Introduce -fprofile-update=prefer-atomic

PR bootstrap/78069
* common.opt: Add prefer-atomic as a new enum value for
-fprofile-update.
* coretypes.h: Likewise.
* doc/invoke.texi: Document the new option value.
* gcc.c: Replace atomic with prefer-atomic.  Remove warning.
* tree-profile.c (tree_profiling): Select default value
of -fprofile-update when 'prefer-atomic' is selected.
PR bootstrap/78069
* gcc.dg/no_profile_instrument_function-attr-1.c: Update test
to match scanned pattern.
* gcc.dg/tree-ssa/ssa-lim-11.c: Likewise.

From-SVN: r242386

The second patch updates the Cortex-A57 scheduler now that we can differentiate between shifts and bitfield inserts.

The second patch updates the Cortex-A57 scheduler now that we can differentiate
between shifts and bitfield inserts.  The Cortex-A57 Software Optimization Guide
indicates that BFM operations use the integer multi-cycle pipeline, while ARM
UXTB/H instructions use the Integer 1 or Integer 0 pipelines, so swap the bfm
and extend reservations.  This results in minor scheduling differences.

* config/arm/cortex-a57.md (cortex_a57_alu): Move extend here, bfm...
(cortex_a57_alu_shift): ...here.

From-SVN: r242385

Currently the SBFM, UBFM and BFM instructions all use the attribute "bfm".

SBFM and UBFM include all shifts on AArch64, which are simpler than bitfield
insert.  Add a new bfx attribute for these instructions so that they can be
modelled more accurately in the future.  There is no difference in code
generation.

* config/aarch64/aarch64.md (aarch64_ashl_sisd_or_int_<mode>3)
Use bfx attribute.
(aarch64_lshr_sisd_or_int_<mode>3): Likewise.
(aarch64_ashr_sisd_or_int_<mode>3): Likewise.
(<optab>si3_insn_uxtw): Likewise.
(<optab><mode>3_insn): Likewise.
(<ANY_EXTEND:optab><GPI:mode>_ashl<SHORT:mode>): Likewise.
(zero_extend<GPI:mode>_lshr<SHORT:mode>): Likewise.
(extend<GPI:mode>_ashr<SHORT:mode>): Likewise.
(<optab><mode>): Likewise.
(insv<mode>): Likewise.
(andim_ashift<mode>_bfiz): Likewise.
* config/aarch64/thunderx.md (thunderx_shift): Add bfx.
* config/arm/cortex-a53.md (cortex_a53_alu_shift): Likewise.
* config/arm/cortex-a57.md (cortex_a57_alu): Add bfx.
* config/arm/exynos-m1.md (exynos_m1_alu): Add bfx.
(exynos_m1_alu_p): Likewise.
* config/arm/types.md: Add bfx.
* config/arm/xgene1.md (xgene1_bfm): Add bfx.

From-SVN: r242384

The existing vector costs stop some beneficial vectorization.

The existing vector costs stop some beneficial vectorization.  This is mostly
due to vector statement cost being set to 3 as well as vector loads having a
higher cost than scalar loads.  This means that even when we vectorize 4x, it
is possible that the cost of a vectorized loop is similar to the scalar
version, and we fail to vectorize.

Using a cost of 3 for a vector operation suggests they are 3 times as
expensive as scalar operations.  Since most vector operations have a
similar throughput as scalar operations, this is not correct.

Using slightly lower values for these heuristics now allows this loop
and many others to be vectorized.  On a proprietary benchmark the gain
from vectorizing this loop is around 15-30% which shows vectorizing it is
indeed beneficial.

* config/aarch64/aarch64.c (cortexa57_vector_cost):
Change vec_stmt_cost, vec_align_load_cost and vec_unalign_load_cost.

From-SVN: r242383

Makefile.in: Replace s-interr-hwint.adb with s-interr-vxworks.adb throughout.

* gcc-interface/Makefile.in: Replace s-interr-hwint.adb with
s-interr-vxworks.adb throughout.

From-SVN: r242382

remove conditional compilation of HAVE_AS_LEB128 code

gcc/ChangeLog:

2016-08-20  Trevor Saunders  <tbsaunde+gcc@tbsaunde.org>

* acinclude.m4 (gcc_GAS_CHECK_FEATURE): Support doing an action
if the feature isn't available.
* configure: Regenerate.
* configure.ac: define HAVE_AS_LEB128 to 0 when not available.
* dwarf2asm.c (dw2_asm_output_data_uleb128): Always compile code
for HAVE_AS_LEB128.
(dw2_asm_output_data_sleb128): Likewise.
(dw2_asm_output_delta_uleb128): Likewise.
(dw2_asm_output_delta_sleb128): Likewise.
* except.c (output_one_function_exception_table): Likewise.
(dw2_size_of_call_site_table): Likewise.
(sjlj_size_of_call_site_table): Likewise.
* dwarf2out.c (output_loc_list): Likewise.
(output_rnglists): Likewise.

From-SVN: r242381

re PR tree-optimization/78312 (wrong code due to ssa-backprop)

2016-11-14  Richard Biener  <rguenther@suse.de>

PR tree-optimization/78312
* gimple-ssa-backprop.c (backprop::prepare_change): Reset
flow-sensitive info.

* gcc.dg/torture/pr78312.c: New testcase.

From-SVN: r242380

re PR target/78093 ([avr] New variable attribute "absdata" and option "-mabsdata" to enable LDS / STS on Reduced Tiny)

gcc/
PR target/78093
* doc/invoke.texi (AVR Options) [-mabsdata]: Document new option.
* config/avr/avr.opt (-mabsdata): New option.
* config/avr/avr-arch.h (avr_device_specific_features): Add AVR_ISA_LDS.
* config/avr/avr.c (avr_encode_section_info) [AVR_TINY]: If
-mabsdata & symbol is not progmem, tag as AVR_SYMBOL_FLAG_TINY_ABSDATA.
* config/avr/avr-mcus.def (attiny4/5/9/10/20): Use AVR_ISA_LDS.
* config/avr/gen-avr-mmcu-specs.c (print_mcu): Print cc1_absdata
spec depending on AVR_ISA_LDS.
* config/avr/specs.h (CC1_SPEC): Enhanced by cc1_absdata spec.
gcc/testsuite/
PR target/78093
* gcc.target/avr/torture/tiny-absdata-2.c: New test.

From-SVN: r242379

Implement P0217R3 - C++17 structured bindings

Implement P0217R3 - C++17 structured bindings
* g++.dg/cpp1z/decomp1.C: New test.
* g++.dg/cpp1z/decomp2.C: New test.
* g++.dg/cpp1z/decomp3.C: New test.
* g++.dg/cpp1z/decomp4.C: New test.
* g++.dg/cpp1z/decomp5.C: New test.
* g++.dg/cpp1z/decomp6.C: New test.
* g++.dg/cpp1z/decomp7.C: New test.
* g++.dg/cpp1z/decomp8.C: New test.
* g++.dg/cpp1z/decomp9.C: New test.
* g++.dg/cpp1z/decomp10.C: New test.

Co-Authored-By: Jason Merrill <jason@redhat.com>
From-SVN: r242378

Implement P0217R3 - C++17 structured bindings

gcc/
* match.pd: Don't try to compare addresses of variables with
DECL_VALUE_EXPR.
gcc/cp/
* cp-tree.h (struct lang_decl_base): Add decomposition_p.
(DECL_DECOMPOSITION_P): New
(enum auto_deduction_context): Add adc_decomp_type.
(enum cp_declarator_kind): Add cdk_decomp.
* constexpr.c (cxx_eval_constant_expression): Look through
DECL_VALUE_EXPR.
(potential_constant_expression_1): Likewise.
* decl.c (reshape_init): Preserve CONSTRUCTOR_IS_DIRECT_INIT.
(check_initializer): Use build_aggr_init for DECL_DECOMPOSITION_P.
(cp_finish_decl): Pass adc_decomp_type for decomposition.
(find_decomp_class_base, get_tuple_size, get_tuple_element_type)
(get_tuple_decomp_init, cp_finish_decomp): New.
(grokdeclarator): Handle decomposition.
* init.c (build_aggr_init): Handle decomposition array.
(build_vec_init): Handle initialization from { array }.
* name-lookup.c (add_function): Always wrap TEMPLATE_DECL in
OVERLOAD.
* parser.c (declarator_can_be_parameter_pack): Handle cdk_decomp.
(function_declarator_p, strip_declarator_types)
(cp_parser_check_declarator_template_parameters): Likewise.
(cp_parser_range_for, cp_convert_range_for): Handle decomposition.
(cp_parser_simple_declaration): Parse decomposition.
(cp_parser_decomposition_declaration): New.
* pt.c (tsubst_decomp_names): New.
(subst_expr) [DECL_EXPR, RANGE_FOR_STMT]: Handle decomposition.
(do_auto_deduction): Handle adc_decomp_type.
* semantics.c (finish_decltype_type): Look through DECL_VALUE_EXPR.
* typeck.c (is_bitfield_expr_with_lowered_type): Likewise.
* tree.c (lvalue_kind): Likewise.
(cp_build_reference_type): Handle reference collapsing.

Co-Authored-By: Jason Merrill <jason@redhat.com>
From-SVN: r242377

Improve various diagnostic issues.

* call.c (build_new_method_call_1): Include template arguments in
error message.
(print_error_for_call_failure): Likewise.
(build_new_function_call): Pass them in.
* name-lookup.c (supplement_binding_1): Don't complain about a
conflict with an erroneous declaration.
* error.c (dump_decl): Fix printing of alias declaration.
* decl.c (make_typename_type): Call cxx_incomplete_type_error.
* parser.c (cp_parser_diagnose_invalid_type_name): Likewise.
* semantics.c (perform_koenig_lookup): Don't wrap an error in
TEMPLATE_ID_EXPR.

From-SVN: r242376

Delete addressof for temporaries (LWG 2598)

* include/bits/move.h (addressof(const _Tp&&)): Add deleted overload,
as per LWG 2598.

From-SVN: r242375

Make std::future::share() noexcept (LWG 2556)

* include/std/future (future::share(), future<R&>::share())
(future<void>::share()): Add noexcept, as per LWG 2556.

From-SVN: r242374

PR78326 fix incorrect access of data member in base class

PR libstdc++/78326
* include/experimental/memory_resource (memory_resource::_S_max_align):
Change access to protected.

From-SVN: r242373

Daily bump.

From-SVN: r242372

Add array support to std::shared_ptr for C++17

* doc/xml/manual/status_cxx2017.xml: Update status.
* doc/html/manual/status.html: Regenerate.
* include/bits/shared_ptr.h (shared_ptr(unique_ptr<_Yp, _Del>)): Add
extension constructor to maintain C++14 behaviour.
* include/bits/shared_ptr_base.h (__sp_array_delete): Add new struct.
(__shared_count(_Ptr, false_type), __shared_count(_Ptr, true_type)):
New constructors.
(__sp_compatible_with, __sp_is_constructible): Add specializations
for array support.
(__sp_is_constructible_arr, __sp_is_constructible_arrN): New helpers.
(__shared_ptr_access): New base class for observer member functions.
(__shared_ptr::element_type): Use remove_extent.
(__shared_ptr::_UniqCompatible): Add __sp_compatible_with check.
(__shared_ptr(_Yp*)): Use tag dispatching to call new __shared_count
constructor.
(__shared_ptr(unique_ptr<_Yp, _Del>)): Add extension constructor.
(__shared_ptr::operator*, __shared_ptr::operator->): Remove and
inherit from __shared_ptr_access base class.
(__shared_ptr::__has_esft_base): Return false for array types.
(__weak_ptr::element_type): Use remove_extent.
* include/experimental/bits/shared_ptr.h (__libfund_v1): Remove.
(__shared_ptr<__libfund_v1<_Tp>>): Remove specializations.
(__wak_ptr<__libfund_v1<_Tp>>): Likewise.
(experimental::__sp_compatible_v): Redefine using
__sp_compatible_with.
(experimental::__sp_is_constructible_v): Redefine using
__sp_is_constructible.
(get_deleter, operator<<): Change argument from __shared_ptr to
shared_ptr.
* testsuite/20_util/shared_ptr/cons/array.cc: New test.
* testsuite/20_util/shared_ptr/cons/unique_ptr_array.cc: Adjust for
new behaviour.
* testsuite/20_util/shared_ptr/observers/array.cc: Test observers for
arrays.
* testsuite/20_util/shared_ptr/observers/array_neg.cc: New test.

From-SVN: r242369

pr78268.C: New test.

gcc/testsuite/ChangeLog:

2016-11-13  Kugan Vivekanandarajah  <kuganv@linaro.org>

* g++.dg/torture/pr78268.C: New test.

gcc/ChangeLog:

2016-11-13  Kugan Vivekanandarajah  <kuganv@linaro.org>

* ipa-cp.c (ipa_get_jf_pass_through_result): Skip unary expressions.
(propagate_vr_accross_jump_function): Handle unary expressions.
* ipa-prop.c (ipa_set_jf_unary_pass_through): New.
(load_from_param_1): New.
(load_from_unmodified_param): Factor common part into load_from_param_1.
(load_from_param): New.
(compute_complex_assign_jump_func): Handle unary expressions.
(update_jump_functions_after_inlining): Likewise.
(ipa_write_jump_function): Likewise.
(ipa_read_jump_function): Likewise.

From-SVN: r242368

Implement P0403R1, Literal suffixes for basic_string_view.

* include/std/string_view
(operator""sv(const char*, size_t)): New.
(operator""sv(const wchar_t*, size_t)): Likewise.
(operator""sv(const char16_t*, size_t)): Likewise.
(operator""sv(const char32_t*, size_t)): Likewise.
* testsuite/21_strings/basic_string_view/literals/types.cc: New.
* testsuite/21_strings/basic_string_view/literals/values.cc: Likewise.
* testsuite/experimental/string_view/literals/values.cc: Add
tests for literals with embedded NULs.

From-SVN: r242367

re PR tree-optimization/35503 (Warning about restricted pointers?)

2016-11-13  Prathamesh Kulkarni  <prathamesh.kulkarni@linaro.org>

PR c/35503
* doc/invoke.texi: Document Wrestrict.
* pretty-print.c (pp_format): Add case for "Z" specifier.
(test_pp_format): Test "Z" specifier.
c-family/
* c-common.h (warn_for_restrict): Declare.
* c-warn.c: Include gcc-rich-location.h.
(warn_for_restrict): New function.
* c-format.c (gcc_tdiag_char_table): Add entry for "Z" specifier.
(gcc_cdiag_char_table): Likewise.
(gcc_cxxdiag_char_table): Likewise.
* c.opt (Wrestrict): New option.
c/
* c-parser.c (c_parser_postfix_expression_after_primary): Call
warn_for_restrict.
cp/
* parser.c (cp_parser_postfix_pexpression): Call warn_for_restrict.
testsuite/
* c-c++-common/pr35503-1.c: New test.
* c-c++-common/pr35503-2.c: Likewise.
* c-c++-common/pr35503-3.c: Likewise.
* gcc.dg/format/gcc_diag-1.c: Add tests for "Z" specifier.

From-SVN: r242366

decl.c (gnat_to_gnu_entity): In assertion about known Esize...

* gcc-interface/decl.c (gnat_to_gnu_entity): In assertion about known
Esize, protect with !is_type and change !Unknown_Esize to Known_Esize.

From-SVN: r242363

re PR rtl-optimization/78232 (FAIL: gcc.dg/torture/pr48124-4.c)

PR rtl-optimization/78232
PR rtl-optimization/78248
* gcc.dg/ubsan/pr78248.c: New test.

From-SVN: r242362

decl.c (gnat_to_gnu_entity): Look at the underlying type for the signedness of the type.

* gcc-interface/decl.c (gnat_to_gnu_entity) <E_Signed_Integer_Subtype>:
Look at the underlying type for the signedness of the type.

From-SVN: r242361

decl.c (annotate_value): Deal specially with negative constants.

* gcc-interface/decl.c (annotate_value) <INTEGER_CST>: Deal specially
with negative constants.

From-SVN: r242360

utils2.c (gnat_protect_expr): Also protect only the address if the expression is the component of a dereference.

* gcc-interface/utils2.c (gnat_protect_expr): Also protect only the
address if the expression is the component of a dereference.
Do not use a reference type for the final temporary reference.

From-SVN: r242358

Makefile.in (NO_OMIT_ADAFLAGS): Define.

* gcc-interface/Makefile.in (NO_OMIT_ADAFLAGS): Define.
(a-except.o): Replace -fno-inline with NO_INLINE_ADAFLAGS.
(s-memory.o): New rule.
(tracebak.o): Replace -fno-omit-frame-pointer with NO_OMIT_ADAFLAGS.

From-SVN: r242357

c-ada-spec.c (print_ada_declaration): For typedef declarations...

* c-ada-spec.c (print_ada_declaration): For typedef declarations, look
for nested types only if the type is a record or union and dump SLOC.

From-SVN: r242356

ipa-icf.c (sem_function::merge): Do not create a wrapper also if the original function needs a static chain.

* ipa-icf.c (sem_function::merge): Do not create a wrapper also if the
original function needs a static chain.

From-SVN: r242354

re PR target/78336 (powerpc-darwin bootstrap broken (probably by 241930))

        PR target/78336
        * config/rs6000/rs6000.c (rs6000_asm_weaken_decl): Protect
        ASM_OUTPUT_DEF.

From-SVN: r242353

re PR fortran/60952 ([F03] Problem using "end" as a type-bound procedure and contained procedure)

2016-11-13  Janus Weil  <janus@gcc.gnu.org>

PR fortran/60952
* decl.c (match_procedure_in_type): Apply the FL_PROCEDURE attribute
to the target procedure.

2016-11-13  Janus Weil  <janus@gcc.gnu.org>

PR fortran/60952
* gfortran.dg/typebound_proc_34.f90: New test.

From-SVN: r242352

re PR fortran/66366 ([OOP] ICE on invalid with non-allocatable CLASS variable)

2016-11-13  Janus Weil  <janus@gcc.gnu.org>

PR fortran/66366
* resolve.c (resolve_component): Move check for C437
to ...
* decl.c (build_struct): ... here. Fix indentation.

2016-11-13  Janus Weil  <janus@gcc.gnu.org>

PR fortran/66366
* gfortran.dg/class_57.f90: Changed error message.
* gfortran.dg/class_60.f90: New test.

From-SVN: r242351

CWG 2233 - default arg and parameter pack

* typeck.c (convert_arguments): Handle default arg followed by none.

From-SVN: r242350

Fix constexpr lvalue use of __real and __imag.

* constexpr.c (potential_constant_expression_1): REALPART_EXPR and
IMAGPART_EXPR can be lvalues.

From-SVN: r242349

DR 374 - specialization in outer namespace

PR c++/56840
* pt.c (check_specialization_namespace): Allow any enclosing
namespace.
(check_unqualified_spec_or_inst): New.
(check_explicit_specialization): Call it.
* parser.c (cp_parser_elaborated_type_specifier)
(cp_parser_class_head): Call it.

From-SVN: r242348

Daily bump.

From-SVN: r242345

* g++.dg/pr78112.C: XFAIL AIX.

From-SVN: r242342

* src/filesystem/ops.cc (is_empty): Fix typo in exception message.

From-SVN: r242341

* include/std/future (future_error): Fix public typo to private.

From-SVN: r242340

* es.po: Update.

From-SVN: r242338

rs6000: Don't forget to initialize the TOC (PR77957)

The code generating traceback tables mistakenly does an early return
if !optional_tbtab, which causes it to miss the code generating the TOC
section.  This only matters if the TOC will be empty since otherwise
the section is created elsewhere.

This patch fixes it.

PR target/77957
* config/rs6000/rs6000.c (rs6000_output_function_epilogue): Don't
return early if !optional_tbtab.

From-SVN: r242336

re PR fortran/77501 ([F03] ICE in gfc_match_generic, at fortran/decl.c:9429)

2016-11-12  Janus Weil  <janus@gcc.gnu.org>

PR fortran/77501
* class.c (gfc_find_typebound_intrinsic_op): Remove an unnecessary
assert and nullification.
* decl.c (gfc_match_decl_type_spec): Use gfc_get_tbp_symtree,
fix indentation.
(gfc_match_generic): Remove an unnecessary assert.
Use gfc_get_tbp_symtree to avoid ICE.

2016-11-12  Janus Weil  <janus@gcc.gnu.org>

PR fortran/77501
* gfortran.dg/typebound_generic_16.f90: New test.

From-SVN: r242335

Add std::future_error constructor from future_errc

* include/std/future (future_error): Make existing constructor
private and add constructor from future_errc.

From-SVN: r242334

Use shared_ptr<T>::element_type in hash specializations

* include/bits/shared_ptr.h (hash<shared_ptr<T>>): Use element_type.
* include/bits/shared_ptr_base.h (hash<__shared_ptr<T, L>>): Likewise.

From-SVN: r242333

Daily bump.

From-SVN: r242331

re PR c++/71225 (Overeager instantiation of members of non-template class nested in class template)

PR c++/71225
* g++.dg/cpp0x/pr71225.C: New test.

From-SVN: r242328

re PR rtl-optimization/59461 (missed zero-extension elimination in the combiner)

PR rtl-optimization/59461
* doc/rtl.texi (paradoxical subregs): Add missing word.
* combine.c (reg_nonzero_bits_for_combine): Do not discard results
in modes with precision larger than that of last_set_mode.
* rtlanal.c (nonzero_bits1) <SUBREG>: If WORD_REGISTER_OPERATIONS is
set and LOAD_EXTEND_OP is appropriate, propagate results from inner
REGs to paradoxical SUBREGs.
(num_sign_bit_copies1) <SUBREG>: Likewise.  Check that the mode is not
larger than a word before invoking LOAD_EXTEND_OP on it.

From-SVN: r242326

* fr.po: Update.

From-SVN: r242324

i386.md (*<shift_insn><mode>3_doubleword): Mark operand 0 as earlyclobber.

* config/i386/i386.md (*<shift_insn><mode>3_doubleword): Mark
operand 0 as earlyclobber.
(*ashl<mode>3_doubleword): Ditto for all operand 0 alternatives.

From-SVN: r242318

re PR target/78243 (incorrect byte offset in vextractuh with -mcpu=power9)

2016-11-11  Michael Meissner  <meissner@linux.vnet.ibm.com>

PR target/78243
* config/rs6000/vsx.md (vsx_extract_<mode>_p9): Correct the
element order for little endian ordering.

* config/rs6000/altivec.md (reduc_plus_scal_<mode>): Use
VECTOR_ELT_ORDER_BIG and not BYTES_BIG_ENDIAN to adjust element
number.

From-SVN: r242317

[i386][musl] Add cpuinfo to static libgcc only on *-musl*

The __cpu_indicator_init and __cpu_model symbols are not safe to use
from shared libgcc_s.so from ifunc resolvers, so since gcc-6, only
the definitions from static libgcc.a are used, however the symbols
are kept in libgcc_s as well for backward compatibility (with
appropriate symbol version).  On targets without such backward
compatibility concern add cpuinfo to the static library only (this
avoids running the ctor, reduces libgcc_s size and elf abi concerns
about the versioned symbols).

libgcc/
2016-11-11  Szabolcs Nagy  <szabolcs.nagy@arm.com>

* config.host (i[3456]86-*-musl*, x86_64-*-musl*): Use
i386/t-cpuinfo-static instead of i386/t-cpuinfo.
* config/i386/t-cpuinfo-static: New.

From-SVN: r242268

re PR target/78310 (ICE: insn does not satisfy its constraints: {*bmi2_rorxdi3_1} with -mbmi2)

PR target/78310
* config/i386/i386.md (rotate to rotatex splitter): Avoid overflow
when calculating operand 2.
(rotate to rotatex zext splitter): Ditto.

testsuite/ChangeLog:

PR target/78310
* gcc.target/i386/pr78310.c: New test.

From-SVN: r242076

gimple-ssa-isolate-paths.c (is_divmod_with_given_divisor): New function.

* gimple-ssa-isolate-paths.c (is_divmod_with_given_divisor): New
function.
(stmt_uses_name_in_undefined_way): New function, extracted from
find_implicit_erroneous_behavior and extended for div/mod case.
(stmt_uses_0_or_null_in_undefined_way): New function, extracted from
find_explicit_erroneous_behavior and extended for div/mod case.
(find_implicit_erroneous_behavior): Use new helper function.
(find_explicit_erroneous_behavior): Use new helper function.

* gcc.dg/tree-ssa/isolate-6.c: New test.
* gcc.dg/tree-ssa/isolate-7.c: New test.

From-SVN: r242075

re PR testsuite/78292 (test case gcc.dg/vect/vect-cond-2.c fails starting with r241967)

gcc/testsuite
PR testsuite/78292
* gcc.dg/vect/vect-cond-2.c: Only drop xfail for targets supporting
vect_max_reduc.

From-SVN: r242073

libgo: fix GOARCH_INT64ALIGN for several targets

    From Andreas Schwab.

    Reviewed-on: https://go-review.googlesource.com/33130

From-SVN: r242072

re PR c++/72774 (ICE on invalid C++ code on x86_64-linux-gnu (tree check: expected tree that contains ‘decl minimal’ structure, have ‘tree_list’ in consider_binding_level, at cp/name-lookup.c:4721))

PR c++/72774
* g++.dg/parse/pr72774.C: New test.

From-SVN: r242070

re PR tree-optimization/71575 ([graphite] internal compiler error: in copy_cond_phi_nodes, at graphite-isl-ast-to-gimple.c:2500)

2016-11-11  Richard Biener  <rguenther@suse.de>

PR tree-optimization/71575
* graphite-isl-ast-to-gimple.c (copy_cond_phi_nodes): Remove
bogus assert.

* gcc.dg/graphite/pr71575-1.c: New testcase.
* gcc.dg/graphite/pr71575-2.c: Likewise.

From-SVN: r242069

re PR middle-end/78295 (Spurious -Wuninitialized warning for vector element assignment)

2016-11-11  Richard Biener  <rguenther@suse.de>

PR middle-end/78295
* tree-ssa-uninit.c (warn_uninitialized_vars): Do not warn
about uninitialized destination arg of BIT_INSERT_EXPR.

* gcc.dg/uninit-pr78295.c: New testcase.

From-SVN: r242068

Daily bump.

From-SVN: r242065

re PR c/37998 (Unclear documentation of -fno-common)

2016-11-10  Sandra Loosemore  <sandra@codesourcery.com>

PR c/37998

gcc/
* doc/invoke.texi (Code Gen Options) [-fno-common]: Use correct
terminology.  Expand to remove ambiguity.

From-SVN: r242062

runtime: copy signal code from Go 1.7 runtime

    Add a little shell script to auto-generate runtime.sigtable from the
    known signal names.

    Force the main package to always import the runtime package.  Otherwise
    some runtime package global variables may never be initialized.

    Set the syscallsp and syscallpc fields of g when entering a syscall, so
    that the runtime package knows when a g is executing a syscall.

    Fix runtime.funcPC to avoid dead store elimination of the interface
    value when the function is inlined.

    Reviewed-on: https://go-review.googlesource.com/33025

From-SVN: r242060

combine: Do not call simplify from inside change_zero_ext (PR78232)

When combine splits a three-insn combination into two instructions it
can reuse i2dest for the temporary result of the first new instruction.
However all information it has in reg_stat about that register will be
stale.  This results in the simplify_gen_binary calls in change_zero_ext
using out-of-date information, which makes it think one of the ANDs
generated there always results in 0, and it doesn't get better from there.

This can also happen if a splitter in the MD uses nonzero_bits (for
example).  I tried to make the splitting code in combine save and restore
the i2dest reg_stat info, but that causes one of the acats tests to fail.

This whole reg_stat thing needs an overhaul, and/or we shouldn't reuse
i2dest for unrelated purposes when splitting.

This patch changes change_zero_ext to do the expected simplifications
itself and not call simplify_gen_*.

PR rtl-optimization/78232
* combine.c (try_combine): Add a big comment about why reusing i2dest
is undesirable.
(change_zero_ext): Do not call simplify_gen_binary, do the
simplifications manually.

From-SVN: r242059

Fix ICE and improve errors for invalid anonymous structure declarations.

PR fortran/78277
* gcc/fortran/decl.c (gfc_match_data_decl): Gracefully handle bad
anonymous structure declarations.

PR fortran/78277
* gcc/testsuite/gfortran.dg/dec_structure_17.f90: New test.

From-SVN: r242058

Fix some whitespace.

gcc/fortran/
* decl.c (get_struct_decl, gfc_match_map, gfc_match_union): Fix
whitespace.
* interface.c (gfc_compare_union_types): Likewise.

From-SVN: r242057

PR c++/77337 - auto return and lambda

* pt.c (tsubst_friend_function): Don't set DECL_INITIAL.
(instantiate_decl): It's OK to defer a constexpr function.
* cp-tree.h (DECL_FRIEND_PSEUDO_TEMPLATE_INSTANTIATION): Check
DECL_LANG_SPECIFIC.
* decl2.c (decl_defined_p): Use it.  No longer static.
* decl.c (redeclaration_error_message): Use decl_defined_p.
* constexpr.c (cxx_eval_call_expression): Set input_location around
call to instantiate_decl.

From-SVN: r242056

debug.cc (format_word): Delete.

2016-11-10  François Dumont  <fdumont@gcc.gnu.org>

* src/c++11/debug.cc (format_word): Delete.
(print_literal): New. Replace call to print_word for literals.

From-SVN: r242055

rs6000.c (rs6000_hard_regno_mode_ok): If ISA 3.0...

[gcc]
2016-11-10  Michael Meissner  <meissner@linux.vnet.ibm.com>

* config/rs6000/rs6000.c (rs6000_hard_regno_mode_ok): If ISA 3.0,
enable HImode and QImode to go in vector registers by default if
the -mvsx-small-integer option is enabled.
(rs6000_secondary_reload_simple_move): Likewise.
(rs6000_preferred_reload_class): Don't force integer constants to
be loaded into vector registers that we can easily make into
memory (or being created in the GPRs and moved over with direct
move).
* config/rs6000/vsx.md (UNSPEC_P9_MEMORY): Delete, no longer
used.
(vsx_extract_<mode>): Rework V4SImode, V8HImode, and V16QImode
vector extraction on ISA 3.0 when the scalar integer can be
allocated in vector registers.  Generate the VEC_SELECT directy,
and don't use UNSPEC's to avoid having the scalar type in a vector
register.  Make the expander target registers, and let the
combiner fold in results storing to memory, if the machine
supports stores.
(vsx_extract_<mode>_di): Likewise.
(vsx_extract_<mode>_p9): Likewise.
(vsx_extract_<mode>_di_p9): Likewise.
(vsx_extract_<mode>_store_p9): Likewise.
(vsx_extract_si): Likewise.
(vsx_extract_<mode>_p8): Likewise.
(p9_lxsi<wd>zx): Delete, no longer used.
(p9_stxsi<wd>x): Likewise.
* config/rs6000/rs6000.md (INT_ISA3): New mode iterator for
integers in vector registers for ISA 3.0.
(QHI): Update comment.
(zero_extendqi<mode>2): Add support for ISA 3.0 scalar load or
vector extract instructions in sign/zero extend.
(zero_extendhi<mode>): Likewise.
(extendqi<mode>): Likewise.
(extendhi<mode>2): Likewise.
(HImode splitter for load/sign extend in vector register):
Likewise.
(float<QHI:mode><FP_ISA3:mode>2): Eliminate old method of
optimizing floating point conversions to/from small data types and
rewrite it to support QImode/HImode being allowed in vector
registers on ISA 3.0.
(float<QHI:mode><FP_ISA3:mode>2_internal): Likewise.
(floatuns<QHI:mode><FP_ISA3:mode>2): Likewise.
(floatuns<QHI:mode><FP_ISA3:mode>2_internal): Likewise.
(fix_trunc<SFDF:mode><QHI:mode>2): Likewise.
(fix_trunc<SFDF:mode><QHI:mode>2_internal): Likewise.
(fixuns_trunc<SFDF:mode><QHI:mode>2): Likewise.
(fixuns_trunc<SFDF:mode><QHI:mode>2_internal): Likewise.
VSPLITISW on ISA 2.07.
(movhi_internal): Combine movhi_internal and movqi_internal into
one mov<mode>_internal with an iterator.  Add support for QImode
and HImode being allowed in vector registers.  Make large number
of attributes and constraints easier to read.
(movqi_internal): Likewise.
(mov<mode>_internal): Likewise.
(movdi_internal64): Fix constraint to allow loading -16..15 with
VSPLITISW on ISA 2.07.
(integer XXSPLTIB splitter): Add support for QI, HI, and SImode as
well as DImode.

[gcc/testsuite]
2016-11-10  Michael Meissner  <meissner@linux.vnet.ibm.com>

* gcc.target/powerpc/vsx-qimode.c: New test for QImode, HImode
being allowed in vector registers.
* gcc.target/powerpc/vsx-qimode2.c: Likewise.
* gcc.target/powerpc/vsx-qimode3.c: Likewise.
* gcc.target/powerpc/vsx-himode.c: Likewise.
* gcc.target/powerpc/vsx-himode2.c: Likewise.
* gcc.target/powerpc/vsx-himode3.c: Likewise.
* gcc.target/powerpc/p9-extract-1.c: Change MFVSRD to just MFVSR,
to allow matching MFVSRD or MFVSRW.

From-SVN: r242048

re PR rtl-optimization/78241 (wrong code with -funroll-loops)

PR rtl-optimization/78241
* loop-unroll.c (unroll_loop_runtime_iterations): Don't adjust 'niter', but
emit initial peel copy if niter expr is not reliable.

* gcc.dg/pr78241.c: New test.

From-SVN: r242047

dwarf2cfi: Dump row differences before asserting

If maybe_record_trace_start fails because the CFI is inconsistent on two
paths into a block it currently just ICEs.  This changes it to also dump
the CFI on those two paths in the dump file; debugging it without that
information is hopeless.

* dwarf2cfi.c (dump_cfi_row): Add forward declaration.
(maybe_record_trace_start): If the CFI is different on the new and
old paths, print out both to the dump file before ICEing.

From-SVN: r242046

cpp.c (cpp_define_builtins): Define _OPENMP to 201511 instead of 201307.

gcc/fortran/
* cpp.c (cpp_define_builtins): Define _OPENMP to 201511 instead
of 201307.
* gfortran.texi: Mention partial OpenMP 4.5 support.
* intrinsic.texi: Update for OpenMP 4.5.
gcc/testsuite/
* gfortran.dg/openmp-define-3.f90: Expect 201511 instead of
201307.
libgomp/
* omp_lib.f90.in (openmp_version): Change to 201511 from 201307.
* omp_lib.h.in (openmp_version): Likewise.
* testsuite/libgomp.fortran/openmp_version-1.f: Expect 201511 instead
of 201307.
* testsuite/libgomp.fortran/openmp_version-2.f90: Likewise.

From-SVN: r242044

target.def (additional_allocno_class_p): New.

2016-11-10  Vladimir Makarov  <vmakarov@redhat.com>

* target.def (additional_allocno_class_p): New.
* hooks.h (hook_bool_reg_class_t_false): New prototype.
* hooks.c (hook_bool_reg_class_t_false): New.
* ira.c (setup_allocno_and_important_classes): Use the new hook.
* doc/tm.texi.in (TARGET_ADDITIONAL_ALLOCNO_CLASS_P): Add it.
* doc/tm.texi: Update.

From-SVN: r242043

mangle.c (mangle_decl): Only emit -Wc++1z-compat warnings for public or external symbols.

* mangle.c (mangle_decl): Only emit -Wc++1z-compat warnings for
public or external symbols.

* g++.dg/cpp1z/noexcept-type14.C: New test.
* g++.dg/asan/asan_test.C: Remove -Wno-c++1z-compat from dg-options.

From-SVN: r242042

[Patch testsuite] Check range of exponent for tests added in r241817

gcc/ChangeLog

2016-11-10  James Greenhalgh  <james.greenhalgh@arm.com>

* gcc.dg/torture/fp-int-convert.h (M_OK2): New, use it in
WVAL0S tests added in r241817.

From-SVN: r242041

gengtype.c (new_structure): Append to structures list.

* gengtype.c (new_structure): Append to structures list.

(find_structure): Likewise.

From-SVN: r242040

Fix loop distribution bug with merging edges that creates too many edges.

gcc/
* tree-loop-distribution.c (pg_add_dependence_edges): Return 2 if
this_dir is 2.  Check for this_dir non-zero before dir != this_dir
check.

From-SVN: r242038

omp-low.c (lower_omp_target): Fix up argument to is_reference.

gcc/
* omp-low.c (lower_omp_target): Fix up argument to is_reference.
(expand_omp_ordered_sink): Handle TREE_PURPOSE of deps being
TRUNC_DIV_EXPR.
* gimplify.c (gimplify_scan_omp_clauses): Likewise.  Set
ctx->target_map_scalars_firstprivate on OMP_TARGET even for Fortran.
Remove omp_no_lastprivate callers.  Propagate lastprivate on combined
teams distribute parallel for simd even to distribute and teams
construct.  For OMP_CLAUSE_DEPEND add missing break at the end of
OMP_CLAUSE_DEPEND_SINK case.
(omp_notice_variable): Use lang_hooks.decls.omp_scalar_p.
(omp_no_lastprivate): Removed.
(gimplify_adjust_omp_clauses): Remove omp_no_lastprivate callers.
(gimplify_omp_for): Likewise.
(computable_teams_clause): Fail for automatic vars from current
function not yet seen in bind expr.
* langhooks.c (lhd_omp_scalar_p): New function.
* langhooks.h (struct lang_hooks_for_decls): Add omp_scalar_p.
* varpool.c (varpool_node::get_create): Set node->offloading
even for DECL_EXTERNAL decls.
* langhooks-def.h (lhd_omp_scalar_p): New prototype.
(LANG_HOOKS_OMP_SCALAR_P): Define.
(LANG_HOOKS_DECLS): Use it.
gcc/fortran/
* openmp.c (gfc_free_omp_clauses): Free critical_name, grainsize,
hint, num_tasks, priority and if_exprs.
(gfc_match_omp_to_link, gfc_match_omp_depend_sink): New functions.
(enum omp_mask1, enum omp_mask2): New enums.
Change all OMP_CLAUSE_* defines into enum values, and change their
values from ((uint64_t) 1 << bit) to just bit.
(omp_mask, omp_inv_mask): New classes.  Add ctors and operators.
(gfc_match_omp_clauses): Change mask argument from uint64_t to
const omp_mask.  Assert OMP_MASK1_LAST and OMP_MASK2_LAST are
at most 64.  Move delete clause handling to where it
alphabetically belongs.  Parse defaultmap, grainsize, hint,
is_device_ptr, nogroup, nowait, num_tasks, priority, simd, threads
and use_device_ptr clauses.  Parse if clause modifier.  Parse map
clause always modifier, and release and delete kinds.  Parse ordered
clause with argument.  Parse schedule clause modifiers.  Differentiate
device clause parsing based on openacc flag.  Guard link clause
parsing with openacc flag.  Add support for parsing
linear clause modifiers.  Parse depend(source) and depend(sink: ...).
Use gfc_match_omp_to_link for to and link clauses in declare target
construct.
(match_acc): Change mask type from uint64_t to const omp_mask.
(OMP_SINGLE_CLAUSES, OMP_ORDERED_CLAUSES,
OMP_DECLARE_TARGET_CLAUSES, OMP_TASKLOOP_CLAUSES,
OMP_TARGET_ENTER_DATA_CLAUSES, OMP_TARGET_EXIT_DATA_CLAUSES): Define.
(OACC_PARALLEL_CLAUSES, OACC_KERNELS_CLAUSES, OACC_DATA_CLAUSES,
OACC_LOOP_CLAUSES, OACC_HOST_DATA_CLAUSES, OACC_DECLARE_CLAUSES,
OACC_ENTER_DATA_CLAUSES, OACC_EXIT_DATA_CLAUSES, OACC_WAIT_CLAUSES,
OACC_ROUTINE_CLAUSES, OMP_PARALLEL_CLAUSES, OMP_DECLARE_SIMD_CLAUSES,
OMP_SECTIONS_CLAUSES, OMP_TEAMS_CLAUSES, OMP_DISTRIBUTE_CLAUSES):
Replace first or only OMP_CLAUSE_* value in bitset with
omp_mask (OMP_CLAUSE_*).
(OMP_DO_CLAUSES): Likewise.  Add OMP_CLAUSE_LINEAR.
(OMP_SIMD_CLAUSES): Replace first or only OMP_CLAUSE_* value in
bitset with omp_mask (OMP_CLAUSE_*).  Add OMP_CLAUSE_SIMDLEN.
(OACC_UPDATE_CLAUSES): Replace first or only OMP_CLAUSE_* value in
bitset with omp_mask (OMP_CLAUSE_*).  Replace OMP_CLAUSE_OACC_DEVICE
with OMP_CLAUSE_DEVICE.
(OMP_TASK_CLAUSES): Replace first or only OMP_CLAUSE_* value in
bitset with omp_mask (OMP_CLAUSE_*).  Add OMP_CLAUSE_PRIORITY.
(OMP_TARGET_CLAUSES): Replace first or only OMP_CLAUSE_* value in
bitset with omp_mask (OMP_CLAUSE_*).  Add OMP_CLAUSE_DEPEND,
OMP_CLAUSE_NOWAIT, OMP_CLAUSE_PRIVATE, OMP_CLAUSE_FIRSTPRIVATE,
OMP_CLAUSE_DEFAULTMAP and OMP_CLAUSE_IS_DEVICE_PTR.
(OMP_TARGET_DATA_CLAUSES): Replace first or only OMP_CLAUSE_* value in
bitset with omp_mask (OMP_CLAUSE_*).  Add OMP_CLAUSE_USE_DEVICE_PTR.
(OMP_TARGET_UPDATE_CLAUSES): Replace first or only OMP_CLAUSE_* value
in bitset with omp_mask (OMP_CLAUSE_*).  Add OMP_CLAUSE_DEPEND and
OMP_CLAUSE_NOWAIT.
(match_omp): Change mask argument from unsigned int to
const omp_mask.
(gfc_match_omp_critical): Parse optional clauses and use omp_clauses
union member instead of omp_name.
(gfc_match_omp_end_critical): New function.
(gfc_match_omp_distribute_parallel_do): Remove ordered and linear
clauses from the mask.
(gfc_match_omp_distribute_parallel_do_simd): Use
& ~(omp_mask (OMP_CLAUSE_*)) instead of & ~OMP_CLAUSE_*.
(gfc_match_omp_target_teams_distribute_parallel_do_simd): Likewise.
(gfc_match_omp_teams_distribute_parallel_do_simd): Likewise.
(gfc_match_omp_do_simd): Likewise.  Don't remove ordered clause from
the mask.
(gfc_match_omp_parallel_do_simd): Likewise.
(gfc_match_omp_target_teams_distribute_parallel_do): Likewise.
(gfc_match_omp_teams_distribute_parallel_do): Likewise.
(gfc_match_omp_declare_simd): If not using the form with
(proc-name), require space before first clause.  Make (proc-name)
optional.  If not present, set proc_name to NULL.
(gfc_match_omp_declare_target): Rewritten for OpenMP 4.5.
(gfc_match_omp_single): Use OMP_SINGLE_CLAUSES.
(gfc_match_omp_task, gfc_match_omp_taskwait, gfc_match_omp_taskyield):
Move around to where they belong alphabetically.
(gfc_match_omp_target_enter_data, gfc_match_omp_target_exit_data,
gfc_match_omp_target_parallel, gfc_match_omp_target_parallel_do,
gfc_match_omp_target_parallel_do_simd, gfc_match_omp_target_simd,
gfc_match_omp_taskloop, gfc_match_omp_taskloop_simd):
New functions.
(gfc_match_omp_ordered): Parse clauses.
(gfc_match_omp_ordered_depend): New function.
(gfc_match_omp_cancel, gfc_match_omp_end_single): Use
omp_mask (OMP_CLAUSE_*) instead of OMP_CLAUSE_*.
(resolve_oacc_scalar_int_expr): Renamed to ...
(resolve_scalar_int_expr): ... this.  Fix up formatting.
(resolve_oacc_positive_int_expr): Renamed to ...
(resolve_positive_int_expr): ... this.  Fix up formatting.
(resolve_nonnegative_int_expr): New function.
(resolve_omp_clauses): Adjust callers, use the above functions
even for OpenMP clauses, add handling of new OpenMP 4.5 clauses.
Require orderedc >= collapse if specified. Handle depend(sink:)
and depend(source) restrictions.  Disallow linear clause when
orderedc is non-zero.  Diagnose linear clause modifiers when not in
declare simd.  Only check for integer type if ref modifier
is not used.  Remove diagnostics for required VALUE attribute.
Diagnose VALUE attribute with ref or uval modifiers.  Allow
non-constant linear-step, if it is a dummy argument alone and is
mentioned in uniform clause.  Diagnose map kinds not allowed
for various constructs.  Diagnose target {enter ,exit ,}data without
any map clauses.  Add dummy OMP_LIST_IS_DEVICE_PTR and
OMP_LIST_USE_DEVICE_PTR cases.
(gfc_resolve_omp_do_blocks): Set omp_current_do_collapse to orderedc
if non-zero.
(gfc_resolve_omp_parallel_blocks): Handle new OpenMP 4.5 constructs,
replace underscores with spaces in a few construct names.
(resolve_omp_do): Set collapse to orderedc if non-zero.  Handle new
OpenMP 4.5 constructs.
(resolve_oacc_loop_blocks): Call resolve_positive_int_expr instead
of resolve_oacc_positive_int_expr.
(gfc_resolve_omp_directive): Handle new OpenMP 4.5 constructs.
(gfc_resolve_omp_declare_simd): Allow ods->proc_name to be NULL.
* trans-openmp.c (gfc_omp_scalar_p): New function.
(doacross_steps): New variable.
(gfc_trans_omp_clauses): Handle new OpenMP 4.5 clauses and new clause
modifiers.
(gfc_trans_omp_critical): Adjust EXEC_OMP_CRITICAL handling.
(gfc_trans_omp_do): Handle doacross loops.  Clear sched_simd flag.
Handle EXEC_OMP_TASKLOOP.
(gfc_trans_omp_ordered): Translate omp clauses, allow NULL
code->block.
(GFC_OMP_SPLIT_TASKLOOP, GFC_OMP_MASK_TASKLOOP): New enum constants.
(gfc_split_omp_clauses): Copy orderedc together with ordered.  Change
firstprivate and lastprivate handling for OpenMP 4.5.
Handle EXEC_OMP_TARGET_SIMD, EXEC_OMP_TARGET_PARALLEL{,_DO,_DO_SIMD}
and EXEC_OMP_TASKLOOP{,_SIMD}.  Add handling for new OpenMP 4.5
clauses and clause modifiers and handle if clause without/with
modifiers.
(gfc_trans_omp_teams): Add omp_clauses argument, add it to other
teams clauses.  Don't wrap into OMP_TEAMS if -fopenmp-simd.
(gfc_trans_omp_target): For -fopenmp, translate num_teams and
thread_limit clauses on combined target teams early and pass to
gfc_trans_omp_teams.  Set OMP_TARGET_COMBINED if needed.
Handle EXEC_OMP_TARGET_PARALLEL{,_DO,_DO_SIMD} and
EXEC_OMP_TARGET_SIMD.
(gfc_trans_omp_taskloop, gfc_trans_omp_target_enter_data,
gfc_trans_omp_target_exit_data): New functions.
(gfc_trans_omp_directive): Handle EXEC_OMP_TARGET_{ENTER,EXIT}_DATA
EXEC_OMP_TASKLOOP{,_SIMD}, EXEC_OMP_TARGET_PARALLEL{,_DO,_DO_SIMD}
and EXEC_OMP_TARGET_SIMD.  Adjust gfc_trans_omp_teams caller.
* symbol.c (check_conflict): Handle omp_declare_target_link.
(gfc_add_omp_declare_target_link): New function.
(gfc_copy_attr): Copy omp_declare_target_link.
* dump-parse-tree.c (show_omp_namelist): Handle OMP_DEPEND_SINK_FIRST
depend_op.  Print linear clause modifiers.
(show_omp_clauses): Adjust for OpenMP 4.5 clause changes.
(show_omp_node): Print clauses for EXEC_OMP_ORDERED.  Allow NULL
c->block for EXEC_OMP_ORDERED.  Formatting fixes.  Adjust handling of
EXEC_OMP_CRITICAL, handle new OpenMP 4.5 constructs and some
forgotten OpenMP 4.0 constructs.
(show_code_node): Handle new OpenMP 4.5 constructs and some forgotten
OpenMP 4.0 constructs.
* gfortran.h (symbol_attribute): Add omp_declare_target_link bitfield.
(struct gfc_omp_namelist): Add u.common and u.linear_op fields.
(struct gfc_common_head): Change omp_declare_target into bitfield.
Add omp_declare_target_link bitfield.
(gfc_add_omp_declare_target_link): New prototype.
(enum gfc_statement): Add ST_OMP_TARGET_PARALLEL,
ST_OMP_END_TARGET_PARALLEL, ST_OMP_TARGET_PARALLEL_DO,
ST_OMP_END_TARGET_PARALLEL_DO, ST_OMP_TARGET_PARALLEL_DO_SIMD,
ST_OMP_END_TARGET_PARALLEL_DO_SIMD, ST_OMP_TARGET_ENTER_DATA,
ST_OMP_TARGET_EXIT_DATA, ST_OMP_TARGET_SIMD, ST_OMP_END_TARGET_SIMD,
ST_OMP_TASKLOOP, ST_OMP_END_TASKLOOP, ST_OMP_TASKLOOP_SIMD,
ST_OMP_END_TASKLOOP_SIMD and ST_OMP_ORDERED_DEPEND.
(enum gfc_omp_depend_op): Add OMP_DEPEND_SINK_FIRST and
OMP_DEPEND_SINK.
(enum gfc_omp_linear_op): New.
(struct gfc_omp_clauses): Add critical_name, depend_source,
orderedc, defaultmap, nogroup, sched_simd, sched_monotonic,
sched_nonmonotonic, simd, threads, grainsize, hint, num_tasks,
priority and if_exprs fields.
(enum gfc_exec_op): Add EXEC_OMP_END_CRITICAL,
EXEC_OMP_TARGET_ENTER_DATA, EXEC_OMP_TARGET_EXIT_DATA,
EXEC_OMP_TARGET_PARALLEL, EXEC_OMP_TARGET_PARALLEL_DO,
EXEC_OMP_TARGET_PARALLEL_DO_SIMD, EXEC_OMP_TARGET_SIMD,
EXEC_OMP_TASKLOOP, EXEC_OMP_TASKLOOP_SIMD.
(enum gfc_omp_map_op): Add OMP_MAP_RELEASE,
OMP_MAP_ALWAYS_TO, OMP_MAP_ALWAYS_FROM and OMP_MAP_ALWAYS_TOFROM.
(OMP_LIST_IS_DEVICE_PTR, OMP_LIST_USE_DEVICE_PTR): New.
(enum gfc_omp_if_kind): New.
* module.c (enum ab_attribute): Add AB_OMP_DECLARE_TARGET_LINK.
(attr_bits): Add AB_OMP_DECLARE_TARGET_LINK entry.
(mio_symbol_attribute): Save and restore omp_declare_target_link bit.
* trans.h (gfc_omp_scalar_p): New prototype.
* frontend-passes.c (gfc_code_walker): Handle new OpenMP 4.5
expressions.
* trans.c (trans_code): Handle new OpenMP 4.5 constructs.
* resolve.c (gfc_resolve_blocks): Likewise.
(gfc_resolve_code): Likewise.
* f95-lang.c (LANG_HOOKS_OMP_SCALAR_P): Redefine to gfc_omp_scalar_p.
(gfc_attribute_table): Add "omp declare target link".
* st.c (gfc_free_statement): Handle EXEC_OMP_END_CRITICAL like
EXEC_OMP_CRITICAL before, free clauses for EXEC_OMP_CRITICAL
and new OpenMP 4.5 constructs.  Free omp clauses even for
EXEC_OMP_ORDERED.
* match.c (match_exit_cycle): Rename collapse variable to count,
set it to orderedc if non-zero, instead of collapse.
* trans-decl.c (add_attributes_to_decl): Add "omp declare target link"
instead of "omp declare target" for omp_declare_target_link.
* trans-common.c (build_common_decl): Likewise.
* match.h (gfc_match_omp_target_enter_data,
gfc_match_omp_target_exit_data, gfc_match_omp_target_parallel,
gfc_match_omp_target_parallel_do,
gfc_match_omp_target_parallel_do_simd, gfc_match_omp_target_simd,
gfc_match_omp_taskloop, gfc_match_omp_taskloop_simd,
gfc_match_omp_end_critical, gfc_match_omp_ordered_depend): New
prototypes.
* parse.c (decode_omp_directive): Use gfc_match_omp_end_critical
instead of gfc_match_omp_critical for !$omp end critical.
Handle new OpenMP 4.5 constructs.  If ordered directive has
depend clause as the first of the clauses, use
gfc_match_omp_ordered_depend and ST_OMP_ORDERED_DEPEND instead of
gfc_match_omp_ordered and ST_OMP_ORDERED.
(case_executable): Add ST_OMP_TARGET_ENTER_DATA,
ST_OMP_TARGET_EXIT_DATA and ST_OMP_ORDERED_DEPEND cases.
(case_exec_markers): Add ST_OMP_TARGET_PARALLEL,
ST_OMP_TARGET_PARALLEL_DO, ST_OMP_TARGET_PARALLEL_DO_SIMD,
ST_OMP_TARGET_SIMD, ST_OMP_TASKLOOP and ST_OMP_TASKLOOP_SIMD cases.
(gfc_ascii_statement): Handle new OpenMP 4.5 constructs.
(parse_omp_do): Handle ST_OMP_TARGET_PARALLEL_DO,
ST_OMP_TARGET_PARALLEL_DO_SIMD, ST_OMP_TASKLOOP and
ST_OMP_TASKLOOP_SIMD.
(parse_omp_structured_block): Handle EXEC_OMP_END_CRITICAL instead
of EXEC_OMP_CRITICAL, adjust for EXEC_OMP_CRITICAL having omp clauses
now.
(parse_executable): Handle ST_OMP_TARGET_PARALLEL,
ST_OMP_TARGET_PARALLEL_DO, ST_OMP_TARGET_PARALLEL_DO_SIMD,
ST_OMP_TASKLOOP and ST_OMP_TASKLOOP_SIMD.
gcc/testsuite/
* gfortran.dg/gomp/pr77516.f90: Add dg-warning.
* gfortran.dg/gomp/target1.f90: Remove ordered clause where it is
no longer allowed and corresponding ordered construct.
* gfortran.dg/gomp/linear-1.f90: New test.
* gfortran.dg/gomp/declare-simd-2.f90: New test.
* gfortran.dg/gomp/declare-target-1.f90: New test.
* gfortran.dg/gomp/declare-target-2.f90: New test.
libgomp/
* testsuite/libgomp.fortran/examples-4/declare_target-1.f90
(fib_wrapper): Add map(from: x) clause.
* testsuite/libgomp.fortran/examples-4/declare_target-2.f90
(e_53_2): Likewise.
* testsuite/libgomp.fortran/examples-4/declare_target-4.f90
(accum): Add map(tmp) clause.
* testsuite/libgomp.fortran/examples-4/declare_target-5.f90
(accum): Add map(tofrom: tmp) clause.
* testsuite/libgomp.fortran/examples-4/target_data-3.f90
(gramSchmidt): Likewise.
* testsuite/libgomp.fortran/examples-4/teams-2.f90 (dotprod): Add
map(tofrom: sum) clause.
* testsuite/libgomp.fortran/nestedfn5.f90 (foo): Add twice
map (alloc: a, l) clause.  Add defaultmap(tofrom: scalar) clause.
* testsuite/libgomp.fortran/pr66199-2.f90: Adjust for linear clause
only allowed on the loop iterator.
* testsuite/libgomp.fortran/target4.f90 (foo): Add map(t) clause.
* testsuite/libgomp.fortran/taskloop2.f90: New test.
* testsuite/libgomp.fortran/taskloop4.f90: New test.
* testsuite/libgomp.fortran/doacross1.f90: New test.
* testsuite/libgomp.fortran/doacross3.f90: New test.
* testsuite/libgomp.fortran/taskloop1.f90: New test.
* testsuite/libgomp.fortran/taskloop3.f90: New test.
* testsuite/libgomp.fortran/doacross2.f90: New test.
* testsuite/libgomp.c/doacross-1.c (main): Add missing
#pragma omp atomic read.
* testsuite/libgomp.c/doacross-2.c (main): Likewise.
* testsuite/libgomp.c/doacross-3.c (main): Likewise.

From-SVN: r242037

Create live_switch_vars conditionally (PR sanitizer/78270)

PR sanitizer/78270
* gcc.dg/asan/pr78270.c: New test.
PR sanitizer/78270
* gimplify.c (gimplify_switch_expr): Create live_switch_vars
only when SWITCH_BODY is a BIND_EXPR.

From-SVN: r242036

debug/78112: remove recent duplicates for DW_TAG_subprogram attributes

Disable unnecessary calls to dwarf2out_decl for the parents of nested
functions. This fixes a libstdc++ build regression on
x86_64-apple-darwin, as these calls used to add duplicate attributes
(for instance DW_AT_inline and DW_AT_object_pointer) to the
corresponding DW_TAG_subprogram DIE.

Bootstrapped and regtested on x86_64-linux.

gcc/
PR debug/78112
* dwarf2out.c (dwarf2out_early_global_decl): Call dwarf2out_decl
on the context only when it has no DIE yet.

testsuite/
PR debug/78112
* g++.dg/pr78112.C: New testcase

Co-Authored-By: Jakub Jelinek <jakub@redhat.com>
From-SVN: r242035

[ARM] Remove duplicated enum type for CPU identifiers

* arm.h (target_cpus): Delete.
* arm-opts.h (enum processor_type): Prefix entires with TARGET_CPU_.
* arm.c (all_cores): Prefix IDENT with TARGET_CPU_.
(all_architectures): Likewise.
(arm_option_override): Adjust use of CPU enums.
(arm_sched_reorder): Likewise.
* vfp.md (movdi_vfp, movdi_vfp_cortexa8): Likewise.
* arm.opt (mcpu, mtune): Adjust use of CPU enums.
* arm/genopt.sh (processor_type): Prefix enumeration entries with
TARGET_CPU_.
* arm-tables.opt: Regenerated.

From-SVN: r242034

[ARM/AArch64] Add Falkor CPU support.

2016-11-10  Siddhesh Poyarekar  <siddhesh.poyarekar@linaro.org>

* config/aarch64/aarch64-cores.def (qdf24xx): Update part
number.
(falkor): New core.
* config/aarch64/aarch64-tune.md: Regenerated.
* config/arm/arm-cores.def (falkor): New core.
* config/arm/arm-tables.opt: Regenerated.
* config/arm/arm-tune.md: Regenerated.
* config/arm/bpabi.h (BE8_LINK_SPEC): Add falkor support.
* config/arm/t-aprofile (MULTILIB_MATCHES): Likewise.
* doc/invoke.texi (AArch64 Options/-mtune): Document it.
(ARM Options/-mtune): Likewise.

From-SVN: r242033

revert: ipa-cp.c (ipa_get_jf_pass_through_result): Handle unary expressions.

2016-11-10  Kugan Vivekanandarajah  <kuganv@linaro.org>

Revert
2016-11-09  Kugan Vivekanandarajah  <kuganv@linaro.org>

* ipa-cp.c (ipa_get_jf_pass_through_result): Handle unary expressions.
(propagate_vr_accross_jump_function): Likewise.
* ipa-prop.c (ipa_set_jf_unary_pass_through): New.
(load_from_param_1): New.
(load_from_unmodified_param): Factor common part into load_from_param_1.
(load_from_param): New.
(compute_complex_assign_jump_func): Handle unary expressions.
(ipa_write_jump_function): Likewise.
(ipa_read_jump_function): Likewise.

From-SVN: r242032

simplify-rtx: Transform (xor (and (xor A B) C) B) with C const

match.pd transforms (A&C)|(B&~C) to ((A^B)&C)^B, which is fewer
operations if C is not const (and it is not on simple tests at least,
this transform is done very early already).

Various processors have "insert" instructions that can do this, but
combine cannot build those from the xor-and-xor, especially it has no
chance at all to do that if A or B or multiple instructions as well
(on PowerPC, the rl[wd]imi instructions can do this with together with
a rotate, or a simple shift with appropriate C; other ISAs have similar
insns).

This patch makes RTL simplify transform (xor (and (xor A B) C) B) back
to (ior (and A C) (and B ~C)) for constant C (and similar with A instead
of B for that last term).

* simplify-rtx.c (simplify_binary_operation_1): Simplify
(xor (and (xor A B) C) B) to (ior (and A C) (and B ~C)) and
(xor (and (xor A B) C) A) to (ior (and A ~C) (and B C)) if C
is a const_int.

From-SVN: r242031

Daily bump.

From-SVN: r242030

asan_test.C: Remove -Wno-format, add -Wno-c++1z-compat.

* g++.dg/asan/asan_test.C: Remove -Wno-format, add
-Wno-c++1z-compat.
* g++.dg/asan/dejagnu-gtest.h (EXPECT_TRUE): Use "%s" format
string in case condition contains % characters.

From-SVN: r242027

re PR c++/78283 (-Wc++1z-compat warning is sticky)

PR c++/78283
* mangle.c (start_mangling): Reset G.need_cxx1z_warning.

* g++.dg/cpp1z/noexcept-type13.C: New test.

From-SVN: r242026

regex.h (regex_iterator::regex_iterator()): Define end() as _M_pregex == nullptr.

* libstdc++-v3/include/bits/regex.h (regex_iterator::regex_iterator()):
Define end() as _M_pregex == nullptr.
* libstdc++-v3/include/bits/regex.tcc (regex_iterator::operator==(),
regex_iterator::operator++()): Fix operator==() and operator++() to
look at null-ness of _M_pregex on both sides.
* testsuite/28_regex/regression.cc: New testcase.

From-SVN: r242025

compiler: rework static initializer code

    Rename is_immutable to is_static_initializer to try to capture what it
    really means.  Be more precise about when an address expression, or a
    binary expression, can be a static initializer.  Don't check whether a
    type has pointers when deciding whether an initializer must be
    read-write, just check whether it is being used to initialize a global
    variable.  To make that work set the Translate_context function to NULL
    for a global variable with a static initializer.

    The effect of this is to let more global variables be initialized
    directly, rather than being initialized in the generated init function.

    Reviewed-on: https://go-review.googlesource.com/32917

From-SVN: r242024

print-rtl-function.c: add (param) directive to dump

gcc/ChangeLog:
* print-rtl-function.c: Include varasm.h.
(print_any_param_name): New function.
(print_param): New function.
(print_rtx_function): Call print_param for each argument.
* print-rtl.c (rtx_writer::finish_directive): New function.
* print-rtl.h (rtx_writer::finish_directive): New decl.

From-SVN: r242023

mips.exp (mips-dg-options): Upgrade to MIPS IV if using (HAS_MOVN) with MIPS III.

2016-11-09  Toma Tabacu  <toma.tabacu@imgtec.com>

        * gcc.target/mips/mips.exp (mips-dg-options): Upgrade to MIPS IV if
        using (HAS_MOVN) with MIPS III.

From-SVN: r242021

[multiple changes]

2016-11-09  Mikael Morin  <mikael@gcc.gnu.org>
    Janus Weil  <janus@gcc.gnu.org>

PR fortran/46459
* interface.c (compare_actual_formal): Add safety checks to avoid ICE.

2016-11-09  Janus Weil  <janus@gcc.gnu.org>

PR fortran/46459
* gfortran.dg/volatile14.f90: New test.

From-SVN: r242020

Fix C++17 template placeholder for template template parm.

* parser.c (cp_parser_simple_type_specifier): Allow placeholder
for template template parameter.
(cp_parser_type_id_1): Improve diagnostic.
* decl.c (grokdeclarator): Handle class deduction diagnostics here.
* pt.c (splice_late_return_type): Not here.
(tsubst) [TEMPLATE_TYPE_PARM]: Substitute into placeholder template.
(do_class_deduction): Handle non-class templates.

From-SVN: r242018

Implement P0127R2, Declaring non-type parameters with auto.

gcc/cp/
* cp-tree.h (enum auto_deduction_context): Add adc_unify.
* decl.c (grokdeclarator): Allow 'auto' in C++17 template non-type
parameter types.
* pt.c (do_auto_deduction): Add outer_targs parameter.
(convert_template_argument): Call do_auto_deduction.  If adc_unify,
don't give up on dependent init.
(unify): Likewise.  In C++17, walk into the type of a
TEMPLATE_PARM_INDEX.
(for_each_template_parm): Add any_fn parameter.
(struct pair_fn_data): Likewise.
(for_each_template_parm_r): Call it for any tree.  In C++17, walk
into the type of a TEMPLATE_PARM_INDEX.
(zero_r, array_deduction_r, try_array_deduction): New.
(type_unification_real): Call try_array_deduction.
(get_partial_spec_bindings): Likewise.
gcc/c-family/
* c-cppbuiltin.c (c_cpp_builtins): Define __cpp_template_auto.

From-SVN: r242017

re PR target/78262 (wrong code with -fschedule-insns)

PR target/78262
* config/i386/i386.md (*<shift_insn><mode>3_doubleword): Mark
operand 0 as earlyclobber.

testsuite/ChangeLog:

PR target/78262
* gcc.target/i386/pr78262.c: New test.

From-SVN: r242013

* fold-const-call.c (fold_const_call): Fix up type of s0 and s1.

From-SVN: r242011

Fix ICE in gfc_trans_subcomponent_assign due to NULL components.

PR fortran/78259
* gcc/fortran/trans-expr.c (gfc_trans_subcomponent_assign): Guard
against NULL values.

PR fortran/78259
* gcc/testsuite/gfortran.dg/pr78259.f90: New test.

From-SVN: r242010

[multiple changes]

2016-11-09  Steve Kargl <kargl@gcc.gnu.org>
    Janus Weil  <janus@gcc.gnu.org>

PR fortran/60777
* expr.c (external_spec_function): Allow recursive specification
functions in F03.

2016-11-09  Janus Weil  <janus@gcc.gnu.org>

PR fortran/60777
* gfortran.dg/spec_expr_7.f90: New test.

From-SVN: r242009

re PR middle-end/77718 (expand_builtin_memcmp swaps args)

PR target/77718
* builtins.c (expand_builtin_memcmp): Formatting fix.

* gcc.c-torture/execute/pr77718.c: New test.

From-SVN: r242007

flag-types.h (enum sanitize_code): Add SANITIZE_SHIFT_BASE and SANITIZE_SHIFT_EXPONENT...

* flag-types.h (enum sanitize_code): Add SANITIZE_SHIFT_BASE
and SANITIZE_SHIFT_EXPONENT, change SANITIZE_SHIFT to bitwise
or of them, renumber other enumerators.
* opts.c (sanitizer_opts): Add shift-base and shift-exponent.
* doc/invoke.texi: Document -fsanitize=shift-base and
-fsanitize-shift-exponent, document -fsanitize=shift as
having those 2 suboptions.
c-family/
* c-ubsan.c (ubsan_instrument_shift): Handle split
-fsanitize=shift-base and -fsanitize=shift-exponent.
testsuite/
* gcc.dg/ubsan/c99-shift-3.c: New test.
* gcc.dg/ubsan/c99-shift-4.c: New test.
* gcc.dg/ubsan/c99-shift-5.c: New test.
* gcc.dg/ubsan/c99-shift-6.c: New test.

From-SVN: r242005

fold-const.c (tree_swap_operands_p): Remove unused arg.

2016-11-09  Richard Biener  <rguenther@suse.de>

* fold-const.c (tree_swap_operands_p): Remove unused arg.
* fold-const.c (tree_swap_operands_p): Likewise.
(fold_binary_loc): Adjust.
(fold_ternary_loc): Likewise.
* genmatch.c (dt_operand::gen_gimple_exp): Likewise.
* gimple-fold.c (fold_stmt_1): Likewise.
* gimple-match-head.c (gimple_resimplify2): Likewise.
(gimple_resimplify3): Likewise.
(gimple_simplify): Likewise.
* tree-ssa-dom.c (record_equality): Likewise.
* tree-ssa-reassoc.c (optimize_range_tests_var_bound): Likewise.
* tree-ssa-sccvn.c (vn_nary_op_compute_hash): Likewise.
* tree-ssa-threadedge.c (simplify_control_stmt_condition_1): Likewise.

From-SVN: r242004

tree-ssa-dom.c (canonicalize_comparison): Remove.

2016-11-09  Richard Biener  <rguenther@suse.de>

* tree-ssa-dom.c (canonicalize_comparison): Remove.
(optimize_stmt): Remove redundant pre-propagation canonicalization
of comparison operand order.

From-SVN: r242003

Fix folding of memcmp("a", "a", 2) (PR

* fold-const-call.c (fold_const_call): Fix the folding.
* gcc.dg/tree-ssa/builtins-folding-generic.c (main): Add new
test-case for memcmp.
* gcc.dg/tree-ssa/builtins-folding-gimple.c: Likewise.

From-SVN: r242000

common.opt (flag_evaluation_order): Remove.

2016-11-09  Richard Biener  <rguenther@suse.de>

* common.opt (flag_evaluation_order): Remove.
* expr.c (expand_operands): Remove code guarded by
flag_evaluation_order.
* fold-const.c (reorder_operands_p): Remove, it always returns
true.
(negate_expr_p): Remove calls to reorder_operands_p.
(fold_negate_expr): Likewise.
(tree_swap_operands_p): Likewise.
(fold_binary_loc): Likewise.

From-SVN: r241998