[AArch64] Fix label mode

This patch fixes a case where we tried to force a VOIDmode label
into a DImode register, which led to invalid rtl that was previously
undiagnosed.

2017-08-22  Richard Sandiford  <richard.sandiford@linaro.org>
    Alan Hayward  <alan.hayward@arm.com>
    David Sherwood  <david.sherwood@arm.com>

gcc/
* config/aarch64/aarch64.md (casesi): Use DImode rather than
VOIDmode for the LABEL_REF.

Co-Authored-By: Alan Hayward <alan.hayward@arm.com>
Co-Authored-By: David Sherwood <david.sherwood@arm.com>
From-SVN: r251266

revert: asan.exp: Switch on *.cc tests.

2017-08-22  Richard Biener  <rguenther@suse.de>

Revert
2017-08-09  Slava Barinov  <v.barinov@samsung.com>

* g++.dg/asan/asan.exp: Switch on *.cc tests.

From-SVN: r251265

tree-cfg.c (gimple_split_edge): Avoid reallocating target PHI nodes.

2017-08-22  Richard Biener  <rguenther@suse.de>

* tree-cfg.c (gimple_split_edge): Avoid reallocating target
PHI nodes.

From-SVN: r251264

common.opt (feliminate-dwarf2-dups): Ignore.

2017-08-22  Richard Biener  <rguenther@suse.de>

* common.opt (feliminate-dwarf2-dups): Ignore.
* doc/invoke.texi (feliminate-dwarf2-dups): Remove documentation.
* dwarf2out.c (push_new_compile_unit, pop_compile_unit,
same_die_p_wrap, compute_section_prefix,
is_symbol_die, assign_symbol_names, break_out_includes): Remove.
(comdat_symbol_id, comdat_symbol_number): Likewise.
(cu_hash_table_entry, cu_hash_table_entry_hasher, cu_hash_type):
Likewise.
(check_duplicate_cu, record_comdat_symbol_number): Likewise.
(output_die): Mark unreachable path unreachable.
(dwarf2out_start_source_file): Do not create DW_TAG_GNU_BINCL.
(dwarf2out_end_source_file): Do not create DW_TAG_GNU_EINCL.
(dwarf2out_init): Remove code handling flag_eliminate_dwarf2_dups.
(dwarf2out_early_finish): Likewise.

* g++.dg/debug/dwarf2/dwarf2-1.C: Remove -feliminate-dwarf2-dups.
* g++.dg/debug/dwarf2/dwarf2-2.C: Likewise.
* g++.dg/debug/dwarf2/pr46123-2.C: Likewise.
* g++.dg/debug/dwarf2/typedef5.C: Likewise.
* gcc.dg/debug/dwarf2/dwarf2-3.c: Likewise.
* gcc.dg/debug/dwarf2/dwarf2-3.h: Likewise.
* gcc.dg/debug/dwarf2/dups-types.c: Remove.
* gcc.dg/debug/dwarf2/dups-types.h: Likewise.

From-SVN: r251262

* wide-int.h (hwi_with_prec::hwi_with_prec): Sign extend.

From-SVN: r251260

re PR target/81910 ([avr] ICE with "address" attribute on type)

gcc/
PR target/81910
* config/avr/avr.c (avr_handle_addr_attribute): Early return if
not VAR_P. Filter attribute warnings with OPT_Wattributes.
(avr_attribute_table) <io, io_low, address>: Initialize
.decl_required with true.

From-SVN: r251256

re PR fortran/81296 (derived type I/o problem)

2017-08-21  Jerry DeLisle  <jvdelisle@gcc.gnu.org>

PR fortran/81296
* trans-io.c (get_dtio_proc): Add check for format label and set
formatted flag accordingly. Reorganize the code a little.

* gfortran.dg/dtio_12.f90: Update test.

From-SVN: r251254

Daily bump.

From-SVN: r251253

rs6000-cpus.def (-mvsx-scalar-float): Delete undocumented debugging options.

2017-08-21  Michael Meissner  <meissner@linux.vnet.ibm.com>

* config/rs6000/rs6000-cpus.def (-mvsx-scalar-float): Delete
undocumented debugging options.
(-mvsx-scalar-double): Likewise.
(-mallow-df-permute): Likewise.
(-mvectorize-builtins): Likewise.
* config/rs6000/rs6000.c (rs6000_init_hard_regno_mode_ok):
Likewise.
(rs6000_builtin_vectorized_function): Likewise.
(rs6000_builtin_md_vectorized_function): Likewise.
(rs6000_opt_vars): Likewise.

From-SVN: r251248

search.c (lookup_field_r): Remove obsolete code for type-named field in PoD.

* search.c (lookup_field_r): Remove obsolete code for type-named
field in PoD.

From-SVN: r251243

search.c (lookup_field_1): Assert TYPE is a class and VFIELD isn't special.

* search.c (lookup_field_1): Assert TYPE is a class and VFIELD
isn't special.
(lookup_field_fuzzy_info::fuzzy_lookup_fnfields): Delete.
(lookup_field_fuzzy_r): Adjust.

From-SVN: r251241

pr81899.C: Fix c++03.

* g++.dg/template/pr81899.C: Fix c++03.
* g++.dg/debug/debug9.C: Add -fno-reorder-blocks-and-partition"

From-SVN: r251240

C: use full locations within c_parser_expr_list's vec<location_t>

The previous patch uncovered a bug in how c_parser_expr_list builds the
vec<location_t>: it was only using the location of the first token
within each assignment-expression in the expr-list.

This shows up in e.g. this -Wformat warning, where only part of the
2nd param is underlined:

   printf("hello %i", (long)0);
                 ~^   ~
                 %li

This patch fixes c_parser_expr_list to use the full range of
each assignment-expression in the list for the vec<location_t>, so
that for the above we print:

   printf("hello %i", (long)0);
                 ~^   ~~~~~~~
                 %li

gcc/c/ChangeLog:
* c-parser.c (c_parser_expr_list): Use c_expr::get_location ()
rather than peeking the location of the first token.
* c-tree.h (c_expr::get_location): New method.

gcc/testsuite/ChangeLog:
* gcc.dg/format/diagnostic-ranges.c (test_mismatching_types):
Update expected result to show all of "(long)0" being underlined.
* gcc.dg/plugin/diagnostic-test-string-literals-1.c
(test_multitoken_macro): Update expected underlining.

From-SVN: r251239

c-family/c/c++: pass optional vec<location_t> to c-format.c

This patch passes along the vec<location_t> of argument locations
at a callsite from the C frontend to check_function_arguments and
from there to c-format.c, so that we can underline the pertinent
argument to mismatched format codes even for tree codes like decls
and constants which lack a location_t for their usage sites.

This takes e.g.:

    printf("hello %i %i %i ", foo, bar, baz);
                     ~^
                     %s

to:

    printf("hello %i %i %i ", foo, bar, baz);
                     ~^            ~~~
                     %s

which is useful for cases where there's more than one variadic argument.

gcc/c-family/ChangeLog:
* c-common.c (check_function_arguments): Add "arglogs" param; pass
it to check_function_format.
* c-common.h (check_function_arguments): Add vec<location_t> *
param.
(check_function_format): Likewise.
* c-format.c (struct format_check_context): Add field "arglocs".
(check_function_format): Add param "arglocs"; pass it to
check_format_info.
(check_format_info):  Add param "arglocs"; use it to initialize
new field of format_ctx.
(check_format_arg): Pass format_ctx->arglocs to new param of
check_format_info_main.
(class argument_parser): New field "arglocs".
(argument_parser::argument_parser): Add "arglocs_" param and use
it to initialize new field.
(argument_parser::check_argument_type): Pass new arglocs field to
check_format_types.
(check_format_info_main): Add param "arglocs", and use it when
constructing arg_parser.
(check_format_types): Add param "arglocs"; use it if non-NULL when
!EXPR_HAS_LOCATION (cur_param) to get at location information.

gcc/c/ChangeLog:
* c-typeck.c (build_function_call_vec): Pass arg_loc to call
to check_function_arguments.

gcc/cp/ChangeLog:
* call.c (build_over_call): Pass NULL for new parameter to
check_function_arguments.
* typeck.c (cp_build_function_call_vec): Likewise.

gcc/testsuite/ChangeLog:
* gcc.dg/format/diagnostic-ranges.c: Update expected results
to show underlining of all pertinent params.
* gcc.dg/format/pr72858.c: Likewise.

From-SVN: r251238

re PR target/46091 (missed optimization: x86 bt/btc/bts instructions)

PR target/46091
* config/i386/i386.md (*btsq_imm): Rename from *btsq.
(*btrq_imm): Rename from *btrq.
(*btcq_imm): Rename from *btcq.
(btsc): New code attribute.
(*<btsc><mode>): New insn pattern.
(*btr<mode>): Ditto.
(*<btsc><mode>_mask): New insn_and_split pattern.
(*btr<mode>_mask): Ditto.

testsuite/ChangeLog:

PR target/46091
* gcc.target/i386/pr46091-4.c: New test.
* gcc.target/i386/pr46091-4a.c: Ditto.
* gcc.target/i386/pr46091-5.c: Ditto.
* gcc.target/i386/pr46091-5a.c: Ditto.

From-SVN: r251235

PR libstdc++/81912 make std::__iterator_category constexpr

PR libstdc++/81912
* include/bits/stl_iterator_base_types.h (__iterator_category): Add
constexpr for C++11 and later.
* testsuite/24_iterators/container_access.cc: Add target selector.
* testsuite/24_iterators/range_access.cc: Fix clause number in
comment.
* testsuite/24_iterators/range_access_cpp14.cc: Likewise.
* testsuite/24_iterators/range_access_cpp17.cc: New.

From-SVN: r251234

Simplify pad_below implementation

This patch simplifies the alignment calculations in pad_below.
The first arm of the "if" was:

- taking GET_MODE_BITSIZE (always equal to GET_MODE_SIZE * BITS_PER_UNIT),
- rounding up to the next multiple of PARM_BOUNDARY
- converting bits to bytes and
- subtracting the GET_MODE_SIZE

so was in effect calculating the number of bytes needed to round
GET_MODE_SIZE up to (PARM_BOUNDARY / BITS_PER_UNIT).  That can be
done more directly as -size & (align - 1), which is easier to
convert to variable-sized modes.

2017-08-21  Richard Sandiford  <richard.sandiford@linaro.org>
    Alan Hayward  <alan.hayward@arm.com>
    David Sherwood  <david.sherwood@arm.com>

gcc/
* function.c (pad_below): Simplify padding calculation.

Co-Authored-By: Alan Hayward <alan.hayward@arm.com>
Co-Authored-By: David Sherwood <david.sherwood@arm.com>
From-SVN: r251233

Remove the frame size argument from function_prologue/epilogue

Later patches will add support for frame sizes that are a run-time
invariant but not a compile-time constant.  This then raised the
question of whether the frame size argument to the function_prologue/
epilogue hooks should be updated accordingly.

It turned out that only two targets actually used this argument, and
even they got other frame-related information from the cached machine
structure.  This patch therefore removes the argument and makes the two
targets use get_frame_size () instead.

2017-08-21  Richard Sandiford  <richard.sandiford@linaro.org>
    Alan Hayward  <alan.hayward@arm.com>
    David Sherwood  <david.sherwood@arm.com>

gcc/
* target.def (function_prologue): Remove frame size argument.
(function_epilogue): Likewise.
* doc/tm.texi: Regenerate.
* final.c (final_start_function): Update call to function_prologue.
(final_end_function): Update call to function_epilogue.
(default_function_pro_epilogue): Remove frame size argument.
* output.h (default_function_pro_epilogue): Likewise.
* config/arm/arm.c (arm_output_function_epilogue): Likewise.
(arm_output_function_prologue): Likewise.
* config/frv/frv.c (frv_function_prologue): Likewise.
(frv_function_epilogue): Likewise.
* config/i386/i386.c (ix86_output_function_epilogue): Likewise.
* config/ia64/ia64.c (ia64_output_function_prologue): Likewise.
(ia64_output_function_epilogue): Likewise.
* config/m32r/m32r.c (m32r_output_function_prologue): Likewise.
(m32r_output_function_epilogue): Likewise.
* config/microblaze/microblaze.c (microblaze_function_prologue)
(microblaze_function_epilogue): Likewise.
* config/mips/mips.c (mips_output_function_prologue): Likewise.
(mips_output_function_epilogue): Likewise.
* config/mmix/mmix.c (mmix_target_asm_function_prologue): Likewise.
(mmix_target_asm_function_epilogue): Likewise.
* config/msp430/msp430.c (msp430_start_function): Likewise.
* config/nds32/nds32.c (nds32_asm_function_prologue): Likewise.
(nds32_asm_function_epilogue): Likewise.
* config/nios2/nios2.c (nios2_asm_function_prologue): Likewise.
* config/pa/pa.c (pa_output_function_prologue): Likewise.
(pa_output_function_epilogue): Likewise.
* config/powerpcspe/powerpcspe.c (rs6000_output_function_prologue)
(rs6000_output_function_epilogue): Likewise.
* config/rl78/rl78.c (rl78_start_function): Likewise.
* config/rs6000/rs6000.c (rs6000_output_function_prologue): Likewise.
(rs6000_output_function_epilogue): Likewise.
* config/rx/rx.c (rx_output_function_prologue): Likewise.
* config/sh/sh.c (sh_output_function_epilogue): Likewise.
* config/sparc/sparc.c (sparc_asm_function_prologue): Likewise.
(sparc_asm_function_epilogue): Likewise.

Co-Authored-By: Alan Hayward <alan.hayward@arm.com>
Co-Authored-By: David Sherwood <david.sherwood@arm.com>
From-SVN: r251232

Add a type_has_mode_precision_p helper function

...to replace instances of:

    TYPE_PRECISION (t) == GET_MODE_PRECISION (TYPE_MODE (t))

These conditions would need to be rewritten with variable-sized
modes anyway.

2017-08-21  Richard Sandiford  <richard.sandiford@linaro.org>

gcc/
* tree.h (type_has_mode_precision_p): New function.
* convert.c (convert_to_integer_1): Use it.
* expr.c (expand_expr_real_2): Likewise.
(expand_expr_real_1): Likewise.
* fold-const.c (fold_single_bit_test_into_sign_test): Likewise.
* match.pd: Likewise.
* tree-ssa-forwprop.c (simplify_rotate): Likewise.
* tree-ssa-math-opts.c (convert_mult_to_fma): Likewise.
* tree-tailcall.c (process_assignment): Likewise.
* tree-vect-loop.c (vectorizable_reduction): Likewise.
* tree-vect-patterns.c (vect_recog_vector_vector_shift_pattern)
(vect_recog_mult_pattern, vect_recog_divmod_pattern): Likewise.
* tree-vect-stmts.c (vectorizable_conversion): Likewise.
(vectorizable_assignment): Likewise.
(vectorizable_shift): Likewise.
(vectorizable_operation): Likewise.
* tree-vrp.c (register_edge_assert_for_2): Likewise.

From-SVN: r251231

This patch simplifies pow (C, x) into exp (x * C1) if C > 0, C1 = log (C).

Do this only for fast-math as accuracy is reduced.  This is much faster
since pow is more complex than exp.

    gcc/
* match.pd: Add pow (C, x) simplification

From-SVN: r251230

* de.po: Update.

From-SVN: r251228

re PR c++/81899 (ICE: Segmentation fault)

PR c++/81899
* pt.c (instantiate_class_template_1):
BOUND_TEMPLATE_TEMPLATE_PARM is never friend-injected.

PR c++/81899
* g++.dg/template/pr81899.C: New.

From-SVN: r251227

re PR tree-optimization/81900 (GCC trunk miscompiles Perl / __sigsetjmp issue)

2017-08-21  Richard Biener  <rguenther@suse.de>

PR tree-optimization/81900
* tree-ssa-pre.c (compute_antic_aux): Properly compute changed
for blocks with abnormal predecessors.
(compute_antic): Do not set visited flag prematurely.

* gcc.dg/torture/pr81900.c: New testcase.

From-SVN: r251226

re PR target/79883 (avr i18n: untranslated "interrupt" or "signal")

PR target/79883
* config/avr/avr.c (avr_set_current_function): Typo in diagnostic.

From-SVN: r251223

Tabify 2017-08-21  Richard Biener  <rguenther@suse.de> entry

From-SVN: r251222

Move vector_type_mode to tree.c

...so that it's possible to use TYPE_MODE in tree.h.

2017-08-21  Richard Sandiford  <richard.sandiford@linaro.org>

gcc/
* stor-layout.h (vector_type_mode): Move to...
* tree.h (vector_type_mode): ...here.
* stor-layout.c (vector_type_mode): Move to...
* tree.c (vector_type_mode): ...here.  Include rtl.h and regs.h.

From-SVN: r251221

re PR go/78628 (GO fails to build a translation unit decl)

2017-08-21  Richard Biener  <rguenther@suse.de>

include/
* simple-object.h (simple_object_copy_lto_debug_sections): New
function.

libiberty/
* simple-object-common.h (struct simple_object_functions): Add
copy_lto_debug_sections hook.
* simple-object.c: Include fcntl.h.
(handle_lto_debug_sections): New helper function.
(simple_object_copy_lto_debug_sections): New function copying
early LTO debug sections to regular debug sections in a new file.
(simple_object_start_write): Handle NULL segment_name.
* simple-object-coff.c (simple_object_coff_functions): Adjust
for not implemented copy_lto_debug_sections hook.
* simple-object-mach-o.c (simple_object_mach_o_functions): Likewise.
* simple-object-xcoff.c (simple_object_xcoff_functions): Likewise.
* simple-object-elf.c (SHT_NULL, SHT_SYMTAB, SHT_RELA, SHT_REL,
SHT_GROUP): Add various sectopn header types.
(SHF_EXCLUDE): Add flag.
(Elf32_External_Sym, Elf64_External_Sym): Add symbol struct.
(ELF_ST_BIND, ELF_ST_TYPE, ELF_ST_INFO): Add accessors.
(STT_OBJECT, STT_FUNC, STT_TLS, STT_GNU_IFUNC): Add Symbol types.
(STV_DEFAULT): Add symbol visibility.
(SHN_COMMON): Add special section index name.
(struct simple_object_elf_write): New.
(simple_object_elf_start_write): Adjust for new private data.
(simple_object_elf_write_shdr): Pass in values for all fields
we write.
(simple_object_elf_write_to_file): Adjust.  Copy from recorded
section headers if requested.
(simple_object_elf_release_write): Release private data.
(simple_object_elf_copy_lto_debug_sections): Copy and rename sections
as denoted by PFN and all their dependences, symbols and relocations
to the empty destination file.
(simple_object_elf_functions): Adjust for copy_lto_debug_sections hook.

        gcc/
        * debug.h (struct gcc_debug_hooks): Add die_ref_for_decl and
        register_external_die hooks.
        (debug_false_tree_charstarstar_uhwistar): Declare.
        (debug_nothing_tree_charstar_uhwi): Likewise.
        * debug.c (do_nothing_debug_hooks): Adjust.
        (debug_false_tree_charstarstar_uhwistar): New do nothing.
        (debug_nothing_tree_charstar_uhwi): Likewise.
        * dbxout.c (dbx_debug_hooks): Adjust.
        (xcoff_debug_hooks): Likewise.
        * sdbout.c (sdb_debug_hooks): Likewise.
        * vmsdbgout.c (vmsdbg_debug_hooks): Likewise.
        * dwarf2out.c (macinfo_label_base): New global.
(dwarf2out_register_external_die): New function for the
register_external_die hook.
        (dwarf2out_die_ref_for_decl): Likewise for die_ref_for_decl.
        (dwarf2_debug_hooks): Use them.
        (dwarf2_lineno_debug_hooks): Adjust.
        (struct die_struct): Add with_offset flag.
        (DEBUG_LTO_DWO_INFO_SECTION, DEBUG_LTO_INFO_SECTION,
        DEBUG_LTO_DWO_ABBREV_SECTION, DEBUG_LTO_ABBREV_SECTION,
        DEBUG_LTO_DWO_MACINFO_SECTION, DEBUG_LTO_MACINFO_SECTION,
        DEBUG_LTO_DWO_MACRO_SECTION, DEBUG_LTO_MACRO_SECTION,
        DEBUG_LTO_LINE_SECTION, DEBUG_LTO_DWO_STR_OFFSETS_SECTION,
        DEBUG_LTO_STR_DWO_SECTION, DEBUG_STR_LTO_SECTION): New macros
        defining section names for the early LTO debug variants.
(reset_indirect_string): New helper.
        (add_AT_external_die_ref): Helper for dwarf2out_register_external_die.
        (print_dw_val): Add support for offsetted symbol references.
(get_ultimate_context): Split out from is_cxx.
(is_cxx): Use get_ultimate_context.
(is_fortran): Add decl overload.
        (compute_comp_unit_symbol): Split out worker from
compute_section_prefix.
        (compute_section_prefix): Call compute_comp_unit_symbol and
set comdat_type_p here.
        (output_die): Skip DIE symbol output for the LTO added one.
        Handle DIE symbol references with offset.
        (output_comp_unit): Guard section name mangling properly.
        For LTO debug sections emit a symbol at the section beginning
        which we use to refer to its DIEs.
        (add_abstract_origin_attribute): For DIEs registered via
        dwarf2out_register_external_die directly refer to the early
        DIE rather than indirectly through the shadow one we created.
Remove obsolete call to dwarf2out_abstract_function for
non-function/block origins.
        (gen_array_type_die): When generating early LTO debug do
        not emit DW_AT_string_length.
        (gen_formal_parameter_die): Do not re-create DIEs for PARM_DECLs
        late when in LTO.  As suggested place a gcc_unreachable for
the DECL_ABSTRACT_P case.
        (gen_subprogram_die): Avoid another specification DIE
        for early built declarations/definitions for the late LTO case.
        (gen_variable_die): Add type references for late duplicated VLA dies
        when in late LTO.
        (gen_inlined_subroutine_die): Do not call dwarf2out_abstract_function,
        we have the abstract instance already.
        (process_scope_var): Adjust decl DIE contexts in LTO which
        first puts them in limbo.
        (gen_decl_die): Do not generate type DIEs late apart from
        types for VLAs or for decls we do not yet have a DIE.  Do not
call dwarf2out_abstract_function late.
        (dwarf2out_early_global_decl): Make sure to create DIEs
        for abstract instances of a decl first.
        (dwarf2out_late_global_decl): Adjust comment.
        (output_macinfo_op): With multiple macro sections use
macinfo_label_base to distinguish labels.
        (output_macinfo): Likewise.  Update macinfo_label_base.
Pass in the line info label.
(note_variable_value_in_expr): When generating LTO resolve
all variable values here by generating DIEs as needed.
        (init_sections_and_labels): Add early LTO debug flag parameter
        and generate different sections and names if set.  Add generation
        counter for the labels so we can have multiple of them.
        (reset_dies): Helper to allow DIEs to be output multiple times.
        (dwarf2out_finish): When outputting DIEs to the fat part of an
LTO object first reset DIEs.
        (dwarf2out_early_finish): Output early DIEs when generating LTO.
(modified_type_die): Check for decl_ultimate_origin being self
before recursing.
(gen_type_die_with_usage): Likewise.
(gen_typedef_die): Allow decl_ultimate_origin being self.
        (set_decl_abstract_flags): Remove.
        (set_block_abstract_flags): Likewise.
        (dwarf2out_abstract_function): Treat the early generated DIEs
        as the abstract copy and only add DW_AT_inline and
        DW_AT_artificial here and call set_decl_origin_self.
If the DIE has an abstract origin don't do anything.
* tree.c (free_lang_data): Build a dummy TRANSLATION_UNIT_DECL
if we have none yet (Go fails to build one, PR78628).
(variably_modified_type_p): Prevent endless recursion for Ada
cyclic pointer types.
        * lto-streamer-in.c: Include debug.h.
        (dref_queue): New global.
        (lto_read_tree_1): Stream in DIE references.
        (lto_input_tree): Register DIE references.
(input_function): Stream DECL_DEBUG_ARGS.
        * lto-streamer-out.c: Include debug.h.
        (lto_write_tree_1): Output DIE references.
        (DFS::DFS_write_tree_body): Follow DECL_ABSTRACT_ORIGIN.
Force a TRANSLATION_UNIT_DECL DECL_CONTEXT for file-scope decls.
(output_function): Stream DECL_DEBUG_ARGS.
        * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
        Stream DECL_ABSTRACT_ORIGIN.
        * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
(write_ts_decl_minimal_tree_pointers): Force a TRANSLATION_UNIT_DECL
DECL_CONTEXT for file-scope decls.
        * lto-streamer.h (struct dref_entry): Declare.
        (dref_queue): Likewise.
* cfgexpand.c (pass_expand::execute): Do not call the
outlining_inline_function hook here.
        * lto-wrapper.c (debug_obj): New global.
        (tool_cleanup): Unlink it if required.
        (debug_objcopy): New function.
        (run_gcc): Handle early debug sections in the IL files by
        extracting them to separate files, partially linkin them and
        feeding the result back as result to the linker.

        * config/darwin.h (DEBUG_LTO_INFO_SECTION, DEBUG_LTO_ABBREV_SECTION,
        DEBUG_LTO_MACINFO_SECTION, DEBUG_LTO_LINE_SECTION,
        DEBUG_STR_LTO_SECTION, DEBUG_LTO_MACRO_SECTION): Put early debug
        sections into a separate segment.
        * config/darwin.c (darwin_asm_named_section): Handle __GNU_DWARF_LTO
        segments.
        (darwin_asm_dwarf_section): Likewise.
        (darwin_asm_output_dwarf_offset): Likewise.

* config/i386/i386.c (make_resolver_func): Set DECL_IGNORED_P.

        lto/
        * lto.c (unify_scc): Truncate DIE reference queue for dropped SCCs.
        (lto_read_decls): Process TRANSLATION_UNIT_DECLs.  Remove
        TYPE_DECL debug processing, register DIE references from
        prevailing SCCs with the debug machinery.
        (lto_section_with_id): Handle LTO debug sections.

libstdc++/
* testsuite/libstdc++-prettyprinters/prettyprinters.exp: Run all
tests with -flto as well if supported.

        testsuite/
* c-c++-common/asan/global-overflow-1.c: Adjust diagnostic location
regex to handle the LTO case.
* c-c++-common/asan/heap-overflow-1.c: Likewise.
* c-c++-common/asan/misalign-1.c: Likewise.
* c-c++-common/asan/misalign-2.c: Likewise.
* c-c++-common/asan/null-deref-1.c: Likewise.
* c-c++-common/asan/stack-overflow-1.c: Likewise.
* c-c++-common/asan/strncpy-overflow-1.c: Likewise.
* c-c++-common/asan/use-after-free-1.c: Likewise.
* c-c++-common/asan/alloca_big_alignment.c: Likewise.
* c-c++-common/asan/alloca_detect_custom_size.c: Likewise.
* c-c++-common/asan/alloca_overflow_partial.c: Likewise.
* c-c++-common/asan/alloca_overflow_right.c: Likewise.
* c-c++-common/asan/alloca_underflow_left.c: Likewise.
* g++.dg/asan/large-func-test-1.C: Likewise.
* gfortran.dg/save_6.f90: Add -flto -g variant of save_5.f90.

From-SVN: r251220

Pass rtx and index to read-md.c iterator routines

The read-md.c iterator callbacks previously used a void * to record the
position at which the iterator value should be installed.  This doesn't
scale easily to the SUBREG_BYTE representation used by a later patch,
so this patch replaces the void * with both an rtx and an operand
number.  The operand number is ignored for modes and codes.

2017-08-21  Richard Sandiford  <richard.sandiford@linaro.org>
    Alan Hayward  <alan.hayward@arm.com>
    David Sherwood  <david.sherwood@arm.com>

gcc/
* read-md.h (md_reader::record_potential_iterator_use): Replace
pointer argument with an rtx and an index.
* read-rtl.c (iterator_group::apply_iterator): Likewise.
(apply_mode_iterator): Likewise.
(apply_code_iterator): Likewise.
(apply_int_iterator): Likewise.
(apply_subst_iterator): Likewise.
(record_iterator_use): Likewise.
(record_attribute_use): Likewise.
(md_reader::record_potential_iterator_use): Likewise.  Update calls
to record_iterator_use and apply_iterator.
(iterator_use): Replace ptr with x and index.
(attribute_use): Likewise.
(apply_attribute_uses): Update calls to apply_iterator.
(apply_iterators): Likewise.  Update initialization of iterator_use.
(rtx_reader::read_rtx_code): Update calls to record_iterator_use
and record_potential_iterator_use.
(rtx_reader::read_rtx_operand): Likewise.

Co-Authored-By: Alan Hayward <alan.hayward@arm.com>
Co-Authored-By: David Sherwood <david.sherwood@arm.com>
From-SVN: r251219

Fix bogus CONST_WIDE_INT hash

The CONST_WIDE_INT case in const_rtx_hash_1 started the hash
with the precision of the mode, but the mode is always VOIDmode.

2017-08-21  Richard Sandiford  <richard.sandiford@linaro.org>
    Alan Hayward  <alan.hayward@arm.com>
    David Sherwood  <david.sherwood@arm.com>

gcc/
* varasm.c (const_rtx_hash_1): Don't hash in the mode of a
CONST_WIDE_INT.

Co-Authored-By: Alan Hayward <alan.hayward@arm.com>
Co-Authored-By: David Sherwood <david.sherwood@arm.com>
From-SVN: r251218

re PR tree-optimization/81884 (Invalid code generation with zero size arrays or flexible array members)

2017-08-21  Richard Biener  <rguenther@suse.de>

PR middle-end/81884
* tree-ssa-alias.c (stmt_kills_ref_p): Handle array accesses
at struct end conservatively when comparing common bases.

* g++.dg/torture/pr81884.C: New testcase.

From-SVN: r251217

tree-ssa-loop-im.c (struct lim_aux_data): Add ref index member.

2017-08-21  Richard Biener  <rguenther@suse.de>

* tree-ssa-loop-im.c (struct lim_aux_data): Add ref index member.
(mem_ref_in_stmt): Remove.
(determine_max_movement): Use ref index to get at the reference.
(invariantness_dom_walker::before_dom_children): Deal with
lim data already initialized.
(gather_mem_refs_stmt): Initialize lim data and record ref index.

From-SVN: r251216

Daily bump.

From-SVN: r251215

re PR c/81854 (weak alias of an incompatible symbol accepted)

libstdc++/ChangeLog:

PR c/81854
* src/c++98/compatibility.cc (_GLIBCXX_3_4_SYMVER): Declare alias
target as a C++ function with no prototype.
(_GLIBCXX_3_4_5_SYMVER): Ditto.

From-SVN: r251211

Fix typo.

From-SVN: r251210

re PR testsuite/81056 (FAIL: 17_intro/names.cc (test for excess errors))

PR testsuite/81056
* testsuite/17_intro/names.cc: Undef 'd' and 'r' on __hpux__

From-SVN: r251208

re PR ipa/77732 (FAIL: gcc.dg/ipa/ipcp-cstagg-7.c scan-ipa-dump-times cp "Discovered an indirect  call to a known target" 3)

PR ipa/77732
* gcc.dg/ipa/ipcp-cstagg-7.c: Xfail on 32-bit hppa.

From-SVN: r251206

Daily bump.

From-SVN: r251205

i386.h (OPTION_MASK_ISA_ROUND): Remove.

* config/i386/i386.h (OPTION_MASK_ISA_ROUND): Remove.
(TARGET_ISA_ROUND): Ditto.
(TARGET_ROUND): Ditto.
* config/i386/i386.c: Substitute TARGET_ROUND with TARGET_SSE4_1.
* config/i386/i386.md: Ditto.
* config/i386/sse.md: Ditto.
* config/i386/i386-builtin.def: Substitute OPTION_MASK_ISA_ROUND
with OPTION_MASK_ISA_SSE4_1.

From-SVN: r251201

re PR target/81894 (Typo in x86 built-in function list)

PR target/81894
* doc/extend.texi (x86 Built-in Functions): Correct the name of
__builtin_ia32_lzcnt_u16.

From-SVN: r251197

Daily bump.

From-SVN: r251196

jit: fix segfault with autovectorization (PR tree-optimization/46805)

libgccjit ran into its own version of PR tree-optimization/46805 (seen
with the Go frontend); this patch fixes it in the same way.

gcc/jit/ChangeLog:
PR tree-optimization/46805
* dummy-frontend.c (jit_langhook_parse_file): Handle vector types.

gcc/testsuite/ChangeLog:
PR tree-optimization/46805
* jit.dg/all-non-failing-tests.h: Add test-autovectorize.c.
* jit.dg/test-autovectorize.c: New test case.

From-SVN: r251192

jit: make simpler reproducers

The C reproducers generated by gcc_jit_context_dump_reproducer_to_file
contain numerous pointer values (from %p) to ensure uniqueness of the
identifiers, but this makes them less readable than they could be.

This patch updates reproducer::make_identifier so that the pointer
is only added if it's necessary for uniqueness.

gcc/jit/ChangeLog:
* jit-recording.c (class gcc::jit::reproducer): Rename field
"m_identifiers" to "m_map_memento_to_identifier".  Add field
"m_set_identifiers" and struct hash_traits for it.
(gcc::jit::reproducer::reproducer): Update for above.
(convert_to_identifier): New function.
(gcc::jit::reproducer::ensure_identifier_is_unique): New method.
(gcc::jit::reproducer::make_identifier): Avoid appending the %p
unless necessary for uniqueness.  Update for field renaming.
(gcc::jit::reproducer::get_identifier): Update for field renaming.

From-SVN: r251191

re PR target/80210 (ICE in in extract_insn, at recog.c:2311 on ppc64 for with __builtin_pow)

gcc/
PR target/80210
* config/rs6000/rs6000.c (rs6000_activate_target_options): New function.
(rs6000_set_current_function): Rewrite function to use it.

gcc/testsuite/
PR target/80210
* gcc.target/powerpc/pr80210.c: New test.

From-SVN: r251190

re PR go/81893 (compilation error in libgo starting with r251127)

PR go/81893
    runtime: only use PPC GNU/Linux register code on little endian

    Reportedly the current code builds on little endian but not big endian.

    Fixes https://gcc.gnu.org/PR81893.

    Reviewed-on: https://go-review.googlesource.com/57110

From-SVN: r251188

Simplify allocator usage in unordered containers

* include/bits/hashtable_policy.h (_ReuseOrAllocNode): Remove
__value_alloc_type and __value_alloc_traits typedefs.
(_ReuseOrAllocNode::operator()): Call construct and destroy on the
node allocator.
(_Hashtable_alloc): Simplify __value_alloc_traits typedef.
(_Hashtable_alloc<_NodeAlloc>::_M_allocate_node(_Args&&...)): Call
construct on the node allocator.
(_Hashtable_alloc<_NodeAlloc>::_M_deallocate_node(__node_type*)): Call
destroy on the node allocator.

From-SVN: r251187

C++: fix ordering of missing std #include suggestion (PR c++/81514)

gcc/cp/ChangeLog:
PR c++/81514
* name-lookup.c (maybe_suggest_missing_header): Convert return
type from void to bool; return true iff a suggestion was offered.
(suggest_alternative_in_explicit_scope): Move call to
maybe_suggest_missing_header to before use of best_match, and
return true if the former offers a suggestion.

gcc/testsuite/ChangeLog:
PR c++/81514
* g++.dg/lookup/empty.h: New file.
* g++.dg/lookup/missing-std-include-2.C: Replace include of
stdio.h with empty.h and a declaration of a "std::sprintf" not based
on a built-in.

From-SVN: r251186

PR libstdc++/81891 fix double-free in hashtable constructor

PR libstdc++/81891
* include/bits/hashtable.h (_Hashtable(_InputIterator, _InputIterator,
size_type, const _H1&, const _H2&, const _Hash&, const _Equal&,
const _ExtractKey&, const allocator_type&)): Let destructor do clean
up if an exception is thrown.
* testsuite/23_containers/unordered_map/cons/81891.cc: New.

From-SVN: r251185

misc/cgo/test: make cgo tests run on AIX

    Reviewed-on: https://go-review.googlesource.com/56910

From-SVN: r251182

c-parser.c (c_parser_postfix_expression): Remove unused code.

* c-parser.c (c_parser_postfix_expression): Remove unused code.  Update
commentary.

From-SVN: r251181

Add warn_if_not_aligned attribute

Add warn_if_not_aligned attribute as well as  command line options:
-Wif-not-aligned and -Wpacked-not-aligned.

__attribute__((warn_if_not_aligned(N))) causes compiler to issue a
warning if the field in a struct or union is not aligned to N:

typedef unsigned long long __u64
  __attribute__((aligned(4),warn_if_not_aligned(8)));

struct foo
{
  int i1;
  int i2;
  __u64 x;
};

__u64 is aligned to 4 bytes.  But inside struct foo, __u64 should be
aligned at 8 bytes.  It is used to define struct foo in such a way that
struct foo has the same layout and x has the same alignment when __u64
is aligned at either 4 or 8 bytes.

Since struct foo is normally aligned to 4 bytes, a warning will be issued:

warning: alignment 4 of 'struct foo' is less than 8

Align struct foo to 8 bytes:

struct foo
{
  int i1;
  int i2;
  __u64 x;
} __attribute__((aligned(8)));

silences the warning.  It also warns the field with misaligned offset:

struct foo
{
  int i1;
  int i2;
  int i3;
  __u64 x;
} __attribute__((aligned(8)));

warning: 'x' offset 12 in 'struct foo' isn't aligned to 8

This warning is controlled by -Wif-not-aligned and is enabled by default.

When -Wpacked-not-aligned is used, the same warning is also issued for
the field with explicitly specified alignment in a packed struct or union:

struct __attribute__ ((aligned (8))) S8 { char a[8]; };
struct __attribute__ ((packed)) S {
  struct S8 s8;
};

warning: alignment 1 of 'struct S' is less than 8

This warning is disabled by default and enabled by -Wall.

gcc/

PR c/53037
* print-tree.c (print_node): Support DECL_WARN_IF_NOT_ALIGN
and TYPE_WARN_IF_NOT_ALIGN.
* stor-layout.c (do_type_align): Merge DECL_WARN_IF_NOT_ALIGN.
(handle_warn_if_not_align): New.
(place_union_field): Call handle_warn_if_not_align.
(place_field): Call handle_warn_if_not_align.  Copy
TYPE_WARN_IF_NOT_ALIGN.
(finish_builtin_struct): Copy TYPE_WARN_IF_NOT_ALIGN.
(layout_type): Likewise.
* tree-core.h (tree_type_common): Add warn_if_not_align.  Set
spare to 18.
(tree_decl_common): Add warn_if_not_align.
* tree.c (build_range_type_1): Copy TYPE_WARN_IF_NOT_ALIGN.
* tree.h (TYPE_WARN_IF_NOT_ALIGN): New.
(SET_TYPE_WARN_IF_NOT_ALIGN): Likewise.
(DECL_WARN_IF_NOT_ALIGN): Likewise.
(SET_DECL_WARN_IF_NOT_ALIGN): Likewise.
* doc/extend.texi: Document warn_if_not_aligned attribute.
* doc/invoke.texi: Document -Wif-not-aligned and
-Wpacked-not-aligned.

gcc/c-family/

PR c/53037
* c-attribs.c (handle_warn_if_not_aligned_attribute): New.
(c_common_attribute_table): Add warn_if_not_aligned.
(handle_aligned_attribute): Renamed to ...
(common_handle_aligned_attribute): Remove argument, name, and add
argument, warn_if_not_aligned.  Handle warn_if_not_aligned.
(handle_aligned_attribute): New.
* c.opt: Add -Wif-not-aligned and -Wpacked-not-aligned.

gcc/c/

PR c/53037
* c-decl.c (merge_decls): Also merge DECL_WARN_IF_NOT_ALIGN.
(check_bitfield_type_and_width): Don't allow bit-field with
warn_if_not_aligned type.

gcc/cp/

PR c/53037
* decl.c (duplicate_decls): Also merge DECL_WARN_IF_NOT_ALIGN.
* decl2.c (grokbitfield): Don't allow bit-field with
warn_if_not_aligned type.

gcc/testsuite/

PR c/53037
* c-c++-common/pr53037-5.c: New test.
* g++.dg/pr53037-1.C: Likewise.
* g++.dg/pr53037-2.C: Likewise.
* g++.dg/pr53037-3.C: Likewise.
* g++.dg/pr53037-4.C: Likewise.
* gcc.dg/pr53037-1.c: Likewise.
* gcc.dg/pr53037-2.c: Likewise.
* gcc.dg/pr53037-3.c: Likewise.
* gcc.dg/pr53037-4.c: Likewise.

From-SVN: r251180

cmd/go: pass -funwind-tables when compiling C code

    Using -funwind-tables is necessary to permit Go code to correctly
    throw a panic through C code.  This hasn't been necessary in the past
    as -funwind-tables is the default on x86.  However, it is not the
    default for PPC AIX.

    Reviewed-on: https://go-review.googlesource.com/56650

From-SVN: r251179

Daily bump.

From-SVN: r251178

Fix build of --enable-gather-detailed-mem-stats (PR bootstrap/81864).

2017-08-17  Martin Liska  <mliska@suse.cz>

PR bootstrap/81864
* tree-loop-distribution.c (ddrs_table): Change type to pointer
type.
(get_data_dependence): Use it as pointer type.
(distribute_loop): Likewise.

From-SVN: r251165

altivec.md (UNSPEC_VMRGOW_DIRECT): New constant.

2017-08-17  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>

* config/rs6000/altivec.md (UNSPEC_VMRGOW_DIRECT): New constant.
(p8_vmrgew_v4sf_direct): Generalize to p8_vmrgew_<mode>_direct.
(p8_vmrgow_<mode>_direct): New define_insn.
* config/rs6000/rs6000.c (altivec_expand_vec_perm_const): Properly
handle endianness for vmrgew and vmrgow permute patterns.

From-SVN: r251161

cp-tree.def (TEMPLATE_TEMPLATE_PARM): Remove stale comment.

* cp-tree.def (TEMPLATE_TEMPLATE_PARM): Remove stale comment.
* cp-tree.h (ENUM_TEMPLATE_INFO): Delete.
(TYPE_TEMPLATE_INFO): Simplify.
(SET_TYPE_TEMPLATE_INFO): Simplify.

From-SVN: r251159

altivec.md (VParity): Remove TARGET_VSX_TIMODE.

gcc/
* config/rs6000/altivec.md (VParity): Remove TARGET_VSX_TIMODE.
* config/rs6000/rs6000-cpus.def: Remove comment.
(ISA_2_7_MASKS_SERVER): Delete OPTION_MASK_VSX_TIMODE;
(POWERPC_MASKS): Likewise.
* config/rs6000/rs6000.c (rs6000_hard_regno_mode_ok): Remove unneeded
use of TARGET_VSX_TIMODE.
(rs6000_setup_reg_addr_masks): Change TARGET_VSX_TIMODE to TARGET_VSX.
(rs6000_init_hard_regno_mode_ok): Remove unneeded uses of
TARGET_VSX_TIMODE.  Change use of TARGET_VSX_TIMODE to TARGET_VSX.
(rs6000_option_override_internal): Remove dead code.
(rs6000_legitimize_address): Change TARGET_VSX_TIMODE to TARGET_VSX.
(rs6000_legitimize_reload_address): Likewise.
(rs6000_legitimate_address_p): Likewise.
(rs6000_opt_masks): Delete "vsx-timode".
(rs6000_disable_incompatible_switches): Remove mention of -mvsx-timode
from function comment.
* config/rs6000/rs6000.h (MASK_VSX_TIMODE): Delete.
* config/rs6000/rs6000.md (FMOVE128_GPR): Remove TARGET_VSX_TIMODE.
(V16QI, V8HI, V4SI, V4SF, V2DI, V2DF, V1TI): Remove useless empty
condition.
* config/rs6000/rs6000.opt (mvsx-timode): Replace with stub.
* config/rs6000/vector.md (VEC_IP): Remove TARGET_VSX_TIMODE.
* config/rs6000/vsx.md (VSX_LE_128): Likewise.
(VSX_TI): Likewise.
(VSX_M): Likewise.
(define_peephole2): Likewise.

gcc/testsuite/
* gcc.target/powerpc/p8vector-int128-1.c: Remove use of -mvsx-timode.
* gcc.target/powerpc/p9-vparity.c: Likewise.
* gcc.target/powerpc/pr68805.c: Likewise.
* gcc.target/powerpc/pr80098-4.c: Remove useless test case.

From-SVN: r251158

PR c/81859 - [8 Regression] valgrind error from warn_about_normalization

gcc/ChangeLog:

PR c/81859
* pretty-print.c (pp_format): Use strnlen in %.*s to avoid reading
past the end of an array.
(test_pp_format): Add test cases.

From-SVN: r251157

Add missing ECF_NOTHROW flags to internal.def

This patch adds missing ECF_NOTHROW flags to the vectorisable
integer internal functions.

2017-08-17  Richard Sandiford  <richard.sandiford@linaro.org>

gcc/
* internal-fn.def (CLRSB, CLZ, CTZ, FFS, PARITY, POPCOUNT): Add
missing ECF_NOTHROW flags.

From-SVN: r251155

lex.c (maybe_add_lang_type_raw): BOUND_TEMPLATE_TEMPLATE_PARMs don't need lang_type.

* lex.c (maybe_add_lang_type_raw): BOUND_TEMPLATE_TEMPLATE_PARMs
don't need lang_type.
(cxx_make_type): Use maybe_add_lang_type_raw return value.
* mangle.c (CLASSTYPE_TEMPLATE_ID_P): Don't rely on
TYPE_LANG_SPECIFIC.

From-SVN: r251154

re PR target/72804 (Poor code gen with -mvsx-timode)

gcc/
PR target/72804
* config/rs6000/vsx.md (*vsx_le_permute_<mode>): Add support for
operands residing in integer registers.
(*vsx_le_perm_load_<mode>): Likewise.
(*vsx_le_perm_store_<mode>): Likewise.
(define_peephole2): Add peepholes to optimize the above.

gcc/testsuite/
PR target/72804
* gcc.target/powerpc/pr72804.c: New test.

From-SVN: r251153

re PR middle-end/81814 (Incorrect behaviour at -O0 (conditional operator))

PR middle-end/81814
* fold-const.c (operand_equal_for_comparison_p): Remove code that used
to mimic what shorten_compare did.  Change the return type to bool.
(fold_cond_expr_with_comparison): Update call to
operand_equal_for_comparison_p.
(fold_ternary_loc): Likewise.

* gcc.dg/torture/pr81814.c: New test.

From-SVN: r251152

re PR ada/81878 (--disable-bootstrap --enable-languages=ada fails)

2017-08-17  Richard Biener  <rguenther@suse.de>

PR ada/81878
* Makefile.in (CXX_LFLAGS): Remove.
(TOOLS_FLAGS_TO_PASS_NATIVE): Pass $(CXX) as CXX.
(TOOLS_FLAGS_TO_PASS_RE): Likewise.

From-SVN: r251150

[AArch64] Improve SIMD store of zero.

This patch changes patterns in aarch64-simd.md to replace

    movi    v0.4s, 0
    str    q0, [x0, 16]

With:

    stp xzr, xzr, [x0, 16]

When we are storing zeros to vectors like this:

    void f(uint32x4_t *p) {
      uint32x4_t x = { 0, 0, 0, 0};
      p[1] = x;
    }

gcc/
2017-08-17  Jackson Woodruff  <jackson.woodruff@arm.com>

* aarch64-simd.md (mov<mode>): No longer force zero immediate into
register.
(*aarch64_simd_mov<mode>): Add new case for stp using zero immediate.

gcc/testsuite/
2017-08-17  Jackson Woodruff  <jackson.woodruff@arm.com>

* gcc.target/aarch64/simd/vect_str_zero.c: New testcase.

From-SVN: r251149

cp-tree.h (struct lang_type): Remove template_info field.

* cp-tree.h (struct lang_type): Remove template_info field.
(CLASSTYPE_TEMPLATE_INFO): Use TYPE_LANG_SLOT_1.
(TEMPLATE_TEMPLATE_PARM_TEMPLATE_INFO): Likewise.

From-SVN: r251148

Add the missing ChangeLog for r251076

From-SVN: r251147

tree-ssa-structalias.c (solve_graph): When propagating to successors update the graphs succ edges and avoid duplicate...

2017-08-17  Richard Biener  <rguenther@suse.de>

* tree-ssa-structalias.c (solve_graph): When propagating
to successors update the graphs succ edges and avoid duplicate work.

From-SVN: r251146

re PR target/81861 (ASan pr64820.c testcase segfaults with LTO and -fstack-protector-strong)

2017-08-17  Maxim Ostapenko  <m.ostapenko@samsung.com>

PR target/81861
* config/i386/i386.c (ix86_option_override_internal): Save target
specific options after ix86_stack_protector_guard_reg was changed.

From-SVN: r251145

Require effective target nonlocal_goto for ipa/pr81696.c

2017-08-17  Tom de Vries  <tom@codesourcery.com>

* gcc.dg/ipa/pr81696.c: Require effective target nonlocal_goto.

From-SVN: r251144

re PR fortran/81827 (Large compile time with derived-type rrays)

2017-08-17  Richard Biener  <rguenther@suse.de>

PR tree-optimization/81827
* tree-ssa-structalias.c (struct variable_info): Add is_reg_var
flag.
(new_var_info): Initialize it conservatively.
(get_call_vi): Mark register vars.
(new_scalar_tmp_constraint_exp): Likewise.
(handle_rhs_call): Likewise.
(handle_const_call): Likewise.
(create_function_info_for): Likewise.
(solve_constraints): Sort varinfos to separate register from
non-register vars to pack points-to solution bitmaps during
iteration.

From-SVN: r251143

gimplify.c (gimplify_adjust_omp_clauses): Compare with 0 instead of 1.

* gimplify.c (gimplify_adjust_omp_clauses): Compare with 0 instead of
1.

From-SVN: r251142

tree-vrp.c (vrp_int_const_binop): Do not set *overflow_p to true when overflow is undefined and we saturated the...

2017-08-17  Richard Biener  <rguenther@suse.de>

* tree-vrp.c (vrp_int_const_binop): Do not set *overflow_p
to true when overflow is undefined and we saturated the
result.

* gcc.dg/tree-ssa/vrp117.c: New testcase.

From-SVN: r251141

[RS6000] PR 80938, Don't emit frame info for regs that don't need saving

It is possible when using out-of-line register saves or store multiple
to save some registers unnecessarily, for example one reg in the block
saved might be unused.  We don't need to emit frame info for those
registers as that just bloats the info, and also can result in an ICE
when shrink-wrap gives multiple paths through the function saving
different sets of registers.  Join points need to have identical frame
register save state regardless of the path taken.

This patch reverts the previous fix for PR80939 "Use SAVE_MULTIPLE
only if we restore what it saves (PR80938)" and instead fixes the PR
by correcting the frame info.  The change to rs6000_savres_strategy
is an optimization, but note that it hides the underlying problem in
the PR testcase.

PR target/80938
* config/rs6000/rs6000.c (rs6000_savres_strategy): Revert 2017-08-09.
Don't use store multiple if only one reg needs saving.
(interesting_frame_related_regno): New function.
(rs6000_frame_related): Don't emit frame info for regs that
don't need saving.
(rs6000_emit_epilogue): Likewise.

From-SVN: r251140

Daily bump.

From-SVN: r251139

runtime: better implementation of netpoll for AIX

    Reviewed-on: https://go-review.googlesource.com/54170

From-SVN: r251133

* fr.po: Update.

From-SVN: r251131

tree-core.h (tree_type_non_common): Rename binfo to lang_1.

* tree-core.h (tree_type_non_common): Rename binfo to lang_1.
* tree.h (TYPE_BINFO): Use type_non_common.maxval.
(TYPE_LANG_SLOT_1): Use type_non_common.lang_1, for any type.
* tree.c (free_lang_data_in_type): Use else-if chain.  Always
clear TYPE_LANG_1.  Remove obsolete member-function stripping.
(find_decls_types_r): Comment about TYPE_MAX_VALUES_RAW.
(verify_type): Adjust for TYPE_BINFO move.
* lto-streamer-out.c (DFS::DFS_write_tree_body): No need to
process TYPE_BINFO directly.
(hash_tree): Likewise.
* tree-streamer-in.c (lto_input_ts_type_non_common_tree_pointers):
Likewise.
* tree-streamer-out.c (write_ts_type_non_common_tree_pointers):
Likewise.

lto/
* lto.c (mentions_vars_p_type): Use TYPE_LANG_SLOT_1.
(compare_tree_sccs_1): No need to compare TYPE_BINFO directly.
(lto_fixup_prevailing_decls): Use TYPE_LANG_SLOT_1.

From-SVN: r251129

diagnostic-show-locus.c: remove unused field from class colorizer

gcc/ChangeLog:
* diagnostic-show-locus.c (colorizer::m_caret): Remove unused
field.

From-SVN: r251128

runtime: improvements for signal registers

    Fix dumpregs on i386, implement dumpregs for PPC Linux/AIX, get PC on AIX.

    Reviewed-on: https://go-review.googlesource.com/56170

From-SVN: r251127

patchable_function_entry-decl.c (dg-final): Adapt scan-assembler-times for alpha*-*-*.

* c-c++-common/patchable_function_entry-decl.c (dg-final): Adapt
scan-assembler-times for alpha*-*-*.
* c-c++-common/patchable_function_entry-default.c (dg-final): Ditto.
* c-c++-common/patchable_function_entry-definition.c (dg-final): Ditto.

From-SVN: r251126

re PR fortran/81116 (Last character of allocatable-length string reset to blank in an assigment)

2017-08-16  Thomas Koenig  <tkoenig@gcc.gnu.org>

PR fortran/81116
* frontend-passes.c (realloc_string_callback): If expression is a
concatenation, also check for dependency.
(constant_string_length): Check for presence of symtree.

2017-08-16  Thomas Koenig  <tkoenig@gcc.gnu.org>

PR fortran/81116
* gfortran.dg/realloc_on_assignment_29.f90:  New test.

From-SVN: r251125

re PR target/46091 (missed optimization: x86 bt/btc/bts instructions)

PR target/46091
* config/i386/i386.md (*anddi_1_btr): Change predicates of
operand 0 and operand 1 to nomimmediate_operand. Add "m" constraint.
Add ix86_binary_operator_ok to insn constraint.
(*iordi_1_bts): Ditto.
(*xordi_1_btc): Ditto.
(*btsq): Change predicate of operand 0 to nonimmediate_operand.
Update corresponding peephole2 pattern.
(*btrq): Ditto.
(*btcq): Ditto.

testsuite/ChangeLog:

PR target/46091
* gcc.target/i386/pr46091-1.c: Update scan-assembler-times.
(testm): New test function.
* gcc.target/i386/pr46091-2.c: Ditto.
* gcc.target/i386/pr46091-3.c: Ditto.

From-SVN: r251124

re PR middle-end/81832 (ICE in expand_LOOP_DIST_ALIAS, at internal-fn.c:2273)

PR tree-optimization/81832
* tree-ssa-loop-ch.c (should_duplicate_loop_header_p): Don't
copy loop header which has IFN_LOOP_DIST_ALIAS call.

gcc/testsuite
* gcc.dg/tree-ssa/pr81832.c: New test.

From-SVN: r251123

re PR middle-end/81695 (internal compiler error: in size_binop_loc, at fold-const.c:1768)

PR middle/81695
* fold-const.c (fold_indirect_ref_1): Restore original behavior
regarding size_zero_node.

From-SVN: r251119

Fix building of cross compiler (PR target/81753).

2017-08-16  Martin Liska  <mliska@suse.cz>

PR target/81753
* config.gcc: Respect previously set extra_objs in case
of darwin target.

From-SVN: r251118

PR81815: Invalid conditional reduction

We weren't checking whether the phi in a conditional reduction was
used by the condition itself (which isn't a case we handle).

2017-08-11  Richard Sandiford  <richard.sandiford@linaro.org>

gcc/
PR tree-optimization/81835
* tree-vect-loop.c (vect_is_simple_reduction): Simply checks for
the phi SSA_NAME.  Check that the condition in a COND_EXPR does
not depend on the phi.

gcc/testsuite/
PR tree-optimization/81835
* gcc.dg/vect/pr81815.c: New test.

From-SVN: r251117

[RS6000] Delete code made dead by r250482

* config/rs6000/rs6000.c (rs6000_init_hard_regno_mode_ok): Delete
dead code.

From-SVN: r251116

[RS6000] Merge rs6000_reg_live_or_pic_offset_p into save_reg_p

rs6000_reg_live_or_pic_offset_p is just save_reg_p with special
handling for the pic register and eh_return.  This merge also
simplifies the eh_return handling.  The intent of
https://gcc.gnu.org/ml/gcc-patches/2010-09/msg01838.html was to say
the PIC reg needed to be saved for eh_return, not all gprs.  Of
course, it doesn't hurt to say all gprs need to be saved for eh_return
as that is what the target-independent code does by setting DF live,
but it's unnecessary in the backend.

* config/rs6000/rs6000.c (rs6000_reg_live_or_pic_offset_p): Merge..
(save_reg_p): ..into this.  Update all callers.
(first_reg_to_save): Simplify.

From-SVN: r251115

[RS6000] Don't restore fixed regs

* config/rs6000/rs6000.c (rs6000_savres_strategy): Don't restore
fixed regs.

From-SVN: r251114

Daily bump.

From-SVN: r251113

Limit SH strncmp inline expansion (PR target/78460).

GCC mainline built for sh4-linux-gnu runs out of memory building a
glibc test, which calls strncmp with very large constant size
argument, resulting in the SH inline strncmp expansion trying to
inline a fully unrolled expansion of strncmp for that size.

This patch limits that fully unrolled expansion to the case of less
than 32 bytes.  This is explicitly *not* trying to be optimal in any
way (very likely a lower threshold makes sense), just to limit enough
to avoid the out-of-memory issue in the glibc testsuite.

I have *not* run the GCC testsuite for SH.  I have verified that this
allows the glibc testsuite to build OK, with both GCC mainline and GCC
7 branch (and that the included test builds quickly with patched GCC,
runs out of memory with unpatched GCC).

PR target/78460
PR target/67712
gcc:
* config/sh/sh-mem.cc (sh_expand_cmpnstr): Only unroll for
constant count if that count is less than 32.

gcc/testsuite:
* gcc.c-torture/compile/string-large-1.c: New test.

From-SVN: r251108

Update .po files.

* be.po, da.po, de.po, el.po, es.po, fi.po, fr.po, hr.po, id.po,
ja.po, nl.po, ru.po, sr.po, sv.po, tr.po, uk.po, vi.po, zh_CN.po,
zh_TW.po: Update.

From-SVN: r251106

gcc.c (execute): Emit friendlier message if inferior is killed by an external cause.

* gcc.c (execute): Emit friendlier message if inferior is killed
by an external cause.

From-SVN: r251104

re PR tree-optimization/81790 (ICE in vn_nary_build_or_lookup_1, at tree-ssa-sccvn.c:1738)

2017-08-15  Richard Biener  <rguenther@suse.de>

PR tree-optimization/81790
* tree-ssa-sccvn.c (vn_lookup_simplify_result): Handle both
CONSTRUCTORs from simplifying and VN.

* gcc.dg/torture/pr81790.c: New testcase.

From-SVN: r251103

builtin-attrs.def: Add comments.

gcc/ChangeLog:

* builtin-attrs.def: Add comments.

From-SVN: r251101

PR c/81117 - Improve buffer overflow checking in strncpy - part 2

gcc/ChangeLog:

PR c/81117
* doc/extend.texi (attribute nonstring): Document new attribute.

gcc/c-family/ChangeLog:

PR c/81117
* c-attribs.c (c_common_attribute_table): Add nonstring entry.
(handle_nonstring_attribute): New function.

gcc/testsuite/ChangeLog:

PR c/81117
* c-c++-common/attr-nonstring-1.c: New test.

From-SVN: r251100

PR c/81117 - Improve buffer overflow checking in strncpy - part 1

gcc/ChangeLog:

        PR c/81117
* tree-diagnostic.c (default_tree_printer): Handle %G.
* gimple-pretty-print.h (percent_G_format): Declare new function.
* gimple-pretty-print.c (percent_G_format): Define.
* tree-pretty-print.c (percent_K_format): Add argument.

gcc/c/ChangeLog:

PR c/81117
* c-objc-common.c (c_objc_common_init): Handle 'G'.

gcc/c-family/ChangeLog:

PR c/81117
* c-format.h (T89_G): New macro.
* c-format.c (local_gcall_ptr_node): New variable.
(init_dynamic_diag_info): Initialize it.

gcc/cp/ChangeLog:

PR c/81117
* error.c (cp_printer): Handle 'G'.

gcc/testsuite/ChangeLog:

PR c/81117
* gcc.dg/format/gcc_diag-10.c: Exercise %G.

From-SVN: r251098

ucnid-5.c: Skip on AIX.

        * gcc.dg/ucnid-5.c: Skip on AIX.
        * gcc.target/powerpc/pr79909.c: Skip on AIX.

From-SVN: r251097

PR translation/79998 - typo in diagnostic "specified bound %wu"

gcc/ChangeLog:
* gimple-ssa-sprintf.c (pass_sprintf_length::handle_gimple_call):
Remove a stray space.

From-SVN: r251096

re PR target/46091 (missed optimization: x86 bt/btc/bts instructions)

PR target/46091
* config/i386/i386.md (*anddi_1_btr): New insn_and_split pattern.
(*iordi_1_bts): Ditto.
(*xordi_1_btc): Ditto.

testsuite/ChangeLog:

PR target/46091
* gcc.target/i386/pr46091-1.c: New test.
* gcc.target/i386/pr46091-2.c: Ditto.
* gcc.target/i386/pr46091-3.c: Ditto.

From-SVN: r251095

[AArch64] Fix longbranch test

Fix longbranch test so it still generates long tbz branches.

    gcc/testsuite/
PR target/81643
* gcc.target/aarch64/long_branch_1.c: Improve testcase.

From-SVN: r251094