ipa-cp.c (ipcp_store_vr_results): Avoid static local var zero.

* ipa-cp.c (ipcp_store_vr_results): Avoid static local
var zero.
* sreal.h (sreal::min, sreal::max): Avoid static local vars,
construct values without normalization.
* tree-ssa-sccvn.c (vn_reference_lookup_3): Don't initialize
static local lhs_ops to vNULL.
cp/
* name-lookup.c (store_bindings, store_class_bindings): Don't
initialize static local bindings_need_stored to vNULL.

From-SVN: r240408

[PATCH 7/17][ARM] Add FP16 data movement instructions.

gcc/
2016-09-23  Matthew Wahab  <matthew.wahab@arm.com>
    Jiong Wang <jiong.wang@arm.com>

* config/arm/arm.c (coproc_secondary_reload_class): Make HFmode
available when FP16 instructions are available.
(output_move_vfp): Add support for 16-bit data moves.
(arm_validize_comparison): Fix some white-space.  Support HFmode
by conversion to SFmode.
* config/arm/arm.md (truncdfhf2): Fix a comment.
(extendhfdf2): Likewise.
(cstorehf4): New.
(movsicc): Fix some white-space.
(movhfcc): New.
(movsfcc): Fix some white-space.
(*cmovhf): New.
* config/arm/vfp.md (*arm_movhi_vfp): Disable when VFP FP16
instructions are available.
(*thumb2_movhi_vfp): Likewise.
(*arm_movhi_fp16): New.
(*thumb2_movhi_fp16): New.
(*movhf_vfp_fp16): New.
(*movhf_vfp_neon): Disable when VFP FP16 instructions are
available.
(*movhf_vfp): Likewise.
(extendhfsf2): Enable when VFP FP16 instructions are available.
(truncsfhf2):  Enable when VFP FP16 instructions are available.

testsuite/
2016-09-23  Matthew Wahab  <matthew.wahab@arm.com>

* gcc.target/arm/armv8_2_fp16-move-1.c: New.
* gcc.target/arm/fp16-aapcs-1.c: Update expected output.

Co-Authored-By: Jiong Wang <jiong.wang@arm.com>
From-SVN: r240407

Fix typos: adress -> address

* config/s390/vx-builtins.md: Replace 'adress' with 'address'.

From-SVN: r240406

gcc-dg.exp (process-message): Support relative line number notation - .+4 or .-1 etc.

* lib/gcc-dg.exp (process-message): Support relative line number
notation - .+4 or .-1 etc.
* gcc.dg/dg-test-1.c: New test.

From-SVN: r240405

[PATCH 6/17][ARM] Add data processing intrinsics for float16_t.

gcc/
2016-09-23  Matthew Wahab  <matthew.wahab@arm.com>

* config/arm/arm.c (arm_evpc_neon_vuzp): Add support for V8HF and
V4HF modes.
(arm_evpc_neon_vtrn): Likewise.
(arm_evpc_neon_vrev): Likewise.
(arm_evpc_neon_vext): Likewise.
* config/arm/arm_neon.h (vbsl_f16): New.
(vbslq_f16): New.
(vdup_n_f16): New.
(vdupq_n_f16): New.
(vdup_lane_f16): New.
(vdupq_lane_f16): New.
(vext_f16): New.
(vextq_f16): New.
(vmov_n_f16): New.
(vmovq_n_f16): New.
(vrev64_f16): New.
(vrev64q_f16): New.
(vtrn_f16): New.
(vtrnq_f16): New.
(vuzp_f16): New.
(vuzpq_f16): New.
(vzip_f16): New.
(vzipq_f16): New.
* config/arm/arm_neon_buillins.def (vdup_n): New (v8hf, v4hf variants).
(vdup_lane): New (v8hf, v4hf variants).
(vext): New (v8hf, v4hf variants).
(vbsl): New (v8hf, v4hf variants).
* config/arm/iterators.md (VDQWH): New.
(VH): New.
(V_double_vector_mode): Add V8HF and V4HF.  Fix white-space.
(Scalar_mul_8_16): Fix white-space.
(Is_d_reg): Add V4HF and V8HF.
* config/arm/neon.md (neon_vdup_lane<mode>_internal): New.
(neon_vdup_lane<mode>): New.
(neon_vtrn<mode>_internal): Replace VDQW with VDQWH.
(*neon_vtrn<mode>_insn): Likewise.
(neon_vzip<mode>_internal): Likewise. Also fix white-space.
(*neon_vzip<mode>_insn): Likewise
(neon_vuzp<mode>_internal): Likewise.
(*neon_vuzp<mode>_insn): Likewise
* config/arm/vec-common.md (vec_perm_const<mode>): New.

testsuite/
2016-09-23  Matthew Wahab  <matthew.wahab@arm.com>

* gcc.target/aarch64/advsimd-intrinsics/arm-neon-ref.h
(FP16_SUPPORTED): New
(expected-hfloat-16x4): Make conditional on __fp16 support.
(expected-hfloat-16x8): Likewise.
(vdup_n_f16): Disable for non-AArch64 targets.
* gcc.target/aarch64/advsimd-intrinsics/vbsl.c: Add __fp16 tests,
conditional on FP16_SUPPORTED.
* gcc.target/aarch64/advsimd-intrinsics/vdup-vmov.c: Likewise.
* gcc.target/aarch64/advsimd-intrinsics/vdup_lane.c: Likewise.
* gcc.target/aarch64/advsimd-intrinsics/vext.c: Likewise.
* gcc.target/aarch64/advsimd-intrinsics/vrev.c: Likewise.
* gcc.target/aarch64/advsimd-intrinsics/vshuffle.inc: Add support
for testing __fp16.
* gcc.target/aarch64/advsimd-intrinsics/vtrn.c: Add __fp16 tests,
conditional on FP16_SUPPORTED.
* gcc.target/aarch64/advsimd-intrinsics/vuzp.c: Likewise.
* gcc.target/aarch64/advsimd-intrinsics/vzip.c: Likewise.

From-SVN: r240404

[PATCH 5/17][ARM] Enable HI mode moves for floating point values.

gcc/
2016-09-23  Jiong Wang  <jiong.wang@arm.com>
    Matthew Wahab  <matthew.wahab@arm.com>

* config/arm/arm.c (output_move_vfp): Weaken assert to allow
HImode.
(arm_hard_regno_mode_ok): Allow HImode values in VFP registers.
* config/arm/arm.md (*movhi_bytes): Disable when VFP registers are
available.  Also fix some white-space.
* config/arm/vfp.md (*arm_movhi_vfp): New.
(*thumb2_movhi_vfp): New.

testsuite/
2016-09-23  Matthew Wahab  <matthew.wahab@arm.com>

* gcc.target/arm/short-vfp-1.c: New.

From-SVN: r240403

[PATCH 4/17][ARM] Define feature macros for FP16.

2016-09-23  Matthew Wahab  <matthew.wahab@arm.com>

* config/arm/arm-c.c (arm_cpu_builtins): Define
"__ARM_FEATURE_FP16_SCALAR_ARITHMETIC" and
"__ARM_FEATURE_FP16_VECTOR_ARITHMETIC".

testsuite/
2016-09-23  Matthew Wahab  <matthew.wahab@arm.com>

* gcc.target/arm/attr-fp16-arith-1.c: New.

From-SVN: r240402

[PATCH 3/17][Testsuite] Add ARM support for ARMv8.2-A with FP16 arithmetic instructions.

gcc/
2016-09-23  Matthew Wahab  <matthew.wahab@arm.com>

* doc/sourcebuild.texi (ARM-specific attributes): Add anchor for
arm_v8_1a_neon_ok.  Add entries for arm_v8_2a_fp16_scalar_ok,
arm_v8_2a_fp16_scalar_hw, arm_v8_2a_fp16_neon_ok and
arm_v8_2a_fp16_neon_hw.
(Add options): Add entries for arm_v8_1a_neon, arm_v8_2a_scalar,
arm_v8_2a_neon.

testsuite/
2016-09-23  Matthew Wahab  <matthew.wahab@arm.com>

* lib/target-supports.exp (add_options_for_arm_v8_2a_fp16_scalar):
New.
(add_options_for_arm_v8_2a_fp16_neon): New.
(check_effective_target_arm_arch_v8_2a_ok): Auto-generate.
(add_options_for_arm_arch_v8_2a): Auto-generate.
(check_effective_target_arm_arch_v8_2a_multilib): Auto-generate.
(check_effective_target_arm_v8_2a_fp16_scalar_ok_nocache): New.
(check_effective_target_arm_v8_2a_fp16_scalar_ok): New.
(check_effective_target_arm_v8_2a_fp16_neon_ok_nocache): New.
(check_effective_target_arm_v8_2a_fp16_neon_ok): New.
(check_effective_target_arm_v8_2a_fp16_scalar_hw): New.
(check_effective_target_arm_v8_2a_fp16_neon_hw): New.

From-SVN: r240401

[PATCH 2/17][Testsuite] Add a selector for ARM FP16 alternative format support.

2016-09-23  Matthew Wahab  <matthew.wahab@arm.com>

* doc/sourcebuild.texi (ARM-specific attributes): Add entries for
arm_fp16_alternative_ok and arm_fp16_none_ok.

testsuite/
2016-09-23  Matthew Wahab  <matthew.wahab@arm.com>

* g++.dg/ext/arm-fp16/arm-fp16-ops-3.C: Use
arm_fp16_alternative_ok.
* g++.dg/ext/arm-fp16/arm-fp16-ops-4.C: Likewise.
* gcc.dg/torture/arm-fp16-int-convert-alt.c: Likewise.
* gcc/testsuite/gcc.dg/torture/arm-fp16-ops-3.c: Likewise.
* gcc/testsuite/gcc.dg/torture/arm-fp16-ops-4.c: Likewise.
* gcc.target/arm/fp16-compile-alt-1.c: Likewise.
* gcc.target/arm/fp16-compile-alt-10.c: Likewise.
* gcc.target/arm/fp16-compile-alt-11.c: Likewise.
* gcc.target/arm/fp16-compile-alt-12.c: Likewise.
* gcc.target/arm/fp16-compile-alt-2.c: Likewise.
* gcc.target/arm/fp16-compile-alt-3.c: Likewise.
* gcc.target/arm/fp16-compile-alt-4.c: Likewise.
* gcc.target/arm/fp16-compile-alt-5.c: Likewise.
* gcc.target/arm/fp16-compile-alt-6.c: Likewise.
* gcc.target/arm/fp16-compile-alt-7.c: Likewise.
* gcc.target/arm/fp16-compile-alt-8.c: Likewise.
* gcc.target/arm/fp16-compile-alt-9.c: Likewise.
* gcc.target/arm/fp16-compile-none-1.c: Use arm_fp16_none_ok.
* gcc.target/arm/fp16-compile-none-2.c: Likewise.
* gcc.target/arm/fp16-rounding-alt-1.c: Use
arm_fp16_alternative_ok.
* lib/target-supports.exp
(check_effective_target_arm_fp16_alternative_ok_nocache): New.
(check_effective_target_arm_fp16_alternative_ok): New.
(check_effective_target_arm_fp16_none_ok_nocache): New.
(check_effective_target_arm_fp16_none_ok): New.

From-SVN: r240400

Fix typo in IPA ICF

* ipa-icf.c (sem_variable::merge): Replace adress
with address.
* gcc.dg/ipa/pr77653.c: Replace adress
with address.

From-SVN: r240399

[PATCH 1/17][ARM] Add ARMv8.2-A command line option and profile.

2016-09-23  Matthew Wahab  <matthew.wahab@arm.com>

* config/arm/arm-arches.def ("armv8.1-a"): Add FL_CRC32.
("armv8.2-a"): New.
("armv8.2-a+fp16"): New.
* config/arm/arm-protos.h (FL2_ARCH8_2): New.
(FL2_FP16INST): New.
(FL2_FOR_ARCH8_2A): New.
* config/arm/arm-tables.opt: Regenerate.
* config/arm/arm.c (arm_arch8_2): New.
(arm_fp16_inst): New.
(arm_option_override): Set arm_arch8_2 and arm_fp16_inst.  Check
for incompatible fp16-format settings.
* config/arm/arm.h (TARGET_VFP_FP16INST): New.
(TARGET_NEON_FP16INST): New.
(arm_arch8_2): Declare.
(arm_fp16_inst): Declare.
* config/arm/bpabi.h (BE8_LINK_SPEC): Add entries for
march=armv8.2-a and march=armv8.2-a+fp16.
* config/arm/t-aprofile (Arch Matches): Add entries for armv8.2-a
and armv8.2-a+fp16.
* doc/invoke.texi (ARM Options): Add "-march=armv8.1-a",
"-march=armv8.2-a" and "-march=armv8.2-a+fp16".

From-SVN: r240398

Remove fused-madd from documentation

* doc/extend.texi: Remove fused-madd from i386 target
options.

From-SVN: r240395

Support movbe as a i386 target optimization node

* gcc.target/i386/movbe-4.c: New test.
* config/i386/i386.c (ix86_valid_target_attribute_inner_p):
Handle movbe.

From-SVN: r240394

Support crc32 as a i386 target optimization node

* config/i386/i386.c (ix86_valid_target_attribute_inner_p):
Handle crc32.
* gcc.target/i386/crc32-5.c: New test.

From-SVN: r240393

re PR target/71652 (ICE in in ix86_target_macros_internal, at config/i386/i386-c.c:187)

Fix PR target/71652

PR target/71652
* config/i386/i386.c (ix86_option_override_internal): Change
signature and return false when there's an error related to
arch string.
(release_options_strings): New function.
(ix86_valid_target_attribute_tree): Call the function.
* gcc.target/i386/pr71652.c: New test.
* gcc.target/i386/pr71652-2.c: New test.
* gcc.target/i386/pr71652-3.c: New test.

From-SVN: r240392

gcc-dg.exp (process-message): Support relative line number notation - .+4 or .-1 etc.

* lib/gcc-dg.exp (process-message): Support relative line number
notation - .+4 or .-1 etc.
* gcc.dg/dg-test-1.c: New test.

From-SVN: r240391

* hsa-gen.c (hsa_op_immed::hsa_op_immed Use CONSTRUCTOR_NELTS (...)
instead of vec_safe_length (CONSTRUCTOR_ELTS (...)).
(gen_hsa_ctor_assignment): Likewise.
* print-tree.c (print_node): Likewise.
* tree-dump.c (dequeue_and_dump): Likewise.
* tree-sra.c (sra_modify_constructor_assign): Likewise.
* expr.c (store_constructor): Likewise.
* fold-const.c (operand_equal_p): Likewise.
* tree-pretty-print.c (dump_generic_node): Likewise.
* hsa-brig.c (hsa_op_immed::emit_to_buffer): Likewise.
* ipa-icf-gimple.c (func_checker::compare_operand): Likewise.
cp/
* typeck2.c (process_init_constructor_record): Use
CONSTRUCTOR_NELTS (...) instead of
vec_safe_length (CONSTRUCTOR_ELTS (...)).
* decl.c (reshape_init_r): Likewise.
(check_initializer): Likewise.
ada/
* gcc-interface/decl.c (gnat_to_gnu_entity): Use
CONSTRUCTOR_NELTS (...) instead of
vec_safe_length (CONSTRUCTOR_ELTS (...)).

From-SVN: r240390

hooks.h (hook_uint_uintp_false): Declare.

2016-09-23  Richard Biener  <rguenther@suse.de>

* hooks.h (hook_uint_uintp_false): Declare.

From-SVN: r240389

Provide right LDD offset bound in avr_address_cost

This patch fixes cost computation in avr_address_cost - instead of the
hardcoded 61, it uses the already existing MAX_LD_OFFSET(mode) macro.

This showed up when investigating a code size regression in the ivopts
pass. That pass computes address_cost with and without an offset to
decide on the right induction variable candidate(s). The legitimate
address target hook returns false for offsets more than 63, so the
pass calls the TARGET_ADDRESS_COST hook with 62 as the offset.

The avr_address_cost hook returns 18 as the cost, and the ivopts pass
concludes that the cost of address with *any* offset is 18, which is not
true - the higher cost is incurred only with offsets bigger than MAX_LD_OFFSET.
This in turn results in a suboptimal choice of induction variables in the
ivopts pass. The patch changes the hardcoded 61 to use the mode
specific MAX_LD_OFFSET instead.

Regression testing with just that fix showed one additional
compilation timeout. That turned out to be the same as
https://lists.nongnu.org/archive/html/avr-gcc-list/2014-03/msg00010.html
- the middle end takes too much time to decide on the best strategy to
multiply DImode values on a 64 bit host. This already causes timeouts
for a few builtin-arith-overflow-* tests (see
https://gcc.gnu.org/ml/gcc-testresults/2016-09/msg02018.html), so it
isn't really related to this fix. Just providing a cost estimate for
DImode mul fixes the timeout though, so the patch does that by scaling
SImode costs by 2 for DImode muls.

With both changes in, there are no regressions, and the
builtin-arith-overflow-* tests now PASS and don't timeout.

gcc/ChangeLog

2016-09-22  Senthil Kumar Selvaraj  <senthil_kumar.selvaraj@atmel.com>

* config/avr/avr.c (avr_rtx_costs_1): Handle DImode MULT.
(avr_address_cost): Replace 61 with MAX_LD_OFFSET(mode).

From-SVN: r240388

[RTEMS] Always use atomic builtins for libstdc++

libstdc++-v3/
* config/cpu/m68k/atomicity.h: Adjust comment.
* acinclude.m4 (GLIBCXX_ENABLE_ATOMIC_BUILTINS): Honor
explicit atomicity_dir setup via configure.host.
* configure.host (rtems-*): Set atomicity_dir.
* configure: Regenerate.

From-SVN: r240387

Daily bump.

From-SVN: r240386

PR target/77676 - powerpc64 and powerpc64le stage2 bootstrap fail

gcc/testsuite/ChangeLog:

PR target/77676
* gcc.dg/tree-ssa/builtin-sprintf-1.c: Define and use wint_t.
* gcc.dg/tree-ssa/builtin-sprintf-2.c: Fix typo.
* gcc.dg/tree-ssa/builtin-sprintf-3.c: New test.
* gcc.dg/tree-ssa/builtin-sprintf-warn-5.c: New test.

gcc/ChangeLog:

PR target/77676
* gimple-ssa-sprintf.c (target_int_min, target_int_max): Use
HOST_BITS_PER_WIDE_INT, make a static local variable auto.
(target_int_min): Correct computation.
(format_integer): Use long long as the argument for the ll length
modifier.
(format_floating): Use target_int_max().
(get_string_length): Same.
(format_string): Avoid setting the bounded flag for strings
of unknown length.
(try_substitute_return_value): Avoid setting range info when
the result isn't bounded.
* varasm.c (assemble_name): Increase buffer size.

From-SVN: r240383

compiler: compile runtime.getcaller{pc,sp} into builtin functions

    The runtime functions runtime.getcallerpc and runtime.getcallersp are
    intended to be efficient ways to get the return and frame address of the
    caller (that is, the caller of runtime.getcallerpc).  In the C code that
    is implemented by simply using C macros:

    This patch essentially implements those macros in the Go code.

    It would be nice if we could just use //extern for this, but it doesn't
    work because the runtime code passes the right argument.  Of course we
    could change the runtime code, but these are common enough that I'd
    prefer to avoid the difference from the gc version of the runtime code.

    This patch corrects the existing declaration of __builtin_return_address
    to use uint32, rather than uint, for the parameter type.  The builtin
    functions take the C type "unsigned int", which for the targets we use
    corresponds to the Go type uint32.  Not that it should matter, really.

    Reviewed-on: https://go-review.googlesource.com/29653

From-SVN: r240382

ifcvt-1.c: Compile also for 64-bit i?86-*-* target.

* gcc.dg/ifcvt-1.c: Compile also for 64-bit i?86-*-* target.
* gcc.dg/ifcvt-2.c: Ditto.
* gcc.dg/zero_bits_compound-1.c: Ditto.
* gcc.dg/zero_bits_compound-1.c: Ditto.
* gcc.dg/pr40550.c: Simplify target selectors.
Use dg-additional-options.
* gcc.dg/pr47893.c: Ditto.
* gcc.dg/pr68435.c: Compile also for i?86-*-* target.  Add -march=i686
additional options for 32-bit x86 targets.
* gcc.dg/pr70955.c: Move to ...
* gcc.target/i386/pr70955.c: ... here.  Simplify target selector.

From-SVN: r240381

re PR c++/61019 (ICE: incomplete type of class template as pseudo-destructor-name)

2016-09-22  Paolo Carlini  <paolo.carlini@oracle.com>

PR c++/61019
* g++.dg/cpp0x/pr61019.C: New.

From-SVN: r240380

[ARM] Add support for -mpure-code option

gcc/ChangeLog:
2016-09-22  Andre Vieira  <andre.simoesdiasvieira@arm.com>
            Terry Guo  <terry.guo@arm.com>

    * target.def (elf_flags_numeric): New target hook.
    * targhooks.h (default_asm_elf_flags_numeric): New.
    * varasm.c (default_asm_elf_flags_numeric): New.
    (default_elf_asm_named_section): Use new target hook.
    * config/arm/arm.opt (mpure-code): New.
    * config/arm/arm.h (SECTION_ARM_PURECODE): New.
    * config/arm/arm.c (arm_asm_init_sections): Add section
    attribute to default text section if -mpure-code.
    (arm_option_check_internal): Diagnose use of option with
    non supported targets and/or options.
    (arm_asm_elf_flags_numeric): New.
    (arm_function_section): New.
    (arm_elf_section_type_flags): New.
    * config/arm/elf.h (JUMP_TABLES_IN_TEXT_SECTION): Disable
    for -mpure-code.
    * gcc/doc/texi (TARGET_ASM_ELF_FLAGS_NUMERIC): New.
    * gcc/doc/texi.in (TARGET_ASM_ELF_FLAGS_NUMERIC): Likewise.

gcc/testsuite/ChangeLog:
2016-09-22  Andre Vieira  <andre.simoesdiasvieira@arm.com>
            Terry Guo  <terry.guo@arm.com>

    * gcc.target/arm/pure-code/ffunction-sections.c: New.
    * gcc.target/arm/pure-code/no-literal-pool.c: New.
    * gcc.target/arm/pure-code/pure-code.exp: New.

Co-Authored-By: Terry Guo <terry.guo@arm.com>
From-SVN: r240379

gofrontend: add missing break in Builtin_call_expression::do_flatten

    Reviewed-on: https://go-review.googlesource.com/29593

From-SVN: r240378

const-2b.c: Also compile for x86_64-*-*.

* gcc.dg/debug/dwarf2/const-2b.c: Also compile for x86_64-*-*.
Remove SSE effective target requirement.

From-SVN: r240377

* MAINTAINERS (Write After Approval): Add myself.

From-SVN: r240375

re PR c++/71979 (ICE with on C++ code with incorrect type in overloaded base class '=' operator: in build_base_path, at cp/class.c:304)

/cp
2016-09-22  Paolo Carlini  <paolo.carlini@oracle.com>

PR c++/71979
* class.c (build_base_path): Allow for lookup_base returning
NULL_TREE.

/testsuite
2016-09-22  Paolo Carlini  <paolo.carlini@oracle.com>

PR c++/71979
* g++.dg/cpp0x/pr71979.C: New.

From-SVN: r240373

Revert accidental commit.

From-SVN: r240369

* regcprop.c (copyprop_hardreg_forward_1): Remove noop moves.

From-SVN: r240365

Implement C++17 node extraction and insertion (P0083R5)

* doc/xml/manual/status_cxx2017.xml: Document status.
* doc/html/*: Regenerate.
* include/Makefile.am: Add bits/node_handle.h and reorder.
* include/Makefile.in: Regenerate.
* include/bits/hashtable.h (_Hashtable::node_type)
(_Hashtable::insert_return_type, _Hashtable::_M_reinsert_node)
(_Hashtable::_M_reinsert_node_multi, _Hashtable::extract)
(_Hashtable::_M_merge_unique, _Hashtable::_M_merge_multi): Define.
(_Hash_merge_helper): Define primary template.
* include/bits/node_handle.h: New header.
* include/bits/stl_map.h (map): Declare _Rb_tree_merge_helper as
friend.
(map::node_type, map::insert_return_type, map::extract, map::merge)
(map::insert(node_type&&), map::insert(const_iterator, node_type&&)):
Define new members.
(_Rb_tree_merge_helper): Specialize for map.
* include/bits/stl_multimap.h (multimap): Declare _Rb_tree_merge_helper
as friend.
(multimap::node_type, multimap::extract, multimap::merge)
(multimap::insert(node_type&&))
(multimap::insert(const_iterator, node_type&&)): Define.
(_Rb_tree_merge_helper): Specialize for multimap.
* include/bits/stl_multiset.h (multiset): Declare _Rb_tree_merge_helper
as friend.
(multiset::node_type, multiset::extract, multiset::merge)
(multiset::insert(node_type&&))
(multiset::insert(const_iterator, node_type&&)): Define.
* include/bits/stl_set.h (set): Declare _Rb_tree_merge_helper as
friend.
(set::node_type, set::insert_return_type, set::extract, set::merge)
(set::insert(node_type&&), set::insert(const_iterator, node_type&&)):
Define.
(_Rb_tree_merge_helper): Specialize for set.
* include/bits/stl_tree.h (_Rb_tree): Declare _Rb_tree<> as friend.
(_Rb_tree::node_type, _Rb_tree::insert_return_type)
(_Rb_tree::_M_reinsert_node_unique, _Rb_tree::_M_reinsert_node_equal)
(_Rb_tree::_M_reinsert_node_hint_unique)
(_Rb_tree::_M_reinsert_node_hint_equal, _Rb_tree::extract)
(_Rb_tree::_M_merge_unique, _Rb_tree::_M_merge_equal): Define.
(_Rb_tree_merge_helper): Specialize for multiset.
* include/bits/unordered_map.h (unordered_map): Declare
unordered_map<> and unordered_multimap<> as friends.
(unordered_map::node_type, unordered_map::insert_return_type)
(unordered_map::extract, unordered_map::merge)
(unordered_map::insert(node_type&&))
(unordered_map::insert(const_iterator, node_type&&))
(unordered_multimap): Declare _Hash_merge_helper as friend.
(unordered_multimap::node_type, unordered_multimap::extract)
(unordered_multimap::merge, unordered_multimap::insert(node_type&&))
(unordered_multimap::insert(const_iterator, node_type&&)): Define.
(_Hash_merge_helper): Specialize for unordered maps and multimaps.
* include/bits/unordered_set.h (unordered_set, unordered_multiset):
Declare _Hash_merge_helper as friend.
(unordered_set::node_type, unordered_set::insert_return_type)
(unordered_set::extract, unordered_set::merge)
(unordered_set::insert(node_type&&))
(unordered_set::insert(const_iterator, node_type&&)): Define.
(unordered_multiset::node_type, unordered_multiset::extract)
(unordered_multiset::merge, unordered_multiset::insert(node_type&&))
(unordered_multiset::insert(const_iterator, node_type&&)): Define.
(_Hash_merge_helper): Specialize for unordered sets and multisets.
* include/debug/map.h (map): Add using declarations or forwarding
functions for new members.
* include/debug/map.h (multimap): Likewise.
* include/debug/map.h (multiset): Likewise.
* include/debug/map.h (set): Likewise.
* include/debug/unordered_map (unordered_map, unordered_multimap):
Likewise.
* include/debug/unordered_set( unordered_set, unordered_multiset):
Likewise.
* python/libstdcxx/v6/printers.py (get_value_from_aligned_membuf): New
helper function.
(get_value_from_list_node, get_value_from_Rb_tree_node): Use helper.
(StdNodeHandlePrinter): Define printer for node handles.
(build_libstdcxx_dictionary): Register StdNodeHandlePrinter.
* testsuite/23_containers/map/modifiers/extract.cc: New.
* testsuite/23_containers/map/modifiers/merge.cc: New.
* testsuite/23_containers/multimap/modifiers/extract.cc: New.
* testsuite/23_containers/multimap/modifiers/merge.cc: New.
* testsuite/23_containers/multiset/modifiers/extract.cc: New.
* testsuite/23_containers/multiset/modifiers/merge.cc: New.
* testsuite/23_containers/set/modifiers/extract.cc: New.
* testsuite/23_containers/set/modifiers/merge.cc: New.
* testsuite/23_containers/unordered_map/modifiers/extract.cc: New.
* testsuite/23_containers/unordered_map/modifiers/merge.cc: New.
* testsuite/23_containers/unordered_multimap/modifiers/extract.cc:
New.
* testsuite/23_containers/unordered_multimap/modifiers/merge.cc: New.
* testsuite/23_containers/unordered_multiset/modifiers/extract.cc:
New.
* testsuite/23_containers/unordered_multiset/modifiers/merge.cc: New.
* testsuite/23_containers/unordered_set/modifiers/extract.cc: New.
* testsuite/23_containers/unordered_set/modifiers/merge.cc: New.
* testsuite/23_containers/unordered_set/instantiation_neg.cc: Adjust
dg-error lineno.
* testsuite/libstdc++-prettyprinters/cxx17.cc: Test node handles.

From-SVN: r240363

make next_cc0_user take rtx_insn *

gcc/ChangeLog:

2016-09-22  Trevor Saunders  <tbsaunde+gcc@tbsaunde.org>

* emit-rtl.c (next_cc0_user): Make argument type rtx_insn *.
* rtl.h: Adjust prototype.

From-SVN: r240362

make next/prev active_insn and active_insn_p take rtx_insn *

gcc/ChangeLog:

2016-09-22  Trevor Saunders  <tbsaunde+gcc@tbsaunde.org>

* emit-rtl.c (next_active_insn): Change argument type to
rtx_insn *.
(prev_active_insn): Likewise.
(active_insn_p): Likewise.
* rtl.h: Adjust prototypes.
* cfgcleanup.c (merge_blocks_move_successor_nojumps): Adjust.
* config/arc/arc.md: Likewise.
* config/pa/pa.c (branch_to_delay_slot_p): Likewise.
(branch_needs_nop_p): Likewise.
(use_skip_p): Likewise.
* config/sh/sh.c (gen_block_redirect): Likewise.
(split_branches): Likewise.
* reorg.c (optimize_skip): Likewise.
(fill_simple_delay_slots): Likewise.
(fill_slots_from_thread): Likewise.
(relax_delay_slots): Likewise.
* resource.c (mark_target_live_regs): Likewise.

From-SVN: r240361

make next/prev nonnote_nondebug_insn take rtx_insn *

gcc/ChangeLog:

2016-09-22  Trevor Saunders  <tbsaunde+gcc@tbsaunde.org>

* config/cris/cris.c (cris_asm_output_case_end): Change argument
type to rtx_insn *.
* emit-rtl.c (next_nonnote_nondebug_insn): Likewise.
(prev_nonnote_nondebug_insn): Likewise.
* config/cris/cris-protos.h: Adjust prototype.
* rtl.h: Likewise.
* jump.c (rtx_renumbered_equal_p): Adjust.

From-SVN: r240360

make prev_real_insn take rtx_insn *

gcc/ChangeLog:

2016-09-22  Trevor Saunders  <tbsaunde+gcc@tbsaunde.org>

* emit-rtl.c (prev_real_insn): Change argument type to rtx_insn *.
* rtl.h: Adjust prototype.
* config/sh/sh.md: Adjust.
* dwarf2out.c (add_var_loc_to_decl): Likewise.

From-SVN: r240359

make next/prev nondebug_insn take rtx_insn *

gcc/ChangeLog:

2016-09-22  Trevor Saunders  <tbsaunde+gcc@tbsaunde.org>

* emit-rtl.c (next_nondebug_insn): Change argument type to
rtx_insn *.
(prev_nondebug_insn): Likewise.
* loop-doloop.c (doloop_condition_get): Likewise.
* rtl.h: Adjust prototype.
* cfgloop.h: Likewise.

From-SVN: r240358

make next/prev _nonnote_insn take rtx_insn *

gcc/ChangeLog:

2016-09-22  Trevor Saunders  <tbsaunde+gcc@tbsaunde.org>

* emit-rtl.c (next_nonnote_insn): Change argument type to
rtx_insn *.
(prev_nonnote_insn): Likewise.
* jump.c (reversed_comparison_code_parts): Likewise.
(reversed_comparison): Likewise.
* rtl.h: Adjust prototypes.
* config/arc/arc.md: Adjust.
* cse.c (find_comparison_args): Likewise.
* reorg.c (redundant_insn): Change return type to rtx_insn *.
(fix_reg_dead_note): Change argument type to rtx_insn *.
(delete_prior_computation): Likewise.
(delete_computation): Likewise.
(fill_slots_from_thread): Adjust.
(relax_delay_slots): Likewise.
* simplify-rtx.c (simplify_unary_operation_1): Likewise.
(simplify_relational_operation_1): Likewise.
(simplify_ternary_operation): Likewise.

From-SVN: r240357

use rtx_insn * more

gcc/ChangeLog:

2016-09-22  Trevor Saunders  <tbsaunde+gcc@tbsaunde.org>

* config/arc/arc-protos.h (arc_label_align): Change type of
variables from rtx to rtx_insn *.
* config/arc/arc.c (arc_label_align): Likewise.
* config/arm/arm.c (any_sibcall_could_use_r3): Likewise.
* config/bfin/bfin.c (workaround_speculation): Likewise.
* config/c6x/c6x.c (find_next_cycle_insn): Likewise.
(find_last_same_clock): Likewise.
(reorg_split_calls): Likewise.
* config/cris/cris-protos.h (cris_cc0_user_requires_cmp): Likewise.
* config/cris/cris.c (cris_cc0_user_requires_cmp): Likewise.
* config/h8300/h8300-protos.h (same_cmp_preceding_p): Likewise.
(same_cmp_following_p): Likewise.
* config/h8300/h8300.c (same_cmp_preceding_p): Likewise.
(same_cmp_following_p): Likwise.
* config/m32r/m32r.c (m32r_expand_epilogue): Likewise.
* config/nds32/nds32-protos.h (nds32_target_alignment): Likewise.
* config/nds32/nds32.c (nds32_target_alignment): Likewise.
* config/rl78/rl78.c (rl78_alloc_physical_registers_op2):
* Likewise.
(rl78_alloc_physical_registers_cmp): Likewise.
(rl78_alloc_physical_registers_umul): Likewise.
(rl78_calculate_death_notes): Likewise.
* config/s390/s390-protos.h (s390_label_align): Likewise.
* config/s390/s390.c (s390_label_align): Likewise.
* config/sh/sh.c (barrier_align): Likewise.
* config/sparc/sparc-protos.h (emit_cbcond_nop): Likewise.
* config/sparc/sparc.c (sparc_asm_function_epilogue): Likewise.
(emit_cbcond_nop): Likewise.

From-SVN: r240356

pr77550.C: Use __SIZE_TYPE__.

2016-09-22  Bernd Edlinger  <bernd.edlinger@hotmail.de>

        * g++.dg/pr77550.C: Use __SIZE_TYPE__.

From-SVN: r240355

re PR ipa/77653 (wrong code at -Os and above on x86_64-linux-gnu (in both 32-bit and 64-bit modes))

Fix PR ipa/77653

PR ipa/77653
* gcc.dg/ipa/pr77653.c: New test.
PR ipa/77653
* ipa-icf.c (sem_variable::merge): Yield merge operation if
alias address matters, not necessarily address of original.

From-SVN: r240354

re PR tree-optimization/77697 (suspicious code in tree-ssa-forwprop.c)

2016-09-22  Richard Biener  <rguenther@suse.de>

PR middle-end/77697
* gimple-fold.c (fold_array_ctor_reference): Turn asserts into
fold fails.

From-SVN: r240353

re PR tree-optimization/77677 (ICE at -O1 and above in both 32-bit and 64-bit modes on x86_64-linux-gnu (internal compiler error: in set_value_range, at tree-vrp.c:361))

2016-09-22  Richard Biener  <rguenther@suse.de>

PR middle-end/77677
* gimple-match-head.c (gimple_resimplify1): Drop TREE_OVERFLOW
from constant folding results.
(gimple_resimplify2): Likewise.
(gimple_resimplify3): Likewise.

From-SVN: r240352

re PR fortran/77678 (ICE in fold_read_from_constant_string, at fold-const.c:13706)

2016-09-22  Richard Biener  <rguenther@suse.de>

PR middle-end/77678
* expr.c (expand_expr_real_1): Guard array access against negative
offset.

From-SVN: r240351

Use GMP_RNDN for portability to MPFR 2.4.2

* gimple-ssa-sprintf.c (format_floating_max): Use GMP_RNDN instead
of MPFR_RNDN.
(format_floating): Likewise.

From-SVN: r240350

re PR fortran/48298 ([F03] User-Defined Derived-Type IO (DTIO))

2016-09-22  Paul Thomas  <pault@gcc.gnu.org>

PR fortran/48298
* gfortran.h : Place the pseudo operators INTRINSIC_FORMATTED
and INTRINSIC_UNFORMATTED after the sentinel in enum
gfc_intrinsic_op so that they do not appear as place holders
in module_write.
* interface.c (dtio_op): Comment on the special nature of the
pseudo operators INTRINSIC FORMATTED and INTRINSIC_UNFORMATTED.

From-SVN: r240349

re PR tree-optimization/77665 (ICE in expand_GOMP_SIMD_VF, at internal-fn.c:172)

PR fortran/77665
* tree-inline.c (remap_gimple_stmt): Set has_simduid_loops
for all IFN_GOMP_SIMD_* internal fns, not just for
IFN_GOMP_SIMD_ORDERED_*.

* gfortran.dg/gomp/pr77665.f90: New test.

From-SVN: r240348

Fix tests on old arm platforms for optional.

* testsuite/20_util/optional/77288.cc: Don't use exception_ptr.

From-SVN: r240347

class.c (push_super_field): Set DECL_CONTEXT.

2016-09-22  Richard Biener  <rguenther@suse.de>

java/
* class.c (push_super_field): Set DECL_CONTEXT.

From-SVN: r240346

Update pretty printer for std::variant

* python/libstdcxx/v6/printers.py (StdVariantPrinter): Adjust for
recent change to _Variant_storage.
* testsuite/libstdc++-prettyprinters/cxx17.cc: Test variant with
reference type.

From-SVN: r240345

Always qualify std::forward in <variant>

* include/bits/uses_allocator.h (__uses_allocator_construct): Qualify
std::forward and ::new. Cast pointer to void*.
* include/std/variant (_Variant_storage, _Union, _Variant_base)
(__access, __visit_invoke, variant, visit): Qualify std::forward.
* testsuite/20_util/variant/compile.cc: Test for ADL problems.

From-SVN: r240344

variant (variant::operator=): Fix assignment on references.

* libstdc++-v3/include/std/variant (variant::operator=): Fix assignment
on references.
* libstdc++-v3/testsuite/20_util/variant/compile.cc: Add test.

From-SVN: r240343

interface.c (check_dtio_interface1): Introduce errors for alternate returns and incorrect numbers of arguments.

2016-09-22  Paul Thomas  <pault@gcc.gnu.org>

* interface.c (check_dtio_interface1): Introduce errors for
alternate returns and incorrect numbers of arguments.
(gfc_find_specific_dtio_proc): Return cleanly if the derived
type either doesn't exist or has no namespace.

2016-09-22  Paul Thomas  <pault@gcc.gnu.org>

* gfortran.dg/dtio_11.f90: Correct for changed error messages.
* gfortran.dg/dtio_13.f90: New test.

From-SVN: r240342

re PR fortran/66107 (ICE on missing parameter value for initialisation (segfault))

2016-09-21  Louis Krupp  <louis.krupp@zoho.com>

PR fortran/66107
* gfortran.dg/pr66107.f90: New test.

2016-09-21  Louis Krupp  <louis.krupp@zoho.com>

PR fortran/66107
* decl.c (add_init_expr_to_sym): Catch variable character length
in parameter array.

From-SVN: r240341

re PR libstdc++/77641 (std::variant copy-initialization fails for type with non-trivial constexpr ctor)

PR libstdc++/77641
* include/std/variant (_Variant_storage::_Variant_storage):
Change _Variant_storage's union to be default constructible.
* testsuite/20_util/variant/compile.cc: New test.

From-SVN: r240340

Daily bump.

From-SVN: r240339

re PR target/77670 (PowerPC64 Spec 2006 fails on 453.povray using -mcpu=power9 -mpower9-minmax)

2016-09-21  Michael Meissner  <meissner@linux.vnet.ibm.com>

PR target/77670
* config/rs6000/predicates.md (invert_fpmask_comparison_operator):
New predicate that matches the ISA 3.0 XSCMP{EQ,GT,GE}DP
instructions when you want to invert the test.
* config/rs6000/rs6000.md (fpmask<mode>): Use the arguments in the
correct order for XXSEL.
(mov<SFDF:mode><SFDF2:mode>cc_invert_p9): Define the inverted test
for using XSCMP{EQ,GT,GE}DP.

From-SVN: r240336

re PR target/77670 (PowerPC64 Spec 2006 fails on 453.povray using -mcpu=power9 -mpower9-minmax)

2016-09-21  Michael Meissner  <meissner@linux.vnet.ibm.com>

PR target/77670
* config/rs6000/predicates.md (invert_fpmask_comparison_operator):
New predicate that matches the ISA 3.0 XSCMP{EQ,GT,GE}DP
instructions when you want to invert the test.
* config/rs6000/rs6000.md (fpmask<mode>): Use the arguments in the
correct order for XXSEL.
(mov<SFDF:mode><SFDF2:mode>cc_invert_p9): Define the inverted test
for using XSCMP{EQ,GT,GE}DP.

From-SVN: r240335

compiler, runtime: replace hashmap code with Go 1.7 hashmap

    This change removes the gccgo-specific hashmap code and replaces it with
    the hashmap code from the Go 1.7 runtime.  The Go 1.7 hashmap code is
    more efficient, does a better job on details like when to update a key,
    and provides some support against denial-of-service attacks.

    The compiler is changed to call the new hashmap functions instead of the
    old ones.

    The compiler now tracks which types are reflexive and which require
    updating when used as a map key, and records the information in map type
    descriptors.

    Map_index_expression is simplified.  The special case for a map index on
    the right hand side of a tuple expression has been unnecessary for some
    time, and is removed.  The support for specially marking a map index as
    an lvalue is removed, in favor of lowering an assignment to a map index
    into a function call.  The long-obsolete support for a map index of a
    pointer to a map is removed.

    The __go_new_map_big function (known to the compiler as
    Runtime::MAKEMAPBIG) is no longer needed, as the new runtime.makemap
    function takes an int64 hint argument.

    The old map descriptor type and supporting expression is removed.

    The compiler was still supporting the long-obsolete syntax `m[k] = 0,
    false` to delete a value from a map.  That is now removed, requiring a
    change to one of the gccgo-specific tests.

    The builtin len function applied to a map or channel p is now compiled
    as `p == nil ? 0 : *(*int)(p)`.  The __go_chan_len function (known to
    the compiler as Runtime::CHAN_LEN) is removed.

    Support for a shared zero value for maps to large value types is
    introduced, along the lines of the gc compiler.  The zero value is
    handled as a common variable.

    The hash function is changed to take a seed argument, changing the
    runtime hash functions and the compiler-generated hash functions.
    Unlike the gc compiler, both the hash and equal functions continue to
    take the type length.

    Types that can not be compared now store nil for the hash and equal
    functions, rather than pointing to functions that throw.  Interface hash
    and comparison functions now check explicitly for nil.  This matches the
    gc compiler and permits a simple implementation for ismapkey.

    The compiler is changed to permit marking struct and array types as
    incomparable, meaning that they have no hash or equal function.  We use
    this for thunk types, removing the existing special code to avoid
    generating hash/equal functions for them.

    The C runtime code adds memclr, memequal, and memmove functions.

    The hashmap code uses go:linkname comments to make the functions
    visible, as otherwise the compiler would discard them.

    The hashmap code comments out the unused reference to the address of the
    first parameter in the race code, as otherwise the compiler thinks that
    the parameter escapes and copies it onto the heap.  This is probably not
    needed when we enable escape analysis.

    Several runtime map tests that ere previously skipped for gccgo are now
    run.

    The Go runtime picks up type kind information and stubs.  The type kind
    information causes the generated runtime header file to define some
    constants, including `empty`, and the C code is adjusted accordingly.

    A Go-callable version of runtime.throw, that takes a Go string, is
    added to be called from the hashmap code.

    Reviewed-on: https://go-review.googlesource.com/29447

* go.go-torture/execute/map-1.go: Replace old map deletion syntax
with call to builtin delete function.

From-SVN: r240334

Introduce class rtx_reader

Bundle up various global variables within gensupport.c into a
class rtx_reader, with a view towards making it easier to run the
code more than once in-process.

gcc/ChangeLog:
* genconstants.c (main): Introduce noop_reader and convert call
to read_md_files to a method call.
* genenums.c (main): Likewise.
* genmddeps.c (main): Likewise.
* genpreds.c (write_tm_constrs_h): Replace use of "in_fname" with
rtx_reader_ptr->get_top_level_filename ().
(write_tm_preds_h): Likewise.
(write_insn_preds_c): Likewise.
* gensupport.c (class gen_reader): New subclass of rtx_reader.
(rtx_handle_directive): Convert to...
(gen_reader::handle_unknown_directive): ...this.
(init_rtx_reader_args_cb): Convert return type from bool to
rtx_reader *.  Create a gen_reader instance, using it for the
call to read_md_files.  Return it if no errors occur.
(init_rtx_reader_args): Convert return type from bool to
rtx_reader *.
* gensupport.h (init_rtx_reader_args_cb): Likewise.
(init_rtx_reader_args_cb): Likewise.
* read-md.c (struct file_name_list): Move to class rtx_reader.
(read_md_file): Delete in favor of rtx_reader::m_read_md_file.
(read_md_filename): Delete in favor of
rtx_reader::m_read_md_filename.
(read_md_lineno): Delete in favor of rtx_reader::m_read_md_lineno.
(in_fname): Delete in favor of rtx_reader::m_toplevel_fname.
(base_dir): Delete in favor of rtx_reader::m_base_dir.
(first_dir_md_include): Delete in favor of
rtx_reader::m_first_dir_md_include.
(last_dir_md_include_ptr): Delete in favor of
rtx_reader::m_last_dir_md_include_ptr.
(max_include_len): Delete.
(rtx_reader_ptr): New.
(fatal_with_file_and_line): Use get_filename and get_lineno
accessors of rtx_reader_ptr.
(require_char_ws): Likewise.
(rtx_reader::read_char): New method, based on ::read_char.
(rtx_reader::unread_char): New method, based on ::unread_char.
(read_escape): Use get_filename and get_lineno accessors of
rtx_reader_ptr.
(read_braced_string): Use get_lineno accessor of rtx_reader_ptr.
(read_string): Use get_filename and get_lineno accessors of
rtx_reader_ptr.
(rtx_reader::rtx_reader): New ctor.
(rtx_reader::~rtx_reader): New dtor.
(handle_include): Convert from a function to...
(rtx_reader::handle_include): ...this method, converting
handle_directive from a callback to a virtual function.
(handle_file): Likewise, converting to...
(rtx_reader::handle_file): ...this method.
(handle_toplevel_file): Likewise, converting to...
(rtx_reader::handle_toplevel_file): ...this method.
(rtx_reader::get_current_location): New method.
(parse_include): Convert from a function to...
(rtx_reader::add_include_path): ...this method, dropping redundant
update to unused max_include_len.
(read_md_files): Convert from a function to...
(rtx_reader::read_md_files): ...this method, converting
handle_directive from a callback to a virtual function.
(noop_reader::handle_unknown_directive): New method.
* read-md.h (directive_handler_t): Delete this typedef.
(in_fname): Delete.
(read_md_file): Delete.
(read_md_lineno): Delete.
(read_md_filename): Delete.
(class rtx_reader): New class.
(rtx_reader_ptr): New decl.
(class noop_reader): New subclass of rtx_reader.
(read_char): Reimplement in terms of rtx_reader::read_char.
(unread_char): Reimplement in terms of rtx_reader::unread_char.
(read_md_files): Delete.
* read-rtl.c (read_rtx_code): Update for deletion of globals
read_md_filename and read_md_lineno.

From-SVN: r240333

Add PR target/71395 marker to 71395 fix

From-SVN: r240332

c-common.c (c_common_truthvalue_conversion): Inhibit Wint-in-bool-context warning with from_macro_definition_at.

2016-09-21  Bernd Edlinger  <bernd.edlinger@hotmail.de>

        * c-common.c (c_common_truthvalue_conversion): Inhibit
        Wint-in-bool-context warning with from_macro_definition_at.
        Mention the expression will always evaluate to true.

From-SVN: r240331

Add from_macro_definition_at predicate for locations.

gcc/
* input.h (from_macro_definition_at): New.
libcpp/
* line-map.c (linemap_location_from_macro_definition_p): New.
* line-map.h: Declare it.

From-SVN: r240330

Core 903

Core 903
* call.c (null_ptr_cst_p): Check char_type_p.

From-SVN: r240329

Add _FloatN, _FloatNx tests for __builtin_fpclassify.

This patch adds tests for _FloatN and _FloatNx arguments to the
__builtin_fpclassify type-generic built-in function, omitted from the
original tests for type-generic functions on these types.

Tested for x86_64-pc-linux-gnu; all the supported new tests pass.

* gcc.dg/torture/float128-tg-3.c, gcc.dg/torture/float128x-tg-3.c,
gcc.dg/torture/float16-tg-3.c, gcc.dg/torture/float32-tg-3.c,
gcc.dg/torture/float32x-tg-3.c, gcc.dg/torture/float64-tg-3.c,
gcc.dg/torture/float64x-tg-3.c, gcc.dg/torture/floatn-tg-3.h: New
tests.

From-SVN: r240328

Fix JUMP_LABEL documentation

* doc/rtl.texi (JUMP_LABEL): Document RETURN and SIMPLE_RETURN values.

From-SVN: r240325

re PR libstdc++/77288 (Std::experimental::optional::operator= implementation is broken in gcc 6.1)

PR libstdc++/77288
* include/std/optional (__is_optional_impl, __is_optional): Remove.
(__converts_from_optional, __assigns_from_optional): New.
(optional(_Up&&)): Use is_same instead of __is_optional.
(optional(const optional<_Up>&)): Constrain with
__converts_from_optional.
(optional(optional<_Up>&&)): Likewise.
(operator=(_Up&&)): Use is_same instead of __is_optional, check
is_same and is_scalar.
(operator=(const optional<_Up>&)): Constrain with
__converts_from_optional and __assigns_from_optional.
(operator=(optional<_Up>&&)): Likewise.
* testsuite/20_util/optional/77288.cc: New.
* testsuite/20_util/optional/cons/value.cc: Adjust.

From-SVN: r240324

Implement LWG 2729 for tuple.

* include/std/tuple (_Tuple_impl(_Tuple_impl&&)):
Suppress conditionally.
(_Tuple_impl(_Tuple_impl<_Idx, _UHead, _UTails...>&&)): Likewise.
(__is_tuple_impl_trait_impl, __is_tuple_impl_trait): New.
(_Tuple_impl(const _Head&)): Constrain.
(_Tuple_impl(_UHead&&)): Likewise.
(_Tuple_impl(_Tuple_impl&&)): Suppress conditionally.
(_Tuple_impl(const _Tuple_impl<_Idx, _UHead>&)): Constrain.
(_Tuple_impl(_Tuple_impl<_Idx, _UHead>&&)): Likewise.
(operator=(const tuple&)): Enable conditionally.
(operator=(tuple&&)): Suppress conditionally.
(operator=(const tuple<_UElements...>&)): Constrain.
(operator=(tuple<_UElements...>&&)): Likewise.
(operator=(const tuple&)): Enable conditionally (2-param tuple).
(operator=(tuple&&)): Suppress conditionally (2-param tuple).
(operator=(const tuple<_U1, _U2>&)): Constrain.
(operator=(tuple<_U1, _U2>&&)): Likewise.
(operator=(const pair<_U1, _U2>&)): Likewise.
(operator=(pair<_U1, _U2>&&)): Likewise.
* testsuite/20_util/tuple/element_access/get_neg.cc: Adjust.
* testsuite/20_util/tuple/tuple_traits.cc: New.

From-SVN: r240323

re PR libstdc++/77537 (pair constructors do not properly SFINAE)

PR libstdc++/77537
Implement LWG 2729 for pair.
* include/bits/stl_pair.h (_PCC): New.
(_ConstructiblePair, _ImplicitlyConvertiblePair):
Turn into static member functions of _PCC.
(_MoveConstructiblePair, _ImplicitlyMoveConvertiblePair): Likewise.
(_PCCP): New.
(pair(const _T1&, const _T2&)): Adjust.
(_PCCFP): New.
(pair(const pair<_U1, _U2>&)): Adjust.
(pair(_U1&&, const _T2&)): Likewise.
(pair(const _T1&, _U2&&)): Likewise.
(pair(_U1&&, _U2&&)): Likewise.
(pair(pair<_U1, _U2>&&)): Likewise.
(operator=(const pair&)): Make conditionally deleted.
(operator=(pair&&)): Make conditionally suppressed.
(operator=(const pair<_U1, _U2>&)): Constrain.
(operator=(pair<_U1, _U2>&&): Likewise.
* include/std/type_traits (__nonesuch): New.
* testsuite/20_util/pair/traits.cc: New.

From-SVN: r240322

PR bootstrap/77676 - powerpc64 and powerpc64le stage2 bootstrap fail

* c.opt (fprintf-return-value): Temporarily initialize to zero
to unblock bootstrap failures.

From-SVN: r240321

dump-noaddr.x: Remove debug statements.

* gcc.c-torture/unsorted/dump-noaddr.x: Remove debug statements.

From-SVN: r240318

re PR c++/77651 (c++ -Wno-aligned-new ICEs)

PR c++/77651
c-family/
* c.opt (Waligned-new=): Add RejectNegative.
(faligned-new=): Likewise.  Spelling fix - change
aligned_new_threshhold to aligned_new_threshold.
* c-cppbuiltin.c (c_cpp_builtins): Change aligned_new_threshhold
to aligned_new_threshold.
cp/
* init.c (build_new_1): Don't suggest to use -faligned-new if
aligned_new_threshold is non-zero.
(type_has_new_extended_alignment): Change aligned_new_threshhold
to aligned_new_threshold.
* call.c (second_parm_is_size_t, aligned_allocation_fn_p,
aligned_deallocation_fn_p, build_op_delete_call): Likewise.
* decl.c (cxx_init_decl_processing): Likewise.
testsuite/
* g++.dg/cpp1z/aligned-new6.C: New test.

From-SVN: r240317

re PR sanitizer/77567 (ASAN: Bugus error "alloc-dealloc-mismatch (malloc vs operator delete [])" with C++17's over-aligned types)

PR sanitizer/77567
* asan/asan_new_delete.cc: Cherry-pick upstream r282019.

From-SVN: r240316

[ARM] FP16 ARM Alternative format variants of AAPCS tests.

testsuite/
2016-09-21  Matthew Wahab  <matthew.wahab@arm.com>

* gcc.target/arm/fp16-aapcs-3.c: New.
* gcc.target/arm/fp16-aapcs-4.c: New.
* gcc.target/arm/aapcs/aapcs/vfp22.c: New.
* gcc.target/arm/aapcs/aapcs/vfp23.c: New.
* gcc.target/arm/aapcs/aapcs/vfp24.c: New.
* gcc.target/arm/aapcs/aapcs/vfp25.c: New.

From-SVN: r240314

re PR tree-optimization/77550 (std::deque with -O3 has infinite std::distance)

gcc:
2016-09-21  Bernd Edlinger  <bernd.edlinger@hotmail.de>

        PR tree-optimization/77550
        * tree-vect-stmts.c (create_array_ref): Change parameters.
        (get_group_alias_ptr_type): New function.
        (vectorizable_store, vectorizable_load): Use get_group_alias_ptr_type.

testsuite:
2016-09-21  Bernd Edlinger  <bernd.edlinger@hotmail.de>

        PR tree-optimization/77550
        * g++.dg/pr77550.C: New test.

From-SVN: r240313

configure.ac: Do not create links, only substitute the filenames.

* configure.ac: Do not create links, only substitute the filenames.
* configure: Regenerate.
* Makefile.in: Assign the substitution results to variables.
(LIBGCC_LINKS): Define.
(enable-execute-stack.c): New rule.
(unwind.h): Likewise.
(md-unwind-support.h): Likewise.
(sfp-machine.h): Likewise.
(gthr-default.h): Likewise.
Add $(LIBGCC_LINKS) to the prerequisites of all object files and
unwind.h as prerequisite of install-unwind_h-forbuild.

From-SVN: r240312

gimple-ssa-sprintf.c (pass_sprintf_length::compute_format_length): Add falls through comment.

* gimple-ssa-sprintf.c (pass_sprintf_length::compute_format_length):
Add falls through comment.

From-SVN: r240311

dwarf2out.c (remove_child_with_prev): Clear child->die_sib.

2016-09-21  Richard Biener  <rguenther@suse.de>

* dwarf2out.c (remove_child_with_prev): Clear child->die_sib.
(replace_child): Likewise.
(remove_child_TAG): Adjust.
(move_marked_base_types): Likewise.
(prune_unused_types_prune): Clear die_sib of removed children.

From-SVN: r240310

Use rand_s on MinGW-w64, fix array bound overflow.

2016-09-21  Janne Blomqvist  <jb@gcc.gnu.org>

* intrinsics/random.c (getosrandom): Use rand_s() on
MinGW-w64. Fix bounds overflow in fallback code.

From-SVN: r240309

re PR target/77326 ([avr] Invalid optimization omits comparison)

gcc/
PR target/77326
* config/avr/avr.c (avr_notice_update_cc) [CC_NONE]: If insn
touches some regs mentioned in cc_status, do CC_STATUS_INIT.
gcc/testsuite/
PR target/77326
* gcc.target/avr/torture/pr77326.c: New test.

From-SVN: r240306

pr68078.f90: Run on x86_64_*_linux* only.

2016-09-21  Louis Krupp  <louis.krupp@zoho.com>

* gfortran.dg/pr68078.f90: Run on x86_64_*_linux* only.

From-SVN: r240304

re PR tree-optimization/77648 (Setting conversion to a integer to double to 0 3/4 through a loop)

2016-09-21  Richard Biener  <rguenther@suse.de>

PR tree-optimization/77648
* tree-ssa-structalias.c (process_constraint): Handle all DEREF
with complex RHS.
(make_transitive_closure_constraints): Adjust comment.
(make_any_offset_constraints): New function.
(handle_rhs_call): Make sure to first expand a pointer to all
subfields before transitively closing it.
(handle_const_call): Likewise.  Properly expand returned
pointers as well.
(handle_pure_call): Likewise.

* gcc.dg/torture/pr77648-1.c: New testcase.
* gcc.dg/torture/pr77648-2.c: Likewise.

From-SVN: r240303

re PR tree-optimization/77621 (Internal compiler error for mtune=atom + msse2)

2016-09-21  Richard Biener  <rguenther@suse.de>
Jakub Jelinek  <jakub@redhat.com>

PR tree-optimization/77621
* tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Split
group at non-vectorizable stmts.

* gcc.dg/pr77621.c: New testcase.

Co-Authored-By: Jakub Jelinek <jakub@redhat.com>
From-SVN: r240302

re PR fortran/77657 (link error with implementation of user-defined derived type input/output (UD-DTIO) in child extending abstract parent)

2016-09-21  Paul Thomas  <pault@gcc.gnu.org>

PR fortran/77657

* interface.c (gfc_find_specific_dtio_proc): Borrow trick from
resolve_typebound_generic_call to find dtio procedures that
over-ride those in the declared type.

2016-09-21  Paul Thomas  <pault@gcc.gnu.org>

PR fortran/77657
* gfortran.dg/dtio_12.f90: New test.

From-SVN: r240301

Make integer size explicit

For the lower vrp bound to be 2/-2, unsigned ints must be 4 bytes wide. This
breaks targets like avr. Explicitly using __UINT32_TYPE__ (behind a typedef)
makes the testcase pass for all targets.

gcc/testsuite/ChangeLog:

2016-09-21  Senthil Kumar Selvaraj  <senthil_kumar.selvaraj@atmel.com>

* gcc.dg/tree-ssa/pr64130.c: Use __UINT32_TYPE__ instead of int.

From-SVN: r240300

Incorrect arithmetic optimization involving bitfield arguments

gcc/ChangeLog:

2016-09-21  Kugan Vivekanandarajah  <kuganv@linaro.org>

PR tree-optimization/72835
* tree-ssa-reassoc.c (make_new_ssa_for_def): New.
(make_new_ssa_for_all_defs): Likewise.
(zero_one_operation): Replace all SSA_NAMEs defined in the chain.

gcc/testsuite/ChangeLog:

2016-09-21  Kugan Vivekanandarajah  <kuganv@linaro.org>

PR tree-optimization/72835
* gcc.dg/tree-ssa/pr72835.c: New test.

From-SVN: r240299

PR middle-end/49905 - Better sanity checking on sprintf src & dest to

gcc/ChangeLog:

PR middle-end/49905
* Makefile.in (OBJS): Add gimple-ssa-sprintf.o.
* config/linux.h (TARGET_PRINTF_POINTER_FORMAT): Redefine.
* config/linux.c (gnu_libc_printf_pointer_format): New function.
* config/sol2.h (TARGET_PRINTF_POINTER_FORMAT): Same.
* config/sol2.c (solaris_printf_pointer_format): New function.
* doc/invoke.texi (-Wformat-length, -fprintf-return-value): New
options.
* doc/tm.texi.in (TARGET_PRINTF_POINTER_FORMAT): Document.
* doc/tm.texi: Regenerate.
* gimple-fold.h (get_range_strlen): New function.
(get_maxval_strlen): Declare existing function.
* gimple-fold.c (get_range_strlen): Add arguments and compute both
maximum and minimum.
 (get_range_strlen): Define overload.
(get_maxval_strlen): Adjust.
* gimple-ssa-sprintf.c: New file and pass.
* passes.def (pass_sprintf_length): Add new pass.
* targhooks.h (default_printf_pointer_format): Declare new function.
(gnu_libc_printf_pointer_format): Same.
(solaris_libc_printf_pointer_format): Same.
* targhooks.c (default_printf_pointer_format): Define new function.
* tree-pass.h (make_pass_sprintf_length): Declare new function.
* print-tree.c: Increase buffer size.

gcc/c-family/ChangeLog:

PR middle-end/49905
* c.opt: Add -Wformat-length and -fprintf-return-value.

gcc/testsuite/ChangeLog:

PR middle-end/49905
* gcc.dg/builtin-stringop-chk-1.c: Adjust.
* gcc.dg/tree-ssa/builtin-sprintf-warn-1.c: New test.
* gcc.dg/tree-ssa/builtin-sprintf-warn-2.c: New test.
* gcc.dg/tree-ssa/builtin-sprintf-warn-3.c: New test.
* gcc.dg/tree-ssa/builtin-sprintf-warn-4.c: New test.
* gcc.dg/tree-ssa/builtin-sprintf.c: New test.
* gcc.dg/tree-ssa/builtin-sprintf-2.c: New test.

From-SVN: r240298

Daily bump.

From-SVN: r240297

Use ipa-vrp result in tree-vrp

gcc/ChangeLog:

2016-09-21  Kugan Vivekanandarajah  <kuganv@linaro.org>

 * tree-vrp.c (get_value_range): Teach PARM_DECL to use ipa-vrp
 results.

gcc/testsuite/ChangeLog:

2016-09-21  Kugan Vivekanandarajah  <kuganv@linaro.org>

* gcc.dg/guality/pr54519-1.c: Add -fno-ipa-vrp. Else constant
arguments/compare will be optimized away.
* gcc.dg/torture/ftrapv-1.c: Likewise.

From-SVN: r240293

Add IPA VRP

gcc/lto/ChangeLog:
2016-09-21  Kugan Vivekanandarajah  <kuganv@linaro.org>

* lto-partition.c: Include tree-vrp.h.
* lto.c: Likewise.

gcc/testsuite/ChangeLog:

2016-09-21  Kugan Vivekanandarajah  <kuganv@linaro.org>

* g++.dg/ipa/pure-const-3.C: Add -fno-ipa-vrp. Else constant arguments
will be optimized away.
* gcc.dg/ipa/vrp1.c: New test.
* gcc.dg/ipa/vrp2.c: New test.
* gcc.dg/ipa/vrp3.c: New test.

gcc/ChangeLog:

2016-09-21  Kugan Vivekanandarajah  <kuganv@linaro.org>

* common.opt: New option -fipa-vrp.
* ipa-cp.c (ipa_get_vr_lat): New.
(ipcp_vr_lattice::print): Likewise.
(print_all_lattices): Call ipcp_vr_lattice::print.
(ipcp_vr_lattice::meet_with): New.
(ipcp_vr_lattice::meet_with_1): Likewise.
(ipcp_vr_lattice::top_p): Likewise.
(ipcp_vr_lattice::bottom_p): Likewsie.
(ipcp_vr_lattice::set_to_bottom): Likewise.
(set_all_contains_variable): Call VR set_to_bottom.
(initialize_node_lattices): Init VR lattices.
(propagate_vr_accross_jump_function): New.
(propagate_constants_accross_call): Call
propagate_vr_accross_jump_function.
(ipcp_store_vr_results): New.
(ipcp_driver): Handle VR.
* ipa-prop.c (ipa_print_node_jump_functions_for_edge): Handle VR.
(ipa_set_jf_unknown): Likewise.
(ipa_compute_jump_functions_for_edge): Likewise.
(ipa_node_params_t::duplicate): Likewise.
(ipa_write_jump_function): Likewise.
(ipa_read_jump_function): Likewise.
(write_ipcp_transformation_info): Likewise.
(read_ipcp_transformation_info): Likewise.
(ipcp_update_vr): New.
(ipcp_transform_function): Handle VR.
* ipa-prop.h (struct ipa_vr): New.
* cgraph.c: Include tree-vrp.h.
* cgraphunit.c: Likewise.
* ipa-utils.c: Likewise.
* ipa.c: Likewise.

From-SVN: r240292

Add Early VRP

gcc/ChangeLog:

2016-09-21  Kugan Vivekanandarajah  <kuganv@linaro.org>

* doc/invoke.texi: Document -fdump-tree-evrp.
* passes.def: Define new pass_early_vrp.
* timevar.def: Define new TV_TREE_EARLY_VRP.
* tree-pass.h (make_pass_early_vrp): New.
* tree-ssa-propagate.c: Make replace_uses_in non static.
* tree-ssa-propagate.h: Export replace_uses_in.
* tree-vrp.c (extract_range_for_var_from_comparison_expr): New.
(extract_range_from_assert): Factor out
extract_range_for_var_from_comparison_expr.
(vrp_initialize_lattice): New.
(vrp_initialize): Factor out vrp_initialize_lattice.
(vrp_valueize): Fix it to reject complex value ranges.
(vrp_free_lattice): New.
(evrp_dom_walker::before_dom_children): Likewise.
(evrp_dom_walker::after_dom_children): Likewise.
(evrp_dom_walker::push_value_range): Likewise.
(evrp_dom_walker::pop_value_range): Likewise.
(execute_early_vrp): Likewise.
(execute_vrp): Call vrp_initialize_lattice and
vrp_free_lattice.
(make_pass_early_vrp): New.

gcc/testsuite/ChangeLog:

2016-09-21  Kugan Vivekanandarajah  <kuganv@linaro.org>

* g++.dg/tree-ssa/pr31146-2.C: Run with -fno-tree-evrp as evrp also
does the same transformation.
* g++.dg/warn/pr33738.C: XFAIL as optimization now happens in ccp.
* gcc.dg/tree-ssa/evrp1.c: New test.
* gcc.dg/tree-ssa/evrp2.c: New test.
* gcc.dg/tree-ssa/evrp3.c: New test.
* gcc.dg/tree-ssa/pr20657.c: Check for the pattern in evrp dump.
* gcc.dg/tree-ssa/pr22117.c: Likewise.
* gcc.dg/tree-ssa/pr61839_2.c: Likewise.
* gcc.dg/tree-ssa/pr64130.c: Likewise.
* gcc.dg/tree-ssa/pr37508.c: Change the pattern to be checked as
foling now happens early.
* gcc.dg/tree-ssa/vrp04.c: Likewise.
* gcc.dg/tree-ssa/vrp06.c: Likewise.
* gcc.dg/tree-ssa/vrp16.c: Likewise.
* gcc.dg/tree-ssa/vrp25.c: Likewise.
* gcc.dg/tree-ssa/vrp67.c: Likewise.

From-SVN: r240291

Report DejaGnu ERROR messages in dg-extract-results.

2016-09-20  Christophe Lyon  <christophe.lyon@linaro.org>

* dg-extract-results.py: Report DejaGnu error in the final
summary.
* dg-extract-results.sh: Likewise.

From-SVN: r240289

Report DejaGnu ERROR messages in compare_tests

2016-09-20  Christophe Lyon  <christophe.lyon@linaro.org>

* compare_tests: Take ERROR messages into account when
          comparing.

From-SVN: r240288

trans-intrinsic.c (conv_expr_ref_to_caf_ref): Adjust fall through comment.

* trans-intrinsic.c (conv_expr_ref_to_caf_ref): Adjust fall through
comment.

From-SVN: r240286

syscall: build export_unix_test.go on solaris

    Patch from Rainer Orth.

    Reviewed-on: https://go-review.googlesource.com/29436

From-SVN: r240285

i386.md (mult->ashift peephole2s): Use pow2p_hwi instead of exact_log2.

* config/i386/i386.md (mult->ashift peephole2s): Use pow2p_hwi
instead of exact_log2.

From-SVN: r240281

re PR go/77625 (go/gofrontend/ast-dump.cc:169:42: error: ‘new’ of type ‘std::ofstr eam {aka std::basic_ofstream<char>}’ with extended alignment 16)

PR go/77625

    compiler: allocate std::ofstream as a local variable

    GCC PR 77625 points out a warning about new std::ofstream.  I don't know
    how that is supposed to work, but in this case the std::ofstream may as
    well be a local variable anyhow.

    Reviewed-on: https://go-review.googlesource.com/29435

From-SVN: r240279

re PR tree-optimization/77621 (Internal compiler error for mtune=atom + msse2)

PR target/77621
* config/i386/i386.c (ix86_preferred_simd_mode) <case DFmode>:
Don't return word_mode for !TARGET_VECTORIZE_DOUBLE.
(ix86_add_stmt_cost): Penalize DFmode vector operations
for !TARGET_VECTORIZE_DOUBLE.

testsuite/ChangeLog:

PR target/77621
* gcc.target/i386/pr77621.c: New test.
* gcc.target/i386/vect-double-2.c: Update scan-tree-dump-times
pattern, loop should vectorize with -mtune=atom.

From-SVN: r240277

re PR go/77642 (GO Bootstrap fail starting with r239872 splitstack signature does not match)

PR go/77642

    runtime: pass correct type to __splitstack_find

    The code was passing uintptr* to a function that expected size_t*.

    Based on patch by Andreas Krebbel.

    Fixes GCC PR 77642.

    Reviewed-on: https://go-review.googlesource.com/29433

From-SVN: r240275