
4684         * config/rs6000/rs6000.c (rs6000_init_builtins): Rename builtin to
4685         __builtin_ppc_speculation_barrier.
4686
4687 2018-03-05  Jakub Jelinek  <jakub@redhat.com>
4688
4689         PR target/84700
4690         * combine.c (combine_simplify_rtx): Don't try to simplify if
4691         if_then_else_cond returned non-NULL, but either true_rtx or false_rtx
4692         are equal to x.
4693
4694 2018-03-05  Segher Boessenkool  <segher@kernel.crashing.org>
4695
4696         * config/rs6000/rs6000.c (rs6000_loop_align): Don't align tiny loops
4697         to 32 bytes when compiling for POWER9.
4698
4699 2018-03-05  Jakub Jelinek  <jakub@redhat.com>
4700
4701         PR target/84564
4702         * config/i386/i386.c (ix86_function_ok_for_sibcall): Check for
4703         regparm >= 3 with no arg reg available also for calls with
4704         flag_force_indirect_call.  Pass decl to ix86_function_regparm.
4705
4706         PR target/84524
4707         * config/i386/sse.md (*<code><mode>3): Replace <mask_prefix3> with
4708         orig,vex.
4709         (*<plusminus_insn><mode>3): Likewise.  Remove <mask_operand3> uses.
4710
4711 2018-03-05  Peter Bergner  <bergner@vnet.ibm.com>
4712
4713         PR target/84264
4714         * config/rs6000/vector.md (mov<mode>): Disallow altivec memory operands.
4715
4716 2018-03-05  Richard Biener  <rguenther@suse.de>
4717
4718         PR tree-optimization/84486
4719         * tree-ssa-pre.c (create_expression_by_pieces): Remove dead code.
4720         When inserting a __builtin_assume_aligned call set the LHS
4721         SSA name alignment info accordingly.
4722
4723 2018-03-05  Wilco Dijkstra  <wdijkstr@arm.com>
4724
4725         PR tree-optimization/84114
4726         * config/aarch64/aarch64.c (aarch64_reassociation_width)
4727         Avoid reassociation of FLOAT_MODE addition.
4728
4729 2018-03-05  Olga Makhotina  <olga.makhotina@intel.com>
4730
4731         * common/config/i386/i386-common.c (OPTION_MASK_ISA_PCONFIG_SET,
4732         OPTION_MASK_ISA_PCONFIG_UNSET, OPTION_MASK_ISA_WBNOINVD_SET,
4733         OPTION_MASK_ISA_WBNOINVD_UNSET): New definitions.
4734         (ix86_handle_option): Handle -mpconfig and -mwbnoinvd.
4735         * config.gcc (pconfigintrin.h, wbnoinvdintrin.h) : Add headers.
4736         * config/i386/cpuid.h (bit_PCONFIG, bit_WBNOINVD): New.
4737         * config/i386/driver-i386.c (host_detect_local_cpu): Detect -mpconfig
4738         and -mwbnoinvd.
4739         * config/i386/i386-builtin.def (__builtin_ia32_wbnoinvd,
4740         __builtin_ia32_wbinvd): New builtins.
4741         (SPECIAL_ARGS2): New.
4742         * config/i386/i386-c.c (__WBNOINVD__, __PCONFIG__): New.
4743         (SPECIAL_ARGS2): New.
4744         * config/i386/i386.c (ix86_target_string): Add -mpconfig and -mwbnoinvd.
4745         (ix86_valid_target_attribute_inner_p): Ditto.
4746         (ix86_init_mmx_sse_builtins): Add special_args2.
4747         * config/i386/i386.h (TARGET_PCONFIG, TARGET_PCONFIG_P, TARGET_WBNOINVD,
4748         TARGET_WBNOINVD_P): New.
4749         * config/i386/i386.md (UNSPECV_WBINVD, UNSPECV_WBNOINVD): New.
4750         (define_insn "wbinvd", define_insn "wbnoinvd"): New.
4751         * config/i386/i386.opt: Add -mpconfig and -mwbnoinvd.
4752         * config/i386/immintrin.h (_wbinvd): New intrinsic.
4753         * config/i386/pconfigintrin.h: New file.
4754         * config/i386/wbnoinvdintrin.h: Ditto.
4755         * config/i386/x86intrin.h: Add headers pconfigintrin.h and wbnoinvdintrin.h.
4756         * doc/invoke.texi (-mpconfig, -mwbnoinvd): New.
4757
4758 2018-03-05  Richard Biener  <rguenther@suse.de>
4759
4760         PR tree-optimization/84670
4761         * tree-ssa-pre.c (struct bb_bitmap_sets): Add visited_with_visited_succs
4762         member.
4763         (BB_VISITED_WITH_VISITED_SUCCS): New define.
4764         (compute_antic): Initialize BB_VISITED_WITH_VISITED_SUCCS.
4765         (compute_antic_aux): Only assert the number of values in ANTIC_IN
4766         doesn't grow if all successors (recursively) were visited at least
4767         once.
4768
4769 2018-03-05  Richard Biener  <rguenther@suse.de>
4770
4771         PR tree-optimization/84650
4772         * tree-ssa-loop-im.c (pass_lim::execute): Reset the SCEV cache
4773         if executed in the loop pipeline.
4774
4775 2018-03-05  Sandra Loosemore  <sandra@codesourcery.com>
4776
4777         * doc/configfiles.texi (Configuration Files): Move info about
4778         conditionalizing $target-protos.h to...
4779         * doc/sourcebuild.texi (Back End): Here.  Explain how $target.h
4780         differs from $target-protos.h.
4781
4782 2018-03-05  Kito Cheng  <kito.cheng@gmail.com>
4783             Chung-Ju Wu  <jasonwucj@gmail.com>
4784
4785         * config/nds32/nds32-protos.h (nds32_expand_setmem): Declare.
4786         * config/nds32/nds32-multiple.md (setmemsi): Define.
4787         * config/nds32/nds32-memory-manipulation.c
4788         (nds32_gen_dup_4_byte_to_word_value): New.
4789         (emit_setmem_word_loop): New.
4790         (emit_setmem_byte_loop): New.
4791         (nds32_expand_setmem_loop): New.
4792         (nds32_expand_setmem_loop_v3m): New.
4793         (nds32_expand_setmem_unroll): New.
4794         (nds32_expand_setmem): New.
4795
4796 2018-03-04  Kito Cheng  <kito.cheng@gmail.com>
4797             Chung-Ju Wu  <jasonwucj@gmail.com>
4798
4799         * config/nds32/nds32-memory-manipulation.c
4800         (nds32_emit_load_store): New.
4801         (nds32_emit_post_inc_load_store): New.
4802         (nds32_emit_mem_move): New.
4803         (nds32_emit_mem_move_block): New.
4804         (nds32_expand_movmemsi_loop_unknown_size): New.
4805         (nds32_expand_movmemsi_loop_known_size): New.
4806         (nds32_expand_movmemsi_loop): New.
4807         (nds32_expand_movmemsi_unroll): New.
4808         (nds32_expand_movmemqi): Rename ...
4809         (nds32_expand_movmemsi): ... to this.
4810         * config/nds32/nds32-multiple.md (movmemqi): Rename ...
4811         (movmemsi): ... to this.
4812         * config/nds32/nds32-protos.h (nds32_expand_movmemqi): Rename ...
4813         (nds32_expand_movmemsi): ... to this.
4814
4815 2018-03-04  Kito Cheng  <kito.cheng@gmail.com>
4816             Monk Chiang  <sh.chiang04@gmail.com>
4817             Chung-Ju Wu  <jasonwucj@gmail.com>
4818
4819         * config/nds32/nds32-protos.h
4820         (nds32_expand_load_multiple): New arguments.
4821         (nds32_expand_store_multiple): Ditto.
4822         (nds32_valid_multiple_load_store): Rename ...
4823         (nds32_valid_multiple_load_store_p): ... to this.
4824         * config/nds32/nds32-memory-manipulation.c
4825         (nds32_expand_load_multiple): Refine implementation.
4826         (nds32_expand_store_multiple): Ditto.
4827         * config/nds32/nds32-multiple.md
4828         (load_multiple): Update nds32_expand_load_multiple interface.
4829         (store_multiple): Update nds32_expand_store_multiple interface.
4830         * config/nds32/nds32-predicates.c
4831         (nds32_valid_multiple_load_store): Rename ...
4832         (nds32_valid_multiple_load_store_p): ... to this and refine
4833         implementation.
4834         * config/nds32/predicates.md
4835         (nds32_load_multiple_and_update_address_operation): New predicate.
4836         (nds32_store_multiple_and_update_address_operation): New predicate.
4837
4838 2018-03-04  Kito Cheng  <kito.cheng@gmail.com>
4839             Chung-Ju Wu  <jasonwucj@gmail.com>
4840
4841         * config/nds32/nds32.md (type): Add load_multiple and store_multiple.
4842         (combo): New attribute.
4843         * config/nds32/nds32-multiple.md: Refine patterns with new attributes.
4844
4845 2018-03-03  Chung-Ju Wu  <jasonwucj@gmail.com>
4846
4847         * config/nds32/nds32.opt: Change -mcmodel= default value.
4848
4849 2018-03-03  Kito Cheng  <kito.cheng@gmail.com>
4850             Monk Chiang  <sh.chiang04@gmail.com>
4851             Chung-Ju Wu  <jasonwucj@gmail.com>
4852
4853         * config/nds32/constants.md (unspec_element): New enum.
4854         * config/nds32/constraints.md (Umw): New constraint.
4855         * config/nds32/nds32-intrinsic.c: Add more builtin functions.
4856         * config/nds32/nds32-intrinsic.md: Likewise.
4857         * config/nds32/nds32-md-auxiliary.c (nds32_regno_to_enable4): New.
4858         (nds32_valid_smw_lwm_base_p): New.
4859         (nds32_output_smw_single_word): New.
4860         (nds32_output_lmw_single_word): New.
4861         (nds32_expand_unaligned_load): New.
4862         (nds32_expand_unaligned_store): New.
4863         * config/nds32/nds32-protos.h (nds32_valid_smw_lwm_base_p): Declare.
4864         (nds32_output_smw_single_word): Declare.
4865         (nds32_output_lmw_single_word): Declare.
4866         (nds32_expand_unaligned_load): Declare.
4867         (nds32_expand_unaligned_store): Declare.
4868         * config/nds32/nds32.h (nds32_builtins): Add NDS32_BUILTIN_UALOAD_HW,
4869         NDS32_BUILTIN_UALOAD_W, NDS32_BUILTIN_UALOAD_DW,
4870         NDS32_BUILTIN_UASTORE_HW, NDS32_BUILTIN_UASTORE_W,
4871         NDS32_BUILTIN_UASTORE_DW.
4872         * config/nds32/predicates.md (nds32_lmw_smw_base_operand): New
4873         predicate.
4874
4875 2018-03-03  Monk Chiang  <sh.chiang04@gmail.com>
4876             Kito Cheng  <kito.cheng@gmail.com>
4877             Chung-Ju Wu  <jasonwucj@gmail.com>
4878
4879         * config/nds32/nds32-intrinsic.c
4880         (nds32_expand_builtin_null_ftype_reg): Delete.
4881         (nds32_expand_builtin_reg_ftype_imm): Ditto.
4882         (nds32_expand_builtin_null_ftype_reg_imm): Ditto.
4883         (nds32_read_argument): New.
4884         (nds32_legitimize_target): Ditto.
4885         (nds32_legitimize_argument): Ditto.
4886         (nds32_check_constant_argument): Ditto.
4887         (nds32_expand_unop_builtin): Ditto.
4888         (nds32_expand_unopimm_builtin): Ditto.
4889         (nds32_expand_binop_builtin): Ditto.
4890         (nds32_builtin_decl_impl): Ditto.
4891         (builtin_description): Ditto.
4892         (nds32_expand_builtin_impl): Rewrite with new infrastructure.
4893         (nds32_init_builtins_impl): Ditto.
4894         * config/nds32/nds32.c (TARGET_BUILTIN_DECL): Define.
4895         (nds32_builtin_decl): New.
4896         * config/nds32/nds32.h (nds32_builtins): Add NDS32_BUILTIN_COUNT.
4897         * config/nds32/nds32-protos.h (nds32_builtin_decl_impl): Declare.
4898
4899 2018-03-02  Jeff Law  <law@redhat.com>
4900
4901         * reorg.c (stop_search_p): Handle DEBUG_INSN.
4902         (redundant_insn, fill_simple_delay_slots): Likewise.
4903         (fill_slots_from_thread): Likewise.
4904         * resource.c (mark_referenced_resources): Likewise.
4905         (mark_set_resources, find_dead_or_set_registers): Likewise.
4906
4907 2018-03-02  Jakub Jelinek  <jakub@redhat.com>
4908
4909         * substring-locations.h (format_warning_va): Formatting fix for
4910         ATTRIBUTE_GCC_DIAG.
4911         (format_warning_at_substring): Fix up ATTRIBUTE_GCC_DIAG second
4912         argument.
4913         (format_warning_n_va, format_warning_at_substring_n): New prototypes.
4914         * substring-locations.c: Include intl.h.
4915         (format_warning_va): Turned into small wrapper around
4916         format_warning_n_va, renamed to ...
4917         (format_warning_n_va): ... this, add N and PLURAL_GMSGID arguments,
4918         rename GMSGID to SINGULAR_GMSGID, if SINGULAR_GMSGID != PLURAL_GMSGID,
4919         use ngettext.
4920         (format_warning_at_substring_n): New function.
4921         * gimple-ssa-sprintf.c: Remove GCC diagnostic ignored pragma.
4922         (fmtwarn): Add ATTRIBUTE_GCC_DIAG.  Turn into a copy of
4923         format_warning_at_substring with just a shorter name instead of
4924         const function pointer.
4925         (fmtwarn_n): New function.
4926         (maybe_warn, format_directive, parse_directive): Use fmtwarn_n where
4927         appropriate, get rid of all the fmtstr temporaries, move conditionals
4928         with G_() wrapped string literals directly into fmtwarn arguments,
4929         cast dir.len to (int), formatting fixes.
4930
4931 2018-03-02  Thomas Schwinge  <thomas@codesourcery.com>
4932
4933         * doc/invoke.texi: Remove "Cilk Plus" references.
4934
4935 2018-03-02  Jakub Jelinek  <jakub@redhat.com>
4936             Richard Biener  <rguenther@suse.de>
4937
4938         PR ipa/84628
4939         * expr.c (expand_expr_real_1) <case CALL_EXPR>: Don't emit diagnostics
4940         for error or warning attributes if CALL_FROM_THUNK_P is set.
4941         Formatting fixes.
4942
4943 2018-03-02  Jakub Jelinek  <jakub@redhat.com>
4944
4945         PR target/56540
4946         * config/pa/pa.h (TARGET_CPU_CPP_BUILTINS): Predefine
4947         __SIZEOF_128__ macro if HPUX_LONG_DOUBLE_LIBRARY.
4948
4949         PR target/56540
4950         * config/ia64/ia64.h (TARGET_CPU_CPP_BUILTINS): Predefine
4951         __SIZEOF_{FPREG,FLOAT{80,128}}__ macros.
4952
4953         * predict.c (test_prediction_value_range): Use PROB_UNINITIALIZED
4954         instead of -1U in last predictors element's probability member.
4955
4956 2018-03-02  Eric Botcazou  <ebotcazou@adacore.com>
4957
4958         PR ipa/83983
4959         * ipa-devirt.c (odr_subtypes_equivalent_p): Get the ODR type of both
4960         arguments if they are comparable.
4961
4962 2018-03-02  Richard Sandiford  <richard.sandiford@linaro.org>
4963
4964         PR tree-optimization/84634
4965         * tree-vect-stmts.c (vectorizable_store, vectorizable_load): Replace
4966         masks and masked_loop_p with a single loop_masks, making sure it's
4967         null for bb vectorization.
4968
4969 2018-03-02  Richard Sandiford  <richard.sandiford@linaro.org>
4970
4971         * tree-vect-data-refs.c (vect_analyze_data_ref_dependence)
4972         (vect_analyze_data_ref_access): Use loop->safe_len rather than
4973         loop->force_vectorize to check whether there is no alias.
4974
4975 2018-03-02  Jakub Jelinek  <jakub@redhat.com>
4976
4977         PR target/84614
4978         * rtl.h (prev_real_nondebug_insn, next_real_nondebug_insn): New
4979         prototypes.
4980         * emit-rtl.c (next_real_insn, prev_real_insn): Fix up function
4981         comments.
4982         (next_real_nondebug_insn, prev_real_nondebug_insn): New functions.
4983         * cfgcleanup.c (try_head_merge_bb): Use prev_real_nondebug_insn
4984         instead of a loop around prev_real_insn.
4985         * combine.c (move_deaths): Use prev_real_nondebug_insn instead of
4986         prev_real_insn.
4987
4988         PR inline-asm/84625
4989         * config/i386/i386.c (ix86_print_operand): Use conditional
4990         output_operand_lossage instead of gcc_assert if CONST_VECTOR is not
4991         zero vector.
4992
4993 2018-03-02  Richard Biener  <rguenther@suse.de>
4994
4995         PR tree-optimization/84427
4996         * tree-ssa-pre.c (bitmap_remove_expr_from_set): Remove.
4997         (bitmap_set_subtract_values): Rewrite to handle multiple
4998         exprs per value.
4999         (clean): Likewise.
5000         (prune_clobbered_mems): Likewise.
5001         (phi_translate): Take edge instead of pred/phiblock.
5002         (phi_translate_1): Likewise.
5003         (phi_translate_set): Likewise.  Insert all translated
5004         exprs for a value into the set, keeping possibly multiple
5005         expressions per value.
5006         (compute_antic_aux): Adjust for phi_translate changes.
5007         When intersecting union the expressions and prune those
5008         not in the final value set, keeping possibly multiple
5009         expressions per value.  Do not use value-insertion
5010         for unioning ANTIC_OUT U EXP_GEN - TMP_GEN but merge
5011         all expressions.  Add verification that the value-sets
5012         only shrink during iteration.
5013         (compute_partial_antic_aux): Adjust for the phi_translate changes.
5014         (do_pre_regular_insertion): Likewise.
5015         (do_pre_partial_partial_insertion): Likewise.
5016
5017 2018-03-02  Richard Biener  <rguenther@suse.de>
5018
5019         PR target/82005
5020         * config/darwin.c (saved_debug_info_level): New static global.
5021         (darwin_asm_lto_start): Disable debug info generation for LTO out.
5022         (darwin_asm_lto_end): Restore debug info generation settings.
5023
5024 2018-03-01  Martin Liska  <mliska@suse.cz>
5025
5026         PR sanitizer/82484
5027         * sanopt.c (sanitize_rewrite_addressable_params): Do not handle
5028         volatile arguments.
5029
5030 2018-03-01  Richard Biener  <rguenther@suse.de>
5031
5032         PR debug/84645
5033         * dwarf2out.c (gen_variable_die): Properly handle late VLA
5034         type annotation with LTO when debug was disabled at compile-time.
5035
5036 2018-03-01  Matthew Fortune  <mfortune@gmail.com>
5037
5038         * config/mips/mips.c (mips_final_prescan_insn): Fix incorrect
5039         XINT with INTVAL.
5040         (mips_final_postscan_insn): Likewise.
5041
5042 2018-03-01  Richard Sandiford  <richard.sandiford@linaro.org>
5043
5044         PR rtl-optimization/84528
5045         * alias.c (init_alias_target): Add commentary.
5046         (init_alias_analysis): Only give HARD_FRAME_POINTER_REGNUM
5047         a unique base value if the frame pointer is not eliminated
5048         to the stack pointer.
5049
5050 2018-03-01  Tom de Vries  <tom@codesourcery.com>
5051
5052         PR rtl-optimization/83327
5053         * lra-int.h (hard_regs_spilled_into): Declare.
5054         * lra.c (hard_regs_spilled_into): Define.
5055         (init_reg_info): Init hard_regs_spilled_into.
5056         * lra-spills.c (assign_spill_hard_regs): Update hard_regs_spilled_into.
5057         * lra-lives.c (make_hard_regno_born, make_hard_regno_dead)
5058         (process_bb_lives): Handle hard_regs_spilled_into.
5059         (lra_create_live_ranges_1): Before doing liveness propagation, clear
5060         regs in all_hard_regs_bitmap if set in hard_regs_spilled_into.
5061
5062 2018-02-28  David Edelsohn  <dje.gcc@gmail.com>
5063
5064         * config.gcc (powerpc-ibm-aix7.1.*): New stanza.
5065         (powerpc-ibm-aix[789]*): Default to AIX 7.2.
5066         * config/rs6000/aix71.h (TARGET_DEFAULT): Revert to Power4 ISA.
5067         * config/rs6000/aix72.h: New file.
5068
5069 2018-02-28  Jakub Jelinek  <jakub@redhat.com>
5070
5071         * gimple-ssa-warn-restrict.c (maybe_diag_overlap): Use warning_n
5072         instead of warning_at with conditional singular and plural messages
5073         where possible.
5074
5075         PR target/52991
5076         * stor-layout.c (update_alignment_for_field): For
5077         targetm.ms_bitfield_layout_p (rli->t), if !is_bitfield
5078         && !DECL_PACKED (field), do the alignment update, just use
5079         only desired_align instead of MAX (type_align, desired_align)
5080         as the alignment.
5081         (place_field): Don't do known_align < desired_align handling
5082         early if targetm.ms_bitfield_layout_p (rli->t) and rli->prev_field
5083         is non-NULL, instead do it after rli->prev_field handling and
5084         only if not within a bitfield word.  For DECL_PACKED (field)
5085         use type_align of BITS_PER_UNIT.
5086
5087 2018-02-28  Eric Botcazou  <ebotcazou@adacore.com>
5088
5089         * config/aarch64/aarch64.c (aarch64_emit_probe_stack_range): Remove
5090         superfluous parentheses and trailing spaces.
5091
5092 2018-02-28  Richard Biener  <rguenther@suse.de>
5093
5094         PR tree-optimization/84584
5095         * graphite-scop-detection.c (scop_detection::add_scop): Discard
5096         SCoPs with fake exit edge.
5097
5098 2018-02-28  Martin Liska  <mliska@suse.cz>
5099
5100         PR testsuite/84597
5101         * timevar.c (timer::print): Fix format to properly print 100%
5102         values.
5103
5104 2018-02-28  Richard Biener  <rguenther@suse.de>
5105
5106         PR middle-end/84607
5107         * genmatch.c (capture_info::walk_match): Do not mark
5108         captured expressions without operands as expr_p given
5109         they act more like predicates and should be subject to
5110         "lost tail" side-effect preserving.
5111
5112 2018-02-28  Alexandre Oliva  <aoliva@redhat.com>
5113
5114         PR rtl-optimization/81611
5115         * auto-inc-dec.c (attempt_change): Move dead note from
5116         mem_insn if it's the next use of regno
5117         (find_address): Take address use of reg holding
5118         non-incremented value.  Add parm to limit search to the named
5119         reg only.
5120         (merge_in_block): Attempt to use a mem insn that is the next
5121         use of the original regno.
5122
5123 2018-02-27  Martin Sebor  <msebor@redhat.com>
5124
5125         PR c++/83871
5126         * gcc/doc/invoke.texi (-Wmissing-attributes): New option.
5127         * gcc/print-tree.c (print_node): Handle DECL_UNINLINABLE.
5128
5129 2018-02-27  Martin Sebor  <msebor@redhat.com>
5130
5131         PR translation/84207
5132         * diagnostic-core.h (warning_n, error_n, inform_n): Change
5133         n argument to unsigned HOST_WIDE_INT.
5134         * diagnostic.c (warning_n, error_n, inform_n): Ditto.
5135         (diagnostic_n_impl): Ditto.  Handle arguments in excess of LONG_MAX.
5136         * gimple-ssa-sprintf.c (format_directive): Simplify inform_n call.
5137         * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Use warning_n.
5138
5139 2018-02-27  Richard Biener  <rguenther@suse.de>
5140
5141         PR tree-optimization/84512
5142         * tree-vect-loop.c (vect_compute_single_scalar_iteration_cost):
5143         Do not use the estimate returned from record_stmt_cost for
5144         the scalar iteration cost but sum properly using add_stmt_cost.
5145
5146 2018-02-27  Richard Biener  <rguenther@suse.de>
5147
5148         PR tree-optimization/84466
5149         * graphite-scop-detection.c (scop_detection::stmt_simple_for_scop_p):
5150         Adjust last change to less strictly validate use operands.
5151
5152 2018-02-27  Martin Liska  <mliska@suse.cz>
5153
5154         PR gcov-profile/84548
5155         * gcov.c (process_file): Allow partial overlap and consider it
5156         also as group functions.
5157         (output_lines): Properly calculate range of lines for a group.
5158
5159 2018-02-27  Martin Liska  <mliska@suse.cz>
5160
5161         * timevar.c (timer::print_row): Remove 'usr', 'sys', 'wall' and
5162         'ggc' suffixes.  Change first column width.
5163         (timer::print): Fix formatting of the column.
5164
5165 2018-02-27  Alexandre Oliva  <aoliva@redhat.com>
5166
5167         * tree-ssa-live.c (remove_unused_scope_block_p): Do not
5168         preserve inline entry blocks for the sake of debug inline
5169         entry point markers alone.
5170         (remove_unused_locals): Suggest in comments a better place to
5171         force the preservation of inline entry blocks that are
5172         otherwise unused, but do not preserve them.
5173
5174 2018-02-26  H.J. Lu  <hongjiu.lu@intel.com>
5175
5176         * config/i386/i386.c (ix86_output_indirect_jmp): Update comments.
5177
5178 2018-02-26  H.J. Lu  <hongjiu.lu@intel.com>
5179
5180         PR target/84039
5181         * config/i386/constraints.md (Bs): Replace
5182         ix86_indirect_branch_register with
5183         TARGET_INDIRECT_BRANCH_REGISTER.
5184         (Bw): Likewise.
5185         * config/i386/i386.md (indirect_jump): Likewise.
5186         (tablejump): Likewise.
5187         (*sibcall_memory): Likewise.
5188         (*sibcall_value_memory): Likewise.
5189         Peepholes of indirect call and jump via memory: Likewise.
5190         (*sibcall_GOT_32): Disallowed for TARGET_INDIRECT_BRANCH_REGISTER.
5191         (*sibcall_value_GOT_32): Likewise.
5192         * config/i386/predicates.md (indirect_branch_operand): Likewise.
5193         (GOT_memory_operand): Likewise.
5194         (call_insn_operand): Likewise.
5195         (sibcall_insn_operand): Likewise.
5196         (GOT32_symbol_operand): Likewise.
5197         * config/i386/i386.h (TARGET_INDIRECT_BRANCH_REGISTER): New.
5198
5199 2018-02-26  Eric Botcazou  <ebotcazou@adacore.com>
5200
5201         PR rtl-optimization/83496
5202         * reorg.c (steal_delay_list_from_target): Change REDUNDANT array from
5203         booleans to RTXes.  Call fix_reg_dead_note on every non-null element.
5204         (steal_delay_list_from_fallthrough): Call fix_reg_dead_note on a
5205         redundant insn, if any.
5206         (relax_delay_slots): Likewise.
5207         (update_reg_unused_notes): Rename REDUNDANT_INSN to OTHER_INSN.
5208
5209 2018-02-26  Richard Sandiford  <richard.sandiford@linaro.org>
5210
5211         PR tree-optimization/83965
5212         * tree-vect-patterns.c (vect_reassociating_reduction_p): Assume
5213         that grouped statements are part of a reduction chain.  Return
5214         true if the statement is not marked as a reduction itself but
5215         is part of a group.
5216         (vect_recog_dot_prod_pattern): Don't check whether the statement
5217         is part of a group here.
5218         (vect_recog_sad_pattern): Likewise.
5219         (vect_recog_widen_sum_pattern): Likewise.
5220
5221 2018-02-26  Eric Botcazou  <ebotcazou@adacore.com>
5222
5223         PR debug/84545
5224         * final.c (rest_of_clean_state): Also look for calls inside sequences.
5225
5226 2018-02-26  H.J. Lu  <hongjiu.lu@intel.com>
5227
5228         PR target/84530
5229         * config/i386/i386-protos.h (ix86_output_indirect_jmp): Remove
5230         the bool argument.
5231         (ix86_output_indirect_function_return): New prototype.
5232         (ix86_split_simple_return_pop_internal): Likewise.
5233         * config/i386/i386.c (indirect_return_via_cx): New.
5234         (indirect_return_via_cx_bnd): Likewise.
5235         (indirect_thunk_name): Handle return va CX_REG.
5236         (output_indirect_thunk_function): Create alias for
5237         __x86_return_thunk_[re]cx and __x86_return_thunk_[re]cx_bnd.
5238         (ix86_output_indirect_jmp): Remove the bool argument.
5239         (ix86_output_indirect_function_return): New function.
5240         (ix86_split_simple_return_pop_internal): Likewise.
5241         * config/i386/i386.md (*indirect_jump): Don't pass false
5242         to ix86_output_indirect_jmp.
5243         (*tablejump_1): Likewise.
5244         (simple_return_pop_internal): Change it to define_insn_and_split.
5245         Call ix86_split_simple_return_pop_internal to split it for
5246         -mfunction-return=.
5247         (simple_return_indirect_internal): Call
5248         ix86_output_indirect_function_return instead of
5249         ix86_output_indirect_jmp.
5250
5251 2018-02-26  Jakub Jelinek  <jakub@redhat.com>
5252
5253         PR bootstrap/84405
5254         * vec.h (vec_default_construct): For BROKEN_VALUE_INITIALIZATION use
5255         memset and value initialization afterwards.
5256
5257 2018-02-26  Christophe Lyon  <christophe.lyon@linaro.org>
5258
5259         * Makefile.in (lto-wrapper): Use ALL_LINKERFLAGS.
5260
5261 2018-02-26  Ramana Radhakrishnan  <ramana.radhakrishnan@arm.com>
5262
5263         PR target/84521
5264         * common/config/aarch64/aarch64-common.c
5265         (aarch_option_optimization_table[]): Switch
5266         off fomit-frame-pointer
5267
5268 2018-02-26  Kito Cheng  <kito.cheng@gmail.com>
5269             Chung-Ju Wu  <jasonwucj@gmail.com>
5270
5271         * config/nds32/nds32-multiple.md (load_multiple): Disallow
5272         volatile memory.
5273         (store_multiple): Ditto.
5274
5275 2018-02-26  Kito Cheng  <kito.cheng@gmail.com>
5276
5277         * config.gcc: Add --with-cpu support for nds32 target.
5278         * config/nds32/nds32-opts.h (nds32_cpu_type): New.
5279         * config/nds32/nds32.opt: Add -mcpu= option.
5280
5281 2018-02-25  Segher Boessenkool  <segher@kernel.crashing.org>
5282
5283         * config/rs6000/rs6000.opt (mvrsave=no, mvrsave=yes, isel=no,
5284         isel=yes): Warn for these deprecated options.
5285
5286 2018-02-23  David Edelsohn  <dje.gcc@gmail.com>
5287
5288         * config/rs6000/aix71.h (TARGET_DEFAULT): Change to
5289         ISA_2_5_MASKS_EMBEDDED.
5290
5291 2018-02-23  Jakub Jelinek  <jakub@redhat.com>
5292
5293         * ipa-prop.c (ipa_vr_ggc_hash_traits::hash): Hash p->min and
5294         p->max as pointers rather than using iterative_hash_expr.
5295
5296 2018-02-23  Carl Love  <cel@us.ibm.com>
5297
5298         * config/rs6000/rs6000-builtin.def: Change VSIGNED2 and VUNSIGNED2
5299         macro expansions from BU_VSX_2 to BU_P8V_VSX_2 and BU_VSX_OVERLOAD_2 to
5300         BU_P8V_OVERLOAD_2.
5301         * config/rs6000/rs6000-c.c: Change VSX_BUILTIN_VEC_VSIGNED2 to
5302         P8V_BUILTIN_VEC_VSIGNED2.  Change VSX_BUILTIN_VEC_VUNSIGNED2 to
5303         P8V_BUILTIN_VEC_VUNSIGNED2.
5304
5305 2018-02-22  Vladimir Makarov  <vmakarov@redhat.com>
5306
5307         PR target/81572
5308         * lra-int.h (LRA_UNKNOWN_ALT, LRA_NON_CLOBBERED_ALT): New macros.
5309         * lra.c (lra_set_insn_recog_data, lra_update_insn_recog_data): Use
5310         LRA_UNKNOWN_ALT.
5311         * lra-constraints.c (curr_insn_transform): Set up
5312         LRA_NON_CLOBBERED_ALT for moves processed on the fast path.  Use
5313         LRA_UNKNOWN_ALT.
5314         (remove_inheritance_pseudos): Use LRA_UNKNOWN_ALT.
5315         * lra-eliminations.c (spill_pseudos): Ditto.
5316         (process_insn_for_elimination): Ditto.
5317         * lra-lives.c (reg_early_clobber_p): Use the new macros.
5318         * lra-spills.c (spill_pseudos): Use LRA_UNKNOWN_ALT and
5319         LRA_NON_CLOBBERED_ALT.
5320
5321 2018-02-22  Martin Sebor  <msebor@redhat.com>
5322
5323         PR tree-optimization/84480
5324         * gimple-fold.c (gimple_fold_builtin_strcpy): Move warnings
5325         to maybe_diag_stxncpy_trunc.  Call it.
5326         * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Integrate warnings
5327         from gimple_fold_builtin_strcpy.  Print inlining stack.
5328         (handle_builtin_stxncpy): Print inlining stack.
5329         * tree-ssa-strlen.h (maybe_diag_stxncpy_trunc): Declare.
5330
5331 2018-02-22  H.J. Lu  <hongjiu.lu@intel.com>
5332
5333         PR target/84176
5334         * config/i386/i386.c (ix86_set_indirect_branch_type): Issue an
5335         error when -mindirect-branch=thunk-extern, -fcf-protection=branch
5336         and -fcheck-pointer-bounds are used together.
5337         (indirect_thunk_prefix): New enum.
5338         (indirect_thunk_need_prefix): New function.
5339         (indirect_thunk_name): Replace need_bnd_p with need_prefix.  Use
5340         "_nt" instead of "_bnd" for NOTRACK prefix.
5341         (output_indirect_thunk): Replace need_bnd_p with need_prefix.
5342         (output_indirect_thunk_function): Likewise.
5343         (): Likewise.
5344         (ix86_code_end): Update output_indirect_thunk_function calls.
5345         (ix86_output_indirect_branch_via_reg): Replace
5346         ix86_bnd_prefixed_insn_p with indirect_thunk_need_prefix.
5347         (ix86_output_indirect_branch_via_push): Likewise.
5348         (ix86_output_function_return): Likewise.
5349         * doc/invoke.texi: Document -mindirect-branch=thunk-extern is
5350         incompatible with -fcf-protection=branch and
5351         -fcheck-pointer-bounds.
5352
5353 2018-02-22  Steve Ellcey  <sellcey@cavium.com>
5354
5355         PR target/83335
5356         * config/aarch64/aarch64.c (aarch64_print_address_internal):
5357         Change gcc_assert call to output_operand_lossage.
5358
5359 2018-02-22  Steve Ellcey  <sellcey@cavium.com>
5360
5361         * doc/extend.texi (__builtin_extend_pointer): Document builtin.
5362
5363 2018-02-22  DJ Delorie  <dj@redhat.com>
5364             Sebastian Perta  <sebastian.perta@renesas.com>
5365             Oleg Endo  <olegendo@gcc.gnu.org>
5366
5367         * config/rx/rx.c (rx_rtx_costs): New function.
5368         (TARGET_RTX_COSTS): Override to use rx_rtx_costs.
5369
5370 2018-02-22  Thomas Preud'homme  <thomas.preudhomme@arm.com>
5371
5372         * config/arm/t-multilib: Map Armv8-R to Armv7 multilibs.
5373
5374 2018-02-22  Martin Liska  <mliska@suse.cz>
5375
5376         PR driver/83193
5377         * common/config/arm/arm-common.c (arm_print_hint_for_cpu_option):
5378         Add "native" as a possible value.
5379
5380 2018-02-22  Martin Liska  <mliska@suse.cz>
5381
5382         PR driver/83193
5383         * config/i386/i386.c (ix86_option_override_internal):
5384         Add "native" as a possible value for -march and -mtune.
5385
5386 2018-02-22  Jakub Jelinek  <jakub@redhat.com>
5387
5388         PR target/84502
5389         * stor-layout.c (finalize_type_size): Propagate TYPE_EMPTY_P flag
5390         to all type variants.
5391
5392         PR tree-optimization/84503
5393         * gimple-ssa-store-merging.c (merged_store_group::merge_into): Compute
5394         width as info->bitpos + info->bitsize - start.
5395         (merged_store_group::merge_overlapping): Simplify width computation.
5396         (check_no_overlap): New function.
5397         (imm_store_chain_info::try_coalesce_bswap): Compute expected
5398         start + width and last_order of the group, fail if check_no_overlap
5399         fails.
5400         (imm_store_chain_info::coalesce_immediate_stores): Don't merge info
5401         to group if check_no_overlap fails.
5402
5403 2018-02-21  Segher Boessenkool  <segher@kernel.crashing.org>
5404
5405         * config/rs6000/altivec.md: Delete contraint arguments to
5406         define_expand, define_split, and define_peephole2, and in
5407         define_insn_and_split if always unused.
5408         * config/rs6000/darwin.md: Ditto.
5409         * config/rs6000/dfp.md: Ditto.
5410         * config/rs6000/rs6000.md: Ditto.
5411         * config/rs6000/sync.md: Ditto.
5412         * config/rs6000/vector.md: Ditto.
5413         * config/rs6000/vsx.md: Ditto.
5414
5415 2018-02-21  Segher Boessenkool  <segher@kernel.crashing.org>
5416
5417         * config/rs6000/altivec.md: Write output control strings as braced
5418         blocks instead of double-quoted strings.
5419         * config/rs6000/darwin.md: Ditto.
5420         * config/rs6000/rs6000.md: Ditto.
5421         * config/rs6000/vector.md: Ditto.
5422         * config/rs6000/vsx.md: Ditto.
5423
5424 2018-02-21  Jason Merrill  <jason@redhat.com>
5425
5426         PR c++/84314 - ICE with templates and fastcall attribute.
5427         * attribs.c (build_type_attribute_qual_variant): Remove assert.
5428
5429 2018-02-21  Jan Hubicka  <hubicka@ucw.cz>
5430
5431         * ipa-cp.c (determine_versionability): Fix comment typos.
5432
5433 2018-02-21  Jan Hubicka  <hubicka@ucw.cz>
5434
5435         PR c/84229
5436         * ipa-cp.c (determine_versionability): Do not version functions caling
5437         va_arg_pack.
5438
5439 2018-02-21  Martin Liska  <mliska@suse.cz>
5440
5441         PR driver/83193
5442         * config/aarch64/aarch64.c (aarch64_print_hint_for_core_or_arch):
5443         Add "native" as a possible value.
5444         * config/aarch64/aarch64.h (HAVE_LOCAL_CPU_DETECT):  Define
5445         the macro when native cpu detection is available.
5446
5447 2018-02-21  Martin Liska  <mliska@suse.cz>
5448
5449         PR driver/83193
5450         * common/config/arm/arm-common.c (arm_print_hint_for_arch_option):
5451         Add "native" as a possible value.
5452         * config/arm/arm.h (HAVE_LOCAL_CPU_DETECT): Define the macro
5453         when native cpu detection is available.
5454
5455 2018-02-21  Jakub Jelinek  <jakub@redhat.com>
5456             Martin Sebor  <msebor@redhat.com>
5457
5458         PR tree-optimization/84478
5459         * gimple-fold.h (get_range_strlen): Add a bool argument defaulted to
5460         false.
5461         * gimple-fold.c (get_range_strlen): Make minlen const and assume it
5462         can't be NULL.  Change FUZZY from bool to int, for 1 add PHI/COND_EXPR
5463         support which is conservatively correct, for 2 only stay conservative
5464         for maxlen.  Formatting and comment capitalization fixes.  Add STRICT
5465         argument to the 2 argument get_range_strlen, adjust 6 arg
5466         get_range_strlen caller and clear minmaxlen[0] and [1] if it returned
5467         false.
5468         (get_maxval_strlen): Adjust 6 arg get_range_strlen caller.
5469         (gimple_fold_builtin_strlen): Pass true as last argument to
5470         get_range_strlen.
5471
5472 2018-02-20  Martin Sebor  <msebor@redhat.com>
5473
5474         PR middle-end/84095
5475         * gimple-ssa-warn-restrict.c (builtin_memref::extend_offset_range): New.
5476         (builtin_memref::set_base_and_offset): Same.  Handle inner references.
5477         (builtin_memref::builtin_memref): Factor out parts into
5478         set_base_and_offset and call it.
5479
5480 2018-02-20  Richard Sandiford  <richard.sandiford@linaro.org>
5481
5482         PR middle-end/84406
5483         * optabs-query.c (find_widening_optab_handler_and_mode): If from_mode
5484         is a scalar_int_mode, assert that to_mode is a scalar_int_mode with
5485         greater precision.  If to_mode is a MODE_PARTIAL_INT, stop the
5486         search at the associated MODE_INT.
5487
5488 2018-02-20  Jeff Law  <law@redhat.com>
5489
5490         PR middle-end/82123
5491         PR tree-optimization/81592
5492         PR middle-end/79257
5493         * gimple-ssa-sprintf.c (format_integer): Query EVRP range analyzer
5494         for range data rather than using global data.
5495         * gimple-ssa-sprintf.c (get_int_range): Query EVRP range analyzer for
5496         range data rather than using global data.
5497         * gimple-ssa-sprintf.c (get_int_range): Accept vr_values parameter
5498         pass it to children as needed.
5499         (struct directive::fmtresult): Similarly.
5500         (struct directive::set_width): Similarly.
5501         (struct directive::set_precision): Similarly.
5502         (format_integer, format_directive, parse_directive): Similarly.
5503         (format_none): Accept unnamed vr_values parameter.
5504         (format_percent, format_floating, format_character): Similarly.
5505         (format_string, format_plain): Similarly.
5506         * gimple-ssa-sprintf.c (sprintf_dom_walker::handle_gimple_call): Query
5507         the EVRP range analyzer for range data rather than using global data.
5508         * gimple-ssa-sprintf.c: Include alloc-pool.h, vr-values.h and
5509         gimple-ssa-evrp-analyze.h
5510         (class sprintf_dom_walker): Add after_dom_children member function.
5511         Add evrp_range_analyzer member.
5512         (sprintf_dom_walker::before_dom_children): Call into the EVRP
5513         range analyzer as needed.
5514         (sprintf_dom_walker::after_dom_children): New member function.
5515         * gimple-ssa-evrp-analyze.c (evrp_range_analyzer::enter): Do nothing
5516         if not optimizing.
5517         (evrp_range_analyzer::record_ranges_from_stmt): Likewise.
5518         (evrp_range_analyzer::pop_to_marker): Likewise.
5519
5520 2018-02-20  Richard Sandiford  <richard.sandiford@linaro.org>
5521
5522         PR tree-optimization/84419
5523         * internal-fn.c (expand_call_mem_ref): Create a TARGET_MEM_REF
5524         with the required type if its current type is compatible but
5525         different.
5526
5527 2018-02-20  Jakub Jelinek  <jakub@redhat.com>
5528
5529         PR middle-end/82004
5530         * match.pd (pow(C,x) -> exp(log(C)*x)): Delay all folding until
5531         after vectorization.
5532
5533 2018-02-20  Martin Liska  <mliska@suse.cz>
5534
5535         PR driver/83193
5536         * config/aarch64/aarch64.c (aarch64_print_hint_for_core_or_arch): Print
5537         possible values if we don't have a hint.
5538
5539 2018-02-20  Martin Liska  <mliska@suse.cz>
5540
5541         PR c/84310
5542         PR target/79747
5543         * final.c (shorten_branches): Build align_tab array with one
5544         more element.
5545         * opts.c (finish_options): Add alignment option limit check.
5546         (MAX_CODE_ALIGN): Likewise.
5547         (MAX_CODE_ALIGN_VALUE): Likewise.
5548         * doc/invoke.texi: Document maximum allowed option value for
5549         all -falign-* options.
5550
5551 2018-02-19  Jakub Jelinek  <jakub@redhat.com>
5552
5553         PR target/84146
5554         * reg-notes.def (REG_CALL_ARG_LOCATION): New reg note.
5555         * insn-notes.def (NOTE_INSN_CALL_ARG_LOCATION): Remove.
5556         * var-tracking.c (emit_note_insn_var_location): Remove all references
5557         to NOTE_INSN_CALL_ARG_LOCATION.
5558         (emit_notes_in_bb): Emit arguments as REG_CALL_ARG_LOCATION note on
5559         the CALL_INSN rather than separate NOTE_INSN_CALL_ARG_LOCATION note.
5560         Use copy_rtx_if_shared.
5561         * dwarf2out.c (gen_subprogram_die): Use XEXP with 0 instead of
5562         NOTE_VAR_LOCATION on ca_loc->call_arg_loc_note.
5563         (dwarf2out_var_location): Remove handling of
5564         NOTE_INSN_CALL_ARG_LOCATION, instead handle REG_CALL_ARG_LOCATION note
5565         on call_insn.
5566         * final.c (final_scan_insn): Remove all references to
5567         NOTE_INSN_CALL_ARG_LOCATION.
5568         (rest_of_clean_state): Likewise.  Remove REG_CALL_ARG_LOCATION notes
5569         before dumping final insns.
5570         * except.c (emit_note_eh_region_end): Remove all references to
5571         NOTE_INSN_CALL_ARG_LOCATION.
5572         * config/alpha/alpha.c (alpha_pad_function_end): Likewise.
5573         * config/c6x/c6x.c (c6x_gen_bundles): Likewise.
5574         * config/arc/arc.c (hwloop_optimize): Likewise.
5575         * config/arm/arm.c (create_fix_barrier): Likewise.
5576         * config/s390/s390.c (s390_chunkify_start): Likewise.
5577         * config/sh/sh.c (find_barrier): Likewise.
5578         * config/i386/i386.c (rest_of_insert_endbranch,
5579         ix86_seh_fixup_eh_fallthru): Likewise.
5580         * config/xtensa/xtensa.c (hwloop_optimize): Likewise.
5581         * config/iq2000/iq2000.c (final_prescan_insn): Likewise.
5582         * config/frv/frv.c (frv_function_prologue): Likewise.
5583         * emit-rtl.c (try_split): Likewise.  Copy over REG_CALL_ARG_LOCATION
5584         reg note.
5585         (note_outside_basic_block_p): Remove all references to
5586         NOTE_INSN_CALL_ARG_LOCATION.
5587         * gengtype.c (adjust_field_rtx_def): Likewise.
5588         * print-rtl.c (rtx_writer::print_rtx_operand_code_0, print_insn):
5589         Likewise.
5590         * jump.c (cleanup_barriers, delete_related_insns): Likewise.
5591         * cfgrtl.c (force_nonfallthru_and_redirect): Likewise.
5592
5593         PR c++/84444
5594         * builtins.c (builtin_mathfn_code): Don't check if CALL_EXPR_FN (t)
5595         is ADDR_EXPR.
5596
5597         PR tree-optimization/84452
5598         * tree-vect-patterns.c (vect_recog_pow_pattern): Don't call
5599         expand_simd_clones if targetm.simd_clone.compute_vecsize_and_simdlen
5600         is NULL.
5601
5602 2018-02-19  Martin Liska  <mliska@suse.cz>
5603
5604         PR sanitizer/82183
5605         * passes.def: Put pass_sancov_O0 before pass_lower_switch with -O0.
5606
5607 2018-02-19  Martin Liska  <mliska@suse.cz>
5608             Richard Sandiford  <richard.sandiford@linaro.org>
5609
5610         PR tree-optimization/82491
5611         * gimple-fold.c (get_base_constructor): Make earlier bail out
5612         to prevent ubsan.
5613
5614 2018-02-19  Carl Love  <cel@us.ibm.com>
5615
5616         * config/rs6000/rs6000-builtin.def: Change NEG macro expansions from
5617         BU_ALTIVEC_A to BU_P8V_AV_1 and BU_ALTIVEC_OVERLOAD_1 to
5618         BU_P8V_OVERLOAD_1.
5619         * config/rs6000/rs6000-c.c: Change ALTIVEC_BUILTIN_VEC_NEG to
5620         P8V_BUILTIN_VEC_NEG.
5621
5622 2018-02-19  Sebastian Perta  <sebastian.perta@renesas.com>
5623
5624         * config/rl78/rl78.md (movdf): New define expand.
5625
5626 2018-02-19  Martin Liska  <mliska@suse.cz>
5627
5628         PR other/80589
5629         * doc/invoke.texi: Fix typo.
5630         * params.def (PARAM_MAX_LOOP_HEADER_INSNS): Likewise.
5631
5632 2018-02-18  Segher Boessenkool  <segher@kernel.crashing.org>
5633
5634         * config/rs6000/rs6000.c (rs6000_option_override_internal): Don't
5635         handle rs6000_single_float and rs6000_double_float specially for
5636         e500 family CPUs.
5637
5638 2018-02-16  Jeff Law  <law@redhat.com>
5639
5640         * config/rx/rx.c (add_pop_cfi_notes): New function.;
5641         (pop_regs): Use it.
5642
5643 2018-02-16  Jakub Jelinek  <jakub@redhat.com>
5644
5645         PR ipa/84425
5646         * ipa-inline.c (inline_small_functions): Fix a typo.
5647
5648 2018-02-16  Nathan Sidwell  <nathan@acm.org>
5649
5650         * doc/extend.texi (Backwards Compatibility): Americanize 'behaviour'.
5651
5652 2018-02-16  Carl Love  <cel@us.ibm.com>
5653
5654         * config/rs6000/rs6000-builtin.def: Add BU_P8V_VSX_2 macro definition.
5655         Change FLOAT2 expansions from BU_VSX_2 to BU_P8V_VSX_2 and
5656         from BU_VSX_OVERLOAD_2 to BU_P8V_OVERLOAD_2.
5657         * config/rs6000/rs6000-c.c: Changed macro VSX_BUILTIN_VEC_FLOAT2
5658         expansion to P8V_BUILTIN_VEC_FLOAT2.
5659
5660 2018-02-16  Vladimir Makarov  <vmakarov@redhat.com>
5661
5662         PR rtl-optimization/70023
5663         * lra-constraints.c (inherit_in_ebb): Take hard reg mode of
5664         src_regno into account.
5665
5666 2018-02-16  Carl Love  <cel@us.ibm.com>
5667
5668         * config/rs6000/altivec.h: Remove vec_vextract4b and vec_vinsert4b.
5669         * config/rs6000/rs6000-builtin.def: Remove macro expansion for
5670         VEXTRACT4B, VINSERT4B, VINSERT4B_DI and VEXTRACT4B.
5671         * config/rs6000/rs6000.c: Remove case statements for
5672         P9V_BUILTIN_VEXTRACT4B, P9V_BUILTIN_VEC_VEXTRACT4B,
5673         P9V_BUILTIN_VINSERT4B, P9V_BUILTIN_VINSERT4B_DI,
5674         and P9V_BUILTIN_VEC_VINSERT4B.
5675         * config/rs6000/rs6000-c.c (altivec_expand_builtin): Remove entries for
5676         P9V_BUILTIN_VEC_VEXTRACT4B and P9V_BUILTIN_VEC_VINSERT4B.
5677         * config/rs6000/vsx.md:
5678         * doc/extend.texi: Remove vec_vextract4b, non ABI definitions for
5679         vec_insert4b.
5680
5681 2018-02-16  Carl Love  <cel@us.ibm.com>
5682
5683         * config/rs6000/altivec.h: Add builtin names vec_extract4b
5684         vec_insert4b.
5685         * config/rs6000/rs6000-builtin.def: Add INSERT4B and EXTRACT4B
5686         definitions.
5687         * config/rs6000/rs6000-c.c: Add the definitions for
5688         P9V_BUILTIN_VEC_EXTRACT4B and P9V_BUILTIN_VEC_INSERT4B.
5689         * config/rs6000/rs6000.c (altivec_expand_builtin): Add
5690         P9V_BUILTIN_EXTRACT4B and P9V_BUILTIN_INSERT4B case statements.
5691         * config/rs6000/vsx.md: Add define_insn extract4b.  Add define_expand
5692         definition for insert4b and define insn *insert3b_internal.
5693         * doc/extend.texi: Add documentation for vec_extract4b.
5694
5695 2018-02-16  Nathan Sidwell  <nathan@acm.org>
5696
5697         * doc/extend.texi (Backwards Compatibility): Mention friend
5698         injection.  Note for-scope is deprecated.
5699         * doc/invoke.texi (-ffriend-injection): Deprecate.
5700
5701 2018-02-16  Segher Boessenkool  <segher@kernel.crashing.org>
5702
5703         * combine.c (try_combine): When adjusting LOG_LINKS for the destination
5704         that moved to I2, also allow destinations that are a paradoxical
5705         subreg (instead of a normal reg).
5706
5707 2018-02-16  Oleg Endo  <olegendo@gcc.gnu.org>
5708
5709         PR target/83831
5710         * config/rx/rx.c (rx_fuse_in_memory_bitop): Convert shift operand
5711         to QImode.
5712
5713 2018-02-16  Richard Biener  <rguenther@suse.de>
5714
5715         PR tree-optimization/84037
5716         PR tree-optimization/84016
5717         PR target/82862
5718         * config/i386/i386.c (ix86_builtin_vectorization_cost):
5719         Adjust vec_construct for the fact we need additional higher latency
5720         128bit inserts for AVX256 and AVX512 vector builds.
5721         (ix86_add_stmt_cost): Scale vector construction cost for
5722         elementwise loads.
5723
5724 2018-02-16  Richard Biener  <rguenther@suse.de>
5725
5726         PR tree-optimization/84417
5727         * tree-ssa.c (non_rewritable_mem_ref_base): Properly constrain
5728         the MEM_REF offset when conversion to BIT_FIELD_REF is desired.
5729         (non_rewritable_lvalue_p): Likewise, use poly-ints.
5730
5731 2018-02-16  Martin Liska  <mliska@suse.cz>
5732
5733         PR sanitizer/84307
5734         * internal-fn.def (ASAN_CHECK): Set proper flags.
5735         (ASAN_MARK): Likewise.
5736
5737 2018-02-16  Julia Koval  <julia.koval@intel.com>
5738
5739         * config/i386/i386.c (ix86_option_override_internal): Remove PTA_CLWB
5740         from PTA_CANNONLAKE.
5741
5742 2018-02-16  Jakub Jelinek  <jakub@redhat.com>
5743
5744         PR target/84272
5745         * config/aarch64/cortex-a57-fma-steering.c (fma_forest::merge_forest):
5746         Use ++iter rather than iter++ for std::list iterators.
5747         (func_fma_steering::dfs): Likewise.  Don't delete nodes right away,
5748         defer deleting them until all nodes in the forest are processed.  Do
5749         free even leaf nodes.  Change to_process into auto_vec.
5750
5751         PR bootstrap/84405
5752         * system.h (BROKEN_VALUE_INITIALIZATION): Define for GCC < 4.3.
5753         * vec.h (vec_default_construct): Use memset instead of placement new
5754         if BROKEN_VALUE_INITIALIZATION is defined.
5755         * hash-table.h (hash_table<Descriptor, Allocator>::empty_slow): Use
5756         memset instead of value initialization if BROKEN_VALUE_INITIALIZATION
5757         is defined.
5758
5759         PR rtl-optimization/83723
5760         * lra-int.h (lra_substitute_pseudo): Add DEBUG_P argument.
5761         * lra.c (lra_substitute_pseudo): Likewise.  If true, use
5762         gen_rtx_raw_SUBREG instead of gen_rtx_SUBREG.  Pass DEBUG_P to
5763         recursive calls.
5764         (lra_substitute_pseudo_within_insn): Adjust lra_substitute_pseudo
5765         callers.
5766         * lra-constraints.c (inherit_reload_reg, split_reg): Likewise.
5767
5768 2018-02-16  Eric Botcazou  <ebotcazou@adacore.com>
5769
5770         PR rtl-optimization/81443
5771         * rtlanal.c (num_sign_bit_copies1) <SUBREG>: Do not propagate results
5772         from inner REGs to paradoxical SUBREGs.
5773
5774 2018-02-16  Richard Biener  <rguenther@suse.de>
5775
5776         PR tree-optimization/84399
5777         * graphite-scop-detection.c (scop_detection::stmt_simple_for_scop_p):
5778         For operands we can analyze at their definition make sure we can
5779         analyze them at each use as well.
5780
5781 2018-02-16  Richard Biener  <rguenther@suse.de>
5782
5783         PR tree-optimization/84190
5784         * tree-ssa.c (non_rewritable_mem_ref_base): Do not touch
5785         volatile accesses if the decl isn't volatile.
5786
5787 2018-02-15  Jason Merrill  <jason@redhat.com>
5788
5789         PR c++/84314 - ICE with templates and fastcall attribute.
5790         * attribs.c (build_type_attribute_qual_variant): Don't clobber
5791         TYPE_CANONICAL on an existing type.
5792
5793 2018-02-15  Jakub Jelinek  <jakub@redhat.com>
5794
5795         PR tree-optimization/84383
5796         * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Don't look at
5797         dstoff nor call operand_equal_p if dstbase is NULL.
5798
5799         PR tree-optimization/84334
5800         * match.pd ((A +- CST1) +- CST2 -> A + CST3): If A is
5801         also a CONSTANT_CLASS_P, punt.
5802
5803 2018-02-14  Jim Wilson  <jimw@sifive.com>
5804
5805         * config/riscv/riscv.c (riscv_first_stack_step): Move locals after
5806         first SMALL_OPERAND check.  New local min_second_step.  Move assert
5807         to where locals are set.  Add TARGET_RVC support.
5808         * config/riscv/riscv.h (C_SxSP_BITS, SWSP_REACH, SDSP_REACH): New.
5809
5810 2018-02-14  Indu Bhagat  <indu.bhagat@oracle.com>
5811
5812         * doc/invoke.texi: Correct -Wformat-overflow code sample.
5813
5814 2018-02-14  Martin Sebor  <msebor@redhat.com>
5815
5816         PR tree-optimization/83698
5817         * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For
5818         arrays constrain the offset range to their bounds.
5819         (builtin_access::strcat_overlap): Adjust the bounds of overlap offset.
5820         (builtin_access::overlap): Avoid setting the size of overlap if it's
5821         already been set.
5822         (maybe_diag_overlap): Also consider arrays when deciding what values
5823         of offsets to include in diagnostics.
5824
5825 2018-02-14  Martin Sebor  <msebor@redhat.com>
5826
5827         PR c/84108
5828         * attribs.c (diag_attr_exclusions): Consider the exclusion(s)
5829         that correspond to the kind of a declaration.
5830
5831 2018-02-14  John David Anglin  <danglin@gcc.gnu.org>
5832
5833         PR target/83984
5834         * config/pa/pa.md: Load address of PIC label using the linkage table
5835         if the label is nonlocal.
5836
5837 2018-02-14  Kelvin Nilsen  <kelvin@gcc.gnu.org>
5838
5839         * config/rs6000/rs6000.c (rs6000_option_override_internal): Issue
5840         warning message if user requests -maltivec=be.
5841         * doc/invoke.texi: Document deprecation of -maltivec=be.
5842
5843 2018-02-14  Will Schmidt  <will_schmidt@vnet.ibm.com>
5844
5845         PR target/84220
5846         * config/rs6000/rs6000-c.c: Update definitions for
5847         ALTIVEC_BUILTIN_VEC_SLD, ALTIVEC_BUILTIN_VEC_SLDW,
5848         VEC_XXSLDWI and ALTIVEC_BUILTIN_VEC_XXPERMDI builtins.
5849
5850 2018-02-14  Igor Tsimbalist  <igor.v.tsimbalist@intel.com>
5851
5852         PR target/84239
5853         * config/i386/cetintrin.h: Remove _rdssp[d|q] and
5854         add _get_ssp intrinsics. Remove argument from
5855         __builtin_ia32_rdssp[d|q].
5856         * config/i386/i386-builtin-types.def: Add UINT_FTYPE_VOID.
5857         * config/i386/i386-builtin.def: Remove argument from
5858         __builtin_ia32_rdssp[d|q].
5859         * config/i386/i386.c: Use UINT_FTYPE_VOID. Use
5860         ix86_expand_special_args_builtin for _rdssp[d|q].
5861         * config/i386/i386.md: Remove argument from rdssp[si|di] insn.
5862         Clear register before usage.
5863         * doc/extend.texi: Remove argument from __builtin_ia32_rdssp[d|q].
5864         Add documentation for new _get_ssp and _inc_ssp intrinsics.
5865
5866 2018-02-14  Richard Sandiford  <richard.sandiford@linaro.org>
5867
5868         PR tree-optimization/84357
5869         * tree-data-ref.c (object_address_invariant_in_loop_p): Check
5870         operand 1 of an ARRAY_REF too.
5871
5872 2018-02-14  Oleg Endo  <olegendo@gcc.gnu.org>
5873
5874         PR target/83831
5875         * config/rx/rx-protos.h (rx_reg_dead_or_unused_after_insn,
5876         rx_copy_reg_dead_or_unused_notes, rx_fuse_in_memory_bitop): New
5877         declarations.
5878         (set_of_reg): New struct.
5879         (rx_find_set_of_reg, rx_find_use_of_reg): New functions.
5880         * config/rx/rx.c (rx_reg_dead_or_unused_after_insn,
5881         rx_copy_reg_dead_or_unused_notes, rx_fuse_in_memory_bitop): New
5882         functions.
5883         * config/rx/rx.md (andsi3, iorsi3, xorsi3): Convert to insn_and_split.
5884         Split into bitclr, bitset, bitinvert patterns if appropriate.
5885         (*bitset, *bitinvert, *bitclr): Convert to named insn_and_split and
5886         use rx_fuse_in_memory_bitop.
5887         (*bitset_in_memory, *bitinvert_in_memory, *bitclr_in_memory): Convert
5888         to named insn, correct maximum insn length.
5889
5890 2018-02-14  Jozef Lawrynowicz  <jozefl.gcc@gmail.com>
5891
5892         PR target/79242
5893         * machmode.def: Define a complex mode for PARTIAL_INT.
5894         * genmodes.c (complex_class): Return MODE_COMPLEX_INT for
5895         MODE_PARTIAL_INT.
5896         * doc/rtl.texi: Document CSPImode.
5897         * config/msp430/msp430.c (msp430_hard_regno_nregs): Add CPSImode
5898         handling.
5899         (msp430_hard_regno_nregs_with_padding): Likewise.
5900
5901 2018-02-13  Peter Bergner  <bergner@vnet.ibm.com>
5902
5903         PR target/84279
5904         * config/rs6000/rs6000.c (mem_operand_gpr): Disallow altivec addresses.
5905
5906 2018-02-13  Segher Boessenkool  <segher@kernel.crashing.org>
5907
5908         PR rtl-optimization/84169
5909         * combine.c (try_combine): New variable split_i2i3.  Set it to true if
5910         we generated a parallel as new i3 and we split that to new i2 and i3
5911         instructions.  Handle split_i2i3 similar to swap_i2i3: scan the
5912         LOG_LINKs of i3 to see which of those need to link to i2 now.  Link
5913         those to i2, not i1.  Partially rewrite this scan code.
5914
5915 2018-02-13  Jakub Jelinek  <jakub@redhat.com>
5916
5917         PR c/82210
5918         * stor-layout.c (place_field): For variable length fields, adjust
5919         offset_align afterwards not just based on the field's alignment,
5920         but also on the size.
5921
5922         PR middle-end/84309
5923         * match.pd (pow(C,x) -> exp(log(C)*x)): Use exp2s and log2s instead
5924         of exps and logs in the use_exp2 case.
5925
5926 2018-02-13  Jeff Law  <law@redhat.com>
5927
5928         * config/rl/rl78.c (rl78_attribute_table): Fix terminator and
5929         entry for "vector".
5930
5931         * config/rl78/rl78.c (rl78_handle_func_attribute): Mark
5932         ARGS as unused.
5933
5934 2018-02-13  Alexandre Oliva  <aoliva@redhat.com>
5935
5936         PR debug/84342
5937         PR debug/84319
5938         * common.opt (gas-loc-support, gas-locview-support): New.
5939         (ginline-points, ginternal-reset-location-views): New.
5940         * doc/invoke.texi: Document them.  Use @itemx where intended.
5941         (gvariable-location-views): Adjust.
5942         * target.def (reset_location_view): New.
5943         * doc/tm.texi.in (DWARF2_ASM_VIEW_DEBUG_INFO): New.
5944         (TARGET_RESET_LOCATION_VIEW): New.
5945         * doc/tm.texi: Rebuilt.
5946         * dwarf2out.c (dwarf2out_default_as_loc_support): New.
5947         (dwarf2out_default_as_locview_support): New.
5948         (output_asm_line_debug_info): Use option variables.
5949         (dwarf2out_maybe_output_loclist_view_pair): Likewise.
5950         (output_loc_list): Likewise.
5951         (add_high_low_attributes): Check option variables.
5952         Don't output entry view attribute in strict mode.
5953         (gen_inlined_subroutine_die): Check option variables.
5954         (dwarf2out_inline_entry): Likewise.
5955         (init_sections_and_labels): Likewise.
5956         (dwarf2out_early_finish): Likewise.
5957         (maybe_reset_location_view): New, from...
5958         (dwarf2out_var_location): ... here.  Call it.
5959         * debug.h (dwarf2out_default_as_loc_support): Declare.
5960         (dwarf2out_default_as_locview_support): Declare.
5961         * hooks.c (hook_int_rtx_insn_0): New.
5962         * hooks.h (hook_int_rtx_insn_0): Declare.
5963         * toplev.c (process_options): Take -gas-loc-support and
5964         -gas-locview-support from dwarf2out.  Enable
5965         -gvariable-location-views by default only with locview
5966         assembler support.  Enable -ginternal-reset-location-views by
5967         default only if the target defines the corresponding hook.
5968         Enable -ginline-points by default if location views are
5969         enabled; force it disabled if statement frontiers are
5970         disabled.
5971         * tree-inline.c (expand_call_inline): Check option variables.
5972         * tree-ssa-live.c (remove_unused_scope_block_p): Likewise.
5973
5974 2018-02-13  Richard Sandiford  <richard.sandiford@linaro.org>
5975
5976         PR tree-optimization/84321
5977         * tree-vrp.c (intersect_range_with_nonzero_bits): Fix VR_ANTI_RANGE
5978         handling.  Also check whether the anti-range contains any values
5979         that satisfy the mask; switch to a VR_RANGE if not.
5980
5981 2018-02-13  Paolo Bonzini  <bonzini@gnu.org>
5982
5983         PR sanitizer/84340
5984         * internal-fn.def (ASAN_CHECK, ASAN_MARK): Revert changes to fnspec.
5985
5986 2018-02-13  Martin Jambor  <mjambor@suse.cz>
5987
5988         PR c++/83990
5989         * ipa-param-manipulation.c (ipa_modify_call_arguments): Use location
5990         of call statements, also set location of a load to a temporary.
5991
5992 2018-02-13  Sebastian Perta  <sebastian.perta@renesas.com>
5993
5994         * config/rl78/rl78.c (add_vector_labels): New function.
5995         * config/rl78/rl78.c (rl78_handle_vector_attribute): New function.
5996         * config/rl78/rl78.c (rl78_start_function): Call add_vector_labels.
5997         * config/rl78/rl78.c (rl78_handle_func_attribute): Removed the assert
5998         which checks that no arguments are passed.
5999         * config/rl78/rl78.c (rl78_attribute_table): Add "vector" attribute.
6000         * doc/extend.texi: Documentation for the new attribute.
6001
6002 2018-02-13  Andreas Schwab  <schwab@suse.de>
6003
6004         * config/riscv/linux.h (CPP_SPEC): Define.
6005
6006 2018-02-13  Jakub Jelinek  <jakub@redhat.com>
6007
6008         PR target/84335
6009         * config/i386/i386.c (ix86_init_mmx_sse_builtins): Pass
6010         OPTION_MASK_ISA_AES | OPTION_MASK_ISA_SSE2 instead of
6011         OPTION_MASK_ISA_AES as first argument to def_builtin_const
6012         for AES builtins.  Pass OPTION_MASK_ISA_PCLMUL | OPTION_MASK_ISA_SSE2
6013         instead of OPTION_MASK_ISA_PCLMUL as first argument to
6014         def_builtin_const for __builtin_ia32_pclmulqdq128 builtin.
6015         * config/i386/wmmintrin.h: If __SSE2__ is not defined, enable it
6016         temporarily for AES and PCLMUL builtins.
6017
6018         PR tree-optimization/84339
6019         * gimple-fold.c (get_range_strlen): Set *FLEXP to true when handling
6020         ARRAY_REF where first operand is array_at_struct_end_p COMPONENT_REF.
6021         Formatting fixes.
6022
6023         PR middle-end/84309
6024         * match.pd (pow(C,x) -> exp(log(C)*x)): Optimize instead into
6025         exp2(log2(C)*x) if C is a power of 2 and c99 runtime is available.
6026         * generic-match-head.c (canonicalize_math_after_vectorization_p): New
6027         inline function.
6028         * gimple-match-head.c (canonicalize_math_after_vectorization_p): New
6029         inline function.
6030         * omp-simd-clone.h: New file.
6031         * omp-simd-clone.c: Include omp-simd-clone.h.
6032         (expand_simd_clones): No longer static.
6033         * tree-vect-patterns.c: Include fold-const-call.h, attribs.h,
6034         cgraph.h and omp-simd-clone.h.
6035         (vect_recog_pow_pattern): Optimize pow(C,x) to exp(log(C)*x).
6036         (vect_recog_widen_shift_pattern): Formatting fix.
6037         (vect_pattern_recog_1): Don't check optab for calls.
6038
6039         PR target/84336
6040         * config/i386/sse.md (<avx512>_vpermi2var<mode>3_mask): Force
6041         operands[2] into a REG before using gen_lowpart on it.
6042
6043 2018-02-12  Jeff Law  <law@redhat.com>
6044
6045         PR target/83760
6046         * config/sh/sh.c (find_barrier): Consider a sibling call
6047         a barrier as well.
6048
6049         * cse.c (try_back_substitute_reg): Move any REG_ARGS_SIZE note when
6050         successfully back substituting a reg.
6051
6052 2018-02-12  Richard Biener  <rguenther@suse.de>
6053
6054         PR tree-optimization/84037
6055         * tree-vect-slp.c (vect_analyze_slp_cost): Add visited
6056         parameter, move visited init to caller.
6057         (vect_slp_analyze_operations): Separate cost from validity
6058         check, initialize visited once for all instances.
6059         (vect_schedule_slp): Analyze map to CSE vectorized nodes once
6060         for all instances.
6061         * tree-vect-stmts.c (vect_model_simple_cost): Make early
6062         out an assert.
6063         (vect_model_promotion_demotion_cost): Likewise.
6064         (vectorizable_bswap): Guard cost modeling with !slp_node
6065         instead of !PURE_SLP_STMT to avoid double-counting on hybrid
6066         SLP stmts.
6067         (vectorizable_call): Likewise.
6068         (vectorizable_conversion): Likewise.
6069         (vectorizable_assignment): Likewise.
6070         (vectorizable_shift): Likewise.
6071         (vectorizable_operation): Likewise.
6072         (vectorizable_store): Likewise.
6073         (vectorizable_load): Likewise.
6074         (vectorizable_condition): Likewise.
6075         (vectorizable_comparison): Likewise.
6076
6077 2018-02-12  Paolo Bonzini  <bonzini@gnu.org>
6078
6079         PR sanitizer/84307
6080         * internal-fn.def (ASAN_CHECK): Fix fnspec to account for return value.
6081         (ASAN_MARK): Fix fnspec to account for return value, change pointer
6082         argument from 'R' to 'W' so that the pointed-to datum is clobbered.
6083
6084 2018-02-08  Jan Hubicka  <hubicka@ucw.cz>
6085
6086         PR middle-end/83665
6087         * params.def (inline-min-speedup): Increase from 8 to 15.
6088         (max-inline-insns-auto): Decrease from 40 to 30.
6089         * ipa-split.c (consider_split): Add some buffer for function to
6090         be considered inlining candidate.
6091         * invoke.texi (max-inline-insns-auto, inline-min-speedup): UPdate
6092         default values.
6093
6094 2018-02-12  Richard Biener  <rguenther@suse.de>
6095
6096         PR tree-optimization/84037
6097         * tree-vect-slp.c (vect_build_slp_tree_2): Try swapping the
6098         matched stmts if we cannot swap the non-matched ones.
6099
6100 2018-02-12  Olga Makhotina  <olga.makhotina@intel.com>
6101
6102         * config/i386/avx512fintrin.h (_mm_mask_scalef_round_sd,
6103         _mm_maskz_scalef_round_sd, _mm_mask_scalef_round_ss,
6104         _mm_maskz_scalef_round_ss): New intrinsics.
6105         (__builtin_ia32_scalefsd_round, __builtin_ia32_scalefss_round): Fix.
6106         * config/i386/i386-builtin.def (__builtin_ia32_scalefsd_round,
6107         __builtin_ia32_scalefss_round): Remove.
6108         (__builtin_ia32_scalefsd_mask_round,
6109         __builtin_ia32_scalefss_mask_round): New intrinsics.
6110         * config/i386/sse.md (vmscalef<mode><round_name>): Renamed to ...
6111         (vmscalef<mode><mask_scalar_name><round_scalar_name>): ... this.
6112         ((match_operand:VF_128 2 "<round_nimm_predicate>"
6113         "<round_constraint>")): Changed to ...
6114         ((match_operand:VF_128 2 "<round_scalar_nimm_predicate>"
6115         "<round_scalar_constraint>")): ... this.
6116         ("vscalef<ssescalarmodesuffix>\t{<round_op3>%2, %1, %0|
6117         %0, %1, %2<round_op3>}"): Changed to ...
6118         ("vscalef<ssescalarmodesuffix>\t{<round_scalar_mask_op3>%2, %1,
6119         %0<mask_scalar_operand3>|%0<mask_scalar_operand3>, %1,
6120         %2<round_scalar_mask_op3>}"): ... this.
6121         * config/i386/subst.md (round_scalar_nimm_predicate): New.
6122
6123 2018-02-12  Olga Makhotina  <olga.makhotina@intel.com>
6124
6125         * config/i386/avx512fintrin.h (_mm_mask_sqrt_round_sd)
6126         (_mm_maskz_sqrt_round_sd, _mm_mask_sqrt_round_ss)
6127         (_mm_maskz_sqrt_round_ss): New intrinsics.
6128         (__builtin_ia32_sqrtsd_round, __builtin_ia32_sqrtss_round): Remove.
6129         (__builtin_ia32_sqrtsd_mask_round)
6130         (__builtin_ia32_sqrtss_mask_round): New builtins.
6131         * config/i386/i386-builtin.def (__builtin_ia32_sqrtsd_round)
6132         (__builtin_ia32_sqrtss_round): Remove.
6133         (__builtin_ia32_sqrtsd_mask_round)
6134         (__builtin_ia32_sqrtss_mask_round): New builtins.
6135         * config/i386/sse.md (vmsqrt<mode>2<round_name>): Renamed to ...
6136         (vmsqrt<mode>2<mask_scalar_name><round_scalar_name>): ... this.
6137         ((match_operand:VF_128 1 "vector_operand"
6138         "xBm,<round_constraint>")): Changed to ...
6139         ((match_operand:VF_128 1 "vector_operand"
6140         "xBm,<round_scalar_constraint>")): ... this.
6141         (vsqrt<ssescalarmodesuffix>\t{<round_op3>%1, %2, %0|
6142         %0, %2, %<iptr>1<round_op3>}): Changed to ...
6143         (vsqrt<ssescalarmodesuffix>\t{<round_scalar_mask_op3>%1, %2,
6144         %0<mask_scalar_operand3>|%0<mask_scalar_operand3>, %2,
6145         %<iptr>1<round_scalar_mask_op3>}): ... this.
6146         ((set_attr "prefix" "<round_prefix>")): Changed to ...
6147         ((set_attr "prefix" "<round_scalar_prefix>")): ... this.
6148
6149 2018-02-11  Steven Munroe  <munroesj@gcc.gnu.org>
6150
6151         PR target/84266
6152         * config/rs6000/mmintrin.h (_mm_cmpeq_pi32 [_ARCH_PWR9]):
6153         Cast vec_cmpeq result to correct type.
6154         * config/rs6000/mmintrin.h (_mm_cmpgt_pi32 [_ARCH_PWR9]):
6155         Cast vec_cmpgt result to correct type.
6156
6157 2018-02-11  Alexandre Oliva  <aoliva@redhat.com>
6158
6159         * final.c (final_scan_insn_1): Renamed from...
6160         (final_scan_insn): ... this.  New wrapper, to recover
6161         seen from the outermost call in recursive ones.
6162         * config/sparc/sparc.c (output_return): Drop seen from call.
6163         (output_sibcall): Likewise.
6164         * config/visium/visium.c (output_branch): Likewise.
6165
6166 2018-02-10  John David Anglin  <danglin@gcc.gnu.org>
6167
6168         * config/pa/pa.c (hppa_profile_hook): Mark SYMBOL_REF for _mcount as
6169         function label.
6170
6171 2018-02-10  Alan Modra  <amodra@gmail.com>
6172
6173         PR target/84300
6174         * config/rs6000/rs6000.md (split_stack_return): Remove (use ..).
6175         Specify LR as an input.
6176
6177 2018-02-10  Jakub Jelinek  <jakub@redhat.com>
6178
6179         PR sanitizer/83987
6180         * omp-low.c (maybe_remove_omp_member_access_dummy_vars,
6181         remove_member_access_dummy_vars): New functions.
6182         (lower_omp_for, lower_omp_taskreg, lower_omp_target,
6183         lower_omp_1, execute_lower_omp): Use them.
6184
6185         PR rtl-optimization/84308
6186         * shrink-wrap.c (spread_components): Release todo vector.
6187
6188 2018-02-09  Vladimir Makarov  <vmakarov@redhat.com>
6189
6190         PR rtl-optimization/57193
6191         * ira-color.c (struct allocno_color_data): Add member
6192         conflict_allocno_hard_prefs.
6193         (update_conflict_allocno_hard_prefs): New.
6194         (bucket_allocno_compare_func): Add a preference based on
6195         conflict_allocno_hard_prefs.
6196         (push_allocno_to_stack): Update conflict_allocno_hard_prefs.
6197         (color_allocnos): Remove a dead code.  Initiate
6198         conflict_allocno_hard_prefs.  Call update_costs_from_prefs.
6199
6200 2018-02-09  Jakub Jelinek  <jakub@redhat.com>
6201
6202         PR target/84226
6203         * config/rs6000/vsx.md (p9_xxbrq_v16qi): Change input operand
6204         constraint from =wa to wa.  Avoid a subreg on the output operand,
6205         instead use a pseudo and subreg it in a move.
6206         (p9_xxbrd_<mode>): Changed to ...
6207         (p9_xxbrd_v2di): ... this insn, without VSX_D iterator.
6208         (p9_xxbrd_v2df): New expander.
6209         (p9_xxbrw_<mode>): Changed to ...
6210         (p9_xxbrw_v4si): ... this insn, without VSX_W iterator.
6211         (p9_xxbrw_v4sf): New expander.
6212
6213 2018-02-09  Sebastian Perta  <sebastian.perta@renesas.com>
6214
6215         * config/rx/rx.md (movsicc): Update expander to be matched by GCC.
6216
6217 2018-02-09  Peter Bergner  <bergner@vnet.ibm.com>
6218
6219         PR target/83926
6220         * config/rs6000/vsx.md (vsx_mul_v2di): Handle generating a 64-bit
6221         multiply in 32-bit mode.
6222         (vsx_div_v2di): Handle generating a 64-bit signed divide in 32-bit mode.
6223         (vsx_udiv_v2di): Handle generating a 64-bit unsigned divide in 32-bit
6224         mode.
6225
6226 2018-02-09  Sebastian Perta  <sebastian.perta@renesas.com>
6227
6228         * config/rx/constraints.md (CALL_OP_SYMBOL_REF): Added new constraint
6229         to allow or block "symbol_ref" depending on the value of TARGET_JSR.
6230         * config/rx/rx.md (call_internal): Use CALL_OP_SYMBOL_REF.
6231         * config/rx/rx.md (call_value_internal): Use CALL_OP_SYMBOL_REF.
6232
6233 2018-02-09  Pierre-Marie de Rodat  <derodat@adacore.com>
6234
6235         PR lto/84213
6236         * dwarf2out.c (is_trivial_indirect_ref): New function.
6237         (dwarf2out_late_global_decl): Do not generate a location
6238         attribute for variables that have a non-trivial DECL_VALUE_EXPR
6239         and that are not defined in the current unit.
6240
6241 2018-02-09  Eric Botcazou  <ebotcazou@adacore.com>
6242
6243         * optabs.c (prepare_cmp_insn): Try harder to emit a direct comparison
6244         instead of a libcall for UNORDERED.
6245
6246 2018-02-09  Tamar Christina  <tamar.christina@arm.com>
6247
6248         PR target/82641
6249         * config/arm/arm-c.c (arm_cpu_builtins): Un-define __ARM_FEATURE_LDREX,
6250         __ARM_ARCH_PROFILE, __ARM_ARCH_ISA_THUMB, __ARM_FP and __ARM_NEON_FP.
6251
6252 2018-02-09  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>
6253
6254         PR target/PR84295
6255         * config/s390/s390.c (s390_set_current_function): Invoke
6256         s390_indirect_branch_settings also if fndecl didn't change.
6257
6258 2018-02-09  Alexandre Oliva  <aoliva@redhat.com>
6259
6260         * config/rs6000/rs6000.md (blockage): Set length to zero.
6261
6262 2018-02-09  Eric Botcazou  <ebotcazou@adacore.com>
6263
6264         * expr.c (optimize_bitfield_assignment_op): Remove obsolete assertion.
6265
6266 2018-02-09  Jakub Jelinek  <jakub@redhat.com>
6267
6268         PR sanitizer/84285
6269         * gcc.c (STATIC_LIBASAN_LIBS, STATIC_LIBTSAN_LIBS,
6270         STATIC_LIBLSAN_LIBS, STATIC_LIBUBSAN_LIBS): Handle -static like
6271         -static-lib*san.
6272
6273         PR debug/84252
6274         * var-tracking.c (vt_add_function_parameter): Punt for non-onepart
6275         PARALLEL incoming that failed vt_get_decl_and_offset check.
6276
6277         PR middle-end/84237
6278         * output.h (bss_initializer_p): Add NAMED argument, defaulted to false.
6279         * varasm.c (bss_initializer_p): Add NAMED argument, if true, ignore
6280         TREE_READONLY bit.
6281         (get_variable_section): For decls in named .bss* sections pass true as
6282         second argument to bss_initializer_p.
6283
6284 2018-02-09  Marek Polacek  <polacek@redhat.com>
6285             Jakub Jelinek  <jakub@redhat.com>
6286
6287         PR c++/83659
6288         * fold-const.c (fold_indirect_ref_1): Use VECTOR_TYPE_P macro.
6289         Formatting fixes.  Verify first that tree_fits_poly_int64_p (op01).
6290         Sync some changes from cxx_fold_indirect_ref.
6291
6292 2018-02-09  Alexandre Oliva  <aoliva@redhat.com>
6293
6294         * cfgexpand.c (expand_gimple_basic_block): Handle inline entry
6295         markers.
6296         * dwarf2out.c (dwarf2_debug_hooks): Enable inline_entry hook.
6297         (BLOCK_INLINE_ENTRY_LABEL): New.
6298         (dwarf2out_var_location): Disregard inline entry markers.
6299         (inline_entry_data): New struct.
6300         (inline_entry_data_hasher): New hashtable type.
6301         (inline_entry_data_hasher::hash): New.
6302         (inline_entry_data_hasher::equal): New.
6303         (inline_entry_data_table): New variable.
6304         (add_high_low_attributes): Add DW_AT_entry_pc and
6305         DW_AT_GNU_entry_view attributes if a pending entry is found
6306         in inline_entry_data_table.  Add old entry_pc attribute only
6307         if debug nonbinding markers are disabled.
6308         (gen_inlined_subroutine_die): Set BLOCK_DIE if nonbinding
6309         markers are enabled.
6310         (block_within_block_p, dwarf2out_inline_entry): New.
6311         (dwarf2out_finish): Check that no entries remained in
6312         inline_entry_data_table.
6313         * final.c (reemit_insn_block_notes): Handle inline entry notes.
6314         (final_scan_insn, notice_source_line): Likewise.
6315         (rest_of_clean_state): Skip inline entry markers.
6316         * gimple-pretty-print.c (dump_gimple_debug): Handle inline entry
6317         markers.
6318         * gimple.c (gimple_build_debug_inline_entry): New.
6319         * gimple.h (enum gimple_debug_subcode): Add
6320         GIMPLE_DEBUG_INLINE_ENTRY.
6321         (gimple_build_debug_inline_entry): Declare.
6322         (gimple_debug_inline_entry_p): New.
6323         (gimple_debug_nonbind_marker_p): Adjust.
6324         * insn-notes.def (INLINE_ENTRY): New.
6325         * print-rtl.c (rtx_writer::print_rtx_operand_code_0): Handle
6326         inline entry marker notes.
6327         (print_insn): Likewise.
6328         * rtl.h (NOTE_MARKER_P): Add INLINE_ENTRY support.
6329         (INSN_DEBUG_MARKER_KIND): Likewise.
6330         (GEN_RTX_DEBUG_MARKER_INLINE_ENTRY_PAT): New.
6331         * tree-inline.c (expand_call_inline): Build and insert
6332         debug_inline_entry stmt.
6333         * tree-ssa-live.c (remove_unused_scope_block_p): Preserve
6334         inline entry blocks early, if nonbind markers are enabled.
6335         (dump_scope_block): Dump fragment info.
6336         * var-tracking.c (reemit_marker_as_note): Handle inline entry note.
6337         * doc/gimple.texi (gimple_debug_inline_entry_p): New.
6338         (gimple_build_debug_inline_entry): New.
6339         * doc/invoke.texi (gstatement-frontiers, gno-statement-frontiers):
6340         Enable/disable inline entry points too.
6341         * doc/rtl.texi (NOTE_INSN_INLINE_ENTRY): New.
6342         (DEBUG_INSN): Describe inline entry markers.
6343
6344         * common.opt (gvariable-location-views): New.
6345         (gvariable-location-views=incompat5): New.
6346         * config.in: Rebuilt.
6347         * configure: Rebuilt.
6348         * configure.ac: Test assembler for view support.
6349         * dwarf2asm.c (dw2_asm_output_symname_uleb128): New.
6350         * dwarf2asm.h (dw2_asm_output_symname_uleb128): Declare.
6351         * dwarf2out.c (var_loc_view): New typedef.
6352         (struct dw_loc_list_struct): Add vl_symbol, vbegin, vend.
6353         (dwarf2out_locviews_in_attribute): New.
6354         (dwarf2out_locviews_in_loclist): New.
6355         (dw_val_equal_p): Compare val_view_list of dw_val_class_view_lists.
6356         (enum dw_line_info_opcode): Add LI_adv_address.
6357         (struct dw_line_info_table): Add view.
6358         (RESET_NEXT_VIEW, RESETTING_VIEW_P): New macros.
6359         (DWARF2_ASM_VIEW_DEBUG_INFO): Define default.
6360         (zero_view_p): New variable.
6361         (ZERO_VIEW_P): New macro.
6362         (output_asm_line_debug_info): New.
6363         (struct var_loc_node): Add view.
6364         (add_AT_view_list, AT_loc_list): New.
6365         (add_var_loc_to_decl): Add view param.  Test it against last.
6366         (new_loc_list): Add view params.  Record them.
6367         (AT_loc_list_ptr): Handle loc and view lists.
6368         (view_list_to_loc_list_val_node): New.
6369         (print_dw_val): Handle dw_val_class_view_list.
6370         (size_of_die): Likewise.
6371         (value_format): Likewise.
6372         (loc_list_has_views): New.
6373         (gen_llsym): Set vl_symbol too.
6374         (maybe_gen_llsym, skip_loc_list_entry): New.
6375         (dwarf2out_maybe_output_loclist_view_pair): New.
6376         (output_loc_list): Output view list or entries too.
6377         (output_view_list_offset): New.
6378         (output_die): Handle dw_val_class_view_list.
6379         (output_dwarf_version): New.
6380         (output_compilation_unit_header): Use it.
6381         (output_skeleton_debug_sections): Likewise.
6382         (output_rnglists, output_line_info): Likewise.
6383         (output_pubnames, output_aranges): Update version comments.
6384         (output_one_line_info_table): Output view numbers in asm comments.
6385         (dw_loc_list): Determine current endview, pass it to new_loc_list.
6386         Call maybe_gen_llsym.
6387         (loc_list_from_tree_1): Adjust.
6388         (add_AT_location_description): Create view list attribute if
6389         needed, check it's absent otherwise.
6390         (convert_cfa_to_fb_loc_list): Adjust.
6391         (maybe_emit_file): Call output_asm_line_debug_info for test.
6392         (dwarf2out_var_location): Reset views as needed.  Precompute
6393         add_var_loc_to_decl args.  Call get_attr_min_length only if we have the
6394         attribute.  Set view.
6395         (new_line_info_table): Reset next view.
6396         (set_cur_line_info_table): Call output_asm_line_debug_info for test.
6397         (dwarf2out_source_line): Likewise.  Output view resets and labels to
6398         the assembler, or select appropriate line info opcodes.
6399         (prune_unused_types_walk_attribs): Handle dw_val_class_view_list.
6400         (optimize_string_length): Catch it.  Adjust.
6401         (resolve_addr): Copy vl_symbol along with ll_symbol.  Handle
6402         dw_val_class_view_list, and remove it if no longer needed.
6403         (hash_loc_list): Hash view numbers.
6404         (loc_list_hasher::equal): Compare them.
6405         (optimize_location_lists): Check whether a view list symbol is
6406         needed, and whether the locview attribute is present, and
6407         whether they match.  Remove the locview attribute if no longer
6408         needed.
6409         (index_location_lists): Call skip_loc_list_entry for test.
6410         (dwarf2out_finish): Call output_asm_line_debug_info for test.
6411         Use output_dwarf_version.
6412         * dwarf2out.h (enum dw_val_class): Add dw_val_class_view_list.
6413         (struct dw_val_node): Add val_view_list.
6414         * final.c (SEEN_NEXT_VIEW): New.
6415         (set_next_view_needed): New.
6416         (clear_next_view_needed): New.
6417         (maybe_output_next_view): New.
6418         (final_start_function): Rename to...
6419         (final_start_function_1): ... this.  Take pointer to FIRST,
6420         add SEEN parameter.  Emit param bindings in the initial view.
6421         (final_start_function): Reintroduce SEEN-less interface.
6422         (final): Rename to...
6423         (final_1): ... this.  Take SEEN parameter.  Output final pending
6424         next view at the end.
6425         (final): Reintroduce seen-less interface.
6426         (final_scan_insn): Output pending next view before switching
6427         sections or ending a block.  Mark the next view as needed when
6428         outputting variable locations.  Notify debug backend of section
6429         changes, and of location view changes.
6430         (rest_of_handle_final): Adjust.
6431         * toplev.c (process_options): Autodetect value for debug variable
6432         location views option.  Warn on incompat5 without -gdwarf-5.
6433         * doc/invoke.texi (gvariable-location-views): New.
6434         (gvariable-location-views=incompat5): New.
6435         (gno-variable-location-views): New.
6436
6437 2018-02-08  David Malcolm  <dmalcolm@redhat.com>
6438
6439         PR tree-optimization/84136
6440         * tree-cfg.c (find_taken_edge_computed_goto): Remove assertion
6441         that the result of find_edge is non-NULL.
6442
6443 2018-02-08  Sergey Shalnov  <sergey.shalnov@intel.com>
6444
6445         PR target/83008
6446         * config/i386/x86-tune-costs.h (skylake_cost): Fix cost of
6447         storing integer register in SImode.  Fix cost of 256 and 512
6448         byte aligned SSE register store.
6449
6450 2018-02-08  Sergey Shalnov  <sergey.shalnov@intel.com>
6451
6452         * config/i386/i386.c (ix86_multiplication_cost): Fix
6453         multiplication cost for TARGET_AVX512DQ.
6454
6455 2018-02-08  Marek Polacek  <polacek@redhat.com>
6456
6457         PR tree-optimization/84238
6458         * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Verify the result of
6459         get_range_strlen.
6460
6461 2018-02-08  Richard Sandiford  <richard.sandiford@linaro.org>
6462
6463         PR tree-optimization/84265
6464         * tree-vect-stmts.c (vectorizable_store): Don't treat
6465         VMAT_CONTIGUOUS accesses as grouped.
6466         (vectorizable_load): Likewise.
6467
6468 2018-02-08  Richard Sandiford  <richard.sandiford@linaro.org>
6469
6470         PR tree-optimization/81635
6471         * wide-int.h (wi::round_down_for_mask, wi::round_up_for_mask): Declare.
6472         * wide-int.cc (wi::round_down_for_mask, wi::round_up_for_mask)
6473         (test_round_for_mask): New functions.
6474         (wide_int_cc_tests): Call test_round_for_mask.
6475         * tree-vrp.h (intersect_range_with_nonzero_bits): Declare.
6476         * tree-vrp.c (intersect_range_with_nonzero_bits): New function.
6477         * tree-data-ref.c (split_constant_offset_1): Use it to refine the
6478         range returned by get_range_info.
6479
6480 2018-02-08  Jan Hubicka  <hubicka@ucw.cz>
6481
6482         PR ipa/81360
6483         * cgraph.h (symtab_node::output_to_lto_symbol_table_p): Declare
6484         * symtab.c: Include builtins.h
6485         (symtab_node::output_to_lto_symbol_table_p): Move here
6486         from lto-streamer-out.c:output_symbol_p.
6487         * lto-streamer-out.c (write_symbol): Turn early exit to assert.
6488         (output_symbol_p): Move all logic to symtab.c
6489         (produce_symtab): Update.
6490
6491 2018-02-08  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>
6492
6493         * config/s390/s390-opts.h (enum indirect_branch): Define.
6494         * config/s390/s390-protos.h (s390_return_addr_from_memory)
6495         (s390_indirect_branch_via_thunk)
6496         (s390_indirect_branch_via_inline_thunk): Add function prototypes.
6497         (enum s390_indirect_branch_type): Define.
6498         * config/s390/s390.c (struct s390_frame_layout, struct
6499         machine_function): Remove.
6500         (indirect_branch_prez10thunk_mask, indirect_branch_z10thunk_mask)
6501         (indirect_branch_table_label_no, indirect_branch_table_name):
6502         Define variables.
6503         (INDIRECT_BRANCH_NUM_OPTIONS): Define macro.
6504         (enum s390_indirect_branch_option): Define.
6505         (s390_return_addr_from_memory): New function.
6506         (s390_handle_string_attribute): New function.
6507         (s390_attribute_table): Add new attribute handler.
6508         (s390_execute_label): Handle UNSPEC_EXECUTE_JUMP patterns.
6509         (s390_indirect_branch_via_thunk): New function.
6510         (s390_indirect_branch_via_inline_thunk): New function.
6511         (s390_function_ok_for_sibcall): When jumping via thunk disallow
6512         sibling call optimization for non z10 compiles.
6513         (s390_emit_call): Force indirect branch target to be a single
6514         register.  Add r1 clobber for non-z10 compiles.
6515         (s390_emit_epilogue): Emit return jump via return_use expander.
6516         (s390_reorg): Handle JUMP_INSNs as execute targets.
6517         (s390_option_override_internal): Perform validity checks for the
6518         new command line options.
6519         (s390_indirect_branch_attrvalue): New function.
6520         (s390_indirect_branch_settings): New function.
6521         (s390_set_current_function): Invoke s390_indirect_branch_settings.
6522         (s390_output_indirect_thunk_function):  New function.
6523         (s390_code_end): Implement target hook.
6524         (s390_case_values_threshold): Implement target hook.
6525         (TARGET_ASM_CODE_END, TARGET_CASE_VALUES_THRESHOLD): Define target
6526         macros.
6527         * config/s390/s390.h (struct s390_frame_layout)
6528         (struct machine_function): Move here from s390.c.
6529         (TARGET_INDIRECT_BRANCH_NOBP_RET)
6530         (TARGET_INDIRECT_BRANCH_NOBP_JUMP)
6531         (TARGET_INDIRECT_BRANCH_NOBP_JUMP_THUNK)
6532         (TARGET_INDIRECT_BRANCH_NOBP_JUMP_INLINE_THUNK)
6533         (TARGET_INDIRECT_BRANCH_NOBP_CALL)
6534         (TARGET_DEFAULT_INDIRECT_BRANCH_TABLE)
6535         (TARGET_INDIRECT_BRANCH_THUNK_NAME_EXRL)
6536         (TARGET_INDIRECT_BRANCH_THUNK_NAME_EX)
6537         (TARGET_INDIRECT_BRANCH_TABLE): Define macros.
6538         * config/s390/s390.md (UNSPEC_EXECUTE_JUMP)
6539         (INDIRECT_BRANCH_THUNK_REGNUM): Define constants.
6540         (mnemonic attribute): Add values which aren't recognized
6541         automatically.
6542         ("*cjump_long", "*icjump_long", "*basr", "*basr_r"): Disable
6543         pattern for branch conversion.  Fix mnemonic attribute.
6544         ("*c<code>", "*sibcall_br", "*sibcall_value_br", "*return"): Emit
6545         indirect branch via thunk if requested.
6546         ("indirect_jump", "<code>"): Expand patterns for branch conversion.
6547         ("*indirect_jump"): Disable for branch conversion using out of
6548         line thunks.
6549         ("indirect_jump_via_thunk<mode>_z10")
6550         ("indirect_jump_via_thunk<mode>")
6551         ("indirect_jump_via_inlinethunk<mode>_z10")
6552         ("indirect_jump_via_inlinethunk<mode>", "*casesi_jump")
6553         ("casesi_jump_via_thunk<mode>_z10", "casesi_jump_via_thunk<mode>")
6554         ("casesi_jump_via_inlinethunk<mode>_z10")
6555         ("casesi_jump_via_inlinethunk<mode>", "*basr_via_thunk<mode>_z10")
6556         ("*basr_via_thunk<mode>", "*basr_r_via_thunk_z10")
6557         ("*basr_r_via_thunk", "return<mode>_prez10"): New pattern.
6558         ("*indirect2_jump"): Disable for branch conversion.
6559         ("casesi_jump"): Turn into expander and expand patterns for branch
6560         conversion.
6561         ("return_use"): New expander.
6562         ("*return"): Emit return via thunk and rename it to ...
6563         ("*return<mode>"): ... this one.
6564         * config/s390/s390.opt: Add new options and and enum for the
6565         option values.
6566
6567 2018-02-08  Richard Sandiford  <richard.sandiford@linaro.org>
6568
6569         * lra-constraints.c (match_reload): Unconditionally use
6570         gen_lowpart_SUBREG, rather than selecting between that
6571         and equivalent gen_rtx_SUBREG code.
6572
6573 2018-02-08  Richard Biener  <rguenther@suse.de>
6574
6575         PR tree-optimization/84233
6576         * tree-ssa-phiprop.c (propagate_with_phi): Use separate
6577         changed flag instead of boguously re-using phi_inserted.
6578
6579 2018-02-08  Martin Jambor  <mjambor@suse.cz>
6580
6581         * hsa-gen.c (get_symbol_for_decl): Set program allocation for
6582         static local variables.
6583
6584 2018-02-08  Richard Biener  <rguenther@suse.de>
6585
6586         PR tree-optimization/84278
6587         * tree-vect-stmts.c (vectorizable_store): When looking for
6588         smaller vector types to perform grouped strided loads/stores
6589         make sure the mode is supported by the target.
6590         (vectorizable_load): Likewise.
6591
6592 2018-02-08  Wilco Dijkstra  <wdijkstr@arm.com>
6593
6594         * config/aarch64/aarch64.c (aarch64_components_for_bb):
6595         Increase LDP/STP opportunities by adding adjacent callee-saves.
6596
6597 2018-02-08  Wilco Dijkstra  <wdijkstr@arm.com>
6598
6599         PR rtl-optimization/84068
6600         PR rtl-optimization/83459
6601         * haifa-sched.c (rank_for_schedule): Fix SCHED_PRESSURE_MODEL sorting.
6602
6603 2018-02-08  Aldy Hernandez  <aldyh@redhat.com>
6604
6605         PR tree-optimization/84224
6606         * gimple-ssa-warn-alloca.c (pass_walloca::execute): Remove assert.
6607         * calls.c (gimple_alloca_call_p): Only return TRUE when we have
6608         non-zero arguments.
6609
6610 2018-02-07  Iain Sandoe  <iain@codesourcery.com>
6611
6612         PR target/84113
6613         * config/rs6000/altivec.md (*restore_world): Remove LR use.
6614         * config/rs6000/predicates.md (restore_world_operation): Adjust op
6615         count, remove one USE.
6616
6617 2018-02-07  Michael Meissner  <meissner@linux.vnet.ibm.com>
6618
6619         * doc/install.texi (Configuration): Document the
6620         --with-long-double-format={ibm,ieee} PowerPC configuration
6621         options.
6622
6623         PR target/84154
6624         * config/rs6000/rs6000.md (fix_trunc<SFDF:mode><QHI:mode>2):
6625         Convert from define_expand to be define_insn_and_split.  Rework
6626         float/double/_Float128 conversions to QI/HI/SImode to work with
6627         both ISA 2.07 (power8) or ISA 3.0 (power9).  Fix regression where
6628         conversions to QI/HImode types did a store and then a load to
6629         truncate the value.  For conversions to VSX registers, don't split
6630         the insn, instead emit the code directly.  Use the code iterator
6631         any_fix to combine signed and unsigned conversions.
6632         (fix<uns>_trunc<SFDF:mode>si2_p8): Likewise.
6633         (fixuns_trunc<SFDF:mode><QHI:mode>2): Likewise.
6634         (fix_trunc<IEEE128:mode><QHI:mode>2): Likewise.
6635         (fix<uns>_trunc<SFDF:mode><QHI:mode>2): Likewise.
6636         (fix_<mode>di2_hw): Likewise.
6637         (fixuns_<mode>di2_hw): Likewise.
6638         (fix_<mode>si2_hw): Likewise.
6639         (fixuns_<mode>si2_hw): Likewise.
6640         (fix<uns>_<IEEE128:mode><SDI:mode>2_hw): Likewise.
6641         (fix<uns>_trunc<IEEE128:mode><QHI:mode>2): Likewise.
6642         (fctiw<u>z_<mode>_smallint): Rename fctiw<u>z_<mode>_smallint to
6643         fix<uns>_trunc<SFDF:mode>si2_p8.
6644         (fix_trunc<SFDF:mode><QHI:mode>2_internal): Delete, no longer
6645         used.
6646         (fixuns_trunc<SFDF:mode><QHI:mode>2_internal): Likewise.
6647         (fix<uns>_<mode>_mem): Likewise.
6648         (fctiw<u>z_<mode>_mem): Likewise.
6649         (fix<uns>_<mode>_mem): Likewise.
6650         (fix<uns>_trunc<SFDF:mode><QHSI:mode>2_mem): On ISA 3.0, prevent
6651         the register allocator from doing a direct move to the GPRs to do
6652         a store, and instead use the ISA 3.0 store byte/half-word from
6653         vector register instruction.  For IEEE 128-bit floating point,
6654         also optimize stores of 32-bit ints.
6655         (fix<uns>_trunc<IEEE128:mode><QHSI:mode>2_mem): Likewise.
6656
6657 2018-02-07  Alan Hayward  <alan.hayward@arm.com>
6658
6659         * genextract.c (push_pathstr_operand): New function to support
6660         [a-zA-Z].
6661         (walk_rtx): Call push_pathstr_operand.
6662         (print_path): Support [a-zA-Z].
6663
6664 2018-02-07  Richard Biener  <rguenther@suse.de>
6665
6666         PR tree-optimization/84037
6667         * tree-vectorizer.h (struct _loop_vec_info): Add ivexpr_map member.
6668         (cse_and_gimplify_to_preheader): Declare.
6669         (vect_get_place_in_interleaving_chain): Likewise.
6670         * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
6671         ivexpr_map.
6672         (_loop_vec_info::~_loop_vec_info): Delete it.
6673         (cse_and_gimplify_to_preheader): New function.
6674         * tree-vect-slp.c (vect_get_place_in_interleaving_chain): Export.
6675         * tree-vect-stmts.c (vectorizable_store): CSE base and steps.
6676         (vectorizable_load): Likewise.  For grouped stores always base
6677         the IV on the first element.
6678         * tree-vect-loop-manip.c (vect_loop_versioning): Unshare versioning
6679         condition before gimplifying.
6680
6681 2018-02-07  Jakub Jelinek  <jakub@redhat.com>
6682
6683         * tree-eh.c (operation_could_trap_helper_p): Ignore honor_trapv for
6684         *DIV_EXPR and *MOD_EXPR.
6685
6686 2018-02-07  H.J. Lu  <hongjiu.lu@intel.com>
6687
6688         PR target/84248
6689         * config/i386/i386.c (ix86_option_override_internal): Mask out
6690         the CF_SET bit when checking -fcf-protection.
6691
6692 2018-02-07  Tom de Vries  <tom@codesourcery.com>
6693
6694         PR libgomp/84217
6695         * omp-expand.c (expand_oacc_collapse_init): Ensure diff_type is large
6696         enough.
6697
6698 2018-02-07  Richard Biener  <rguenther@suse.de>
6699
6700         PR tree-optimization/84204
6701         * tree-chrec.c (chrec_fold_plus_1): Remove size limiting in
6702         this place.
6703
6704         PR tree-optimization/84205
6705         * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Also
6706         special-case isl_ast_op_zdiv_r.
6707
6708         PR tree-optimization/84223
6709         * graphite-scop-detection.c (gather_bbs::before_dom_children):
6710         Only add conditions from within the region.
6711         (gather_bbs::after_dom_children): Adjust.
6712
6713 2018-02-07  Georg-Johann Lay  <avr@gjlay.de>
6714
6715         PR target/84209
6716         * config/avr/avr.h (GENERAL_REGNO_P, GENERAL_REG_P): New macros.
6717         * config/avr/avr.md: Only post-reload split REG-REG moves if
6718         either register is GENERAL_REG_P.
6719
6720 2018-02-07  Jakub Jelinek  <jakub@redhat.com>
6721
6722         PR tree-optimization/84235
6723         * tree-ssa-scopedtables.c
6724         (avail_exprs_stack::simplify_binary_operation): Fir MINUS_EXPR, punt
6725         if the subtraction is performed in floating point type where NaNs are
6726         honored.  For *DIV_EXPR, punt for ALL_FRACT_MODE_Ps where we can't
6727         build 1.  Formatting fix.
6728
6729 2018-02-06  Jakub Jelinek  <jakub@redhat.com>
6730
6731         PR target/84146
6732         * config/i386/i386.c (rest_of_insert_endbranch): Only skip
6733         NOTE_INSN_CALL_ARG_LOCATION after a call, not anything else,
6734         and skip it regardless of bb boundaries.  Use CALL_P macro,
6735         don't test INSN_P (insn) together with CALL_P or JUMP_P check
6736         unnecessarily, formatting fix.
6737
6738 2018-02-06  Michael Collison  <michael.collison@arm.com>
6739
6740         * config/arm/thumb2.md:
6741         (*thumb2_mov_negscc): Split only if TARGET_THUMB2 && !arm_restrict_it.
6742         (*thumb_mov_notscc): Ditto.
6743
6744 2018-02-06  Michael Meissner  <meissner@linux.vnet.ibm.com>
6745
6746         PR target/84154
6747         * config/rs6000/rs6000.md (su code attribute): Use "u" for
6748         unsigned_fix, not "s".
6749
6750 2018-02-06  Rainer Orth  <ro@CeBiTec.Uni-Bielefeld.DE>
6751
6752         * configure.ac (gcc_fn_eh_frame_ro): New function.
6753         (gcc_cv_as_cfi_directive): Check both 32 and 64-bit assembler for
6754         correct .eh_frame permissions.
6755         * configure: Regenerate.
6756
6757 2018-02-06  Andrew Jenner  <andrew@codeourcery.com>
6758
6759         * doc/invoke.texi: Add section for the PowerPC SPE backend. Remove
6760         irrelevant options.
6761
6762 2018-02-06  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
6763
6764         * config/rs6000/rs6000.c (rs6000_option_override_internal):
6765         Display warning message for -mno-speculate-indirect-jumps.
6766
6767 2018-02-06  Andrew Jenner  <andrew@codesourcery.com>
6768
6769         * config/powerpcspe/powerpcspe.opt: (msimple-fpu, mfpu) Add
6770         Undocumented.
6771         * config/powerpcspe/sysv4.opt (mbit-align): Likewise.
6772
6773 2018-02-06  Aldy Hernandez  <aldyh@redhat.com>
6774
6775         PR tree-optimization/84225
6776         * tree-eh.c (find_trapping_overflow): Only call
6777         operation_no_trapping_overflow when ANY_INTEGRAL_TYPE_P.
6778
6779 2018-02-06  Igor Tsimbalist  <igor.v.tsimbalist@intel.com>
6780
6781         PR target/84145
6782         * config/i386/i386.c: Reimplement the check of possible options
6783         -mibt/-mshstk conbination. Change error messages.
6784         * doc/invoke.texi: Fix a typo: remove extra '='.
6785
6786 2018-02-06  Marek Polacek  <polacek@redhat.com>
6787
6788         PR tree-optimization/84228
6789         * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Skip debug statements.
6790
6791 2018-02-06  Tamar Christina  <tamar.christina@arm.com>
6792
6793         PR target/82641
6794         * config/arm/arm.c (arm_print_asm_arch_directives): Record already
6795         emitted arch directives.
6796         * config/arm/arm-c.c (arm_cpu_builtins): Undefine __ARM_ARCH and
6797         __ARM_FEATURE_COPROC before changing architectures.
6798
6799 2018-02-06  Richard Biener  <rguenther@suse.de>
6800
6801         * config/i386/i386.c (print_reg): Fix typo.
6802         (ix86_loop_unroll_adjust): Do not unroll beyond the original nunroll.
6803
6804 2018-02-06  Eric Botcazou  <ebotcazou@adacore.com>
6805
6806         * configure: Regenerate.
6807
6808 2018-02-05  Martin Sebor  <msebor@redhat.com>
6809
6810         PR tree-optimization/83369
6811         * tree-ssa-ccp.c (pass_post_ipa_warn::execute): Use %G to print
6812         inlining context.
6813
6814 2018-02-05  Martin Liska  <mliska@suse.cz>
6815
6816         * doc/invoke.texi: Cherry-pick upstream r323995.
6817
6818 2018-02-05  Richard Sandiford  <richard.sandiford@linaro.org>
6819
6820         * ira.c (ira_init_register_move_cost): Adjust comment.
6821
6822 2018-02-05  Martin Liska  <mliska@suse.cz>
6823
6824         PR gcov-profile/84137
6825         * doc/gcov.texi: Fix typo in documentation.
6826
6827 2018-02-05  Martin Liska  <mliska@suse.cz>
6828
6829         PR gcov-profile/83879
6830         * doc/gcov.texi: Document necessity of --dynamic-list-data when
6831         using dlopen functionality.
6832
6833 2018-02-05  Olga Makhotina  <olga.makhotina@intel.com>
6834
6835         * config/i386/avx512dqintrin.h (_mm_mask_range_sd, _mm_maskz_range_sd,
6836         _mm_mask_range_round_sd, _mm_maskz_range_round_sd, _mm_mask_range_ss,
6837         _mm_maskz_range_ss, _mm_mask_range_round_ss,
6838         _mm_maskz_range_round_ss): New intrinsics.
6839         (__builtin_ia32_rangesd128_round)
6840         (__builtin_ia32_rangess128_round): Remove.
6841         (__builtin_ia32_rangesd128_mask_round,
6842         __builtin_ia32_rangess128_mask_round): New builtins.
6843         * config/i386/i386-builtin.def (__builtin_ia32_rangesd128_round,
6844         __builtin_ia32_rangess128_round): Remove.
6845         (__builtin_ia32_rangesd128_mask_round,
6846         __builtin_ia32_rangess128_mask_round): New builtins.
6847         * config/i386/sse.md (ranges<mode><round_saeonly_name>): Renamed to ...
6848         (ranges<mode><mask_scalar_name><round_saeonly_scalar_name>): ... this.
6849         ((match_operand:VF_128 2 "<round_saeonly_nimm_predicate>"
6850         "<round_saeonly_constraint>")): Changed to ...
6851         ((match_operand:VF_128 2 "<round_saeonly_scalar_nimm_predicate>"
6852         "<round_saeonly_scalar_constraint>")): ... this.
6853         ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_op4>%2, %1, %0|
6854         %0, %1, %2<round_saeonly_op4>, %3}"): Changed to ...
6855         ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_scalar_mask_op4>%2,
6856         %1, %0<mask_scalar_operand4>|%0<mask_scalar_operand4>, %1,
6857         %2<round_saeonly_scalar_mask_op4>, %3}"): ... this.
6858
6859 2018-02-02  Andrew Jenner  <andrew@codesourcery.com>
6860
6861         * config/powerpcspe/powerpcspe.opt: Add Undocumented to irrelevant
6862         options.
6863         * config/powerpcspe/powerpcspe-tables.opt (rs6000_cpu_opt_value):
6864         Remove all values except native, 8540 and 8548.
6865
6866 2018-02-02  H.J. Lu  <hongjiu.lu@intel.com>
6867
6868         * config/i386/i386.c (ix86_output_function_return): Pass
6869         INVALID_REGNUM, instead of -1, as invalid register number to
6870         indirect_thunk_name and output_indirect_thunk.
6871
6872 2018-02-02  Julia Koval  <julia.koval@intel.com>
6873
6874         * config.gcc: Add -march=icelake.
6875         * config/i386/driver-i386.c (host_detect_local_cpu): Detect icelake.
6876         * config/i386/i386-c.c (ix86_target_macros_internal): Handle icelake.
6877         * config/i386/i386.c (processor_costs): Add m_ICELAKE.
6878         (PTA_ICELAKE, PTA_AVX512VNNI, PTA_GFNI, PTA_VAES, PTA_AVX512VBMI2,
6879         PTA_VPCLMULQDQ, PTA_RDPID, PTA_AVX512BITALG): New.
6880         (processor_target_table): Add icelake.
6881         (ix86_option_override_internal): Handle new PTAs.
6882         (get_builtin_code_for_version): Handle icelake.
6883         (M_INTEL_COREI7_ICELAKE): New.
6884         (fold_builtin_cpu): Handle icelake.
6885         * config/i386/i386.h (TARGET_ICELAKE, PROCESSOR_ICELAKE): New.
6886         * doc/invoke.texi: Add -march=icelake.
6887
6888 2018-02-02  Julia Koval  <julia.koval@intel.com>
6889
6890         * config/i386/i386.c (ix86_option_override_internal): Change flags type
6891         to wide_int_bitmask.
6892         * wide-int-bitmask.h: New.
6893
6894 2018-02-02  Igor Tsimbalist  <igor.v.tsimbalist@intel.com>
6895
6896         PR target/84066
6897         * config/i386/i386.md: Replace Pmode with word_mode in
6898         builtin_setjmp_setup and builtin_longjmp to support x32.
6899
6900 2018-02-01  Peter Bergner  <bergner@vnet.ibm.com>
6901
6902         PR target/56010
6903         PR target/83743
6904         * config/rs6000/driver-rs6000.c: #include "diagnostic.h".
6905         #include "opts.h".
6906         (rs6000_supported_cpu_names): New static variable.
6907         (linux_cpu_translation_table): Likewise.
6908         (elf_platform) <cpu>: Define new static variable and use it.
6909         Translate kernel AT_PLATFORM name to canonical name if needed.
6910         Error if platform name is unknown.
6911
6912 2018-02-01  Aldy Hernandez  <aldyh@redhat.com>
6913
6914         PR target/84089
6915         * config/pa/predicates.md (base14_operand): Handle E_VOIDmode.
6916
6917 2018-02-01  Jeff Law  <law@redhat.com>
6918
6919         PR target/84128
6920         * config/i386/i386.c (release_scratch_register_on_entry): Add new
6921         OFFSET and RELEASE_VIA_POP arguments.  Use SP+OFFSET to restore
6922         the scratch if RELEASE_VIA_POP is false.
6923         (ix86_adjust_stack_and_probe_stack_clash): Un-constify SIZE.
6924         If we have to save a temporary register, decrement SIZE appropriately.
6925         Pass new arguments to release_scratch_register_on_entry.
6926         (ix86_adjust_stack_and_probe): Likewise.
6927         (ix86_emit_probe_stack_range): Pass new arguments to
6928         release_scratch_register_on_entry.
6929
6930 2018-02-01  Uros Bizjak  <ubizjak@gmail.com>
6931
6932         PR rtl-optimization/84157
6933         * combine.c (change_zero_ext): Use REG_P predicate in
6934         front of HARD_REGISTER_P predicate.
6935
6936 2018-02-01  Georg-Johann Lay  <avr@gjlay.de>
6937
6938         * config/avr/avr.c (avr_option_override): Move disabling of
6939         -fdelete-null-pointer-checks to...
6940         * common/config/avr/avr-common.c (avr_option_optimization_table):
6941         ...here.
6942
6943 2018-02-01  Richard Sandiford  <richard.sandiford@linaro.org>
6944
6945         PR tree-optimization/81635
6946         * tree-data-ref.c (split_constant_offset_1): For types that
6947         wrap on overflow, try to use range info to prove that wrapping
6948         cannot occur.
6949
6950 2018-02-01  Renlin Li  <renlin.li@arm.com>
6951
6952         PR target/83370
6953         * config/aarch64/aarch64.c (aarch64_class_max_nregs): Handle
6954         TAILCALL_ADDR_REGS.
6955         (aarch64_register_move_cost): Likewise.
6956         * config/aarch64/aarch64.h (reg_class): Rename CALLER_SAVE_REGS to
6957         TAILCALL_ADDR_REGS.
6958         (REG_CLASS_NAMES): Likewise.
6959         (REG_CLASS_CONTENTS): Rename CALLER_SAVE_REGS to
6960         TAILCALL_ADDR_REGS. Remove IP registers.
6961         * config/aarch64/aarch64.md (Ucs): Update register constraint.
6962
6963 2018-02-01  Richard Biener  <rguenther@suse.de>
6964
6965         * domwalk.h (dom_walker::dom_walker): Add additional constructor
6966         for specifying RPO order and allow NULL for that.
6967         * domwalk.c (dom_walker::dom_walker): Likewise.
6968         (dom_walker::walk): Handle NULL RPO order.
6969         * tree-into-ssa.c (rewrite_dom_walker): Do not walk dom children
6970         in RPO order.
6971         (rewrite_update_dom_walker): Likewise.
6972         (mark_def_dom_walker): Likewise.
6973
6974 2018-02-01  Richard Sandiford  <richard.sandiford@linaro.org>
6975
6976         * config/aarch64/aarch64-protos.h (aarch64_split_sve_subreg_move)
6977         (aarch64_maybe_expand_sve_subreg_move): Declare.
6978         * config/aarch64/aarch64.md (UNSPEC_REV_SUBREG): New unspec.
6979         * config/aarch64/predicates.md (aarch64_any_register_operand): New
6980         predicate.
6981         * config/aarch64/aarch64-sve.md (mov<mode>): Optimize subreg moves
6982         that are semantically a reverse operation.
6983         (*aarch64_sve_mov<mode>_subreg_be): New pattern.
6984         * config/aarch64/aarch64.c (aarch64_maybe_expand_sve_subreg_move):
6985         (aarch64_replace_reg_mode, aarch64_split_sve_subreg_move): New
6986         functions.
6987         (aarch64_can_change_mode_class): For big-endian, forbid changes
6988         between two SVE modes if they have different element sizes.
6989
6990 2018-02-01  Richard Sandiford  <richard.sandiford@linaro.org>
6991
6992         * config/aarch64/aarch64.c (aarch64_expand_sve_const_vector): Prefer
6993         the TImode handling for big-endian targets.
6994
6995 2018-02-01  Richard Sandiford  <richard.sandiford@linaro.org>
6996
6997         * config/aarch64/aarch64-sve.md (sve_ld1rq): Replace with...
6998         (*sve_ld1rq<Vesize>): ... this new pattern.  Handle all element sizes,
6999         not just bytes.
7000         * config/aarch64/aarch64.c (aarch64_expand_sve_widened_duplicate):
7001         Remove BSWAP handing for big-endian targets and use the form of
7002         LD1RQ appropariate for the mode.
7003
7004 2018-02-01  Richard Sandiford  <richard.sandiford@linaro.org>
7005
7006         * config/aarch64/aarch64.c (aarch64_simd_valid_immediate): Handle
7007         all CONST_VECTOR_DUPLICATE_P vectors, not just those with a single
7008         duplicated element.
7009
7010 2018-02-01  Richard Sandiford  <richard.sandiford@linaro.org>
7011
7012         PR tearget/83845
7013         * config/aarch64/aarch64.c (aarch64_secondary_reload): Tighten
7014         check for operands that need to go through aarch64_sve_reload_be.
7015
7016 2018-02-01  Jakub Jelinek  <jakub@redhat.com>
7017
7018         PR tree-optimization/81661
7019         PR tree-optimization/84117
7020         * tree-eh.h (rewrite_to_non_trapping_overflow): Declare.
7021         * tree-eh.c: Include gimplify.h.
7022         (find_trapping_overflow, replace_trapping_overflow,
7023         rewrite_to_non_trapping_overflow): New functions.
7024         * tree-vect-loop.c: Include tree-eh.h.
7025         (vect_get_loop_niters): Use rewrite_to_non_trapping_overflow.
7026         * tree-data-ref.c: Include tree-eh.h.
7027         (get_segment_min_max): Use rewrite_to_non_trapping_overflow.
7028
7029 2018-01-31  Uros Bizjak  <ubizjak@gmail.com>
7030
7031         PR rtl-optimization/84123
7032         * combine.c (change_zero_ext): Check if hard register satisfies
7033         can_change_dest_mode before calling gen_lowpart_SUBREG.
7034
7035 2018-01-31  Vladimir Makarov  <vmakarov@redhat.com>
7036
7037         PR target/82444
7038         * ira.c (ira_init_register_move_cost): Remove assert.
7039
7040 2018-01-31  Eric Botcazou  <ebotcazou@adacore.com>
7041
7042         PR rtl-optimization/84071
7043         * doc/tm.texi.in (WORD_REGISTER_OPERATIONS): Add explicit case.
7044         * doc/tm.texi: Regenerate.
7045
7046 2018-01-31  Richard Biener  <rguenther@suse.de>
7047
7048         PR tree-optimization/84132
7049         * tree-data-ref.c (analyze_miv_subscript): Properly
7050         check whether evolution_function_is_affine_multivariate_p
7051         before calling gcd_of_steps_may_divide_p.
7052
7053 2018-01-31  Julia Koval  <julia.koval@intel.com>
7054
7055         PR target/83618
7056         * config/i386/i386.c (ix86_expand_builtin): Handle IX86_BUILTIN_RDPID.
7057         * config/i386/i386.md (rdpid_rex64) New.
7058         (rdpid): Make 32bit only.
7059
7060 2018-01-29  Aldy Hernandez  <aldyh@redhat.com>
7061
7062         PR lto/84105
7063         * tree-pretty-print.c (dump_generic_node): Handle a TYPE_NAME with
7064         an IDENTIFIER_NODE for FUNCTION_TYPE's.
7065
7066 2018-01-31  Eric Botcazou  <ebotcazou@adacore.com>
7067
7068         Revert
7069         2018-01-12  Eric Botcazou  <ebotcazou@adacore.com>
7070
7071         * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
7072
7073 2018-01-31  Eric Botcazou  <ebotcazou@adacore.com>
7074
7075         PR rtl-optimization/84071
7076         * combine.c (record_dead_and_set_regs_1): Record the source unmodified
7077         for a paradoxical SUBREG on a WORD_REGISTER_OPERATIONS target.
7078
7079 2018-01-31  Claudiu Zissulescu  <claziss@synopsys.com>
7080
7081         * config/arc/arc.c (arc_handle_aux_attribute): New function.
7082         (arc_attribute_table): Add 'aux' attribute.
7083         (arc_in_small_data_p): Consider aux like variables.
7084         (arc_is_aux_reg_p): New function.
7085         (arc_asm_output_aligned_decl_local): Ignore 'aux' like variables.
7086         (arc_get_aux_arg): New function.
7087         (prepare_move_operands): Handle aux-register access.
7088         (arc_handle_aux_attribute): New function.
7089         * doc/extend.texi (ARC Variable attributes): Add subsection.
7090
7091 2018-01-31  Claudiu Zissulescu  <claziss@synopsys.com>
7092
7093         * config/arc/arc-protos.h (arc_is_uncached_mem_p): Function proto.
7094         * config/arc/arc.c (arc_handle_uncached_attribute): New function.
7095         (arc_attribute_table): Add 'uncached' attribute.
7096         (arc_print_operand): Print '.di' flag for uncached memory
7097         accesses.
7098         (arc_in_small_data_p): Do not consider for small data the uncached
7099         types.
7100         (arc_is_uncached_mem_p): New function.
7101         * config/arc/predicates.md (compact_store_memory_operand): Check
7102         for uncached memory accesses.
7103         (nonvol_nonimm_operand): Likewise.
7104         * gcc/doc/extend.texi (ARC Type Attribute): New subsection.
7105
7106 2018-01-31  Jakub Jelinek  <jakub@redhat.com>
7107
7108         PR c/84100
7109         * common.opt (falign-functions=, falign-jumps=, falign-labels=,
7110         falign-loops=): Add Optimization flag.
7111
7112 2018-01-30  Jeff Law  <law@redhat.com>
7113
7114         PR target/84064
7115         * i386.c (ix86_adjust_stack_and_probe_stack_clash): New argument
7116         INT_REGISTERS_SAVED.  Check it prior to calling
7117         get_scratch_register_on_entry.
7118         (ix86_adjust_stack_and_probe): Similarly.
7119         (ix86_emit_probe_stack_range): Similarly.
7120         (ix86_expand_prologue): Corresponding changes.
7121
7122 2018-01-30  Rainer Orth  <ro@CeBiTec.Uni-Bielefeld.DE>
7123
7124         PR target/40411
7125         * config/sol2.h (STARTFILE_ARCH_SPEC): Use -std=c*,
7126         -std=iso9899:199409 instead of -pedantic to select values-Xc.o.
7127
7128 2018-01-30  Vladimir Makarov  <vmakarov@redhat.com>
7129
7130         PR target/84112
7131         * lra-constraints.c (curr_insn_transform): Process AND in the
7132         address.
7133
7134 2018-01-30  Jakub Jelinek  <jakub@redhat.com>
7135
7136         PR rtl-optimization/83986
7137         * sched-deps.c (sched_analyze_insn): For frame related insns, add anti
7138         dependence against last_pending_memory_flush in addition to
7139         pending_jump_insns.
7140
7141 2018-01-30  Alexandre Oliva  <aoliva@redhat.com>
7142
7143         PR tree-optimization/81611
7144         * tree-ssa-dom.c (simple_iv_increment_p): Skip intervening
7145         copies.
7146
7147 2018-01-30  Aaron Sawdey  <acsawdey@linux.vnet.ibm.com>
7148
7149         PR target/83758
7150         * config/rs6000/rs6000.c (rs6000_internal_arg_pointer): Only return
7151         a reg rtx.
7152
7153 2018-01-30  Richard Biener  <rguenther@suse.de>
7154             Jakub Jelinek  <jakub@redhat.com>
7155
7156         PR tree-optimization/84111
7157         * tree-ssa-loop-ivcanon.c (tree_unroll_loops_completely_1): Skip
7158         inner loops added during recursion, as they don't have up-to-date
7159         SSA form.
7160
7161 2018-01-30  Jan Hubicka  <hubicka@ucw.cz>
7162
7163         PR ipa/81360
7164         * ipa-inline.c (can_inline_edge_p): Break out late tests to...
7165         (can_inline_edge_by_limits_p): ... here.
7166         (can_early_inline_edge_p, check_callers,
7167         update_caller_keys, update_callee_keys, recursive_inlining,
7168         add_new_edges_to_heap, speculation_useful_p,
7169         inline_small_functions,
7170         inline_small_functions, flatten_function,
7171         inline_to_all_callers_1): Update.
7172
7173 2018-01-30  Jan Hubicka  <hubicka@ucw.cz>
7174
7175         * profile-count.c (profile_count::combine_with_ipa_count): Handle
7176         zeros correctly.
7177
7178 2018-01-30  Richard Biener  <rguenther@suse.de>
7179
7180         PR tree-optimization/83008
7181         * tree-vect-slp.c (vect_analyze_slp_cost_1): Properly cost
7182         invariant and constant vector uses in stmts when they need
7183         more than one stmt.
7184
7185 2018-01-30  Rainer Orth  <ro@CeBiTec.Uni-Bielefeld.DE>
7186
7187         PR bootstrap/84017
7188         * configure.ac (gcc_cv_as_shf_merge): Disable on Solaris 10/x86.
7189         * configure: Regenerate.
7190
7191 2018-01-30  Richard Sandiford  <richard.sandiford@linaro.org>
7192
7193         * config/aarch64/aarch64-sve.md (*vec_extract<mode><Vel>_0): New
7194         pattern.
7195         (*vec_extract<mode><Vel>_v128): Require a nonzero lane number.
7196         Use gen_rtx_REG rather than gen_lowpart.
7197
7198 2018-01-30  Richard Sandiford  <richard.sandiford@linaro.org>
7199
7200         * lra-constraints.c (match_reload): Use subreg_lowpart_offset
7201         rather than 0 when creating partial subregs.
7202
7203 2018-01-30  Richard Sandiford  <richard.sandiford@linaro.org>
7204
7205         * vec-perm-indices.c (vec_perm_indices::series_p): Give examples
7206         of usage.
7207
7208 2018-01-29  Michael Meissner  <meissner@linux.vnet.ibm.com>
7209
7210         PR target/81550
7211         * config/rs6000/rs6000.c (rs6000_setup_reg_addr_masks): If DFmode
7212         and SFmode can go in Altivec registers (-mcpu=power7 for DFmode,
7213         -mcpu=power8 for SFmode) don't set the PRE_INCDEC or PRE_MODIFY
7214         flags.  This restores the settings used before the 2017-07-24.
7215         Turning off pre increment/decrement/modify allows IVOPTS to
7216         optimize DF/SF loops where the index is an int.
7217
7218 2018-01-29  Richard Biener  <rguenther@suse.de>
7219             Kelvin Nilsen  <kelvin@gcc.gnu.org>
7220
7221         PR bootstrap/80867
7222         * tree-vect-stmts.c (vectorizable_call): Don't call
7223         targetm.vectorize_builtin_md_vectorized_function if callee is
7224         NULL.
7225
7226 2018-01-22  Carl Love  <cel@us.ibm.com>
7227
7228         * doc/extend.tex: Fix typo in second arg in
7229         __builtin_bcdadd_{lt|eq|gt|ov} and __builtin_bcdsub_{lt|eq|gt|ov}.
7230
7231 2018-01-29  Richard Biener  <rguenther@suse.de>
7232
7233         PR tree-optimization/84086
7234         * tree-ssanames.c: Include cfgloop.h and tree-scalar-evolution.h.
7235         (flush_ssaname_freelist): When SSA names were released reset
7236         the SCEV hash table.
7237
7238 2018-01-29  Richard Biener  <rguenther@suse.de>
7239
7240         PR tree-optimization/84057
7241         * tree-ssa-loop-ivcanon.c (unloop_loops): Deal with already
7242         removed paths when removing edges.
7243
7244 2018-01-27  H.J. Lu  <hongjiu.lu@intel.com>
7245
7246         * doc/invoke.texi: Replace -mfunction-return==@var{choice} with
7247         -mfunction-return=@var{choice}.
7248
7249 2018-01-27  Bernd Edlinger  <bernd.edlinger@hotmail.de>
7250
7251         PR diagnostic/84034
7252         * diagnostic-show-locus.c (get_line_width_without_trailing_whitespace):
7253         Handle CR like TAB.
7254         (layout::print_source_line): Likewise.
7255         (test_get_line_width_without_trailing_whitespace): Add test cases.
7256
7257 2018-01-27  Jakub Jelinek  <jakub@redhat.com>
7258
7259         PR middle-end/84040
7260         * sched-deps.c (sched_macro_fuse_insns): Return immediately for
7261         debug insns.
7262
7263 2018-01-26  Jim Wilson  <jimw@sifive.com>
7264
7265         * config/riscv/riscv.h (MAX_FIXED_MODE_SIZE): New.
7266
7267         * config/riscv/elf.h (LIB_SPEC): Don't include -lgloss when nosys.specs
7268         specified.
7269
7270 2018-01-26  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
7271
7272         * config/aarch64/aarch64.md: Add peepholes for CMP + SUB -> SUBS
7273         and CMP + SUB-immediate -> SUBS.
7274
7275 2018-01-26  Martin Sebor  <msebor@redhat.com>
7276
7277         PR tree-optimization/83896
7278         * tree-ssa-strlen.c (get_string_len): Rename...
7279         (get_string_cst_length): ...to this.  Return HOST_WIDE_INT.
7280         Avoid assuming length is constant.
7281         (handle_char_store): Use HOST_WIDE_INT for string length.
7282
7283 2018-01-26  Uros Bizjak  <ubizjak@gmail.com>
7284
7285         PR target/81763
7286         * config/i386/i386.md (*andndi3_doubleword): Add earlyclobber
7287         to (=&r,r,rm) alternative. Add (=r,0,rm) and (=r,r,0) alternatives.
7288
7289 2018-01-26  Richard Biener  <rguenther@suse.de>
7290
7291         PR rtl-optimization/84003
7292         * dse.c (record_store): Only record redundant stores when
7293         the earlier store aliases at least all accesses the later one does.
7294
7295 2018-01-26  Jakub Jelinek  <jakub@redhat.com>
7296
7297         PR rtl-optimization/83985
7298         * dce.c (deletable_insn_p): Return false for separate shrink wrapping
7299         REG_CFA_RESTORE insns.
7300         (delete_unmarked_insns): Don't ignore separate shrink wrapping
7301         REG_CFA_RESTORE insns here.
7302
7303         PR c/83989
7304         * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Don't
7305         use SSA_NAME_VAR as base for SSA_NAMEs with non-NULL SSA_NAME_VAR.
7306
7307 2018-01-26  Claudiu Zissulescu  <claziss@synopsys.com>
7308
7309         * config/arc/arc-arch.h (arc_tune_attr): Add ARC_TUNE_CORE_3.
7310         * config/arc/arc.c (arc_sched_issue_rate): Use ARC_TUNE_... .
7311         (arc_init): Likewise.
7312         (arc_override_options): Likewise.
7313         (arc_file_start): Choose Tag_ARC_CPU_variation based on arc_tune
7314         value.
7315         (hwloop_fail): Use TARGET_DBNZ when we want to check for dbnz insn
7316         support.
7317         * config/arc/arc.h (TARGET_DBNZ): Define.
7318         * config/arc/arc.md (attr tune): Add core_3, use ARC_TUNE_... to
7319         properly set the tune attribute.
7320         (dbnz): Use TARGET_DBNZ guard.
7321         * config/arc/arc.opt (mtune): Add core3 option.
7322
7323 2018-01-26  Claudiu Zissulescu  <claziss@synopsys.com>
7324
7325         * config/arc/arc.c (arc_delegitimize_address_0): Refactored to
7326         recognize new pic like addresses.
7327         (arc_delegitimize_address): Clean up.
7328
7329 2018-01-26  Claudiu Zissulescu  <claziss@synopsys.com>
7330
7331         * config/arc/arc-arches.def: Option mrf16 valid for all
7332         architectures.
7333         * config/arc/arc-c.def (__ARC_RF16__): New predefined macro.
7334         * config/arc/arc-cpus.def (em_mini): New cpu with rf16 on.
7335         * config/arc/arc-options.def (FL_RF16): Add mrf16 option.
7336         * config/arc/arc-tables.opt: Regenerate.
7337         * config/arc/arc.c (arc_conditional_register_usage): Handle
7338         reduced register file case.
7339         (arc_file_start): Set must have build attributes.
7340         * config/arc/arc.h (MAX_ARC_PARM_REGS): Conditional define using
7341         mrf16 option value.
7342         * config/arc/arc.opt (mrf16): Add new option.
7343         * config/arc/elf.h (ATTRIBUTE_PCS): Define.
7344         * config/arc/genmultilib.awk: Handle new mrf16 option.
7345         * config/arc/linux.h (ATTRIBUTE_PCS): Define.
7346         * config/arc/t-multilib: Regenerate.
7347         * doc/invoke.texi (ARC Options): Document mrf16 option.
7348
7349 2018-01-26  Claudiu Zissulescu  <claziss@synopsys.com>
7350
7351         * config/arc/arc-protos.h: Add arc_is_secure_call_p proto.
7352         * config/arc/arc.c (arc_handle_secure_attribute): New function.
7353         (arc_attribute_table): Add 'secure_call' attribute.
7354         (arc_print_operand): Print secure call operand.
7355         (arc_function_ok_for_sibcall): Don't optimize tail calls when
7356         secure.
7357         (arc_is_secure_call_p): New function.  * config/arc/arc.md
7358         (call_i): Add support for sjli instruction.
7359         (call_value_i): Likewise.
7360         * config/arc/constraints.md (Csc): New constraint.
7361
7362 2018-01-26  Claudiu Zissulescu  <claziss@synopsys.com>
7363             John Eric Martin  <John.Martin@emmicro-us.com>
7364
7365         * config/arc/arc-protos.h: Add arc_is_jli_call_p proto.
7366         * config/arc/arc.c (_arc_jli_section): New struct.
7367         (arc_jli_section): New type.
7368         (rc_jli_sections): New static variable.
7369         (arc_handle_jli_attribute): New function.
7370         (arc_attribute_table): Add jli_always and jli_fixed attribute.
7371         (arc_file_end): New function.
7372         (TARGET_ASM_FILE_END): Define.
7373         (arc_print_operand): Reuse 'S' letter for JLI output instruction.
7374         (arc_add_jli_section): New function.
7375         (jli_call_scan): Likewise.
7376         (arc_reorg): Call jli_call_scan.
7377         (arc_output_addsi): Remove 'S' from printing asm operand.
7378         (arc_is_jli_call_p): New function.
7379         * config/arc/arc.md (movqi_insn): Remove 'S' from printing asm
7380         operand.
7381         (movhi_insn): Likewise.
7382         (movsi_insn): Likewise.
7383         (movsi_set_cc_insn): Likewise.
7384         (loadqi_update): Likewise.
7385         (load_zeroextendqisi_update): Likewise.
7386         (load_signextendqisi_update): Likewise.
7387         (loadhi_update): Likewise.
7388         (load_zeroextendhisi_update): Likewise.
7389         (load_signextendhisi_update): Likewise.
7390         (loadsi_update): Likewise.
7391         (loadsf_update): Likewise.
7392         (movsicc_insn): Likewise.
7393         (bset_insn): Likewise.
7394         (bxor_insn): Likewise.
7395         (bclr_insn): Likewise.
7396         (bmsk_insn): Likewise.
7397         (bicsi3_insn): Likewise.
7398         (cmpsi_cc_c_insn): Likewise.
7399         (movsi_ne): Likewise.
7400         (movsi_cond_exec): Likewise.
7401         (clrsbsi2): Likewise.
7402         (norm_f): Likewise.
7403         (normw): Likewise.
7404         (swap): Likewise.
7405         (divaw): Likewise.
7406         (flag): Likewise.
7407         (sr): Likewise.
7408         (kflag): Likewise.
7409         (ffs): Likewise.
7410         (ffs_f): Likewise.
7411         (fls): Likewise.
7412         (call_i): Remove 'S' asm letter, add jli instruction.
7413         (call_value_i): Likewise.
7414         * config/arc/arc.op (mjli-always): New option.
7415         * config/arc/constraints.md (Cji): New constraint.
7416         * config/arc/fpx.md (addsf3_fpx): Remove 'S' from printing asm
7417         operand.
7418         (subsf3_fpx): Likewise.
7419         (mulsf3_fpx): Likewise.
7420         * config/arc/simdext.md (vendrec_insn): Remove 'S' from printing
7421         asm operand.
7422         * doc/extend.texi (ARC): Document 'jli-always' and 'jli-fixed'
7423         function attrbutes.
7424         * doc/invoke.texi (ARC): Document mjli-always option.
7425
7426 2018-01-26  Sebastian Perta  <sebastian.perta@renesas.com>
7427
7428         * config/rl78/rl78.c (rl78_addsi3_internal): If operand 2 is const
7429         avoid addition with 0 and use incw and decw where possible.
7430
7431 2018-01-26  Richard Biener  <rguenther@suse.de>
7432
7433         PR tree-optimization/81082
7434         * fold-const.c (fold_plusminus_mult_expr): Do not perform the
7435         association if it requires casting to unsigned.
7436         * match.pd ((A * C) +- (B * C) -> (A+-B)): New patterns derived
7437         from fold_plusminus_mult_expr to catch important cases late when
7438         range info is available.
7439
7440 2018-01-26  Rainer Orth  <ro@CeBiTec.Uni-Bielefeld.DE>
7441
7442         * config/i386/sol2.h (USE_HIDDEN_LINKONCE): Remove.
7443         * configure.ac (hidden_linkonce): New test.
7444         * configure: Regenerate.
7445         * config.in: Regenerate.
7446
7447 2018-01-26  Julia Koval  <julia.koval@intel.com>
7448
7449         * config/i386/avx512bitalgintrin.h (_mm512_bitshuffle_epi64_mask,
7450         _mm512_mask_bitshuffle_epi64_mask, _mm256_bitshuffle_epi64_mask,
7451         _mm256_mask_bitshuffle_epi64_mask, _mm_bitshuffle_epi64_mask,
7452         _mm_mask_bitshuffle_epi64_mask): Fix type.
7453         * config/i386/i386-builtin-types.def (UHI_FTYPE_V2DI_V2DI_UHI,
7454         USI_FTYPE_V4DI_V4DI_USI): Remove.
7455         * config/i386/i386-builtin.def (__builtin_ia32_vpshufbitqmb512_mask,
7456         __builtin_ia32_vpshufbitqmb256_mask,
7457         __builtin_ia32_vpshufbitqmb128_mask): Fix types.
7458         * config/i386/i386.c (ix86_expand_args_builtin): Remove old types.
7459         * config/i386/sse.md (VI1_AVX512VLBW): Change types.
7460
7461 2018-01-26  Alan Modra  <amodra@gmail.com>
7462
7463         PR target/84033
7464         * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Exclude
7465         UNSPEC_VBPERMQ.  Sort other unspecs.
7466
7467 2018-01-25  David Edelsohn  <dje.gcc@gmail.com>
7468
7469         * doc/invoke.texi (PowerPC Options): Document 'native' cpu type.
7470
7471 2018-01-25  Jan Hubicka  <hubicka@ucw.cz>
7472
7473         PR middle-end/83055
7474         * predict.c (drop_profile): Do not push/pop cfun; update also
7475         node->count.
7476         (handle_missing_profiles): Fix logic looking for zero profiles.
7477
7478 2018-01-25  Jakub Jelinek  <jakub@redhat.com>
7479
7480         PR middle-end/83977
7481         * ipa-fnsummary.c (compute_fn_summary): Clear can_change_signature
7482         on functions with #pragma omp declare simd or functions with simd
7483         attribute.
7484         * omp-simd-clone.c (expand_simd_clones): Revert 2018-01-24 change.
7485         * config/i386/i386.c (ix86_simd_clone_compute_vecsize_and_simdlen):
7486         Remove trailing \n from warning_at calls.
7487
7488 2018-01-25  Tom de Vries  <tom@codesourcery.com>
7489
7490         PR target/84028
7491         * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
7492         for neutered workers.
7493
7494 2018-01-24  Joseph Myers  <joseph@codesourcery.com>
7495
7496         PR target/68467
7497         * config/m68k/m68k.c (m68k_promote_function_mode): New function.
7498         (TARGET_PROMOTE_FUNCTION_MODE): New macro.
7499
7500 2018-01-24  Jeff Law  <law@redhat.com>
7501
7502         PR target/83994
7503         * i386.c (get_probe_interval): Move to earlier point.
7504         (ix86_compute_frame_layout): If -fstack-clash-protection and
7505         the frame is larger than the probe interval, then use pushes
7506         to save registers rather than reg->mem moves.
7507         (ix86_expand_prologue): Remove conditional for int_registers_saved
7508         assertion.
7509
7510 2018-01-24  Vladimir Makarov  <vmakarov@redhat.com>
7511
7512         PR target/84014
7513         * ira-build.c (setup_min_max_allocno_live_range_point): Set up
7514         min/max for never referenced object.
7515
7516 2018-01-24  Jakub Jelinek  <jakub@redhat.com>
7517
7518         PR middle-end/83977
7519         * tree.c (free_lang_data_in_decl): Don't clear DECL_ABSTRACT_ORIGIN
7520         here.
7521         * omp-low.c (create_omp_child_function): Remove "omp declare simd"
7522         attributes from DECL_ATTRIBUTES (decl) without affecting
7523         DECL_ATTRIBUTES (current_function_decl).
7524         * omp-simd-clone.c (expand_simd_clones): Ignore DECL_ARTIFICIAL
7525         functions with non-NULL DECL_ABSTRACT_ORIGIN.
7526
7527 2018-01-24  Richard Sandiford  <richard.sandiford@linaro.org>
7528
7529         PR tree-optimization/83979
7530         * fold-const.c (fold_comparison): Use constant_boolean_node
7531         instead of boolean_{true,false}_node.
7532
7533 2018-01-24  Jan Hubicka  <hubicka@ucw.cz>
7534
7535         * ipa-profile.c (ipa_propagate_frequency_1): Fix logic skipping calls
7536         with zero counts.
7537
7538 2018-01-24  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
7539
7540         * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
7541         Simplify the clause that sets the length attribute.
7542         (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
7543         (*sibcall_nonlocal_sysv<mode>): Clean up code block; simplify the
7544         clause that sets the length attribute.
7545         (*sibcall_value_nonlocal_sysv<mode>): Likewise.
7546
7547 2018-01-24  Tom de Vries  <tom@codesourcery.com>
7548
7549         PR target/83589
7550         * config/nvptx/nvptx.c (WORKAROUND_PTXJIT_BUG_2): Define to 1.
7551         (nvptx_pc_set, nvptx_condjump_label): New function. Copy from jump.c.
7552         Add strict parameter.
7553         (prevent_branch_around_nothing): Insert dummy insn between branch to
7554         label and label with no ptx insn inbetween.
7555         * config/nvptx/nvptx.md (define_insn "fake_nop"): New insn.
7556
7557 2018-01-24  Tom de Vries  <tom@codesourcery.com>
7558
7559         PR target/81352
7560         * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
7561         for neutered threads in warp.
7562         * config/nvptx/nvptx.md (define_insn "exit"): New insn.
7563
7564 2018-01-24  Richard Biener  <rguenther@suse.de>
7565
7566         PR tree-optimization/83176
7567         * tree-chrec.c (chrec_fold_plus_1): Handle (signed T){(T) .. }
7568         operands.
7569
7570 2018-01-24  Richard Biener  <rguenther@suse.de>
7571
7572         PR tree-optimization/82819
7573         * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Avoid
7574         code generating pluses that are no-ops in the target precision.
7575
7576 2018-01-24  Richard Biener  <rguenther@suse.de>
7577
7578         PR middle-end/84000
7579         * tree-cfg.c (replace_loop_annotate): Handle annot_expr_parallel_kind.
7580
7581 2018-01-23  Jan Hubicka  <hubicka@ucw.cz>
7582
7583         * cfgcleanup.c (try_crossjump_to_edge): Use combine_with_count
7584         to merge probabilities.
7585         * predict.c (probably_never_executed): Also mark as cold functions
7586         with global 0 profile and guessed local profile.
7587         * profile-count.c (profile_probability::combine_with_count): New
7588         member function.
7589         * profile-count.h (profile_probability::operator*,
7590         profile_probability::operator*=, profile_probability::operator/,
7591         profile_probability::operator/=): Reduce precision to adjusted
7592         and set value to guessed on contradictory divisions.
7593         (profile_probability::combine_with_freq): Remove.
7594         (profile_probability::combine_wiht_count): Declare.
7595         (profile_count::force_nonzero):: Set to adjusted.
7596         (profile_count::probability_in):: Set quality to adjusted.
7597         * tree-ssa-tail-merge.c (replace_block_by): Use
7598         combine_with_count.
7599
7600 2018-01-23  Andrew Waterman  <andrew@sifive.com>
7601             Jim Wilson  <jimw@sifive.com>
7602
7603         * config/riscv/riscv.c (riscv_stack_boundary): New.
7604         (riscv_option_override): Set riscv_stack_boundary.  Handle
7605         riscv_preferred_stack_boundary_arg.
7606         * config/riscv/riscv.h (MIN_STACK_BOUNDARY, ABI_STACK_BOUNDARY): New.
7607         (BIGGEST_ALIGNMENT): Set to STACK_BOUNDARY.
7608         (STACK_BOUNDARY): Set to riscv_stack_boundary.
7609         (RISCV_STACK_ALIGN): Use STACK_BOUNDARY.
7610         * config/riscv/riscv.opt (mpreferred-stack-boundary): New.
7611         * doc/invoke.tex (RISC-V Options): Add -mpreferred-stack-boundary.
7612
7613 2018-01-23  H.J. Lu  <hongjiu.lu@intel.com>
7614
7615         PR target/83905
7616         * config/i386/i386.c (ix86_expand_prologue): Use cost reference
7617         of struct ix86_frame.
7618         (ix86_expand_epilogue): Likewise.  Add a local variable for
7619         the reg_save_offset field in struct ix86_frame.
7620
7621 2018-01-23  Bin Cheng  <bin.cheng@arm.com>
7622
7623         PR tree-optimization/82604
7624         * tree-loop-distribution.c (enum partition_kind): New enum item
7625         PKIND_PARTIAL_MEMSET.
7626         (partition_builtin_p): Support above new enum item.
7627         (generate_code_for_partition): Ditto.
7628         (compute_access_range): Differentiate cases that equality can be
7629         proven at all loops, the innermost loops or no loops.
7630         (classify_builtin_st, classify_builtin_ldst): Adjust call to above
7631         function.  Set PKIND_PARTIAL_MEMSET for partition appropriately.
7632         (finalize_partitions, distribute_loop): Don't fuse partition of
7633         PKIND_PARTIAL_MEMSET kind when distributing 3-level loop nest.
7634         (prepare_perfect_loop_nest): Distribute 3-level loop nest only if
7635         parloop is enabled.
7636
7637 2018-01-23  Martin Liska  <mliska@suse.cz>
7638
7639         * predict.def (PRED_INDIR_CALL): Set probability to PROB_EVEN in
7640         order to ignore the predictor.
7641         (PRED_POLYMORPHIC_CALL): Likewise.
7642         (PRED_RECURSIVE_CALL): Likewise.
7643
7644 2018-01-23  Martin Liska  <mliska@suse.cz>
7645
7646         * tree-profile.c (tree_profiling): Print function header to
7647         aware reader which function we are working on.
7648         * value-prof.c (gimple_find_values_to_profile): Do not print
7649         not interesting value histograms.
7650
7651 2018-01-23  Martin Liska  <mliska@suse.cz>
7652
7653         * profile-count.h (enum profile_quality): Add
7654         profile_uninitialized as the first value. Do not number values
7655         as they are zero based.
7656         (profile_count::verify): Update sanity check.
7657         (profile_probability::verify): Likewise.
7658
7659 2018-01-23  Nathan Sidwell  <nathan@acm.org>
7660
7661         * doc/invoke.texi (ffor-scope): Deprecate.
7662
7663 2018-01-23  David Malcolm  <dmalcolm@redhat.com>
7664
7665         PR tree-optimization/83510
7666         * domwalk.c (set_all_edges_as_executable): New function.
7667         (dom_walker::dom_walker): Convert bool param
7668         "skip_unreachable_blocks" to enum reachability.  Move setup of
7669         edge flags to set_all_edges_as_executable and only do it when
7670         reachability is REACHABLE_BLOCKS.
7671         * domwalk.h (enum dom_walker::reachability): New enum.
7672         (dom_walker::dom_walker): Convert bool param
7673         "skip_unreachable_blocks" to enum reachability.
7674         (set_all_edges_as_executable): New decl.
7675         * graphite-scop-detection.c  (gather_bbs::gather_bbs): Convert
7676         from false for "skip_unreachable_blocks" to ALL_BLOCKS for
7677         "reachability".
7678         * tree-ssa-dom.c (dom_opt_dom_walker::dom_opt_dom_walker): Likewise,
7679         but converting true to REACHABLE_BLOCKS.
7680         * tree-ssa-sccvn.c (sccvn_dom_walker::sccvn_dom_walker): Likewise.
7681         * tree-vrp.c
7682         (check_array_bounds_dom_walker::check_array_bounds_dom_walker):
7683         Likewise, but converting it to REACHABLE_BLOCKS_PRESERVING_FLAGS.
7684         (vrp_dom_walker::vrp_dom_walker): Likewise, but converting it to
7685         REACHABLE_BLOCKS.
7686         (vrp_prop::vrp_finalize): Call set_all_edges_as_executable
7687         if check_all_array_refs will be called.
7688
7689 2018-01-23  David Malcolm  <dmalcolm@redhat.com>
7690
7691         * tree.c (selftest::test_location_wrappers): Add more test
7692         coverage.
7693
7694 2018-01-23  David Malcolm  <dmalcolm@redhat.com>
7695
7696         * sbitmap.c (selftest::test_set_range): Fix memory leaks.
7697         (selftest::test_bit_in_range): Likewise.
7698
7699 2018-01-23  Richard Sandiford  <richard.sandiford@linaro.org>
7700
7701         PR testsuite/83888
7702         * doc/sourcebuild.texi (vect_float): Say that the selector
7703         only describes the situation when -funsafe-math-optimizations is on.
7704         (vect_float_strict): Document.
7705
7706 2018-01-23  Richard Sandiford  <richard.sandiford@linaro.org>
7707
7708         PR tree-optimization/83965
7709         * tree-vect-patterns.c (vect_reassociating_reduction_p): New function.
7710         (vect_recog_dot_prod_pattern, vect_recog_sad_pattern): Use it
7711         instead of checking only for a reduction.
7712         (vect_recog_widen_sum_pattern): Likewise.
7713
7714 2018-01-23  Jan Hubicka  <hubicka@ucw.cz>
7715
7716         * predict.c (probably_never_executed): Only use precise profile info.
7717         (compute_function_frequency): Skip after inlining hack since we now
7718         have quality checking.
7719
7720 2018-01-23  Jan Hubicka  <hubicka@ucw.cz>
7721
7722         * profile-count.h (profile_probability::very_unlikely,
7723         profile_probability::unlikely, profile_probability::even): Set
7724         precision to guessed.
7725
7726 2018-01-23  Richard Biener  <rguenther@suse.de>
7727
7728         PR tree-optimization/83963
7729         * graphite-scop-detection.c (scop_detection::harmful_loop_in_region):
7730         Properly terminate dominator walk when crossing the exit edge not
7731         when visiting its source block.
7732
7733 2018-01-23  Jakub Jelinek  <jakub@redhat.com>
7734
7735         PR c++/83918
7736         * tree.c (maybe_wrap_with_location): Use NON_LVALUE_EXPR rather than
7737         VIEW_CONVERT_EXPR to wrap CONST_DECLs.
7738
7739 2018-01-22  Jakub Jelinek  <jakub@redhat.com>
7740
7741         PR tree-optimization/83957
7742         * omp-expand.c (expand_omp_for_generic): Ignore virtual PHIs.  Remove
7743         semicolon after for body surrounded by braces.
7744
7745         PR tree-optimization/83081
7746         * profile-count.h (profile_probability::split): New method.
7747         * dojump.c (do_jump_1) <case TRUTH_ANDIF_EXPR, case TRUTH_ORIF_EXPR>:
7748         Use profile_probability::split.
7749         (do_compare_rtx_and_jump): Fix adjustment of probabilities
7750         when splitting a single conditional jump into 2.
7751
7752 2018-01-22  David Malcolm  <dmalcolm@redhat.com>
7753
7754         PR tree-optimization/69452
7755         * tree-ssa-loop-im.c (class move_computations_dom_walker): Remove
7756         decl.
7757
7758 2018-01-22  Sebastian Perta  <sebastian.perta@renesas.com>
7759
7760         * config/rl78/rl78-expand.md (bswaphi2): New define_expand.
7761         * config/rl78/rl78-virt.md (*bswaphi2_virt): New define_insn.
7762         * config/rl78/rl78-real.md (*bswaphi2_real): New define_insn.
7763
7764 2018-01-22  Sebastian Perta  <sebastian.perta@renesas.com>
7765
7766         * config/rl78/rl78-protos.h (rl78_split_movdi): New function declaration.
7767         * config/rl78/rl78.md (movdi): New define_expand.
7768         * config/rl78/rl78.c (rl78_split_movdi): New function.
7769
7770 2018-01-22  Michael Meissner  <meissner@linux.vnet.ibm.com>
7771
7772         PR target/83862
7773         * config/rs6000/rs6000-protos.h (rs6000_split_signbit): Delete,
7774         no longer used.
7775         * config/rs6000/rs6000.c (rs6000_split_signbit): Likewise.
7776         * config/rs6000/rs6000.md (signbit<mode>2): Change code for IEEE
7777         128-bit to produce an UNSPEC move to get the double word with the
7778         signbit and then a shift directly to do signbit.
7779         (signbit<mode>2_dm): Replace old IEEE 128-bit signbit
7780         implementation with a new version that just does either a direct
7781         move or a regular move.  Move memory interface to separate insns.
7782         Move insns so they are next to the expander.
7783         (signbit<mode>2_dm_mem_be): New combiner insns to combine load
7784         with signbit move.  Split big and little endian case.
7785         (signbit<mode>2_dm_mem_le): Likewise.
7786         (signbit<mode>2_dm_<su>ext): Delete, no longer used.
7787         (signbit<mode>2_dm2): Likewise.
7788
7789 2018-01-22  Sebastian Perta  <sebastian.perta@renesas.com>
7790
7791         * config/rl78/rl78.md (anddi3): New define_expand.
7792
7793 2018-01-22  Sebastian Perta  <sebastian.perta@renesas.com>
7794
7795         * config/rl78/rl78.md (umindi3): New define_expand.
7796
7797 2018-01-22  Sebastian Perta  <sebastian.perta@renesas.com>
7798
7799         * config/rl78/rl78.md (smindi3): New define_expand.
7800
7801 2018-01-22  Sebastian Perta  <sebastian.perta@renesas.com>
7802
7803         * config/rl78/rl78.md (smaxdi3): New define_expand.
7804
7805 2018-01-22  Carl Love  <cel@us.ibm.com>
7806
7807         * config/rs6000/rs6000-builtin.def (ST_ELEMREV_V1TI, LD_ELEMREV_V1TI,
7808         LVX_V1TI): Add macro expansion.
7809         * config/rs6000/rs6000-c.c (altivec_builtin_types): Add argument
7810         definitions for VSX_BUILTIN_VEC_XST_BE, VSX_BUILTIN_VEC_ST,
7811         VSX_BUILTIN_VEC_XL, LD_ELEMREV_V1TI builtins.
7812         * config/rs6000/rs6000-p8swap.c (insn_is_swappable_p);
7813         Change check to determine if the instruction is a byte reversing
7814         entry.  Fix typo in comment.
7815         * config/rs6000/rs6000.c (altivec_expand_builtin): Add case entry
7816         for VSX_BUILTIN_ST_ELEMREV_V1TI and VSX_BUILTIN_LD_ELEMREV_V1TI.
7817         Add def_builtin calls for new builtins.
7818         * config/rs6000/vsx.md (vsx_st_elemrev_v1ti, vsx_ld_elemrev_v1ti):
7819         Add define_insn expansion.
7820
7821 2018-01-22  Sebastian Perta  <sebastian.perta@renesas.com>
7822
7823         * config/rl78/rl78.md (umaxdi3): New define_expand.
7824
7825 2018-01-22  Sebastian Perta  <sebastian.perta@renesas.com>
7826
7827         * config/rl78/rl78.c (rl78_note_reg_set): Fixed dead reg check
7828         for non-QImode registers.
7829
7830 2018-01-22  Richard Biener  <rguenther@suse.de>
7831
7832         PR tree-optimization/83963
7833         * graphite-scop-detection.c (scop_detection::get_sese): Delay
7834         including the loop exit block.
7835         (scop_detection::merge_sese): Likewise.
7836         (scop_detection::add_scop): Do it here instead.
7837
7838 2018-01-22  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
7839
7840         * doc/sourcebuild.texi (arm_softfloat): Document.
7841
7842 2018-01-21  John David Anglin  <danglin@gcc.gnu.org>
7843
7844         PR gcc/77734
7845         * config/pa/pa.c (pa_function_ok_for_sibcall): Use
7846         targetm.binds_local_p instead of TREE_PUBLIC to check local binding.
7847         Move TARGET_PORTABLE_RUNTIME check after TARGET_64BIT check.
7848
7849 2018-01-21  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
7850             David Edelsohn  <dje.gcc@gmail.com>
7851
7852         PR target/83946
7853         * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
7854         Change "crset eq" to "crset 2".
7855         (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
7856         (*call_indirect_aix<mode>_nospec): Likewise.
7857         (*call_value_indirect_aix<mode>_nospec): Likewise.
7858         (*call_indirect_elfv2<mode>_nospec): Likewise.
7859         (*call_value_indirect_elfv2<mode>_nospec): Likewise.
7860         (*sibcall_nonlocal_sysv<mode>): Change "crset eq" to "crset 2";
7861         change assembly output from . to $.
7862         (*sibcall_value_nonlocal_sysv<mode>): Likewise.
7863         (indirect_jump<mode>_nospec): Change assembly output from . to $.
7864         (*tablejump<mode>_internal1_nospec): Likewise.
7865
7866 2018-01-21  Oleg Endo  <olegendo@gcc.gnu.org>
7867
7868         PR target/80870
7869         * config/sh/sh_optimize_sett_clrt.cc:
7870         Use INCLUDE_ALGORITHM and INCLUDE_VECTOR instead of direct includes.
7871
7872 2018-01-20  Richard Sandiford  <richard.sandiford@linaro.org>
7873
7874         PR tree-optimization/83940
7875         * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): Set
7876         offset_dt to vect_constant_def rather than vect_unknown_def_type.
7877         (vect_check_load_store_mask): Add a mask_dt_out parameter and
7878         use it to pass back the definition type.
7879         (vect_check_store_rhs): Likewise rhs_dt_out.
7880         (vect_build_gather_load_calls): Add a mask_dt argument and use
7881         it instead of a call to vect_is_simple_use.
7882         (vectorizable_store): Update calls to vect_check_load_store_mask
7883         and vect_check_store_rhs.  Use the dt returned by the latter instead
7884         of scatter_src_dt.  Use the cached mask_dt and gs_info.offset_dt
7885         instead of calls to vect_is_simple_use.  Pass the scalar rather
7886         than the vector operand to vect_is_simple_use when handling
7887         second and subsequent copies of an rhs value.
7888         (vectorizable_load): Update calls to vect_check_load_store_mask
7889         and vect_build_gather_load_calls.  Use the cached mask_dt and
7890         gs_info.offset_dt instead of calls to vect_is_simple_use.
7891
7892 2018-01-20  Jakub Jelinek  <jakub@redhat.com>
7893
7894         PR middle-end/83945
7895         * tree-emutls.c: Include gimplify.h.
7896         (lower_emutls_2): New function.
7897         (lower_emutls_1): If ADDR_EXPR is a gimple invariant and walk_tree
7898         with lower_emutls_2 callback finds some TLS decl in it, unshare_expr
7899         it before further processing.
7900
7901         PR target/83930
7902         * simplify-rtx.c (simplify_binary_operation_1) <case UMOD>: Use
7903         UINTVAL (trueop1) instead of INTVAL (op1).
7904
7905 2018-01-19  Jakub Jelinek  <jakub@redhat.com>
7906
7907         PR debug/81570
7908         PR debug/83728
7909         * dwarf2cfi.c (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define to
7910         INCOMING_FRAME_SP_OFFSET if not defined.
7911         (scan_trace): Add ENTRY argument.  If true and
7912         DEFAULT_INCOMING_FRAME_SP_OFFSET != INCOMING_FRAME_SP_OFFSET,
7913         emit a note to adjust the CFA offset.
7914         (create_cfi_notes): Adjust scan_trace callers.
7915         (create_cie_data): Use DEFAULT_INCOMING_FRAME_SP_OFFSET rather than
7916         INCOMING_FRAME_SP_OFFSET in the CIE.
7917         * config/i386/i386.h (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define.
7918         * config/stormy16/stormy16.h (DEFAULT_INCOMING_FRAME_SP_OFFSET):
7919         Likewise.
7920         * doc/tm.texi.in (DEFAULT_INCOMING_FRAME_SP_OFFSET): Document.
7921         * doc/tm.texi: Regenerated.
7922
7923 2018-01-19  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>
7924
7925         PR rtl-optimization/83147
7926         * lra-constraints.c (remove_inheritance_pseudos): Use
7927         lra_substitute_pseudo_within_insn.
7928
7929 2018-01-19  Tom de Vries  <tom@codesourcery.com>
7930             Cesar Philippidis  <cesar@codesourcery.com>
7931
7932         PR target/83920
7933         * config/nvptx/nvptx.c (nvptx_single): Fix jit workaround.
7934
7935 2018-01-19  Cesar Philippidis  <cesar@codesourcery.com>
7936
7937         PR target/83790
7938         * config/nvptx/nvptx.c (output_init_frag): Don't use generic address
7939         spaces for function labels.
7940
7941 2018-01-19  Martin Liska  <mliska@suse.cz>
7942
7943         * predict.def (PRED_LOOP_EXIT): Change from 85 to 89.
7944         (PRED_LOOP_EXIT_WITH_RECURSION): Change from 72 to 78.
7945         (PRED_LOOP_EXTRA_EXIT): Change from 83 to 67.
7946         (PRED_OPCODE_POSITIVE): Change from 64 to 59.
7947         (PRED_TREE_OPCODE_POSITIVE): Change from 64 to 59.
7948         (PRED_CONST_RETURN): Change from 69 to 65.
7949         (PRED_NULL_RETURN): Change from 91 to 71.
7950         (PRED_LOOP_IV_COMPARE_GUESS): Change from 98 to 64.
7951         (PRED_LOOP_GUARD): Change from 66 to 73.
7952
7953 2018-01-19  Martin Liska  <mliska@suse.cz>
7954
7955         * predict.c (predict_insn_def): Add new assert.
7956         (struct branch_predictor): Change type to signed integer.
7957         (test_prediction_value_range): Amend test to cover
7958         PROB_UNINITIALIZED.
7959         * predict.def (PRED_LOOP_ITERATIONS): Use the new constant.
7960         (PRED_LOOP_ITERATIONS_GUESSED): Likewise.
7961         (PRED_LOOP_ITERATIONS_MAX): Likewise.
7962         (PRED_LOOP_IV_COMPARE): Likewise.
7963         * predict.h (PROB_UNINITIALIZED): Define new constant.
7964
7965 2018-01-19  Martin Liska  <mliska@suse.cz>
7966
7967         * predict.c (dump_prediction): Add new format for
7968         analyze_brprob.py script which is enabled with -details
7969         suboption.
7970         * profile-count.h (precise_p): New function.
7971
7972 2018-01-19  Richard Sandiford  <richard.sandiford@linaro.org>
7973
7974         PR tree-optimization/83922
7975         * tree-vect-loop.c (vect_verify_full_masking): Return false if
7976         there are no statements that need masking.
7977         (vect_active_double_reduction_p): New function.
7978         (vect_analyze_loop_operations): Use it when handling phis that
7979         are not in the loop header.
7980
7981 2018-01-19  Richard Sandiford  <richard.sandiford@linaro.org>
7982
7983         PR tree-optimization/83914
7984         * tree-vect-loop.c (vectorizable_induction): Don't convert
7985         init_expr or apply the peeling adjustment for inductions
7986         that are nested within the vectorized loop.
7987
7988 2018-01-19  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
7989
7990         * config/arm/thumb2.md (*thumb2_negsi2_short): Use RSB mnemonic
7991         instead of NEG.
7992
7993 2018-01-18  Jakub Jelinek  <jakub@redhat.com>
7994
7995         PR sanitizer/81715
7996         PR testsuite/83882
7997         * function.h (gimplify_parameters): Add gimple_seq * argument.
7998         * function.c: Include gimple.h and options.h.
7999         (gimplify_parameters): Add cleanup argument, add CLOBBER stmts
8000         for the added local temporaries if needed.
8001         * gimplify.c (gimplify_body): Adjust gimplify_parameters caller,
8002         if there are any parameter cleanups, wrap whole body into a
8003         try/finally with the cleanups.
8004
8005 2018-01-18  Wilco Dijkstra  <wdijkstr@arm.com>
8006
8007         PR target/82964
8008         * config/aarch64/aarch64.c (aarch64_legitimate_constant_p):
8009         Use GET_MODE_CLASS for scalar floating point.
8010
8011 2018-01-18  Jan Hubicka  <hubicka@ucw.cz>
8012
8013         PR ipa/82256
8014         patch by PaX Team
8015         * cgraphclones.c (cgraph_node::create_version_clone_with_body):
8016         Fix call of call_cgraph_insertion_hooks.
8017
8018 2018-01-18  Martin Sebor  <msebor@redhat.com>
8019
8020         * doc/invoke.texi (-Wclass-memaccess): Tweak text.
8021
8022 2018-01-18  Jan Hubicka  <hubicka@ucw.cz>
8023
8024         PR ipa/83619
8025         * cgraph.c (cgraph_edge::redirect_call_stmt_to_callee): Update edge
8026         frequencies.
8027
8028 2018-01-18  Boris Kolpackov  <boris@codesynthesis.com>
8029
8030         PR other/70268
8031         * common.opt: (-ffile-prefix-map): New option.
8032         * opts.c (common_handle_option): Defer it.
8033         * opts-global.c (handle_common_deferred_options): Handle it.
8034         * debug.h (remap_debug_filename, add_debug_prefix_map): Move to...
8035         * file-prefix-map.h: New file.
8036         (remap_debug_filename, add_debug_prefix_map): ...here.
8037         (add_macro_prefix_map, add_file_prefix_map, remap_macro_filename): New.
8038         * final.c (debug_prefix_map, add_debug_prefix_map
8039         remap_debug_filename): Move to...
8040         * file-prefix-map.c: New file.
8041         (file_prefix_map, add_prefix_map, remap_filename) ...here and rename,
8042         generalize, get rid of alloca(), use strrchr() instead of strchr().
8043         (add_macro_prefix_map, add_debug_prefix_map, add_file_prefix_map):
8044         Implement in terms of add_prefix_map().
8045         (remap_macro_filename, remap_debug_filename): Implement in term of
8046         remap_filename().
8047         * Makefile.in (OBJS, PLUGIN_HEADERS): Add new files.
8048         * builtins.c (fold_builtin_FILE): Call remap_macro_filename().
8049         * dbxout.c: Include file-prefix-map.h.
8050         * varasm.c: Likewise.
8051         * vmsdbgout.c: Likewise.
8052         * xcoffout.c: Likewise.
8053         * dwarf2out.c: Likewise plus omit new options from DW_AT_producer.
8054         * doc/cppopts.texi (-fmacro-prefix-map): Document.
8055         * doc/invoke.texi (-ffile-prefix-map): Document.
8056         (-fdebug-prefix-map): Update description.
8057
8058 2018-01-18  Martin Liska  <mliska@suse.cz>
8059
8060         * config/i386/i386.c (indirect_thunk_name): Document that also
8061         lfence is emitted.
8062         (output_indirect_thunk): Document why both instructions
8063         (pause and lfence) are generated.
8064
8065 2018-01-18  Richard Biener  <rguenther@suse.de>
8066
8067         PR tree-optimization/83887
8068         * graphite-scop-detection.c
8069         (scop_detection::get_nearest_dom_with_single_entry): Remove.
8070         (scop_detection::get_nearest_pdom_with_single_exit): Likewise.
8071         (scop_detection::merge_sese): Re-implement with a flood-fill
8072         algorithm that properly finds a SESE region if it exists.
8073
8074 2018-01-18  Jakub Jelinek  <jakub@redhat.com>
8075
8076         PR c/61240
8077         * match.pd ((P + A) - P, P - (P + A), (P + A) - (P + B)): For
8078         pointer_diff optimizations use view_convert instead of convert.
8079
8080 2018-01-17  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
8081
8082         * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
8083         Generate different code for -mno-speculate-indirect-jumps.
8084         (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
8085         (*call_indirect_aix<mode>): Disable for
8086         -mno-speculate-indirect-jumps.
8087         (*call_indirect_aix<mode>_nospec): New define_insn.
8088         (*call_value_indirect_aix<mode>): Disable for
8089         -mno-speculate-indirect-jumps.
8090         (*call_value_indirect_aix<mode>_nospec): New define_insn.
8091         (*sibcall_nonlocal_sysv<mode>): Generate different code for
8092         -mno-speculate-indirect-jumps.
8093         (*sibcall_value_nonlocal_sysv<mode>): Likewise.
8094
8095 2018-01-17  Michael Meissner  <meissner@linux.vnet.ibm.com>
8096
8097         * config/rs6000/rs6000.c (rs6000_emit_move): If we load or store a
8098         long double type, set the flags for noting the default long double
8099         type, even if we don't pass or return a long double type.
8100
8101 2018-01-17  Jan Hubicka  <hubicka@ucw.cz>
8102
8103         PR ipa/83051
8104         * ipa-inline.c (flatten_function): Do not overwrite final inlining
8105         failure.
8106
8107 2018-01-17  Will Schmidt  <will_schmidt@vnet.ibm.com>
8108
8109         * config/rs6000/rs6000.c (rs6000_gimple_builtin): Add gimple folding
8110         support for merge[hl].
8111         (fold_mergehl_helper): New helper function.
8112         (tree-vector-builder.h): New #include for tree_vector_builder usage.
8113         * config/rs6000/altivec.md (altivec_vmrghw_direct): Add xxmrghw insn.
8114         (altivec_vmrglw_direct): Add xxmrglw insn.
8115
8116 2018-01-17  Andrew Waterman  <andrew@sifive.com>
8117
8118         * config/riscv/riscv.c (riscv_conditional_register_usage): If
8119         UNITS_PER_FP_ARG is 0, set call_used_regs to 1 for all FP regs.
8120
8121 2018-01-17  David Malcolm  <dmalcolm@redhat.com>
8122
8123         PR lto/83121
8124         * ipa-devirt.c (add_type_duplicate): When comparing memory layout,
8125         call the lto_location_cache before reading the
8126         DECL_SOURCE_LOCATION of the types.
8127
8128 2018-01-17  Wilco Dijkstra  <wdijkstr@arm.com>
8129             Richard Sandiford  <richard.sandiford@linaro.org>
8130
8131         * config/aarch64/aarch64.md (movti_aarch64): Use Uti constraint.
8132         * config/aarch64/aarch64.c (aarch64_mov128_immediate): New function.
8133         (aarch64_legitimate_constant_p): Just support CONST_DOUBLE
8134         SF/DF/TF mode to avoid creating illegal CONST_WIDE_INT immediates.
8135         * config/aarch64/aarch64-protos.h (aarch64_mov128_immediate):
8136         Add declaration.
8137         * config/aarch64/constraints.md (aarch64_movti_operand):
8138         Limit immediates.
8139         * config/aarch64/predicates.md (Uti): Add new constraint.
8140
8141 2018-01-17 Carl Love  <cel@us.ibm.com>
8142
8143         * config/rs6000/vsx.md (define_expand xl_len_r,
8144         define_expand stxvl, define_expand *stxvl): Add match_dup argument.
8145         (define_insn): Add, match_dup 1 argument to define_insn stxvll and
8146         lxvll.
8147         (define_expand, define_insn): Move the shift left from  the
8148         define_insn to the define_expand for lxvl and stxvl instructions.
8149         * config/rs6000/rs6000-builtin.def (BU_P9V_64BIT_VSX_2): Change LXVL
8150         and XL_LEN_R definitions to PURE.
8151
8152 2018-01-17  Uros Bizjak  <ubizjak@gmail.com>
8153
8154         * config/i386/i386.c (indirect_thunk_name): Declare regno
8155         as unsigned int.  Compare regno with INVALID_REGNUM.
8156         (output_indirect_thunk): Ditto.
8157         (output_indirect_thunk_function): Ditto.
8158         (ix86_code_end): Declare regno as unsigned int.  Use INVALID_REGNUM
8159         in the call to output_indirect_thunk_function.
8160
8161 2018-01-17  Richard Sandiford  <richard.sandiford@linaro.org>
8162
8163         PR middle-end/83884
8164         * expr.c (expand_expr_real_1): Use the size of GET_MODE (op0)
8165         rather than the size of inner_type to determine the stack slot size
8166         when handling VIEW_CONVERT_EXPRs on strict-alignment targets.
8167
8168 2018-01-16  Sebastian Peryt  <sebastian.peryt@intel.com>
8169
8170         PR target/83546
8171         * config/i386/i386.c (ix86_option_override_internal): Add PTA_RDRND
8172         to PTA_SILVERMONT.
8173
8174 2018-01-16  Michael Meissner  <meissner@linux.vnet.ibm.com>
8175
8176         * config.gcc (powerpc*-linux*-*): Add support for 64-bit little
8177         endian Linux systems to optionally enable multilibs for selecting
8178         the long double type if the user configured an explicit type.
8179         * config/rs6000/rs6000.h (TARGET_IEEEQUAD_MULTILIB): Indicate we
8180         have no long double multilibs if not defined.
8181         * config/rs6000/rs6000.c (rs6000_option_override_internal): Do not
8182         warn if the user used -mabi={ieee,ibm}longdouble and we built
8183         multilibs for long double.
8184         * config/rs6000/linux64.h (MULTILIB_DEFAULTS_IEEE): Define as the
8185         appropriate multilib option.
8186         (MULTILIB_DEFAULTS): Add MULTILIB_DEFAULTS_IEEE to the default
8187         multilib options.
8188         * config/rs6000/t-ldouble-linux64le-ibm: New configuration files
8189         for building long double multilibs.
8190         * config/rs6000/t-ldouble-linux64le-ieee: Likewise.
8191
8192 2018-01-16  John David Anglin  <danglin@gcc.gnu.org>
8193
8194         * config.gcc (hppa*-*-linux*): Change callee copies ABI to caller
8195         copies.
8196
8197         * config/pa.h (MALLOC_ABI_ALIGNMENT): Set 32-bit alignment default to
8198         64 bits.
8199         * config/pa/pa32-linux.h (MALLOC_ABI_ALIGNMENT): Set alignment to
8200         128 bits.
8201
8202         * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Cleanup type and mode
8203         variables.
8204
8205         * config/pa/pa.c (pa_function_arg_size): Apply CEIL to GET_MODE_SIZE
8206         return value.
8207
8208 2018-01-16  Eric Botcazou  <ebotcazou@adacore.com>
8209
8210         * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For an
8211         ADDR_EXPR, do not count the offset of a COMPONENT_REF twice.
8212
8213 2018-01-16  Kelvin Nilsen  <kelvin@gcc.gnu.org>
8214
8215         * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Generate
8216         different rtl trees depending on TARGET_64BIT.
8217         (rs6000_gen_lvx): Likewise.
8218
8219 2018-01-16  Eric Botcazou  <ebotcazou@adacore.com>
8220
8221         * config/visium/visium.md (nop): Tweak comment.
8222         (hazard_nop): Likewise.
8223
8224 2018-01-16  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
8225
8226         * config/rs6000/rs6000.c (rs6000_opt_vars): Add entry for
8227         -mspeculate-indirect-jumps.
8228         * config/rs6000/rs6000.md (*call_indirect_elfv2<mode>): Disable
8229         for -mno-speculate-indirect-jumps.
8230         (*call_indirect_elfv2<mode>_nospec): New define_insn.
8231         (*call_value_indirect_elfv2<mode>): Disable for
8232         -mno-speculate-indirect-jumps.
8233         (*call_value_indirect_elfv2<mode>_nospec): New define_insn.
8234         (indirect_jump): Emit different RTL for
8235         -mno-speculate-indirect-jumps.
8236         (*indirect_jump<mode>): Disable for
8237         -mno-speculate-indirect-jumps.
8238         (*indirect_jump<mode>_nospec): New define_insn.
8239         (tablejump): Emit different RTL for
8240         -mno-speculate-indirect-jumps.
8241         (tablejumpsi): Disable for -mno-speculate-indirect-jumps.
8242         (tablejumpsi_nospec): New define_expand.
8243         (tablejumpdi): Disable for -mno-speculate-indirect-jumps.
8244         (tablejumpdi_nospec): New define_expand.
8245         (*tablejump<mode>_internal1): Disable for
8246         -mno-speculate-indirect-jumps.
8247         (*tablejump<mode>_internal1_nospec): New define_insn.
8248         * config/rs6000/rs6000.opt (mspeculate-indirect-jumps): New
8249         option.
8250
8251 2018-01-16  Artyom Skrobov tyomitch@gmail.com
8252
8253         * caller-save.c (insert_save): Drop unnecessary parameter.  All
8254         callers updated.
8255
8256 2018-01-16  Jakub Jelinek  <jakub@redhat.com>
8257             Richard Biener  <rguenth@suse.de>
8258
8259         PR libgomp/83590
8260         * gimplify.c (gimplify_one_sizepos): For is_gimple_constant (expr)
8261         return early, inline manually is_gimple_sizepos.  Make sure if we
8262         call gimplify_expr we don't end up with a gimple constant.
8263         * tree.c (variably_modified_type_p): Don't return true for
8264         is_gimple_constant (_t).  Inline manually is_gimple_sizepos.
8265         * gimplify.h (is_gimple_sizepos): Remove.
8266
8267 2018-01-16  Richard Sandiford  <richard.sandiford@linaro.org>
8268
8269         PR tree-optimization/83857
8270         * tree-vect-loop.c (vect_analyze_loop_operations): Don't call
8271         vectorizable_live_operation for pure SLP statements.
8272         (vectorizable_live_operation): Handle PHIs.
8273
8274 2018-01-16  Richard Biener  <rguenther@suse.de>
8275
8276         PR tree-optimization/83867
8277         * tree-vect-stmts.c (vect_transform_stmt): Precompute
8278         nested_in_vect_loop_p since the scalar stmt may get invalidated.
8279
8280 2018-01-16  Jakub Jelinek  <jakub@redhat.com>
8281
8282         PR c/83844
8283         * stor-layout.c (handle_warn_if_not_align): Use byte_position and
8284         multiple_of_p instead of unchecked tree_to_uhwi and UHWI check.
8285         If off is not INTEGER_CST, issue a may not be aligned warning
8286         rather than isn't aligned.  Use isn%'t rather than isn't.
8287         * fold-const.c (multiple_of_p) <case BIT_AND_EXPR>: Don't fall through
8288         into MULT_EXPR.
8289         <case MULT_EXPR>: Improve the case when bottom and one of the
8290         MULT_EXPR operands are INTEGER_CSTs and bottom is multiple of that
8291         operand, in that case check if the other operand is multiple of
8292         bottom divided by the INTEGER_CST operand.
8293
8294 2018-01-16  Richard Sandiford  <richard.sandiford@linaro.org>
8295
8296         PR target/83858
8297         * config/pa/pa.h (FUNCTION_ARG_SIZE): Delete.
8298         * config/pa/pa-protos.h (pa_function_arg_size): Declare.
8299         * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Use
8300         pa_function_arg_size instead of FUNCTION_ARG_SIZE.
8301         * config/pa/pa.c (pa_function_arg_advance): Likewise.
8302         (pa_function_arg, pa_arg_partial_bytes): Likewise.
8303         (pa_function_arg_size): New function.
8304
8305 2018-01-16  Richard Sandiford  <richard.sandiford@linaro.org>
8306
8307         * fold-const.c (fold_ternary_loc): Construct the vec_perm_indices
8308         in a separate statement.
8309
8310 2018-01-16  Richard Sandiford  <richard.sandiford@linaro.org>
8311
8312         PR tree-optimization/83847
8313         * tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Don't
8314         group gathers and scatters.
8315
8316 2018-01-16  Jakub Jelinek  <jakub@redhat.com>
8317
8318         PR rtl-optimization/86620
8319         * params.def (max-sched-ready-insns): Bump minimum value to 1.
8320
8321         PR rtl-optimization/83213
8322         * recog.c (peep2_attempt): Copy over CROSSING_JUMP_P from peepinsn
8323         to last if both are JUMP_INSNs.
8324
8325         PR tree-optimization/83843
8326         * gimple-ssa-store-merging.c
8327         (imm_store_chain_info::output_merged_store): Handle bit_not_p on
8328         store_immediate_info for bswap/nop orig_stores.
8329
8330 2018-01-15  Andrew Waterman  <andrew@sifive.com>
8331
8332         * config/riscv/riscv.c (riscv_rtx_costs) <MULT>: Increase cost if
8333         !TARGET_MUL.
8334         <UDIV>: Increase cost if !TARGET_DIV.
8335
8336 2018-01-15  Segher Boessenkool  <segher@kernel.crashing.org>
8337
8338         * config/rs6000/rs6000.md (define_attr "type"): Remove delayed_cr.
8339         (define_attr "cr_logical_3op"): New.
8340         (cceq_ior_compare): Adjust.
8341         (cceq_ior_compare_complement): Adjust.
8342         (*cceq_rev_compare): Adjust.
8343         * config/rs6000/rs6000.c (rs6000_adjust_cost): Adjust.
8344         (is_cracked_insn): Adjust.
8345         (insn_must_be_first_in_group): Adjust.
8346         * config/rs6000/40x.md: Adjust.
8347         * config/rs6000/440.md: Adjust.
8348         * config/rs6000/476.md: Adjust.
8349         * config/rs6000/601.md: Adjust.
8350         * config/rs6000/603.md: Adjust.
8351         * config/rs6000/6xx.md: Adjust.
8352         * config/rs6000/7450.md: Adjust.
8353         * config/rs6000/7xx.md: Adjust.
8354         * config/rs6000/8540.md: Adjust.
8355         * config/rs6000/cell.md: Adjust.
8356         * config/rs6000/e300c2c3.md: Adjust.
8357         * config/rs6000/e500mc.md: Adjust.
8358         * config/rs6000/e500mc64.md: Adjust.
8359         * config/rs6000/e5500.md: Adjust.
8360         * config/rs6000/e6500.md: Adjust.
8361         * config/rs6000/mpc.md: Adjust.
8362         * config/rs6000/power4.md: Adjust.
8363         * config/rs6000/power5.md: Adjust.
8364         * config/rs6000/power6.md: Adjust.
8365         * config/rs6000/power7.md: Adjust.
8366         * config/rs6000/power8.md: Adjust.
8367         * config/rs6000/power9.md: Adjust.
8368         * config/rs6000/rs64.md: Adjust.
8369         * config/rs6000/titan.md: Adjust.
8370
8371 2018-01-15  H.J. Lu  <hongjiu.lu@intel.com>
8372
8373         * config/i386/predicates.md (indirect_branch_operand): Rewrite
8374         ix86_indirect_branch_register logic.
8375
8376 2018-01-15  H.J. Lu  <hongjiu.lu@intel.com>
8377
8378         * config/i386/constraints.md (Bs): Update
8379         ix86_indirect_branch_register check.  Don't check
8380         ix86_indirect_branch_register with GOT_memory_operand.
8381         (Bw): Likewise.
8382         * config/i386/predicates.md (GOT_memory_operand): Don't check
8383         ix86_indirect_branch_register here.
8384         (GOT32_symbol_operand): Likewise.
8385
8386 2018-01-15  H.J. Lu  <hongjiu.lu@intel.com>
8387
8388         * config/i386/predicates.md (constant_call_address_operand):
8389         Rewrite ix86_indirect_branch_register logic.
8390         (sibcall_insn_operand): Likewise.
8391
8392 2018-01-15  H.J. Lu  <hongjiu.lu@intel.com>
8393
8394         * config/i386/constraints.md (Bs): Replace
8395         ix86_indirect_branch_thunk_register with
8396         ix86_indirect_branch_register.
8397         (Bw): Likewise.
8398         * config/i386/i386.md (indirect_jump): Likewise.
8399         (tablejump): Likewise.
8400         (*sibcall_memory): Likewise.
8401         (*sibcall_value_memory): Likewise.
8402         Peepholes of indirect call and jump via memory: Likewise.
8403         * config/i386/i386.opt: Likewise.
8404         * config/i386/predicates.md (indirect_branch_operand): Likewise.
8405         (GOT_memory_operand): Likewise.
8406         (call_insn_operand): Likewise.
8407         (sibcall_insn_operand): Likewise.
8408         (GOT32_symbol_operand): Likewise.
8409
8410 2018-01-15  Jakub Jelinek  <jakub@redhat.com>
8411
8412         PR middle-end/83837
8413         * omp-expand.c (expand_omp_atomic_pipeline): Use loaded_val
8414         type rather than type addr's type points to.
8415         (expand_omp_atomic_mutex): Likewise.
8416         (expand_omp_atomic): Likewise.
8417
8418 2018-01-15  H.J. Lu  <hongjiu.lu@intel.com>
8419
8420         PR target/83839
8421         * config/i386/i386.c (output_indirect_thunk_function): Use
8422         ASM_OUTPUT_LABEL, instead of ASM_OUTPUT_DEF, for TARGET_MACHO
8423         for  __x86_return_thunk.
8424
8425 2018-01-15  Richard Biener  <rguenther@suse.de>
8426
8427         PR middle-end/83850
8428         * expmed.c (extract_bit_field_1): Fix typo.
8429
8430 2018-01-15  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
8431
8432         PR target/83687
8433         * config/arm/iterators.md (VF): New mode iterator.
8434         * config/arm/neon.md (neon_vabd<mode>_2): Use the above.
8435         Remove integer-related logic from pattern.
8436         (neon_vabd<mode>_3): Likewise.
8437
8438 2018-01-15  Jakub Jelinek  <jakub@redhat.com>
8439
8440         PR middle-end/82694
8441         * common.opt (fstrict-overflow): No longer an alias.
8442         (fwrapv-pointer): New option.
8443         * tree.h (TYPE_OVERFLOW_WRAPS, TYPE_OVERFLOW_UNDEFINED): Define
8444         also for pointer types based on flag_wrapv_pointer.
8445         * opts.c (common_handle_option) <case OPT_fstrict_overflow>: Set
8446         opts->x_flag_wrap[pv] to !value, clear opts->x_flag_trapv if
8447         opts->x_flag_wrapv got set.
8448         * fold-const.c (fold_comparison, fold_binary_loc): Revert 2017-08-01
8449         changes, just use TYPE_OVERFLOW_UNDEFINED on pointer type instead of
8450         POINTER_TYPE_OVERFLOW_UNDEFINED.
8451         * match.pd: Likewise in address comparison pattern.
8452         * doc/invoke.texi: Document -fwrapv and -fstrict-overflow.
8453
8454 2018-01-15  Richard Biener  <rguenther@suse.de>
8455
8456         PR lto/83804
8457         * tree.c (free_lang_data_in_type): Always unlink TYPE_DECLs
8458         from TYPE_FIELDS.  Free TYPE_BINFO if not used by devirtualization.
8459         Reset type names to their identifier if their TYPE_DECL doesn't
8460         have linkage (and thus is used for ODR and devirt).
8461         (save_debug_info_for_decl): Remove.
8462         (save_debug_info_for_type): Likewise.
8463         (add_tree_to_fld_list): Adjust.
8464         * tree-pretty-print.c (dump_generic_node): Make dumping of
8465         type names more robust.
8466
8467 2018-01-15  Richard Biener  <rguenther@suse.de>
8468
8469         * BASE-VER: Bump to 8.0.1.
8470
8471 2018-01-14  Martin Sebor  <msebor@redhat.com>
8472
8473         PR other/83508
8474         * builtins.c (check_access): Avoid warning when the no-warning bit
8475         is set.
8476
8477 2018-01-14  Cory Fields  <cory-nospam-@coryfields.com>
8478
8479         * tree-ssa-loop-im.c (sort_bbs_in_loop_postorder_cmp): Stabilize sort.
8480         * ira-color (allocno_hard_regs_compare): Likewise.
8481
8482 2018-01-14  Nathan Rossi  <nathan@nathanrossi.com>
8483
8484         PR target/83013
8485         * config/microblaze/microblaze.c (microblaze_asm_output_ident):
8486         Use .pushsection/.popsection.
8487
8488 2018-01-14  Martin Sebor  <msebor@redhat.com>
8489
8490         PR c++/81327
8491         * doc/invoke.texi (-Wlass-memaccess): Document suppression by casting.
8492
8493 2018-01-14  Jakub Jelinek  <jakub@redhat.com>
8494
8495         * config.gcc (i[34567]86-*-*): Remove one duplicate gfniintrin.h
8496         entry from extra_headers.
8497         (x86_64-*-*): Remove two duplicate gfniintrin.h entries from
8498         extra_headers, make the list bitwise identical to the i?86-*-* one.
8499
8500 2018-01-14  H.J. Lu  <hongjiu.lu@intel.com>
8501
8502         * config/i386/i386.c (ix86_set_indirect_branch_type): Disallow
8503         -mcmodel=large with -mindirect-branch=thunk,
8504         -mindirect-branch=thunk-extern, -mfunction-return=thunk and
8505         -mfunction-return=thunk-extern.
8506         * doc/invoke.texi: Document -mcmodel=large is incompatible with
8507         -mindirect-branch=thunk, -mindirect-branch=thunk-extern,
8508         -mfunction-return=thunk and -mfunction-return=thunk-extern.
8509
8510 2018-01-14  H.J. Lu  <hongjiu.lu@intel.com>
8511
8512         * config/i386/i386.c (print_reg): Print the name of the full
8513         integer register without '%'.
8514         (ix86_print_operand): Handle 'V'.
8515          * doc/extend.texi: Document 'V' modifier.
8516
8517 2018-01-14  H.J. Lu  <hongjiu.lu@intel.com>
8518
8519         * config/i386/constraints.md (Bs): Disallow memory operand for
8520         -mindirect-branch-register.
8521         (Bw): Likewise.
8522         * config/i386/predicates.md (indirect_branch_operand): Likewise.
8523         (GOT_memory_operand): Likewise.
8524         (call_insn_operand): Likewise.
8525         (sibcall_insn_operand): Likewise.
8526         (GOT32_symbol_operand): Likewise.
8527         * config/i386/i386.md (indirect_jump): Call convert_memory_address
8528         for -mindirect-branch-register.
8529         (tablejump): Likewise.
8530         (*sibcall_memory): Likewise.
8531         (*sibcall_value_memory): Likewise.
8532         Disallow peepholes of indirect call and jump via memory for
8533         -mindirect-branch-register.
8534         (*call_pop): Replace m with Bw.
8535         (*call_value_pop): Likewise.
8536         (*sibcall_pop_memory): Replace m with Bs.
8537         * config/i386/i386.opt (mindirect-branch-register): New option.
8538         * doc/invoke.texi: Document -mindirect-branch-register option.
8539
8540 2018-01-14  H.J. Lu  <hongjiu.lu@intel.com>
8541
8542         * config/i386/i386-protos.h (ix86_output_function_return): New.
8543         * config/i386/i386.c (ix86_set_indirect_branch_type): Also
8544         set function_return_type.
8545         (indirect_thunk_name): Add ret_p to indicate thunk for function
8546         return.
8547         (output_indirect_thunk_function): Pass false to
8548         indirect_thunk_name.
8549         (ix86_output_indirect_branch_via_reg): Likewise.
8550         (ix86_output_indirect_branch_via_push): Likewise.
8551         (output_indirect_thunk_function): Create alias for function
8552         return thunk if regno < 0.
8553         (ix86_output_function_return): New function.
8554         (ix86_handle_fndecl_attribute): Handle function_return.
8555         (ix86_attribute_table): Add function_return.
8556         * config/i386/i386.h (machine_function): Add
8557         function_return_type.
8558         * config/i386/i386.md (simple_return_internal): Use
8559         ix86_output_function_return.
8560         (simple_return_internal_long): Likewise.
8561         * config/i386/i386.opt (mfunction-return=): New option.
8562         (indirect_branch): Mention -mfunction-return=.
8563         * doc/extend.texi: Document function_return function attribute.
8564         * doc/invoke.texi: Document -mfunction-return= option.
8565
8566 2018-01-14  H.J. Lu  <hongjiu.lu@intel.com>
8567
8568         * config/i386/i386-opts.h (indirect_branch): New.
8569         * config/i386/i386-protos.h (ix86_output_indirect_jmp): Likewise.
8570         * config/i386/i386.c (ix86_using_red_zone): Disallow red-zone
8571         with local indirect jump when converting indirect call and jump.
8572         (ix86_set_indirect_branch_type): New.
8573         (ix86_set_current_function): Call ix86_set_indirect_branch_type.
8574         (indirectlabelno): New.
8575         (indirect_thunk_needed): Likewise.
8576         (indirect_thunk_bnd_needed): Likewise.
8577         (indirect_thunks_used): Likewise.
8578         (indirect_thunks_bnd_used): Likewise.
8579         (INDIRECT_LABEL): Likewise.
8580         (indirect_thunk_name): Likewise.
8581         (output_indirect_thunk): Likewise.
8582         (output_indirect_thunk_function): Likewise.
8583         (ix86_output_indirect_branch_via_reg): Likewise.
8584         (ix86_output_indirect_branch_via_push): Likewise.
8585         (ix86_output_indirect_branch): Likewise.
8586         (ix86_output_indirect_jmp): Likewise.
8587         (ix86_code_end): Call output_indirect_thunk_function if needed.
8588         (ix86_output_call_insn): Call ix86_output_indirect_branch if
8589         needed.
8590         (ix86_handle_fndecl_attribute): Handle indirect_branch.
8591         (ix86_attribute_table): Add indirect_branch.
8592         * config/i386/i386.h (machine_function): Add indirect_branch_type
8593         and has_local_indirect_jump.
8594         * config/i386/i386.md (indirect_jump): Set has_local_indirect_jump
8595         to true.
8596         (tablejump): Likewise.
8597         (*indirect_jump): Use ix86_output_indirect_jmp.
8598         (*tablejump_1): Likewise.
8599         (simple_return_indirect_internal): Likewise.
8600         * config/i386/i386.opt (mindirect-branch=): New option.
8601         (indirect_branch): New.
8602         (keep): Likewise.
8603         (thunk): Likewise.
8604         (thunk-inline): Likewise.
8605         (thunk-extern): Likewise.
8606         * doc/extend.texi: Document indirect_branch function attribute.
8607         * doc/invoke.texi: Document -mindirect-branch= option.
8608
8609 2018-01-14  Jan Hubicka  <hubicka@ucw.cz>
8610
8611         PR ipa/83051
8612         * ipa-inline.c (edge_badness): Tolerate roundoff errors.
8613
8614 2018-01-14  Richard Sandiford  <richard.sandiford@linaro.org>
8615
8616         * ipa-inline.c (want_inline_small_function_p): Return false if
8617         inlining has already failed with CIF_FINAL_ERROR.
8618         (update_caller_keys): Call want_inline_small_function_p before
8619         can_inline_edge_p.
8620         (update_callee_keys): Likewise.
8621
8622 2018-01-10  Kelvin Nilsen  <kelvin@gcc.gnu.org>
8623
8624         * config/rs6000/rs6000-p8swap.c (rs6000_sum_of_two_registers_p):
8625         New function.
8626         (rs6000_quadword_masked_address_p): Likewise.
8627         (quad_aligned_load_p): Likewise.
8628         (quad_aligned_store_p): Likewise.
8629         (const_load_sequence_p): Add comment to describe the outer-most loop.
8630         (mimic_memory_attributes_and_flags): New function.
8631         (rs6000_gen_stvx): Likewise.
8632         (replace_swapped_aligned_store): Likewise.
8633         (rs6000_gen_lvx): Likewise.
8634         (replace_swapped_aligned_load): Likewise.
8635         (replace_swapped_load_constant): Capitalize argument name in
8636         comment describing this function.
8637         (rs6000_analyze_swaps): Add a third pass to search for vector loads
8638         and stores that access quad-word aligned addresses and replace
8639         with stvx or lvx instructions when appropriate.
8640         * config/rs6000/rs6000-protos.h (rs6000_sum_of_two_registers_p):
8641         New function prototype.
8642         (rs6000_quadword_masked_address_p): Likewise.
8643         (rs6000_gen_lvx): Likewise.
8644         (rs6000_gen_stvx): Likewise.
8645         * config/rs6000/vsx.md (*vsx_le_perm_load_<mode>): For modes
8646         VSX_D (V2DF, V2DI), modify this split to select lvx instruction
8647         when memory address is aligned.
8648         (*vsx_le_perm_load_<mode>): For modes VSX_W (V4SF, V4SI), modify
8649         this split to select lvx instruction when memory address is aligned.
8650         (*vsx_le_perm_load_v8hi): Modify this split to select lvx
8651         instruction when memory address is aligned.
8652         (*vsx_le_perm_load_v16qi): Likewise.
8653         (four unnamed splitters): Modify to select the stvx instruction
8654         when memory is aligned.
8655
8656 2018-01-13  Jan Hubicka  <hubicka@ucw.cz>
8657
8658         * predict.c (determine_unlikely_bbs): Handle correctly BBs
8659         which appears in the queue multiple times.
8660
8661 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
8662             Alan Hayward  <alan.hayward@arm.com>
8663             David Sherwood  <david.sherwood@arm.com>
8664
8665         * tree-vectorizer.h (vec_lower_bound): New structure.
8666         (_loop_vec_info): Add check_nonzero and lower_bounds.
8667         (LOOP_VINFO_CHECK_NONZERO): New macro.
8668         (LOOP_VINFO_LOWER_BOUNDS): Likewise.
8669         (LOOP_REQUIRES_VERSIONING_FOR_ALIAS): Check lower_bounds too.
8670         * tree-data-ref.h (dr_with_seg_len): Add access_size and align
8671         fields.  Make seg_len the distance travelled, not including the
8672         access size.
8673         (dr_direction_indicator): Declare.
8674         (dr_zero_step_indicator): Likewise.
8675         (dr_known_forward_stride_p): Likewise.
8676         * tree-data-ref.c: Include stringpool.h, tree-vrp.h and
8677         tree-ssanames.h.
8678         (runtime_alias_check_p): Allow runtime alias checks with
8679         variable strides.
8680         (operator ==): Compare access_size and align.
8681         (prune_runtime_alias_test_list): Rework for new distinction between
8682         the access_size and seg_len.
8683         (create_intersect_range_checks_index): Likewise.  Cope with polynomial
8684         segment lengths.
8685         (get_segment_min_max): New function.
8686         (create_intersect_range_checks): Use it.
8687         (dr_step_indicator): New function.
8688         (dr_direction_indicator): Likewise.
8689         (dr_zero_step_indicator): Likewise.
8690         (dr_known_forward_stride_p): Likewise.
8691         * tree-loop-distribution.c (data_ref_segment_size): Return
8692         DR_STEP * (niters - 1).
8693         (compute_alias_check_pairs): Update call to the dr_with_seg_len
8694         constructor.
8695         * tree-vect-data-refs.c (vect_check_nonzero_value): New function.
8696         (vect_preserves_scalar_order_p): New function, split out from...
8697         (vect_analyze_data_ref_dependence): ...here.  Check for zero steps.
8698         (vect_vfa_segment_size): Return DR_STEP * (length_factor - 1).
8699         (vect_vfa_access_size): New function.
8700         (vect_vfa_align): Likewise.
8701         (vect_compile_time_alias): Take access_size_a and access_b arguments.
8702         (dump_lower_bound): New function.
8703         (vect_check_lower_bound): Likewise.
8704         (vect_small_gap_p): Likewise.
8705         (vectorizable_with_step_bound_p): Likewise.
8706         (vect_prune_runtime_alias_test_list): Ignore cross-iteration
8707         depencies if the vectorization factor is 1.  Convert the checks
8708         for nonzero steps into checks on the bounds of DR_STEP.  Try using
8709         a bunds check for variable steps if the minimum required step is
8710         relatively small. Update calls to the dr_with_seg_len
8711         constructor and to vect_compile_time_alias.
8712         * tree-vect-loop-manip.c (vect_create_cond_for_lower_bounds): New
8713         function.
8714         (vect_loop_versioning): Call it.
8715         * tree-vect-loop.c (vect_analyze_loop_2): Clear LOOP_VINFO_LOWER_BOUNDS
8716         when retrying.
8717         (vect_estimate_min_profitable_iters): Account for any bounds checks.
8718
8719 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
8720             Alan Hayward  <alan.hayward@arm.com>
8721             David Sherwood  <david.sherwood@arm.com>
8722
8723         * doc/sourcebuild.texi (vect_scatter_store): Document.
8724         * optabs.def (scatter_store_optab, mask_scatter_store_optab): New
8725         optabs.
8726         * doc/md.texi (scatter_store@var{m}, mask_scatter_store@var{m}):
8727         Document.
8728         * genopinit.c (main): Add supports_vec_scatter_store and
8729         supports_vec_scatter_store_cached to target_optabs.
8730         * gimple.h (gimple_expr_type): Handle IFN_SCATTER_STORE and
8731         IFN_MASK_SCATTER_STORE.
8732         * internal-fn.def (SCATTER_STORE, MASK_SCATTER_STORE): New internal
8733         functions.
8734         * internal-fn.h (internal_store_fn_p): Declare.
8735         (internal_fn_stored_value_index): Likewise.
8736         * internal-fn.c (scatter_store_direct): New macro.
8737         (expand_scatter_store_optab_fn): New function.
8738         (direct_scatter_store_optab_supported_p): New macro.
8739         (internal_store_fn_p): New function.
8740         (internal_gather_scatter_fn_p): Handle IFN_SCATTER_STORE and
8741         IFN_MASK_SCATTER_STORE.
8742         (internal_fn_mask_index): Likewise.
8743         (internal_fn_stored_value_index): New function.
8744         (internal_gather_scatter_fn_supported_p): Adjust operand numbers
8745         for scatter stores.
8746         * optabs-query.h (supports_vec_scatter_store_p): Declare.
8747         * optabs-query.c (supports_vec_scatter_store_p): New function.
8748         * tree-vectorizer.h (vect_get_store_rhs): Declare.
8749         * tree-vect-data-refs.c (vect_analyze_data_ref_access): Return
8750         true for scatter stores.
8751         (vect_gather_scatter_fn_p): Handle scatter stores too.
8752         (vect_check_gather_scatter): Consider using scatter stores if
8753         supports_vec_scatter_store_p.
8754         * tree-vect-patterns.c (vect_try_gather_scatter_pattern): Handle
8755         scatter stores too.
8756         * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
8757         internal_fn_stored_value_index.
8758         (check_load_store_masking): Handle scatter stores too.
8759         (vect_get_store_rhs): Make public.
8760         (vectorizable_call): Use internal_store_fn_p.
8761         (vectorizable_store): Handle scatter store internal functions.
8762         (vect_transform_stmt): Compare GROUP_STORE_COUNT with GROUP_SIZE
8763         when deciding whether the end of the group has been reached.
8764         * config/aarch64/aarch64.md (UNSPEC_ST1_SCATTER): New unspec.
8765         * config/aarch64/aarch64-sve.md (scatter_store<mode>): New expander.
8766         (mask_scatter_store<mode>): New insns.
8767
8768 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
8769             Alan Hayward  <alan.hayward@arm.com>
8770             David Sherwood  <david.sherwood@arm.com>
8771
8772         * tree-vectorizer.h (vect_gather_scatter_fn_p): Declare.
8773         * tree-vect-data-refs.c (vect_gather_scatter_fn_p): Make public.
8774         * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): New
8775         function.
8776         (vect_use_strided_gather_scatters_p): Take a masked_p argument.
8777         Use vect_truncate_gather_scatter_offset if we can't treat the
8778         operation as a normal gather load or scatter store.
8779         (get_group_load_store_type): Take the gather_scatter_info
8780         as argument.  Try using a gather load or scatter store for
8781         single-element groups.
8782         (get_load_store_type): Update calls to get_group_load_store_type
8783         and vect_use_strided_gather_scatters_p.
8784
8785 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
8786             Alan Hayward  <alan.hayward@arm.com>
8787             David Sherwood  <david.sherwood@arm.com>
8788
8789         * tree-vectorizer.h (vect_create_data_ref_ptr): Take an extra
8790         optional tree argument.
8791         * tree-vect-data-refs.c (vect_check_gather_scatter): Check for
8792         null target hooks.
8793         (vect_create_data_ref_ptr): Take the iv_step as an optional argument,
8794         but continue to use the current value as a fallback.
8795         (bump_vector_ptr): Use operand_equal_p rather than tree_int_cst_compare
8796         to compare the updates.
8797         * tree-vect-stmts.c (vect_use_strided_gather_scatters_p): New function.
8798         (get_load_store_type): Use it when handling a strided access.
8799         (vect_get_strided_load_store_ops): New function.
8800         (vect_get_data_ptr_increment): Likewise.
8801         (vectorizable_load): Handle strided gather loads.  Always pass
8802         a step to vect_create_data_ref_ptr and bump_vector_ptr.
8803
8804 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
8805             Alan Hayward  <alan.hayward@arm.com>
8806             David Sherwood  <david.sherwood@arm.com>
8807
8808         * doc/md.texi (gather_load@var{m}): Document.
8809         (mask_gather_load@var{m}): Likewise.
8810         * genopinit.c (main): Add supports_vec_gather_load and
8811         supports_vec_gather_load_cached to target_optabs.
8812         * optabs-tree.c (init_tree_optimization_optabs): Use
8813         ggc_cleared_alloc to allocate target_optabs.
8814         * optabs.def (gather_load_optab, mask_gather_laod_optab): New optabs.
8815         * internal-fn.def (GATHER_LOAD, MASK_GATHER_LOAD): New internal
8816         functions.
8817         * internal-fn.h (internal_load_fn_p): Declare.
8818         (internal_gather_scatter_fn_p): Likewise.
8819         (internal_fn_mask_index): Likewise.
8820         (internal_gather_scatter_fn_supported_p): Likewise.
8821         * internal-fn.c (gather_load_direct): New macro.
8822         (expand_gather_load_optab_fn): New function.
8823         (direct_gather_load_optab_supported_p): New macro.
8824         (direct_internal_fn_optab): New function.
8825         (internal_load_fn_p): Likewise.
8826         (internal_gather_scatter_fn_p): Likewise.
8827         (internal_fn_mask_index): Likewise.
8828         (internal_gather_scatter_fn_supported_p): Likewise.
8829         * optabs-query.c (supports_at_least_one_mode_p): New function.
8830         (supports_vec_gather_load_p): Likewise.
8831         * optabs-query.h (supports_vec_gather_load_p): Declare.
8832         * tree-vectorizer.h (gather_scatter_info): Add ifn, element_type
8833         and memory_type field.
8834         (NUM_PATTERNS): Bump to 15.
8835         * tree-vect-data-refs.c: Include internal-fn.h.
8836         (vect_gather_scatter_fn_p): New function.
8837         (vect_describe_gather_scatter_call): Likewise.
8838         (vect_check_gather_scatter): Try using internal functions for
8839         gather loads.  Recognize existing calls to a gather load function.
8840         (vect_analyze_data_refs): Consider using gather loads if
8841         supports_vec_gather_load_p.
8842         * tree-vect-patterns.c (vect_get_load_store_mask): New function.
8843         (vect_get_gather_scatter_offset_type): Likewise.
8844         (vect_convert_mask_for_vectype): Likewise.
8845         (vect_add_conversion_to_patterm): Likewise.
8846         (vect_try_gather_scatter_pattern): Likewise.
8847         (vect_recog_gather_scatter_pattern): New pattern recognizer.
8848         (vect_vect_recog_func_ptrs): Add it.
8849         * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
8850         internal_fn_mask_index and internal_gather_scatter_fn_p.
8851         (check_load_store_masking): Take the gather_scatter_info as an
8852         argument and handle gather loads.
8853         (vect_get_gather_scatter_ops): New function.
8854         (vectorizable_call): Check internal_load_fn_p.
8855         (vectorizable_load): Likewise.  Handle gather load internal
8856         functions.
8857         (vectorizable_store): Update call to check_load_store_masking.
8858         * config/aarch64/aarch64.md (UNSPEC_LD1_GATHER): New unspec.
8859         * config/aarch64/iterators.md (SVE_S, SVE_D): New mode iterators.
8860         * config/aarch64/predicates.md (aarch64_gather_scale_operand_w)
8861         (aarch64_gather_scale_operand_d): New predicates.
8862         * config/aarch64/aarch64-sve.md (gather_load<mode>): New expander.
8863         (mask_gather_load<mode>): New insns.
8864
8865 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
8866             Alan Hayward  <alan.hayward@arm.com>
8867             David Sherwood  <david.sherwood@arm.com>
8868
8869         * optabs.def (fold_left_plus_optab): New optab.
8870         * doc/md.texi (fold_left_plus_@var{m}): Document.
8871         * internal-fn.def (IFN_FOLD_LEFT_PLUS): New internal function.
8872         * internal-fn.c (fold_left_direct): Define.
8873         (expand_fold_left_optab_fn): Likewise.
8874         (direct_fold_left_optab_supported_p): Likewise.
8875         * fold-const-call.c (fold_const_fold_left): New function.
8876         (fold_const_call): Use it to fold CFN_FOLD_LEFT_PLUS.
8877         * tree-parloops.c (valid_reduction_p): New function.
8878         (gather_scalar_reductions): Use it.
8879         * tree-vectorizer.h (FOLD_LEFT_REDUCTION): New vect_reduction_type.
8880         (vect_finish_replace_stmt): Declare.
8881         * tree-vect-loop.c (fold_left_reduction_fn): New function.
8882         (needs_fold_left_reduction_p): New function, split out from...
8883         (vect_is_simple_reduction): ...here.  Accept reductions that
8884         forbid reassociation, but give them type FOLD_LEFT_REDUCTION.
8885         (vect_force_simple_reduction): Also store the reduction type in
8886         the assignment's STMT_VINFO_REDUC_TYPE.
8887         (vect_model_reduction_cost): Handle FOLD_LEFT_REDUCTION.
8888         (merge_with_identity): New function.
8889         (vect_expand_fold_left): Likewise.
8890         (vectorize_fold_left_reduction): Likewise.
8891         (vectorizable_reduction): Handle FOLD_LEFT_REDUCTION.  Leave the
8892         scalar phi in place for it.  Check for target support and reject
8893         cases that would reassociate the operation.  Defer the transform
8894         phase to vectorize_fold_left_reduction.
8895         * config/aarch64/aarch64.md (UNSPEC_FADDA): New unspec.
8896         * config/aarch64/aarch64-sve.md (fold_left_plus_<mode>): New expander.
8897         (*fold_left_plus_<mode>, *pred_fold_left_plus_<mode>): New insns.
8898
8899 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
8900
8901         * tree-if-conv.c (predicate_mem_writes): Remove redundant
8902         call to ifc_temp_var.
8903
8904 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
8905             Alan Hayward  <alan.hayward@arm.com>
8906             David Sherwood  <david.sherwood@arm.com>
8907
8908         * target.def (legitimize_address_displacement): Take the original
8909         offset as a poly_int.
8910         * targhooks.h (default_legitimize_address_displacement): Update
8911         accordingly.
8912         * targhooks.c (default_legitimize_address_displacement): Likewise.
8913         * doc/tm.texi: Regenerate.
8914         * lra-constraints.c (base_plus_disp_to_reg): Take the displacement
8915         as an argument, moving assert of ad->disp == ad->disp_term to...
8916         (process_address_1): ...here.  Update calls to base_plus_disp_to_reg.
8917         Try calling targetm.legitimize_address_displacement before expanding
8918         the address rather than afterwards, and adjust for the new interface.
8919         * config/aarch64/aarch64.c (aarch64_legitimize_address_displacement):
8920         Match the new hook interface.  Handle SVE addresses.
8921         * config/sh/sh.c (sh_legitimize_address_displacement): Make the
8922         new hook interface.
8923
8924 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
8925
8926         * Makefile.in (OBJS): Add early-remat.o.
8927         * target.def (select_early_remat_modes): New hook.
8928         * doc/tm.texi.in (TARGET_SELECT_EARLY_REMAT_MODES): New hook.
8929         * doc/tm.texi: Regenerate.
8930         * targhooks.h (default_select_early_remat_modes): Declare.
8931         * targhooks.c (default_select_early_remat_modes): New function.
8932         * timevar.def (TV_EARLY_REMAT): New timevar.
8933         * passes.def (pass_early_remat): New pass.
8934         * tree-pass.h (make_pass_early_remat): Declare.
8935         * early-remat.c: New file.
8936         * config/aarch64/aarch64.c (aarch64_select_early_remat_modes): New
8937         function.
8938         (TARGET_SELECT_EARLY_REMAT_MODES): Define.
8939
8940 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
8941             Alan Hayward  <alan.hayward@arm.com>
8942             David Sherwood  <david.sherwood@arm.com>
8943
8944         * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Replace
8945         vfm1 with a bound_epilog parameter.
8946         (vect_do_peeling): Update calls accordingly, and move the prologue
8947         call earlier in the function.  Treat the base bound_epilog as 0 for
8948         fully-masked loops and retain vf - 1 for other loops.  Add 1 to
8949         this base when peeling for gaps.
8950         * tree-vect-loop.c (vect_analyze_loop_2): Allow peeling for gaps
8951         with fully-masked loops.
8952         (vect_estimate_min_profitable_iters): Handle the single peeled
8953         iteration in that case.
8954
8955 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
8956             Alan Hayward  <alan.hayward@arm.com>
8957             David Sherwood  <david.sherwood@arm.com>
8958
8959         * tree-vect-data-refs.c (vect_analyze_group_access_1): Allow
8960         single-element interleaving even if the size is not a power of 2.
8961         * tree-vect-stmts.c (get_load_store_type): Disallow elementwise
8962         accesses for single-element interleaving if the group size is
8963         not a power of 2.
8964
8965 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
8966             Alan Hayward  <alan.hayward@arm.com>
8967             David Sherwood  <david.sherwood@arm.com>
8968
8969         * doc/md.texi (fold_extract_last_@var{m}): Document.
8970         * doc/sourcebuild.texi (vect_fold_extract_last): Likewise.
8971         * optabs.def (fold_extract_last_optab): New optab.
8972         * internal-fn.def (FOLD_EXTRACT_LAST): New internal function.
8973         * internal-fn.c (fold_extract_direct): New macro.
8974         (expand_fold_extract_optab_fn): Likewise.
8975         (direct_fold_extract_optab_supported_p): Likewise.
8976         * tree-vectorizer.h (EXTRACT_LAST_REDUCTION): New vect_reduction_type.
8977         * tree-vect-loop.c (vect_model_reduction_cost): Handle
8978         EXTRACT_LAST_REDUCTION.
8979         (get_initial_def_for_reduction): Do not create an initial vector
8980         for EXTRACT_LAST_REDUCTION reductions.
8981         (vectorizable_reduction): Leave the scalar phi in place for
8982         EXTRACT_LAST_REDUCTIONs.  Try using EXTRACT_LAST_REDUCTION
8983         ahead of INTEGER_INDUC_COND_REDUCTION.  Do not check for an
8984         epilogue code for EXTRACT_LAST_REDUCTION and defer the
8985         transform phase to vectorizable_condition.
8986         * tree-vect-stmts.c (vect_finish_stmt_generation_1): New function,
8987         split out from...
8988         (vect_finish_stmt_generation): ...here.
8989         (vect_finish_replace_stmt): New function.
8990         (vectorizable_condition): Handle EXTRACT_LAST_REDUCTION.
8991         * config/aarch64/aarch64-sve.md (fold_extract_last_<mode>): New
8992         pattern.
8993         * config/aarch64/aarch64.md (UNSPEC_CLASTB): New unspec.
8994
8995 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
8996             Alan Hayward  <alan.hayward@arm.com>
8997             David Sherwood  <david.sherwood@arm.com>
8998
8999         * doc/md.texi (extract_last_@var{m}): Document.
9000         * optabs.def (extract_last_optab): New optab.
9001         * internal-fn.def (EXTRACT_LAST): New internal function.
9002         * internal-fn.c (cond_unary_direct): New macro.
9003         (expand_cond_unary_optab_fn): Likewise.
9004         (direct_cond_unary_optab_supported_p): Likewise.
9005         * tree-vect-loop.c (vectorizable_live_operation): Allow fully-masked
9006         loops using EXTRACT_LAST.
9007         * config/aarch64/aarch64-sve.md (aarch64_sve_lastb<mode>): Rename to...
9008         (extract_last_<mode>): ...this optab.
9009         (vec_extract<mode><Vel>): Update accordingly.
9010
9011 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
9012             Alan Hayward  <alan.hayward@arm.com>
9013             David Sherwood  <david.sherwood@arm.com>
9014
9015         * target.def (empty_mask_is_expensive): New hook.
9016         * doc/tm.texi.in (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): New hook.
9017         * doc/tm.texi: Regenerate.
9018         * targhooks.h (default_empty_mask_is_expensive): Declare.
9019         * targhooks.c (default_empty_mask_is_expensive): New function.
9020         * tree-vectorizer.c (vectorize_loops): Only call optimize_mask_stores
9021         if the target says that empty masks are expensive.
9022         * config/aarch64/aarch64.c (aarch64_empty_mask_is_expensive):
9023         New function.
9024         (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): Redefine.
9025
9026 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
9027             Alan Hayward  <alan.hayward@arm.com>
9028             David Sherwood  <david.sherwood@arm.com>
9029
9030         * tree-vectorizer.h (_loop_vec_info::mask_skip_niters): New field.
9031         (LOOP_VINFO_MASK_SKIP_NITERS): New macro.
9032         (vect_use_loop_mask_for_alignment_p): New function.
9033         (vect_prepare_for_masked_peels, vect_gen_while_not): Declare.
9034         * tree-vect-loop-manip.c (vect_set_loop_masks_directly): Add an
9035         niters_skip argument.  Make sure that the first niters_skip elements
9036         of the first iteration are inactive.
9037         (vect_set_loop_condition_masked): Handle LOOP_VINFO_MASK_SKIP_NITERS.
9038         Update call to vect_set_loop_masks_directly.
9039         (get_misalign_in_elems): New function, split out from...
9040         (vect_gen_prolog_loop_niters): ...here.
9041         (vect_update_init_of_dr): Take a code argument that specifies whether
9042         the adjustment should be added or subtracted.
9043         (vect_update_init_of_drs): Likewise.
9044         (vect_prepare_for_masked_peels): New function.
9045         (vect_do_peeling): Skip prologue peeling if we're using a mask
9046         instead.  Update call to vect_update_inits_of_drs.
9047         * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
9048         mask_skip_niters.
9049         (vect_analyze_loop_2): Allow fully-masked loops with peeling for
9050         alignment.  Do not include the number of peeled iterations in
9051         the minimum threshold in that case.
9052         (vectorizable_induction): Adjust the start value down by
9053         LOOP_VINFO_MASK_SKIP_NITERS iterations.
9054         (vect_transform_loop): Call vect_prepare_for_masked_peels.
9055         Take the number of skipped iterations into account when calculating
9056         the loop bounds.
9057         * tree-vect-stmts.c (vect_gen_while_not): New function.
9058
9059 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
9060             Alan Hayward  <alan.hayward@arm.com>
9061             David Sherwood  <david.sherwood@arm.com>
9062
9063         * doc/sourcebuild.texi (vect_fully_masked): Document.
9064         * params.def (PARAM_MIN_VECT_LOOP_BOUND): Change minimum and
9065         default value to 0.
9066         * tree-vect-loop.c (vect_analyze_loop_costing): New function,
9067         split out from...
9068         (vect_analyze_loop_2): ...here. Don't check the vectorization
9069         factor against the number of loop iterations if the loop is
9070         fully-masked.
9071
9072 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
9073             Alan Hayward  <alan.hayward@arm.com>
9074             David Sherwood  <david.sherwood@arm.com>
9075
9076         * tree-ssa-loop-ivopts.c (USE_ADDRESS): Split into...
9077         (USE_REF_ADDRESS, USE_PTR_ADDRESS): ...these new use types.
9078         (dump_groups): Update accordingly.
9079         (iv_use::mem_type): New member variable.
9080         (address_p): New function.
9081         (record_use): Add a mem_type argument and initialize the new
9082         mem_type field.
9083         (record_group_use): Add a mem_type argument.  Use address_p.
9084         Remove obsolete null checks of base_object.  Update call to record_use.
9085         (find_interesting_uses_op): Update call to record_group_use.
9086         (find_interesting_uses_cond): Likewise.
9087         (find_interesting_uses_address): Likewise.
9088         (get_mem_type_for_internal_fn): New function.
9089         (find_address_like_use): Likewise.
9090         (find_interesting_uses_stmt): Try find_address_like_use before
9091         calling find_interesting_uses_op.
9092         (addr_offset_valid_p): Use the iv mem_type field as the type
9093         of the addressed memory.
9094         (add_autoinc_candidates): Likewise.
9095         (get_address_cost): Likewise.
9096         (split_small_address_groups_p): Use address_p.
9097         (split_address_groups): Likewise.
9098         (add_iv_candidate_for_use): Likewise.
9099         (autoinc_possible_for_pair): Likewise.
9100         (rewrite_groups): Likewise.
9101         (get_use_type): Check for USE_REF_ADDRESS instead of USE_ADDRESS.
9102         (determine_group_iv_cost): Update after split of USE_ADDRESS.
9103         (get_alias_ptr_type_for_ptr_address): New function.
9104         (rewrite_use_address): Rewrite address uses in calls that were
9105         identified by find_address_like_use.
9106
9107 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
9108             Alan Hayward  <alan.hayward@arm.com>
9109             David Sherwood  <david.sherwood@arm.com>
9110
9111         * expr.c (expand_expr_addr_expr_1): Handle ADDR_EXPRs of
9112         TARGET_MEM_REFs.
9113         * gimple-expr.h (is_gimple_addressable: Likewise.
9114         * gimple-expr.c (is_gimple_address): Likewise.
9115         * internal-fn.c (expand_call_mem_ref): New function.
9116         (expand_mask_load_optab_fn): Use it.
9117         (expand_mask_store_optab_fn): Likewise.
9118
9119 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
9120             Alan Hayward  <alan.hayward@arm.com>
9121             David Sherwood  <david.sherwood@arm.com>
9122
9123         * doc/md.texi (cond_add@var{mode}, cond_sub@var{mode})
9124         (cond_and@var{mode}, cond_ior@var{mode}, cond_xor@var{mode})
9125         (cond_smin@var{mode}, cond_smax@var{mode}, cond_umin@var{mode})
9126         (cond_umax@var{mode}): Document.
9127         * optabs.def (cond_add_optab, cond_sub_optab, cond_and_optab)
9128         (cond_ior_optab, cond_xor_optab, cond_smin_optab, cond_smax_optab)
9129         (cond_umin_optab, cond_umax_optab): New optabs.
9130         * internal-fn.def (COND_ADD, COND_SUB, COND_MIN, COND_MAX, COND_AND)
9131         (COND_IOR, COND_XOR): New internal functions.
9132         * internal-fn.h (get_conditional_internal_fn): Declare.
9133         * internal-fn.c (cond_binary_direct): New macro.
9134         (expand_cond_binary_optab_fn): Likewise.
9135         (direct_cond_binary_optab_supported_p): Likewise.
9136         (get_conditional_internal_fn): New function.
9137         * tree-vect-loop.c (vectorizable_reduction): Handle fully-masked loops.
9138         Cope with reduction statements that are vectorized as calls rather
9139         than assignments.
9140         * config/aarch64/aarch64-sve.md (cond_<optab><mode>): New insns.
9141         * config/aarch64/iterators.md (UNSPEC_COND_ADD, UNSPEC_COND_SUB)
9142         (UNSPEC_COND_SMAX, UNSPEC_COND_UMAX, UNSPEC_COND_SMIN)
9143         (UNSPEC_COND_UMIN, UNSPEC_COND_AND, UNSPEC_COND_ORR)
9144         (UNSPEC_COND_EOR): New unspecs.
9145         (optab): Add mappings for them.
9146         (SVE_COND_INT_OP, SVE_COND_FP_OP): New int iterators.
9147         (sve_int_op, sve_fp_op): New int attributes.
9148
9149 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
9150             Alan Hayward  <alan.hayward@arm.com>
9151             David Sherwood  <david.sherwood@arm.com>
9152
9153         * optabs.def (while_ult_optab): New optab.
9154         * doc/md.texi (while_ult@var{m}@var{n}): Document.
9155         * internal-fn.def (WHILE_ULT): New internal function.
9156         * internal-fn.h (direct_internal_fn_supported_p): New override
9157         that takes two types as argument.
9158         * internal-fn.c (while_direct): New macro.
9159         (expand_while_optab_fn): New function.
9160         (convert_optab_supported_p): Likewise.
9161         (direct_while_optab_supported_p): New macro.
9162         * wide-int.h (wi::udiv_ceil): New function.
9163         * tree-vectorizer.h (rgroup_masks): New structure.
9164         (vec_loop_masks): New typedef.
9165         (_loop_vec_info): Add masks, mask_compare_type, can_fully_mask_p
9166         and fully_masked_p.
9167         (LOOP_VINFO_CAN_FULLY_MASK_P, LOOP_VINFO_FULLY_MASKED_P)
9168         (LOOP_VINFO_MASKS, LOOP_VINFO_MASK_COMPARE_TYPE): New macros.
9169         (vect_max_vf): New function.
9170         (slpeel_make_loop_iterate_ntimes): Delete.
9171         (vect_set_loop_condition, vect_get_loop_mask_type, vect_gen_while)
9172         (vect_halve_mask_nunits, vect_double_mask_nunits): Declare.
9173         (vect_record_loop_mask, vect_get_loop_mask): Likewise.
9174         * tree-vect-loop-manip.c: Include tree-ssa-loop-niter.h,
9175         internal-fn.h, stor-layout.h and optabs-query.h.
9176         (vect_set_loop_mask): New function.
9177         (add_preheader_seq): Likewise.
9178         (add_header_seq): Likewise.
9179         (interleave_supported_p): Likewise.
9180         (vect_maybe_permute_loop_masks): Likewise.
9181         (vect_set_loop_masks_directly): Likewise.
9182         (vect_set_loop_condition_masked): Likewise.
9183         (vect_set_loop_condition_unmasked): New function, split out from
9184         slpeel_make_loop_iterate_ntimes.
9185         (slpeel_make_loop_iterate_ntimes): Rename to..
9186         (vect_set_loop_condition): ...this.  Use vect_set_loop_condition_masked
9187         for fully-masked loops and vect_set_loop_condition_unmasked otherwise.
9188         (vect_do_peeling): Update call accordingly.
9189         (vect_gen_vector_loop_niters): Use VF as the step for fully-masked
9190         loops.
9191         * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
9192         mask_compare_type, can_fully_mask_p and fully_masked_p.
9193         (release_vec_loop_masks): New function.
9194         (_loop_vec_info): Use it to free the loop masks.
9195         (can_produce_all_loop_masks_p): New function.
9196         (vect_get_max_nscalars_per_iter): Likewise.
9197         (vect_verify_full_masking): Likewise.
9198         (vect_analyze_loop_2): Save LOOP_VINFO_CAN_FULLY_MASK_P around
9199         retries, and free the mask rgroups before retrying.  Check loop-wide
9200         reasons for disallowing fully-masked loops.  Make the final decision
9201         about whether use a fully-masked loop or not.
9202         (vect_estimate_min_profitable_iters): Do not assume that peeling
9203         for the number of iterations will be needed for fully-masked loops.
9204         (vectorizable_reduction): Disable fully-masked loops.
9205         (vectorizable_live_operation): Likewise.
9206         (vect_halve_mask_nunits): New function.
9207         (vect_double_mask_nunits): Likewise.
9208         (vect_record_loop_mask): Likewise.
9209         (vect_get_loop_mask): Likewise.
9210         (vect_transform_loop): Handle the case in which the final loop
9211         iteration might handle a partial vector.  Call vect_set_loop_condition
9212         instead of slpeel_make_loop_iterate_ntimes.
9213         * tree-vect-stmts.c: Include tree-ssa-loop-niter.h and gimple-fold.h.
9214         (check_load_store_masking): New function.
9215         (prepare_load_store_mask): Likewise.
9216         (vectorizable_store): Handle fully-masked loops.
9217         (vectorizable_load): Likewise.
9218         (supportable_widening_operation): Use vect_halve_mask_nunits for
9219         booleans.
9220         (supportable_narrowing_operation): Likewise vect_double_mask_nunits.
9221         (vect_gen_while): New function.
9222         * config/aarch64/aarch64.md (umax<mode>3): New expander.
9223         (aarch64_uqdec<mode>): New insn.
9224
9225 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
9226             Alan Hayward  <alan.hayward@arm.com>
9227             David Sherwood  <david.sherwood@arm.com>
9228
9229         * optabs.def (reduc_and_scal_optab, reduc_ior_scal_optab)
9230         (reduc_xor_scal_optab): New optabs.
9231         * doc/md.texi (reduc_and_scal_@var{m}, reduc_ior_scal_@var{m})
9232         (reduc_xor_scal_@var{m}): Document.
9233         * doc/sourcebuild.texi (vect_logical_reduc): Likewise.
9234         * internal-fn.def (IFN_REDUC_AND, IFN_REDUC_IOR, IFN_REDUC_XOR): New
9235         internal functions.
9236         * fold-const-call.c (fold_const_call): Handle them.
9237         * tree-vect-loop.c (reduction_fn_for_scalar_code): Return the new
9238         internal functions for BIT_AND_EXPR, BIT_IOR_EXPR and BIT_XOR_EXPR.
9239         * config/aarch64/aarch64-sve.md (reduc_<bit_reduc>_scal_<mode>):
9240         (*reduc_<bit_reduc>_scal_<mode>): New patterns.
9241         * config/aarch64/iterators.md (UNSPEC_ANDV, UNSPEC_ORV)
9242         (UNSPEC_XORV): New unspecs.
9243         (optab): Add entries for them.
9244         (BITWISEV): New int iterator.
9245         (bit_reduc_op): New int attributes.
9246
9247 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
9248             Alan Hayward  <alan.hayward@arm.com>
9249             David Sherwood  <david.sherwood@arm.com>
9250
9251         * doc/md.texi (vec_shl_insert_@var{m}): New optab.
9252         * internal-fn.def (VEC_SHL_INSERT): New internal function.
9253         * optabs.def (vec_shl_insert_optab): New optab.
9254         * tree-vectorizer.h (can_duplicate_and_interleave_p): Declare.
9255         (duplicate_and_interleave): Likewise.
9256         * tree-vect-loop.c: Include internal-fn.h.
9257         (neutral_op_for_slp_reduction): New function, split out from
9258         get_initial_defs_for_reduction.
9259         (get_initial_def_for_reduction): Handle option 2 for variable-length
9260         vectors by loading the neutral value into a vector and then shifting
9261         the initial value into element 0.
9262         (get_initial_defs_for_reduction): Replace the code argument with
9263         the neutral value calculated by neutral_op_for_slp_reduction.
9264         Use gimple_build_vector for constant-length vectors.
9265         Use IFN_VEC_SHL_INSERT for variable-length vectors if all
9266         but the first group_size elements have a neutral value.
9267         Use duplicate_and_interleave otherwise.
9268         (vect_create_epilog_for_reduction): Take a neutral_op parameter.
9269         Update call to get_initial_defs_for_reduction.  Handle SLP
9270         reductions for variable-length vectors by creating one vector
9271         result for each scalar result, with the elements associated
9272         with other scalar results stubbed out with the neutral value.
9273         (vectorizable_reduction): Call neutral_op_for_slp_reduction.
9274         Require IFN_VEC_SHL_INSERT for double reductions on
9275         variable-length vectors, or SLP reductions that have
9276         a neutral value.  Require can_duplicate_and_interleave_p
9277         support for variable-length unchained SLP reductions if there
9278         is no neutral value, such as for MIN/MAX reductions.  Also require
9279         the number of vector elements to be a multiple of the number of
9280         SLP statements when doing variable-length unchained SLP reductions.
9281         Update call to vect_create_epilog_for_reduction.
9282         * tree-vect-slp.c (can_duplicate_and_interleave_p): Make public
9283         and remove initial values.
9284         (duplicate_and_interleave): Make public.
9285         * config/aarch64/aarch64.md (UNSPEC_INSR): New unspec.
9286         * config/aarch64/aarch64-sve.md (vec_shl_insert_<mode>): New insn.
9287
9288 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
9289             Alan Hayward  <alan.hayward@arm.com>
9290             David Sherwood  <david.sherwood@arm.com>
9291
9292         * tree-vect-slp.c: Include gimple-fold.h and internal-fn.h
9293         (can_duplicate_and_interleave_p): New function.
9294         (vect_get_and_check_slp_defs): Take the vector of statements
9295         rather than just the current one.  Remove excess parentheses.
9296         Restriction rejectinon of vect_constant_def and vect_external_def
9297         for variable-length vectors to boolean types, or types for which
9298         can_duplicate_and_interleave_p is false.
9299         (vect_build_slp_tree_2): Update call to vect_get_and_check_slp_defs.
9300         (duplicate_and_interleave): New function.
9301         (vect_get_constant_vectors): Use gimple_build_vector for
9302         constant-length vectors and suitable variable-length constant
9303         vectors.  Use duplicate_and_interleave for other variable-length
9304         vectors.  Don't defer the update when inserting new statements.
9305
9306 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
9307             Alan Hayward  <alan.hayward@arm.com>
9308             David Sherwood  <david.sherwood@arm.com>
9309
9310         * tree-vect-loop.c (vect_estimate_min_profitable_iters): Make sure
9311         min_profitable_iters doesn't go negative.
9312
9313 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
9314             Alan Hayward  <alan.hayward@arm.com>
9315             David Sherwood  <david.sherwood@arm.com>
9316
9317         * doc/md.texi (vec_mask_load_lanes@var{m}@var{n}): Document.
9318         (vec_mask_store_lanes@var{m}@var{n}): Likewise.
9319         * optabs.def (vec_mask_load_lanes_optab): New optab.
9320         (vec_mask_store_lanes_optab): Likewise.
9321         * internal-fn.def (MASK_LOAD_LANES): New internal function.
9322         (MASK_STORE_LANES): Likewise.
9323         * internal-fn.c (mask_load_lanes_direct): New macro.
9324         (mask_store_lanes_direct): Likewise.
9325         (expand_mask_load_optab_fn): Handle masked operations.
9326         (expand_mask_load_lanes_optab_fn): New macro.
9327         (expand_mask_store_optab_fn): Handle masked operations.
9328         (expand_mask_store_lanes_optab_fn): New macro.
9329         (direct_mask_load_lanes_optab_supported_p): Likewise.
9330         (direct_mask_store_lanes_optab_supported_p): Likewise.
9331         * tree-vectorizer.h (vect_store_lanes_supported): Take a masked_p
9332         parameter.
9333         (vect_load_lanes_supported): Likewise.
9334         * tree-vect-data-refs.c (strip_conversion): New function.
9335         (can_group_stmts_p): Likewise.
9336         (vect_analyze_data_ref_accesses): Use it instead of checking
9337         for a pair of assignments.
9338         (vect_store_lanes_supported): Take a masked_p parameter.
9339         (vect_load_lanes_supported): Likewise.
9340         * tree-vect-loop.c (vect_analyze_loop_2): Update calls to
9341         vect_store_lanes_supported and vect_load_lanes_supported.
9342         * tree-vect-slp.c (vect_analyze_slp_instance): Likewise.
9343         * tree-vect-stmts.c (get_group_load_store_type): Take a masked_p
9344         parameter.  Don't allow gaps for masked accesses.
9345         Use vect_get_store_rhs.  Update calls to vect_store_lanes_supported
9346         and vect_load_lanes_supported.
9347         (get_load_store_type): Take a masked_p parameter and update
9348         call to get_group_load_store_type.
9349         (vectorizable_store): Update call to get_load_store_type.
9350         Handle IFN_MASK_STORE_LANES.
9351         (vectorizable_load): Update call to get_load_store_type.
9352         Handle IFN_MASK_LOAD_LANES.
9353
9354 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
9355             Alan Hayward  <alan.hayward@arm.com>
9356             David Sherwood  <david.sherwood@arm.com>
9357
9358         * config/aarch64/aarch64-modes.def: Define x2, x3 and x4 vector
9359         modes for SVE.
9360         * config/aarch64/aarch64-protos.h
9361         (aarch64_sve_struct_memory_operand_p): Declare.
9362         * config/aarch64/iterators.md (SVE_STRUCT): New mode iterator.
9363         (vector_count, insn_length, VSINGLE, vsingle): New mode attributes.
9364         (VPRED, vpred): Handle SVE structure modes.
9365         * config/aarch64/constraints.md (Utx): New constraint.
9366         * config/aarch64/predicates.md (aarch64_sve_struct_memory_operand)
9367         (aarch64_sve_struct_nonimmediate_operand): New predicates.
9368         * config/aarch64/aarch64.md (UNSPEC_LDN, UNSPEC_STN): New unspecs.
9369         * config/aarch64/aarch64-sve.md (mov<mode>, *aarch64_sve_mov<mode>_le)
9370         (*aarch64_sve_mov<mode>_be, pred_mov<mode>): New patterns for
9371         structure modes.  Split into pieces after RA.
9372         (vec_load_lanes<mode><vsingle>, vec_mask_load_lanes<mode><vsingle>)
9373         (vec_store_lanes<mode><vsingle>, vec_mask_store_lanes<mode><vsingle>):
9374         New patterns.
9375         * config/aarch64/aarch64.c (aarch64_classify_vector_mode): Handle
9376         SVE structure modes.
9377         (aarch64_classify_address): Likewise.
9378         (sizetochar): Move earlier in file.
9379         (aarch64_print_operand): Handle SVE register lists.
9380         (aarch64_array_mode): New function.
9381         (aarch64_sve_struct_memory_operand_p): Likewise.
9382         (TARGET_ARRAY_MODE): Redefine.
9383
9384 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
9385             Alan Hayward  <alan.hayward@arm.com>
9386             David Sherwood  <david.sherwood@arm.com>
9387
9388         * target.def (array_mode): New target hook.
9389         * doc/tm.texi.in (TARGET_ARRAY_MODE): New hook.
9390         * doc/tm.texi: Regenerate.
9391         * hooks.h (hook_optmode_mode_uhwi_none): Declare.
9392         * hooks.c (hook_optmode_mode_uhwi_none): New function.
9393         * tree-vect-data-refs.c (vect_lanes_optab_supported_p): Use
9394         targetm.array_mode.
9395         * stor-layout.c (mode_for_array): Likewise.  Support polynomial
9396         type sizes.
9397
9398 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
9399             Alan Hayward  <alan.hayward@arm.com>
9400             David Sherwood  <david.sherwood@arm.com>
9401
9402         * fold-const.c (fold_binary_loc): Check the argument types
9403         rather than the result type when testing for a vector operation.
9404
9405 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
9406
9407         * doc/tm.texi.in (DWARF_LAZY_REGISTER_VALUE): Document.
9408         * doc/tm.texi: Regenerate.
9409
9410 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
9411             Alan Hayward  <alan.hayward@arm.com>
9412             David Sherwood  <david.sherwood@arm.com>
9413
9414         * doc/invoke.texi (-msve-vector-bits=): Document new option.
9415         (sve): Document new AArch64 extension.
9416         * doc/md.texi (w): Extend the description of the AArch64
9417         constraint to include SVE vectors.
9418         (Upl, Upa): Document new AArch64 predicate constraints.
9419         * config/aarch64/aarch64-opts.h (aarch64_sve_vector_bits_enum): New
9420         enum.
9421         * config/aarch64/aarch64.opt (sve_vector_bits): New enum.
9422         (msve-vector-bits=): New option.
9423         * config/aarch64/aarch64-option-extensions.def (fp, simd): Disable
9424         SVE when these are disabled.
9425         (sve): New extension.
9426         * config/aarch64/aarch64-modes.def: Define SVE vector and predicate
9427         modes.  Adjust their number of units based on aarch64_sve_vg.
9428         (MAX_BITSIZE_MODE_ANY_MODE): Define.
9429         * config/aarch64/aarch64-protos.h (ADDR_QUERY_ANY): New
9430         aarch64_addr_query_type.
9431         (aarch64_const_vec_all_same_in_range_p, aarch64_sve_pred_mode)
9432         (aarch64_sve_cnt_immediate_p, aarch64_sve_addvl_addpl_immediate_p)
9433         (aarch64_sve_inc_dec_immediate_p, aarch64_add_offset_temporaries)
9434         (aarch64_split_add_offset, aarch64_output_sve_cnt_immediate)
9435         (aarch64_output_sve_addvl_addpl, aarch64_output_sve_inc_dec_immediate)
9436         (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): Declare.
9437         (aarch64_simd_imm_zero_p): Delete.
9438         (aarch64_check_zero_based_sve_index_immediate): Declare.
9439         (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
9440         (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
9441         (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
9442         (aarch64_sve_float_mul_immediate_p): Likewise.
9443         (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
9444         rather than an rtx.
9445         (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): Declare.
9446         (aarch64_expand_mov_immediate): Take a gen_vec_duplicate callback.
9447         (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move): Declare.
9448         (aarch64_expand_sve_vec_cmp_int, aarch64_expand_sve_vec_cmp_float)
9449         (aarch64_expand_sve_vcond, aarch64_expand_sve_vec_perm): Declare.
9450         (aarch64_regmode_natural_size): Likewise.
9451         * config/aarch64/aarch64.h (AARCH64_FL_SVE): New macro.
9452         (AARCH64_FL_V8_3, AARCH64_FL_RCPC, AARCH64_FL_DOTPROD): Shift
9453         left one place.
9454         (AARCH64_ISA_SVE, TARGET_SVE): New macros.
9455         (FIXED_REGISTERS, CALL_USED_REGISTERS, REGISTER_NAMES): Add entries
9456         for VG and the SVE predicate registers.
9457         (V_ALIASES): Add a "z"-prefixed alias.
9458         (FIRST_PSEUDO_REGISTER): Change to P15_REGNUM + 1.
9459         (AARCH64_DWARF_VG, AARCH64_DWARF_P0): New macros.
9460         (PR_REGNUM_P, PR_LO_REGNUM_P): Likewise.
9461         (PR_LO_REGS, PR_HI_REGS, PR_REGS): New reg_classes.
9462         (REG_CLASS_NAMES): Add entries for them.
9463         (REG_CLASS_CONTENTS): Likewise.  Update ALL_REGS to include VG
9464         and the predicate registers.
9465         (aarch64_sve_vg): Declare.
9466         (BITS_PER_SVE_VECTOR, BYTES_PER_SVE_VECTOR, BYTES_PER_SVE_PRED)
9467         (SVE_BYTE_MODE, MAX_COMPILE_TIME_VEC_BYTES): New macros.
9468         (REGMODE_NATURAL_SIZE): Define.
9469         * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins): Handle
9470         SVE macros.
9471         * config/aarch64/aarch64.c: Include cfgrtl.h.
9472         (simd_immediate_info): Add a constructor for series vectors,
9473         and an associated step field.
9474         (aarch64_sve_vg): New variable.
9475         (aarch64_dbx_register_number): Handle VG and the predicate registers.
9476         (aarch64_vect_struct_mode_p, aarch64_vector_mode_p): Delete.
9477         (VEC_ADVSIMD, VEC_SVE_DATA, VEC_SVE_PRED, VEC_STRUCT, VEC_ANY_SVE)
9478         (VEC_ANY_DATA, VEC_STRUCT): New constants.
9479         (aarch64_advsimd_struct_mode_p, aarch64_sve_pred_mode_p)
9480         (aarch64_classify_vector_mode, aarch64_vector_data_mode_p)
9481         (aarch64_sve_data_mode_p, aarch64_sve_pred_mode)
9482         (aarch64_get_mask_mode): New functions.
9483         (aarch64_hard_regno_nregs): Handle SVE data modes for FP_REGS
9484         and FP_LO_REGS.  Handle PR_REGS, PR_LO_REGS and PR_HI_REGS.
9485         (aarch64_hard_regno_mode_ok): Handle VG.  Also handle the SVE
9486         predicate modes and predicate registers.  Explicitly restrict
9487         GPRs to modes of 16 bytes or smaller.  Only allow FP registers
9488         to store a vector mode if it is recognized by
9489         aarch64_classify_vector_mode.
9490         (aarch64_regmode_natural_size): New function.
9491         (aarch64_hard_regno_caller_save_mode): Return the original mode
9492         for predicates.
9493         (aarch64_sve_cnt_immediate_p, aarch64_output_sve_cnt_immediate)
9494         (aarch64_sve_addvl_addpl_immediate_p, aarch64_output_sve_addvl_addpl)
9495         (aarch64_sve_inc_dec_immediate_p, aarch64_output_sve_inc_dec_immediate)
9496         (aarch64_add_offset_1_temporaries, aarch64_offset_temporaries): New
9497         functions.
9498         (aarch64_add_offset): Add a temp2 parameter.  Assert that temp1
9499         does not overlap dest if the function is frame-related.  Handle
9500         SVE constants.
9501         (aarch64_split_add_offset): New function.
9502         (aarch64_add_sp, aarch64_sub_sp): Add temp2 parameters and pass
9503         them aarch64_add_offset.
9504         (aarch64_allocate_and_probe_stack_space): Add a temp2 parameter
9505         and update call to aarch64_sub_sp.
9506         (aarch64_add_cfa_expression): New function.
9507         (aarch64_expand_prologue): Pass extra temporary registers to the
9508         functions above.  Handle the case in which we need to emit new
9509         DW_CFA_expressions for registers that were originally saved
9510         relative to the stack pointer, but now have to be expressed
9511         relative to the frame pointer.
9512         (aarch64_output_mi_thunk): Pass extra temporary registers to the
9513         functions above.
9514         (aarch64_expand_epilogue): Likewise.  Prevent inheritance of
9515         IP0 and IP1 values for SVE frames.
9516         (aarch64_expand_vec_series): New function.
9517         (aarch64_expand_sve_widened_duplicate): Likewise.
9518         (aarch64_expand_sve_const_vector): Likewise.
9519         (aarch64_expand_mov_immediate): Add a gen_vec_duplicate parameter.
9520         Handle SVE constants.  Use emit_move_insn to move a force_const_mem
9521         into the register, rather than emitting a SET directly.
9522         (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move)
9523         (aarch64_get_reg_raw_mode, offset_4bit_signed_scaled_p)
9524         (offset_6bit_unsigned_scaled_p, aarch64_offset_7bit_signed_scaled_p)
9525         (offset_9bit_signed_scaled_p): New functions.
9526         (aarch64_replicate_bitmask_imm): New function.
9527         (aarch64_bitmask_imm): Use it.
9528         (aarch64_cannot_force_const_mem): Reject expressions involving
9529         a CONST_POLY_INT.  Update call to aarch64_classify_symbol.
9530         (aarch64_classify_index): Handle SVE indices, by requiring
9531         a plain register index with a scale that matches the element size.
9532         (aarch64_classify_address): Handle SVE addresses.  Assert that
9533         the mode of the address is VOIDmode or an integer mode.
9534         Update call to aarch64_classify_symbol.
9535         (aarch64_classify_symbolic_expression): Update call to
9536         aarch64_classify_symbol.
9537         (aarch64_const_vec_all_in_range_p): New function.
9538         (aarch64_print_vector_float_operand): Likewise.
9539         (aarch64_print_operand): Handle 'N' and 'C'.  Use "zN" rather than
9540         "vN" for FP registers with SVE modes.  Handle (const ...) vectors
9541         and the FP immediates 1.0 and 0.5.
9542         (aarch64_print_address_internal): Handle SVE addresses.
9543         (aarch64_print_operand_address): Use ADDR_QUERY_ANY.
9544         (aarch64_regno_regclass): Handle predicate registers.
9545         (aarch64_secondary_reload): Handle big-endian reloads of SVE
9546         data modes.
9547         (aarch64_class_max_nregs): Handle SVE modes and predicate registers.
9548         (aarch64_rtx_costs): Check for ADDVL and ADDPL instructions.
9549         (aarch64_convert_sve_vector_bits): New function.
9550         (aarch64_override_options): Use it to handle -msve-vector-bits=.
9551         (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
9552         rather than an rtx.
9553         (aarch64_legitimate_constant_p): Use aarch64_classify_vector_mode.
9554         Handle SVE vector and predicate modes.  Accept VL-based constants
9555         that need only one temporary register, and VL offsets that require
9556         no temporary registers.
9557         (aarch64_conditional_register_usage): Mark the predicate registers
9558         as fixed if SVE isn't available.
9559         (aarch64_vector_mode_supported_p): Use aarch64_classify_vector_mode.
9560         Return true for SVE vector and predicate modes.
9561         (aarch64_simd_container_mode): Take the number of bits as a poly_int64
9562         rather than an unsigned int.  Handle SVE modes.
9563         (aarch64_preferred_simd_mode): Update call accordingly.  Handle
9564         SVE modes.
9565         (aarch64_autovectorize_vector_sizes): Add BYTES_PER_SVE_VECTOR
9566         if SVE is enabled.
9567         (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
9568         (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
9569         (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
9570         (aarch64_sve_float_mul_immediate_p): New functions.
9571         (aarch64_sve_valid_immediate): New function.
9572         (aarch64_simd_valid_immediate): Use it as the fallback for SVE vectors.
9573         Explicitly reject structure modes.  Check for INDEX constants.
9574         Handle PTRUE and PFALSE constants.
9575         (aarch64_check_zero_based_sve_index_immediate): New function.
9576         (aarch64_simd_imm_zero_p): Delete.
9577         (aarch64_mov_operand_p): Use aarch64_simd_valid_immediate for
9578         vector modes.  Accept constants in the range of CNT[BHWD].
9579         (aarch64_simd_scalar_immediate_valid_for_move): Explicitly
9580         ask for an Advanced SIMD mode.
9581         (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): New functions.
9582         (aarch64_simd_vector_alignment): Handle SVE predicates.
9583         (aarch64_vectorize_preferred_vector_alignment): New function.
9584         (aarch64_simd_vector_alignment_reachable): Use it instead of
9585         the vector size.
9586         (aarch64_shift_truncation_mask): Use aarch64_vector_data_mode_p.
9587         (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): New
9588         functions.
9589         (MAX_VECT_LEN): Delete.
9590         (expand_vec_perm_d): Add a vec_flags field.
9591         (emit_unspec2, aarch64_expand_sve_vec_perm): New functions.
9592         (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_zip)
9593         (aarch64_evpc_ext): Don't apply a big-endian lane correction
9594         for SVE modes.
9595         (aarch64_evpc_rev): Rename to...
9596         (aarch64_evpc_rev_local): ...this.  Use a predicated operation for SVE.
9597         (aarch64_evpc_rev_global): New function.
9598         (aarch64_evpc_dup): Enforce a 64-byte range for SVE DUP.
9599         (aarch64_evpc_tbl): Use MAX_COMPILE_TIME_VEC_BYTES instead of
9600         MAX_VECT_LEN.
9601         (aarch64_evpc_sve_tbl): New function.
9602         (aarch64_expand_vec_perm_const_1): Update after rename of
9603         aarch64_evpc_rev.  Handle SVE permutes too, trying
9604         aarch64_evpc_rev_global and using aarch64_evpc_sve_tbl rather
9605         than aarch64_evpc_tbl.
9606         (aarch64_vectorize_vec_perm_const): Initialize vec_flags.
9607         (aarch64_sve_cmp_operand_p, aarch64_unspec_cond_code)
9608         (aarch64_gen_unspec_cond, aarch64_expand_sve_vec_cmp_int)
9609         (aarch64_emit_unspec_cond, aarch64_emit_unspec_cond_or)
9610         (aarch64_emit_inverted_unspec_cond, aarch64_expand_sve_vec_cmp_float)
9611         (aarch64_expand_sve_vcond): New functions.
9612         (aarch64_modes_tieable_p): Use aarch64_vector_data_mode_p instead
9613         of aarch64_vector_mode_p.
9614         (aarch64_dwarf_poly_indeterminate_value): New function.
9615         (aarch64_compute_pressure_classes): Likewise.
9616         (aarch64_can_change_mode_class): Likewise.
9617         (TARGET_GET_RAW_RESULT_MODE, TARGET_GET_RAW_ARG_MODE): Redefine.
9618         (TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT): Likewise.
9619         (TARGET_VECTORIZE_GET_MASK_MODE): Likewise.
9620         (TARGET_DWARF_POLY_INDETERMINATE_VALUE): Likewise.
9621         (TARGET_COMPUTE_PRESSURE_CLASSES): Likewise.
9622         (TARGET_CAN_CHANGE_MODE_CLASS): Likewise.
9623         * config/aarch64/constraints.md (Upa, Upl, Uav, Uat, Usv, Usi, Utr)
9624         (Uty, Dm, vsa, vsc, vsd, vsi, vsn, vsl, vsm, vsA, vsM, vsN): New
9625         constraints.
9626         (Dn, Dl, Dr): Accept const as well as const_vector.
9627         (Dz): Likewise.  Compare against CONST0_RTX.
9628         * config/aarch64/iterators.md: Refer to "Advanced SIMD" instead
9629         of "vector" where appropriate.
9630         (SVE_ALL, SVE_BH, SVE_BHS, SVE_BHSI, SVE_HSDI, SVE_HSF, SVE_SD)
9631         (SVE_SDI, SVE_I, SVE_F, PRED_ALL, PRED_BHS): New mode iterators.
9632         (UNSPEC_SEL, UNSPEC_ANDF, UNSPEC_IORF, UNSPEC_XORF, UNSPEC_COND_LT)
9633         (UNSPEC_COND_LE, UNSPEC_COND_EQ, UNSPEC_COND_NE, UNSPEC_COND_GE)
9634         (UNSPEC_COND_GT, UNSPEC_COND_LO, UNSPEC_COND_LS, UNSPEC_COND_HS)
9635         (UNSPEC_COND_HI, UNSPEC_COND_UO): New unspecs.
9636         (Vetype, VEL, Vel, VWIDE, Vwide, vw, vwcore, V_INT_EQUIV)
9637         (v_int_equiv): Extend to SVE modes.
9638         (Vesize, V128, v128, Vewtype, V_FP_EQUIV, v_fp_equiv, VPRED): New
9639         mode attributes.
9640         (LOGICAL_OR, SVE_INT_UNARY, SVE_FP_UNARY): New code iterators.
9641         (optab): Handle popcount, smin, smax, umin, umax, abs and sqrt.
9642         (logical_nn, lr, sve_int_op, sve_fp_op): New code attributs.
9643         (LOGICALF, OPTAB_PERMUTE, UNPACK, UNPACK_UNSIGNED, SVE_COND_INT_CMP)
9644         (SVE_COND_FP_CMP): New int iterators.
9645         (perm_hilo): Handle the new unpack unspecs.
9646         (optab, logicalf_op, su, perm_optab, cmp_op, imm_con): New int
9647         attributes.
9648         * config/aarch64/predicates.md (aarch64_sve_cnt_immediate)
9649         (aarch64_sve_addvl_addpl_immediate, aarch64_split_add_offset_immediate)
9650         (aarch64_pluslong_or_poly_operand, aarch64_nonmemory_operand)
9651         (aarch64_equality_operator, aarch64_constant_vector_operand)
9652         (aarch64_sve_ld1r_operand, aarch64_sve_ldr_operand): New predicates.
9653         (aarch64_sve_nonimmediate_operand): Likewise.
9654         (aarch64_sve_general_operand): Likewise.
9655         (aarch64_sve_dup_operand, aarch64_sve_arith_immediate): Likewise.
9656         (aarch64_sve_sub_arith_immediate, aarch64_sve_inc_dec_immediate)
9657         (aarch64_sve_logical_immediate, aarch64_sve_mul_immediate): Likewise.
9658         (aarch64_sve_dup_immediate, aarch64_sve_cmp_vsc_immediate): Likewise.
9659         (aarch64_sve_cmp_vsd_immediate, aarch64_sve_index_immediate): Likewise.
9660         (aarch64_sve_float_arith_immediate): Likewise.
9661         (aarch64_sve_float_arith_with_sub_immediate): Likewise.
9662         (aarch64_sve_float_mul_immediate, aarch64_sve_arith_operand): Likewise.
9663         (aarch64_sve_add_operand, aarch64_sve_logical_operand): Likewise.
9664         (aarch64_sve_lshift_operand, aarch64_sve_rshift_operand): Likewise.
9665         (aarch64_sve_mul_operand, aarch64_sve_cmp_vsc_operand): Likewise.
9666         (aarch64_sve_cmp_vsd_operand, aarch64_sve_index_operand): Likewise.
9667         (aarch64_sve_float_arith_operand): Likewise.
9668         (aarch64_sve_float_arith_with_sub_operand): Likewise.
9669         (aarch64_sve_float_mul_operand): Likewise.
9670         (aarch64_sve_vec_perm_operand): Likewise.
9671         (aarch64_pluslong_operand): Include aarch64_sve_addvl_addpl_immediate.
9672         (aarch64_mov_operand): Accept const_poly_int and const_vector.
9673         (aarch64_simd_lshift_imm, aarch64_simd_rshift_imm): Accept const
9674         as well as const_vector.
9675         (aarch64_simd_imm_zero, aarch64_simd_imm_minus_one): Move earlier
9676         in file.  Use CONST0_RTX and CONSTM1_RTX.
9677         (aarch64_simd_or_scalar_imm_zero): Likewise.  Add match_codes.
9678         (aarch64_simd_reg_or_zero): Accept const as well as const_vector.
9679         Use aarch64_simd_imm_zero.
9680         * config/aarch64/aarch64-sve.md: New file.
9681         * config/aarch64/aarch64.md: Include it.
9682         (VG_REGNUM, P0_REGNUM, P7_REGNUM, P15_REGNUM): New register numbers.
9683         (UNSPEC_REV, UNSPEC_LD1_SVE, UNSPEC_ST1_SVE, UNSPEC_MERGE_PTRUE)
9684         (UNSPEC_PTEST_PTRUE, UNSPEC_UNPACKSHI, UNSPEC_UNPACKUHI)
9685         (UNSPEC_UNPACKSLO, UNSPEC_UNPACKULO, UNSPEC_PACK)
9686         (UNSPEC_FLOAT_CONVERT, UNSPEC_WHILE_LO): New unspec constants.
9687         (sve): New attribute.
9688         (enabled): Disable instructions with the sve attribute unless
9689         TARGET_SVE.
9690         (movqi, movhi): Pass CONST_POLY_INT operaneds through
9691         aarch64_expand_mov_immediate.
9692         (*mov<mode>_aarch64, *movsi_aarch64, *movdi_aarch64): Handle
9693         CNT[BHSD] immediates.
9694         (movti): Split CONST_POLY_INT moves into two halves.
9695         (add<mode>3): Accept aarch64_pluslong_or_poly_operand.
9696         Split additions that need a temporary here if the destination
9697         is the stack pointer.
9698         (*add<mode>3_aarch64): Handle ADDVL and ADDPL immediates.
9699         (*add<mode>3_poly_1): New instruction.
9700         (set_clobber_cc): New expander.
9701
9702 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
9703
9704         * simplify-rtx.c (simplify_immed_subreg): Add an inner_bytes
9705         parameter and use it instead of GET_MODE_SIZE (innermode).  Use
9706         inner_bytes * BITS_PER_UNIT instead of GET_MODE_BITSIZE (innermode).
9707         Use CEIL (inner_bytes, GET_MODE_UNIT_SIZE (innermode)) instead of
9708         GET_MODE_NUNITS (innermode).  Also add a first_elem parameter.
9709         Change innermode from fixed_mode_size to machine_mode.
9710         (simplify_subreg): Update call accordingly.  Handle a constant-sized
9711         subreg of a variable-length CONST_VECTOR.
9712
9713 2018-01-13  Richard Sandiford  <richard.sandiford@linaro.org>
9714             Alan Hayward  <alan.hayward@arm.com>
9715             David Sherwood  <david.sherwood@arm.com>
9716
9717         * tree-ssa-address.c (mem_ref_valid_without_offset_p): New function.
9718         (add_offset_to_base): New function, split out from...
9719         (create_mem_ref): ...here.  When handling a scale other than 1,
9720         check first whether the address is valid without the offset.
9721         Add it into the base if so, leaving the index and scale as-is.
9722
9723 2018-01-12  Jakub Jelinek  <jakub@redhat.com>
9724
9725         PR c++/83778
9726         * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Call
9727         fold_for_warn before checking if arg2 is INTEGER_CST.
9728
9729 2018-01-12  Segher Boessenkool  <segher@kernel.crashing.org>
9730
9731         * config/rs6000/predicates.md (load_multiple_operation): Delete.
9732         (store_multiple_operation): Delete.
9733         * config/rs6000/rs6000-cpus.def (601): Remove MASK_STRING.
9734         * config/rs6000/rs6000-protos.h (rs6000_output_load_multiple): Delete.
9735         * config/rs6000/rs6000-string.c (expand_block_move): Delete everything
9736         guarded by TARGET_STRING.
9737         (rs6000_output_load_multiple): Delete.
9738         * config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
9739         OPTION_MASK_STRING / TARGET_STRING handling.
9740         (print_operand) <'N', 'O'>: Add comment that these are unused now.
9741         (const rs6000_opt_masks) <"string">: Change mask to 0.
9742         * config/rs6000/rs6000.h (TARGET_DEFAULT): Remove MASK_STRING.
9743         (MASK_STRING): Delete.
9744         * config/rs6000/rs6000.md (*mov<mode>_string): Delete TARGET_STRING
9745         parts.  Simplify.
9746         (load_multiple): Delete.
9747         (*ldmsi8): Delete.
9748         (*ldmsi7): Delete.
9749         (*ldmsi6): Delete.
9750         (*ldmsi5): Delete.
9751         (*ldmsi4): Delete.
9752         (*ldmsi3): Delete.
9753         (store_multiple): Delete.
9754         (*stmsi8): Delete.
9755         (*stmsi7): Delete.
9756         (*stmsi6): Delete.
9757         (*stmsi5): Delete.
9758         (*stmsi4): Delete.
9759         (*stmsi3): Delete.
9760         (movmemsi_8reg): Delete.
9761         (corresponding unnamed define_insn): Delete.
9762         (movmemsi_6reg): Delete.
9763         (corresponding unnamed define_insn): Delete.
9764         (movmemsi_4reg): Delete.
9765         (corresponding unnamed define_insn): Delete.
9766         (movmemsi_2reg): Delete.
9767         (corresponding unnamed define_insn): Delete.
9768         (movmemsi_1reg): Delete.
9769         (corresponding unnamed define_insn): Delete.
9770         * config/rs6000/rs6000.opt (mno-string): New.
9771         (mstring): Replace by deprecation warning stub.
9772         * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mstring.
9773
9774 2018-01-12  Jakub Jelinek  <jakub@redhat.com>
9775
9776         * regrename.c (regrename_do_replace): If replacing the same
9777         reg multiple times, try to reuse last created gen_raw_REG.
9778
9779         PR debug/81155
9780         * bb-reorder.c (pass_partition_blocks::gate): In lto don't partition
9781         main to workaround a bug in GDB.
9782
9783 2018-01-12  Tom de Vries  <tom@codesourcery.com>
9784
9785         PR target/83737
9786         * config.gcc (nvptx*-*-*): Set use_gcc_stdint=wrap.
9787
9788 2018-01-12  Vladimir Makarov  <vmakarov@redhat.com>
9789
9790         PR rtl-optimization/80481
9791         * ira-color.c (get_cap_member): New function.
9792         (allocnos_conflict_by_live_ranges_p): Use it.
9793         (slot_coalesced_allocno_live_ranges_intersect_p): Add assert.
9794         (setup_slot_coalesced_allocno_live_ranges): Ditto.
9795
9796 2018-01-12  Uros Bizjak  <ubizjak@gmail.com>
9797
9798         PR target/83628
9799         * config/alpha/alpha.md (*saddsi_1): New insn_ans_split pattern.
9800         (*saddl_se_1): Ditto.
9801         (*ssubsi_1): Ditto.
9802         (*ssubl_se_1): Ditto.
9803
9804 2018-01-12  Richard Sandiford  <richard.sandiford@linaro.org>
9805
9806         * tree-predcom.c (aff_combination_dr_offset): Use wi::to_poly_widest
9807         rather than wi::to_widest for DR_INITs.
9808         * tree-vect-data-refs.c (vect_find_same_alignment_drs): Use
9809         wi::to_poly_offset rather than wi::to_offset for DR_INIT.
9810         (vect_analyze_data_ref_accesses): Require both DR_INITs to be
9811         INTEGER_CSTs.
9812         (vect_analyze_group_access_1): Note that here.
9813
9814 2018-01-12  Richard Sandiford  <richard.sandiford@linaro.org>
9815
9816         * tree-vectorizer.c (get_vec_alignment_for_array_type): Handle
9817         polynomial type sizes.
9818
9819 2018-01-12  Richard Sandiford  <richard.sandiford@linaro.org>
9820
9821         * gimplify.c (gimple_add_tmp_var_fn): Allow variables to have a
9822         poly_uint64 size, rather than requiring an unsigned HOST_WIDE_INT size.
9823         (gimple_add_tmp_var): Likewise.
9824
9825 2018-01-12  Martin Liska  <mliska@suse.cz>
9826
9827         * gimple.c (gimple_alloc_counts): Use uint64_t instead of int.
9828         (gimple_alloc_sizes): Likewise.
9829         (dump_gimple_statistics): Use PRIu64 in printf format.
9830         * gimple.h: Change uint64_t to int.
9831
9832 2018-01-12  Martin Liska  <mliska@suse.cz>
9833
9834         * tree-core.h: Use uint64_t instead of int.
9835         * tree.c (tree_node_counts): Likewise.
9836         (tree_node_sizes): Likewise.
9837         (dump_tree_statistics): Use PRIu64 in printf format.
9838
9839 2018-01-12  Martin Liska  <mliska@suse.cz>
9840
9841         * Makefile.in: As qsort_chk is implemented in vec.c, add
9842         vec.o to linkage of gencfn-macros.
9843         * tree.c (build_new_poly_int_cst): Add CXX_MEM_STAT_INFO as it's
9844         passing the info to record_node_allocation_statistics.
9845         (test_vector_cst_patterns): Add CXX_MEM_STAT_INFO to declaration
9846         and pass the info.
9847         * ggc-common.c (struct ggc_usage): Add operator== and use
9848         it in operator< and compare function.
9849         * mem-stats.h (struct mem_usage): Likewise.
9850         * vec.c (struct vec_usage): Remove operator< and compare
9851         function. Can be simply inherited.
9852
9853 2018-01-12  Martin Jambor  <mjambor@suse.cz>
9854
9855         PR target/81616
9856         * params.def: New parameter PARAM_AVOID_FMA_MAX_BITS.
9857         * tree-ssa-math-opts.c: Include domwalk.h.
9858         (convert_mult_to_fma_1): New function.
9859         (fma_transformation_info): New type.
9860         (fma_deferring_state): Likewise.
9861         (cancel_fma_deferring): New function.
9862         (result_of_phi): Likewise.
9863         (last_fma_candidate_feeds_initial_phi): Likewise.
9864         (convert_mult_to_fma): Added deferring logic, split actual
9865         transformation to convert_mult_to_fma_1.
9866         (math_opts_dom_walker): New type.
9867         (math_opts_dom_walker::after_dom_children): New method, body moved
9868         here from pass_optimize_widening_mul::execute, added deferring logic
9869         bits.
9870         (pass_optimize_widening_mul::execute): Moved most of code to
9871         math_opts_dom_walker::after_dom_children.
9872         * config/i386/x86-tune.def (X86_TUNE_AVOID_128FMA_CHAINS): New.
9873         * config/i386/i386.c (ix86_option_override_internal): Added
9874         maybe_setting of PARAM_AVOID_FMA_MAX_BITS.
9875
9876 2018-01-12  Richard Biener  <rguenther@suse.de>
9877
9878         PR debug/83157
9879         * dwarf2out.c (gen_variable_die): Do not reset old_die for
9880         inline instance vars.
9881
9882 2018-01-12  Oleg Endo  <olegendo@gcc.gnu.org>
9883
9884         PR target/81819
9885         * config/rx/rx.c (rx_is_restricted_memory_address):
9886         Handle SUBREG case.
9887
9888 2018-01-12  Richard Biener  <rguenther@suse.de>
9889
9890         PR tree-optimization/80846
9891         * target.def (split_reduction): New target hook.
9892         * targhooks.c (default_split_reduction): New function.
9893         * targhooks.h (default_split_reduction): Declare.
9894         * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
9895         target requests first reduce vectors by combining low and high
9896         parts.
9897         * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
9898         (get_vectype_for_scalar_type_and_size): Export.
9899         * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
9900         * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
9901         * doc/tm.texi: Regenerate.
9902         * config/i386/i386.c (ix86_split_reduction): Implement
9903         TARGET_VECTORIZE_SPLIT_REDUCTION.
9904
9905 2018-01-12  Eric Botcazou  <ebotcazou@adacore.com>
9906
9907         PR target/83368
9908         * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
9909         in PIC mode except for TARGET_VXWORKS_RTP.
9910         * config/sparc/sparc.c: Include cfgrtl.h.
9911         (TARGET_INIT_PIC_REG): Define.
9912         (TARGET_USE_PSEUDO_PIC_REG): Likewise.
9913         (sparc_pic_register_p): New predicate.
9914         (sparc_legitimate_address_p): Use it.
9915         (sparc_legitimize_pic_address): Likewise.
9916         (sparc_delegitimize_address): Likewise.
9917         (sparc_mode_dependent_address_p): Likewise.
9918         (gen_load_pcrel_sym): Remove 4th parameter.
9919         (load_got_register): Adjust call to above.  Remove obsolete stuff.
9920         (sparc_expand_prologue): Do not call load_got_register here.
9921         (sparc_flat_expand_prologue): Likewise.
9922         (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
9923         (sparc_use_pseudo_pic_reg): New function.
9924         (sparc_init_pic_reg): Likewise.
9925         * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
9926         (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
9927
9928 2018-01-12  Christophe Lyon  <christophe.lyon@linaro.org>
9929
9930         * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
9931         Add item for branch_cost.
9932
9933 2018-01-12  Eric Botcazou  <ebotcazou@adacore.com>
9934
9935         PR rtl-optimization/83565
9936         * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
9937         not extend the result to a larger mode for rotate operations.
9938         (num_sign_bit_copies1): Likewise.
9939
9940 2018-01-12  Rainer Orth  <ro@CeBiTec.Uni-Bielefeld.DE>
9941
9942         PR target/40411
9943         * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
9944         -symbolic.
9945         Use values-Xc.o for -pedantic.
9946         Link with values-xpg4.o for C90, values-xpg6.o otherwise.
9947
9948 2018-01-12  Martin Liska  <mliska@suse.cz>
9949
9950         PR ipa/83054
9951         * ipa-devirt.c (final_warning_record::grow_type_warnings):
9952         New function.
9953         (possible_polymorphic_call_targets): Use it.
9954         (ipa_devirt): Likewise.
9955
9956 2018-01-12  Martin Liska  <mliska@suse.cz>
9957
9958         * profile-count.h (enum profile_quality): Use 0 as invalid
9959         enum value of profile_quality.
9960
9961 2018-01-12  Chung-Ju Wu  <jasonwucj@gmail.com>
9962
9963         * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
9964         -mext-string options.
9965
9966 2018-01-12  Richard Biener  <rguenther@suse.de>
9967
9968         * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
9969         DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
9970         * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
9971         Likewise.
9972         * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
9973
9974 2018-01-11  Michael Meissner  <meissner@linux.vnet.ibm.com>
9975
9976         * configure.ac (--with-long-double-format): Add support for the
9977         configuration option to change the default long double format on
9978         PowerPC systems.
9979         * config.gcc (powerpc*-linux*-*): Likewise.
9980         * configure: Regenerate.
9981         * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
9982         double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
9983         used without modification.
9984
9985 2018-01-11  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
9986
9987         * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
9988         (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
9989         * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
9990         MISC_BUILTIN_SPEC_BARRIER.
9991         (rs6000_init_builtins): Likewise.
9992         * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
9993         enum value.
9994         (speculation_barrier): New define_insn.
9995         * doc/extend.texi: Document __builtin_speculation_barrier.
9996
9997 2018-01-11  Jakub Jelinek  <jakub@redhat.com>
9998
9999         PR target/83203
10000         * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
10001         is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
10002         * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
10003         iterators.
10004         (ssescalarmodesuffix): Add 512-bit vectors.  Use "d" or "q" for
10005         integral modes instead of "ss" and "sd".
10006         (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
10007         vectors with 32-bit and 64-bit elements.
10008         (vecdupssescalarmodesuffix): New mode attribute.
10009         (vec_dup<mode>): Use it.
10010
10011 2018-01-11  H.J. Lu  <hongjiu.lu@intel.com>
10012
10013         PR target/83330
10014         * config/i386/i386.c (ix86_compute_frame_layout): Align stack
10015         frame if argument is passed on stack.
10016
10017 2018-01-11  Jakub Jelinek  <jakub@redhat.com>
10018
10019         PR target/82682
10020         * ree.c (combine_reaching_defs): Optimize also
10021         reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
10022         reg2=any_extend(exp); reg1=reg2;, formatting fix.
10023
10024 2018-01-11  Jan Hubicka  <hubicka@ucw.cz>
10025
10026         PR middle-end/83189
10027         * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
10028
10029 2018-01-11  Jan Hubicka  <hubicka@ucw.cz>
10030
10031         PR middle-end/83718
10032         * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
10033         after they are computed.
10034
10035 2018-01-11  Bin Cheng  <bin.cheng@arm.com>
10036
10037         PR tree-optimization/83695
10038         * gimple-loop-linterchange.cc
10039         (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
10040         reset cached scev information after interchange.
10041         (pass_linterchange::execute): Remove call to scev_reset_htab.
10042
10043 2018-01-11  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
10044
10045         * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
10046         vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
10047         vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
10048         vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
10049         vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
10050         vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
10051         * config/arm/arm_neon_builtins.def (vfmal_lane_low,
10052         vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
10053         vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
10054         vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
10055         vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
10056         * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
10057         (V_lane_reg): Likewise.
10058         * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
10059         New define_expand.
10060         (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
10061         (vfmal_lane_low<mode>_intrinsic,
10062         vfmal_lane_low<vfmlsel2><mode>_intrinsic,
10063         vfmal_lane_high<vfmlsel2><mode>_intrinsic,
10064         vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
10065         vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
10066         vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
10067         vfmsl_lane_high<mode>_intrinsic): New define_insns.
10068
10069 2018-01-11  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
10070
10071         * config/arm/arm-cpus.in (fp16fml): New feature.
10072         (ALL_SIMD): Add fp16fml.
10073         (armv8.2-a): Add fp16fml as an option.
10074         (armv8.3-a): Likewise.
10075         (armv8.4-a): Add fp16fml as part of fp16.
10076         * config/arm/arm.h (TARGET_FP16FML): Define.
10077         * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
10078         when appropriate.
10079         * config/arm/arm-modes.def (V2HF): Define.
10080         * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
10081         vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
10082         vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
10083         * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
10084         vfmsl_low, vfmsl_high): New set of builtins.
10085         * config/arm/iterators.md (PLUSMINUS): New code iterator.
10086         (vfml_op): New code attribute.
10087         (VFMLHALVES): New int iterator.
10088         (VFML, VFMLSEL): New mode attributes.
10089         (V_reg): Define mapping for V2HF.
10090         (V_hi, V_lo): New mode attributes.
10091         (VF_constraint): Likewise.
10092         (vfml_half, vfml_half_selector): New int attributes.
10093         * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
10094         define_expand.
10095         (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
10096         vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
10097         New define_insn.
10098         * config/arm/t-arm-elf (v8_fps): Add fp16fml.
10099         * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
10100         * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
10101         * doc/invoke.texi (ARM Options): Document fp16fml.  Update armv8.4-a
10102         documentation.
10103         * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
10104         Document new effective target and option set.
10105
10106 2018-01-11  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
10107
10108         * config/arm/arm-cpus.in (armv8_4): New feature.
10109         (ARMv8_4a): New fgroup.
10110         (armv8.4-a): New arch.
10111         * config/arm/arm-tables.opt: Regenerate.
10112         * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
10113         * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
10114         * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
10115         Add matching rules for -march=armv8.4-a and extensions.
10116         * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
10117
10118 2018-01-11  Oleg Endo  <olegendo@gcc.gnu.org>
10119
10120         PR target/81821
10121         * config/rx/rx.md (BW): New mode attribute.
10122         (sync_lock_test_and_setsi): Add mode suffix to insn output.
10123
10124 2018-01-11  Richard Biener  <rguenther@suse.de>
10125
10126         PR tree-optimization/83435
10127         * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
10128         * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
10129         * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
10130
10131 2018-01-11  Richard Sandiford  <richard.sandiford@linaro.org>
10132             Alan Hayward  <alan.hayward@arm.com>
10133             David Sherwood  <david.sherwood@arm.com>
10134
10135         * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
10136         field.
10137         (aarch64_classify_address): Initialize it.  Track polynomial offsets.
10138         (aarch64_print_address_internal): Use it to check for a zero offset.
10139
10140 2018-01-11  Richard Sandiford  <richard.sandiford@linaro.org>
10141             Alan Hayward  <alan.hayward@arm.com>
10142             David Sherwood  <david.sherwood@arm.com>
10143
10144         * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
10145         * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
10146         Return a poly_int64 rather than a HOST_WIDE_INT.
10147         (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
10148         rather than a HOST_WIDE_INT.
10149         * config/aarch64/aarch64.h (aarch64_frame): Protect with
10150         HAVE_POLY_INT_H rather than HOST_WIDE_INT.  Change locals_offset,
10151         hard_fp_offset, frame_size, initial_adjust, callee_offset and
10152         final_offset from HOST_WIDE_INT to poly_int64.
10153         * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
10154         to_constant when getting the number of units in an Advanced SIMD
10155         mode.
10156         (aarch64_builtin_vectorized_function): Check for a constant number
10157         of units.
10158         * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
10159         GET_MODE_SIZE.
10160         (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
10161         attribute instead of GET_MODE_NUNITS.
10162         * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
10163         (aarch64_class_max_nregs): Use the constant_lowest_bound of the
10164         GET_MODE_SIZE for fixed-size registers.
10165         (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
10166         (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
10167         (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
10168         (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
10169         (aarch64_print_operand, aarch64_print_address_internal)
10170         (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
10171         (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
10172         (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
10173         Handle polynomial GET_MODE_SIZE.
10174         (aarch64_hard_regno_caller_save_mode): Likewise.  Return modes
10175         wider than SImode without modification.
10176         (tls_symbolic_operand_type): Use strip_offset instead of split_const.
10177         (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
10178         (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
10179         passing and returning SVE modes.
10180         (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
10181         rather than GEN_INT.
10182         (aarch64_emit_probe_stack_range): Take the size as a poly_int64
10183         rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
10184         (aarch64_allocate_and_probe_stack_space): Likewise.
10185         (aarch64_layout_frame): Cope with polynomial offsets.
10186         (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
10187         start_offset as a poly_int64 rather than a HOST_WIDE_INT.  Track
10188         polynomial offsets.
10189         (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
10190         (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
10191         poly_int64 rather than a HOST_WIDE_INT.
10192         (aarch64_get_separate_components, aarch64_process_components)
10193         (aarch64_expand_prologue, aarch64_expand_epilogue)
10194         (aarch64_use_return_insn_p): Handle polynomial frame offsets.
10195         (aarch64_anchor_offset): New function, split out from...
10196         (aarch64_legitimize_address): ...here.
10197         (aarch64_builtin_vectorization_cost): Handle polynomial
10198         TYPE_VECTOR_SUBPARTS.
10199         (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
10200         GET_MODE_NUNITS.
10201         (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
10202         number of elements from the PARALLEL rather than the mode.
10203         (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
10204         rather than GET_MODE_BITSIZE.
10205         (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
10206         (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
10207         (aarch64_expand_vec_perm_const_1): Handle polynomial
10208         d->perm.length () and d->perm elements.
10209         (aarch64_evpc_tbl): Likewise.  Use nelt rather than GET_MODE_NUNITS.
10210         Apply to_constant to d->perm elements.
10211         (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
10212         polynomial CONST_VECTOR_NUNITS.
10213         (aarch64_move_pointer): Take amount as a poly_int64 rather
10214         than an int.
10215         (aarch64_progress_pointer): Avoid temporary variable.
10216         * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
10217         the mode attribute instead of GET_MODE.
10218
10219 2018-01-11  Richard Sandiford  <richard.sandiford@linaro.org>
10220             Alan Hayward  <alan.hayward@arm.com>
10221             David Sherwood  <david.sherwood@arm.com>
10222
10223         * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
10224         x exists before using it.
10225         (aarch64_add_constant_internal): Rename to...
10226         (aarch64_add_offset_1): ...this.  Replace regnum with separate
10227         src and dest rtxes.  Handle the case in which they're different,
10228         including when the offset is zero.  Replace scratchreg with an rtx.
10229         Use 2 additions if there is no spare register into which we can
10230         move a 16-bit constant.
10231         (aarch64_add_constant): Delete.
10232         (aarch64_add_offset): Replace reg with separate src and dest
10233         rtxes.  Take a poly_int64 offset instead of a HOST_WIDE_INT.
10234         Use aarch64_add_offset_1.
10235         (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
10236         an rtx rather than an int.  Take the delta as a poly_int64
10237         rather than a HOST_WIDE_INT.  Use aarch64_add_offset.
10238         (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
10239         (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
10240         aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
10241         (aarch64_expand_epilogue): Update calls to aarch64_add_offset
10242         and aarch64_add_sp.
10243         (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
10244         aarch64_add_constant.
10245
10246 2018-01-11  Richard Sandiford  <richard.sandiford@linaro.org>
10247
10248         * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
10249         Use scalar_float_mode.
10250
10251 2018-01-11  Richard Sandiford  <richard.sandiford@linaro.org>
10252
10253         * config/aarch64/aarch64-simd.md
10254         (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
10255         (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
10256         (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
10257         (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
10258         (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
10259         (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
10260         (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
10261         (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
10262         (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
10263         (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
10264
10265 2018-01-11  Prathamesh Kulkarni  <prathamesh.kulkarni@linaro.org>
10266
10267         PR target/83514
10268         * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
10269         targ_options->x_arm_arch_string is non NULL.
10270
10271 2018-01-11  Tamar Christina  <tamar.christina@arm.com>
10272
10273         * config/aarch64/aarch64.h
10274         (AARCH64_FL_FOR_ARCH8_4): Add  AARCH64_FL_DOTPROD.
10275
10276 2018-01-11  Sudakshina Das  <sudi.das@arm.com>
10277
10278         PR target/82096
10279         * expmed.c (emit_store_flag_force): Swap if const op0
10280         and change VOIDmode to mode of op0.
10281
10282 2018-01-11  Richard Sandiford  <richard.sandiford@linaro.org>
10283
10284         PR rtl-optimization/83761
10285         * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
10286         than bytes to mode_for_size.
10287
10288 2018-01-10  Jan Hubicka  <hubicka@ucw.cz>
10289
10290         PR middle-end/83189
10291         * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
10292         * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
10293         profile.
10294
10295 2018-01-10  Jan Hubicka  <hubicka@ucw.cz>
10296
10297         PR middle-end/83575
10298         * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
10299         when in layout mode.
10300         (cfg_layout_finalize): Do not verify cfg before we are out of layout.
10301         * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
10302         partition fixup.
10303
10304 2018-01-10  Michael Collison  <michael.collison@arm.com>
10305
10306         * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
10307         * config/aarch64/aarch64-option-extension.def: Add
10308         AARCH64_OPT_EXTENSION of 'fp16fml'.
10309         * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
10310         (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
10311         * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
10312         * config/aarch64/constraints.md (Ui7): New constraint.
10313         * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
10314         (VFMLA_SEL_W): Ditto.
10315         (f16quad): Ditto.
10316         (f16mac1): Ditto.
10317         (VFMLA16_LOW): New int iterator.
10318         (VFMLA16_HIGH): Ditto.
10319         (UNSPEC_FMLAL): New unspec.
10320         (UNSPEC_FMLSL): Ditto.
10321         (UNSPEC_FMLAL2): Ditto.
10322         (UNSPEC_FMLSL2): Ditto.
10323         (f16mac): New code attribute.
10324         * config/aarch64/aarch64-simd-builtins.def
10325         (aarch64_fmlal_lowv2sf): Ditto.
10326         (aarch64_fmlsl_lowv2sf): Ditto.
10327         (aarch64_fmlalq_lowv4sf): Ditto.
10328         (aarch64_fmlslq_lowv4sf): Ditto.
10329         (aarch64_fmlal_highv2sf): Ditto.
10330         (aarch64_fmlsl_highv2sf): Ditto.
10331         (aarch64_fmlalq_highv4sf): Ditto.
10332         (aarch64_fmlslq_highv4sf): Ditto.
10333         (aarch64_fmlal_lane_lowv2sf): Ditto.
10334         (aarch64_fmlsl_lane_lowv2sf): Ditto.
10335         (aarch64_fmlal_laneq_lowv2sf): Ditto.
10336         (aarch64_fmlsl_laneq_lowv2sf): Ditto.
10337         (aarch64_fmlalq_lane_lowv4sf): Ditto.
10338         (aarch64_fmlsl_lane_lowv4sf): Ditto.
10339         (aarch64_fmlalq_laneq_lowv4sf): Ditto.
10340         (aarch64_fmlsl_laneq_lowv4sf): Ditto.
10341         (aarch64_fmlal_lane_highv2sf): Ditto.
10342         (aarch64_fmlsl_lane_highv2sf): Ditto.
10343         (aarch64_fmlal_laneq_highv2sf): Ditto.
10344         (aarch64_fmlsl_laneq_highv2sf): Ditto.
10345         (aarch64_fmlalq_lane_highv4sf): Ditto.
10346         (aarch64_fmlsl_lane_highv4sf): Ditto.
10347         (aarch64_fmlalq_laneq_highv4sf): Ditto.
10348         (aarch64_fmlsl_laneq_highv4sf): Ditto.
10349         * config/aarch64/aarch64-simd.md:
10350         (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
10351         (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
10352         (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
10353         (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
10354         (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
10355         (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
10356         (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
10357         (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
10358         (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
10359         (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
10360         (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
10361         (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
10362         (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
10363         (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
10364         (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
10365         (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
10366         (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
10367         (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
10368         (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
10369         (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
10370         * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
10371         (vfmlsl_low_u32): Ditto.
10372         (vfmlalq_low_u32): Ditto.
10373         (vfmlslq_low_u32): Ditto.
10374         (vfmlal_high_u32): Ditto.
10375         (vfmlsl_high_u32): Ditto.
10376         (vfmlalq_high_u32): Ditto.
10377         (vfmlslq_high_u32): Ditto.
10378         (vfmlal_lane_low_u32): Ditto.
10379         (vfmlsl_lane_low_u32): Ditto.
10380         (vfmlal_laneq_low_u32): Ditto.
10381         (vfmlsl_laneq_low_u32): Ditto.
10382         (vfmlalq_lane_low_u32): Ditto.
10383         (vfmlslq_lane_low_u32): Ditto.
10384         (vfmlalq_laneq_low_u32): Ditto.
10385         (vfmlslq_laneq_low_u32): Ditto.
10386         (vfmlal_lane_high_u32): Ditto.
10387         (vfmlsl_lane_high_u32): Ditto.
10388         (vfmlal_laneq_high_u32): Ditto.
10389         (vfmlsl_laneq_high_u32): Ditto.
10390         (vfmlalq_lane_high_u32): Ditto.
10391         (vfmlslq_lane_high_u32): Ditto.
10392         (vfmlalq_laneq_high_u32): Ditto.
10393         (vfmlslq_laneq_high_u32): Ditto.
10394         * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
10395         (AARCH64_FL_FOR_ARCH8_4): New.
10396         (AARCH64_ISA_F16FML): New ISA flag.
10397         (TARGET_F16FML): New feature flag for fp16fml.
10398         (doc/invoke.texi): Document new fp16fml option.
10399
10400 2018-01-10  Michael Collison  <michael.collison@arm.com>
10401
10402         * config/aarch64/aarch64-builtins.c:
10403         (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
10404         * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
10405         (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
10406         * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
10407         (AARCH64_ISA_SHA3): New ISA flag.
10408         (TARGET_SHA3): New feature flag for sha3.
10409         * config/aarch64/iterators.md (sha512_op): New int attribute.
10410         (CRYPTO_SHA512): New int iterator.
10411         (UNSPEC_SHA512H): New unspec.
10412         (UNSPEC_SHA512H2): Ditto.
10413         (UNSPEC_SHA512SU0): Ditto.
10414         (UNSPEC_SHA512SU1): Ditto.
10415         * config/aarch64/aarch64-simd-builtins.def
10416         (aarch64_crypto_sha512hqv2di): New builtin.
10417         (aarch64_crypto_sha512h2qv2di): Ditto.
10418         (aarch64_crypto_sha512su0qv2di): Ditto.
10419         (aarch64_crypto_sha512su1qv2di): Ditto.
10420         (aarch64_eor3qv8hi): Ditto.
10421         (aarch64_rax1qv2di): Ditto.
10422         (aarch64_xarqv2di): Ditto.
10423         (aarch64_bcaxqv8hi): Ditto.
10424         * config/aarch64/aarch64-simd.md:
10425         (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
10426         (aarch64_crypto_sha512su0qv2di): Ditto.
10427         (aarch64_crypto_sha512su1qv2di): Ditto.
10428         (aarch64_eor3qv8hi): Ditto.
10429         (aarch64_rax1qv2di): Ditto.
10430         (aarch64_xarqv2di): Ditto.
10431         (aarch64_bcaxqv8hi): Ditto.
10432         * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
10433         (vsha512h2q_u64): Ditto.
10434         (vsha512su0q_u64): Ditto.
10435         (vsha512su1q_u64): Ditto.
10436         (veor3q_u16): Ditto.
10437         (vrax1q_u64): Ditto.
10438         (vxarq_u64): Ditto.
10439         (vbcaxq_u16): Ditto.
10440         * config/arm/types.md (crypto_sha512): New type attribute.
10441         (crypto_sha3): Ditto.
10442         (doc/invoke.texi): Document new sha3 option.
10443
10444 2018-01-10  Michael Collison  <michael.collison@arm.com>
10445
10446         * config/aarch64/aarch64-builtins.c:
10447         (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
10448         * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
10449         (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
10450         (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
10451         * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
10452         (AARCH64_ISA_SM4): New ISA flag.
10453         (TARGET_SM4): New feature flag for sm4.
10454         * config/aarch64/aarch64-simd-builtins.def
10455         (aarch64_sm3ss1qv4si): Ditto.
10456         (aarch64_sm3tt1aq4si): Ditto.
10457         (aarch64_sm3tt1bq4si): Ditto.
10458         (aarch64_sm3tt2aq4si): Ditto.
10459         (aarch64_sm3tt2bq4si): Ditto.
10460         (aarch64_sm3partw1qv4si): Ditto.
10461         (aarch64_sm3partw2qv4si): Ditto.
10462         (aarch64_sm4eqv4si): Ditto.
10463         (aarch64_sm4ekeyqv4si): Ditto.
10464         * config/aarch64/aarch64-simd.md:
10465         (aarch64_sm3ss1qv4si): Ditto.
10466         (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
10467         (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
10468         (aarch64_sm4eqv4si): Ditto.
10469         (aarch64_sm4ekeyqv4si): Ditto.
10470         * config/aarch64/iterators.md (sm3tt_op): New int iterator.
10471         (sm3part_op): Ditto.
10472         (CRYPTO_SM3TT): Ditto.
10473         (CRYPTO_SM3PART): Ditto.
10474         (UNSPEC_SM3SS1): New unspec.
10475         (UNSPEC_SM3TT1A): Ditto.
10476         (UNSPEC_SM3TT1B): Ditto.
10477         (UNSPEC_SM3TT2A): Ditto.
10478         (UNSPEC_SM3TT2B): Ditto.
10479         (UNSPEC_SM3PARTW1): Ditto.
10480         (UNSPEC_SM3PARTW2): Ditto.
10481         (UNSPEC_SM4E): Ditto.
10482         (UNSPEC_SM4EKEY): Ditto.
10483         * config/aarch64/constraints.md (Ui2): New constraint.
10484         * config/aarch64/predicates.md (aarch64_imm2): New predicate.
10485         * config/arm/types.md (crypto_sm3): New type attribute.
10486         (crypto_sm4): Ditto.
10487         * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
10488         (vsm3tt1aq_u32): Ditto.
10489         (vsm3tt1bq_u32): Ditto.
10490         (vsm3tt2aq_u32): Ditto.
10491         (vsm3tt2bq_u32): Ditto.
10492         (vsm3partw1q_u32): Ditto.
10493         (vsm3partw2q_u32): Ditto.
10494         (vsm4eq_u32): Ditto.
10495         (vsm4ekeyq_u32): Ditto.
10496         (doc/invoke.texi): Document new sm4 option.
10497
10498 2018-01-10  Michael Collison  <michael.collison@arm.com>
10499
10500         * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
10501         * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
10502         (AARCH64_FL_FOR_ARCH8_4): New.
10503         (AARCH64_FL_V8_4): New flag.
10504         (doc/invoke.texi): Document new armv8.4-a option.
10505
10506 2018-01-10  Michael Collison  <michael.collison@arm.com>
10507
10508         * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
10509         (__ARM_FEATURE_AES): Define if TARGET_AES is true.
10510         (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
10511         * config/aarch64/aarch64-option-extension.def: Add
10512         AARCH64_OPT_EXTENSION of 'sha2'.
10513         (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
10514         (crypto): Disable sha2 and aes if crypto disabled.
10515         (crypto): Enable aes and sha2 if enabled.
10516         (simd): Disable sha2 and aes if simd disabled.
10517         * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
10518         New flags.
10519         (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
10520         (TARGET_SHA2): New feature flag for sha2.
10521         (TARGET_AES): New feature flag for aes.
10522         * config/aarch64/aarch64-simd.md:
10523         (aarch64_crypto_aes<aes_op>v16qi): Make pattern
10524         conditional on TARGET_AES.
10525         (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
10526         (aarch64_crypto_sha1hsi): Make pattern conditional
10527         on TARGET_SHA2.
10528         (aarch64_crypto_sha1hv4si): Ditto.
10529         (aarch64_be_crypto_sha1hv4si): Ditto.
10530         (aarch64_crypto_sha1su1v4si): Ditto.
10531         (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
10532         (aarch64_crypto_sha1su0v4si): Ditto.
10533         (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
10534         (aarch64_crypto_sha256su0v4si): Ditto.
10535         (aarch64_crypto_sha256su1v4si): Ditto.
10536         (doc/invoke.texi): Document new aes and sha2 options.
10537
10538 2018-01-10  Martin Sebor  <msebor@redhat.com>
10539
10540         PR tree-optimization/83781
10541         * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
10542         as string arrays.
10543
10544 2018-01-11  Martin Sebor  <msebor@gmail.com>
10545             Prathamesh Kulkarni  <prathamesh.kulkarni@linaro.org>
10546
10547         PR tree-optimization/83501
10548         PR tree-optimization/81703
10549
10550         * tree-ssa-strlen.c (get_string_cst): Rename...
10551         (get_string_len): ...to this.  Handle global constants.
10552         (handle_char_store): Adjust.
10553
10554 2018-01-10  Kito Cheng  <kito.cheng@gmail.com>
10555             Jim Wilson  <jimw@sifive.com>
10556
10557         * config/riscv/riscv-protos.h (riscv_output_return): New.
10558         * config/riscv/riscv.c (struct machine_function): New naked_p field.
10559         (riscv_attribute_table, riscv_output_return),
10560         (riscv_handle_fndecl_attribute, riscv_naked_function_p),
10561         (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
10562         (riscv_compute_frame_info): Only compute frame->mask if not a naked
10563         function.
10564         (riscv_expand_prologue): Add early return for naked function.
10565         (riscv_expand_epilogue): Likewise.
10566         (riscv_function_ok_for_sibcall): Return false for naked function.
10567         (riscv_set_current_function): New.
10568         (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
10569         (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
10570         * config/riscv/riscv.md (simple_return): Call riscv_output_return.
10571         * doc/extend.texi (RISC-V Function Attributes): New.
10572
10573 2018-01-10  Michael Meissner  <meissner@linux.vnet.ibm.com>
10574
10575         * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
10576         check for 128-bit long double before checking TCmode.
10577         * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
10578         128-bit long doubles before checking TFmode or TCmode.
10579         (FLOAT128_IBM_P): Likewise.
10580
10581 2018-01-10  Martin Sebor  <msebor@redhat.com>
10582
10583         PR tree-optimization/83671
10584         * builtins.c (c_strlen): Unconditionally return zero for the empty
10585         string.
10586         Use -Warray-bounds for warnings.
10587         * gimple-fold.c (get_range_strlen): Handle non-constant lengths
10588         for non-constant array indices with COMPONENT_REF, arrays of
10589         arrays, and pointers to arrays.
10590         (gimple_fold_builtin_strlen): Determine and set length range for
10591         non-constant character arrays.
10592
10593 2018-01-10  Aldy Hernandez  <aldyh@redhat.com>
10594
10595         PR middle-end/81897
10596         * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
10597         empty blocks.
10598
10599 2018-01-10  Eric Botcazou  <ebotcazou@adacore.com>
10600
10601         * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
10602
10603 2018-01-10  Peter Bergner  <bergner@vnet.ibm.com>
10604
10605         PR target/83399
10606         * config/rs6000/rs6000.c (print_operand) <'y'>: Use
10607         VECTOR_MEM_ALTIVEC_OR_VSX_P.
10608         * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
10609         indexed_or_indirect_operand predicate.
10610         (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
10611         (*vsx_le_perm_load_v8hi): Likewise.
10612         (*vsx_le_perm_load_v16qi): Likewise.
10613         (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
10614         (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
10615         (*vsx_le_perm_store_v8hi): Likewise.
10616         (*vsx_le_perm_store_v16qi): Likewise.
10617         (eight unnamed splitters): Likewise.
10618
10619 2018-01-10  Peter Bergner  <bergner@vnet.ibm.com>
10620
10621         * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
10622         * config/rs6000/emmintrin.h: Likewise.
10623         * config/rs6000/mmintrin.h: Likewise.
10624         * config/rs6000/xmmintrin.h: Likewise.
10625
10626 2018-01-10  David Malcolm  <dmalcolm@redhat.com>
10627
10628         PR c++/43486
10629         * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
10630         "public_flag".
10631         * tree.c (tree_nop_conversion): Return true for location wrapper
10632         nodes.
10633         (maybe_wrap_with_location): New function.
10634         (selftest::check_strip_nops): New function.
10635         (selftest::test_location_wrappers): New function.
10636         (selftest::tree_c_tests): Call it.
10637         * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
10638         (maybe_wrap_with_location): New decl.
10639         (EXPR_LOCATION_WRAPPER_P): New macro.
10640         (location_wrapper_p): New inline function.
10641         (tree_strip_any_location_wrapper): New inline function.
10642
10643 2018-01-10  H.J. Lu  <hongjiu.lu@intel.com>
10644
10645         PR target/83735
10646         * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
10647         stack_realign_offset for the largest alignment of stack slot
10648         actually used.
10649         (ix86_find_max_used_stack_alignment): New function.
10650         (ix86_finalize_stack_frame_flags): Use it.  Set
10651         max_used_stack_alignment if we don't realign stack.
10652         * config/i386/i386.h (machine_function): Add
10653         max_used_stack_alignment.
10654
10655 2018-01-10  Christophe Lyon  <christophe.lyon@linaro.org>
10656
10657         * config/arm/arm.opt (-mbranch-cost): New option.
10658         * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
10659         account.
10660
10661 2018-01-10  Segher Boessenkool  <segher@kernel.crashing.org>
10662
10663         PR target/83629
10664         * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
10665         load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
10666
10667 2018-01-10  Richard Biener  <rguenther@suse.de>
10668
10669         PR debug/83765
10670         * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
10671         early out so it also covers the case where we have a non-NULL
10672         origin.
10673
10674 2018-01-10  Richard Sandiford  <richard.sandiford@linaro.org>
10675
10676         PR tree-optimization/83753
10677         * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
10678         for non-strided grouped accesses if the number of elements is 1.
10679
10680 2018-01-10  Jan Hubicka  <hubicka@ucw.cz>
10681
10682         PR target/81616
10683         * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
10684         * i386.h (TARGET_USE_GATHER): Define.
10685         * x86-tune.def (X86_TUNE_USE_GATHER): New.
10686
10687 2018-01-10  Martin Liska  <mliska@suse.cz>
10688
10689         PR bootstrap/82831
10690         * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
10691         * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
10692         partitioning.
10693         * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
10694         CLEANUP_NO_PARTITIONING is not set.
10695
10696 2018-01-10  Richard Sandiford  <richard.sandiford@linaro.org>
10697
10698         * doc/rtl.texi: Remove documentation of (const ...) wrappers
10699         for vectors, as a partial revert of r254296.
10700         * rtl.h (const_vec_p): Delete.
10701         (const_vec_duplicate_p): Don't test for vector CONSTs.
10702         (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
10703         * expmed.c (make_tree): Likewise.
10704
10705         Revert:
10706         * common.md (E, F): Use CONSTANT_P instead of checking for
10707         CONST_VECTOR.
10708         * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
10709         checking for CONST_VECTOR.
10710
10711 2018-01-09  Jan Hubicka  <hubicka@ucw.cz>
10712
10713         PR middle-end/83575
10714         * predict.c (force_edge_cold): Handle in more sane way edges
10715         with no prediction.
10716
10717 2018-01-09  Carl Love  <cel@us.ibm.com>
10718
10719         * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
10720         V4SI, V4SF types.
10721         (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
10722         * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
10723         VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
10724         VMRGOW_V2DI, VMRGOW_V2DF.  Remove definition for VMRGOW.
10725         * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
10726         P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW):  Add definitions.
10727         * config/rs6000/rs6000-protos.h: Add extern defition for
10728         rs6000_generate_float2_double_code.
10729         * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
10730         function.
10731         * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
10732         (float2_v2df): Add define_expand.
10733
10734 2018-01-09  Uros Bizjak  <ubizjak@gmail.com>
10735
10736         PR target/83628
10737         * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
10738         op_mode in the force_to_mode call.
10739
10740 2018-01-09  Richard Sandiford  <richard.sandiford@linaro.org>
10741
10742         * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
10743         instead of checking each element individually.
10744         (aarch64_evpc_uzp): Likewise.
10745         (aarch64_evpc_zip): Likewise.
10746         (aarch64_evpc_ext): Likewise.
10747         (aarch64_evpc_rev): Likewise.
10748         (aarch64_evpc_dup): Test the encoding for a single duplicated element,
10749         instead of checking each element individually.  Return true without
10750         generating rtl if
10751         (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
10752         whether all selected elements come from the same input, instead of
10753         checking each element individually.  Remove calls to gen_rtx_REG,
10754         start_sequence and end_sequence and instead assert that no rtl is
10755         generated.
10756
10757 2018-01-09  Richard Sandiford  <richard.sandiford@linaro.org>
10758
10759         * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
10760         order of HIGH and CONST checks.
10761
10762 2018-01-09  Richard Sandiford  <richard.sandiford@linaro.org>
10763
10764         * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
10765         if the destination isn't an SSA_NAME.
10766
10767 2018-01-09  Richard Biener  <rguenther@suse.de>
10768
10769         PR tree-optimization/83668
10770         * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
10771         move prologue...
10772         (canonicalize_loop_form): ... here, renamed from ...
10773         (canonicalize_loop_closed_ssa_form): ... this and amended to
10774         swap successor edges for loop exit blocks to make us use
10775         the RPO order we need for initial schedule generation.
10776
10777 2018-01-09  Joseph Myers  <joseph@codesourcery.com>
10778
10779         PR tree-optimization/64811
10780         * match.pd: When optimizing comparisons with Inf, avoid
10781         introducing or losing exceptions from comparisons with NaN.
10782
10783 2018-01-09  Martin Liska  <mliska@suse.cz>
10784
10785         PR sanitizer/82517
10786         * asan.c (shadow_mem_size): Add gcc_assert.
10787
10788 2018-01-09  Georg-Johann Lay  <avr@gjlay.de>
10789
10790         Don't save registers in main().
10791
10792         PR target/83738
10793         * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
10794         * config/avr/avr.opt (-mmain-is-OS_task): New target option.
10795         * config/avr/avr.c (avr_set_current_function): Don't error if
10796         naked, OS_task or OS_main are specified at the same time.
10797         (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
10798         OS_main.
10799         (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
10800         attribute.
10801         * common/config/avr/avr-common.c (avr_option_optimization_table):
10802         Switch on -mmain-is-OS_task for optimizing compilations.
10803
10804 2018-01-09  Richard Biener  <rguenther@suse.de>
10805
10806         PR tree-optimization/83572
10807         * graphite.c: Include cfganal.h.
10808         (graphite_transform_loops): Connect infinite loops to exit
10809         and remove fake edges at the end.
10810
10811 2018-01-09  Jan Hubicka  <hubicka@ucw.cz>
10812
10813         * ipa-inline.c (edge_badness): Revert accidental checkin.
10814
10815 2018-01-09  Jan Hubicka  <hubicka@ucw.cz>
10816
10817         PR ipa/80763
10818         * ipa-comdats.c (set_comdat_group): Only set comdat group of real
10819         symbols; not inline clones.
10820
10821 2018-01-09  Jakub Jelinek  <jakub@redhat.com>
10822
10823         PR target/83507
10824         * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
10825         hard registers.  Formatting fixes.
10826
10827         PR preprocessor/83722
10828         * gcc.c (try_generate_repro): Pass
10829         &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
10830         &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
10831         do_report_bug.
10832
10833 2018-01-08  Monk Chiang  <sh.chiang04@gmail.com>
10834             Kito Cheng  <kito.cheng@gmail.com>
10835
10836         * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
10837         (riscv_leaf_function_p): Delete.
10838         (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
10839
10840 2018-01-08  Aaron Sawdey  <acsawdey@linux.vnet.ibm.com>
10841
10842         * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
10843         function.
10844         (do_ifelse): New function.
10845         (do_isel): New function.
10846         (do_sub3): New function.
10847         (do_add3): New function.
10848         (do_load_mask_compare): New function.
10849         (do_overlap_load_compare): New function.
10850         (expand_compare_loop): New function.
10851         (expand_block_compare): Call expand_compare_loop() when appropriate.
10852         * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
10853         option description.
10854         (-mblock-compare-inline-loop-limit): New option.
10855
10856 2018-01-08  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
10857
10858         PR target/83677
10859         * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
10860         Reverse order of second and third operands in first alternative.
10861         * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
10862         of first and second elements in UNSPEC_VPERMR vector.
10863         (altivec_expand_vec_perm_le): Likewise.
10864
10865 2018-01-08  Jeff Law  <law@redhat.com>
10866
10867         PR rtl-optimizatin/81308
10868         * tree-switch-conversion.c (cfg_altered): New file scoped static.
10869         (process_switch): If group_case_labels makes a change, then set
10870         cfg_altered.
10871         (pass_convert_switch::execute): If a switch is converted, then
10872         set cfg_altered.  Return TODO_cfg_cleanup if cfg_altered is true.
10873
10874         PR rtl-optimization/81308
10875         * recog.c (split_all_insns): Conditionally cleanup the CFG after
10876         splitting insns.
10877
10878 2018-01-08  Vidya Praveen  <vidyapraveen@arm.com>
10879
10880         PR target/83663 - Revert r255946
10881         * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
10882         generation for cases where splatting a value is not useful.
10883         * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
10884         across a vec_duplicate and a paradoxical subreg forming a vector
10885         mode to a vec_concat.
10886
10887 2018-01-08  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
10888
10889         * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
10890         -march=armv8.3-a variants.
10891         * config/arm/t-multilib: Likewise.
10892         * config/arm/t-arm-elf: Likewise.  Handle dotprod extension.
10893
10894 2018-01-08  Aaron Sawdey  <acsawdey@linux.vnet.ibm.com>
10895
10896         * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
10897         to generate rtl.
10898         (cceq_ior_compare_complement): Give it a name so I can use it, and
10899         change boolean_or_operator predicate to boolean_operator so it can
10900         be used to generate a crand.
10901         (eqne): New code iterator.
10902         (bd/bd_neg): New code_attrs.
10903         (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
10904         a single define_insn.
10905         (<bd>tf_<mode>): A new insn pattern for the conditional form branch
10906         decrement (bdnzt/bdnzf/bdzt/bdzf).
10907         * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
10908         with the new names of the branch decrement patterns, and added the
10909         names of the branch decrement conditional patterns.
10910
10911 2018-01-08  Richard Biener  <rguenther@suse.de>
10912
10913         PR tree-optimization/83563
10914         * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
10915         cache.
10916
10917 2018-01-08  Richard Biener  <rguenther@suse.de>
10918
10919         PR middle-end/83713
10920         * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
10921
10922 2018-01-08  Richard Biener  <rguenther@suse.de>
10923
10924         PR tree-optimization/83685
10925         * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
10926         references to abnormals.
10927
10928 2018-01-08  Richard Biener  <rguenther@suse.de>
10929
10930         PR lto/83719
10931         * dwarf2out.c (output_indirect_strings): Handle empty
10932         skeleton_debug_str_hash.
10933         (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
10934
10935 2018-01-08  Claudiu Zissulescu  <claziss@synopsys.com>
10936
10937         * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
10938         (emit_store_direct): Likewise.
10939         (arc_trampoline_adjust_address): Likewise.
10940         (arc_asm_trampoline_template): New function.
10941         (arc_initialize_trampoline): Use asm_trampoline_template.
10942         (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
10943         * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
10944         * config/arc/arc.md (flush_icache): Delete pattern.
10945
10946 2018-01-08  Claudiu Zissulescu  <claziss@synopsys.com>
10947
10948         * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
10949         * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
10950         munaligned-access.
10951
10952 2018-01-08  Sebastian Huber  <sebastian.huber@embedded-brains.de>
10953
10954         PR target/83681
10955         * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
10956         by not USED_FOR_TARGET.
10957         (make_pass_resolve_sw_modes): Likewise.
10958
10959 2018-01-08  Sebastian Huber  <sebastian.huber@embedded-brains.de>
10960
10961         * config/nios2/nios2.h (nios2_section_threshold): Guard by not
10962         USED_FOR_TARGET.
10963
10964 2018-01-08  Richard Biener  <rguenther@suse.de>
10965
10966         PR middle-end/83580
10967         * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
10968
10969 2018-01-08  Richard Biener  <rguenther@suse.de>
10970
10971         PR middle-end/83517
10972         * match.pd ((t * 2) / 2) -> t): Add missing :c.
10973
10974 2018-01-06  Aldy Hernandez  <aldyh@redhat.com>
10975
10976         PR middle-end/81897
10977         * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
10978         basic blocks with a small number of successors.
10979         (convert_control_dep_chain_into_preds): Improve handling of
10980         forwarder blocks.
10981         (dump_predicates): Split apart into...
10982         (dump_pred_chain): ...here...
10983         (dump_pred_info): ...and here.
10984         (can_one_predicate_be_invalidated_p): Add debugging printfs.
10985         (can_chain_union_be_invalidated_p): Improve check for invalidation
10986         of paths.
10987         (uninit_uses_cannot_happen): Avoid unnecessary if
10988         convert_control_dep_chain_into_preds yielded nothing.
10989
10990 2018-01-06  Martin Sebor  <msebor@redhat.com>
10991
10992         PR tree-optimization/83640
10993         * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
10994         subtracting negative offset from size.
10995         (builtin_access::overlap): Adjust offset bounds of the access to fall
10996         within the size of the object if possible.
10997
10998 2018-01-06  Richard Sandiford  <richard.sandiford@linaro.org>
10999
11000         PR rtl-optimization/83699
11001         * expmed.c (extract_bit_field_1): Restrict the vector usage of
11002         extract_bit_field_as_subreg to cases in which the extracted
11003         value is also a vector.
11004
11005         * lra-constraints.c (process_alt_operands): Test for the equivalence
11006         substitutions when detecting a possible reload cycle.
11007
11008 2018-01-06  Jakub Jelinek  <jakub@redhat.com>
11009
11010         PR debug/83480
11011         * toplev.c (process_options): Don't enable debug_nonbind_markers_p
11012         by default if flag_selective_schedling{,2}.  Formatting fixes.
11013
11014         PR rtl-optimization/83682
11015         * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
11016         if it has non-VECTOR_MODE element mode.
11017         (vec_duplicate_p): Likewise.
11018
11019         PR middle-end/83694
11020         * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
11021         and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
11022
11023 2018-01-05  Jakub Jelinek  <jakub@redhat.com>
11024
11025         PR target/83604
11026         * config/i386/i386-builtin.def
11027         (__builtin_ia32_vgf2p8affineinvqb_v64qi,
11028         __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
11029         Require also OPTION_MASK_ISA_AVX512F in addition to
11030         OPTION_MASK_ISA_GFNI.
11031         (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
11032         __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
11033         OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
11034         to OPTION_MASK_ISA_GFNI.
11035         (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
11036         OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
11037         OPTION_MASK_ISA_AVX512BW.
11038         (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
11039         OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
11040         addition to OPTION_MASK_ISA_GFNI.
11041         (__builtin_ia32_vgf2p8affineinvqb_v16qi,
11042         __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
11043         Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
11044         to OPTION_MASK_ISA_GFNI.
11045         * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
11046         a requirement for all ISAs rather than any of them with a few
11047         exceptions.
11048         (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
11049         processing.
11050         (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
11051         bitmasks to be enabled with 3 exceptions, instead of requiring any
11052         enabled ISA with lots of exceptions.
11053         * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
11054         vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
11055         Change avx512bw in isa attribute to avx512f.
11056         * config/i386/sgxintrin.h: Add license boilerplate.
11057         * config/i386/vaesintrin.h: Likewise.  Fix macro spelling __AVX512F
11058         to __AVX512F__ and __AVX512VL to __AVX512VL__.
11059         (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
11060         _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
11061         defined.
11062         * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
11063         _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
11064         temporarily sse2 rather than sse if not enabled already.
11065
11066         PR target/83604
11067         * config/i386/sse.md (VI248_VLBW): Rename to ...
11068         (VI248_AVX512VL): ... this.  Don't guard V32HI with TARGET_AVX512BW.
11069         (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
11070         vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
11071         vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
11072         vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
11073         mode iterator instead of VI248_VLBW.
11074
11075 2018-01-05  Jan Hubicka  <hubicka@ucw.cz>
11076
11077         * ipa-fnsummary.c (record_modified_bb_info): Add OP.
11078         (record_modified): Skip clobbers; add debug output.
11079         (param_change_prob): Use sreal frequencies.
11080
11081 2018-01-05  Richard Sandiford  <richard.sandiford@linaro.org>
11082
11083         * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
11084         punt for user-aligned variables.
11085
11086 2018-01-05  Richard Sandiford  <richard.sandiford@linaro.org>
11087
11088         * tree-chrec.c (chrec_contains_symbols): Return true for
11089         POLY_INT_CST.
11090
11091 2018-01-05  Sudakshina Das  <sudi.das@arm.com>
11092
11093         PR target/82439
11094         * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
11095         of (x|y) == x for BICS pattern.
11096
11097 2018-01-05  Jakub Jelinek  <jakub@redhat.com>
11098
11099         PR tree-optimization/83605
11100         * gimple-ssa-strength-reduction.c: Include tree-eh.h.
11101         (find_candidates_dom_walker::before_dom_children): Ignore stmts that
11102         can throw.
11103
11104 2018-01-05  Sebastian Huber  <sebastian.huber@embedded-brains.de>
11105
11106         * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
11107         * config/epiphany/rtems.h: New file.
11108
11109 2018-01-04  Jakub Jelinek  <jakub@redhat.com>
11110             Uros Bizjak  <ubizjak@gmail.com>
11111
11112         PR target/83554
11113         * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
11114         QIreg_operand instead of register_operand predicate.
11115         * config/i386/i386.c (ix86_rop_should_change_byte_p,
11116         set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
11117         comments instead of -fmitigate[-_]rop.
11118
11119 2018-01-04  Rainer Orth  <ro@CeBiTec.Uni-Bielefeld.DE>
11120
11121         PR bootstrap/81926
11122         * cgraphunit.c (symbol_table::compile): Switch to text_section
11123         before calling assembly_start debug hook.
11124         * run-rtl-passes.c (run_rtl_passes): Likewise.
11125         Include output.h.
11126
11127 2018-01-04  Richard Sandiford  <richard.sandiford@linaro.org>
11128
11129         * tree-vrp.c (extract_range_from_binary_expr_1): Check
11130         range_int_cst_p rather than !symbolic_range_p before calling
11131         extract_range_from_multiplicative_op_1.
11132
11133 2018-01-04  Jeff Law  <law@redhat.com>
11134
11135         * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
11136         redundant test in assertion.
11137
11138 2018-01-04  Richard Sandiford  <richard.sandiford@linaro.org>
11139
11140         * doc/rtl.texi: Document machine_mode wrapper classes.
11141
11142 2018-01-04  Richard Sandiford  <richard.sandiford@linaro.org>
11143
11144         * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
11145         using tree_to_uhwi.
11146
11147 2018-01-04  Richard Sandiford  <richard.sandiford@linaro.org>
11148
11149         * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
11150         the VEC_PERM_EXPR fold to fail.
11151
11152 2018-01-04  Jakub Jelinek  <jakub@redhat.com>
11153
11154         PR debug/83585
11155         * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
11156         to switched_sections.
11157
11158 2018-01-04  Richard Sandiford  <richard.sandiford@linaro.org>
11159
11160         PR target/83680
11161         * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
11162         test for d.testing.
11163
11164 2018-01-04  Peter Bergner  <bergner@vnet.ibm.com>
11165
11166         PR target/83387
11167         * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
11168         allow arguments in FP registers if TARGET_HARD_FLOAT is false.
11169
11170 2018-01-04  Jakub Jelinek  <jakub@redhat.com>
11171
11172         PR debug/83666
11173         * cfgexpand.c (expand_debug_expr) <case BIT_FIELD_REF>: Punt if mode
11174         is BLKmode and bitpos not zero or mode change is needed.
11175
11176 2018-01-04  Richard Sandiford  <richard.sandiford@linaro.org>
11177
11178         PR target/83675
11179         * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
11180         TARGET_VIS2.
11181
11182 2018-01-04  Uros Bizjak  <ubizjak@gmail.com>
11183
11184         PR target/83628
11185         * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
11186         instead of MULT rtx.  Update all corresponding splitters.
11187         (*saddl_se): Ditto.
11188         (*ssub<modesuffix>): Ditto.
11189         (*ssubl_se): Ditto.
11190         (*cmp_sadd_di): Update split patterns.
11191         (*cmp_sadd_si): Ditto.
11192         (*cmp_sadd_sidi): Ditto.
11193         (*cmp_ssub_di): Ditto.
11194         (*cmp_ssub_si): Ditto.
11195         (*cmp_ssub_sidi): Ditto.
11196         * config/alpha/predicates.md (const23_operand): New predicate.
11197         * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
11198         Look for ASHIFT, not MULT inner operand.
11199         (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
11200
11201 2018-01-04  Martin Liska  <mliska@suse.cz>
11202
11203         PR gcov-profile/83669
11204         * gcov.c (output_intermediate_file): Add version to intermediate
11205         gcov file.
11206         * doc/gcov.texi: Document new field 'version' in intermediate
11207         file format. Fix location of '-k' option of gcov command.
11208
11209 2018-01-04  Martin Liska  <mliska@suse.cz>
11210
11211         PR ipa/82352
11212         * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
11213
11214 2018-01-04  Jakub Jelinek  <jakub@redhat.com>
11215
11216         * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
11217
11218 2018-01-03  Martin Sebor  <msebor@redhat.com>
11219
11220         PR tree-optimization/83655
11221         * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
11222         checking calls with invalid arguments.
11223
11224 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11225
11226         * tree-vect-stmts.c (vect_get_store_rhs): New function.
11227         (vectorizable_mask_load_store): Delete.
11228         (vectorizable_call): Return false for masked loads and stores.
11229         (vectorizable_store): Handle IFN_MASK_STORE.  Use vect_get_store_rhs
11230         instead of gimple_assign_rhs1.
11231         (vectorizable_load): Handle IFN_MASK_LOAD.
11232         (vect_transform_stmt): Don't set is_store for call_vec_info_type.
11233
11234 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11235
11236         * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
11237         split out from..,
11238         (vectorizable_mask_load_store): ...here.
11239         (vectorizable_load): ...and here.
11240
11241 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11242
11243         * tree-vect-stmts.c (vect_build_all_ones_mask)
11244         (vect_build_zero_merge_argument): New functions, split out from...
11245         (vectorizable_load): ...here.
11246
11247 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11248
11249         * tree-vect-stmts.c (vect_check_store_rhs): New function,
11250         split out from...
11251         (vectorizable_mask_load_store): ...here.
11252         (vectorizable_store): ...and here.
11253
11254 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11255
11256         * tree-vect-stmts.c (vect_check_load_store_mask): New function,
11257         split out from...
11258         (vectorizable_mask_load_store): ...here.
11259
11260 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11261
11262         * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
11263         (vect_model_store_cost): Take a vec_load_store_type instead of a
11264         vect_def_type.
11265         * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
11266         (vect_model_store_cost): Take a vec_load_store_type instead of a
11267         vect_def_type.
11268         (vectorizable_mask_load_store): Update accordingly.
11269         (vectorizable_store): Likewise.
11270         * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
11271
11272 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11273
11274         * tree-vect-loop.c (vect_transform_loop): Stub out scalar
11275         IFN_MASK_LOAD calls here rather than...
11276         * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
11277
11278 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11279             Alan Hayward  <alan.hayward@arm.com>
11280             David Sherwood  <david.sherwood@arm.com>
11281
11282         * expmed.c (extract_bit_field_1): For vector extracts,
11283         fall back to extract_bit_field_as_subreg if vec_extract
11284         isn't available.
11285
11286 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11287             Alan Hayward  <alan.hayward@arm.com>
11288             David Sherwood  <david.sherwood@arm.com>
11289
11290         * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
11291         they are variable or constant sized.
11292         (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
11293         slots for constant-sized data.
11294
11295 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11296             Alan Hayward  <alan.hayward@arm.com>
11297             David Sherwood  <david.sherwood@arm.com>
11298
11299         * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
11300         handling COND_EXPRs with boolean comparisons, try to find a better
11301         basis for the mask type than the boolean itself.
11302
11303 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11304
11305         * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
11306         is calculated and how it can be overridden.
11307         * genmodes.c (max_bitsize_mode_any_mode): New variable.
11308         (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
11309         if defined.
11310         (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
11311         if nonzero.
11312
11313 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11314             Alan Hayward  <alan.hayward@arm.com>
11315             David Sherwood  <david.sherwood@arm.com>
11316
11317         * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
11318         Remove the mode argument.
11319         (aarch64_simd_valid_immediate): Remove the mode and inverse
11320         arguments.
11321         * config/aarch64/iterators.md (bitsize): New iterator.
11322         * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
11323         (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
11324         * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
11325         aarch64_simd_valid_immediate.
11326         * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
11327         (aarch64_reg_or_bic_imm): Likewise.
11328         * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
11329         with an insn_type enum and msl with a modifier_type enum.
11330         Replace element_width with a scalar_mode.  Change the shift
11331         to unsigned int.  Add constructors for scalar_float_mode and
11332         scalar_int_mode elements.
11333         (aarch64_vect_float_const_representable_p): Delete.
11334         (aarch64_can_const_movi_rtx_p)
11335         (aarch64_simd_scalar_immediate_valid_for_move)
11336         (aarch64_simd_make_constant): Update call to
11337         aarch64_simd_valid_immediate.
11338         (aarch64_advsimd_valid_immediate_hs): New function.
11339         (aarch64_advsimd_valid_immediate): Likewise.
11340         (aarch64_simd_valid_immediate): Remove mode and inverse
11341         arguments.  Rewrite to use the above.  Use const_vec_duplicate_p
11342         to detect duplicated constants and use aarch64_float_const_zero_rtx_p
11343         and aarch64_float_const_representable_p on the result.
11344         (aarch64_output_simd_mov_immediate): Remove mode argument.
11345         Update call to aarch64_simd_valid_immediate and use of
11346         simd_immediate_info.
11347         (aarch64_output_scalar_simd_mov_immediate): Update call
11348         accordingly.
11349
11350 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11351             Alan Hayward  <alan.hayward@arm.com>
11352             David Sherwood  <david.sherwood@arm.com>
11353
11354         * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
11355         (mode_nunits): Likewise CONST_MODE_NUNITS.
11356         * machmode.def (ADJUST_NUNITS): Document.
11357         * genmodes.c (mode_data::need_nunits_adj): New field.
11358         (blank_mode): Update accordingly.
11359         (adj_nunits): New variable.
11360         (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
11361         parameter.
11362         (emit_mode_size_inline): Set need_bytesize_adj for all modes
11363         listed in adj_nunits.
11364         (emit_mode_nunits_inline): Set need_nunits_adj for all modes
11365         listed in adj_nunits.  Don't emit case statements for such modes.
11366         (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
11367         and CONST_MODE_PRECISION.  Make CONST_MODE_SIZE expand to
11368         nothing if adj_nunits is nonnull.
11369         (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
11370         (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
11371         (emit_mode_fbit): Update use of print_maybe_const_decl.
11372         (emit_move_size): Likewise.  Treat the array as non-const
11373         if adj_nunits.
11374         (emit_mode_adjustments): Handle adj_nunits.
11375
11376 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11377
11378         * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
11379         * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
11380         (VECTOR_MODES): Use it.
11381         (make_vector_modes): Take the prefix as an argument.
11382
11383 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11384             Alan Hayward  <alan.hayward@arm.com>
11385             David Sherwood  <david.sherwood@arm.com>
11386
11387         * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
11388         * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
11389         for MODE_VECTOR_BOOL.
11390         * machmode.def (VECTOR_BOOL_MODE): Document.
11391         * genmodes.c (VECTOR_BOOL_MODE): New macro.
11392         (make_vector_bool_mode): New function.
11393         (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
11394         MODE_VECTOR_BOOL.
11395         * lto-streamer-in.c (lto_input_mode_table): Likewise.
11396         * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
11397         Likewise.
11398         * stor-layout.c (int_mode_for_mode): Likewise.
11399         * tree.c (build_vector_type_for_mode): Likewise.
11400         * varasm.c (output_constant_pool_2): Likewise.
11401         * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
11402         CONSTM1_RTX (BImode) are the same thing.  Initialize const_tiny_rtx
11403         for MODE_VECTOR_BOOL.
11404         * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
11405         of mode class checks.
11406         * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
11407         instead of a list of mode class checks.
11408         (expand_vector_scalar_condition): Likewise.
11409         (type_for_widest_vector_mode): Handle BImode as an inner mode.
11410
11411 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11412             Alan Hayward  <alan.hayward@arm.com>
11413             David Sherwood  <david.sherwood@arm.com>
11414
11415         * machmode.h (mode_size): Change from unsigned short to
11416         poly_uint16_pod.
11417         (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
11418         (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
11419         or if measurement_type is not polynomial.
11420         (fixed_size_mode::includes_p): Check for constant-sized modes.
11421         * genmodes.c (emit_mode_size_inline): Make mode_size_inline
11422         return a poly_uint16 rather than an unsigned short.
11423         (emit_mode_size): Change the type of mode_size from unsigned short
11424         to poly_uint16_pod.  Use ZERO_COEFFS for the initializer.
11425         (emit_mode_adjustments): Cope with polynomial vector sizes.
11426         * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
11427         for GET_MODE_SIZE.
11428         * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
11429         for GET_MODE_SIZE.
11430         * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
11431         * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
11432         * caller-save.c (setup_save_areas): Likewise.
11433         (replace_reg_with_saved_mem): Likewise.
11434         * calls.c (emit_library_call_value_1): Likewise.
11435         * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
11436         * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
11437         (gen_lowpart_for_combine): Likewise.
11438         * convert.c (convert_to_integer_1): Likewise.
11439         * cse.c (equiv_constant, cse_insn): Likewise.
11440         * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
11441         (cselib_subst_to_values): Likewise.
11442         * dce.c (word_dce_process_block): Likewise.
11443         * df-problems.c (df_word_lr_mark_ref): Likewise.
11444         * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
11445         * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
11446         (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
11447         (rtl_for_decl_location): Likewise.
11448         * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
11449         * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
11450         * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
11451         (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
11452         (expand_expr_real_1): Likewise.
11453         * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
11454         (pad_below): Likewise.
11455         * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
11456         * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
11457         * ira.c (get_subreg_tracking_sizes): Likewise.
11458         * ira-build.c (ira_create_allocno_objects): Likewise.
11459         * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
11460         (ira_sort_regnos_for_alter_reg): Likewise.
11461         * ira-costs.c (record_operand_costs): Likewise.
11462         * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
11463         (resolve_simple_move): Likewise.
11464         * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
11465         (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
11466         (lra_constraints): Likewise.
11467         (CONST_POOL_OK_P): Reject variable-sized modes.
11468         * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
11469         (add_pseudo_to_slot, lra_spill): Likewise.
11470         * omp-low.c (omp_clause_aligned_alignment): Likewise.
11471         * optabs-query.c (get_best_extraction_insn): Likewise.
11472         * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
11473         * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
11474         (expand_mult_highpart, valid_multiword_target_p): Likewise.
11475         * recog.c (offsettable_address_addr_space_p): Likewise.
11476         * regcprop.c (maybe_mode_change): Likewise.
11477         * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
11478         * regrename.c (build_def_use): Likewise.
11479         * regstat.c (dump_reg_info): Likewise.
11480         * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
11481         (find_reloads, find_reloads_subreg_address): Likewise.
11482         * reload1.c (eliminate_regs_1): Likewise.
11483         * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
11484         * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
11485         (simplify_binary_operation_1, simplify_subreg): Likewise.
11486         * targhooks.c (default_function_arg_padding): Likewise.
11487         (default_hard_regno_nregs, default_class_max_nregs): Likewise.
11488         * tree-cfg.c (verify_gimple_assign_binary): Likewise.
11489         (verify_gimple_assign_ternary): Likewise.
11490         * tree-inline.c (estimate_move_cost): Likewise.
11491         * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
11492         * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
11493         (get_address_cost_ainc): Likewise.
11494         * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
11495         (vect_supportable_dr_alignment): Likewise.
11496         * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
11497         (vectorizable_reduction): Likewise.
11498         * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
11499         (vectorizable_operation, vectorizable_load): Likewise.
11500         * tree.c (build_same_sized_truth_vector_type): Likewise.
11501         * valtrack.c (cleanup_auto_inc_dec): Likewise.
11502         * var-tracking.c (emit_note_insn_var_location): Likewise.
11503         * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
11504         (ADDR_VEC_ALIGN): Likewise.
11505
11506 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11507             Alan Hayward  <alan.hayward@arm.com>
11508             David Sherwood  <david.sherwood@arm.com>
11509
11510         * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
11511         unsigned short.
11512         (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
11513         or if measurement_type is polynomial.
11514         * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
11515         * combine.c (make_extraction): Likewise.
11516         * dse.c (find_shift_sequence): Likewise.
11517         * dwarf2out.c (mem_loc_descriptor): Likewise.
11518         * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
11519         (extract_bit_field, extract_low_bits): Likewise.
11520         * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
11521         (optimize_bitfield_assignment_op, expand_assignment): Likewise.
11522         (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
11523         * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
11524         * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
11525         * reload.c (find_reloads): Likewise.
11526         * reload1.c (alter_reg): Likewise.
11527         * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
11528         * targhooks.c (default_secondary_memory_needed_mode): Likewise.
11529         * tree-if-conv.c (predicate_mem_writes): Likewise.
11530         * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
11531         * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
11532         * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
11533         * valtrack.c (dead_debug_insert_temp): Likewise.
11534         * varasm.c (mergeable_constant_section): Likewise.
11535         * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
11536
11537 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11538             Alan Hayward  <alan.hayward@arm.com>
11539             David Sherwood  <david.sherwood@arm.com>
11540
11541         * expr.c (expand_assignment): Cope with polynomial mode sizes
11542         when assigning to a CONCAT.
11543
11544 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11545             Alan Hayward  <alan.hayward@arm.com>
11546             David Sherwood  <david.sherwood@arm.com>
11547
11548         * machmode.h (mode_precision): Change from unsigned short to
11549         poly_uint16_pod.
11550         (mode_to_precision): Return a poly_uint16 rather than an unsigned
11551         short.
11552         (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
11553         or if measurement_type is not polynomial.
11554         (HWI_COMPUTABLE_MODE_P): Turn into a function.  Optimize the case
11555         in which the mode is already known to be a scalar_int_mode.
11556         * genmodes.c (emit_mode_precision): Change the type of mode_precision
11557         from unsigned short to poly_uint16_pod.  Use ZERO_COEFFS for the
11558         initializer.
11559         * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
11560         for GET_MODE_PRECISION.
11561         * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
11562         for GET_MODE_PRECISION.
11563         * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
11564         as polynomial.
11565         (try_combine, find_split_point, combine_simplify_rtx): Likewise.
11566         (expand_field_assignment, make_extraction): Likewise.
11567         (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
11568         (get_last_value): Likewise.
11569         * convert.c (convert_to_integer_1): Likewise.
11570         * cse.c (cse_insn): Likewise.
11571         * expr.c (expand_expr_real_1): Likewise.
11572         * lra-constraints.c (simplify_operand_subreg): Likewise.
11573         * optabs-query.c (can_atomic_load_p): Likewise.
11574         * optabs.c (expand_atomic_load): Likewise.
11575         (expand_atomic_store): Likewise.
11576         * ree.c (combine_reaching_defs): Likewise.
11577         * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
11578         * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
11579         * tree.h (type_has_mode_precision_p): Likewise.
11580         * ubsan.c (instrument_si_overflow): Likewise.
11581
11582 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11583             Alan Hayward  <alan.hayward@arm.com>
11584             David Sherwood  <david.sherwood@arm.com>
11585
11586         * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
11587         polynomial numbers of units.
11588         (SET_TYPE_VECTOR_SUBPARTS): Likewise.
11589         (valid_vector_subparts_p): New function.
11590         (build_vector_type): Remove temporary shim and take the number
11591         of units as a poly_uint64 rather than an int.
11592         (build_opaque_vector_type): Take the number of units as a
11593         poly_uint64 rather than an int.
11594         * tree.c (build_vector_from_ctor): Handle polynomial
11595         TYPE_VECTOR_SUBPARTS.
11596         (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
11597         (uniform_vector_p, vector_type_mode, build_vector): Likewise.
11598         (build_vector_from_val): If the number of units is variable,
11599         use build_vec_duplicate_cst for constant operands and
11600         VEC_DUPLICATE_EXPR otherwise.
11601         (make_vector_type): Remove temporary is_constant ().
11602         (build_vector_type, build_opaque_vector_type): Take the number of
11603         units as a poly_uint64 rather than an int.
11604         (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
11605         VECTOR_CST_NELTS.
11606         * cfgexpand.c (expand_debug_expr): Likewise.
11607         * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
11608         (store_constructor, expand_expr_real_1): Likewise.
11609         (const_scalar_mask_from_tree): Likewise.
11610         * fold-const-call.c (fold_const_reduction): Likewise.
11611         * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
11612         (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
11613         (native_encode_vector, vec_cst_ctor_to_array): Likewise.
11614         (fold_relational_const): Likewise.
11615         (native_interpret_vector): Likewise.  Change the size from an
11616         int to an unsigned int.
11617         * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
11618         TYPE_VECTOR_SUBPARTS.
11619         (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
11620         (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
11621         duplicating a non-constant operand into a variable-length vector.
11622         * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
11623         TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
11624         * ipa-icf.c (sem_variable::equals): Likewise.
11625         * match.pd: Likewise.
11626         * omp-simd-clone.c (simd_clone_subparts): Likewise.
11627         * print-tree.c (print_node): Likewise.
11628         * stor-layout.c (layout_type): Likewise.
11629         * targhooks.c (default_builtin_vectorization_cost): Likewise.
11630         * tree-cfg.c (verify_gimple_comparison): Likewise.
11631         (verify_gimple_assign_binary): Likewise.
11632         (verify_gimple_assign_ternary): Likewise.
11633         (verify_gimple_assign_single): Likewise.
11634         * tree-pretty-print.c (dump_generic_node): Likewise.
11635         * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
11636         (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
11637         * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
11638         (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
11639         (vect_shift_permute_load_chain): Likewise.
11640         * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
11641         (expand_vector_condition, optimize_vector_constructor): Likewise.
11642         (lower_vec_perm, get_compute_type): Likewise.
11643         * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
11644         (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
11645         * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
11646         (vect_recog_mask_conversion_pattern): Likewise.
11647         * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
11648         (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
11649         * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
11650         (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
11651         (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
11652         (vectorizable_shift, vectorizable_operation, vectorizable_store)
11653         (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
11654         (supportable_widening_operation): Likewise.
11655         (supportable_narrowing_operation): Likewise.
11656         * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
11657         Likewise.
11658         * varasm.c (output_constant): Likewise.
11659
11660 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11661             Alan Hayward  <alan.hayward@arm.com>
11662             David Sherwood  <david.sherwood@arm.com>
11663
11664         * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
11665         so that both the length == 3 and length != 3 cases set up their
11666         own permute vectors.  Add comments explaining why we know the
11667         number of elements is constant.
11668         (vect_permute_load_chain): Likewise.
11669
11670 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11671             Alan Hayward  <alan.hayward@arm.com>
11672             David Sherwood  <david.sherwood@arm.com>
11673
11674         * machmode.h (mode_nunits): Change from unsigned char to
11675         poly_uint16_pod.
11676         (ONLY_FIXED_SIZE_MODES): New macro.
11677         (pod_mode::measurement_type, scalar_int_mode::measurement_type)
11678         (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
11679         (complex_mode::measurement_type, fixed_size_mode::measurement_type):
11680         New typedefs.
11681         (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
11682         (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
11683         or if measurement_type is not polynomial.
11684         * genmodes.c (ZERO_COEFFS): New macro.
11685         (emit_mode_nunits_inline): Make mode_nunits_inline return a
11686         poly_uint16.
11687         (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
11688         Use ZERO_COEFFS when emitting initializers.
11689         * data-streamer.h (bp_pack_poly_value): New function.
11690         (bp_unpack_poly_value): Likewise.
11691         * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
11692         for GET_MODE_NUNITS.
11693         * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
11694         for GET_MODE_NUNITS.
11695         * tree.c (make_vector_type): Remove temporary shim and make
11696         the real function take the number of units as a poly_uint64
11697         rather than an int.
11698         (build_vector_type_for_mode): Handle polynomial nunits.
11699         * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
11700         * emit-rtl.c (const_vec_series_p_1): Likewise.
11701         (gen_rtx_CONST_VECTOR): Likewise.
11702         * fold-const.c (test_vec_duplicate_folding): Likewise.
11703         * genrecog.c (validate_pattern): Likewise.
11704         * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
11705         * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
11706         * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
11707         (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
11708         (expand_vec_cond_expr, expand_mult_highpart): Likewise.
11709         * rtlanal.c (subreg_get_info): Likewise.
11710         * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
11711         (vect_grouped_load_supported): Likewise.
11712         * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
11713         * tree-vect-loop.c (have_whole_vector_shift): Likewise.
11714         * simplify-rtx.c (simplify_unary_operation_1): Likewise.
11715         (simplify_const_unary_operation, simplify_binary_operation_1)
11716         (simplify_const_binary_operation, simplify_ternary_operation)
11717         (test_vector_ops_duplicate, test_vector_ops): Likewise.
11718         (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
11719         instead of CONST_VECTOR_NUNITS.
11720         * varasm.c (output_constant_pool_2): Likewise.
11721         * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
11722         explicit-encoded elements in the XVEC for variable-length vectors.
11723
11724 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11725
11726         * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
11727
11728 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11729             Alan Hayward  <alan.hayward@arm.com>
11730             David Sherwood  <david.sherwood@arm.com>
11731
11732         * coretypes.h (fixed_size_mode): Declare.
11733         (fixed_size_mode_pod): New typedef.
11734         * builtins.h (target_builtins::x_apply_args_mode)
11735         (target_builtins::x_apply_result_mode): Change type to
11736         fixed_size_mode_pod.
11737         * builtins.c (apply_args_size, apply_result_size, result_vector)
11738         (expand_builtin_apply_args_1, expand_builtin_apply)
11739         (expand_builtin_return): Update accordingly.
11740
11741 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11742
11743         * cse.c (hash_rtx_cb): Hash only the encoded elements.
11744         * cselib.c (cselib_hash_rtx): Likewise.
11745         * expmed.c (make_tree): Build VECTOR_CSTs directly from the
11746         CONST_VECTOR encoding.
11747
11748 2018-01-03  Jakub Jelinek  <jakub@redhat.com>
11749             Jeff Law  <law@redhat.com>
11750
11751         PR target/83641
11752         * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
11753         noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
11754         only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
11755         and add REG_CFA_ADJUST_CFA notes in that case to both insns.
11756
11757         PR target/83641
11758         * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
11759         explicitly probe *sp in a noreturn function if there were any callee
11760         register saves or frame pointer is needed.
11761
11762 2018-01-03  Jakub Jelinek  <jakub@redhat.com>
11763
11764         PR debug/83621
11765         * cfgexpand.c (expand_debug_expr): Return NULL if mode is
11766         BLKmode for ternary, binary or unary expressions.
11767
11768         PR debug/83645
11769         * var-tracking.c (delete_vta_debug_insn): New inline function.
11770         (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
11771         insns from get_insns () to NULL instead of each bb separately.
11772         Use delete_vta_debug_insn.  No longer static.
11773         (vt_debug_insns_local, variable_tracking_main_1): Adjust
11774         delete_vta_debug_insns callers.
11775         * rtl.h (delete_vta_debug_insns): Declare.
11776         * final.c (rest_of_handle_final): Call delete_vta_debug_insns
11777         instead of variable_tracking_main.
11778
11779 2018-01-03  Martin Sebor  <msebor@redhat.com>
11780
11781         PR tree-optimization/83603
11782         * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
11783         arguments past the endof the argument list in functions declared
11784         without a prototype.
11785         * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
11786         Avoid checking when arguments are null.
11787
11788 2018-01-03  Martin Sebor  <msebor@redhat.com>
11789
11790         PR c/83559
11791         * doc/extend.texi (attribute const): Fix a typo.
11792         * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
11793         issuing -Wsuggest-attribute for void functions.
11794
11795 2018-01-03  Martin Sebor  <msebor@redhat.com>
11796
11797         * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
11798         offset_int::from instead of wide_int::to_shwi.
11799         (maybe_diag_overlap): Remove assertion.
11800         Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
11801         * gimple-ssa-sprintf.c (format_directive): Same.
11802         (parse_directive): Same.
11803         (sprintf_dom_walker::compute_format_length): Same.
11804         (try_substitute_return_value): Same.
11805
11806 2018-01-03  Jeff Law  <law@redhat.com>
11807
11808         PR middle-end/83654
11809         * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
11810         non-constant residual for zero at runtime and avoid probing in
11811         that case.  Reorganize code for trailing problem to mirror handling
11812         of the residual.
11813
11814 2018-01-03  Prathamesh Kulkarni  <prathamesh.kulkarni@linaro.org>
11815
11816         PR tree-optimization/83501
11817         * tree-ssa-strlen.c (get_string_cst): New.
11818         (handle_char_store): Call get_string_cst.
11819
11820 2018-01-03  Martin Liska  <mliska@suse.cz>
11821
11822         PR tree-optimization/83593
11823         * tree-ssa-strlen.c: Include tree-cfg.h.
11824         (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
11825         (strlen_dom_walker): Add new member variable m_cleanup_cfg.
11826         (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
11827         to false.
11828         (strlen_dom_walker::before_dom_children): Call
11829         gimple_purge_dead_eh_edges. Dump tranformation with details
11830         dump flags.
11831         (strlen_dom_walker::before_dom_children): Update call by adding
11832         new argument cleanup_eh.
11833         (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
11834
11835 2018-01-03  Martin Liska  <mliska@suse.cz>
11836
11837         PR ipa/83549
11838         * cif-code.def (VARIADIC_THUNK): New enum value.
11839         * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
11840         thunks.
11841
11842 2018-01-03  Jan Beulich  <jbeulich@suse.com>
11843
11844         * sse.md (mov<mode>_internal): Tighten condition for when to use
11845         vmovdqu<ssescalarsize> for TI and OI modes.
11846
11847 2018-01-03  Jakub Jelinek  <jakub@redhat.com>
11848
11849         Update copyright years.
11850
11851 2018-01-03  Martin Liska  <mliska@suse.cz>
11852
11853         PR ipa/83594
11854         * ipa-visibility.c (function_and_variable_visibility): Skip
11855         functions with noipa attribure.
11856
11857 2018-01-03  Jakub Jelinek  <jakub@redhat.com>
11858
11859         * gcc.c (process_command): Update copyright notice dates.
11860         * gcov-dump.c (print_version): Ditto.
11861         * gcov.c (print_version): Ditto.
11862         * gcov-tool.c (print_version): Ditto.
11863         * gengtype.c (create_file): Ditto.
11864         * doc/cpp.texi: Bump @copying's copyright year.
11865         * doc/cppinternals.texi: Ditto.
11866         * doc/gcc.texi: Ditto.
11867         * doc/gccint.texi: Ditto.
11868         * doc/gcov.texi: Ditto.
11869         * doc/install.texi: Ditto.
11870         * doc/invoke.texi: Ditto.
11871
11872 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11873
11874         * vector-builder.h (vector_builder::m_full_nelts): Change from
11875         unsigned int to poly_uint64.
11876         (vector_builder::full_nelts): Update prototype accordingly.
11877         (vector_builder::new_vector): Likewise.
11878         (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
11879         (vector_builder::operator ==): Likewise.
11880         (vector_builder::finalize): Likewise.
11881         * int-vector-builder.h (int_vector_builder::int_vector_builder):
11882         Take the number of elements as a poly_uint64 rather than an
11883         unsigned int.
11884         * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
11885         from unsigned int to poly_uint64.
11886         (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
11887         (vec_perm_indices::new_vector): Likewise.
11888         (vec_perm_indices::length): Likewise.
11889         (vec_perm_indices::nelts_per_input): Likewise.
11890         (vec_perm_indices::input_nelts): Likewise.
11891         * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
11892         number of elements per input as a poly_uint64 rather than an
11893         unsigned int.  Use the original encoding for variable-length
11894         vectors, rather than clamping each individual element.
11895         For the second and subsequent elements in each pattern,
11896         clamp the step and base before clamping their sum.
11897         (vec_perm_indices::series_p): Handle polynomial element counts.
11898         (vec_perm_indices::all_in_range_p): Likewise.
11899         (vec_perm_indices_to_tree): Likewise.
11900         (vec_perm_indices_to_rtx): Likewise.
11901         * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
11902         * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
11903         (tree_vector_builder::new_binary_operation): Handle polynomial
11904         element counts.  Return false if we need to know the number
11905         of elements at compile time.
11906         * fold-const.c (fold_vec_perm): Punt if the number of elements
11907         isn't known at compile time.
11908
11909 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11910
11911         * vec-perm-indices.h (vec_perm_builder): Change element type
11912         from HOST_WIDE_INT to poly_int64.
11913         (vec_perm_indices::element_type): Update accordingly.
11914         (vec_perm_indices::clamp): Handle polynomial element_types.
11915         * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
11916         (vec_perm_indices::all_in_range_p): Likewise.
11917         (tree_to_vec_perm_builder): Check for poly_int64 trees rather
11918         than shwi trees.
11919         * vector-builder.h (vector_builder::stepped_sequence_p): Handle
11920         polynomial vec_perm_indices element types.
11921         * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
11922         * fold-const.c (fold_vec_perm): Likewise.
11923         * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
11924         * tree-vect-generic.c (lower_vec_perm): Likewise.
11925         * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
11926         * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
11927         element type to HOST_WIDE_INT.
11928
11929 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11930             Alan Hayward  <alan.hayward@arm.com>
11931             David Sherwood  <david.sherwood@arm.com>
11932
11933         * alias.c (addr_side_effect_eval): Take the size as a poly_int64
11934         rather than an int.  Use plus_constant.
11935         (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
11936         Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
11937
11938 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11939             Alan Hayward  <alan.hayward@arm.com>
11940             David Sherwood  <david.sherwood@arm.com>
11941
11942         * calls.c (emit_call_1, expand_call): Change struct_value_size from
11943         a HOST_WIDE_INT to a poly_int64.
11944
11945 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11946             Alan Hayward  <alan.hayward@arm.com>
11947             David Sherwood  <david.sherwood@arm.com>
11948
11949         * calls.c (load_register_parameters): Cope with polynomial
11950         mode sizes.  Require a constant size for BLKmode parameters
11951         that aren't described by a PARALLEL.  If BLOCK_REG_PADDING
11952         forces a parameter to be padded at the lsb end in order to
11953         fill a complete number of words, require the parameter size
11954         to be ordered wrt UNITS_PER_WORD.
11955
11956 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11957             Alan Hayward  <alan.hayward@arm.com>
11958             David Sherwood  <david.sherwood@arm.com>
11959
11960         * reload1.c (spill_stack_slot_width): Change element type
11961         from unsigned int to poly_uint64_pod.
11962         (alter_reg): Treat mode sizes as polynomial.
11963
11964 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11965             Alan Hayward  <alan.hayward@arm.com>
11966             David Sherwood  <david.sherwood@arm.com>
11967
11968         * reload.c (complex_word_subreg_p): New function.
11969         (reload_inner_reg_of_subreg, push_reload): Use it.
11970
11971 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11972             Alan Hayward  <alan.hayward@arm.com>
11973             David Sherwood  <david.sherwood@arm.com>
11974
11975         * lra-constraints.c (process_alt_operands): Reject matched
11976         operands whose sizes aren't ordered.
11977         (match_reload): Refer to this check here.
11978
11979 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11980             Alan Hayward  <alan.hayward@arm.com>
11981             David Sherwood  <david.sherwood@arm.com>
11982
11983         * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
11984         that the mode size is in the set {1, 2, 4, 8, 16}.
11985
11986 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11987             Alan Hayward  <alan.hayward@arm.com>
11988             David Sherwood  <david.sherwood@arm.com>
11989
11990         * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
11991         Use plus_constant instead of gen_rtx_PLUS.
11992
11993 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
11994             Alan Hayward  <alan.hayward@arm.com>
11995             David Sherwood  <david.sherwood@arm.com>
11996
11997         * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
11998         * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
11999         * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
12000         * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
12001         * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
12002         * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
12003         * config/i386/i386-protos.h (ix86_push_rounding): Declare.
12004         * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
12005         * config/i386/i386.c (ix86_push_rounding): ...this new function.
12006         * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
12007         a poly_int64.
12008         * config/m32c/m32c.c (m32c_push_rounding): Likewise.
12009         * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
12010         * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
12011         * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
12012         * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
12013         * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
12014         * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
12015         * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
12016         * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
12017         * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
12018         function.
12019         * expr.c (emit_move_resolve_push): Treat the input and result
12020         of PUSH_ROUNDING as a poly_int64.
12021         (emit_move_complex_push, emit_single_push_insn_1): Likewise.
12022         (emit_push_insn): Likewise.
12023         * lra-eliminations.c (mark_not_eliminable): Likewise.
12024         * recog.c (push_operand): Likewise.
12025         * reload1.c (elimination_effects): Likewise.
12026         * rtlanal.c (nonzero_bits1): Likewise.
12027         * calls.c (store_one_arg): Likewise.  Require the padding to be
12028         known at compile time.
12029
12030 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12031             Alan Hayward  <alan.hayward@arm.com>
12032             David Sherwood  <david.sherwood@arm.com>
12033
12034         * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
12035         Use plus_constant instead of gen_rtx_PLUS.
12036
12037 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12038             Alan Hayward  <alan.hayward@arm.com>
12039             David Sherwood  <david.sherwood@arm.com>
12040
12041         * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
12042         rather than an int.
12043
12044 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12045             Alan Hayward  <alan.hayward@arm.com>
12046             David Sherwood  <david.sherwood@arm.com>
12047
12048         * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
12049         instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
12050         via stack temporaries.  Treat the mode size as polynomial too.
12051
12052 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12053             Alan Hayward  <alan.hayward@arm.com>
12054             David Sherwood  <david.sherwood@arm.com>
12055
12056         * expr.c (expand_expr_real_2): When handling conversions involving
12057         unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
12058         multiplying int_size_in_bytes by BITS_PER_UNIT.  Treat GET_MODE_BISIZE
12059         as a poly_uint64 too.
12060
12061 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12062             Alan Hayward  <alan.hayward@arm.com>
12063             David Sherwood  <david.sherwood@arm.com>
12064
12065         * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
12066
12067 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12068             Alan Hayward  <alan.hayward@arm.com>
12069             David Sherwood  <david.sherwood@arm.com>
12070
12071         * combine.c (can_change_dest_mode): Handle polynomial
12072         REGMODE_NATURAL_SIZE.
12073         * expmed.c (store_bit_field_1): Likewise.
12074         * expr.c (store_constructor): Likewise.
12075         * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
12076         and polynomial REGMODE_NATURAL_SIZE.
12077         (gen_lowpart_common): Likewise.
12078         * reginfo.c (record_subregs_of_mode): Likewise.
12079         * rtlanal.c (read_modify_subreg_p): Likewise.
12080
12081 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12082             Alan Hayward  <alan.hayward@arm.com>
12083             David Sherwood  <david.sherwood@arm.com>
12084
12085         * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
12086         numbers of elements.
12087
12088 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12089             Alan Hayward  <alan.hayward@arm.com>
12090             David Sherwood  <david.sherwood@arm.com>
12091
12092         * match.pd: Cope with polynomial numbers of vector elements.
12093
12094 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12095             Alan Hayward  <alan.hayward@arm.com>
12096             David Sherwood  <david.sherwood@arm.com>
12097
12098         * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
12099         in a POINTER_PLUS_EXPR.
12100
12101 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12102             Alan Hayward  <alan.hayward@arm.com>
12103             David Sherwood  <david.sherwood@arm.com>
12104
12105         * omp-simd-clone.c (simd_clone_subparts): New function.
12106         (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
12107         (ipa_simd_modify_function_body): Likewise.
12108
12109 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12110             Alan Hayward  <alan.hayward@arm.com>
12111             David Sherwood  <david.sherwood@arm.com>
12112
12113         * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
12114         (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
12115         (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
12116         (expand_vector_condition, vector_element): Likewise.
12117         (subparts_gt): New function.
12118         (get_compute_type): Use subparts_gt.
12119         (count_type_subparts): Delete.
12120         (expand_vector_operations_1): Use subparts_gt instead of
12121         count_type_subparts.
12122
12123 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12124             Alan Hayward  <alan.hayward@arm.com>
12125             David Sherwood  <david.sherwood@arm.com>
12126
12127         * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
12128         (vect_compile_time_alias): ...this new function.  Do the calculation
12129         on poly_ints rather than trees.
12130         (vect_prune_runtime_alias_test_list): Update call accordingly.
12131
12132 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12133             Alan Hayward  <alan.hayward@arm.com>
12134             David Sherwood  <david.sherwood@arm.com>
12135
12136         * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
12137         numbers of units.
12138         (vect_schedule_slp_instance): Likewise.
12139
12140 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12141             Alan Hayward  <alan.hayward@arm.com>
12142             David Sherwood  <david.sherwood@arm.com>
12143
12144         * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
12145         constant and extern definitions for variable-length vectors.
12146         (vect_get_constant_vectors): Note that the number of units
12147         is known to be constant.
12148
12149 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12150             Alan Hayward  <alan.hayward@arm.com>
12151             David Sherwood  <david.sherwood@arm.com>
12152
12153         * tree-vect-stmts.c (vectorizable_conversion): Treat the number
12154         of units as polynomial.  Choose between WIDE and NARROW based
12155         on multiple_p.
12156
12157 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12158             Alan Hayward  <alan.hayward@arm.com>
12159             David Sherwood  <david.sherwood@arm.com>
12160
12161         * tree-vect-stmts.c (simd_clone_subparts): New function.
12162         (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
12163
12164 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12165             Alan Hayward  <alan.hayward@arm.com>
12166             David Sherwood  <david.sherwood@arm.com>
12167
12168         * tree-vect-stmts.c (vectorizable_call): Treat the number of
12169         vectors as polynomial.  Use build_index_vector for
12170         IFN_GOMP_SIMD_LANE.
12171
12172 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12173             Alan Hayward  <alan.hayward@arm.com>
12174             David Sherwood  <david.sherwood@arm.com>
12175
12176         * tree-vect-stmts.c (get_load_store_type): Treat the number of
12177         units as polynomial.  Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
12178         for variable-length vectors.
12179         (vectorizable_mask_load_store): Treat the number of units as
12180         polynomial, asserting that it is constant if the condition has
12181         already been enforced.
12182         (vectorizable_store, vectorizable_load): Likewise.
12183
12184 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12185             Alan Hayward  <alan.hayward@arm.com>
12186             David Sherwood  <david.sherwood@arm.com>
12187
12188         * tree-vect-loop.c (vectorizable_live_operation): Treat the number
12189         of units as polynomial.  Punt if we can't tell at compile time
12190         which vector contains the final result.
12191
12192 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12193             Alan Hayward  <alan.hayward@arm.com>
12194             David Sherwood  <david.sherwood@arm.com>
12195
12196         * tree-vect-loop.c (vectorizable_induction): Treat the number
12197         of units as polynomial.  Punt on SLP inductions.  Use an integer
12198         VEC_SERIES_EXPR for variable-length integer reductions.  Use a
12199         cast of such a series for variable-length floating-point
12200         reductions.
12201
12202 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12203             Alan Hayward  <alan.hayward@arm.com>
12204             David Sherwood  <david.sherwood@arm.com>
12205
12206         * tree.h (build_index_vector): Declare.
12207         * tree.c (build_index_vector): New function.
12208         * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
12209         of units as polynomial, forcibly converting it to a constant if
12210         vectorizable_reduction has already enforced the condition.
12211         (vect_create_epilog_for_reduction): Likewise.  Use build_index_vector
12212         to create a {1,2,3,...} vector.
12213         (vectorizable_reduction): Treat the number of units as polynomial.
12214         Choose vectype_in based on the largest scalar element size rather
12215         than the smallest number of units.  Enforce the restrictions
12216         relied on above.
12217
12218 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12219             Alan Hayward  <alan.hayward@arm.com>
12220             David Sherwood  <david.sherwood@arm.com>
12221
12222         * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
12223         number of units as polynomial.
12224
12225 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12226             Alan Hayward  <alan.hayward@arm.com>
12227             David Sherwood  <david.sherwood@arm.com>
12228
12229         * target.h (vector_sizes, auto_vector_sizes): New typedefs.
12230         * target.def (autovectorize_vector_sizes): Return the vector sizes
12231         by pointer, using vector_sizes rather than a bitmask.
12232         * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
12233         * targhooks.c (default_autovectorize_vector_sizes): Likewise.
12234         * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
12235         Likewise.
12236         * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
12237         * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
12238         * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
12239         * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
12240         * omp-general.c (omp_max_vf): Likewise.
12241         * omp-low.c (omp_clause_aligned_alignment): Likewise.
12242         * optabs-query.c (can_vec_mask_load_store_p): Likewise.
12243         * tree-vect-loop.c (vect_analyze_loop): Likewise.
12244         * tree-vect-slp.c (vect_slp_bb): Likewise.
12245         * doc/tm.texi: Regenerate.
12246         * tree-vectorizer.h (current_vector_size): Change from an unsigned int
12247         to a poly_uint64.
12248         * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
12249         the vector size as a poly_uint64 rather than an unsigned int.
12250         (current_vector_size): Change from an unsigned int to a poly_uint64.
12251         (get_vectype_for_scalar_type): Update accordingly.
12252         * tree.h (build_truth_vector_type): Take the size and number of
12253         units as a poly_uint64 rather than an unsigned int.
12254         (build_vector_type): Add a temporary overload that takes
12255         the number of units as a poly_uint64 rather than an unsigned int.
12256         * tree.c (make_vector_type): Likewise.
12257         (build_truth_vector_type): Take the number of units as a poly_uint64
12258         rather than an unsigned int.
12259
12260 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12261             Alan Hayward  <alan.hayward@arm.com>
12262             David Sherwood  <david.sherwood@arm.com>
12263
12264         * target.def (get_mask_mode): Take the number of units and length
12265         as poly_uint64s rather than unsigned ints.
12266         * targhooks.h (default_get_mask_mode): Update accordingly.
12267         * targhooks.c (default_get_mask_mode): Likewise.
12268         * config/i386/i386.c (ix86_get_mask_mode): Likewise.
12269         * doc/tm.texi: Regenerate.
12270
12271 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12272             Alan Hayward  <alan.hayward@arm.com>
12273             David Sherwood  <david.sherwood@arm.com>
12274
12275         * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
12276         * omp-general.c (omp_max_vf): Likewise.
12277         * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
12278         (expand_omp_simd): Handle polynomial safelen.
12279         * omp-low.c (omplow_simd_context): Add a default constructor.
12280         (omplow_simd_context::max_vf): Change from int to poly_uint64.
12281         (lower_rec_simd_input_clauses): Update accordingly.
12282         (lower_rec_input_clauses): Likewise.
12283
12284 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12285             Alan Hayward  <alan.hayward@arm.com>
12286             David Sherwood  <david.sherwood@arm.com>
12287
12288         * tree-vectorizer.h (vect_nunits_for_cost): New function.
12289         * tree-vect-loop.c (vect_model_reduction_cost): Use it.
12290         * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
12291         (vect_analyze_slp_cost): Likewise.
12292         * tree-vect-stmts.c (vect_model_store_cost): Likewise.
12293         (vect_model_load_cost): Likewise.
12294
12295 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12296             Alan Hayward  <alan.hayward@arm.com>
12297             David Sherwood  <david.sherwood@arm.com>
12298
12299         * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
12300         (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
12301         from an unsigned int * to a poly_uint64_pod *.
12302         (calculate_unrolling_factor): New function.
12303         (vect_analyze_slp_instance): Use it.  Track polynomial max_nunits.
12304
12305 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12306             Alan Hayward  <alan.hayward@arm.com>
12307             David Sherwood  <david.sherwood@arm.com>
12308
12309         * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
12310         from an unsigned int to a poly_uint64.
12311         (_loop_vec_info::slp_unrolling_factor): Likewise.
12312         (_loop_vec_info::vectorization_factor): Change from an int
12313         to a poly_uint64.
12314         (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
12315         (vect_get_num_vectors): New function.
12316         (vect_update_max_nunits, vect_vf_for_cost): Likewise.
12317         (vect_get_num_copies): Use vect_get_num_vectors.
12318         (vect_analyze_data_ref_dependences): Change max_vf from an int *
12319         to an unsigned int *.
12320         (vect_analyze_data_refs): Change min_vf from an int * to a
12321         poly_uint64 *.
12322         (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
12323         than an unsigned HOST_WIDE_INT.
12324         * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
12325         (vect_analyze_data_ref_dependence): Change max_vf from an int *
12326         to an unsigned int *.
12327         (vect_analyze_data_ref_dependences): Likewise.
12328         (vect_compute_data_ref_alignment): Handle polynomial vf.
12329         (vect_enhance_data_refs_alignment): Likewise.
12330         (vect_prune_runtime_alias_test_list): Likewise.
12331         (vect_shift_permute_load_chain): Likewise.
12332         (vect_supportable_dr_alignment): Likewise.
12333         (dependence_distance_ge_vf): Take the vectorization factor as a
12334         poly_uint64 rather than an unsigned HOST_WIDE_INT.
12335         (vect_analyze_data_refs): Change min_vf from an int * to a
12336         poly_uint64 *.
12337         * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
12338         vfm1 as a poly_uint64 rather than an int.  Make the same change
12339         for the returned bound_scalar.
12340         (vect_gen_vector_loop_niters): Handle polynomial vf.
12341         (vect_do_peeling): Likewise.  Update call to
12342         vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
12343         (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
12344         be constant.
12345         * tree-vect-loop.c (vect_determine_vectorization_factor)
12346         (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
12347         (vect_get_known_peeling_cost): Likewise.
12348         (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
12349         (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
12350         (vect_transform_loop): Likewise.  Use the lowest possible VF when
12351         updating the upper bounds of the loop.
12352         (vect_min_worthwhile_factor): Make static.  Return an unsigned int
12353         rather than an int.
12354         * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
12355         polynomial unroll factors.
12356         (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
12357         (vect_make_slp_decision): Likewise.
12358         (vect_supported_load_permutation_p): Likewise, and polynomial
12359         vf too.
12360         (vect_analyze_slp_cost): Handle polynomial vf.
12361         (vect_slp_analyze_node_operations): Likewise.
12362         (vect_slp_analyze_bb_1): Likewise.
12363         (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
12364         than an unsigned HOST_WIDE_INT.
12365         * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
12366         (vectorizable_load): Handle polynomial vf.
12367         * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
12368         a poly_uint64.
12369         (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
12370
12371 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12372             Alan Hayward  <alan.hayward@arm.com>
12373             David Sherwood  <david.sherwood@arm.com>
12374
12375         * match.pd: Handle bit operations involving three constants
12376         and try to fold one pair.
12377
12378 2018-01-03  Richard Sandiford  <richard.sandiford@linaro.org>
12379
12380         * tree-vect-loop-manip.c: Include gimple-fold.h.
12381         (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
12382         niters_maybe_zero parameters.  Handle other cases besides a step of 1.
12383         (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
12384         Add a path that uses a step of VF instead of 1, but disable it
12385         for now.
12386         (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
12387         and niters_no_overflow parameters.  Update calls to
12388         slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
12389         Create a new SSA name if the latter choses to use a ste other
12390         than zero, and return it via niters_vector_mult_vf_var.
12391         * tree-vect-loop.c (vect_transform_loop): Update calls to
12392         vect_do_peeling, vect_gen_vector_loop_niters and
12393         slpeel_make_loop_iterate_ntimes.
12394         * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
12395         (vect_gen_vector_loop_niters): Update declarations after above changes.
12396
12397 2018-01-02  Michael Meissner  <meissner@linux.vnet.ibm.com>
12398
12399         * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
12400         128-bit round to integer instructions.
12401         (ceil<mode>2): Likewise.
12402         (btrunc<mode>2): Likewise.
12403         (round<mode>2): Likewise.
12404
12405 2018-01-02  Aaron Sawdey  <acsawdey@linux.vnet.ibm.com>
12406
12407         * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
12408         unaligned VSX load/store on P8/P9.
12409         (expand_block_clear): Allow the use of unaligned VSX
12410         load/store on P8/P9.
12411
12412 2018-01-02  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
12413
12414         * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
12415         New function.
12416         (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
12417         swap associated with both a load and a store.
12418
12419 2018-01-02  Andrew Waterman  <andrew@sifive.com>
12420
12421         * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
12422         * config/riscv/riscv.md (clear_cache): Use it.
12423
12424 2018-01-02  Artyom Skrobov  <tyomitch@gmail.com>
12425
12426         * web.c: Remove out-of-date comment.
12427
12428 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
12429
12430         * expr.c (fixup_args_size_notes): Check that any existing
12431         REG_ARGS_SIZE notes are correct, and don't try to re-add them.
12432         (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
12433         (emit_single_push_insn): ...here.
12434
12435 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
12436
12437         * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
12438         (const_vector_encoded_nelts): New function.
12439         (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
12440         (const_vector_int_elt, const_vector_elt): Declare.
12441         * emit-rtl.c (const_vector_int_elt_1): New function.
12442         (const_vector_elt): Likewise.
12443         * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
12444         of CONST_VECTOR_ELT.
12445
12446 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
12447
12448         * expr.c: Include rtx-vector-builder.h.
12449         (const_vector_mask_from_tree): Use rtx_vector_builder and operate
12450         directly on the tree encoding.
12451         (const_vector_from_tree): Likewise.
12452         * optabs.c: Include rtx-vector-builder.h.
12453         (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
12454         sequence of "u" values.
12455         * vec-perm-indices.c: Include rtx-vector-builder.h.
12456         (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
12457         directly on the vec_perm_indices encoding.
12458
12459 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
12460
12461         * doc/rtl.texi (const_vector): Describe new encoding scheme.
12462         * Makefile.in (OBJS): Add rtx-vector-builder.o.
12463         * rtx-vector-builder.h: New file.
12464         * rtx-vector-builder.c: Likewise.
12465         * rtl.h (rtx_def::u2): Add a const_vector field.
12466         (CONST_VECTOR_NPATTERNS): New macro.
12467         (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
12468         (CONST_VECTOR_DUPLICATE_P): Likewise.
12469         (CONST_VECTOR_STEPPED_P): Likewise.
12470         (CONST_VECTOR_ENCODED_ELT): Likewise.
12471         (const_vec_duplicate_p): Check for a duplicated vector encoding.
12472         (unwrap_const_vec_duplicate): Likewise.
12473         (const_vec_series_p): Check for a non-duplicated vector encoding.
12474         Say that the function only returns true for integer vectors.
12475         * emit-rtl.c: Include rtx-vector-builder.h.
12476         (gen_const_vec_duplicate_1): Delete.
12477         (gen_const_vector): Call gen_const_vec_duplicate instead of
12478         gen_const_vec_duplicate_1.
12479         (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
12480         (gen_const_vec_duplicate): Use rtx_vector_builder.
12481         (gen_const_vec_series): Likewise.
12482         (gen_rtx_CONST_VECTOR): Likewise.
12483         * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
12484         (swap_const_vector_halves): Take an rtx pointer rather than rtx.
12485         Build a new vector rather than modifying a CONST_VECTOR in-place.
12486         (handle_special_swappables): Update call accordingly.
12487         * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
12488         (swap_const_vector_halves): Take an rtx pointer rather than rtx.
12489         Build a new vector rather than modifying a CONST_VECTOR in-place.
12490         (handle_special_swappables): Update call accordingly.
12491
12492 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
12493
12494         * simplify-rtx.c (simplify_const_binary_operation): Use
12495         CONST_VECTOR_ELT instead of XVECEXP.
12496
12497 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
12498
12499         * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
12500         the selector elements to be different from the data elements
12501         if the selector is a VECTOR_CST.
12502         * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
12503         ssizetype for the selector.
12504
12505 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
12506
12507         * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
12508         before testing each element individually.
12509         * tree-vect-generic.c (lower_vec_perm): Likewise.
12510
12511 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
12512
12513         * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
12514         * selftest-run-tests.c (selftest::run_tests): Call it.
12515         * vector-builder.h (vector_builder::operator ==): New function.
12516         (vector_builder::operator !=): Likewise.
12517         * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
12518         (vec_perm_indices::all_from_input_p): New function.
12519         * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
12520         (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
12521         * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
12522         instead of reading the VECTOR_CST directly.  Detect whether both
12523         vector inputs are the same before constructing the vec_perm_indices,
12524         and update the number of inputs argument accordingly.  Use the
12525         utility functions added above.  Only construct sel2 if we need to.
12526
12527 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
12528
12529         * optabs.c (expand_vec_perm_var): Use an explicit encoding for
12530         the broadcast of the low byte.
12531         (expand_mult_highpart): Use an explicit encoding for the permutes.
12532         * optabs-query.c (can_mult_highpart_p): Likewise.
12533         * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
12534         * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
12535         (vectorizable_bswap): Likewise.
12536         * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
12537         explicit encoding for the power-of-2 permutes.
12538         (vect_permute_store_chain): Likewise.
12539         (vect_grouped_load_supported): Likewise.
12540         (vect_permute_load_chain): Likewise.
12541
12542 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
12543
12544         * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
12545         * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
12546         * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
12547         * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
12548         * tree-vect-stmts.c (vectorizable_bswap): Likewise.
12549         (vect_gen_perm_mask_any): Likewise.
12550
12551 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
12552
12553         * int-vector-builder.h: New file.
12554         * vec-perm-indices.h: Include int-vector-builder.h.
12555         (vec_perm_indices): Redefine as an int_vector_builder.
12556         (auto_vec_perm_indices): Delete.
12557         (vec_perm_builder): Redefine as a stand-alone class.
12558         (vec_perm_indices::vec_perm_indices): New function.
12559         (vec_perm_indices::clamp): Likewise.
12560         * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
12561         (vec_perm_indices::new_vector): New function.
12562         (vec_perm_indices::new_expanded_vector): Update for new
12563         vec_perm_indices class.
12564         (vec_perm_indices::rotate_inputs): New function.
12565         (vec_perm_indices::all_in_range_p): Operate directly on the
12566         encoded form, without computing elided elements.
12567         (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
12568         encoding.  Update for new vec_perm_indices class.
12569         * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
12570         the given vec_perm_builder.
12571         (expand_vec_perm_var): Update vec_perm_builder constructor.
12572         (expand_mult_highpart): Use vec_perm_builder instead of
12573         auto_vec_perm_indices.
12574         * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
12575         vec_perm_indices instead of auto_vec_perm_indices.  Use a single
12576         or double series encoding as appropriate.
12577         * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
12578         vec_perm_indices instead of auto_vec_perm_indices.
12579         * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
12580         * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
12581         (vect_permute_store_chain): Likewise.
12582         (vect_grouped_load_supported): Likewise.
12583         (vect_permute_load_chain): Likewise.
12584         (vect_shift_permute_load_chain): Likewise.
12585         * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
12586         (vect_transform_slp_perm_load): Likewise.
12587         (vect_schedule_slp_instance): Likewise.
12588         * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
12589         (vectorizable_mask_load_store): Likewise.
12590         (vectorizable_bswap): Likewise.
12591         (vectorizable_store): Likewise.
12592         (vectorizable_load): Likewise.
12593         * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
12594         vec_perm_indices instead of auto_vec_perm_indices.  Use
12595         tree_to_vec_perm_builder to read the vector from a tree.
12596         * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
12597         vec_perm_builder instead of a vec_perm_indices.
12598         (have_whole_vector_shift): Use vec_perm_builder and
12599         vec_perm_indices instead of auto_vec_perm_indices.  Leave the
12600         truncation to calc_vec_perm_mask_for_shift.
12601         (vect_create_epilog_for_reduction): Likewise.
12602         * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
12603         from auto_vec_perm_indices to vec_perm_indices.
12604         (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
12605         instead of changing individual elements.
12606         (aarch64_vectorize_vec_perm_const): Use new_vector to install
12607         the vector in d.perm.
12608         * config/arm/arm.c (expand_vec_perm_d::perm): Change
12609         from auto_vec_perm_indices to vec_perm_indices.
12610         (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
12611         instead of changing individual elements.
12612         (arm_vectorize_vec_perm_const): Use new_vector to install
12613         the vector in d.perm.
12614         * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
12615         Update vec_perm_builder constructor.
12616         (rs6000_expand_interleave): Likewise.
12617         * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
12618         (rs6000_expand_interleave): Likewise.
12619
12620 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
12621
12622         * optabs-query.c (can_vec_perm_var_p): Check whether lowering
12623         to qimode could truncate the indices.
12624         * optabs.c (expand_vec_perm_var): Likewise.
12625
12626 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
12627
12628         * Makefile.in (OBJS): Add vec-perm-indices.o.
12629         * vec-perm-indices.h: New file.
12630         * vec-perm-indices.c: Likewise.
12631         * target.h (vec_perm_indices): Replace with a forward class
12632         declaration.
12633         (auto_vec_perm_indices): Move to vec-perm-indices.h.
12634         * optabs.h: Include vec-perm-indices.h.
12635         (expand_vec_perm): Delete.
12636         (selector_fits_mode_p, expand_vec_perm_var): Declare.
12637         (expand_vec_perm_const): Declare.
12638         * target.def (vec_perm_const_ok): Replace with...
12639         (vec_perm_const): ...this new hook.
12640         * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
12641         (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
12642         * doc/tm.texi: Regenerate.
12643         * optabs.def (vec_perm_const): Delete.
12644         * doc/md.texi (vec_perm_const): Likewise.
12645         (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
12646         * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
12647         expand_vec_perm for constant permutation vectors.  Assert that
12648         the mode of variable permutation vectors is the integer equivalent
12649         of the mode that is being permuted.
12650         * optabs-query.h (selector_fits_mode_p): Declare.
12651         * optabs-query.c: Include vec-perm-indices.h.
12652         (selector_fits_mode_p): New function.
12653         (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
12654         is defined, instead of checking whether the vec_perm_const_optab
12655         exists.  Use targetm.vectorize.vec_perm_const instead of
12656         targetm.vectorize.vec_perm_const_ok.  Check whether the indices
12657         fit in the vector mode before using a variable permute.
12658         * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
12659         vec_perm_indices instead of an rtx.
12660         (expand_vec_perm): Replace with...
12661         (expand_vec_perm_const): ...this new function.  Take the selector
12662         as a vec_perm_indices rather than an rtx.  Also take the mode of
12663         the selector.  Update call to shift_amt_for_vec_perm_mask.
12664         Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
12665         Use vec_perm_indices::new_expanded_vector to expand the original
12666         selector into bytes.  Check whether the indices fit in the vector
12667         mode before using a variable permute.
12668         (expand_vec_perm_var): Make global.
12669         (expand_mult_highpart): Use expand_vec_perm_const.
12670         * fold-const.c: Includes vec-perm-indices.h.
12671         * tree-ssa-forwprop.c: Likewise.
12672         * tree-vect-data-refs.c: Likewise.
12673         * tree-vect-generic.c: Likewise.
12674         * tree-vect-loop.c: Likewise.
12675         * tree-vect-slp.c: Likewise.
12676         * tree-vect-stmts.c: Likewise.
12677         * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
12678         Delete.
12679         * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
12680         * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
12681         (aarch64_vectorize_vec_perm_const_ok): Fuse into...
12682         (aarch64_vectorize_vec_perm_const): ...this new function.
12683         (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
12684         (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
12685         * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
12686         * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
12687         * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
12688         (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
12689         (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
12690         into...
12691         (arm_vectorize_vec_perm_const): ...this new function.  Explicitly
12692         check for NEON modes.
12693         * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
12694         * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
12695         * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
12696         (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
12697         into...
12698         (ix86_vectorize_vec_perm_const): ...this new function.  Incorporate
12699         the old VEC_PERM_CONST conditions.
12700         * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
12701         * config/ia64/vect.md (vec_perm_const<mode>): Delete.
12702         * config/ia64/ia64.c (ia64_expand_vec_perm_const)
12703         (ia64_vectorize_vec_perm_const_ok): Merge into...
12704         (ia64_vectorize_vec_perm_const): ...this new function.
12705         * config/mips/loongson.md (vec_perm_const<mode>): Delete.
12706         * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
12707         * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
12708         * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
12709         * config/mips/mips.c (mips_expand_vec_perm_const)
12710         (mips_vectorize_vec_perm_const_ok): Merge into...
12711         (mips_vectorize_vec_perm_const): ...this new function.
12712         * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
12713         * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
12714         * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
12715         * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
12716         * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
12717         (rs6000_expand_vec_perm_const): Delete.
12718         * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
12719         Delete.
12720         (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
12721         (altivec_expand_vec_perm_const_le): Take each operand individually.
12722         Operate on constant selectors rather than rtxes.
12723         (altivec_expand_vec_perm_const): Likewise.  Update call to
12724         altivec_expand_vec_perm_const_le.
12725         (rs6000_expand_vec_perm_const): Delete.
12726         (rs6000_vectorize_vec_perm_const_ok): Delete.
12727         (rs6000_vectorize_vec_perm_const): New function.
12728         (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
12729         an element count and rtx array.
12730         (rs6000_expand_extract_even): Update call accordingly.
12731         (rs6000_expand_interleave): Likewise.
12732         * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
12733         * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
12734         * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
12735         * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
12736         (rs6000_expand_vec_perm_const): Delete.
12737         * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
12738         (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
12739         (altivec_expand_vec_perm_const_le): Take each operand individually.
12740         Operate on constant selectors rather than rtxes.
12741         (altivec_expand_vec_perm_const): Likewise.  Update call to
12742         altivec_expand_vec_perm_const_le.
12743         (rs6000_expand_vec_perm_const): Delete.
12744         (rs6000_vectorize_vec_perm_const_ok): Delete.
12745         (rs6000_vectorize_vec_perm_const): New function.  Remove stray
12746         reference to the SPE evmerge intructions.
12747         (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
12748         an element count and rtx array.
12749         (rs6000_expand_extract_even): Update call accordingly.
12750         (rs6000_expand_interleave): Likewise.
12751         * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
12752         * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
12753         new function.
12754         (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
12755
12756 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
12757
12758         * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
12759         vector mode and that that mode matches the mode of the data
12760         being permuted.
12761         (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
12762         out into expand_vec_perm_var.  Do all CONST_VECTOR handling here,
12763         directly using expand_vec_perm_1 when forcing selectors into
12764         registers.
12765         (expand_vec_perm_var): New function, split out from expand_vec_perm.
12766
12767 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
12768
12769         * optabs-query.h (can_vec_perm_p): Delete.
12770         (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
12771         * optabs-query.c (can_vec_perm_p): Split into...
12772         (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
12773         (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
12774         particular selector is valid.
12775         * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
12776         * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
12777         (vect_grouped_load_supported): Likewise.
12778         (vect_shift_permute_load_chain): Likewise.
12779         * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
12780         (vect_transform_slp_perm_load): Likewise.
12781         * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
12782         (vectorizable_bswap): Likewise.
12783         (vect_gen_perm_mask_checked): Likewise.
12784         * fold-const.c (fold_ternary_loc): Likewise.  Don't take
12785         implementations of variable permutation vectors into account
12786         when deciding which selector to use.
12787         * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
12788         vec_perm_const_optab is supported; instead use can_vec_perm_const_p
12789         with a false third argument.
12790         * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
12791         to test whether the constant selector is valid and can_vec_perm_var_p
12792         to test whether a variable selector is valid.
12793
12794 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
12795
12796         * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
12797         * optabs-query.c (can_vec_perm_p): Likewise.
12798         * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
12799         instead of vec_perm_indices.
12800         * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
12801         (vect_gen_perm_mask_checked): Likewise,
12802         * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
12803         (vect_gen_perm_mask_checked): Likewise,
12804
12805 2018-01-02  Richard Sandiford  <richard.sandiford@linaro.org>
12806
12807         * optabs-query.h (qimode_for_vec_perm): Declare.
12808         * optabs-query.c (can_vec_perm_p): Split out qimode search to...
12809         (qimode_for_vec_perm): ...this new function.
12810         * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
12811
12812 2018-01-02  Aaron Sawdey  <acsawdey@linux.vnet.ibm.com>
12813
12814         * rtlanal.c (canonicalize_condition): Return 0 if final rtx
12815         does not have a conditional at the top.
12816
12817 2018-01-02  Richard Biener  <rguenther@suse.de>
12818
12819         * ipa-inline.c (big_speedup_p): Fix expression.
12820
12821 2018-01-02  Jan Hubicka  <hubicka@ucw.cz>
12822
12823         PR target/81616
12824         * config/i386/x86-tune-costs.h: Increase cost of integer load costs
12825         for generic 4->6.
12826
12827 2018-01-02  Jan Hubicka  <hubicka@ucw.cz>
12828
12829         PR target/81616
12830         Generic tuning.
12831         * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
12832         cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
12833         and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
12834         cond_taken_branch_cost 3->4.
12835
12836 2018-01-01  Jakub Jelinek  <jakub@redhat.com>
12837
12838         PR tree-optimization/83581
12839         * tree-loop-distribution.c (pass_loop_distribution::execute): Return
12840         TODO_cleanup_cfg if any changes have been made.
12841
12842         PR middle-end/83608
12843         * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
12844         convert_modes if target mode has the right side, but different mode
12845         class.
12846
12847         PR middle-end/83609
12848         * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
12849         last argument when extracting from CONCAT.  If either from_real or
12850         from_imag is NULL, use expansion through memory.  If result is not
12851         a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
12852         the parts directly to inner mode, if even that fails, use expansion
12853         through memory.
12854
12855         PR middle-end/83623
12856         * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
12857         check for bswap in mode rather than HImode and use that in expand_unop
12858         too.
12859 \f
12860 Copyright (C) 2018 Free Software Foundation, Inc.
12861
12862 Copying and distribution of this file, with or without modification,
12863 are permitted in any medium without royalty provided the copyright
12864 notice and this notice are preserved.