re PR c++/71826 (ICE on valid C++ code with ambiguous member lookup: tree check: expected baselink, have error_mark in tsubst_baselink, at cp/pt.c:13737)

PR c++/71826
* pt.c (tsubst_baselink): Only set BASELINK_OPTYPE for BASELINK_P.

* g++.dg/template/pr71826.C: New test.

From-SVN: r238441

re PR c++/71822 (ICE: in gimplify_expr, at gimplify.c:11025)

PR c++/71822
* cp-gimplify.c (cp_gimplify_expr) <case VEC_INIT_EXPR>: Recursively
fold *expr_p before genericizing it.

* g++.dg/template/defarg21.C: New test.

From-SVN: r238440

re PR c++/71871 (ICE on mixing templates and vector extensions ternary operator)

PR c++/71871
* typeck.c (build_x_conditional_expr): Revert the 2012-10-25 change.

* g++.dg/ext/vector31.C: New test.

From-SVN: r238439

pr70017.c: Do not check for warning on alpha*-*-*.

* gcc.dg/pr70017.c: Do not check for warning on alpha*-*-*.

From-SVN: r238438

Demangle C++17 fold-expressions.

* cp-demangle.c (cplus_demangle_operators): Add f[lrLR].
(d_expression_1): Handle them.
(d_maybe_print_fold_expression): New.
(d_print_comp_inner): Use it.
(d_index_template_argument): Handle negative index.

From-SVN: r238437

Clean up optional's comments.

From-SVN: r238436

re PR middle-end/71734 (FAIL: libgomp.fortran/simd4.f90   -O3 -g  execution test)

gcc/

2016-07-18  Yuri Rumyantsev  <ysrumyan@gmail.com>

PR tree-optimization/71734
* tree-ssa-loop-im.c (ref_indep_loop_p_1): Add REF_LOOP argument which
contains REF, use it to check safelen, assume that safelen value
must be greater 1, fix style.
(ref_indep_loop_p_2): Add REF_LOOP argument.
(ref_indep_loop_p): Pass LOOP as additional argument to
ref_indep_loop_p_2.

gcc/testsuite/

2016-07-18  Yuri Rumyantsev  <ysrumyan@gmail.com>

PR tree-optimization/71734
* g++.dg/vect/pr70729.cc: Delete redundant dg options, fix style.

From-SVN: r238435

Allocate constant size dynamic stack space in the prologue

The attached patch fixes a warning during Linux kernel compilation
on S/390 due to -mwarn-dynamicstack and runtime alignment of stack
variables with constant size causing cfun->calls_alloca to be set
(even if alloca is not used at all).  The patched code places
constant size runtime aligned variables in the "virtual stack
vars" area instead of creating a "virtual stack dynamic" area.

This behaviour is activated by defining

  #define ALLOCATE_DYNAMIC_STACK_SPACE_IN_PROLOGUE 1

in the backend; otherwise the old logic is used.

The kernel uses runtime alignment for the page structure (aligned
to 16 bytes), and apart from triggereing the alloca warning
(-mwarn-dynamicstack), the current Gcc also generates inefficient
code like

  aghi %r15,-160  # prologue: create stack frame
    lgr %r11,%r15   # prologue: generate frame pointer
      aghi %r15,-32   # space for dynamic stack

which could be simplified to

  aghi %r15,-192

(if later optimization passes are able to get rid of the frame
pointer).  Is there a specific reason why the patched behaviour
shouldn't be used for all platforms?

--

As the placement of runtime aligned stack variables with constant
size is done completely in the middleend, I don't see a way to fix
this in the backend.

gcc/ChangeLog:

2016-07-18  Dominik Vogt  <vogt@linux.vnet.ibm.com>

* cfgexpand.c (expand_stack_vars): Implement synamic stack space
allocation in the prologue.
* explow.c (get_dynamic_stack_base): New function to return an address
expression for the dynamic stack base.
(get_dynamic_stack_size): New function to do the required dynamic stack
space size calculations.
(allocate_dynamic_stack_space): Use new functions.
(align_dynamic_address): Move some code from
allocate_dynamic_stack_space to new function.
* explow.h (get_dynamic_stack_base, get_dynamic_stack_size): Export.

gcc/testsuite/ChangeLog:

2016-07-18  Dominik Vogt  <vogt@linux.vnet.ibm.com>

* gcc.target/s390/warn-dynamicstack-1.c: New test.
* gcc.dg/stack-usage-2.c (foo3): Adapt expected warning.
stack-layout-dynamic-1.c: New test.

From-SVN: r238432

S/390: Fix alignment check for literal pool references.

gcc/ChangeLog:

2016-07-18  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>

* config/s390/s390.c (s390_encode_section_info): Always set
notaligned marker if mode size is 0 or no MEM_ALIGN info could be
found.

gcc/testsuite/ChangeLog:

2016-07-18  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>

* gcc.target/s390/nolrl-1.c: New test.

From-SVN: r238427

re PR tree-optimization/71893 (gfortran.dg ICEs in gcc/tree-ssa-pre.c; -fcode-hoisting?)

2016-07-18  Richard Biener  <rguenther@suse.de>

PR tree-optimization/71893
* tree-ssa-pre.c (create_component_ref_by_pieces_1): Compensate
for sizetype cast added by array_ref_element_size.
* tree-ssa-sccvn.c (copy_reference_ops_from_ref): Likewise.

From-SVN: r238426

Daily bump.

From-SVN: r238424

re PR fortran/71523 (Static variables given automatic initializers with -finit-* and -fmax-stack-var-size)

2016-07-17  Fritz Reese  <fritzoreese@gmail.com>

PR fortran/71523
* trans-decl.c (gfc_finish_var_decl): Replace automatic initializer with
a static one.

* gfortran.dg/pr71523_1.f90: New test.
* gfortran.dg/pr71523_2.f90: New test.

From-SVN: r238420

Daily bump.

From-SVN: r238419

pa.c (hppa_profile_hook): Allocate stack space for register parameters.

* config/pa/pa.c (hppa_profile_hook): Allocate stack space for
register parameters.  Remove code to initialize argument pointer
on TARGET_64BIT.  Optimize call to _mcount when it can be reached
using a pc-relative branch.  Cleanup conditional code.
* config/pa/pa.md (call_mcount): New expander.
(call_mcount_nonpic): New insn.
(call_mcount_pic): New insn and split.
(call_mcount_pic_post_reload): New insn.
(call_mcount_64bit): New insn and split.
(call_mcount_64bit_post_reload): New insn.

From-SVN: r238414

nullptr35.C (caught): Fix typo.

* g++.dg/cpp0x/nullptr35.C (caught): Fix typo.

From-SVN: r238413

Daily bump.

From-SVN: r238409

Replace references to C++0x with C++11 in comments

* include/bits/algorithmfwd.h: Change C++0x to C++11 in comments.
* include/bits/move.h: Likewise.
* include/bits/postypes.h: Likewise.
* include/debug/bitset: Likewise.
* include/ext/pb_ds/detail/type_utils.hpp: Likewise.
* include/ext/string_conversions.h: Change C++0x to __cxx11 in
comment.
* testsuite/27_io/fpos/14320-1.cc: Change C++0x to C++11 in comment.
* testsuite/util/thread/all.h: Likewise.

From-SVN: r238402

re PR fortran/62125 (Nested select type not accepted (rejects valid))

2016-07-15  Jerry DeLisle  <jvdelisle@gcc.gnu.org>
    Marco Restelli <mrestelli@gmail.com>

PR fortran/62125
* symbol.c (select_type_insert_tmp): Recursively call self to take care
of nested select type.

* gfortran.dg/pr62125.f90: New test.

Co-Authored-By: Marco Restelli <mrestelli@gmail.com>
From-SVN: r238400

Remove redundant std::move in std::for_each

* include/bits/stl_algo.h (for_each): Remove redundant _GLIBCXX_MOVE
and adjust comment.

From-SVN: r238399

divkc3-1.c: Require p8vector support.

2016-07-15  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>

* gcc.target/powerpc/divkc3-1.c: Require p8vector support.
* gcc.target/powerpc/mulkc3-1.c: Likewise.

From-SVN: r238398

PR c++/71495 - spurious note during SFINAE.

* call.c (convert_like_real): Mask complain.
* semantics.c (perform_koenig_lookup): Likewise.

From-SVN: r238397

c++/58796 Make nullptr match exception handlers of pointer type

libstdc++-v3:

PR c++/58796
* libsupc++/pbase_type_info.cc (__pbase_type_info::__do_catch): Make
nullptr match handlers of pointer type.

gcc/testsuite:

PR c++/58796
* g++.dg/cpp0x/nullptr21.C: Remove void* handlers.
* g++.dg/cpp0x/nullptr35.C: New test.

From-SVN: r238396

PR c++/71092 - ICE with array and constexpr.

* constexpr.c (cxx_eval_call_expression): Fail quietly when cgraph
threw away DECL_SAVED_TREE.

From-SVN: r238395

PR c++/71117 - core 2189 and generic lambda

* call.c (add_template_conv_candidate): Disable if there are
viable candidates.

From-SVN: r238394

PR c++/71511 - ICE on decltype scope in declaration.

* typeck2.c (cxx_incomplete_type_diagnostic): Handle DECLTYPE_TYPE.

From-SVN: r238393

PR c++/71513 - alignas on member enum in template

* pt.c (tsubst_attributes): Fix loop logic.

From-SVN: r238392

PR c++/71604 - type definition in range-based for

PR c++/54430
* parser.c (cp_parser_range_for): Modify IDENTIFIER_BINDING directly.
(cp_parser_simple_declaration): Diagnose type definition in
for-range-declaration.

From-SVN: r238391

PR c++/71711 - mangle C++1z fold-expressions.

* operators.def: Add *_FOLD_EXPR.
* cp-tree.h (FOLD_EXPR_P): Parenthesize.
* mangle.c (write_expression): Handle fold-expressions.
* pt.c (tsubst_unary_left_fold, tsubst_binary_left_fold)
(tsubst_unary_right_fold, tsubst_binary_right_fold): Handle
partial instantiation.

From-SVN: r238390

PR c++/71814 - mangling sizeof... (sP and sZ)

gcc/cp/
* mangle.c (write_expression): Handle sizeof... an argument pack.
libiberty/
* cp-demangle.c (cplus_demangle_operators): Add sP and sZ.
(d_print_comp_inner): Handle them.
(d_template_args_1): Split out from d_template_args.
(d_args_length): New.

From-SVN: r238389

Update -fabi-version default to 11.

* c-opts.c (c_common_post_options): Update -fabi-version default to 11.

From-SVN: r238388

PR c++/71718 - infinite recursion and alias template

* pt.c (push_tinst_level_loc): Set at_eof before fatal_error.

From-SVN: r238387

PR c++/70824 - initializer_list in template

* init.c (constant_value_1): Don't instantiated DECL_INITIAL of
artificial variables.

From-SVN: r238386

Implement C++17 P0025 clamp.

2016-07-15  Edward Smith-Rowland  <3dw4rd@verizon.net>

Implement C++17 P0025 clamp.
* include/bits/algorithmfwd.h: Declare clamp overloads.
* include/bits/stl_algo.h: Implement clamp.  Feature __cpp_lib_clamp.
* testsuite/25_algorithms/clamp/1.cc: New test.
* testsuite/25_algorithms/clamp/2.cc: New test.
* testsuite/25_algorithms/clamp/constexpr.cc: New test.
* testsuite/25_algorithms/clamp/requirements/explicit_instantiation/
1.cc: New test.
* testsuite/25_algorithms/clamp/requirements/explicit_instantiation/
pod.cc: New test.

From-SVN: r238383

Typo.

From-SVN: r238382

predicates.md (const_m255_to_m1_operand): New.

* config/avr/predicates.md (const_m255_to_m1_operand): New.
* config/avr/constraints.md (Cn8, Ca1, Co1, Yx2): New constraints.
* config/avr/avr.md (add<mode>3) <ALL1>: Make "r,0,r" more
expensive.
(*cmphi.zero-extend.0, *cmphi.zero-extend.1)
(*usum_widenqihi3, *udiff_widenqihi3)
(*addhi3_zero_extend.const): New combiner insns.
(andqi3, iorqi3): Provide "l" (NO_LD_REGS) alternative if
just 1 bit is affected.
* config/avr/avr.c (avr_out_bitop) <QImode>: Don't access xop[3].
(avr_out_compare) [EQ,NE]: Tweak comparing d-regs against -1.

From-SVN: r238381

Fix up ChangeLog entry.

From-SVN: r238378

scev-8.c: Update test string.

gcc/testsuite
* gcc.dg/tree-ssa/scev-8.c: Update test string.

From-SVN: r238377

c-parser.c (c_parser_oacc_declare): Don't scan for GOMP_MAP_POINTER.

gcc/c/
* c-parser.c (c_parser_oacc_declare): Don't scan for
GOMP_MAP_POINTER.
* c-typeck.c (handle_omp_array_sections): Mark data clauses with
GOMP_MAP_FORCE_{PRESENT,TO,FROM,TOFROM} as potentially having
zero-length subarrays.

gcc/cp/
* parser.c (cp_parser_oacc_declare): Don't scan for
GOMP_MAP_POINTER.
* semantics.c (handle_omp_array_sections): Mark data clauses with
GOMP_MAP_FORCE_{PRESENT,TO,FROM,TOFROM} as potentially having
zero-length subarrays.

gcc/
* omp-low.c (lower_omp_target): Mark data clauses with
GOMP_MAP_FORCE_{PRESENT,TO,FROM,TOFROM} as potentially having
zero-length subarrays.

libgomp/
* testsuite/libgomp.oacc-c-c++-common/zero_length_subarrays.c: New
test.

From-SVN: r238376

openmp.c (gfc_match_omp_clauses): Scan for clause vector_length before vector.

gcc/fortran/
* openmp.c (gfc_match_omp_clauses): Scan for clause vector_length
before vector.

gcc/testsuite/
* gfortran.dg/goacc/vector_length.f90: New test.

From-SVN: r238375

re PR tree-optimization/71881 (ICE on valid code at -O3 with -g enabled on x86_64-linux-gnu: cannot update SSA form)

2016-07-15  Richard Biener  <rguenther@suse.de>

PR tree-optimization/71881
* tree-loop-distribution.c (destroy_loop): Remove blocks in
reverse DOM order to make debug temp generation happy.

* gcc.dg/torture/pr71881.c: New testcase.

From-SVN: r238374

re PR tree-optimization/71887 (wrong code (SIGFPE) at -O1 and above on x86_64-linux-gnu (in both 32-bit and 64-bit modes))

2016-07-15  Richard Biener  <rguenther@suse.de>

PR tree-optimization/71887
* tree-ssa-phiopt.c (absorbing_element_p): Add rhs arg and
verify it is not zero for division / modulo handling.
(value_replacement): Adjust.

* gcc.dg/torture/pr71887.c: New testcase.

From-SVN: r238373

[PATCH/AARCH64] Add rtx_costs routine for vulcan.

gcc/ChangeLog:

2016-07-15  Virendra Pathak  <virendra.pathak@broadcom.com>
    Julian Brown  <julian@codesourcery.com>

* config/aarch64/aarch64-cores.def: Update vulcan COSTS.
* config/aarch64/aarch64-cost-tables.h
(vulcan_extra_costs): New variable.
* config/aarch64/aarch64.c
(vulcan_addrcost_table): Likewise.
(vulcan_regmove_cost): Likewise.
(vulcan_vector_cost): Likewise.
(vulcan_branch_cost): Likewise.
(vulcan_tunings): Likewise.

Co-Authored-By: Julian Brown <julian@codesourcery.com>
From-SVN: r238372

handle undefined extern vars in output_in_order

* cgraphunit.c (cgraph_order_sort_kind): New entry ORDER_VAR_UNDEF.
(output_in_order): Loop over undefined variables too.  Output them
via assemble_undefined_decl.  Skip variables that correspond to hard
registers or have value-exprs.
* varpool.c (symbol_table::output_variables): Handle undefined
variables together with defined ones.

From-SVN: r238371

tree-ssa-pre.c (get_representative_for): Make sure to return the value number of SSA names.

2016-07-15  Richard Biener  <rguenther@suse.de>

* tree-ssa-pre.c (get_representative_for): Make sure to return
the value number of SSA names.
(phi_translate_1): get_representative_for cannot return NULL.
(do_pre_regular_insertion): Remove redundant call to
fully_constant_expression.
(do_pre_partial_partial_insertion): Likewise.

From-SVN: r238370

re PR c/71858 (Surprising suggestions for misspellings)

PR c/71858
* c-common.h (enum lookup_name_fuzzy_kind): Add
FUZZY_LOOKUP_FUNCTION_NAME.

* c-decl.c (implicit_decl_warning): Use FUZZY_LOOKUP_FUNCTION_NAME
instead of FUZZY_LOOKUP_NAME.
(lookup_name_fuzzy): For FUZZY_LOOKUP_FUNCTION_NAME consider
FUNCTION_DECLs, {VAR,PARM}_DECLs function pointers and macros.

* gcc.dg/spellcheck-identifiers-3.c: New test.

From-SVN: r238369

re PR fortran/71807 (Internal compiler error with NULL() reference in structure constructor)

gcc/fortran/ChangeLog:

2016-07-15  Andre Vehreschild  <vehre@gcc.gnu.org>

PR fortran/71807
* trans-expr.c (gfc_trans_subcomponent_assign): Special casing
when allocatable component is set to null() in initializer.

gcc/testsuite/ChangeLog:

2016-07-15  Andre Vehreschild  <vehre@gcc.gnu.org>

PR fortran/71807
* gfortran.dg/null_9.f90: New test.

From-SVN: r238368

tree-scalar-evolution.c (simple_iv_with_niters): New funcion.

* tree-scalar-evolution.c (simple_iv_with_niters): New funcion.
(derive_simple_iv_with_niters): New function.
(simple_iv): Rewrite using simple_iv_with_niters.
* tree-scalar-evolution.h (simple_iv_with_niters): New decl.
* tree-ssa-loop-niter.c (number_of_iterations_exit_assumptions): New
function.
(number_of_iterations_exit): Rewrite using above function.
* tree-ssa-loop-niter.h (number_of_iterations_exit_assumptions): New
Decl.

gcc/testsuite
* gcc.dg/tree-ssa/loop-41.c: New test.

From-SVN: r238367

re PR tree-optimization/71347 (Performance drop after r235513 on x86-64 in 32-bit mode.)

gcc/testsuite
PR tree-optimization/71347
* gcc.dg/tree-ssa/pr71347.c: XFAIL on ia64, arm, m68k and sparc.

From-SVN: r238366

Fix the constraints for any's assignment operator template to properly
reject assignment from a non-copyable lvalue.

From-SVN: r238365

i386.c (ix86_builtin_vectorization_cost): Adjust vec_construct cost.

2016-07-15  Richard Biener  <rguenther@suse.de>

* config/i386/i386.c (ix86_builtin_vectorization_cost): Adjust
vec_construct cost.

From-SVN: r238364

Daily bump.

From-SVN: r238362

re PR fortran/29819 (Error/warning message should ignore comments for "1" in %C output)

2016-07-14  Steven G. Kargl  <kargl@gcc.gnu.org>

PR fortran/29819
* parse.c (parse_contained): Use proper locus.

2016-07-14  Steven G. Kargl  <kargl@gcc.gnu.org>

PR fortran/29819
* gfortran.dg/bind_c_usage_9.f03: Move dg-error.
* gfortran.dg/contains.f90: Ditto.
* gfortran.dg/contains_empty_1.f03: Ditto.
* gfortran.dg/submodule_3.f08: Ditto.

From-SVN: r238354

re PR testsuite/71865 (test case gcc.dg/diagnostic-token-ranges.c fails starting with r237714)

PR testsuite/71865
* gcc.dg/diagnostic-token-ranges.c: Add -std=c11 to dg-options.
(wide_string_literal_in_asm): Use __asm instead of asm, adjust
expected diagnostics.

From-SVN: r238353

re PR c/71858 (Surprising suggestions for misspellings)

PR c/71858
* c-decl.c (lookup_name_fuzzy): Ignore binding->invisible.

* gcc.dg/spellcheck-identifiers.c (snprintf): Declare.
* gcc.dg/spellcheck-identifiers-2.c: New test.
* gcc.dg/diagnostic-token-ranges.c (nanl): Declare.
* c-c++-common/attributes-1.c: Adjust dg-prune-output.

From-SVN: r238352

re PR tree-optimization/71872 (ICE in inchash::add_expr, at tree.c:7782 - OEP_ADDRESS_OF asserted for ADDR_EXPR applied to constant)

PR tree-optimization/71872
* tree-data-ref.c (get_references_in_stmt): Ignore references
with is_gimple_constant get_base_address.

* gcc.c-torture/compile/pr71872.c: New test.

From-SVN: r238351

arm.h (TARGET_HAVE_LDACQ): Enable for ARMv8-M Mainline.

2016-07-14  Thomas Preud'homme  <thomas.preudhomme@arm.com>

    gcc/
    * config/arm/arm.h (TARGET_HAVE_LDACQ): Enable for ARMv8-M Mainline.
    (TARGET_HAVE_LDACQD): New macro.
    * config/arm/sync.md (atomic_loaddi): Use TARGET_HAVE_LDACQD rather
    than TARGET_HAVE_LDACQ.
    (arm_load_acquire_exclusivedi): Likewise.
    (arm_store_release_exclusivedi): Likewise.

    libgcc/
    * gcc.target/arm/atomic-comp-swap-release-acquire.c: Rename into ...
    * gcc.target/arm/atomic-comp-swap-release-acquire-1.c: This.
    * gcc.target/arm/atomic-op-acq_rel.c: Rename into ...
    * gcc.target/arm/atomic-op-acq_rel-1.c: This.
    * gcc.target/arm/atomic-op-acquire.c: Rename into ...
    * gcc.target/arm/atomic-op-acquire-1.c: This.
    * gcc.target/arm/atomic-op-char.c: Rename into ...
    * gcc.target/arm/atomic-op-char-1.c: This.
    * gcc.target/arm/atomic-op-consume.c: Rename into ...
    * gcc.target/arm/atomic-op-consume-1.c: This.
    * gcc.target/arm/atomic-op-int.c: Rename into ...
    * gcc.target/arm/atomic-op-int-1.c: This.
    * gcc.target/arm/atomic-op-relaxed.c: Rename into ...
    * gcc.target/arm/atomic-op-relaxed-1.c: This.
    * gcc.target/arm/atomic-op-release.c: Rename into ...
    * gcc.target/arm/atomic-op-release-1.c: This.
    * gcc.target/arm/atomic-op-seq_cst.c: Rename into ...
    * gcc.target/arm/atomic-op-seq_cst-1.c: This.
    * gcc.target/arm/atomic-op-short.c: Rename into ...
    * gcc.target/arm/atomic-op-short-1.c: This.
    * gcc.target/arm/atomic-comp-swap-release-acquire-2.c: New test.
    * gcc.target/arm/atomic-op-acq_rel-2.c: Likewise.
    * gcc.target/arm/atomic-op-acquire-2.c: Likewise.
    * gcc.target/arm/atomic-op-char-2.c: Likewise.
    * gcc.target/arm/atomic-op-consume-2.c: Likewise.
    * gcc.target/arm/atomic-op-int-2.c: Likewise.
    * gcc.target/arm/atomic-op-relaxed-2.c: Likewise.
    * gcc.target/arm/atomic-op-release-2.c: Likewise.
    * gcc.target/arm/atomic-op-seq_cst-2.c: Likewise.
    * gcc.target/arm/atomic-op-short-2.c: Likewise.

From-SVN: r238348

re PR fortran/70842 (internal compiler error with character members within a polymorphic pointer)

gcc/testsuite/ChangeLog:

2016-07-14  Andre Vehreschild  <vehre@gcc.gnu.org>

PR fortran/70842
* gfortran.dg/select_type_35.f03: New test.

gcc/fortran/ChangeLog:

2016-07-14  Andre Vehreschild  <vehre@gcc.gnu.org>

PR fortran/70842
* simplify.c (gfc_simplify_len): Only for unlimited polymorphic
types replace the expression's _data ref with a _len ref.

From-SVN: r238347

re PR rtl-optimization/71878 (ICE in cselib_record_set)

2016-07-14  Thomas Preud'homme  <thomas.preudhomme@arm.com>

    gcc/
    PR rtl-optimization/71878
    * lra-constraints.c (match_reload): Pass information about other
    output operands.  Create new unique register value if matching input
    operand shares same register value as output operand being considered.
    (curr_insn_transform): Record output operands already processed.

From-SVN: r238346

Include header for std::__parallel::search

* include/experimental/functional: Include <parallel/algorithm> in
Parallel Mode.

From-SVN: r238342

Add header for std::search to testcase

* testsuite/experimental/functional/searchers.cc: Include <algorithm>
for std::search.

From-SVN: r238341

[vectorizer][2/2] Hook up mult synthesis logic into vectorisation of mult-by-constant

PR target/65951
PR tree-optimization/70923
* tree-vect-patterns.c: Include mult-synthesis.h.
(target_supports_mult_synth_alg): New function.
(synth_lshift_by_additions): Likewise.
(apply_binop_and_append_stmt): Likewise.
(vect_synth_mult_by_constant): Likewise.
(target_has_vecop_for_code): Likewise.
(vect_recog_mult_pattern): Use above functions to synthesize vector
multiplication by integer constants.

* gcc.dg/vect/vect-mult-const-pattern-1.c: New test.
* gcc.dg/vect/vect-mult-const-pattern-2.c: Likewise.
* gcc.dg/vect/pr65951.c: Likewise.
* gcc.dg/vect/vect-iv-9.c: Remove ! vect_int_mult-specific scan.

From-SVN: r238340

[RS6000] Disparage altivec_mov gpr alternatives

and fix constraint '*' placement.

* gcc/config/rs6000/altivec.md (altivec_mov<mode>): Disparage
gpr alternatives.  Correct '*' placement on Y,r alternative.
Add '*' on operand 1 of r,r alternative.

From-SVN: r238339

[1/2] Move choose_mult_variant declaration and dependent declarations to expmed.h

* expmed.c (mult_variant, choose_mult_variant): Move declaration to...
* expmed.h: ... Here.

From-SVN: r238337

gimple.h (stmt_can_terminate_bb_p): New function.

* gimple.h (stmt_can_terminate_bb_p): New function.
* tree-cfg.c (need_fake_edge_p): Rename to ...
(stmt_can_terminate_bb_p): ... this; return true if stmt can
throw external; handle const and pure calls.
* tree-ssa-loop-niter.c (loop_only_exit_p): Use it.

From-SVN: r238336

pr60226.c: Expect maximum object file alignment error on AIX.

        * c-c++-common/pr60226.c: Expect maximum object file alignment
        error on AIX.

From-SVN: r238335

re PR tree-optimization/71866 (gcc locks up after fix for PR70159)

2016-07-14  Richard Biener  <rguenther@suse.de>

PR tree-optimization/71866
* tree-ssa-pre.c (get_constant_for_value_id): Remove.
(do_hoist_insertion): Avoid endless recursion when we
didn't insert anything because we managed to simplify
things down to a constant or SSA name.
(fully_constant_expression): Re-write in terms of ...
* tree-ssa-sccvn.h (vn_nary_simplify): ... this.  Declare.
* tree-ssa-sccvn.c (vn_nary_simplify): New wrapper around
vn_nary_build_or_lookup_1.
(vn_nary_build_or_lookup_1): Added flag and renamed from ...
(vn_nary_build_or_lookup): ... this which now wraps it.

* gcc.dg/torture/pr71866.c: New testcase.

From-SVN: r238334

decl.c (gnat_to_gnu_entity): Also use the void pointer type if the designated type is incomplete and has no...

* gcc-interface/decl.c (gnat_to_gnu_entity) <E_Access_Type>: Also use
the void pointer type if the designated type is incomplete and has no
full view in LTO mode.
<E_Access_Protected_Subprogram_Type>: Adjust comment.
<E_Incomplete_Type>: Likewise.
* gcc-interface/trans.c (Call_to_gnu): Do not convert to the type of
the actual if it is a dummy type.

From-SVN: r238333

Improve doxygen comments for allocators in containers

PR libstdc++/70716
* include/bits/forward_list.h (forward_list): Update doxygen comments
to reflect allocator propagation semantics. Remove ambiguous
statements about data being lost.
* include/bits/stl_deque.h (deque): Likewise.
* include/bits/stl_list.h (list): Likewise.
* include/bits/stl_map.h (map): Likewise.
* include/bits/stl_multimap.h (multimap): Likewise.
* include/bits/stl_multiset.h (multiset): Likewise.
* include/bits/stl_set.h (set): Likewise.
* include/bits/stl_vector.h (vector): Likewise.
* include/bits/unordered_map.h (unordered_map, unordered_multimap):
Likewise.
* include/bits/unordered_set.h (unordered_set, unordered_multiset):
Likewise.

From-SVN: r238332

pr42574.c: Add missing target keyword for the dg-do selector and enclose boolean...

2016-07-14  Thomas Preud'homme  <thomas.preudhomme@arm.com>

    gcc/testsuite/
    * gcc.target/arm/pr42574.c: Add missing target keyword for the dg-do
    selector and enclose boolean expression in curly braces.

From-SVN: r238331

Added forgotten PR testsuite/71489 reference.

From-SVN: r238330

Implement P0032R3, Homogeneous interface for variant, any and optional,

for the parts concerning any and optional.
* include/std/any (_Storage()): Make constexpr and have it
initialize _M_ptr.
(any()): Make constexpr.
(any(const any&)): Adjust.
(any(any&&)): Likewise.
(__any_constructible_t): New.
(any(_ValueType&&)): Constrain.
(any(in_place_type_t<_Tp>, _Args&&...)): New.
(any(in_place_type_t<_Tp>, initializer_list<_Up>, _Args&&...)):
Likewise.
(~any()): Adjust.
(operator=(const any&)): Likewise.
(operator=(any&&)): Likewise.
(operator=(_ValueType&&)): Constrain.
(emplace(_Args&&...)): New.
(emplace(initializer_list<_Up>, _Args&&...)): Likewise.
(clear()): Remove.
(reset()): New.
(swap(any&)): Adjust.
(empty()): Remove.
(has_value()): New.
(type()): Adjust.
(_Manager_internal::_S_create(_Storage&, _Args&&...)): New.
(_Manager_external::_S_create(_Storage&, _Args&&...)): Likewise.
(make_any(_Args&&...)): Likewise.
(make_any(initializer_list<_Up>, _Args&&...)): Likewise.
* include/std/optional (in_place_t, in_place): Remove.
(bad_optional_access): Add a comment referring to LEWG 72.
(emplace(_Args&&...)): Constrain.
(has_value()): New.
(reset()): Likewise.
(make_optional(_Args&&...)): Likewise.
(make_optional(initializer_list<_Up>, _Args&&...)): Likewise.
* include/std/utility (in_place_tag): New.
(__in_place, __in_place_type, __in_place_index): Likewise.
(in_place_t, in_place_type_t, in_place_index_t): Likewise.
(in_place(__in_place*)): Likewise.
(in_place(__in_place_type<_Tp>*)): Likewise.
(in_place(__in_place_index<_Idx>*)): Likewise.
* testsuite/20_util/any/assign/1.cc: Adjust.
* testsuite/20_util/any/assign/emplace.cc: New.
* testsuite/20_util/any/assign/self.cc: Adjust.
* testsuite/20_util/any/cons/1.cc: Likewise.
* testsuite/20_util/any/cons/in_place.cc: New.
* testsuite/20_util/any/make_any.cc: Likewise.
* testsuite/20_util/any/misc/any_cast_neg.cc: Adjust.
* testsuite/20_util/any/misc/swap.cc: Likewise.
* testsuite/20_util/any/modifiers/1.cc: Likewise.
* testsuite/20_util/any/requirements.cc: New.
* testsuite/20_util/in_place/requirements.cc: Likewise.
* testsuite/20_util/optional/constexpr/in_place.cc: Adjust.
* testsuite/20_util/optional/in_place.cc: Likewise.
* testsuite/20_util/optional/make_optional.cc: Add tests for
the new overloads of make_optional.

From-SVN: r238329

Fix ChangeLog.

From-SVN: r238328

[RS6000] -mno-vsx should force -mno-power9-dform

PR target/71733
gcc/
* config/rs6000/rs6000.c (rs6000_option_override_internal): Deal
with p9_vector override before power9-dform override.
gcc/testsuite/
* gcc.target/powerpc/p9-novsx.c: New.

From-SVN: r238326

Some fixes for profile test cases for autofdo

This fixes some basic issues with the profile test cases with autofdo.

- Disable checking for value transformations that autofdo does not
  support.
- Disable checking for fixed hit counts which autofdo does not support
- Enable dumping of afdo log file and check right log file.
- Increase run time of test cases to 1M iterations because autofdo needs
  a few samples to make sense of a program. The test case don't run
  noticeable slower with that.

There are still failures unfortunately, especially the indirect call
transformations do not trigger because autofdo thinks they are not hot.
This can be addressed later.

gcc/testsuite/:

2016-07-13  Andi Kleen  <ak@linux.intel.com>

* g++.dg/tree-prof/indir-call-prof.C: Basic fixes for autofdo.
* g++.dg/tree-prof/morefunc.C: Dito.
* g++.dg/tree-prof/pr35545.C: Dito.
* g++.dg/tree-prof/reorder.C: Dito.
* gcc.dg/tree-prof/20050826-2.c: Dito.
* gcc.dg/tree-prof/cmpsf-1.c: Dito.
* gcc.dg/tree-prof/cold_partition_label.c: Dito.
* gcc.dg/tree-prof/ic-misattribution-1.c: Dito.
* gcc.dg/tree-prof/indir-call-prof.c: Dito.
* gcc.dg/tree-prof/inliner-1.c: Dito.
* gcc.dg/tree-prof/merge_block.c: Dito.
* gcc.dg/tree-prof/stringop-1.c: Dito.
* gcc.dg/tree-prof/stringop-2.c: Dito.
* gcc.dg/tree-prof/switch-case-1.c: Dito.
* gcc.dg/tree-prof/switch-case-2.c: Dito.
* gcc.dg/tree-prof/time-profiler-1.c: Dito.
* gcc.dg/tree-prof/time-profiler-2.c: Dito.
* gcc.dg/tree-prof/update-loopch.c: Dito.
* gcc.dg/tree-prof/val-prof-1.c: Dito.
* gcc.dg/tree-prof/val-prof-2.c: Dito.
* gcc.dg/tree-prof/val-prof-3.c: Dito.
* gcc.dg/tree-prof/val-prof-4.c: Dito.
* gcc.dg/tree-prof/val-prof-5.c: Dito.
* gcc.dg/tree-prof/val-prof-6.c: Dito.
* gcc.dg/tree-prof/val-prof-7.c: Dito.

From-SVN: r238325

Clean up imports files in test suite

autofdo create_gcov creates an extra .imports file. Always remove that
too when running an autofdo test case.

gcc/testsuite/:

2016-07-13  Andi Kleen  <ak@linux.intel.com>

* lib/profopt.exp (profopt-execute): Remove .imports files.

From-SVN: r238324

Always print gcc-auto-profile line in dump file

not just when verbose. This ensures all command lines needed to
reproduce the test case are always logged

gcc/testsuite/:

2016-07-13  Andi Kleen  <ak@linux.intel.com>

* lib/profopt.exp (profopt-execute): Always log profiler
command line.

From-SVN: r238323

Add dg-final-scan-autofdo and dg-final-scan-not-autofdo

Autofdo outputs to different dump files and doesn't support some
transformation that normal profiling. Add dg-final-scan-autofdo
and dg-final-scan-not-autofdo statements to the test suite
so that the test cases can hande those cases separately.

gcc/testsuite/:

2016-07-13  Andi Kleen  <ak@linux.intel.com>

* lib/profopt.exp (dg-final-scan-autofdo,
dg-final-scan-not-autofdo): New functions.

From-SVN: r238322

Don't run instrumented value profiler changes with afdo

The pass to transform gimple based on value profiling runs with autofdo
on, but currently every transformation fails. For indirect calls autofdo
does it on its own, and it doesn't suppport other value profiling. So don't
run this pass when autofdo is active. This also avoids bogus
dump file entries.

gcc/:

2016-07-13  Andi Kleen  <ak@linux.intel.com>

* value-prof.c (gimple_value_profile_transformations): Don't run
when auto_profile is on.

From-SVN: r238321

Print indirect call changes in afdo dump file

Print some information about indirect call promotions in the afdo dump
file. Do it in the same format as the instrumented profiler so that
the test suite can match on it.

gcc/:

2016-07-13  Andi Kleen  <ak@linux.intel.com>

* auto-profile.c (update_inlined_ind_target,
afdo_indirect_call): Print information to dump file.

From-SVN: r238320

Daily bump.

From-SVN: r238319

genrecog.c (special_predicate_operand_p): New function.

* genrecog.c (special_predicate_operand_p): New function.
(predicate_name): Move function.
(validate_pattern): Don't warn about missing mode for all
define_special_predicate predicates.

From-SVN: r238314

re PR c++/70926 (Libiberty Demangler segfaults (5))

PR c++/70926
* cplus-dem.c: Handle large values and overflow when demangling
length variables.
(demangle_template_value_parm): Read only until end of mangled string.
(do_hpacc_template_literal): Likewise.
(do_type): Handle overflow when demangling array indices.

From-SVN: r238313

Fix return type handling for Proxy classes

libjava/

* java/lang/reflect/natVMProxy.cc (unbox): Use ffi_arg for
integer return types smaller than a word.
* testsuite/libjava.jar/ReturnInvocationHandler.java: New file.
* testsuite/libjava.jar/ReturnProxyTest.jar: Likewise.
* testsuite/libjava.jar/ReturnProxyTest.java: Likewise.
* testsuite/libjava.jar/ReturnProxyTest.out: Likewise.
* testsuite/libjava.jar/ReturnProxyTest.xfail: Likewise.
* testsuite/libjava.jar/ReturnTypes.java: Likewise.
* testsuite/libjava.jar/ReturnTypesImpl.java: Likewise.

From-SVN: r238312

Fix the use of FFI closures in the java interpreter for integer returns

libjava/

* interpret-run.cc: Use ffi_arg for FFI integer return types.
* testsuite/libjava.jar/arraysort.java: New file.
* testsuite/libjava.jar/arraysort.jar: New file.
* testsuite/libjava.jar/arraysort.out: New file.
* testsuite/libjava.jar/arraysort.xfail: New file.

From-SVN: r238311

libstdc++/71856 Define _GLIBCXX_PARALLEL_ASSERTIONS

PR libstdc++/71856
* include/bits/c++config (_GLIBCXX_ASSERTIONS): Define to 1 not empty.
* include/parallel/compiletime_settings.h (_GLIBCXX_ASSERTIONS):
Rename to _GLIBCXX_PARALLEL_ASSERTIONS and make default value depend
on _GLIBCXX_ASSERTIONS.
* include/parallel/balanced_quicksort.h: Rename _GLIBCXX_ASSERTIONS.
Include <unistd.h> for sleep.
* include/parallel/losertree.h: Rename _GLIBCXX_ASSERTIONS.
* include/parallel/merge.h: Likewise.
* include/parallel/multiway_merge.h: Likewise.
* include/parallel/partition.h: Likewise.
* include/parallel/queue.h: Likewise.
* include/parallel/sort.h: Likewise.
* testsuite/25_algorithms/headers/algorithm/
parallel_algorithm_assert.cc: New.

From-SVN: r238307

Make basic_string::replace forward to different overload

* include/bits/basic_string.h [_GLIBCXX_USE_CXX11_ABI]
(replace(__const_iterator, __const_iterator, initializer_list<C>)):
Forward to different overload.

From-SVN: r238306

Make __allocated_ptr::_S_raw_ptr static

* include/bits/allocated_ptr.h (__allocated_ptr::_S_raw_ptr): Make
static.

From-SVN: r238305

Revert "Add non-const overload of std::string::data()"

This reverts r238291 aka 4300aa2087d90cd2d55852fa38f082a89bc7e72d.

From-SVN: r238303

tree-vect-data-refs.c (vect_no_alias_p): New function.

* tree-vect-data-refs.c (vect_no_alias_p): New function.
(vect_prune_runtime_alias_test_list): Call vect_no_alias_p to
resolve alias checks which are known at compilation time.
Truncate vector LOOP_VINFO_MAY_ALIAS_DDRS(loop_vinfo) if all
alias checks are resolved.  Move dump info for too many runtime
alias checks to here...
* tree-vect-loop.c (vect_analyze_loop_2): ...From here.

gcc/testsuite
* gcc.dg/vect/vect-35-big-array.c: Refine comment and test.
* gcc.dg/vect/vect-35.c: Ditto.
* gcc.dg/vect/vect-alias-check-2.c: New test.

From-SVN: r238301

Fix Fortran DO loop fallback

* gfortran.dg/ldist-1.f90: Revert change introduces in r238114.
* gfortran.dg/vect/pr62283.f: Add -fno-ipa-icf to additional
options.
* gfortran.dg/pr42108.f90: Update expected dump scan.

From-SVN: r238300

re PR tree-optimization/24574 (a!=0?a/10:0 is not reduced to a/10)

2016-07-13  Richard Biener  <rguenther@suse.de>

PR tree-optimization/24574
* tree-ssa-phiopt.c (absorbing_element_p): Pass in argument
position and add shift, rotate, divison and modulo support
for left zero.
(value_replacement): Pass in argument position to absorbing_element_p.

* gcc.dg/pr24574.c: New testcase.

From-SVN: r238299

Fix some bogus testsuite failures for avr.

gcc/testsuite/

        * c-c++-common/Wduplicated-cond-3.c (fn10): Use smaller
        const literal.
        * c-c++-common/builtin-arith-overflow-2.c: Skip for avr.
        * c-c++-common/pr68833-1.c: Require int32plus.
        * gcc.dg/ipa/pr63551.c: Likewise.
        * gcc.dg/ipa/pr63595.c: Require ptr32plus.
        * gcc.dg/ipa/pr64041.c: Require int32plus.

From-SVN: r238294

Fix ChangeLog.

From-SVN: r238293

Implement P0307R2, Making Optional Greater Equal Again.

* include/experimental/optional (_Has_addressof): Fix the comment.
* include/std/optional (_Has_addressof): Likewise.
(operator=(_Up&&)): Constrain.
(operator=(const optional<_Up>&)): Likewise.
(operator=(optional<_Up>&&)): Likewise.
(__optional_relop_t): New.
(operator==(const optional<_Tp>&, const optional<_Tp>&)): Constrain.
(operator!=(const optional<_Tp>&, const optional<_Tp>&)):
Constrain and make transparent.
(operator<(const optional<_Tp>&, const optional<_Tp>&)): Constrain.
(operator>(const optional<_Tp>&, const optional<_Tp>&)):
Constrain and make transparent.
(operator<=(const optional<_Tp>&, const optional<_Tp>&)): Likewise.
(operator>=(const optional<_Tp>&, const optional<_Tp>&)): Likewise.
(operator==(const optional<_Tp>&, const _Tp&): Constrain.
(operator==(const _Tp&, const optional<_Tp>&)): Likewise.
(operator!=(const optional<_Tp>&, _Tp const&)):
Constrain and make transparent.
(operator!=(const _Tp&, const optional<_Tp>&)): Likewise.
(operator<(const optional<_Tp>&, const _Tp&)): Constrain.
(operator<(const _Tp&, const optional<_Tp>&)): Likewise.
(operator>(const optional<_Tp>&, const _Tp&)):
Constrain and make transparent.
(operator>(const _Tp&, const optional<_Tp>&)): Likewise.
(operator<=(const optional<_Tp>&, const _Tp&)): Likewise.
(operator<=(const _Tp&, const optional<_Tp>&)): Likewise.
(operator>=(const optional<_Tp>&, const _Tp&)): Likewise.
(operator>=(const _Tp&, const optional<_Tp>&)): Likewise.
* testsuite/20_util/optional/constexpr/relops/2.cc: Adjust.
* testsuite/20_util/optional/constexpr/relops/4.cc: Likewise.
* testsuite/20_util/optional/relops/1.cc: Likewise.
* testsuite/20_util/optional/relops/2.cc: Likewise.
* testsuite/20_util/optional/relops/3.cc: Likewise.
* testsuite/20_util/optional/relops/4.cc: Likewise.
* testsuite/20_util/optional/requirements.cc: Add tests to verify
that optional's relops are transparent and don't synthesize
operators. Also test that assignment sfinaes.

From-SVN: r238292

Add non-const overload of std::string::data()

Also fix confusion between pointer and _CharT*, so that allocators with
fancy pointers work correctly.

* include/bits/basic_string.h [_GLIBCXX_USE_CXX11_ABI] (_M_c_str):
New function.
(_M_disjunct, basic_string(const basic_string&, size_t)): Use data()
instead of _M_data().
(basic_string(const basic_string&, size_t, size_t, const _Alloc&)):
Likewise.
(append(const basic_string&)): Likewise.
(append(const basic_string&, size_type, size_type)): Likewise.
(assign(const basic_string&, size_type, size_type)): Likewise.
(insert(size_type, const basic_string&)): Likewise.
(insert(size_type, const basic_string&, size_type, size_type)):
Likewise.
(replace(size_type, size_type, const basic_string&, size_type,
size_type)): Likewise.
(replace(__const_iterator, __const_iterator, const basic_string&)):
Likewise.
(c_str(), data()): Use c_str() instead of _M_data().
(data()): Add non-const overload as per LWG 2391 and P0272R1.
(compare(const basic_string&)): Use data() instead of _M_data().
[!_GLIBCXX_USE_CXX11_ABI] (data()): Add non-const overload.
* include/bits/basic_string.tcc [_GLIBCXX_USE_CXX11_ABI] (_M_mutate):
Pass raw pointers to _S_copy.
(_M_erase, _M_replace_aux): Pass raw pointers to _S_move and
_S_assign.
(find(const _CharT*, size_type, size_type)): Use data instead of
_M_data().
* testsuite/21_strings/basic_string/allocator/char/ext_ptr.cc: New.
* testsuite/21_strings/basic_string/operations/data/char/2.cc: New.
* testsuite/21_strings/basic_string/operations/data/wchar_t/2.cc: New.

From-SVN: r238291

re PR ipa/71633 ([CHKP] internal compiler error: in inline_call)

gcc/

PR ipa/71633
* ipa-inline-transform.c (inline_call): Support
instrumented thunks.

gcc/testsuite/

PR ipa/71633
* g++.dg/pr71633.C: New test.

From-SVN: r238290

arm.h (TARGET_HAVE_CBZ): Define.

2016-07-13  Thomas Preud'homme  <thomas.preudhomme@arm.com>

    gcc/
    * config/arm/arm.h (TARGET_HAVE_CBZ): Define.
    (TARGET_IDIV): Set for all Thumb targets provided they have hardware
    divide feature.
    * config/arm/arm.md (divsi3): New unpredicable alternative for ARMv8-M
    Baseline.  Make initial alternative TARGET_32BIT only.
    (udivsi3): Likewise.
    * config/arm/thumb1.md (thumb1_cbz): New define_insn.
    * doc/sourcebuild.texi (arm_thumb1_cbz_ok): Document new effective
    target.

    gcc/testsuite/
    * lib/target-supports.exp (check_effective_target_arm_thumb1_cbz_ok):
    Add new arm_thumb1_cbz_ok effective target.
    * gcc.target/arm/cbz.c: New test.

From-SVN: r238289

arm.h (TARGET_HAVE_MOVT): Include ARMv8-M as having MOVT.

2016-07-13  Thomas Preud'homme  <thomas.preudhomme@arm.com>

    gcc/
    * config/arm/arm.h (TARGET_HAVE_MOVT): Include ARMv8-M as having MOVT.
    * config/arm/arm.c (arm_arch_name): (const_ok_for_op): Check MOVT/MOVW
    availability with TARGET_HAVE_MOVT.
    (thumb_legitimate_constant_p): Strip the high part of a label_ref.
    (thumb1_rtx_costs): Also return 0 if setting a half word constant and
    MOVW is available and replace (unsigned HOST_WIDE_INT) INTVAL by
    UINTVAL.
    (thumb1_size_rtx_costs): Make set of half word constant also cost 1
    extra instruction if MOVW is available.  Use a cost variable
    incremented by COSTS_N_INSNS (1) when the condition match rather than
    returning an arithmetic expression based on COSTS_N_INSNS.  Make
    constant with bottom half word zero cost 2 instruction if MOVW is
    available.
    * config/arm/arm.md (define_attr "arch"): Add v8mb.
    (define_attr "arch_enabled"): Set to yes if arch value is v8mb and
    target is ARMv8-M Baseline.
    (arm_movt): New unpredicable alternative for ARMv8-M Baseline.
    (arm_movtas_ze): Likewise.
    * config/arm/thumb1.md (thumb1_movdi_insn): Add ARMv8-M Baseline only
    alternative for constants satisfying j constraint.
    (thumb1_movsi_insn): Likewise.
    (movsi splitter for K alternative): Tighten condition to not trigger
    if movt is available and j constraint is satisfied.
    (Pe immediate splitter): Likewise.
    (thumb1_movhi_insn): Add ARMv8-M Baseline only alternative for
    constant fitting in an halfword to use MOVW.
    * doc/sourcebuild.texi (arm_thumb1_movt_ok): Document new ARM
    effective target.

    gcc/testsuite/
    * lib/target-supports.exp (check_effective_target_arm_thumb1_movt_ok):
    Define effective target.
    * gcc.target/arm/pr42574.c: Require arm_thumb1_ok and
    !arm_thumb1_movt_ok to exclude ARMv8-M Baseline.
    * gcc.target/arm/movhi_movw.c: New test.
    * gcc.target/arm/movsi_movw.c: Likewise.
    * gcc.target/arm/movdi_movw.c: Likewise.

From-SVN: r238288

re PR tree-optimization/71104 (ICE: verify_ssa failed (with vfork /  error: definition in block 3 does not dominate use in block 7 ))

2016-07-13  Richard Biener  <rguenther@suse.de>

PR middle-end/71104
* gimplify.c (gimplify_modify_expr): Gimplify the RHS before
gimplifying the LHS.  Make sure to gimplify a returning twice
call LHS without using SSA names.

* gcc.dg/pr71104-1.c: New testcase.
* gcc.dg/pr71104-2.c: Likewise.

From-SVN: r238287

remove unnecessary calls to vec::release

There's no point in calling release () on an auto_vec just before it goes
out of scope.

gcc/ChangeLog:

2016-07-12  Trevor Saunders  <tbsaunde+gcc@tbsaunde.org>

* tree-data-ref.c (find_data_references_in_stmt): Remove
unnecessary call to vec::release.
(graphite_find_data_references_in_stmt): Likewise.
* tree-ssa-alias.c (nonoverlapping_component_refs_of_decl_p): Likewise.
* tree-vect-stmts.c (vectorizable_condition): Likewise.

From-SVN: r238286