[PR64164] Drop copyrename, use coalescible partition as base when optimizing.

for  gcc/ChangeLog

PR rtl-optimization/64164
PR bootstrap/66978
PR middle-end/66983
PR rtl-optimization/67000
PR middle-end/67034
PR middle-end/67035
* Makefile.in (OBJS): Drop tree-ssa-copyrename.o.
* tree-ssa-copyrename.c: Removed.
* opts.c (default_options_table): Drop -ftree-copyrename.  Add
-ftree-coalesce-vars.
* passes.def: Drop all occurrences of pass_rename_ssa_copies.
* common.opt (ftree-copyrename): Ignore.
(ftree-coalesce-inlined-vars): Likewise.
* doc/invoke.texi: Remove the ignored options above.
* gimple-expr.h (gimple_can_coalesce_p): Move declaration
* tree-ssa-coalesce.h: ... here.
* tree-ssa-uncprop.c: Include tree-ssa-coalesce.h and other
headers required by it.
* gimple-expr.c (gimple_can_coalesce_p): Allow coalescing
across variables when flag_tree_coalesce_vars.  Check register
use and promoted modes to allow coalescing.  Do not coalesce
maybe-byref parms with SSA_NAMEs of other variables, or
anonymous SSA_NAMEs.  Moved to tree-ssa-coalesce.c.
* tree-ssa-live.c (struct tree_int_map_hasher): Move along
with its member functions to tree-ssa-coalesce.c.
(var_map_base_init): Likewise.  Renamed to
compute_samebase_partition_bases.
(partition_view_normal): Drop want_bases parameter.
(partition_view_bitmap): Likewise.
* tree-ssa-live.h: Adjust declarations.
* tree-ssa-coalesce.c: Include explow.h and cfgexpand.h.
(build_ssa_conflict_graph): Process PARM_ and RESULT_DECLs's
default defs at the entry point.
(dump_part_var_map): New.
(compute_optimized_partition_bases): New, called by...
(coalesce_ssa_name): ... when flag_tree_coalesce_vars, instead
of compute_samebase_partition_bases.  Adjust.
* alias.c (nonoverlapping_memrefs_p): Disregard gimple-regs.
* cfgexpand.c (leader_merge, parm_maybe_byref_p): New.
(ssa_default_def_partition): New.
(get_rtl_for_parm_ssa_default_def): New.
(align_local_variable, add_stack_var): Support anonymous SSA
names.
(defer_stack_allocation): Likewise.  Declare earlier.
(set_rtl): Merge exprs and attrs, even for MEMs and non-SSA
vars.  Update DECL_RTL for PARM_DECLs and RESULT_DECLs too.
Do no record deferred-allocation marker in
SA.partition_to_pseudo.
(expand_stack_vars): Adjust check for the marker in it.
(expand_one_stack_var_at): Handle anonymous SSA_NAMEs.  Drop
redundant MEM attr setting.
(expand_one_stack_var_1): Handle anonymous SSA_NAMEs.  Renamed
from...
(expand_one_stack_var): ... this.  New wrapper to check and
skip already expanded SSA partitions.
(record_alignment_for_reg_var): New, factored out of...
(expand_one_var): ... this.
(expand_one_ssa_partition): New.
(adjust_one_expanded_partition_var): New.
(expand_one_register_var): Check and skip already expanded SSA
partitions.
(expand_used_vars): Don't create DECLs for anonymous SSA
names.  Expand all SSA partitions, then adjust all SSA names.
(pass::execute): Replace the loops that set
SA.partition_to_pseudo from partition leaders and cleared
DECL_RTL for multi-location variables, and that which used to
rename vars and set attrs, with one that clears DECL_RTL and
checks that PARMs and RESULTs default_defs match DECL_RTL.
* cfgexpand.h (get_rtl_for_parm_ssa_default_def): Declare.
* emit-rtl.c: Include stor-layout.h.
(set_reg_attrs_for_parm): Handle NULL decl.
(set_reg_attrs_for_decl_rtl): Take mode from expression if
it's not a DECL.
* stmt.c (emit_case_decision_tree): Pass it the SSA_NAME
rather than its possibly-NULL DECL.
* explow.c (promote_ssa_mode): New.
* explow.h (promote_ssa_mode): Declare.
* expr.c (expand_expr_real_1): Handle anonymous SSA_NAMEs.
(read_complex_part): Export.
* expr.h (read_complex_part): Declare.
* cfgexpand.h (parm_maybe_byref_p): Declare.
* function.c: Include cfgexpand.h.
(use_register_for_decl): Handle SSA_NAMEs, anonymous or not.
(use_register_for_parm_decl): Wrapper for the above to
special-case the result_ptr.
(rtl_for_parm): Ditto for get_rtl_for_parm_ssa_default_def.
(split_complex_args): Take assign_parm_data_all argument.
Pass it to rtl_for_parm.  Set up rtl and context for split
args.  Reset complex parm before fetching its default decl
rtl.
(assign_parms_unsplit_complex): Use the default-def complex
parm rtl if it matches the components.
(assign_parms_augmented_arg_list): Adjust.
(maybe_reset_rtl_for_parm): Reset DECL_RTL of parms with
multiple locations.  Recognize split complex args.
(assign_parm_adjust_stack_rtl): Add all and parm arguments,
for rtl_for_parm.  For SSA-assigned parms, zero stack_parm.
(assign_parm_setup_block): Prefer SSA-assigned location, and
fill in its address if the memory location of a maybe-byref
parm was not assigned by cfgexpand.
(assign_parm_setup_reg): Likewise.  Adjust its mode as
needed.  Use entry_parm for equiv if stack_parm is NULL.  Make
sure passed_pointer parms don't need conversion.  Copy address
or value as needed.
(assign_parm_setup_stack): Prefer SSA-assigned location.
(assign_parms): Maybe reset DECL_RTL of params.  Adjust stack
rtl before testing for pointer bounds.  Special-case result_ptr.
(expand_function_start): Maybe reset DECL_RTL of result.
Prefer SSA-assigned location for result and static chain.
Factor out DECL_RESULT and SET_DECL_RTL.  Convert static chain
to Pmode if needed, from H.J. Lu  <hongjiu.lu@intel.com>.
* tree-outof-ssa.c (insert_value_copy_on_edge): Handle
anonymous SSA names.  Use promote_ssa_mode.
(get_temp_reg): Likewise.
(remove_ssa_form): Adjust.
* stor-layout.c (layout_decl): Don't set mem attributes of
non-MEMs.
* var-tracking.c (dataflow_set_clear_at_call): Take call_insn
and get its reg_usage for reg invalidation.
(compute_bb_dataflow): Pass it insn.
(emit_notes_in_bb): Likewise.

for  gcc/testsuite/ChangeLog

* gcc.dg/guality/pr54200.c: Add -fno-tree-coalesce-vars.
* gcc.dg/ssp-1.c: Make counter a register.
* gcc.dg/ssp-2.c: Likewise.
* gcc.dg/torture/parm-coalesce.c: New.

From-SVN: r226901

* tree-core.h (tree_base): Fix typo.

From-SVN: r226900

compiler: Report unnamed receiver types.

    gccgo used to crash when presented with an invalid receiver type in a
    method.  Instead, unnamed receiver types should report an error.

    Fixes golang/go#11557.

    Reviewed-on: https://go-review.googlesource.com/13245

From-SVN: r226899

re PR middle-end/67133 (ICE at -Os and above on x86_64-linux-gnu in gimple_op, at gimple.h:2274)

PR middle-end/67133
* gimple.c (infer_nonnull_range_by_attribute): Check that the
nonnull argument position is not outside function arguments.

* gcc.dg/torture/pr67133.c: New test.

From-SVN: r226896

re PR target/67143 (ICE (could not split insn) on aarch64-linux-gnu)

gcc/
2015-08-14  Matthew Wahab  <matthew.wahab@arm.com>

PR target/67143
* config/aarch64/atomics.md (atomic_<optab><mode>): Replace
'lconst_atomic' with 'const_atomic'.
(atomic_fetch_<optab><mode>): Likewise.
(atomic_<optab>_fetch<mode>): Likewise.
* config/aarch64/iterators.md (lconst-atomic): Move below
'const_atomic'.
(const_atomic): New.

gcc/testsuite/
2015-08-14  Matthew Wahab  <matthew.wahab@arm.com>
    Matthias Klose  <doko@debian.org>

PR target/67143
* gcc.c-torture/compile/pr67143.c: New
* gcc.target/aarch64/atomic-op-imm.c
(atomic_fetch_add_negative_RELAXED): New.
(atomic_fetch_sub_negative_ACQUIRE): New.

Co-Authored-By: Matthias Klose <doko@debian.org>
From-SVN: r226895

any (any::operator=(const any&)): Move check for self-assignment.

* include/experimental/any (any::operator=(const any&)): Move check
for self-assignment.
(any::operator=(any&&)): Add check for self-assignment.
(any::operator=(_ValueType&&)): Constrain template argument.
(any::swap(any&)): Add check for self-swap.
* testsuite/experimental/any/assign/self.cc: Test move and swap.
* testsuite/experimental/any/misc/any_cast_neg.cc: Update dg-error.

From-SVN: r226894

array: Add feature-test macro.

* include/experimental/array: Add feature-test macro.
* testsuite/experimental/array/neg.cc: Update dg-error.

From-SVN: r226893

inclhack.def (aix_stdio_inline): New fix.

        * inclhack.def (aix_stdio_inline): New fix.
        (aix_strtof_const): Limit to *-*-aix*.
        (aix_sysmachine): Same.
        (aix_syswait_2): Same.
        (aix_volatile): Same.
        * fixincl.x: Regenerated.
        * test/base/stdio.h [AIX_STDIO_INLINE]: New test.

From-SVN: r226891

Add DWARF2_LINENO_DEBUGGING_INFO, and use it for nvptx

gcc/
* config/nvptx/nvptx.c (nvptx_option_override): Don't override
debug options.
* config/nvptx/nvptx.h (DWARF2_LINENO_DEBUGGING_INFO): Define.
(DWARF2_DEBUGGING_INFO): Don't define.
* debug.h (dwarf2_lineno_debug_hooks): Declare.
* toplev.c (process_options): Add a case for it.
* dwarf2out.c (dwarf2_lineno_debug_hooks): New variable.
(dwarf2out_init): Skip most initializations if
DWARF2_LINENO_DEBUGGING_INFO, but set cur_line_info_table in that
case.
* defaults.h (PREFERRED_DEBUGGING_TYPE): Also use DWARF2_DEBUG if
DWARF2_LINENO_DEBUGGING_INFO.
* opts.c (set_debug_level): Likewise.

Co-Authored-By: Bernd Schmidt <bernds@codesourcery.com>
From-SVN: r226890

[AArch64][obvious] Fix some target attribute inlining tests for -fPIC

* gcc.target/aarch64/target_attr_5.c: Add static modifier to bar.
* gcc.target/aarch64/target_attr_8.c: Likewise.
* gcc.target/aarch64/target_attr_14.c: Likewise.

From-SVN: r226889

DWARF2: add extensions to handle biased types

include/ChangeLog:

* dwarf2.def (DW_AT_GNU_bias): New attribute.

From-SVN: r226888

DWARF2: add extensions to handle fixed point types

include/ChangeLog:

* dwarf2.def (DW_AT_GNU_numerator, DW_AT_GNU_denominator): New
attributes.

From-SVN: r226887

[AArch64][obvious] Mark target_attr_1.c as compile-only

* gcc.target/aarch64/target_attr_1.c: Make test compile-only.

From-SVN: r226886

[Patch ARM/AArch64] Add missing is_neon_type types

gcc/

* config/arm/types.md (is_neon_type): Add missing types.

From-SVN: r226885

driver-i386.c (host_detect_local_cpu): Add support for skylake.

gcc/

* config/i386/driver-i386.c (host_detect_local_cpu): Add support
for skylake.
* config/i386/i386.c (PTA_SKYLAKE): New macros.
(processor_alias_table): Add skylake description.
(enum processor_model): Add skylake processor.
(arch_names_table): Add skylake record.
* doc/invoke.texi: Add skylake item.

gcc/testsuite/

* gcc.target/i386/builtin_target.c: Add skylake check.

libgcc/

* config/i386/cpuinfo.c (enum processor_subtypes): Add skylake.
(get_intel_cpu): Likewise.

From-SVN: r226884

jit: on OS X, add -Wl,-undefined,dynamic_lookup to driver invocation

gcc/jit/ChangeLog:
* jit-playback.c (invoke_driver): On OS X, add
"-Wl,-undefined,dynamic_lookup" to the driver arguments.

From-SVN: r226882

Daily bump.

From-SVN: r226881

dwarf.c (read_function_entry): Add vec_inlined parameter.

* dwarf.c (read_function_entry): Add vec_inlined parameter.
Change all callers.

From-SVN: r226878

pr66314.c: Moved from here to ..

* gcc.dg/pr66314.c: Moved from here to ..
* gcc.dg/asan/pr66314.c: Here.  Use -fno-sanitize=all
to clear all sanitizers before -fsanitize=kernel-address.

From-SVN: r226877

ira-int.h: Include recog.h.

* ira-int.h: Include recog.h.
* ira-build.c: Don't include recog.h.
* ira-color.c: Likewise.
* ira-conflicts.c: Likewise.
* ira-costs.c: Likewise.
* ira-emit.c: Likewise.
* ira-lives.c: Likewise.
* ira.c: Likewise.
* sched-deps.c: Likewise.
* sel-sched.c: Likewise.
* target-globals.c: Likewise.

From-SVN: r226876

re PR middle-end/55035 (reload1.c:3766:41: error: ‘orig_dup[0]’ may be used uninitialized in this function (for fr30, microblaze, moxie, rl78))

gcc/
PR bootstrap/55035
* reload1.c (elimination_costs_in_insn): Make it obvious to the
compiler that the n_dups and n_operands loop bounds are invariant.

From-SVN: r226874

[RTL-ifcvt] Allow PLUS+immediate expression in noce_try_store_flag_constants

* ifcvt.c (noce_try_store_flag_constants): Handle PLUS-immediate
expressions in A and B.

* gcc.target/aarch64/cinc_common_1.c: New test.

From-SVN: r226869

re PR c++/53421 (__attribute__((__may_alias__)) prevents taking address of conversion operator member function)

2015-08-13  Paolo Carlini  <paolo.carlini@oracle.com>

PR c++/53421
* g++.dg/ext/attribute-may-alias-1.C: New.
* g++.dg/ext/attribute-may-alias-2.C: Likewise.

From-SVN: r226868

neg.s: Remove.

2015-08-13  Ville Voutilainen  <ville.voutilainen@gmail.com>

        * testsuite/experimental/array/neg.s: Remove.
        * testsuite/experimental/array/neg.ii.gz: Remove.

From-SVN: r226866

Implement std::experimental::fundamentals_v2::make_array and std::experimental::fundamentals_v2::to_array.

2015-08-13  Ville Voutilainen  <ville.voutilainen@gmail.com>

Implement std::experimental::fundamentals_v2::make_array and
std::experimental::fundamentals_v2::to_array.
* include/Makefile.am: Add array.
* include/Makefile.in: Add array.
* include/experimental/array: New.
* testsuite/experimental/array/make_array.cc: Likewise.
* testsuite/experimental/array/neg.cc: Likewise.
* doc/html/manual/status.html: Mark make_array as Y.
* doc/xml/manual/status_cxx2017.xml: Likewise.

From-SVN: r226865

alloc_traits.h: Add feature-test macro.

* include/bits/alloc_traits.h: Add feature-test macro.
* include/bits/allocator.h: Likewise.
* include/bits/range_access.h: Likewise.
* include/bits/stl_map.h: Likewise.
* include/bits/unordered_map.h: Likewise.

From-SVN: r226864

2.cc: New.

* testsuite/30_threads/mutex/unlock/2.cc: New.
* testsuite/30_threads/recursive_mutex/unlock/2.cc: New.
* testsuite/30_threads/recursive_timed_mutex/unlock/2.cc: New.
* testsuite/30_threads/shared_mutex/unlock/1.cc: New.
* testsuite/30_threads/shared_timed_mutex/unlock/1.cc: New.
* testsuite/30_threads/timed_mutex/unlock/2.cc: New.

From-SVN: r226863

shared_mutex (shared_mutex::unlock()): Fix typo.

2015-08-13  Eelis van der Weegen  <eelis@eelis.net>

* include/std/shared_mutex (shared_mutex::unlock()): Fix typo.

From-SVN: r226862

tree.c (nonnull_arg_p): Move from ...

2015-08-13  Richard Biener  <rguenther@suse.de>

* tree.c (nonnull_arg_p): Move from ...
* tree-vrp.c (nonnull_arg_p): ... here.
* tree.h (nonnull_arg_p): Declare.
* tree-ssa-sccvn.c (init_scc_vn): Perform all lattice init
here, register ptr != 0 for nonnull_arg_p pointer arguments.
Properly initialize static chain and by-reference result pointer.
(run_scc_vn): Adjust.

From-SVN: r226861

[MIPS] Enable load/store bonding for I6400

gcc/
* config/mips/mips.h (ENABLE_LD_ST_PAIRS): Enable load/store pairs for
I6400.

From-SVN: r226860

atomic-inst-cas.c: New.

* gcc.target/aarch64/atomic-inst-cas.c: New.
* gcc.target/aarch64/atomic-inst-ops.inc: New.

From-SVN: r226859

* config/aarch64/aarch64-protos.h
(aarch64_gen_atomic_cas): Declare.
* config/aarch64/aarch64.c (aarch64_expand_compare_and_swap):
Choose appropriate instruction pattern for the target.
(aarch64_gen_atomic_cas): New.
* config/aarch64/atomics.md (UNSPECV_ATOMIC_CAS): New.
(atomic_compare_and_swap<mode>_1): Rename to
aarch64_compare_and_swap<mode>.  Fix some indentation.
(aarch64_compare_and_swap<mode>_lse): New.
(aarch64_atomic_cas<mode>): New.

From-SVN: r226858

aarch64.h (AARCH64_ISA_LSE): New.

* config/aarch64/aarch64.h (AARCH64_ISA_LSE): New.
(TARGET_LSE): New.

From-SVN: r226857

* gcc.dg/tree-ssa/vrp33.c: Add -fno-tree-fre.

From-SVN: r226856

re PR c++/62164 (5.0: ICE: error: Both section and comdat group is set)

2015-08-13  Paolo Carlini  <paolo.carlini@oracle.com>

PR c++/62164
* g++.dg/torture/pr62164.C: New.

From-SVN: r226855

re PR tree-optimization/67191 (ICE: in before_dom_children, at tree-ssa-sccvn.c:4372)

2015-08-13  Richard Biener  <rguenther@suse.de>

PR tree-optimization/67191
* tree-ssa-sccvn.c (sccvn_dom_walker::before_dom_children): Remove
assert we value-numbered last stmts operand because it can validly
trigger for unreachable code.

* gcc.dg/torture/pr67191.c: New testcase.
* g++.dg/torture/pr67191.C: Likewise.

From-SVN: r226854

[RTL-ifcvt] Improve conditional select ops on immediates (fix failing x86_64 cmov tests)

PR rtl-optimization/67103
* ifcvt.c (noce_try_store_flag_constants): Move
x = (-(test != 0) & (b - a)) + a transformation to...
(noce_try_cmove): ... Here.  Try it if normal conditional
move fails.

From-SVN: r226853

tree-ssa-sccvn.c (vn_nary_op_compute_hash): Also canonicalize comparison operand order and commutative ternary op operand order.

2015-08-12  Richard Biener  <rguenther@suse.de>

* tree-ssa-sccvn.c (vn_nary_op_compute_hash): Also canonicalize
comparison operand order and commutative ternary op operand order.
(sccvn_dom_walker::cond_stack): New state to track temporary
expressions.
(sccvn_dom_walker::after_dom_children): Remove tempoary expressions
no longer valid.
(sccvn_dom_walker::record_cond): Add a single temporary conditional
expression.
(sccvn_dom_walker::record_conds): Add a temporary conditional
expressions and all related expressions also true/false.
(sccvn_dom_walker::before_dom_children): Record temporary
expressions based on the controlling condition of a single
predecessor.  When trying to simplify a conditional statement
lookup expressions we might have inserted earlier.

* testsuite/gcc.dg/tree-ssa/ssa-fre-47.c: New testcase.
* testsuite/gcc.dg/tree-ssa/ssa-fre-48.c: Likewise.
* testsuite/gcc.dg/tree-ssa/ssa-fre-49.c: Likewise.
* testsuite/g++.dg/tree-ssa/pr61034.C: Adjust.
* testsuite/gcc.dg/fold-compare-2.c: Likewise.
* testsuite/gcc.dg/pr50763.c: Likewise.
* testsuite/gcc.dg/predict-3.c: Likewise.
* testsuite/gcc.dg/tree-ssa/20030709-2.c: Likewise.
* testsuite/gcc.dg/tree-ssa/pr19831-3.c: Likewise.
* testsuite/gcc.dg/tree-ssa/pr20657.c: Likewise.
* testsuite/gcc.dg/tree-ssa/pr21001.c: Likewise.
* testsuite/gcc.dg/tree-ssa/pr37508.c: Likewise.
* testsuite/gcc.dg/tree-ssa/vrp04.c: Likewise.
* testsuite/gcc.dg/tree-ssa/vrp07.c: Likewise.
* testsuite/gcc.dg/tree-ssa/vrp09.c: Likewise.
* testsuite/gcc.dg/tree-ssa/vrp16.c: Likewise.
* testsuite/gcc.dg/tree-ssa/vrp20.c: Likewise.
* testsuite/gcc.dg/tree-ssa/vrp25.c: Likewise.
* testsuite/gcc.dg/tree-ssa/vrp87.c: Likewise.

From-SVN: r226852

[MIPS] Remove W32 and W64 pseudo-processors.

gcc/
* config/mips/mips.c (mips_rtx_cost_data): Remove costs for W32 and W64
pseudo-processors.
* config/mips/mips.md (processor): Remove w32 and w64.

From-SVN: r226851

re PR tree-optimization/66502 (SCCVN can't handle PHIs optimistically optimally)

2015-08-13  Richard Biener  <rguenther@suse.de>

PR tree-optimization/66502
PR tree-optimization/67167
* tree-ssa-sccvn.c (vn_phi_compute_hash): Do not include
backedge arguments.
(vn_phi_lookup): Adjust.
(vn_phi_insert): Likewise.
(visit_phi): Prefer to value-number to another PHI node
over value-numbering to a PHI argument.
(init_scc_vn): Mark DFS back edges.

* gcc.dg/tree-ssa/ssa-fre-46.c: New testcase.

From-SVN: r226850

gimple.h (gcall::code_): New constant static member.

2015-08-13  Richard Biener  <rguenther@suse.de>

* gimple.h (gcall::code_): New constant static member.
(gcond::code_): Likewise.
* gimple.c (gcall::code_): Define.
(gcond::code_): Likewise.
(is_a_helper <const gcond *>): Add.
(gimple_call_lhs): Use GIMPLE_CHECK2 in the gimple overload
and forward to a new gcall overload with less checking and a
cheaper way to access the operand.
(gimple_call_lhs_ptr): Likewise.
(gimple_call_set_lhs): Likewise.
(gimple_call_internal_p): Likewise.
(gimple_call_with_bounds_p): Likewise.
(gimple_call_set_with_bounds): Likewise.
(gimple_call_internal_fn): Likewise.
(gimple_call_set_ctrl_altering): Likewise.
(gimple_call_ctrl_altering_p): Likewise.
(gimple_call_fntype): Likewise.
(gimple_call_fn): Likewise.
(gimple_call_fn_ptr): Likewise.
(gimple_call_set_fndecl): Likewise.
(gimple_call_fndecl): Likewise.
(gimple_call_chain): Likewise.
(gimple_call_num_args): Likewise.
(gimple_call_arg): Likewise.
(gimple_call_arg_ptr): Likewise.
(gimple_call_set_arg): Likewise.
(gimple_call_noreturn_p): Likewise.
(gimple_cond_code): Likewise.
(gimple_cond_lhs): Likewise.
(gimple_cond_rhs): Likewise.
(gimple_has_lhs): Reduce checking.

From-SVN: r226849

re PR middle-end/25529 ((unsigned * 2)/2 is not changed into unsigned &0x7FFFFFFF)

2015-08-13  Naveen H.S  <Naveen.Hurugalawadi@caviumnetworks.com>

PR middle-end/25529
* match.pd (div (mult @0 @1) @1) : New simplifier.

From-SVN: r226847

compiler: Flatten erroneous subtrees into errors.

    Between the lowering and flattening passes of the compiler, there are
    several passes that modify the lowered Go parse tree and as errors are
    discovered, several nodes transform into error nodes.  However, for a
    higher level node such as a construction expression, the erroneous
    nodes in the subtrees might not propagate their error.  The flatten
    phase for a node now looks for errors in the subtree and flattens the
    node into an error node if any are found.

    Fixes golang/go#11559, golang/go#11536, golang/go#11558.

    Reviewed-on: https://go-review.googlesource.com/13097

From-SVN: r226845

Daily bump.

From-SVN: r226844

re PR c++/53330 (new() operator can return NULL on a zero-length allocation)

2015-08-12  Paolo Carlini  <paolo.carlini@oracle.com>

PR c++/53330
* g++.dg/init/new42.C: New.

From-SVN: r226840

re PR target/67071 (GCC misses an optimization to load vector constants)

[gcc]
2015-08-12  Michael Meissner  <meissner@linux.vnet.ibm.com>

PR target/67071
* config/rs6000/predicates.md (easy_vector_constant_vsldoi): New
predicate to allow construction of vector constants using the
VSLDOI vector shift instruction.

* config/rs6000/rs6000-protos.h (vspltis_shifted): Add
declaration.

* config/rs6000/rs6000.c (vspltis_shifted): New function to return
the number of bytes to be shifted left and filled in with either
all zero or all one bits.
(gen_easy_altivec_constant): Call vsplitis_shifted if no other
methods exist.
(output_vec_const_move): On power8, generate XXLORC to generate
a vector constant with all 1's. Do a split if we need to use a
VSLDOI instruction.

* config/rs6000/rs6000.h (EASY_VECTOR_MSB): Use mode mask to
properly test for the MSB.

* config/rs6000/altivec.md (VSLDOI splitter): Add splitter for
vector constants that can be created with VSLDOI.

[gcc/testsuite]
2015-08-12  Michael Meissner  <meissner@linux.vnet.ibm.com>

PR target/67071
* gcc.target/powerpc/pr67071-1.c: New file to test PR 67071 new
vector constants.
* gcc.target/powerpc/pr67071-2.c: Likewise.
* gcc.target/powerpc/pr67071-3.c: Likewise.

From-SVN: r226836

Revert "replace several uses of the anon namespace with GCC_FINAL"

This reverts commit daa5a8a3cf9b04cd9af5544c61e12e6dca14f870.

From-SVN: r226834

re PR c++/67104 (Constant expression factory function initializes std::array with static storage duration strangely)

PR c++/67104
* constexpr.c (cxx_eval_array_reference): Fix typo.

From-SVN: r226833

re PR c++/67104 (Constant expression factory function initializes std::array with static storage duration strangely)

PR c++/67104
* constexpr.c (cxx_eval_array_reference): Handle sparse
CONSTRUCTORs.

From-SVN: r226830

re PR c++/67108 (ICE: in cxx_eval_call_expression, at cp/constexpr.c:1345 when dumping)

PR c++/67108
* decl2.c (c_parse_final_cleanups): Set at_eof to 2 at end.
* error.c (dump_template_bindings): Don't tsubst in that case.

From-SVN: r226829

xtensa: add -mauto-litpools option

With support from assembler this option allows compiling huge functions,
where single literal pool at the beginning of a function may not be
reachable by L32R instructions at its end.

Currently assembler --auto-litpools option cannot deal with literals
used from multiple locations separated by more than 256 KBytes of code.
Don't turn constants into literals, instead use MOVI instruction to load
them into registers and let the assembler turn them into literals as
necessary.

2015-08-12  Max Filippov  <jcmvbkbc@gmail.com>
gcc/
* config/xtensa/constraints.md (define_constraint "Y"): New
constraint.
* config/xtensa/elf.h (ASM_SPEC): Add m(no-)auto-litpools.
* config/xtensa/linux.h (ASM_SPEC): Likewise.
* config/xtensa/predicates.md (move_operand): Match constants
and symbols in the presence of TARGET_AUTO_LITPOOLS.
* config/xtensa/xtensa.c (xtensa_valid_move): Don't allow
immediate references to TLS data.
(xtensa_emit_move_sequence): Don't force constants to memory in
the presence of TARGET_AUTO_LITPOOLS.
(print_operand): Add 'y' format, same as default, but capable of
printing SF mode constants as well.
* config/xtensa/xtensa.md (movsi_internal, movhi_internal)
(movsf_internal): Add movi pattern that loads literal.
(movsf, movdf): Don't force constants to memory in the presence
of TARGET_AUTO_LITPOOLS.
(movdf_internal): Add 'Y' constraint.
* config/xtensa/xtensa.opt (mauto-litpools): New option.
* doc/invoke.text (Xtensa options): Document -mauto-litpools.

From-SVN: r226828

re PR c++/67161 (ICE with a static_assert using our internal __not/__or/__and traits)

PR c++/67161
* error.c (dump_decl) [TEMPLATE_ID_EXPR]: Pass
TFF_NO_OMIT_DEFAULT_TEMPLATE_ARGUMENTS.

From-SVN: r226827

re PR c++/55095 (Wshift-overflow)

PR c++/55095
* c-common.c (maybe_warn_shift_overflow): Properly handle
left-shifting 1 into the sign bit.

* c-c++-common/Wshift-overflow-6.c: New test.
* c-c++-common/Wshift-overflow-7.c: New test.
* g++.dg/cpp1y/left-shift-2.C: New test.

From-SVN: r226826

compiler: Don't make gc symbol for bad array type.

    When parsing a malformed array type, i.e. invalid length, gccgo
    would loop indefinitely based off of a uninitialized length variable.

    Fixes golang/go#11539.

    Reviewed-on: https://go-review.googlesource.com/13066

From-SVN: r226825

re PR c++/52742 (Initializing an array using brace initializer and template parameters)

2015-08-12  Paolo Carlini  <paolo.carlini@oracle.com>

PR c++/52742
* g++.dg/cpp0x/pr52742.C: New.

From-SVN: r226824

arm-fpus.def: Replace booleans with feature flags.

* config/arm/arm-fpus.def: Replace booleans with feature flags.
Update comment.
* config/arm/arm.c (ARM_FPU): Update macro.
* config/arm/arm.h (TARGET_NEON_FP16): Update feature test.
(TARGET_FP16): Likewise.
(TARGET_CRYPTO): Likewise.
(TARGET_NEON): Likewise.
(struct arm_fpu_desc): Remove fields neon, fp16 and crypto.  Add
field features.

From-SVN: r226821

Fix warnings when bootstrapping on darwin with vtable verification enabled.

libstdc++-v3/ChangeLog:

2015-08-11  Caroline Tice <cmtice@google.com>

        PR 66521, Contributed by Eric Gallager
        * acinclude.m4 (VTV_CXXLINKFLAGS): Make this variable OS-specific, and
        fix the rpath flag to work properly for darwin.
        * configure: Regenerated.

From-SVN: r226820

Remove --with-host-libstdcxx

2015-08-12  Tom de Vries  <tom@codesourcery.com>

PR other/67092
PR other/67098
* configure.ac: Remove --with_host_libstdcxx support.
* configure: Regenerate.

* doc/install.texi: Remove --with_host_libstdcxx item.  Update
--with-stage1-libs, --with-boot-ldflags and --with-boot-libs items
accordingly.  Mention default for --with-stage1-ldflags.

From-SVN: r226819

arm.h (arm_fpu_feature_set): New.

* config/arm/arm.h (arm_fpu_feature_set): New.
(ARM_FPU_FSET_HAS): New.
(FPU_FL_NONE): New.
(FPU_FL_NEON): New.
(FPU_FL_FP16): New.
(FPU_FL_CRYPTO): New.

From-SVN: r226818

Add Knights Landing support to __builtin_cpu_is

This patch adds Knights Landing support to __builtin_cpu_is.

gcc/testsuite/

* gcc.target/i386/builtin_target.c (check_intel_cpu_model):
Check Knights Landing support.

libgcc/

* config/i386/cpuinfo.c (processor_types): Add INTEL_KNL.
(get_intel_cpu): Add Knights Landing support.

From-SVN: r226817

re PR c++/47461 (warn_unused_result attribute ignored for templates)

2015-08-12  Paolo Carlini  <paolo.carlini@oracle.com>

PR c++/47461
* g++.dg/warn/Wunused-result-1.C: New.

From-SVN: r226816

[AArch64] Do not ICE after apologising for -mcmodel=large -fPIC

* config/aarch64/aarch64.c (initialize_aarch64_code_model): Break
after -mcmodel=large -fPIC sorry.

From-SVN: r226815

tree-ssa-sccvn.c (vn_nary_op_compute_hash): Also canonicalize comparison operand order and commutative ternary op operand order.

2015-08-12  Richard Biener  <rguenther@suse.de>

* tree-ssa-sccvn.c (vn_nary_op_compute_hash): Also canonicalize
comparison operand order and commutative ternary op operand order.
(sccvn_dom_walker::cond_stack): New state to track temporary
expressions.
(sccvn_dom_walker::after_dom_children): Remove tempoary expressions
no longer valid.
(sccvn_dom_walker::record_cond): Add a single temporary conditional
expression.
(sccvn_dom_walker::record_conds): Add a temporary conditional
expressions and all related expressions also true/false.
(sccvn_dom_walker::before_dom_children): Record temporary
expressions based on the controlling condition of a single
predecessor.  When trying to simplify a conditional statement
lookup expressions we might have inserted earlier.

* gcc.dg/tree-ssa/ssa-fre-47.c: New testcase.
* gcc.dg/tree-ssa/ssa-fre-48.c: Likewise.
* gcc.dg/tree-ssa/ssa-fre-49.c: Likewise.
* g++.dg/tree-ssa/pr61034.C: Adjust.
* gcc.dg/fold-compare-2.c: Likewise.
* gcc.dg/pr50763.c: Likewise.
* gcc.dg/predict-3.c: Likewise.
* gcc.dg/tree-ssa/20030709-2.c: Likewise.
* gcc.dg/tree-ssa/pr19831-3.c: Likewise.
* gcc.dg/tree-ssa/pr20657.c: Likewise.
* gcc.dg/tree-ssa/pr21001.c: Likewise.
* gcc.dg/tree-ssa/pr37508.c: Likewise.
* gcc.dg/tree-ssa/vrp04.c: Likewise.
* gcc.dg/tree-ssa/vrp07.c: Likewise.
* gcc.dg/tree-ssa/vrp09.c: Likewise.
* gcc.dg/tree-ssa/vrp16.c: Likewise.
* gcc.dg/tree-ssa/vrp20.c: Likewise.
* gcc.dg/tree-ssa/vrp25.c: Likewise.
* gcc.dg/tree-ssa/vrp87.c: Likewise.

From-SVN: r226814

re PR target/67127 ([ARM] Avoiding odd-number ldrd/strd in movdi introduced a regression on armeb-linux-gnueabihf)

2015-08-12  Yvan Roux  <yvan.roux@linaro.org>

PR target/67127
* config/arm/arm.md (movdi): Restrict illegitimate ldrd/strd checking
to ARM core registers.

From-SVN: r226811

tree-vrp.c (simplify_min_or_max_using_ranges): New.

* tree-vrp.c (simplify_min_or_max_using_ranges): New.
(simplify_stmt_using_ranges): Simplify MIN and MAX exprs.

testsuite/
* gcc.dg/vrp-min-max-1.c: New.
* gcc.dg/vrp-min-max-2.c: New.

From-SVN: r226809

c-decl.c (grokdeclarator): Call error_at instead of error and pass LOC to it.

* c-decl.c (grokdeclarator): Call error_at instead of error and pass
LOC to it.

From-SVN: r226807

appendix_contributing.xml: Update patch email policy.

* doc/xml/manual/appendix_contributing.xml: Update patch email policy.
* doc/xml/manual/status_cxx2017.xml: Update status table.
* doc/html/manual/*: Regenerate.

From-SVN: r226806

[MIPS] Scheduler fix for the 74k & 24k.

2015-08-12  Simon Dardis  <simon.dardis@imgtec.com>

gcc/
* config/mips/mips.c (mips_store_data_bypass_p): Bring code into
line with comments.
* config/mips/sb1.md: Update usage of mips_store_data_bypass_p.

From-SVN: r226805

gimple.h (remove_pointer): New trait.

2015-08-12  Richard Biener  <rguenther@suse.de>

* gimple.h (remove_pointer): New trait.
(GIMPLE_CHECK2): New inline template function.
(gassign::code_): New constant static member.
(is_a_helper<const gassign *>): Add.
(gimple_assign_lhs): Use GIMPLE_CHECK2 in the gimple overload
and forward to a new gassign overload with less checking and a
cheaper way to access the operand.
(gimple_assign_lhs_ptr): Likewise.
(gimple_assign_set_lhs): Likewise.
(gimple_assign_rhs1, gimple_assign_rhs1_ptr, gimple_assign_set_rhs1):
Likewise.
(gimple_assign_rhs2, gimple_assign_rhs2_ptr, gimple_assign_set_rhs2):
Likewise.
(gimple_assign_rhs3, gimple_assign_rhs3_ptr, gimple_assign_set_rhs3):
Likewise.
(gimple_assign_rhs_code): Likewise.
* gimple.c (gassign::code_): Define.

From-SVN: r226802

tree-ssa-pre.c (eliminate_dom_walker::before_dom_children): Eliminate edges marked as not executable by SCCVN.

2015-08-12  Richard Biener  <rguenther@suse.de>

* tree-ssa-pre.c (eliminate_dom_walker::before_dom_children):
Eliminate edges marked as not executable by SCCVN.
* tree-ssa-sccvn.c: Include gimple-iterator.h.
(cond_dom_walker): Rename to sccvn_dom_walker.
(sccvn_dom_walker::before_dom_children): Value-number defs
of all stmts.
(run_scc_vn): Remove loop value-numbering all SSA names.
Drop not visited SSA names to varying.

* gcc.dg/tree-ssa/ssa-fre-43.c: Adjust.

From-SVN: r226801

Fix vect/trapv-vect-reduc-4.c for sparc

2015-08-12  Tom de Vries  <tom@codesourcery.com>

PR testsuite/67175
* gcc.dg/vect/trapv-vect-reduc-4.c: Add reduction detection scan.
Handle absence of vector min/max for sparc.

From-SVN: r226800

Daily bump.

From-SVN: r226798

compiler: Check for EOF in malformed signatures.

    When parsing a malformed function declaration with invalid parameters,
    gccgo would infinitely loop looking for the end of the function
    declaration.

    Fixes golang/go#11530, golang/go#11531.

    Reviewed-on: https://go-review.googlesource.com/13065

From-SVN: r226795

compiler: Handle newlines in general comments.

    On comments, the specification says
    (http://golang.org/ref/spec#Comments): General comments start with the
    character sequence /* and continue through the character sequence */.
    A general comment containing one or more newlines acts like a newline,
    otherwise it acts like a space.

    Fixes golang/go#11528.

    Reviewed-on: https://go-review.googlesource.com/13064

From-SVN: r226794

replace several uses of the anon namespace with GCC_FINAL

gcc/ChangeLog:

2015-08-11  Trevor Saunders  <tbsaunde@tbsaunde.org>

* compare-elim.c, dce.c, dse.c, gimple-ssa-isolate-paths.c,
gimple-ssa-strength-reduction.c, graphite.c, init-regs.c,
ipa-pure-const.c, ipa-visibility.c, ipa.c, mode-switching.c,
omp-low.c, reorg.c, sanopt.c, trans-mem.c, tree-eh.c,
tree-if-conv.c, tree-ssa-copyrename.c, tree-ssa-dce.c,
tree-ssa-dom.c, tree-ssa-dse.c, tree-ssa-forwprop.c,
tree-ssa-sink.c, tree-ssanames.c, tree-stdarg.c, tree-tailcall.c,
tree-vect-generic.c, tree.c, ubsan.c, var-tracking.c,
vtable-verify.c, web.c: Use GCC_FINAL instead of the anonymous
namespace.

From-SVN: r226793

add GCC_FINAL to ansidecl.h

include/ChangeLog:

2015-08-11  Trevor Saunders  <tbsaunde+gcc@tbsaunde.org>

* ansidecl.h (GCC_FINAL): New macro.

From-SVN: r226792

compiler: Don't make void-typed temporaries.

    Fixes golang/go#11568.

    Reviewed-on: https://go-review.googlesource.com/12653

From-SVN: r226788

* gcc.target/i386/avx2-vect-shift.c: Remove unneeded dg directives.

From-SVN: r226787

configure.ac (tools/gjdoc): Depend on CREATE_WRAPPERS.

* configure.ac (tools/gjdoc): Depend on CREATE_WRAPPERS.
* configure: Regenerate.
* tools/Makefile.am: Remove unneeded dependencies for Automake 1.11.
* tools/Makefile.in: Regenerate.

From-SVN: r226785

re PR target/66954 (function multiversioning fails for target "aes")

libgcc/ChangeLog:

PR target/66954
* config/i386/cpuinfo.c (enum processor_features): Add FEATURE_PCLMUL.
(get_available_features): Handle FEATURE_PCLMUL.

gcc/ChangeLog:

PR target/66954
* config/i386/i386.c (get_builtin_code_for_version): Add P_PCLMUL
to enum feature_priority and feature_list.
(fold_builtin_cpu): Add F_PCLMUL to enum processor_features
and isa_names_table.

gcc/testsuite/ChangeLog:

PR target/66954
* g++.dg/ext/mv25.C: New test.

From-SVN: r226784

tree-vect-stmts.c (vectorizable_shift): Add missed test on vect_induction_def.

gcc/

* tree-vect-stmts.c (vectorizable_shift): Add missed test on
vect_induction_def.

gcc/testsuite:

* gcc.target/i386/avx2-vect-shift.c: New test.

From-SVN: r226781

re PR c/66098 (#pragma diagnostic 'ignored' not fully undone by pop for strict-overflow)

gcc/ChangeLog:

2015-08-11  Manuel López-Ibáñez  <manu@gcc.gnu.org>

PR c/66098
PR c/66711
* diagnostic.c (diagnostic_classify_diagnostic): Take -Werror into
account when deciding what was the command-line status.

gcc/testsuite/ChangeLog:

2015-08-11  Manuel López-Ibáñez  <manu@gcc.gnu.org>

PR c/66098
PR c/66711
* gcc.dg/pragma-diag-3.c: New test.
* gcc.dg/pragma-diag-4.c: New test.

From-SVN: r226780

* tree-vrp.c (simplify_abs_using_ranges): Simplify.

From-SVN: r226779

tree-ssa-phiopt.c (minmax_replacement): Create new ssa name if we're not the only contributor to target phi.

* tree-ssa-phiopt.c (minmax_replacement): Create new ssa name if
we're not the only contributor to target phi.

testsuite/
* c-c++-common/dfp/operator-comma.c: Call init function.
* c-c++-common/dfp/convert-dfp-2.c: New test.

From-SVN: r226778

[AArch64][Obivious] Add the missing ',' for enumeration element

2015-08-11  Jiong Wang  <jiong.wang@arm.com>

gcc/
  * config/aarch64/aarch64.h (REG_CLASS_NAMES): Add the missing ',' after
  FIXED_REG0.

From-SVN: r226777

Add todo comment for move_sese_region_to_fn

2015-08-11  Tom de Vries  <tom@codesourcery.com>

* tree-cfg.c (move_sese_region_to_fn): Add todo comment.

From-SVN: r226776

Fix test failure on Solaris 9 where strtod() does not accept hexadecimals

2015-08-11  Iain Buclaw  <ibuclaw@gdcproject.org>

libiberty/

* d-demangle.c (dlang_parse_real): Remove call to strtod.
(strtod): Remove declaration.
* testsuite/d-demangle-expected: Update float and complex literal
tests to check correct hexadecimal demangling.

From-SVN: r226774

Daily bump.

From-SVN: r226773

re PR libfortran/67140 (undefined reference to _gfortran_mvbits_i16)

2015-08-10  Steven G. Kargl  <kargl@gcc.gnu.org>

PR libfortran/67140
* intrinsics/mvbits.c: Fix build for paltforms without c_int128_t.

From-SVN: r226769

Treat model == 0x4f as Broadwell

gcc/testsuite/

* gcc.target/i386/builtin_target.c (check_intel_cpu_model):
Treat model == 0x4f as Broadwell.

libgcc/

* config/i386/cpuinfo.c (get_intel_cpu): Treat model == 0x4f as
Broadwell.

From-SVN: r226766

Update -mtune=knl for Knights Landing

From:

https://software.intel.com/sites/default/files/managed/e9/b5/Knights-Corner-is-your-path-to-Knights-Landing.pdf

Knights Landing is “Based on Intel Atom core (based on Silvermont
microarchitecture) with many HPC enhancements.”

This patch replaces CPU_KNL with CPU_SLM to tune for Knights Landing.

* config/i386/i386.c (processor_alias_table): Replace CPU_KNL
with CPU_SLM.
* config/i386/i386.md (cpu): Remove knl.

From-SVN: r226765

[multiple changes]

2015-08-10  Steven G. Kargl  <kargl@gcc.gnu.org>

PR libfortran/67140
* gfortran.map: Add _gfortran_mvbits_i16.
* intrinsics/mvbits.c: Generate mvbits for c_int128_t.

2015-08-10  Francois-Xavier Coudert  <fxcoudert@gcc.gnu.org>

PR libfortran/67140
* gfortran.dg/pr67140.f90: New test.

From-SVN: r226764

[PR libgomp/65742, PR middle-end/66332] libgomp: Remove plugin for non-shared memory host execution

gcc/
* builtins.c (expand_builtin_acc_on_device) [ACCEL_COMPILER]: Emit
open-coded sequence.
* omp-low.c (oacc_process_reduction_data): Remove handline of
GOMP_DEVICE_HOST_NONSHM.
gcc/testsuite/
* c-c++-common/goacc/acc_on_device-2.c: Remove XFAIL for C.
include/
* gomp-constants.c (GOMP_DEVICE_HOST_NONSHM): Remove.
libgomp/
* libgomp-plugin.h (enum offload_target_type): Remove
OFFLOAD_TARGET_TYPE_HOST_NONSHM.
* openacc.f90 (openacc_kinds): Remove acc_device_host_nonshm.
* openacc.h (enum acc_device_t): Likewise.
* openacc_lib.h: Likewise.
* oacc-init.c (name_of_acc_device_t): Don't handle it.
(acc_on_device): Just use __builtin_acc_on_device.
* testsuite/libgomp.oacc-c-c++-common/if-1.c: Don't forbid usage
of acc_on_device builtin.
* plugin/plugin-host.h: Remove file.
* plugin/plugin-host.c: Likewise, but salvage some content into...
* oacc-host.c: ... this file.
* plugin/Makefrag.am: Don't build libgomp-plugin-host_nonshm.la.
* plugin/configfrag.ac (offload_targets): Don't add host_nonshm.
* Makefile.in: Regenerate.
* configure: Likewise.
* testsuite/lib/libgomp.exp
(check_effective_target_openacc_host_nonshm_selected): Remove.
* testsuite/libgomp.oacc-c++/c++.exp: Don't handle
ACC_DEVICE_TYPE=host_nonshm.
* testsuite/libgomp.oacc-c/c.exp: Likewise.
* testsuite/libgomp.oacc-fortran/fortran.exp: Likewise.
* testsuite/libgomp.oacc-c-c++-common/acc_on_device-1.c: Likewise.
* testsuite/libgomp.oacc-fortran/acc_on_device-1-1.f90: Likewise.
* testsuite/libgomp.oacc-fortran/acc_on_device-1-2.f: Likewise.
* testsuite/libgomp.oacc-fortran/acc_on_device-1-3.f: Likewise.

From-SVN: r226763

Empty libgomp for nvptx

* configure.ac (noconfigdirs): Don't add "target-libgomp" for target
nvptx*-*-*.
* configure: Regenerate.
libgomp/
* config/nvptx/affinity.c: New file.
* config/nvptx/alloc.c: Likewise.
* config/nvptx/bar.c: Likewise.
* config/nvptx/barrier.c: Likewise.
* config/nvptx/critical.c: Likewise.
* config/nvptx/env.c: Likewise.
* config/nvptx/error.c: Likewise.
* config/nvptx/fortran.c: Likewise.
* config/nvptx/iter.c: Likewise.
* config/nvptx/iter_ull.c: Likewise.
* config/nvptx/libgomp-plugin.c: Likewise.
* config/nvptx/lock.c: Likewise.
* config/nvptx/loop.c: Likewise.
* config/nvptx/loop_ull.c: Likewise.
* config/nvptx/mutex.c: Likewise.
* config/nvptx/oacc-async.c: Likewise.
* config/nvptx/oacc-cuda.c: Likewise.
* config/nvptx/oacc-host.c: Likewise.
* config/nvptx/oacc-init.c: Likewise.
* config/nvptx/oacc-mem.c: Likewise.
* config/nvptx/oacc-parallel.c: Likewise.
* config/nvptx/oacc-plugin.c: Likewise.
* config/nvptx/omp-lock.h: Likewise.
* config/nvptx/ordered.c: Likewise.
* config/nvptx/parallel.c: Likewise.
* config/nvptx/proc.c: Likewise.
* config/nvptx/ptrlock.c: Likewise.
* config/nvptx/sections.c: Likewise.
* config/nvptx/sem.c: Likewise.
* config/nvptx/single.c: Likewise.
* config/nvptx/splay-tree.c: Likewise.
* config/nvptx/target.c: Likewise.
* config/nvptx/task.c: Likewise.
* config/nvptx/team.c: Likewise.
* config/nvptx/time.c: Likewise.
* config/nvptx/work.c: Likewise.
* configure.ac: Don't probe pthreads support for host nvptx*-*-*.
* configure: Regenerate.
* configure.tgt (config_path): Set to "nvptx" for target
nvptx*-*-*.

Co-Authored-By: Jakub Jelinek <jakub@redhat.com>
From-SVN: r226760

Fix offloading machine mode stream reading

... in context of the GET_MODE_INNER changes applied in r226328.

gcc/
* lto-streamer-in.c (lto_input_mode_table): Adjust to
GET_MODE_INNER changes.
libgomp/
* testsuite/libgomp.oacc-c-c++-common/vector-type-1.c: New file.

From-SVN: r226759

Correctly advance iterator in offloading machine mode stream reading

gcc/
* lto-streamer-in.c (lto_input_mode_table): Correctly advance
iterator.

Co-Authored-By: Ilya Verbin <ilya.verbin@intel.com>
From-SVN: r226758

[AArch64] Recommit correct version for improving TLS descriptor pattern

2015-08-06    Ramana Radhakrishnan  <ramana.radhakrishnan@arm.com>
      Jiong Wang  <jiong.wang@arm.com>

gcc/
  * config/aarch64/aarch64.d (tlsdesc_small_pseudo_<mode>): New pattern.
  * config/aarch64/aarch64.h (reg_class): New enumeration FIXED_REG0.
  (REG_CLASS_NAMES): Likewise.
  (REG_CLASS_CONTENTS): Likewise.
  * config/aarch64/aarch64.c (aarch64_class_max_nregs): Likewise.
  (aarch64_register_move_cost): Likewise.
  (aarch64_load_symref_appropriately): Invoke the new added pattern if
  possible.
  * config/aarch64/constraints.md (Uc0): New constraint.

gcc/testsuite/
  * gcc.target/aarch64/tlsdesc_hoist.c: New testcase.

From-SVN: r226757

[AArch64] Revert wrong commit of "Improve TLS Descriptor pattern to release RTL loop IV opt"

Revert the wrong commit of

  2015-08-06    Ramana Radhakrishnan  <ramana.radhakrishnan@arm.com>
      Jiong Wang  <jiong.wang@arm.com>

* config/aarch64/aarch64.d (tlsdesc_small_pseudo_<mode>): New pattern.
* config/aarch64/aarch64.h (reg_class): New enumeration FIXED_REG0.
(REG_CLASS_NAMES): Likewise.
(REG_CLASS_CONTENTS): Likewise.
* config/aarch64/aarch64.c (aarch64_class_max_nregs): Likewise.
(aarch64_register_move_cost): Likewise.
(aarch64_load_symref_appropriately): Invoke the new added pattern if
possible.
* config/aarch64/constraints.md (Uc0): New constraint.

From-SVN: r226756

re PR fortran/64022 ([F2003][IEEE] ieee_support_flag does not handle kind=10 and kind=16 REAL variables)

PR fortran/64022
* gfortran.dg/ieee/large_4.f90: New test.

From-SVN: r226755

Daily bump.

From-SVN: r226754