Use loop->safelen rather than loop->force_vectorize

...since the latter doesn't guarantee independence by itself.

2018-03-02  Richard Sandiford  <richard.sandiford@linaro.org>

gcc/
* tree-vect-data-refs.c (vect_analyze_data_ref_dependence)
(vect_analyze_data_ref_access): Use loop->safe_len rather than
loop->force_vectorize to check whether there is no alias.

gcc/testsuite/
* gcc.dg/vect/vect-alias-check-13.c: New test.

From-SVN: r258130

re PR rtl-optimization/84614 (wrong code with u16->u128 extension at aarch64 -fno-split-wide-types -g3 --param=max-combine-insns=3)

PR target/84614
* rtl.h (prev_real_nondebug_insn, next_real_nondebug_insn): New
prototypes.
* emit-rtl.c (next_real_insn, prev_real_insn): Fix up function
comments.
(next_real_nondebug_insn, prev_real_nondebug_insn): New functions.
* cfgcleanup.c (try_head_merge_bb): Use prev_real_nondebug_insn
instead of a loop around prev_real_insn.
* combine.c (move_deaths): Use prev_real_nondebug_insn instead of
prev_real_insn.

* gcc.dg/pr84614.c: New test.

From-SVN: r258129

re PR fortran/84219 (Failure to generate error for IO of transfer intrinsic, when MOLD has derived type components.)

2018-03-02  Paul Thomas  <pault@gcc.gnu.org>

PR fortran/84219
* gfortran.dg/coarray_47.f90: Use the correct test.

From-SVN: r258128

re PR inline-asm/84625 (ICE with empty constraint and vector constant)

PR inline-asm/84625
* config/i386/i386.c (ix86_print_operand): Use conditional
output_operand_lossage instead of gcc_assert if CONST_VECTOR is not
zero vector.

* gcc.target/i386/pr84625.c: New test.

From-SVN: r258125

re PR tree-optimization/84427 (gcc ICE at -O3 on x86_64-linux-gnu in compute_antic, at tree-ssa-pre.c:2356)

2018-03-02  Richard Biener  <rguenther@suse.de>

PR tree-optimization/84427
* tree-ssa-pre.c (bitmap_remove_expr_from_set): Remove.
(bitmap_set_subtract_values): Rewrite to handle multiple
exprs per value.
(clean): Likewise.
(prune_clobbered_mems): Likewise.
(phi_translate): Take edge instead of pred/phiblock.
(phi_translate_1): Likewise.
(phi_translate_set): Likewise.  Insert all translated
exprs for a value into the set, keeping possibly multiple
expressions per value.
(compute_antic_aux): Adjust for phi_translate changes.
When intersecting union the expressions and prune those
not in the final value set, keeping possibly multiple
expressions per value.  Do not use value-insertion
for unioning ANTIC_OUT U EXP_GEN - TMP_GEN but merge
all expressions.  Add verification that the value-sets
only shrink during iteration.
(compute_partial_antic_aux): Adjust for the phi_translate changes.
(do_pre_regular_insertion): Likewise.
(do_pre_partial_partial_insertion): Likewise.

* gcc.dg/torture/pr84427.c: New testcase.

From-SVN: r258124

re PR target/82005 (Early lto debug not implemented on Darwin)

2018-03-02  Richard Biener  <rguenther@suse.de>

PR target/82005
* config/darwin.c (saved_debug_info_level): New static global.
(darwin_asm_lto_start): Disable debug info generation for LTO out.
(darwin_asm_lto_end): Restore debug info generation settings.

From-SVN: r258123

re PR target/84534 (several powerpc test cases fail starting with r257915)

PR target/84534
* gcc.target/powerpc/vec-setup-be-long.c: Add dg-xfail-run-if on
powerpc64le*-*-linux*.
* gcc.target/powerpc/vsx-vector-6-le.c: Do not count xxlor's.
* gcc.target/powerpc/vsx-vector-6-le.p9.c: Likewise.

From-SVN: r258122

PR c++/84294 - attributes on a function template redeclaration silently discarded

gcc/cp/ChangeLog:

PR c++/84294
* decl.c (check_redeclaration_no_default_args): Merge attributes
specified on redeclarations of the same function template.
Remove dead code.

gcc/testsuite/ChangeLog:

PR c++/84294
* g++.dg/ext/attr-const.C: Remove xfail.
* g++.dg/ext/attr-malloc-3.C: New test.
* g++.dg/ext/attr-noinline-3.C: New test.
* g++.dg/ext/attr-noreturn-3.C: New test.
* g++.dg/ext/attr-nothrow-3.C: New test.
* g++.dg/ext/attr-pure.C: Remove xfail.

From-SVN: r258121

Daily bump.

From-SVN: r258120

re PR sanitizer/70875 (ICE in get_ubsan_type_info_for_type with -fsanitize=undefined)

PR sanitizer/70875
* gcc.dg/ubsan/bounds-3.c: Add -fno-sanitize-recover=bounds to
dg-options and dg-shouldfail "ubsan" directive.

From-SVN: r258117

re PR c++/84582 (Rejected valid C++ code since r257961)

PR c++/84582
* semantics.c (force_paren_expr): Create a PAREN_EXPR when in
a template.
(maybe_undo_parenthesized_ref): Unwrap PAREN_EXPR.
* typeck2.c (store_init_value): Call fold_non_dependent_expr instead
of instantiate_non_dependent_expr.
* tree.c (lvalue_kind): Handle PAREN_EXPR like NON_DEPENDENT_EXPR.

* g++.dg/cpp1y/auto-fn15.C: Extend testing.
* g++.dg/cpp1z/static1.C: New test.
* g++.dg/template/static37.C: New test.

Co-Authored-By: Jason Merrill <jason@redhat.com>
From-SVN: r258116

[PR c++/84434] ICE with deduction guide

https://gcc.gnu.org/ml/gcc-patches/2018-03/msg00063.html
PR c++/84434
* name-lookup.c (member_vec_dedup): Remove manually peeled
iteration.  Ignore dependent ctor inheritance.

PR c++/84434
* g++.dg/template/pr84434.C: New.

From-SVN: r258114

compiler: avoid crashing on bad self-referential type

    The compiler was crashing partway through emitting an error for a bad
    self-referential struct type (which refers to one of its own fields
    via an unsafe.Offset expression). Tweak the offset logic to check for
    this class of type error and exit cleanly.

    Fixes golang/go#24180

    Reviewed-on: https://go-review.googlesource.com/97855

From-SVN: r258112

PR c++/71569 - decltype of template.

* parser.c (cp_parser_decltype_expr): Handle missing template args.

From-SVN: r258110

re PR c++/84639 (gcc/c-family/c-attribs.c:1822:27: runtime error: shift exponent -1 is negative)

PR c++/84639
* c-attribs.c (common_handle_aligned_attribute): Don't use invalid
alignment in computation.

From-SVN: r258109

[ARM] Fix copysign_softfloat_1.c option directives

gcc.target/arm/copysign_softfloat_1.c's use of arm_arch_v6t2 in
dg-add-option changes the architecture to -march=armv6t2. Since the test
only requires Thumb-2 capable architecture, we just need to add -mthumb
on the command line since arm_thumb2_ok guarantees by definition that
doing that is enough to select Thumb-2. This fixes warning on the
command line when having -mcpu=cortex-m3 in RUNTESTFLAGS for instance.

2018-03-01  Thomas Preud'homme  <thomas.preudhomme@arm.com

    gcc/testsuite/
    * gcc.target/arm/copysign_softfloat_1.c: Remove dg-add-options and add
    -mthumb to the command-line.

From-SVN: r258108

re PR c++/84596 (internal compiler error: unexpected expression '(bool)c' of kind implicit_conv_expr (cxx_eval_constant_expression))

PR c++/84596
* constexpr.c (require_rvalue_constant_expression): New function.
* cp-tree.h: Declare it.
* semantics.c (finish_static_assert): Use it instead of
require_potential_rvalue_constant_expression.

* g++.dg/cpp0x/static_assert14.C: New test.

From-SVN: r258107

re PR c++/79410 (internal compiler error: in gimplify_init_ctor_preeval, at gimplify.c:3489)

2018-03-01  Paolo Carlini  <paolo.carlini@oracle.com>

PR c++/79410
* g++.dg/torture/pr79410.C: New.

From-SVN: r258106

PR c++/71569 - ICE with redundant args on member variable template.

* decl.c (start_decl): Handle partial specialization of member
variable template.
* pt.c (determine_specialization): Allow partial specialization
of member variable template without specializing enclosing class.
(process_partial_specialization): Improve error message.

Co-Authored-By: Alexandre Oliva <aoliva@redhat.com>
From-SVN: r258102

Do not handled volatile arguments (PR sanitizer/82484).

2018-03-01  Martin Liska  <mliska@suse.cz>

PR sanitizer/82484
* sanopt.c (sanitize_rewrite_addressable_params): Do not handle
volatile arguments.
2018-03-01  Martin Liska  <mliska@suse.cz>

PR sanitizer/82484
* gcc.dg/asan/pr82484.c: New test.

From-SVN: r258101

re PR debug/84645 (-flto -g0 at compile-time vs. -flto -g at link time ICEs in add_dwarf_attr, at dwarf2out.c:4353)

2018-03-01  Richard Biener  <rguenther@suse.de>

PR debug/84645
* dwarf2out.c (gen_variable_die): Properly handle late VLA
type annotation with LTO when debug was disabled at compile-time.

* gfortran.dg/lto/pr84645_0.f90: New testcase.

From-SVN: r258100

MIPS: Fix wrong use of XINT instead of INTVAL

This issue was caught with assert checking enabled but is not a
functional bug as XINT(x, 0) happens to overlay INTVAL(x) anyway.

gcc/
* config/mips/mips.c (mips_final_prescan_insn): Fix incorrect
XINT with INTVAL.
(mips_final_postscan_insn): Likewise.

From-SVN: r258099

re PR fortran/84219 (Failure to generate error for IO of transfer intrinsic, when MOLD has derived type components.)

2018-03-01  Paul Thomas  <pault@gcc.gnu.org>

PR fortran/84219
* target-memory.c (gfc_interpret_derived): Assert that BT_VOID
components are caf tokens.
(gfc_target_interpret_expr): Treat BT_VOID expressions as
integers.

2018-03-01  Paul Thomas  <pault@gcc.gnu.org>

PR fortran/84219
* gfortran.dg/coarray_47.f90: New test.

From-SVN: r258098

re PR fortran/84538 (Array of derived type elements incorrectly accessed in function)

2018-03-01  Paul Thomas  <pault@gcc.gnu.org>

PR fortran/84538
* class.c (class_array_ref_detected): Remove the condition that
there be no reference after the array reference.
(find_intrinsic_vtab): Remove excess whitespace.
* trans-array.c (gfc_conv_scalarized_array_ref): Rename 'tmp'
as 'base and call build_class_array_ref earlier.

2018-03-01  Paul Thomas  <pault@gcc.gnu.org>

PR fortran/84538
* gfortran.dg/class_array_23.f03: New test.

From-SVN: r258097

Fix PR number in changelog entry

From-SVN: r258096

re PR target/84528 (gcc.c-torture/execute/960419-2.c -O3 fails with -fno-omit-frame-pointer)

2018-03-01  Richard Sandiford  <richard.sandiford@linaro.org>

gcc/testsuite/
PR rtl-optimization/84528
* gcc.dg/torture/pr84538.c: Rename to...
* gcc.dg/torture/pr84528.c: ...this.

From-SVN: r258095

Tighten use of HARD_FRAME_POINTER_REGNUM in alias.c (PR 84538)

RTL code needs to be consistent about whether it uses the stack
pointer, the frame pointer or the argument pointer to access a
given part of the frame.  alias.c used this to divide accesses
into three independent areas.

The problem in the PR is that we did this for HARD_FRAME_POINTER_REGNUM
even when the register wasn't being used as a frame pointer.  We can't
do that because the frame pointer is then just any old allocatable
register and could certainly point to info accessed through the
argument pointer or stack pointer.

2018-03-01  Richard Sandiford  <richard.sandiford@linaro.org>

gcc/
PR rtl-optimization/84538
* alias.c (init_alias_target): Add commentary.
(init_alias_analysis): Only give HARD_FRAME_POINTER_REGNUM
a unique base value if the frame pointer is not eliminated
to the stack pointer.

gcc/testsuite/
PR rtl-optimization/84538
* gcc.dg/torture/pr84538.c: New test.

From-SVN: r258094

Fix liveness analysis in lra for spilled-into hard regs

2018-03-01  Tom de Vries  <tom@codesourcery.com>

PR rtl-optimization/83327
* lra-int.h (hard_regs_spilled_into): Declare.
* lra.c (hard_regs_spilled_into): Define.
(init_reg_info): Init hard_regs_spilled_into.
* lra-spills.c (assign_spill_hard_regs): Update hard_regs_spilled_into.
* lra-lives.c (make_hard_regno_born, make_hard_regno_dead)
(process_bb_lives): Handle hard_regs_spilled_into.
(lra_create_live_ranges_1): Before doing liveness propagation, clear
regs in all_hard_regs_bitmap if set in hard_regs_spilled_into.

From-SVN: r258093

Daily bump.

From-SVN: r258092

Warray-bounds-2.c: Declare helper static to avoid -fpic test failures.

gcc/testsuite/ChangeLog:

* c-c++-common/Warray-bounds-2.c: Declare helper static to avoid
-fpic test failures.
* c-c++-common/Wrestrict-2.c: Same.
* c-c++-common/Wstringop-truncation.c: Same.
* gcc.dg/Warray-bounds-22.c: Same.
* gcc.dg/Wstringop-overflow.c: Same.
* gcc.dg/attr-alloc_size-3.c (unsigned_range): Same.
(signed_range): Same.
(unsigned_anti_range): Same.
* gcc.dg/attr-alloc_size-4.c (unsigned_range): Same.
(signed_range): Same.
(unsigned_anti_range): Same.
* gcc.dg/attr-alloc_size-7.c: Same.
* gcc.dg/attr-alloc_size-8.c: Same.
* gcc.dg/builtin-alloc-size.c: Same.
* gcc.dg/builtin-stpncpy.c: Same.
* gcc.dg/builtins-nonnull.c: Same.
* gcc.dg/nonnull-4.c (g16_1_3_5_7_11_13): Same.
* gcc.dg/pr79214.c: Same.
* gcc.dg/tree-ssa/builtin-snprintf-warn-1.c: Same.
* gcc.dg/tree-ssa/builtin-snprintf-warn-2.c: Same.
* gcc.dg/tree-ssa/builtin-sprintf-5.c: Same.
* gcc.dg/tree-ssa/builtin-sprintf-warn-11.c: Same.
* gcc.dg/tree-ssa/builtin-sprintf-warn-12.c: Same.
* gcc.dg/tree-ssa/builtin-sprintf-warn-13.c: Same.
* gcc.dg/tree-ssa/builtin-sprintf-warn-14.c: Same.
* gcc.dg/tree-ssa/builtin-sprintf-warn-15.c: Same.
* gcc.dg/tree-ssa/builtin-sprintf-warn-19.c: Same.
* gcc.dg/tree-ssa/builtin-sprintf-warn-3.c: Same.
* gcc.dg/tree-ssa/builtin-sprintf-warn-9.c: Same.

From-SVN: r258089

PR c++/71784 - ICE with ref-qualifier and explicit specialization.

* pt.c (determine_specialization): Check ref-qualifier.

From-SVN: r258085

config.gcc (powerpc-ibm-aix7.1.*): New stanza.

* config.gcc (powerpc-ibm-aix7.1.*): New stanza.
(powerpc-ibm-aix[789]*): Default to AIX 7.2.
* config/rs6000/aix71.h (TARGET_DEFAULT): Revert to Power4 ISA.
* config/rs6000/aix72.h: New file.

From-SVN: r258082

re PR target/84014 (ICE in setup_min_max_allocno_live_range_point, at ira-build.c:2762)

        PR target/84014
        * gcc.target/powerpc/pr84014.c: Use ilp32, not -m32 option.

From-SVN: r258081

re PR c++/84609 (internal compiler error: in cp_parser_abort_tentative_parse, at cp/parser.c:28960 (cp_parser_member_declaration()))

PR c++/84609
* parser.c (cp_parser_attributes_opt): Formatting fix.
(cp_parser_skip_balanced_tokens, cp_parser_skip_gnu_attributes_opt,
cp_parser_skip_std_attribute_spec_seq, cp_parser_skip_attributes_opt):
New functions.
(cp_parser_member_declaration): Use cp_parser_skip_attributes_opt
instead of tentative parse to peek over optional attribute tokens
to check for CPP_COLON after them.

* g++.dg/cpp0x/pr84609.C: New test.

From-SVN: r258080

re PR c++/83871 (wrong code for attribute const and pure on distinct template specializations)

PR c++/83871
PR c++/83503
* pt.c (INCLUDE_STRING): Remove define.
(warn_spec_missing_attributes): Use pretty_printer instead of
std::string.  Fix up inform call so that the list of attributes
is in %s argument.

From-SVN: r258079

gimple-ssa-warn-restrict.c (maybe_diag_overlap): Use warning_n instead of warning_at with conditional singular and plural...

* gimple-ssa-warn-restrict.c (maybe_diag_overlap): Use warning_n
instead of warning_at with conditional singular and plural messages
where possible.

From-SVN: r258078

PR testsuite/84617 - new test cases g++.dg/ext/attr-const.C and g++.dg/ext/attr-pure.C fail

gcc/cp/ChangeLog:

* decl.c (duplicate_decls): Fully merge attributes const, pure,
and malloc.

gcc/testsuite/ChangeLog:

* g++.dg/ext/attr-malloc-3.C: New test.
* g++.dg/ext/attr-const.C: Adjust.  Xfail assertions failing due
to pre-existing problems.
* g++.dg/ext/attr-pure.C: Same.

From-SVN: r258077

re PR fortran/83901 (ICE in fold_convert_loc, at fold-const.c:2402)

2018-02-28  Paul Thomas  <pault@gcc.gnu.org>

PR fortran/83901
* trans-stmt.c (trans_associate_var): Make sure that the se
expression is a pointer type before converting it to the symbol
backend_decl type.

2018-02-28  Paul Thomas  <pault@gcc.gnu.org>

PR fortran/83901
* gfortran.dg/associate_37.f90: New test.

PR fortran/83344
* gfortran.dg/associate_36.f90: Add Steve Kargl as contributer.

From-SVN: r258076

re PR target/52991 (attribute packed broken on mingw32?)

PR target/52991
* stor-layout.c (update_alignment_for_field): For
targetm.ms_bitfield_layout_p (rli->t), if !is_bitfield
&& !DECL_PACKED (field), do the alignment update, just use
only desired_align instead of MAX (type_align, desired_align)
as the alignment.
(place_field): Don't do known_align < desired_align handling
early if targetm.ms_bitfield_layout_p (rli->t) and rli->prev_field
is non-NULL, instead do it after rli->prev_field handling and
only if not within a bitfield word.  For DECL_PACKED (field)
use type_align of BITS_PER_UNIT.

* gcc.dg/bf-ms-layout.c: Revert 2012-04-26 changes.
* gcc.dg/bf-ms-layout-2.c: Revert 2012-02-23 changes.
* gcc.dg/bf-ms-layout-4.c: New test.
* gcc.dg/bf-ms-layout-5.c: New test.

From-SVN: r258075

aarch64.c (aarch64_emit_probe_stack_range): Remove superfluous parentheses and trailing spaces.

* config/aarch64/aarch64.c (aarch64_emit_probe_stack_range): Remove
superfluous parentheses and trailing spaces.

From-SVN: r258074

Remove config/i386/i386.opt from revision 258001 ChangeLog

Revision 258001 doesn't change config/i386/i386.opt.

From-SVN: r258073

re PR target/83399 (Power8 ICE During LRA with 2-op rtl pattern for lvx instruction)

PR target/83399
* gcc.target/powerpc/pr83399.c: Only run on Linux.

From-SVN: r258071

re PR tree-optimization/84584 ([graphite] ICE: Segmentation fault (in dominated_by_p))

2018-02-28  Richard Biener  <rguenther@suse.de>

PR tree-optimization/84584
* graphite-scop-detection.c (scop_detection::add_scop): Discard
SCoPs with fake exit edge.

* gcc.dg/graphite/pr84584.c: New testcase.

From-SVN: r258070

PR preprocessor/84517 allow double-underscore macros after string literals

gcc/testsuite:

PR preprocessor/84517
* g++.dg/cpp0x/udlit-macros.C: Expect a warning for ""__FILE__.

libcpp:

PR preprocessor/84517
* lex.c (is_macro_not_literal_suffix): New function.
(lex_raw_string, lex_string): Use is_macro_not_literal_suffix to
decide when to issue -Wliteral-suffix warnings.

From-SVN: r258069

i-cexten.ads (Float_128): New type.

ada/
* libgnat/i-cexten.ads (Float_128): New type.
c-family/
* c-ada-spec.c (dump_ada_node) <NULLPTR_TYPE>: New case.
<REAL_TYPE>: Deal specifically with _Float128/__float128.

From-SVN: r258068

c-ada-spec.c (dump_ada_double_name): New case.

* c-ada-spec.c (dump_ada_double_name) <ENUMERAL_TYPE>: New case.
(is_char_array): Take a type instead of a declaration.
(dump_ada_array_type): Likewise.
(is_simple_enum): Minor tweak.
(dump_ada_enum_type): New function extracted from...
(dump_ada_node) <ENUMERAL_TYPE>: ...here.  Invoke it.
<INTEGER_TYPE>: Remove unreachable code.
<RECORD_TYPE>: Likewise.  Minor tweaks.
(dump_nested_type) <ARRAY_TYPE>: Adjust to above changes.
<ENUMERAL_TYPE>: New case.
<RECORD_TYPE>: Factor out common code.
(dump_ada_declaration) <ARRAY_TYPE>: Adjust to above changes.
Minor tweaks.  Deal with enumeral types.
(dump_ada_structure): Minor tweaks.

From-SVN: r258067

c-ada-spec.c (dump_ada_node): Do not use generic address for incomplete structures.

* c-ada-spec.c (dump_ada_node) <POINTER_TYPE>: Do not use generic
address for incomplete structures.
(dump_forward_type): Do not bail out for incomplete structures.
(dump_ada_declaration): Do not special-case incomplete structures
for subtypes.  Dump them as null records for types.

From-SVN: r258066

c-ada-spec.c (dump_ada_import): Use boolean and fix formatting.

* c-ada-spec.c (dump_ada_import): Use boolean and fix formatting.
(is_char_array): Fix formatting.
(dump_template_types): Likewise.
(dump_generic_ada_node): Rename into...
(dump_ada_node): ...this.
<POINTER_TYPE>: Remove superfluous space.  Use generic address for
incomplete structures and not for empty structures.  Do not use it
when forward declarations are needed.
(dump_forward_type): New function.
(dump_nested_types): Remove FORWARD parameter.  Do not consider
TREE_VISITED and do not generate a forward declaration.  Only dump
original nested types for nested declaration.
(dump_nested_type) <POINTER_TYPE>: Call dump_forward_type.
<ARRAY_TYPE>: Likewise if the component type is an anonymous pointer.
<RECORD_TYPE>: Do not consider TREE_VISITED.
(dump_ada_declaration): Use booleans and fix formatting throughout.
<TYPE_DECL>: Skip incomplete structures and not empty structures.
Call dump_forward_type instead of dump_nested_types for a typedef.
Remove superfluous check and adjust call to dump_nested_types.
<POINTER_TYPE>: Call dump_forward_type and fall through.
(dump_ada_struct_decl): Rename into...
(dump_ada_structure): ...this.  Do not special-case empty structures.

From-SVN: r258064

Fix timevar format and fix tests (PR testsuite/84597).

2018-02-28  Martin Liska  <mliska@suse.cz>

PR testsuite/84597
* timevar.c (timer::print): Fix format to properly print 100%
values.
2018-02-28  Martin Liska  <mliska@suse.cz>

PR testsuite/84597
* g++.dg/ext/timevar1.C: Fix pruned patterns.
* g++.dg/ext/timevar2.C: Likewise.

From-SVN: r258063

c-ada-spec.c (dump_ada_import): Use boolean and fix formatting.

* c-ada-spec.c (dump_ada_import): Use boolean and fix formatting.
(is_char_array): Fix formatting.
(dump_template_types): Likewise.
(dump_generic_ada_node): Rename into...
(dump_ada_node): ...this.
<POINTER_TYPE>: Remove superfluous space.  Use generic address for
incomplete structures and not for empty structures.  Do not use it
when forward declarations are needed.
(dump_forward_type): New function.
(dump_nested_types): Remove FORWARD parameter.  Do not consider
TREE_VISITED and do not generate a forward declaration.  Only dump
original nested types for nested declaration.
(dump_nested_type) <POINTER_TYPE>: Call dump_forward_type.
<ARRAY_TYPE>: Likewise if the component type is an anonymous pointer.
<RECORD_TYPE>: Do not consider TREE_VISITED.
(dump_ada_declaration): Use booleans and fix formatting throughout.
<TYPE_DECL>: Skip incomplete structures and not empty structures.
Call dump_forward_type instead of dump_nested_types for a typedef.
Remove superfluous check and adjust call to dump_nested_types.
<POINTER_TYPE>: Call dump_forward_type and fall through.
(dump_ada_struct_decl): Rename into...
(dump_ada_structure): ...this.  Do not special-case empty structures.

From-SVN: r258062

re PR c/84607 (Side effects discarded in address computation inside 'if')

2018-02-28  Richard Biener  <rguenther@suse.de>

PR middle-end/84607
* genmatch.c (capture_info::walk_match): Do not mark
captured expressions without operands as expr_p given
they act more like predicates and should be subject to
"lost tail" side-effect preserving.

* gcc.dg/pr84607.c: New testcase.

From-SVN: r258061

[PR c++/84602] ICE with anon-struct

https://gcc.gnu.org/ml/gcc-patches/2018-02/msg01577.html
PR c++/84602
* name-lookup.h (search_anon_aggr): Add defaulted WANT_TYPE arg.
* name-lookup.c (fields_linear_search): Look in an anon-aggr
regardless of want_type.
(search_anon_aggr): Just use get_class_binding_direct.

PR c++/84602
* g++.dg/lookup/pr84602.C: New.

From-SVN: r258060

re PR c++/83871 (wrong code for attribute const and pure on distinct template specializations)

PR c++/83871
PR c++/83503
* g++.dg/ext/attr-warning.C: Remove -fdump-tree-optimized from
dg-options.
* g++.dg/ext/attr-nonnull.C: Likewise.
* g++.dg/ext/attr-noinline.C: Fix syntax in scan-tree-dump-not directives.
* g++.dg/ext/attr-noinline-2.C: Likewise.
* g++.dg/ext/attr-noreturn-2.C: Use -fdump-tree-optimized instead of
-fdump-tree-eh in dg-options.

From-SVN: r258059

decl.c (cp_finish_decomp): Don't adjust eltscnt when calling inform_n.

* decl.c (cp_finish_decomp): Don't adjust eltscnt when calling
inform_n.

From-SVN: r258058

re PR debug/83917 (with -mcall-ms2sysv-xlogues, stepping into x86 tail-call restore stub gives bad backtrace)

PR debug/83917
* configure.ac (AS_HIDDEN_DIRECTIVE): AC_DEFINE_UNQUOTED this to
$asm_hidden_op if visibility ("hidden") attribute works.
(HAVE_AS_CFI_SECTIONS): New AC_DEFINE.
* config/i386/i386-asm.h: Don't include auto-host.h.
(PACKAGE_VERSION, PACKAGE_NAME, PACKAGE_STRING, PACKAGE_TARNAME,
PACKAGE_URL): Don't undefine.
(USE_GAS_CFI_DIRECTIVES): Don't use nor define this macro, instead
guard cfi_startproc only on ifdef __GCC_HAVE_DWARF2_CFI_ASM.
(FN_HIDDEN): Change guard from #ifdef HAVE_GAS_HIDDEN to
#ifdef AS_HIDDEN_DIRECTIVE, use AS_HIDDEN_DIRECTIVE macro in the
definition instead of hardcoded .hidden.
* config/i386/cygwin.S: Include i386-asm.h first before .cfi_sections
directive.  Use #ifdef HAVE_AS_CFI_SECTIONS rather than
#ifdef HAVE_GAS_CFI_SECTIONS_DIRECTIVE to guard .cfi_sections.
(USE_GAS_CFI_DIRECTIVES): Don't define.
* configure: Regenerated.
* config.in: Likewise.

From-SVN: r258057

gcc testsuite changes for new linker messages

GNU ld error messages have changed to comply with the GNU coding
standards.

* lib/prune.exp (prune_gcc_output): Match lower case "in function"
GNU ld message.
* g++.dg/other/anon5.C: Match lower case "bad value" GNU ld message.

From-SVN: r258054

[PR81611] turn inc-and-use-of-dead-orig into auto-inc

When the addressing modes available on the machine don't allow offsets
in addresses, odds are that post-increments will be represented in
trees and RTL as:

  y <= x + 1
  ... *(x) ...
  x <= y

so deal with it by turning such RTL as:

  (set y (plus x n))
  ... (mem x) ...

without intervening uses of y into

  (set y x)
  ... (mem (post_add y n)) ...

so as to create auto-inc addresses that we'd otherwise miss.

for  gcc/ChangeLog

PR rtl-optimization/81611
* auto-inc-dec.c (attempt_change): Move dead note from
mem_insn if it's the next use of regno
(find_address): Take address use of reg holding
non-incremented value.  Add parm to limit search to the named
reg only.
(merge_in_block): Attempt to use a mem insn that is the next
use of the original regno.

From-SVN: r258053

runtime: update AIX memory allocation for new versions

    Reviewed-on: https://go-review.googlesource.com/97357

From-SVN: r258052

libgo: update to final Go 1.10 release

    Reviewed-on: https://go-review.googlesource.com/97517

From-SVN: r258051

Daily bump.

From-SVN: r258050

Avoid including <string> directly.

From-SVN: r258046

PR c++/83871 - wrong code for attribute const and pure on distinct template specializations

PR c++/83871 - wrong code for attribute const and pure on distinct template specializations
PR c++/83503 - [8 Regression] bogus -Wattributes for const and pure on function template specialization

gcc/ChangeLog:

PR c++/83871
* gcc/doc/invoke.texi (-Wmissing-attributes): New option.
* gcc/print-tree.c (print_node): Handle DECL_UNINLINABLE.

gcc/c-family/ChangeLog:

PR c++/83871
* c.opt (-Wmissing-attributes): New option.

gcc/cp/ChangeLog:

PR c++/83871
PR c++/83503
* cp-tree.h (warn_spec_missing_attributes): New function.
((check_explicit_specialization): Add an argument.  Call the above
function.
* decl.c (duplicate_decls): Avoid applying primary function template's
attributes to its explicit specializations.
cp/pt.c (warn_spec_missing_attributes): Define.

gcc/testsuite/ChangeLog:

PR c++/83871
PR c++/83503
* g++.dg/Wmissing-attributes.C: New test.
* g++.dg/ext/attr-const-pure.C: New test.
* g++.dg/ext/attr-const.C: New test.
* g++.dg/ext/attr-deprecated-2.C: New test.
* g++.dg/ext/attr-malloc-2.C: New test.
* g++.dg/ext/attr-malloc.C: New test.
* g++.dg/ext/attr-noinline-2.C: New test.
* g++.dg/ext/attr-noinline.C: New test.
* g++.dg/ext/attr-nonnull.C: New test.
* g++.dg/ext/attr-noreturn-2.C: New test.
* g++.dg/ext/attr-noreturn.C: New test.
* g++.dg/ext/attr-nothrow-2.C: New test.
* g++.dg/ext/attr-nothrow.C: New test.
* g++.dg/ext/attr-optimize.C: New test.
* g++.dg/ext/attr-pure.C: New test.
* g++.dg/ext/attr-returns-nonnull.C: New test.
* g++.dg/ext/attr-warning.C: New test.

From-SVN: r258045

PR translation/84207 - Hard coded plural in gimple-fold.c

gcc/ChangeLog:

PR translation/84207
* diagnostic-core.h (warning_n, error_n, inform_n): Change
n argument to unsigned HOST_WIDE_INT.
* diagnostic.c (warning_n, error_n, inform_n): Ditto.
(diagnostic_n_impl): Ditto.  Handle arguments in excess of LONG_MAX.
* gimple-ssa-sprintf.c (format_directive): Simplify inform_n call.
* tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Use warning_n.

From-SVN: r258044

PR c++/71546 - lambda init-capture with qualified-id.

* parser.c (cp_parser_lambda_introducer): Clear scope after
each lambda capture.

From-SVN: r258043

[PR c++/84426] ICE after conflicting member decl

https://gcc.gnu.org/ml/gcc-patches/2018-02/msg01529.html
PR c++/84426
* name-lookup.h (get_member_slot): Rename ...
(find_member_slot): ... here.
(add_member_slot): New.
* name-lookup.c (member_vec_linear_search): No need to check for
NULL slot.
(get_member_slot): Rename ...
(find_member_slot): ... here.  Don't add slot for incomplete class.
(add_member_slot): New.
* class.c (add_method): Adjust get_member_slot rename.  Bail out
if push_class_level_binding fails.  Create slot and grok
properties once we're committed to insertion.

PR c++/84426
* g++.dg/lookup/pr84426.C: New.

From-SVN: r258042

* cp-tree.h: Adjust comment.

From-SVN: r258040

PR c++/84489 - dependent default template argument

* pt.c (type_unification_real): Handle early substitution failure.

From-SVN: r258039

re PR tree-optimization/84512 (Missed optimization: should be precalculated in compile-time)

2018-02-27  Richard Biener  <rguenther@suse.de>

PR tree-optimization/84512
* tree-vect-loop.c (vect_compute_single_scalar_iteration_cost):
Do not use the estimate returned from record_stmt_cost for
the scalar iteration cost but sum properly using add_stmt_cost.

* gcc.dg/tree-ssa/pr84512.c: New testcase.

From-SVN: r258036

re PR libgomp/84466 (libgomp.graphite/force-parallel-8.c fails starting with r257723)

2018-02-27  Richard Biener  <rguenther@suse.de>

PR tree-optimization/84466
* graphite-scop-detection.c (scop_detection::stmt_simple_for_scop_p):
Adjust last change to less strictly validate use operands.

From-SVN: r258035

Make groups more generic (PR gcov-profile/84548).

2018-02-27  Martin Liska  <mliska@suse.cz>

PR gcov-profile/84548
* gcov.c (process_file): Allow partial overlap and consider it
also as group functions.
(output_lines): Properly calculate range of lines for a group.
2018-02-27  Martin Liska  <mliska@suse.cz>

PR gcov-profile/84548
* g++.dg/gcov/pr84548.C: New test.

From-SVN: r258033

Implement the missing bits of LWG 2769

* include/std/any (any_cast(const any&)): Add static_assert.
(any_cast(any&)): Likewise.
(any_cast(any&&)): Likewise, and remove the handling
for copyable-but-not-movable type.
* testsuite/20_util/any/misc/any_cast.cc: Adjust.
* testsuite/20_util/any/misc/any_cast_neg.cc: Likewise, and
add new tests.

From-SVN: r258031

re PR target/84575 (gcc.target/i386/pr84309.c fail)

PR target/84575
* gcc.target/i386/pr84309.c: Add -mno-avx2 to dg-options.

From-SVN: r258030

Fix formatting of -ftime-report.

2018-02-27  Martin Liska  <mliska@suse.cz>

* timevar.c (timer::print_row): Remove 'usr', 'sys', 'wall' and
'ggc' suffixes.  Change first column width.
(timer::print): Fix formatting of the column.

From-SVN: r258029

[IEPM] don't preserve lexical blocks just for debug inline markers

This patch stops preserving scope blocks just because they are inlined
function scopes, when cleaning up unused scope blocks.  This change
was introduced along with IEPM, but it preserved lots of blocks, and
output debug information for them, although no code from the inlined
function remained after optimization.

The additional preserved blocks took up compile-time memory, and
significant disk space and link time, in some cases more than 25%.
This is deemed excessive, compared with the reasonably small benefit
of allowing one to single-step into an inlined function using a
view-capable debugger.

There was another way of marking inlined function scopes as unused,
based on the markers referencing them during stmt scanning, but that
still preserved too much.

So, this patch restores the pre-IEPM logic of preservation of scopes.
Should a scope block referenced by an inline entry marker be found to
be unused in remove_unused_scope_block_p, the marker will be cleaned
up right after that, in clear_unused_block_pointer, so we won't keep
a dangling reference to a dropped block.

for  gcc/ChangeLog

* tree-ssa-live.c (remove_unused_scope_block_p): Do not
preserve inline entry blocks for the sake of debug inline
entry point markers alone.
(remove_unused_locals): Suggest in comments a better place to
force the preservation of inline entry blocks that are
otherwise unused, but do not preserve them.

From-SVN: r258026

PR c++/84560 - ICE capturing multi-dimensional VLA.

* tree.c (array_of_runtime_bound_p): False if the element is
variably-modified.

From-SVN: r258023

PR c++/84441 - ICE with base initialized from ?:

* call.c (unsafe_copy_elision_p): Handle COND_EXPR.

From-SVN: r258022

PR c++/84520 - ICE with generic lambda in NSDMI.

* lambda.c (lambda_expr_this_capture): Don't look for fake NSDMI
'this' in a generic lambda instantiation.

From-SVN: r258021

Daily bump.

From-SVN: r258020

* es.po, sv.po: Update.

From-SVN: r258016

PR c++/84559 - ICE with constexpr VLA.

* constexpr.c (ensure_literal_type_for_constexpr_object): Check
for constexpr variable with VLA type.

From-SVN: r258015

re PR c++/84558 (ICE with invalid constexpr constructor)

PR c++/84558
* constexpr.c (cxx_eval_vec_init_1): For reuse, treat NULL eltinit like
a valid constant initializer.  Formatting fixes.

* g++.dg/cpp1y/pr84558.C: New test.

From-SVN: r258014

re PR c++/84540 (ICE with alignas in variadic template)

/cp
2018-02-26  Paolo Carlini  <paolo.carlini@oracle.com>

PR c++/84540
* pt.c (tsubst_attributes): Handle correctly tsubst_attribute
returning NULL_TREE.
(apply_late_template_attributes): Likewise.

/testsuite
2018-02-26  Paolo Carlini  <paolo.carlini@oracle.com>

PR c++/84540
* g++.dg/cpp0x/alignas14.C: New.
* g++.dg/cpp0x/alignas15.C: Likewise.

From-SVN: r258012

re PR c++/84557 (ICE with invalid firstprivate variable)

PR c++/84557
* parser.c (cp_parser_omp_var_list_no_open): Only call
cp_parser_lookup_name_simple on names satisfying identifier_p.
(cp_parser_oacc_routine): Likewise.

* g++.dg/gomp/pr84557.C: New test.

From-SVN: r258011

re PR debug/83917 (with -mcall-ms2sysv-xlogues, stepping into x86 tail-call restore stub gives bad backtrace)

PR debug/83917
* config/i386/i386-asm.h (PACKAGE_VERSION, PACKAGE_NAME,
PACKAGE_STRING, PACKAGE_TARNAME, PACKAGE_URL): Undefine between
inclusion of auto-target.h and auto-host.h.
(USE_GAS_CFI_DIRECTIVES): Define if not defined already based on
__GCC_HAVE_DWARF2_CFI_ASM.
(cfi_startproc, cfi_endproc, cfi_adjust_cfa_offset,
cfi_def_cfa_register, cfi_def_cfa, cfi_register, cfi_offset, cfi_push,
cfi_pop): Define.
* config/i386/cygwin.S: Don't include auto-host.h here, just
define USE_GAS_CFI_DIRECTIVES to 1 or 0 and include i386-asm.h.
(cfi_startproc, cfi_endproc, cfi_adjust_cfa_offset,
cfi_def_cfa_register, cfi_register, cfi_push, cfi_pop): Remove.
* config/i386/resms64fx.h: Add cfi_* directives.
* config/i386/resms64x.h: Likewise.

From-SVN: r258010

PR c++/84551 - ICE with concepts and -g.

* parser.c (add_debug_begin_stmt): Do nothing in a concept.

From-SVN: r258009

re PR c++/84325 (internal compiler error, in cxx_eval_constant_expression gcc/cp/constexpr.c:4740)

PR c++/84325
* tree.c (replace_placeholders_r): Only check TREE_CONSTANT on
non-types.

* g++.dg/cpp1z/pr84325.C: New test.

From-SVN: r258008

builtins-3.c: Move vec_neg builtin tests to Power 8 test file.

gcc/testsuite/ChangeLog:

2018-02-26  Carl Love  <cel@us.ibm.com>
* gcc.target/powerpc/builtins-3.c: Move vec_neg builtin tests to
Power 8 test file.
* gcc.target/powerpc/builtins-3-p8.c: Add vec_neg builtin tests.
* gcc.target/powerpc/fold-vec-neg-char.c(dg-options): Add -mcpu=power8.
* gcc.target/powerpc/fold-vec-neg-floatdouble.c(dg-options): Add
-mcpu=power8.
* gcc.target/powerpc/fold-vec-neg-int.c(dg-options): Remove file.
* gcc.target/powerpc/fold-vec-neg-short.c(dg-options): Add
-mcpu=power8.

From-SVN: r258006

i386: Update comments for ix86_output_indirect_jmp

Revision 257992 removed the bool argument from ix86_output_indirect_jmp.
Update comments to reflect it.

* config/i386/i386.c (ix86_output_indirect_jmp): Update comments.

From-SVN: r258005

PR c++/84447 - ICE with deleted inherited ctor with default arg.

* call.c (build_over_call): Handle deleted functions in one place.

From-SVN: r258003

PR c++/81589 - error with is_trivially_constructible

* g++.dg/ext/is_trivially_constructible6.C: New.

From-SVN: r258002

i386: Add TARGET_INDIRECT_BRANCH_REGISTER

For

---
struct C {
  virtual ~C();
  virtual void f();
};

void
f (C *p)
{
  p->f();
  p->f();
}
---

-mindirect-branch=thunk-extern -O2 on x86-64 GNU/Linux generates:

_Z1fP1C:
.LFB0:
        .cfi_startproc
        pushq   %rbx
        .cfi_def_cfa_offset 16
        .cfi_offset 3, -16
        movq    (%rdi), %rax
        movq    %rdi, %rbx
        jmp     .LIND1
.LIND0:
        pushq   16(%rax)
        jmp     __x86_indirect_thunk
.LIND1:
        call    .LIND0
        movq    (%rbx), %rax
        movq    %rbx, %rdi
        popq    %rbx
        .cfi_def_cfa_offset 8
        movq    16(%rax), %rax
        jmp     __x86_indirect_thunk_rax
        .cfi_endproc

x86-64 is supposed to have asynchronous unwind tables by default, but
there is nothing that reflects the change in the (relative) frame
address after .LIND0.  That region really has to be moved outside of
the .cfi_startproc/.cfi_endproc bracket.

This patch adds TARGET_INDIRECT_BRANCH_REGISTER to force indirect
branch via register whenever -mindirect-branch= is used.  Now,
-mindirect-branch=thunk-extern -O2 on x86-64 GNU/Linux generates:

_Z1fP1C:
.LFB0:
.cfi_startproc
pushq %rbx
.cfi_def_cfa_offset 16
.cfi_offset 3, -16
movq (%rdi), %rax
movq %rdi, %rbx
movq 16(%rax), %rax
call __x86_indirect_thunk_rax
movq (%rbx), %rax
movq %rbx, %rdi
popq %rbx
.cfi_def_cfa_offset 8
movq 16(%rax), %rax
jmp __x86_indirect_thunk_rax
.cfi_endproc

so that "-mindirect-branch=thunk-extern" is equivalent to
"-mindirect-branch=thunk-extern -mindirect-branch-register", which is
used by Linux kernel.

gcc/

PR target/84039
* config/i386/constraints.md (Bs): Replace
ix86_indirect_branch_register with
TARGET_INDIRECT_BRANCH_REGISTER.
(Bw): Likewise.
* config/i386/i386.md (indirect_jump): Likewise.
(tablejump): Likewise.
(*sibcall_memory): Likewise.
(*sibcall_value_memory): Likewise.
Peepholes of indirect call and jump via memory: Likewise.
(*sibcall_GOT_32): Disallowed for TARGET_INDIRECT_BRANCH_REGISTER.
(*sibcall_value_GOT_32): Likewise.
* config/i386/i386.opt: Likewise.
* config/i386/predicates.md (indirect_branch_operand): Likewise.
(GOT_memory_operand): Likewise.
(call_insn_operand): Likewise.
(sibcall_insn_operand): Likewise.
(GOT32_symbol_operand): Likewise.
* config/i386/i386.h (TARGET_INDIRECT_BRANCH_REGISTER): New.

gcc/testsuite/

PR target/84039
* gcc.target/i386/indirect-thunk-1.c: Updated.
* gcc.target/i386/indirect-thunk-2.c: Likewise.
* gcc.target/i386/indirect-thunk-3.c: Likewise.
* gcc.target/i386/indirect-thunk-4.c: Likewise.
* gcc.target/i386/indirect-thunk-5.c: Likewise.
* gcc.target/i386/indirect-thunk-6.c: Likewise.
* gcc.target/i386/indirect-thunk-7.c: Likewise.
* gcc.target/i386/indirect-thunk-attr-1.c: Likewise.
* gcc.target/i386/indirect-thunk-attr-2.c: Likewise.
* gcc.target/i386/indirect-thunk-attr-3.c: Likewise.
* gcc.target/i386/indirect-thunk-attr-4.c: Likewise.
* gcc.target/i386/indirect-thunk-attr-5.c: Likewise.
* gcc.target/i386/indirect-thunk-attr-6.c: Likewise.
* gcc.target/i386/indirect-thunk-attr-7.c: Likewise.
* gcc.target/i386/indirect-thunk-bnd-1.c: Likewise.
* gcc.target/i386/indirect-thunk-bnd-2.c: Likewise.
* gcc.target/i386/indirect-thunk-bnd-3.c: Likewise.
* gcc.target/i386/indirect-thunk-bnd-4.c: Likewise.
* gcc.target/i386/indirect-thunk-extern-1.c: Likewise.
* gcc.target/i386/indirect-thunk-extern-2.c: Likewise.
* gcc.target/i386/indirect-thunk-extern-3.c: Likewise.
* gcc.target/i386/indirect-thunk-extern-4.c: Likewise.
* gcc.target/i386/indirect-thunk-extern-5.c: Likewise.
* gcc.target/i386/indirect-thunk-extern-6.c: Likewise.
* gcc.target/i386/indirect-thunk-extern-7.c: Likewise.
* gcc.target/i386/indirect-thunk-inline-1.c: Likewise.
* gcc.target/i386/indirect-thunk-inline-2.c: Likewise.
* gcc.target/i386/indirect-thunk-inline-3.c: Likewise.
* gcc.target/i386/indirect-thunk-inline-4.c: Likewise.
* gcc.target/i386/indirect-thunk-inline-5.c: Likewise.
* gcc.target/i386/indirect-thunk-inline-6.c: Likewise.
* gcc.target/i386/indirect-thunk-inline-7.c: Likewise.
* gcc.target/i386/ret-thunk-9.c: Likewise.
* gcc.target/i386/ret-thunk-10.c: Likewise.
* gcc.target/i386/ret-thunk-11.c: Likewise.
* gcc.target/i386/ret-thunk-12.c: Likewise.
* gcc.target/i386/ret-thunk-13.c: Likewise.
* gcc.target/i386/ret-thunk-14.c: Likewise.
* gcc.target/i386/ret-thunk-15.c: Likewise.

From-SVN: r258001

Fix typo

From-SVN: r257999

Re-add test.

From-SVN: r257998

Remove bogus test.

From-SVN: r257997

re PR rtl-optimization/83496 (wrong code generated with -Os -mbranch-cost=1)

PR rtl-optimization/83496
* reorg.c (steal_delay_list_from_target): Change REDUNDANT array from
booleans to RTXes.  Call fix_reg_dead_note on every non-null element.
(steal_delay_list_from_fallthrough): Call fix_reg_dead_note on a
redundant insn, if any.
(relax_delay_slots): Likewise.
(update_reg_unused_notes): Rename REDUNDANT_INSN to OTHER_INSN.

From-SVN: r257996

Make fix for PR 83965 handle SLP reduction chains

This patch prevents pattern-matching of fold-left SLP reduction chains,
which the previous patch for 83965 didn't handle properly.  It only
stops the last statement in the group from being matched, but that's
enough to cause the group to be dissolved later.

A better fix would be to put all the information about the reduction
on the the first statement in the reduction chain, so that every
statement in the group can tell what the group is doing.  That doesn't
seem like stage 4 material though.

2018-02-26  Richard Sandiford  <richard.sandiford@linaro.org>

gcc/
PR tree-optimization/83965
* tree-vect-patterns.c (vect_reassociating_reduction_p): Assume
that grouped statements are part of a reduction chain.  Return
true if the statement is not marked as a reduction itself but
is part of a group.
(vect_recog_dot_prod_pattern): Don't check whether the statement
is part of a group here.
(vect_recog_sad_pattern): Likewise.
(vect_recog_widen_sum_pattern): Likewise.

gcc/testsuite/
PR tree-optimization/83965
* gcc.dg/vect/pr83965-2.c: New test.

From-SVN: r257995

[testsuite] Add missing function decl to regs-arg-size.c

2018-02-26  Tom de Vries  <tom@codesourcery.com>

* gcc.c-torture/compile/regs-arg-size.c (swprintf): Declare.

From-SVN: r257994

re PR debug/84545 (FAIL: g++.dg/debug/pr44182.C -gdwarf-2 -O2 (test for excess errors))

PR debug/84545
* final.c (rest_of_clean_state): Also look for calls inside sequences.

From-SVN: r257993

i386: Update -mfunction-return= for return with pop

When -mfunction-return= is used, simple_return_pop_internal should pop
return address into ECX register, adjust stack by bytes to pop from stack
and jump to the return thunk via ECX register.

Tested on i686 and x86-64.

PR target/84530
* config/i386/i386-protos.h (ix86_output_indirect_jmp): Remove
the bool argument.
(ix86_output_indirect_function_return): New prototype.
(ix86_split_simple_return_pop_internal): Likewise.
* config/i386/i386.c (indirect_return_via_cx): New.
(indirect_return_via_cx_bnd): Likewise.
(indirect_thunk_name): Handle return va CX_REG.
(output_indirect_thunk_function): Create alias for
__x86_return_thunk_[re]cx and __x86_return_thunk_[re]cx_bnd.
(ix86_output_indirect_jmp): Remove the bool argument.
(ix86_output_indirect_function_return): New function.
(ix86_split_simple_return_pop_internal): Likewise.
* config/i386/i386.md (*indirect_jump): Don't pass false
to ix86_output_indirect_jmp.
(*tablejump_1): Likewise.
(simple_return_pop_internal): Change it to define_insn_and_split.
Call ix86_split_simple_return_pop_internal to split it for
-mfunction-return=.
(simple_return_indirect_internal): Call
ix86_output_indirect_function_return instead of
ix86_output_indirect_jmp.

gcc/testsuite/

PR target/84530
* gcc.target/i386/ret-thunk-22.c: New test.
* gcc.target/i386/ret-thunk-23.c: Likewise.
* gcc.target/i386/ret-thunk-24.c: Likewise.
* gcc.target/i386/ret-thunk-25.c: Likewise.
* gcc.target/i386/ret-thunk-26.c: Likewise.

From-SVN: r257992

re PR c++/84533 (ICE with duplicate enum value)

/cp
2018-02-26  Paolo Carlini  <paolo.carlini@oracle.com>

PR c++/84533
* decl.c (redeclaration_error_message): Don't try to use
DECL_DECLARED_CONSTEXPR_P on CONST_DECLs.

/testsuite
2018-02-26  Paolo Carlini  <paolo.carlini@oracle.com>

PR c++/84533
* g++.dg/cpp1z/pr84533.C: New.

From-SVN: r257991