fptr.c (SIGN_EXTEND): Cast -1 to unsigned.

* config/pa/fptr.c (SIGN_EXTEND): Cast -1 to unsigned.

From-SVN: r227717

install.texi (Downloading the source): Mark up contrib/download_prerequisites properly and drop leading "./".

* doc/install.texi (Downloading the source): Mark up
contrib/download_prerequisites properly and drop leading "./".

From-SVN: r227716

Daily bump.

From-SVN: r227715

stop including tm.h in sendmsg.c

libobjc/ChangeLog:

2015-09-12  Trevor Saunders  <tbsaunde+gcc@tbsaunde.org>

PR libobjc/24775
* sendmsg.c (tm.h): Remove include.

From-SVN: r227712

remove unused defines from sendmsg.c

libobjc/ChangeLog:

2015-09-12  Trevor Saunders  <tbsaunde+gcc@tbsaunde.org>

PR libobjc/24775
* sendmsg.c (gen_rtx): Remove macro.
(gen_rtx_MEM): Likewise.
(gen_rtx_REG): Likewise.
(rtx): Likewise.

From-SVN: r227711

remove STRUCT_VALUE macro

This macro was converted to the TARGET_STRUCT_VALUE_RTX hook many years
ago, however there are still some lingering definitions, and a use in
libobjc.  All the remaining definitions define the macro to 0, which
libobjc treats the same as undefined, so it won't break anything else to
have libobjc stop checking the macro.  However it may be that this part
of libobjc has been broken for a long time on targets that only define
the hook, but that is a separate issue.

gcc/ChangeLog:

2015-09-12  Trevor Saunders  <tbsaunde+gcc@tbsaunde.org>

* config/arc/arc.h: Remove define of STRUCT_VALUE.
* config/lm32/lm32.h: Likewise.
* config/mep/mep.h: Likewise.
* config/visium/visium.h: Likewise.
* system.h: Poison STRUCT_VALUE macro.

libobjc/ChangeLog:

2015-09-12  Trevor Saunders  <tbsaunde+gcc@tbsaunde.org>

PR libobjc/24775
* sendmsg.c: Remove check of STRUCT_VALUE macro.

From-SVN: r227710

re PR ada/66965 (gnat.dg/specs/addr1.ads obsolete -- failing on trunk)

PR ada/66965
* gnat.dg/specs/addr1.ads: Remove.

From-SVN: r227709

pa.c (pa_output_move_double): Enhance to handle HIGH CONSTANT_P operands.

* config/pa/pa.c (pa_output_move_double): Enhance to handle HIGH
CONSTANT_P operands.

From-SVN: r227706

re PR libfortran/67527 (io.h sanitizer complains on 1 << 31)

PR libfortran/67527
PR libfortran/67535
PR libfortran/67536
* io/io.h: Use unsigned values for 31-bit left shifts.
* io/unix.c (buf_read): Do not call memcpy() with NULL pointer arg.
* io/write.c (nml_write_obj): Likewise.

From-SVN: r227705

Daily bump.

From-SVN: r227704

Convert SPARC backend over to LRA.

gcc/

* config/sparc/constraints.md: Make "U" constraint a real register
constraint.
* config/sparc/sparc.c (TARGET_LRA_P): Define.
(D_MODES, DF_MODES): Add missing cast.
(TF_MODES, TF_MODES_NO_S): Include T_MODE.
(OF_MODES, OF_MODES_NO_S): Include O_MODE.
(sparc_register_move_cost): Decrease Niagara/UltrsSPARC memory
cost to 8.
* config/sparc/sparc.h (PROMOTE_MODE): Define.
* config/sparc/sparc.md (*movsi_lo_sum, *movsi_high): Do not
provide these insn when flag_pic.

From-SVN: r227701

compiler: Avoid unsafe memcmp for nointerface comments.

    Fixes golang/go#11577.

    Reviewed-on: https://go-review.googlesource.com/14182

From-SVN: r227699

[PATCH] Another small cleanup to the const_and_copies stack

2015-09-11  Jeff Law  <law@redhat.com>

PR tree-optimization/47679
* tree-ssa-dom.c (struct cond_equivalence): Update comment.
* tree-ssa-scopedtables.h (class const_and_copies): Prefix data
member with m_.  Update inline member functions as necessary.  Add
toplevel comment.
* tree-ssa-scopedtables.c: Update const_and_copies's member
functions to use m_ prefix to access the stack.

From-SVN: r227697

Remove dead code from graphite-optimize-isl.c

The variable `static bool enable_polly_vector' is always assigned to false.
This results in dead code in optimize-isl.c.  No functional change intended.

Passes bootstrap and regtest.

gcc/ChangeLog:

2015-09-11  Aditya Kumar  <aditya.k7@samsung.com>

* graphite-optimize-isl.c (disable_tiling): Remove.
(get_schedule_for_band): Do not use disable_tiling.
(get_prevector_map): Delete function.
(enable_polly_vector): Remove.
(get_schedule_for_band_list): Remove dead code.

From-SVN: r227696

Refactor optimize isl

Refactor graphite-optimize-isl.c. Renamed function name, variable names etc.,
and indented the source according to gcc style guidelines.  Modified comments
accordingly. No functional change intended.

Passes regtest and bootstap on x86_64.

gcc/ChangeLog:

2015-09-10  Aditya Kumar  <aditya.k7@samsung.com>

        * graphite-optimize-isl.c (get_tile_map): Refactor.
        (get_schedule_for_band): Same.
        (getScheduleForBand): Same.
        (get_prevector_map): Same.
        (get_schedule_for_band_list): Same.
        (get_schedule_map): Same.
        (get_single_map): Same.
        (apply_schedule_map_to_scop): Same.
        (optimize_isl): Same.

From-SVN: r227695

Makefile.am (backtrace.lo): Depend on internal.h.

* Makefile.am (backtrace.lo): Depend on internal.h.
(sort.lo, stest.lo): Add explicit dependencies.
* Makefile.in: Rebuild.

From-SVN: r227691

Fix filesystem::canonical on Solaris 10.

PR libstdc++/67173
* src/filesystem/ops.cc (filesystem::canonical): Allocate buffer for
realpath on Solaris 10.

From-SVN: r227689

Check read() result in std::random_device.

PR libstdc++/65142
* src/c++11/random.cc (random_device::_M_getval()): Check read result.

From-SVN: r227687

Fix invalid UTF-8 in wchar_t tests.

2015-09-11  John Marino  <gnugcc@marino.st>
    Jonathan Wakely  <jwakely@redhat.com>

PR libstdc++/67096
* testsuite/22_locale/codecvt/in/wchar_t/4.cc: Do not test code points
above U+10FFFF.
* testsuite/22_locale/codecvt/in/wchar_t/8.cc: Likewise.
* testsuite/22_locale/codecvt/in/wchar_t/9.cc: Likewise.
* testsuite/22_locale/codecvt/length/wchar_t/4.cc: Likewise.
* testsuite/22_locale/codecvt/out/wchar_t/4.cc: Likewise.
* testsuite/22_locale/codecvt/unshift/wchar_t/4.cc: Likewise.
* testsuite/27_io/basic_filebuf/seekoff/wchar_t/1.cc: Likewise.
* testsuite/27_io/basic_filebuf/seekpos/wchar_t/9874.cc: Likewise.
* testsuite/27_io/basic_filebuf/underflow/wchar_t/1.cc: Likewise.
* testsuite/27_io/basic_filebuf/underflow/wchar_t/2.cc: Likewise.
* testsuite/27_io/basic_filebuf/underflow/wchar_t/3.cc: Likewise.
* testsuite/27_io/objects/wchar_t/10.cc: Likewise.
* testsuite/27_io/objects/wchar_t/11.cc: Likewise.
* testsuite/27_io/objects/wchar_t/12.cc: Likewise.
* testsuite/27_io/objects/wchar_t/13.cc: Likewise.

Co-Authored-By: Jonathan Wakely <jwakely@redhat.com>
From-SVN: r227686

Rationalise PCH headers and 17_intro/headers tests.

PR libstdc++/64857
* doc/xml/manual/using.xml: Improve aggregate header documentation.
* doc/html/manual/*: Regenerate.
* include/precompiled/extc++.h: Include <bits/stdc++.h> for C++11
and later and include more extension headers.
* testsuite/17_intro/headers/c++1998/all_attributes.cc: Remove
redundant header.
* testsuite/17_intro/headers/c++200x/: Rename to c++2011.
* testsuite/17_intro/headers/c++2014/all_attributes.cc: Remove
redundant headers.
* testsuite/17_intro/headers/c++2014/all_no_exceptions.cc: New.
* testsuite/17_intro/headers/c++2014/all_no_rtti.cc: New.
* testsuite/17_intro/headers/c++2014/all_pedantic_errors.cc: New.
* testsuite/17_intro/headers/c++2014/operator_names.cc: New.
* testsuite/17_intro/headers/c++2014/stdc++.cc: New.
* testsuite/17_intro/headers/c++2014/stdc++_multiple_inclusion.cc:
New.

From-SVN: r227684

Link gcc.dg/pie-link.c with -pie

* gcc.dg/pie-link.c: Add -pie to dg-options.

From-SVN: r227682

Implement N4258 noexcept for std::basic_string.

PR libstdc++/58265
* doc/xml/manual/intro.xml: Document LWG 2063 and 2064 resolutions.
* doc/html/manual/bugs.html: Regenerate.
* include/bits/basic_string.h (basic_string): Implement N4258. Add
correct exception-specifications and propagate allocators correctly.
* include/bits/basic_string.tcc (basic_string::swap): Propagate
allocators correctly.
* include/debug/string (__gnu_debug::basic_string): Add correct
exceptions-specifications and allcoator-extended constructors.
* testsuite/21_strings/basic_string/allocator/char/copy.cc: New.
* testsuite/21_strings/basic_string/allocator/char/copy_assign.cc:
New.
* testsuite/21_strings/basic_string/allocator/char/minimal.cc: New.
* testsuite/21_strings/basic_string/allocator/char/move.cc: New.
* testsuite/21_strings/basic_string/allocator/char/move_assign.cc:
New.
* testsuite/21_strings/basic_string/allocator/char/noexcept.cc: New.
* testsuite/21_strings/basic_string/allocator/char/swap.cc: New.
* testsuite/21_strings/basic_string/allocator/wchar_t/copy.cc: New.
* testsuite/21_strings/basic_string/allocator/wchar_t/copy_assign.cc:
New.
* testsuite/21_strings/basic_string/allocator/wchar_t/minimal.cc: New.
* testsuite/21_strings/basic_string/allocator/wchar_t/move.cc: New.
* testsuite/21_strings/basic_string/allocator/wchar_t/move_assign.cc:
New.
* testsuite/21_strings/basic_string/allocator/wchar_t/noexcept.cc: New.
* testsuite/21_strings/basic_string/allocator/wchar_t/swap.cc: New.
* testsuite/util/testsuite_allocator.h (tracker_allocator): Define
defaulted assignment operators.

From-SVN: r227681

Allocator-extended constructors for container adaptors.

PR libstdc++/65092
* include/bits/stl_queue.h (queue, priority_queue): Add
allocator-extended constructors.
* include/bits/stl_stack.h (stack): Likewise.
* testsuite/23_containers/priority_queue/requirements/
uses_allocator.cc: Test allocator-extended constructors.
* testsuite/23_containers/queue/requirements/uses_allocator.cc:
Likewise.
* testsuite/23_containers/stack/requirements/uses_allocator.cc:
Likewise.

From-SVN: r227680

Remove separate movtf pattern - Use an iterator for all FP modes.

movtf is unnecessary as a separate expander. Move this to be with
the standard scalar floating point expanders.

Achieved by adding a new iterator and then using the same.

Tested cross aarch64-none-elf and no regressions.

Rebased version from https://gcc.gnu.org/ml/gcc-patches/2015-09/msg00767.html

2015-09-10  Ramana Radhakrishnan  <ramana.radhakrishnan@arm.com>

PR target/63304
        * config/aarch64/aarch.md (mov<mode>:GPF_F16): Use GPF_TF_F16.
        (movtf): Delete.
        * config/aarch64/iterators.md (GPF_TF_F16): New.
        (GPF_F16): Delete.

From-SVN: r227679

pr63210.c (dg-skip-if): Skip armv4t.

gcc/testsuite

* gcc.target/arm/pr63210.c (dg-skip-if): Skip armv4t.
(dg-additional-options): Add -march=armv5t if arm_arch_v5t_ok.

From-SVN: r227677

Make sure that contrib/download_prerequisites is run from correct place

* download_prerequisites: Make sure that script is run from
top level source directory.

From-SVN: r227674

libgo/runtime: return 0, not NULL, from main

    Reviewed-on: https://go-review.googlesource.com/13421

From-SVN: r227673

Daily bump.

From-SVN: r227671

nvptx.c (nvptx_expand_call): Add spacing.

* config/nvptx/nvptx.c (nvptx_expand_call): Add spacing.
(nvptx_reorg): Adjust comments.

From-SVN: r227667

re PR bootstrap/67363 (r227188 breaks build for mingw-w64)

PR bootstrap/67363
* configure.ac: Check if setenv and unsetenv are declared.
* configure: Rebuild.
* config.in: Rebuild.
* system.h: Declare setenv and unsetenv if not declared.

From-SVN: r227666

libgo: Use stat_atim.go on Solaris 12+

    From Rainer Orth.

    Solaris 12 changes the stat_[amc]tim members of struct stat from
    timestruc_t to timespec_t for XPG7 compatiblity, thus breaking the libgo
    build.  The following patch checks for this change and uses the common
    stat_atim.go if appropriate.

    Reviewed-on: https://go-review.googlesource.com/14495

From-SVN: r227665

rs6000.c (swap_web_entry): Update preceding commentary to simplify permute mask adjustment equation.

[gcc]

2015-09-10  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>

* config/rs6000/rs6000.c (swap_web_entry): Update preceding
commentary to simplify permute mask adjustment equation.
(special_handling_values): Add SH_VPERM.
(const_load_sequence_p): New function.
(insn_is_swappable_p): Add logic to recognize an UNSPEC_VPERM with
the mask loaded from the constant pool.
(adjust_vperm): New function.
(handle_special_swappables): Call adjust_vperm.
(dump_swap_insn_table): Handle SH_VPERM.

[gcc/testsuite]

2015-09-10  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>

* gcc.target/powerpc/swaps-p8-20.c: New test.
* gcc.target/powerpc/swaps-p8-21.c: New test.

From-SVN: r227664

* testsuite/util/testsuite_allocator.h (PointerBase::operator[]): Add.

From-SVN: r227661

1.cc: Verify the string contents.

* testsuite/21_strings/basic_string/operators/char/1.cc: Verify the
string contents.

From-SVN: r227660

Put back requires_stack_frame_p used by i386.c

* shrink-wrap.c (requires_stack_frame_p): Remove static.
* shrink-wrap.h (requires_stack_frame_p): Put back.

From-SVN: r227659

Stop reload1.c warning from breaking bootstrap

Some host-target combinations get bogus warnings about orig_dup
being used uninitialized.  I tried to coerce the current uninit
pass to handle this case, but the patch I ended up with only worked
by accident because of the strange way in which the pass handles
limit cases.  (If we have more than MAX_NUM_CHAINS chains, it silently
drops the excess chains and continues regardless, so it's quite easy
to come up with cases where the predicates for either the definition
or the use consider an arbitrary subset of the actual conditions.)

For now this patch turns -Wmaybe-uninitialized into a warning for
the affacted function.  It will be a warning even if someone turns
off warnings on the command line, but I don't think that's important.

Bootstrapped and regression-tested on x86_64-linux-gnu.  Also tested
with a cross-compiler to sparc-linux-gnu (which also triggered the
warning for me).

gcc/
* reload1.c (elimination_costs_in_insn): Locally turn
-Wmaybe-uninitialized into a warning.

From-SVN: r227658

compiler: Report errors from very large types.

    The gcc backend throws an internal error when trying to get the size
    of a type which is larger than the amount of address space on the
    machine.  This patch catches this error and reports it in a user
    friendly way.

    Fixes golang/go#11554.

    Reviewed-on: https://go-review.googlesource.com/13684

* go-gcc.cc (Gcc_backend::type_size): Return -1 for
unrepresentable size.

From-SVN: r227656

shrink-wrap: Header hygiene

2015-09-10  Segher Boessenkool  <segher@kernel.crashing.org>

* shrink-wrap.c (requires_stack_frame_p): Make static.
(prepare_shrink_wrap): Likewise.
(dup_block_and_redirect): Likewise.
* shrink-wrap.h: Remove declarations of those functions.

From-SVN: r227652

re PR fortran/67526 (ICE on missing end parenthesis in substring construct)

2015-09-09  Steven G. Kargl  <kargl@gcc.gnu.org>

PR fortran/67526
* gfortran.dg/pr67526.f90: New test.

2015-09-09  Steven G. Kargl  <kargl@gcc.gnu.org>

PR fortran/67526
* expr.c (gfc_check_init_expr): Do not dereference a NULL pointer.

From-SVN: r227651

re PR c++/67318 (Parsing error when using abbreviated integral type names in template parameter pack declaration)

/cp
2015-09-10  Paolo Carlini  <paolo.carlini@oracle.com>

PR c++/67318
* parser.c (cp_parser_parameter_declaration): Consume the ellipsis
and set template_parameter_pack_p also when the type is null.

/testsuite
2015-09-10  Paolo Carlini  <paolo.carlini@oracle.com>

PR c++/67318
* g++.dg/cpp0x/variadic166.C: New.

From-SVN: r227650

Warn when comparing nonnull arguments to NULL in a function.

If an argument is marked as nonnull then passing in a NULL argument
will produce bad results even if the code checks against NULL. GCC
might optimize such checks away so warn the user when the function
contains such comparisons.

nn.c: In function ‘foo’:
nn.c:6:27: warning: nonnull argument ‘bar’ compared to NULL [-Wnonnull]
 void foo(void *bar) { if (!bar) abort(); }
                           ^
gcc/ChangeLog

* doc/invoke.texi (Wnonnull): Also warns when comparing against NULL.

gcc/c/ChangeLog

       * c-typeck.c (build_binary_op): Check and warn when nonnull arg
       parm against NULL.

gcc/cp/ChangeLog

       * typeck.c (cp_build_binary_op): Check and warn when nonnull arg
       parm against NULL.

gcc/testsuite/ChangeLog

       * c-c++-common/nonnull-1.c: New test.

From-SVN: r227649

re PR fortran/66993 (Spurious ambiguous symbol error with submodules)

2015-09-10  Paul Thomas  <pault@gcc.gnu.org>

PR fortran/66993
* module.c (read_module): If a symtree exists and the symbol has
been associated in a submodule from a parent (sub)module, attach
the symbol to a 'unique symtree' and the new symbol to the
existing symtree.

2015-09-10  Paul Thomas  <pault@gcc.gnu.org>

PR fortran/66993
* gfortran.dg/submodule_11.f08: New test.

From-SVN: r227648

re PR target/67506 ([SH]: error: unrecognizable insn when compiling texlive-binaries)

gcc/
PR target/67506
* config/sh/sh.c (sh_extending_set_of_reg::use_as_extended_reg): Add
missing simplify_gen_subreg.

gcc/testsuite/
PR target/67506
* gcc.c-torture/compile/pr67506.c: New test.

From-SVN: r227646

S/390: Don't use vgm for v1ti and v1tf.

gcc/ChangeLog:

2015-09-10  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>

* config/s390/s390.c (s390_contiguous_bitmask_vector_p): Reject if
the vector element is bigger than 64 bit.

gcc/testsuite/ChangeLog:

2015-09-10  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>

* gcc.target/s390/vector/vec-genbytemask-1.c: Add check for V1TI
initialization with a byte mask.  No change expected here.
* gcc.target/s390/vector/vec-genmask-1.c: Fix whitespace.
* gcc.target/s390/vector/vec-genmask-2.c: Add check for V1TI
initialization with contigious bitmask.  Literal pool is expectd
to be used here.

From-SVN: r227637

S/390: Fix mode iterators vmal, vmah, and vmalh.

gcc/ChangeLog:

2015-09-10  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>

* config/s390/vx-builtins.md ("vec_vmal<mode>", "vec_vmah<mode>")
("vec_vmalh<mode>"): Change mode iterator from VI_HW to VI_HW_QHS.

From-SVN: r227636

S/390: Add V1TImode to constant pool modes.

gcc/ChangeLog:

2015-09-10  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>

* config/s390/s390.c: Add V1TImode to constant pool modes.

From-SVN: r227635

[ARM] PR 67439: Allow matching of *arm32_movhf when -mrestrict-it is on

PR target/67439
* config/arm/arm.md (*arm32_movhf): Remove !arm_restrict_it from
predicate.  Set predicable_short_it attr to "no".

* gcc.target/arm/pr67439_1.c: New test.

From-SVN: r227630

[Patch/expand] Cost instruction sequences when doing left wide shift

Patch background details:

    https://gcc.gnu.org/ml/gcc-patches/2015-08/msg01147.html

gcc/
  PR rtl-optimization/67421
  * expr.c (expand_expr_real_2): Cost instrcution sequences when doing left wide
  shift tranformation.

From-SVN: r227629

arc-common.c: Remove references to A5.

2015-09-10  Claudiu Zissulescu  <claziss@synopsys.com>

        * common/config/arc/arc-common.c: Remove references to A5.
        * config/arc/arc-opts.h: Likewise.
        * config/arc/arc.c, config/arc/arc.h, config/arc/arc.md: Likewise.
        * config/arc/arc.opt, config/arc/constraints.md: Likewise.
        * config/arc/t-arc-newlib: Likewise.

From-SVN: r227627

arc.md (length): Fix attribute length for conditional executed instructions with long immediate.

2015-09-10  Claudiu Zissulescu <claziss@synopsys.com>

        * config/arc/arc.md (length): Fix attribute length for conditional
        executed instructions with long immediate.

(posted as: https://gcc.gnu.org/ml/gcc/2015-03/msg00268.html)

From-SVN: r227626

[AArch64] Skip tiny and large code model on gcc.target/aarch64/pic-small.c

gcc/testsuite/
  * gcc.target/aarch64/pic-small.c (dg-skip-if): Skip tiny and large code model.

From-SVN: r227624

[AArch64] Use logics_imm type for 2nd alternative of *and<mode>3nr_compare0

* config/aarch64/aarch64.md (*and<mode>3nr_compare0): Use logics_imm
type for second alternative.

From-SVN: r227623

Fix typo in doc/install.texi

From-SVN: r227620

doc/install.texi: Mention contrib/download_prerequisites

Mention ./contrib/download_prerequisites on the "Downloading GCC" page.

* doc/invoke.texi (Downloading GCC): Mention
contrib/download_prerequisites script.

From-SVN: r227619

re PR c++/67523 (ICE with invalid combined simd inside of a template)

PR c++/67523
* gimplify.c (gimplify_omp_for): If inner stmt is not found
for combined loop, assert seen_error () and return GS_ERROR.

* g++.dg/gomp/pr67523.C: New test.

From-SVN: r227611

re PR c++/67522 (OpenMP ICE in type_dependent_expression_p)

PR c++/67522
* semantics.c (handle_omp_array_sections_1): Only run
type_dependent_expression_p on VAR_DECL/PARM_DECLs.
(finish_omp_clauses) <case OMP_CLAUSE_LINEAR>: Likewise.
Don't adjust OMP_CLAUSE_LINEAR_STEP if OMP_CLAUSE_DECL
is not a VAR_DECL/PARM_DECL.

* g++.dg/gomp/pr67522.C: New test.

From-SVN: r227610

re PR middle-end/67521 (ICE when OpenMP loop expressions mention the IV)

PR middle-end/67521
* gimplify.c (gimplify_omp_for): Don't call omp_add_variable
if decl is already in outer->variables.

* c-c++-common/gomp/pr67521.c: New test.

From-SVN: r227609

re PR middle-end/67517 (ICE in gimplify_scan_omp_clauses)

PR middle-end/67517
* gimplify.c (gimplify_scan_omp_clauses): Instead of
asserting that decl is not specified in octx->variables,
break out of the loop if it is.

* c-c++-common/gomp/pr67517.c: New test.

From-SVN: r227608

re PR c++/67514 (ICE in omp_add_variable)

PR c++/67514
* gimplify.c (gimplify_omp_for): For loop SIMD construct, if
iterator is not explicitly determined, but is defined inside
of the combined workshare region, handle it like if it has
DECL_EXPR in OMP_FOR_PRE_BODY.

* g++.dg/gomp/pr67514.C: New test.

From-SVN: r227607

re PR c++/67511 (ICE with invalid OpenMP random access iterator)

PR c++/67511
* semantics.c (handle_omp_for_class_iterator): Don't wrap
error_mark_node into a NOP_EXPR to void_type_node.

* g++.dg/gomp/pr67511.C: New test.

From-SVN: r227606

re PR c/67502 (ICE with collapsed for simd loop inside of parallel)

PR c/67502
* c-parser.c (c_parser_omp_for_loop): Emit DECL_EXPR stmts
into OMP_FOR_PRE_BODY rather than before the loop.

* c-c++-common/gomp/pr67502.c: New test.

From-SVN: r227605

compiler: Don't allow shifts with non-integers.

    Fixes golang/go#11616.

    Reviewed-on: https://go-review.googlesource.com/13688

From-SVN: r227604

Daily bump.

From-SVN: r227603

nvptx.md (call_operation): Move bound out of loop.

* config/nvptx/nvptx.md (call_operation): Move bound out of loop.
(*cmp<mode>): Add assembler spacing.
(setcc_int<mode>, set_cc_float<mode>): Likewise.
* config/nvptx/nvptx.c (nvptx_option_override): Override debug
level.
(write_func_decl_from_insn): Refactor argument loops & comma emission.
(nvptx_expand_call): Likewise.
(nvptx_output_call_insn): Likewise.
(nvptx_reorg_subreg): Add spacing.

From-SVN: r227597

Enable lightweight checks with _GLIBCXX_ASSERTIONS.

* doc/xml/manual/using.xml (_GLIBCXX_ASSERTIONS): Document.
* doc/html/manual/using_macros.html: Regenerate.
* include/bits/c++config: Define _GLIBCXX_ASSERTIONS when
_GLIBCXX_DEBUG is defined. Disable std::string extern templates when
(_GLIBCXX_EXTERN_TEMPLATE, __glibcxx_assert): Depend on
_GLIBCXX_ASSERTIONS instead of _GLIBCXX_DEBUG.
* include/debug/debug.h [!_GLIBCXX_DEBUG]: Define
__glibcxx_requires_non_empty_range and __glibcxx_requires_nonempty.
* include/backward/auto_ptr.h (auto_ptr::operator*,
auto_ptr::operator->): Replace _GLIBCXX_DEBUG_ASSERT with
__glibcxx_assert.
* include/bits/basic_string.h (basic_string::operator[],
basic_string::front, basic_string::back, basic_string::pop_back):
Likewise.
* include/bits/random.h
(uniform_int_distribution::param_type::param_type,
uniform_real_distribution::param_type::param_type,
normal_distribution::param_type::param_type,
gamma_distribution::param_type::param_type,
bernoulli_distribution::param_type::param_type,
binomial_distribution::param_type::param_type,
geometric_distribution::param_type::param_type,
negative_binomial_distribution::param_type::param_type,
poisson_distribution::param_type::param_type,
exponential_distribution::param_type::param_type): Likewise.
* include/bits/regex.h (match_results::operator[],
match_results::prefix, match_results::suffix): Likewise.
* include/bits/regex.tcc (format, regex_iterator::operator++):
Likewise.
* include/bits/regex_automaton.tcc (_StateSeq::_M_clone): Likewise.
* include/bits/regex_compiler.tcc (_Compiler::_Compiler,
_Compiler::_M_insert_character_class_matcher): Likewise.
* include/bits/regex_executor.tcc (_Executor::_M_dfs): Likewise.
* include/bits/regex_scanner.tcc (_Scanner::_M_advance,
_Scanner::_M_scan_normal): Likewise.
* include/bits/shared_ptr_base.h (__shared_ptr::_M_reset,
__shared_ptr::operator*): Likewise.
* include/bits/stl_iterator_base_funcs.h (__advance): Likewise.
* include/bits/unique_ptr.h (unique_ptr::operator*,
unique_ptr::operator[]): Likewise.
* include/experimental/fs_path.h (path::path(string_type, _Type),
path::iterator::operator++, path::iterator::operator--,
path::iterator::operator*): Likewise.
* include/experimental/string_view (basic_string_view::operator[],
basic_string_view::front, basic_string_view::back,
basic_string_view::remove_prefix): Likewise.
* include/ext/random (beta_distribution::param_type::param_type,
normal_mv_distribution::param_type::param_type,
rice_distribution::param_type::param_type,
pareto_distribution::param_type::param_type,
k_distribution::param_type::param_type,
arcsine_distribution::param_type::param_type,
hoyt_distribution::param_type::param_type,
triangular_distribution::param_type::param_type,
von_mises_distribution::param_type::param_type,
hypergeometric_distribution::param_type::param_type,
logistic_distribution::param_type::param_type): Likewise.
* include/ext/vstring.h (__versa_string::operator[]): Likewise.
* include/std/complex (polar): Likewise.
* include/std/mutex [!_GTHREAD_USE_MUTEX_TIMEDLOCK]
(timed_mutex::~timed_mutex, timed_mutex::unlock,
(recursive_timed_mutex::~timed_mutex, recursive_timed_mutex::unlock):
Likewise.
* include/std/shared_mutex [!PTHREAD_RWLOCK_INITIALIZER]
(__shared_mutex_pthread::__shared_mutex_pthread,
__shared_mutex_pthread::~__shared_mutex_pthread): Likewise.
(__shared_mutex_pthread::lock, __shared_mutex_pthread::try_lock,
__shared_mutex_pthread::unlock, __shared_mutex_pthread::lock_shared,
__shared_mutex_pthread::try_lock_shared): Likewise.
(__shared_mutex_cv::~__shared_mutex_cv, __shared_mutex_cv::unlock,
__shared_mutex_cv::unlock_shared): Likewise.
(shared_timed_mutex::try_lock_until,
shared_timed_mutex::try_lock_shared_until): Likewise.
* include/std/valarray (valarray::valarray(const _Tp*, size_t),
valarray::operator=, valarray::sum, valarray::min, valarray::max,
_DEFINE_VALARRAY_AUGMENTED_ASSIGNMENT, _DEFINE_BINARY_OPERATOR):
Likewise.

From-SVN: r227595

re PR middle-end/67512 (internal compiler error: in invert_tree_comparison, at fold-const.c:2456)

PR middle-end/67512
* tree-ssa-uninit.c (pred_equal_p): Only call invert_tree_comparison
for comparisons.

* gcc.dg/pr67512.c: New test.

From-SVN: r227594

re PR c++/53184 (Unnecessary anonymous namespace warnings)

2015-09-09  Paolo Carlini  <paolo.carlini@oracle.com>

PR c++/53184
* doc/invoke.texi ([Wsubobject-linkage]): Document.

/c-family
2015-09-09  Paolo Carlini  <paolo.carlini@oracle.com>

PR c++/53184
* c.opt ([Wsubobject-linkage]): Add.

/cp
2015-09-09  Paolo Carlini  <paolo.carlini@oracle.com>

PR c++/53184
* decl2.c (constrain_class_visibility): Use Wsubobject-linkage.

/testsuite
2015-09-09  Paolo Carlini  <paolo.carlini@oracle.com>

PR c++/53184
* g++.dg/warn/Wsubobject-linkage-1.C: New.
* g++.dg/warn/Wsubobject-linkage-2.C: Likewise.
* g++.dg/warn/Wsubobject-linkage-3.C: Likewise.
* g++.dg/warn/Wsubobject-linkage-4.C: Likewise.

From-SVN: r227593

Fix sanitizer/67258 by cherry picking upstream patch

PR sanitizer/67258
* ubsan/ubsan_type_hash.cc: Cherry pick upstream r244101.

Upstraem patch:
commit 1d2477faafda9ad2cc19927b3c31efd22747f013
Author: Alexey Samsonov <vonosmas@gmail.com>
Date:   Wed Aug 5 19:35:46 2015 +0000

    [UBSan] Fix UBSan-vptr false positive.

    Offset from vptr to the start of most-derived object can actually
    be positive in some virtual base class vtables.

    Patch by Stephan Bergmann!

    git-svn-id: https://llvm.org/svn/llvm-project/compiler-rt/trunk@244101 91177308-0d34-0410-b5e6-96231b3b80d8

From-SVN: r227591

Add copyright in gcc/params-list.h

2015-09-09  Tom de Vries  <tom@codesourcery.com>

* params-list.h: Add missing copyright notice.

From-SVN: r227590

nvptx.md (atomic_compare_and_swap<mode>): Use sel_truesi, not andsi.

* config/nvptx/nvptx.md (atomic_compare_and_swap<mode>): Use
sel_truesi, not andsi.

From-SVN: r227587

[ARM][3/3] Expand mod by power of 2

* config/arm/arm.md (*subsi3_compare0): Rename to...
(subsi3_compare0): ... This.
(modsi3): New define_expand.
* config/arm/arm.c (arm_new_rtx_costs, MOD case): Handle case
when operand is power of 2.

* gcc.target/aarch64/mod_2.x: New file.
* gcc.target/aarch64/mod_256.x: Likewise.
* gcc.target/arm/mod_2.c: New test.
* gcc.target/arm/mod_256.c: Likewise.
* gcc.target/aarch64/mod_2.c: Likewise.
* gcc.target/aarch64/mod_256.c: Likewise.

From-SVN: r227586

[AArch64][1/3] Expand signed mod by power of 2 using CSNEG

* config/aarch64/aarch64.md (mod<mode>3): New define_expand.
(*neg<mode>2_compare0): Rename to...
(neg<mode>2_compare0): ... This.
* config/aarch64/aarch64.c (aarch64_rtx_costs, MOD case):
Move check for speed inside the if-then-elses.  Reflect
CSNEG sequence in MOD by power of 2 case.

From-SVN: r227585

re PR c++/67504 (ICE with type dependent collapse argument)

PR c++/67504
* parser.c (cp_parser_omp_clause_collapse): Test tree_fits_shwi_p
before INTEGRAL_TYPE_P test.

* g++.dg/gomp/pr67504.C: New test.

From-SVN: r227579

re PR c/67501 (Bad error recovery for invalid OpenMP clauses in C FE)

PR c/67501
* c-parser.c (c_parser_oacc_all_clauses,
c_parser_omp_all_clauses): Remove invalid clause from
list of clauses even if parser->error is set.

* c-c++-common/gomp/pr67501.c: New test.

From-SVN: r227578

re PR c/67500 (OpenMP ICE with invalid safelen/simdlen/alignment expressions)

PR c/67500
* c-parser.c (c_parser_omp_clause_aligned,
c_parser_omp_clause_safelen, c_parser_omp_clause_simdlen): Fix up
test for errors.
* c-decl.c (temp_pop_parm_decls): Allow b->decl equal to
error_mark_node.

* gcc.dg/gomp/pr67500.c: New test.

From-SVN: r227577

re PR c/67495 (#pragma omp atomic ICEs)

PR c/67495
* c-parser.c (c_parser_omp_atomic): Use c_parser_cast_expression
instead of c_parser_unary_expression.  If the result is !lvalue_p,
wrap the result of c_fully_fold into NON_LVALUE_EXPR.

* gcc.dg/gomp/pr67495.c: New test.

From-SVN: r227576

Fix PowerPC ICE due to secondary_reload ignoring reload replacements

The reason for this PR is that insns emitted by secondary reload
patterns are being generated without taking into account other reloads
that may have occurred.  We run into this problem when an insn has a
pseudo that doesn't get a hard reg, and the pseudo is used in a way
that requires a secondary reload.  In this case the secondary reload
is needed due to gcc generating a 64-bit gpr load from memory insn
with an address offset not a multiple of 4.

PR target/67378
* config/rs6000/rs6000.c (rs6000_secondary_reload_gpr): Find
reload replacement for PRE_MODIFY address reg.

From-SVN: r227573

fix PR53852: stop ISL after a given number of operations

2015-09-02  Sebastian Pop  <s.pop@samsung.com>

            * config.in: Regenerate.
            * configure: Regenerate.
            * configure.ac (HAVE_ISL_CTX_MAX_OPERATIONS): Detect.
            * graphite-optimize-isl.c (optimize_isl): Stop computation when
            PARAM_MAX_ISL_OPERATIONS is reached.
            * params.def (PARAM_MAX_ISL_OPERATIONS): Add.

            * graphite-dependences.c (extend_schedule): Remove gcc_asserts on
            result equal to isl_stat_ok as the status now can be isl_error_quota.
            (subtract_commutative_associative_deps): Same.
            (compute_deps): Same.

testsuite/
            * gcc.dg/graphite/uns-interchange-12.c: Adjust pattern to pass with
            both isl-0.12 and isl-0.15.
            * gcc.dg/graphite/uns-interchange-14.c: Same.
            * gcc.dg/graphite/uns-interchange-15.c: Same.
            * gcc.dg/graphite/uns-interchange-mvt.c: Same.

From-SVN: r227572

Daily bump.

From-SVN: r227571

* backtrace.c: #include <sys/types.h>.

From-SVN: r227568

Remove limit_scops

This patch removes graphite-scop-detection.c:limit_scops function and fix
related issues arising because of that. The functionality limit_scop was added
as an intermediate step to discard the loops which graphite could not
handle. Removing limit_scop required handling of different cases of loops and
surrounding code.  The scop is now larger so most test cases required 'number of
scops detected' to be fixed. By increasing the size of scop we can now optimize
loops which are 'siblings' of each other. This could enable loop fusion on a
number of loops. Since in the graphite framework we mostly want to opimize
loop-nests/adjacent-loops, we now discard scops with less than 2 loops. We
also discard scops without any data references.

Essentially:
 - Remove limite_scops.
 - Only select scops when there are at least two loops (loop nest or, side by side).
 - Discard loops without data-refs.
 - Fix test cases.

Passes bootstrap and reg-test.

gcc/ChangeLog:

2015-09-02  Aditya Kumar  <hiraditya@msn.com>
            Sebastian Pop  <s.pop@samsung.com>

        * graphite-isl-ast-to-gimple.c (gcc_expression_from_isl_ast_expr_id):
        Return the parameter if it was saved in corresponding
        parameter_rename_map of the region.
        (copy_def): Copy def from sese region to the newly created region.
        (copy_internal_parameters): Copy all the internal parameters defined
        within a region to the newly created region.
        (graphite_regenerate_ast_isl): Copy parameters to the new region before
        translating isl to gimple.
        * graphite-scop-detection.c (graphite_can_represent_loop): Bail out if
          the loop-nest does not have any data-references.
        (build_graphite_scops): Create a scop only when there is at least one
        loop inside it.
        (contains_only_close_phi_nodes): Deleted.
        (print_graphite_scop_statistics): Deleted
        (print_graphite_statistics): Deleted
        (limit_scops): Deleted.
        (build_scops): Removed call to limit_scops.
        * sese.c (new_sese): Construct.
        (free_sese): Destruct.
        (sese_add_exit_phis_edge): update_stmt after exit phi edge has been
        added.
        (set_rename): Pass sese region so that parameters inside the region can
        be added to its parameter_rename_map.
        (rename_uses): Pass sese region.
        (graphite_copy_stmts_from_block): Do not copy parameters that have been
        generated in the header of the scop. For each SSA_NAME in the
        parameter_rename_map rename its usage.
        (invariant_in_sese_p_rec): Return false if tree t is defined outside
        sese region.
        (scalar_evolution_in_region): If the tree t is invariant just return t.
        * sese.h: Added a parameter renamne map (parameter_rename_map_t) to
          struct sese to keep track of all the parameters which need renaming.
        * tree-data-ref.c (loop_nest_has_data_refs): Check if a loop nest has
          any data-refs.
        * tree-data-ref.h: Declaration of loop_nest_has_data_refs.

gcc/testsuite/ChangeLog:

2015-09-02  Aditya Kumar  <hiraditya@msn.com>
            Sebastian Pop  <s.pop@samsung.com>

        * gcc.dg/graphite/block-0.c: Modifed test case to match current output.
        * gcc.dg/graphite/block-1.c: Same.
        * gcc.dg/graphite/block-5.c: Same.
        * gcc.dg/graphite/block-6.c: Same.
        * gcc.dg/graphite/interchange-1.c: Same.
        * gcc.dg/graphite/interchange-10.c: Same.
        * gcc.dg/graphite/interchange-11.c: Same.
        * gcc.dg/graphite/interchange-13.c: Same.
        * gcc.dg/graphite/interchange-14.c: Same.
        * gcc.dg/graphite/interchange-3.c: Same.
        * gcc.dg/graphite/interchange-4.c: Same.
        * gcc.dg/graphite/interchange-7.c: Same.
        * gcc.dg/graphite/interchange-8.c: Same.
        * gcc.dg/graphite/interchange-9.c: Same.
        * gcc.dg/graphite/isl-codegen-loop-dumping.c: Same.
        * gcc.dg/graphite/pr35356-1.c (foo): Same.
        * gcc.dg/graphite/pr37485.c: Same.
        * gcc.dg/graphite/scop-0.c (int toto): Same.
        * gcc.dg/graphite/scop-1.c: Same.
        * gcc.dg/graphite/scop-10.c: Same.
        * gcc.dg/graphite/scop-11.c: Same.
        * gcc.dg/graphite/scop-12.c: Same.
        * gcc.dg/graphite/scop-13.c: Same.
        * gcc.dg/graphite/scop-16.c: Same.
        * gcc.dg/graphite/scop-17.c: Same.
        * gcc.dg/graphite/scop-18.c: Same.
        * gcc.dg/graphite/scop-2.c: Same.
        * gcc.dg/graphite/scop-21.c (int test): Same.
        * gcc.dg/graphite/scop-22.c (void foo): Same.
        * gcc.dg/graphite/scop-4.c: Same.
        * gcc.dg/graphite/scop-5.c: Same.
        * gcc.dg/graphite/scop-6.c: Same.
        * gcc.dg/graphite/scop-7.c: Same.
        * gcc.dg/graphite/scop-8.c: Same.
        * gcc.dg/graphite/scop-9.c: Same.
        * gcc.dg/graphite/scop-mvt.c (void mvt): Introduced dependency so that
          data-refs remain inside the inner loop.
        * gcc.dg/graphite/uns-block-1.c: Modifed test case to match o/p.
        * gcc.dg/graphite/uns-interchange-14.c: Same.
        * gcc.dg/graphite/uns-interchange-9.c: Same.
        * gfortran.dg/graphite/interchange-3.f90

libgomp/ChangeLog:

2015-09-04  Aditya Kumar  <hiraditya@msn.com>
            Sebastian Pop  <s.pop@samsung.com>

        * testsuite/libgomp.graphite/bounds.c (int foo): Modifed test case to
          match o/p.
        * testsuite/libgomp.graphite/force-parallel-1.c (void parloop): Same.
        * testsuite/libgomp.graphite/force-parallel-4.c: Same.
        * testsuite/libgomp.graphite/force-parallel-5.c: Same.
        * testsuite/libgomp.graphite/force-parallel-7.c: Same.
        * testsuite/libgomp.graphite/force-parallel-8.c: Same.

Co-Authored-By: Sebastian Pop <s.pop@samsung.com>
From-SVN: r227567

Prevent unnecessary recompilation for trivial params.def changes

2015-09-08  Tom de Vries  <tom@codesourcery.com>

* Makefile.in (generated_files): Add params.list.
(params.list, s-params.list): Add rule.
* params.h (enum compiler_param): Include params-list.h.  Move define
DEFPARAM, include params.def and undef DEFPARAM ...
* params-list.h: ... here.  New file.

From-SVN: r227566

Trivial typo fix in pretty-print.h

gcc/ChangeLog:
* pretty-print.h (printer_fn): Fix typo in comment.

From-SVN: r227562

Re: [PATCH] Minor cleanup of const_and_copies stack

Re: [PATCH] Minor cleanup of const_and_copies stack
        * tree-ssa-scopedtables.h (class const_and_copies): Fix comment typo.

From-SVN: r227559

correct ChangeLog dates+address

From-SVN: r227558

ARM/AArch64 Testsuite] Add float16 lane_f16_indices tests

PR target/63870
* gcc.target/aarch64/advsimd-intrinsics/vld2_lane_f16_indices_1.c: New.
* gcc.target/aarch64/advsimd-intrinsics/vld2q_lane_f16_indices_1.c: New.
* gcc.target/aarch64/advsimd-intrinsics/vld3_lane_f16_indices_1.c: New.
* gcc.target/aarch64/advsimd-intrinsics/vld3q_lane_f16_indices_1.c: New.
* gcc.target/aarch64/advsimd-intrinsics/vld4_lane_f16_indices_1.c: New.
* gcc.target/aarch64/advsimd-intrinsics/vld4q_lane_f16_indices_1.c: New.
* gcc.target/aarch64/advsimd-intrinsics/vst2_lane_f16_indices_1.c: New.
* gcc.target/aarch64/advsimd-intrinsics/vst2q_lane_f16_indices_1.c: New.
* gcc.target/aarch64/advsimd-intrinsics/vst3_lane_f16_indices_1.c: New.
* gcc.target/aarch64/advsimd-intrinsics/vst3q_lane_f16_indices_1.c: New.
* gcc.target/aarch64/advsimd-intrinsics/vst4_lane_f16_indices_1.c: New.
* gcc.target/aarch64/advsimd-intrinsics/vst4q_lane_f16_indices_1.c: New.

From-SVN: r227557

[ARM] Update sourcebuild.texi with testsuite/effective-target hooks

* doc/sourcebuild.texi (arm_neon_fp16): Correct cross-reference.
(arm_neon_fp16_ok): Document adding of -mfp16-format=ieee flag.
(arm_neon_fp16_hw): New.

From-SVN: r227556

[ARM/AArch64 Testsuite]Add test of vcvt{,_high}_i{f32_f16,f16_f32}

* gcc.target/aarch64/advsimd-intrinsics/vcvt_f16.c: New.
* lib/target-supports.exp
(check_effective_target_arm_neon_fp16_hw): New.

From-SVN: r227555

[ARM/AArch64 Testsuite] Add float16 tests to advsimd-intrinsics testsuite

* gcc.target/aarch64/advsimd-intrinsics/advsimd-intrinsics.exp:
Set additional_flags for neon-fp16 if supported, else fallback to neon.

* gcc.target/aarch64/advsimd-intrinsics/arm-neon-ref.h
(hfloat16_t): New.
(result, expected, clean_results, DECL_VARIABLE_64BITS_VARIANTS,
DECL_VARIABLE_128BITS_VARIANTS): Add float16x4_t and float16x8_t cases
if supported.
(CHECK_RESULTS): Redefine using CHECK_RESULTS_NAMED.
(CHECK_RESULTS_NAMED): Move body to CHECK_RESULTS_NAMED_NO_FP16;
redefine in terms of CHECK_RESULTS_NAMED_NO_FP16 with float16 variants
when those are supported.
(CHECK_RESULTS_NAMED_NO_FP16, CHECK_RESULTS_NO_FP16): New.
(vdup_n_f16): New.

* gcc.target/aarch64/advsimd-intrinsics/compute-ref-data.h (buffer,
buffer_pad, buffer_dup, buffer_dup_pad): Add float16x4 and float16x8_t
cases if supported.

* gcc.target/aarch64/advsimd-intrinsics/vbsl.c (exec_vbsl):
Use CHECK_RESULTS_NO_FP16 in place of CHECK_RESULTS.
* gcc.target/aarch64/advsimd-intrinsics/vdup-vmov.c (exec_vdup_vmov):
Likewise.
* gcc.target/aarch64/advsimd-intrinsics/vdup_lane.c (exec_vdup_lane):
Likewise.
* gcc.target/aarch64/advsimd-intrinsics/vext.c (exec_vext): Likewise.

* gcc.target/aarch64/advsimd-intrinsics/vcombine.c (expected):
Add float16x8_t case.
(main, exec_vcombine): test float16x4_t -> float16x8_t, if supported.
* gcc.target/aarch64/advsimd-intrinsics/vcreate.c (expected,
main, exec_vcreate): Likewise.
* gcc.target/aarch64/advsimd-intrinsics/vget_high (expected,
 exec_vget_high): Likewise.
* gcc.target/aarch64/advsimd-intrinsics/vget_low.c (expected,
exec_vget_low): Likewise.
* gcc.target/aarch64/advsimd-intrinsics/vld1.c (expected, exec_vld1):
Likewise.
* gcc.target/aarch64/advsimd-intrinsics/vld1_dup.c (expected,
exec_vld1_dup): Likewise.
* gcc.target/aarch64/advsimd-intrinsics/vld1_lane.c (expected,
exec_vld1_lane): Likewise.
* gcc.target/aarch64/advsimd-intrinsics/vldX.c (expected, exec_vldX):
Likewise.
* gcc.target/aarch64/advsimd-intrinsics/vldX_dup.c (expected,
exec_vldX_dup): Likewise.
* gcc.target/aarch64/advsimd-intrinsics/vldX_lane.c (expected,
exec_vldX_lane): Likewise.
* gcc.target/aarch64/advsimd-intrinsics/vset_lane.c (expected,
exec_vset_lane): Likewise.
* gcc.target/aarch64/advsimd-intrinsics/vst1_lane.c (expected,
 exec_vst1_lane): Likewise.

From-SVN: r227554

re PR c++/67041 ([C++14] Variable template initialized by call to lambda does not compile)

PR c++/67041
* pt.c (tsubst_copy_and_build): Handle variables like functions.

From-SVN: r227553

[fold-const.c] Fix bigendian HFmode in native_interpret_real

* fold-const.c (native_interpret_real): Fix HFmode for bigendian where
UNITS_PER_WORD >= 4.

From-SVN: r227552

[AArch64] Add vcvt(_high)?_f32_f16 intrinsics, with BE RTL fix

gcc/:

* config/aarch64/aarch64-simd.md (aarch64_simd_vec_unpacks_lo_<mode>,
aarch64_simd_vec_unpacks_hi_<mode>): New insn.
(vec_unpacks_lo_v4sf, vec_unpacks_hi_v4sf): Delete insn.
(vec_unpacks_lo_<mode>, vec_unpacks_hi_<mode>): New expand.
(aarch64_float_extend_lo_v2df): Rename to...
(aarch64_float_extend_lo_<Vwide>): this, using VDF and so adding V4SF.

* config/aarch64/aarch64-simd-builtins.def (vec_unpacks_hi): Add v8hf.
(float_extend_lo): Add v4sf.

* config/aarch64/arm_neon.h (vcvt_f32_f16, vcvt_high_f32_f16): New.
* config/aarch64/iterators.md (VQ_HSF): New iterator.
(VWIDE, Vwtype, Vhalftype): Add V8HF, V4SF.
(Vwide): New mode_attr.

From-SVN: r227551

[AArch64] Improve code generation for float16 vector code

gcc/:

* config/aarch64/aarch64-simd.md (aarch64_simd_dup<mode>,
aarch64_dup_lane<mode>, aarch64_dup_lane_<vswap_width_name><mode>,
aarch64_simd_vec_set<mode>, vec_set<mode>, vec_perm_const<mode>,
vec_init<mode>, *aarch64_simd_ld1r<mode>, vec_extract<mode>): Add
V4HF and V8HF variants to iterator.

* config/aarch64/aarch64.c (aarch64_evpc_dup): Add V4HF and V8HF cases.

* config/aarch64/iterators.md (VDQF_F16): New.
(VSWAP_WIDTH, vswap_width_name): Add V4HF and V8HF cases.

From-SVN: r227550

[AArch64] vreinterpret(q?), vget_(low|high), vld1(q?)_dup

gcc/:

* config/aarch64/arm_neon.h (vreinterpret_p8_f16, vreinterpret_p16_f16,
vreinterpret_f16_f64, vreinterpret_f16_s8, vreinterpret_f16_s16,
vreinterpret_f16_s32, vreinterpret_f16_s64, vreinterpret_f16_f32,
vreinterpret_f16_u8, vreinterpret_f16_u16, vreinterpret_f16_u32,
vreinterpret_f16_u64, vreinterpret_f16_p8, vreinterpret_f16_p16,
vreinterpretq_f16_f64, vreinterpretq_f16_s8, vreinterpretq_f16_s16,
vreinterpretq_f16_s32, vreinterpretq_f16_s64, vreinterpretq_f16_f32,
vreinterpretq_f16_u8, vreinterpretq_f16_u16, vreinterpretq_f16_u32,
vreinterpretq_f16_u64, vreinterpretq_f16_p8, vreinterpretq_f16_p16,
vreinterpret_f32_f16, vreinterpret_f64_f16, vreinterpret_s64_f16,
vreinterpret_u64_f16, vreinterpretq_u64_f16, vreinterpret_s8_f16,
vreinterpret_s16_f16, vreinterpret_s32_f16, vreinterpret_u8_f16,
vreinterpret_u16_f16, vreinterpret_u32_f16, vreinterpretq_p8_f16,
vreinterpretq_p16_f16, vreinterpretq_f32_f16, vreinterpretq_f64_f16,
vreinterpretq_s64_f16, vreinterpretq_s8_f16, vreinterpretq_s16_f16,
vreinterpretq_s32_f16, vreinterpretq_u8_f16, vreinterpretq_u16_f16,
vreinterpretq_u32_f16, vget_low_f16, vget_high_f16, vld1_dup_f16,
vld1q_dup_f16): New.

gcc/testsuite/:

* gcc.target/aarch64/vget_high_1.c: Add float16x8->float16x4 case.
* gcc.target/aarch64/vget_low_1.c: Likewise.

From-SVN: r227546

[AArch64] Implement vcvt_{,high_}f16_f32

* config/aarch64/aarch64-simd.md (aarch64_float_truncate_lo_v2sf):
Reparameterize to...
(aarch64_float_truncate_lo_<mode>): ...this, for both V2SF and V4HF.
(aarch64_float_truncate_hi_v4sf): Reparameterize to...
(aarch64_float_truncate_hi_<Vdbl>): ...this, for both V4SF and V8HF.

* config/aarch64/aarch64-simd-builtins.def (float_truncate_hi_): Add
v8hf variant.
(float_truncate_lo_): Use BUILTIN_VDF iterator.

* config/aarch64/arm_neon.h (vcvt_f16_f32, vcvt_high_f16_f32): New.

* config/aarch64/iterators.md (VDF, Vdtype): New.
(VWIDE, Vmwtype): Add cases for V4HF and V2SF.

From-SVN: r227545

[AArch64] vld{2,3,4}{,_lane,_dup}, vcombine, vcreate

gcc/:

* config/aarch64/aarch64.c (aarch64_split_simd_combine): Add V4HFmode.
* config/aarch64/aarch64-builtins.c (VAR13, VAR14): New.
(aarch64_scalar_builtin_types, aarch64_init_simd_builtin_scalar_types):
Add __builtin_aarch64_simd_hf.
* config/aarch64/arm_neon.h (float16x4x2_t, float16x8x2_t,
float16x4x3_t, float16x8x3_t, float16x4x4_t, float16x8x4_t,
vcombine_f16, vst2_lane_f16, vst2q_lane_f16, vst3_lane_f16,
vst3q_lane_f16, vst4_lane_f16, vst4q_lane_f16, vld2_f16, vld2q_f16,
vld3_f16, vld3q_f16, vld4_f16, vld4q_f16, vld2_dup_f16, vld2q_dup_f16,
vld3_dup_f16, vld3q_dup_f16, vld4_dup_f16, vld4q_dup_f16,
vld2_lane_f16, vld2q_lane_f16, vld3_lane_f16, vld3q_lane_f16,
vld4_lane_f16, vld4q_lane_f16, vst2_f16, vst2q_f16, vst3_f16,
vst3q_f16, vst4_f16, vst4q_f16, vcreate_f16): New.

* config/aarch64/iterators.md (VALLDIF, Vtype, Vetype, Vbtype,
V_cmp_result, v_cmp_result): Add cases for V4HF and V8HF.
(VDC, Vdbl): Add V4HF.

gcc/testsuite/:

* gcc.target/aarch64/vldN_1.c: Add float16x4_t and float16x8_t cases.
* gcc.target/aarch64/vldN_dup_1.c: Likewise.
* gcc.target/aarch64/vldN_lane_1.c: Likewise.
(main): update orig_data to avoid float16 NaN on bigendian.

From-SVN: r227543

[AArch64] Add support for float16x{4,8}_t vectors/builtins

gcc/:

* config/aarch64/aarch64.c (aarch64_vector_mode_supported_p): Support
V4HFmode and V8HFmode.
(aarch64_split_simd_move): Add case for V8HFmode.
* config/aarch64/aarch64-builtins.c (v4hf_UP, v8hf_UP): Define.
(aarch64_simd_builtin_std_type): Handle HFmode.
(aarch64_init_simd_builtin_types): Include Float16x4_t and Float16x8_t.

* config/aarch64/aarch64-simd.md (mov<mode>, aarch64_get_lane<mode>,
aarch64_ld1<VALL:mode>, aarch64_st1<VALL:mode): Use VALL_F16 iterator.
(aarch64_be_ld1<mode>, aarch64_be_st1<mode>): Use VALLDI_F16 iterator.

* config/aarch64/aarch64-simd-builtin-types.def: Add Float16x4_t,
Float16x8_t.

* config/aarch64/aarch64-simd-builtins.def (ld1, st1): Use VALL_F16.
* config/aarch64/arm_neon.h (float16x4_t, float16x8_t, float16_t):
New typedefs.
(vget_lane_f16, vgetq_lane_f16, vset_lane_f16, vsetq_lane_f16,
vld1_f16, vld1q_f16, vst1_f16, vst1q_f16, vst1_lane_f16,
vst1q_lane_f16): New.
* config/aarch64/iterators.md (VD, VQ, VQ_NO2E): Add vectors of HFmode.
(VALLDI_F16, VALL_F16): New.
(Vmtype, VEL, VCONQ, VHALF, V_TWO_ELEM, V_THREE_ELEM, V_FOUR_ELEM, q):
Add cases for V4HF and V8HF.
(VDBL, VRL2, VRL3, VRL4): Add V4HF case.

gcc/testsuite/:

* g++.dg/abi/mangle-neon-aarch64.C: Add cases for float16x4_t and
float16x8_t.
* gcc.target/aarch64/vset_lane_1.c: Likewise.
* gcc.target/aarch64/vld1-vst1_1.c: Likewise.
* gcc.target/aarch64/vld1_lane.c: Likewise.

From-SVN: r227542

[ARM] Remaining intrinsics

* config/arm/arm-builtins.c (VAR11, VAR12): New.
* config/arm/arm_neon_builtins.def (vcombine, vld2_dup, vld3_dup,
vld4_dup): Add v4hf variant.
(vget_high, vget_low): Add v8hf variant.
(vld1, vst1, vst1_lane, vld2, vld2_lane, vst2, vst2_lane, vld3,
vld3_lane, vst3, vst3_lane, vld4, vld4_lane, vst4, vst4_lane): Add
v4hf and v8hf variants.

* config/arm/iterators.md (VD_LANE, VD_RE, VQ2, VQ_HS): New.
(VDX): Add V4HF.
(V_DOUBLE): Add case for V4HF.
(VQX): Add V8HF.
(V_HALF): Add case for V8HF.
(VDQX): Add V4HF, V8HF.
(V_elem, V_two_elem, V_three_elem, V_four_elem, V_cmp_result,
V_uf_sclr, V_sz_elem, V_mode_nunits, q): Add cases for V4HF & V8HF.

* config/arm/neon.md (vec_set<mode>internal, vec_extract<mode>,
neon_vget_lane<mode>_sext_internal, neon_vget_lane<mode>_zext_internal,
vec_load_lanesoi<mode>, neon_vld2<mode>, vec_store_lanesoi<mode>,
neon_vst2<mode>, vec_load_lanesci<mode>, neon_vld3<mode>,
neon_vld3qa<mode>, neon_vld3qb<mode>, vec_store_lanesci<mode>,
neon_vst3<mode>, neon_vst3qa<mode>, neon_vst3qb<mode>,
vec_load_lanesxi<mode>, neon_vld4<mode>, neon_vld4qa<mode>,
neon_vld4qb<mode>, vec_store_lanesxi<mode>, neon_vst4<mode>,
neon_vst4qa<mode>, neon_vst4qb<mode>): Change VQ iterator to VQ2.

(neon_vcreate, neon_vreinterpretv8qi<mode>,
neon_vreinterpretv4hi<mode>, neon_vreinterpretv2si<mode>,
neon_vreinterpretv2sf<mode>, neon_vreinterpretdi<mode>):
Change VDX to VD_RE.

(neon_vld2_lane<mode>, neon_vst2_lane<mode>, neon_vld3_lane<mode>,
neon_vst3_lane<mode>, neon_vld4_lane<mode>, neon_vst4_lane<mode>):
Change VD iterator to VD_LANE, and VMQ iterator to VQ_HS.

* config/arm/arm_neon.h (float16x4x2_t, float16x8x2_t, float16x4x3_t,
float16x8x3_t, float16x4x4_t, float16x8x4_t, vcombine_f16,
vget_high_f16, vget_low_f16, vld1_f16, vld1q_f16, vst1_f16, vst1q_f16,
vst1_lane_f16, vst1q_lane_f16, vld2_f16, vld2q_f16, vld2_lane_f16,
vld2q_lane_f16, vld2_dup_f16, vst2_f16, vst2q_f16, vst2_lane_f16,
vst2q_lane_f16, vld3_f16, vld3q_f16, vld3_lane_f16, vld3q_lane_f16,
vld3_dup_f16, vst3_f16, vst3q_f16, vst3_lane_f16, vst3q_lane_f16,
vld4_f16, vld4q_f16, vld4_lane_f16, vld4q_lane_f16, vld4_dup_f16,
vst4_f16, vst4q_f16, vst4_lane_f16, vst4q_lane_f16): New.

From-SVN: r227541

[ARM] float16x8_t intrinsics in arm_neon.h

* config/arm/arm_neon.h (vgetq_lane_f16, vsetq_lane_f16, vld1q_lane_f16,
vld1q_dup_f16, vreinterpretq_p8_f16, vreinterpretq_p16_f16,
vreinterpretq_f16_p8, vreinterpretq_f16_p16, vreinterpretq_f16_f32,
vreinterpretq_f16_p64, vreinterpretq_f16_p128, vreinterpretq_f16_s64,
vreinterpretq_f16_u64, vreinterpretq_f16_s8, vreinterpretq_f16_s16,
vreinterpretq_f16_s32, vreinterpretq_f16_u8, vreinterpretq_f16_u16,
vreinterpretq_f16_u32, vreinterpretq_f32_f16, vreinterpretq_p64_f16,
vreinterpretq_p128_f16, vreinterpretq_s64_f16, vreinterpretq_u64_f16,
vreinterpretq_s8_f16, vreinterpretq_s16_f16, vreinterpretq_s32_f16,
vreinterpretq_u8_f16, vreinterpretq_u16_f16, vreinterpretq_u32_f16):
New.

From-SVN: r227538

fix ChangeLog whitespace

From-SVN: r227537