runtime: copy Go 1.7 runtime semaphore code

    This triggered a check in releaseSudog that g.param not nil, because
    libgo uses the param field when starting a goroutine.  Fixed by clearing
    g->param in kickoff in proc.c.

    Reviewed-on: https://go-review.googlesource.com/30951

From-SVN: r241067

shrink-wrap: Testcases for separate shrink-wrapping

A few testcases for separate shrink-wrapping: test whether it works in a
trivial case; whether it creates more than one prologue where that is
useful; whether it puts prologues inside a loop if that is cheaper.

gcc/testsuite/
* gcc.target/powerpc/shrink-wrap-separate-0.c: New testcase.
* gcc.target/powerpc/shrink-wrap-separate-1.c: New testcase.
* gcc.target/powerpc/shrink-wrap-separate-2.c: New testcase.

From-SVN: r241066

rs6000: Separate shrink-wrapping

This implements the hooks for separate shrink-wrapping for rs6000.
It handles GPRs and LR.  The GPRs get a component number corresponding
to their register number; LR gets component number 0.

* config/rs6000/rs6000.c (machine_function): Add new fields
gpr_is_wrapped_separately and lr_is_wrapped_separately.
(TARGET_SHRINK_WRAP_GET_SEPARATE_COMPONENTS,
TARGET_SHRINK_WRAP_COMPONENTS_FOR_BB,
TARGET_SHRINK_WRAP_DISQUALIFY_COMPONENTS,
TARGET_SHRINK_WRAP_EMIT_PROLOGUE_COMPONENTS,
TARGET_SHRINK_WRAP_EMIT_EPILOGUE_COMPONENTS,
TARGET_SHRINK_WRAP_SET_HANDLED_COMPONENTS): Define.
(rs6000_get_separate_components): New function.
(rs6000_components_for_bb): New function.
(rs6000_disqualify_components): New function.
(rs6000_emit_prologue_components): New function.
(rs6000_emit_epilogue_components): New function.
(rs6000_set_handled_components): New function.
(rs6000_emit_prologue): Don't emit LR save if lr_is_wrapped_separately.
Don't emit GPR saves if gpr_is_wrapped_separately for that register.
(restore_saved_lr): Don't restore LR if lr_is_wrapped_separately.
(rs6000_emit_epilogue): Don't emit GPR restores if
gpr_is_wrapped_separately for that register.  Don't make a
REG_CFA_RESTORE note for registers we did not restore, either.

From-SVN: r241065

Minor reformatting.

From-SVN: r241064

shrink-wrap: Shrink-wrapping for separate components

This is the main substance of this patch series.

Instead of doing all of the prologue and epilogue in one spot, it often
is better to do components of it at different places, so that they are
executed less frequently.

What exactly is a component is completely up to the target; this code
treats it all abstractly, and uses hooks for the target to handle the
more concrete things.  Commonly there is one component for each callee-
saved register, for example.

Components can be executed more than once per function execution.  This
pass makes sure that a component's epilogue is not called more often
than the corresponding prologue has been, at any point in time; that the
prologue is called more often, wherever the prologue's effect is needed;
and that the epilogue is called as often as the prologue has been, when
the function exits.  It does this by first deciding which blocks need
which components active, and then placing prologue and epilogue
components to make that exactly true.

Deciding what blocks should run with a certain component active so that
the total cost of executing the prologues (and epilogues) is optimal, is
not a computationally feasible problem.  Instead, for each basic block,
we estimate the cost of putting a prologue right before the block, and
if that is cheaper than the total cost of putting prologues optimally
(according to the estimated cost) in the dominator subtrees strictly
dominated by this first block, place it at the first block instead.
This simple procedure places the components optimally for any dominator
sub tree where the root node's cost does not depend on anything outside
its subtree.

The cost is the execution frequency of all edges into the block coming
from blocks that do not have this component active.  The estimated cost
is the execution frequency of the block, minus the execution frequency
of any backedges (which by definition are coming from subtrees, so if
the "head" block gets a prologue, the source block of any backedge has
that component active as well).

Currently, the epilogues are placed as late as possible, given the
constraints.  This does not matter for execution cost, but we could
save a little bit of code size by placing the epilogues in a smarter
way.  This is a possible future optimisation.

Now all that is left is inserting prologues and epilogues on all edges
that jump into resp. out of the "active" set of blocks.  Often we need
to insert some components' prologues (or epilogues) on all edges into
(or out of) a block.  In theory cross-jumping can unify all such, but
in practice that often fails; besides, that is a lot of work.  So in
this case we insert the prologue and epilogue components at the "head"
or "tail" of a block, instead.

As a final optimisation, if a block needs a prologue and its immediate
dominator has the block as a post-dominator, that immediate dominator
gets the prologue as well.

* function.c (thread_prologue_and_epilogue_insns): Call
try_shrink_wrapping_separate.  Compute the prologue_seq afterwards,
if it has possibly changed.  Compute the split_prologue_seq and
epilogue_seq later, too.
* shrink-wrap.c: #include cfgbuild.h and insn-config.h.
(dump_components): New function.
(struct sw): New struct.
(SW): New function.
(init_separate_shrink_wrap): New function.
(fini_separate_shrink_wrap): New function.
(place_prologue_for_one_component): New function.
(spread_components): New function.
(disqualify_problematic_components): New function.
(emit_common_heads_for_components): New function.
(emit_common_tails_for_components): New function.
(insert_prologue_epilogue_for_components): New function.
(try_shrink_wrapping_separate): New function.
* shrink-wrap.h: Declare try_shrink_wrapping_separate.

From-SVN: r241063

Define std::sample for C++17

* doc/xml/manual/status_cxx2017.xml: Add std::sample status.
* doc/html/*: Regenerate.
* include/experimental/algorithm (__sample): Move to bits/stl_algo.h
and into namespace std.
* include/bits/stl_algo.h (__sample): Define here. Fix invalid use
of input iterator. Defend against overloaded comma operator.
(sample): Define for C++17.
* testsuite/25_algorithms/sample/1.cc: New test.

From-SVN: r241062

regrename: Don't rename restores

A restore is supposed to restore some certain register.  Restoring it
into some other register will not work.  Don't.

* regrename.c (build_def_use): Invalidate chains that have a
REG_CFA_RESTORE on some instruction.

From-SVN: r241061

dce: Don't dead-code delete separately wrapped restores

If there is a separately wrapped register restore on some path that
is dead (say, control goes into an endless loop after it), then we
cannot delete that restore because that would confuse the DWARF CFI
(if there is another path joining).
This happens with gcc.dg/torture/pr53168.c, for example.

* dce.c (delete_unmarked_insns): Don't delete instructions with
a REG_CFA_RESTORE note.

From-SVN: r241060

separate shrink-wrap: New command-line flag, status flag, hooks, and doc

This patch adds a new command-line flag "-fshrink-wrap-separate", a status
flag "shrink_wrapped_separate", hooks for abstracting the target components,
and documentation for all those.

* common.opt (-fshrink-wrap-separate): New flag.
* doc/invoke.texi: Document it.
* doc/tm.texi.in (Shrink-wrapping separate components): New subsection.
* doc/tm.texi: Regenerate.
* emit-rtl.h (struct rtl_data): New field shrink_wrapped_separate.
* target.def (shrink_wrap): New hook vector.
(get_separate_components, components_for_bb, disqualify_components,
emit_prologue_components, emit_epilogue_components,
set_handled_components): New hooks.

From-SVN: r241059

rs6000: Warn for "GCC vector passed by reference" only with -Wpsabi

* config/rs6000/rs6000.c (rs6000_return_in_memory): Warn for
vector return by reference only if -Wpsabi.
(rs6000_pass_by_reference): Similarly, for argument passing.

From-SVN: r241058

print_rtx_function: integrate dumping of the CFG into the insn chain

gcc/ChangeLog:
* function-tests.c: Include "print-rtl.h".
(selftest::test_expansion_to_rtl): Call print_rtx_function on the
function, and verify what is dumped.
* print-rtl-function.c (print_edge): New function.
(begin_any_block): New function.
(end_any_block): New function.
(can_have_basic_block_p): New function.
(print_rtx_function): Track the basic blocks of insns in the
chain, wrapping those that are within blocks within "(block)"
directives.  Remove the "(cfg)" directive.

From-SVN: r241057

rs6000: Use "dg-warning ... 0" in a few places

GCC used to give warnings for line # 1 for these, but it prints no
line number anymore.  Adjust.

gcc/testsuite/
* gcc.target/powerpc/warn-1.c: Change line number in dg-warning.
* gcc.target/powerpc/warn-2.c: Ditto.

From-SVN: r241056

Add selftest::read_file

gcc/ChangeLog:
* selftest.c (selftest::read_file): New function.
(selftest::test_read_file): New function.
(selftest::selftest_c_tests): Call test_read_file.
* selftest.h (selftest::read_file): New decl.

From-SVN: r241055

[MIPS] Add MSA tests

gcc/testsuite/ChangeLog:

* gcc.dg/vect/slp-26.c: Check if vectorized for MIPS MSA.
* gcc.dg/vect/tree-vect.h (check_vect): Check for MIPS SIMD support.
* gcc.target/mips/mips.exp: Add support for -mmsa.  Imply -mno-mips16
for -mmsa.
* gcc.target/mips/msa.c: New test.
* gcc.target/mips/msa-builtins.c: Likewise.
* lib/target-supports.exp (check_mips_msa_hw_available): New.
(check_effective_target_mips_msa_runtime): Likewise.
(check_effective_target_mips_msa): Likewise.
(add_options_for_mips_msa): Likewise.
(check_effective_target_vect_int): Return TRUE for MIPS MSA.
(check_effective_target_vect_intfloat_cvt): Likewise.
(check_effective_target_vect_uintfloat_cvt): Likewise.
(check_effective_target_vect_floatint_cvt): Likewise.
(check_effective_target_vect_floatuint_cvt): Likewise.
(check_effective_target_vect_shift): ewise.
(check_effective_target_vect_shift_char): Likewise.
(check_effective_target_vect_long): Likewise.
(check_effective_target_vect_float): Likewise.
(check_effective_target_vect_double): Likewise.
(check_effective_target_vect_long_long): Likewise.
(check_effective_target_vect_perm): Likewise.
(check_effective_target_vect_perm_byte): Likewise.
(check_effective_target_vect_perm_short): Likewise.
(check_effective_target_vect_sdot_qi): Likewise.
(check_effective_target_vect_udot_qi): Likewise.
(check_effective_target_vect_sdot_hi): Likewise.
(check_effective_target_vect_udot_hi): Likewise.
(check_effective_target_vect_pack_trunc): Likewise.
(check_effective_target_vect_unpack): Likewise.
(check_effective_target_vect_hw_misalign): Likewise.
(check_effective_target_vect_condition): Likewise.
(check_effective_target_vect_cond_mixed): Likewise.
(check_effective_target_vect_char_mult): Likewise.
(check_effective_target_vect_short_mult): Likewise.
(check_effective_target_vect_int_mult): Likewise.
(check_effective_target_vect_extract_even_odd): Likewise.
(check_effective_target_vect_interleave): Likewise.
(check_vect_support_and_set_flags): Check if the target supports MSA
and append to the list of EFFECTIVE_TARGETS.

From-SVN: r241054

re PR debug/77947 (ICE with -g and -O2 in strip_naming_typedef)

2016-10-12  Richard Biener  <rguenther@suse.de>

PR debug/77947
* cgraphunit.c (analyze_functions): Preserve cgraph nodes
function context.

* g++.dg/torture/pr77947.C: New testcase.

From-SVN: r241053

[multiple changes]

2016-10-12  Yannick Moy  <moy@adacore.com>

* einfo.adb, einfo.ads (Partial_Refinement_Constituents): Take
into account constituents that are themselves abstract states
with full or partial refinement visible.
* sem_prag.adb (Find_Encapsulating_State): Move function
to library-level, to share between subprograms.
(Analyze_Refined_Global_In_Decl_Part): Use
Find_Encapsulating_State to get relevant encapsulating state.

2016-10-12  Arnaud Charlet  <charlet@adacore.com>

* gnat1drv.adb: Fix minor typo.

From-SVN: r241052

syscall, internal/syscall/unix: Fix getrandom, clone on sparc64

    Since sparc is a valid architecture, the name of
    getrandom_linux_sparc.go means that it will be ignored on sparc64,
    even though it's whitelisted with a +build line.

    On SPARC, clone has a unique return value convention which requires
    some inline assembly to convert it to the normal convention.

    Reviewed-on: https://go-review.googlesource.com/30873

From-SVN: r241051

sem_prag.adb (Analyze_Refined_Depends_In_Decl_Part): Adapt checking for optional refinement of abstract state with partial visible...

2016-10-12  Yannick Moy  <moy@adacore.com>

* sem_prag.adb (Analyze_Refined_Depends_In_Decl_Part): Adapt checking
for optional refinement of abstract state with partial
visible refinement.
(Analyze_Refined_Global_In_Decl_Part): Adapt checking for optional
refinement of abstract state with partial visible refinement. Implement
new rules in SPARK RM 7.2.4 related to optional refinement.
Also fix the missing detection of missing items.

From-SVN: r241050

[multiple changes]

2016-10-12  Hristian Kirtchev  <kirtchev@adacore.com>

* einfo.adb Add new usage for Elist29 and Node35.
(Anonymous_Designated_Type): New routine.
(Anonymous_Master): Removed.
(Anonymous_Masters): New routine.
(Set_Anonymous_Designated_Type): New routine.
(Set_Anonymous_Master): Removed.
(Set_Anonymous_Masters): New routine.
(Write_Field29_Name): Add output for Anonymous_Masters.
(Write_Field35_Name): Remove the output for Anonymous_Master. Add
output for Anonymous_Designated_Type.
* einfo.ads Remove attribute Anonymous_Master along with
usage in entities. Add attributes Anonymous_Designated_Type
and Anonymous_Masters along with usage in entities.
(Anonymous_Designated_Type): New routine along with pragma Inline.
(Anonymous_Master): Removed along with pragma Inline.
(Anonymous_Masters): New routine along with pragma Inline.
(Set_Anonymous_Designated_Type): New routine along with pragma Inline.
(Set_Anonymous_Master): Removed along with pragma Inline.
(Set_Anonymous_Masters): New routine along with pragma Inline.
* exp_ch7.adb (Build_Anonymous_Master): Reuse an anonymous master
defined in the same unit if it services the same designated
type, otherwise create a new one.
(Create_Anonymous_Master): Reimplemented.
(Current_Anonymous_Master): New routine.
(In_Subtree): Removed.

2016-10-12  Ed Schonberg  <schonberg@adacore.com>

* sem_prag.adb (Analyze_Pragma, case Dynamic_Predicate):
Check properly whether there is an explicit assertion policy
for predicate checking, even in the presence of a general Ignore
assertion policy.

2016-10-12  Steve Baird  <baird@adacore.com>

* sem.adb (Walk_Library_Items): Cope with ignored ghost units.

From-SVN: r241049

[multiple changes]

2016-10-12  Ed Schonberg  <schonberg@adacore.com>

* lib-writ.adb (Write_ALI): Removal of unused file entries from
dependency list must be performed before the list is sorted,
so that the dependency number of other files is properly set-up
for use in tools that relate entity information to the unit in
which they are declared.

2016-10-12  Hristian Kirtchev  <kirtchev@adacore.com>

* exp_aggr.adb (Initialize_Ctrl_Array_Component):
Create a copy of the initialization expression to avoid sharing
it between multiple components.

2016-10-12  Yannick Moy  <moy@adacore.com>

* einfo.adb, einfo.ads (Has_Partial_Visible_Refinement): New flag
in abtract states.
(Has_Non_Null_Visible_Refinement): Return true for patial refinement.
(Partial_Refinement_Constituents): New function returns the full or
partial refinement constituents depending on scope.
* sem_ch3.adb (Analyze_Declarations): Remove partial visible
refinements when exiting the scope of a package spec or body
and those partial refinements are not in scope afterwards.
* sem_ch7.adb, sem_ch7.ads (Install_Partial_Declarations): Mark
abstract states of parent units with partial refinement so that
it is visible.
* sem_prag.adb (Analyze_Part_Of_In_Decl_Part): Mark enclosing
abstract state if any as having partial refinement in that scope.
(Analyze_Refined_Global_In_Decl_Part): Check constituent usage
based on full or partial refinement depending on scope.

2016-10-12  Ed Schonberg  <schonberg@adacore.com>

* exp_ch4.adb (Expand_N_Type_Conversion): If the target type
has an invariant aspect, insert invariant call at the proper
place in the code rather than rewriting the expression as an
expression with actions, to prevent spurious semantic errors on
the rewritten conversion when it is the object in a renaming.

From-SVN: r241048

Make test helper work with non-copyable types

* testsuite/util/testsuite_common_types.h
(bitwise_assignment_operators): Use direct-initialization for C++11
and later, to avoid CopyConstructible requirement.
* testsuite/29_atomics/atomic/cons/assign_neg.cc: Adjust dg-error
line numbers.
* testsuite/29_atomics/atomic/cons/copy_neg.cc: Likewise.
* testsuite/29_atomics/atomic_integral/cons/assign_neg.cc: Likewise.
* testsuite/29_atomics/atomic_integral/cons/copy_neg.cc: Likewise.
* testsuite/29_atomics/atomic_integral/operators/bitwise_neg.cc:
Adjust expected errors and line numbers.

From-SVN: r241047

Fix LTO_STREAMER_DEBUG build

gcc/
* lto-streamer.c: Fix LTO_STREAMER_DEBUG build.

From-SVN: r241046

Un-break dwarf2out for DWARF2_LINENO_DEBUGGING_INFO configurations

gcc/
* dwarf2out.c (dwarf2_lineno_debug_hooks): Use
dwarf2out_assembly_start.

From-SVN: r241045

Make GCC selftests work for *-wrs-vxworks-* targets

gcc/
* Makefile.in (SELFTEST_FLAGS): Add -nostdinc.

From-SVN: r241044

In gcc/Makefile.in, factor out SELFTEST_FLAGS

gcc/
* Makefile.in (SELFTEST_FLAGS): New variable.
(s-selftest, selftest-gdb, selftest-valgrind): Use it.

From-SVN: r241043

Fix gcc/vmsdbgout.c:vmsdbg_debug_hooks for "More trivial bits from early LTO debug merge"

gcc/
* vmsdbgout.c (vmsdbg_debug_hooks): Add filename parameter to
early_finish hook.

From-SVN: r241042

exp_ch5.adb, [...]: Minor reformatting.

2016-10-12  Hristian Kirtchev  <kirtchev@adacore.com>

* exp_ch5.adb, sem_ch3.adb, exp_ch9.adb, a-tags.adb, sem_prag.adb,
sem_ch12.adb, xref_lib.adb, a-strunb-shared.adb, rtsfind.adb,
freeze.adb, sem_attr.adb, sem_case.adb, exp_ch4.adb, ghost.adb,
exp_ch6.adb, sem_ch4.adb, restrict.adb, s-os_lib.adb: Minor
reformatting.

From-SVN: r241041

[multiple changes]

2016-10-12  Justin Squirek  <squirek@adacore.com>

* sem_ch10.adb (Remove_Limited_With_Clause): Add a check to
detect accidental visibility.

2016-10-12  Ed Schonberg  <schonberg@adacore.com>

* exp_ch4.adb (Expand_Allocator): If the expression is a qualified
expression, add a predicate check after the constraint check.
* sem_res.adb (Resolve_Qualified_Expression): If context is an
allocator, do not apply predicate check, as it will be done when
allocator is expanded.

From-SVN: r241040

[multiple changes]

2016-10-12  Bob Duff  <duff@adacore.com>

* xref_lib.adb: Use renamings-of-slices to ensure
that all references to Tables are properly bounds checked (when
checks are turned on).
* g-dyntab.ads, g-dyntab.adb: Default-initialize the array
components, so we don't get uninitialized pointers in case
of Tables containing access types.  Misc cleanup of the code
and comments.

2016-10-12  Ed Schonberg  <schonberg@adacore.com>

* sem_attr.adb (Analyze_Attribute, case 'Type_Key): Implement
functionality of attribute, to provide a reasonably unique key
for a given type and detect any changes in the semantics of the
type or any of its subcomponents from version to version.

2016-10-12  Bob Duff  <duff@adacore.com>

* sem_case.adb (Check_Choice_Set): Separate
checking for duplicates out into a separate pass from checking
full coverage, because the check for duplicates does not depend
on predicates. Therefore, we shouldn't do it separately for the
predicate vs. no-predicate case; we should share code. The code
for the predicate case was wrong.

From-SVN: r241039

rtl.h (struct rtx_def): Comment how RTX_FLAGS will be dumped in RTL dumps.

* rtl.h (struct rtx_def): Comment how RTX_FLAGS will be
dumped in RTL dumps.

From-SVN: r241037

[multiple changes]

2016-10-12  Jerome Lambourg  <lambourg@adacore.com>

* init.c: Make sure to call finit on x86_64-vx7 to reinitialize
the FPU unit.

2016-10-12  Arnaud Charlet  <charlet@adacore.com>

* lib-load.adb (Load_Unit): Generate an error message even when
Error_Node is null.

2016-10-12  Ed Schonberg  <schonberg@adacore.com>

* lib-writ.adb (Write_ALI): Disable optimization related to transitive
limited_with clauses for now.

2016-10-12  Javier Miranda  <miranda@adacore.com>

* sem_attr.adb (Analyze_Attribute_Old_Result): Generating C
code handle 'old located in inlined _postconditions procedures.
(Analyze_Attribute [Attribute_Result]): Handle 'result when
rewriting the attribute as a reference to the formal parameter
_Result of inlined _postconditions procedures.

2016-10-12  Tristan Gingold  <gingold@adacore.com>

* s-rident.ads (Profile_Info): Remove
Max_Protected_Entries restriction from GNAT_Extended_Ravenscar
* sem_ch9.adb (Analyze_Protected_Type_Declaration):
Not a controlled type on restricted runtimes.

2016-10-12  Gary Dismukes  <dismukes@adacore.com>

* sem_ch3.adb (Derive_Subprogram): Add test
for Is_Controlled of Parent_Type when determining whether an
inherited subprogram with one of the special names Initialize,
Adjust, or Finalize should be derived with its normal name even
when inherited as a private operation (which would normally
result in the inherited operation having a special "hidden" name).

2016-10-12  Ed Schonberg  <schonberg@adacore.com>

* sem_res.adb (Resolve_Call): If a function call returns a
limited view of a type replace it with the non-limited view,
which must be available when compiling call.  This was already
done elsewhere for non-overloaded calls, but needs to be done
after resolution if function name is overloaded.

2016-10-12  Javier Miranda  <miranda@adacore.com>

* a-tags.adb (IW_Membership [private]): new overloaded
subprogram that factorizes the code needed to check if a
given type implements an interface type.
(IW_Membership
[public]): invoke the new internal IW_Membership function.
(Is_Descendant_At_Same_Level): Fix this routine to implement RM
3.9 (12.3/3)

From-SVN: r241036

[multiple changes]

2016-10-12  Tristan Gingold  <gingold@adacore.com>

* exp_ch9.adb (Expand_N_Delay_Relative_Statement): Add support
for a secondary procedure in case of missing Ada.Calendar.Delays
* rtsfind.ads (RTU_Id): Add System_Relative_Delays.
(RE_Id): Add RO_RD_Delay_For.
* rtsfind.adb (Output_Entity_Name): Handle correctly units RO_XX.
* s-rident.ads: Remove No_Relative_Delays
restriction for GNAT_Extended_Ravenscar.

2016-10-12  Ed Schonberg  <schonberg@adacore.com>

* sem_elab.adb (Within_Initial_Condition):  When deternining
the context of the expression, use the original node if it is
a pragma, because Check pragmas are rewritten as conditionals
when assertions are not enabled.

2016-10-12  Bob Duff  <duff@adacore.com>

* spitbol_table.ads, spitbol_table.adb (Adjust, Finalize): Add
"overriding".

2016-10-12  Bob Duff  <duff@adacore.com>

* a-strunb-shared.ads, a-strunb-shared.adb (Finalize):
Make sure Finalize is idempotent.
(Unreference): Check for
Empty_Shared_String, in case the reference count of the empty
string wraps around.
Also add "not null" in various places that can't be null.

2016-10-12  Jerome Lambourg  <lambourg@adacore.com>

* init.c: Fix sigtramp with the x86_64-vx7-vxsim target on
Windows host.

2016-10-12  Vadim Godunko  <godunko@adacore.com>

* s-os_lib.ads (Is_Owner_Readable_File): Renamed from
Is_Readable_File.
(Is_Owner_Writable_File): Renamed from Is_Writable_File.
(Is_Readable_File): Renames Is_Read_Accessible_File.
(Is_Writable_File): Renames Is_Write_Accessible_File.

From-SVN: r241035

Simplify std::call_once implementation

* include/std/mutex [_GLIBCXX_HAVE_TLS] (_Once_call): Remove.
(call_once) [_GLIBCXX_HAVE_TLS]: Simplify by removing _Once_call.

From-SVN: r241034

Add missing return to uninitialized_default_construct_n

* include/bits/stl_uninitialized.h
(__uninitialized_default_novalue_n_1<true>): Add missing return.
* testsuite/20_util/specialized_algorithms/memory_management_tools/
1.cc: Check return values of uninitialized_xxx_n algorithms.

From-SVN: r241033

Remove 'return' from noreturn function

* libsupc++/nested_exception.h (throw_with_nested): Remove return.

From-SVN: r241032

Do not copy std:call_once arguments (LWG 2442)

* doc/xml/manual/intro.xml: Document LWG 2442 status.
* include/std/mutex [_GLIBCXX_HAVE_TLS] (__once_call_impl): Remove.
[_GLIBCXX_HAVE_TLS] (_Once_call): Declare primary template and define
partial specialization to unpack args and forward to std::invoke.
(call_once) [_GLIBCXX_HAVE_TLS]: Use forward_as_tuple and _Once_call
instead of __bind_simple and __once_call_impl.
(call_once) [!_GLIBCXX_HAVE_TLS]: Use __invoke instead of
__bind_simple.
* testsuite/30_threads/call_once/dr2442.cc: New test.

From-SVN: r241031

Add a helper function: create_tmp_reg_or_ssa_name

* gimple-fold.c (create_tmp_reg_or_ssa_name): New function.
(gimple_fold_builtin_memory_op): Use the function.
(gimple_fold_builtin_strchr): Likewise.
(gimple_fold_builtin_strcat): Likewise.
(gimple_build): Likewise.

From-SVN: r241030

sem_ch12.adb (Check_Formal_Package_Instance): Skip an internal formal entity without a parent only if...

2016-10-12  Ed Schonberg  <schonberg@adacore.com>

* sem_ch12.adb (Check_Formal_Package_Instance): Skip an internal
formal entity without a parent only if the corresponding actual
entity has a different kind.
* exp_ch9.adb (Build_Class_Wide_Master): If the master is
declared locally, insert the renaming declaration after the
master declaration, to prevent access before elaboration in gigi.

From-SVN: r241029

contracts.adb (Analyze_Contracts): For a type declaration, analyze an iterable aspect when present.

2016-10-12  Ed Schonberg  <schonberg@adacore.com>

* contracts.adb (Analyze_Contracts): For a type declaration, analyze
an iterable aspect when present.

From-SVN: r241028

* diagnostic.c (diagnostc_report_diagnostic): Fix formatting.

From-SVN: r241027

[multiple changes]

2016-10-12  Ed Schonberg  <schonberg@adacore.com>

* sem_ch12.adb (Check_Formal_Package_Instance): Handle properly
an instance of a formal package with defaults, when defaulted
parameters include tagged private types and array types.

2016-10-12  Tristan Gingold  <gingold@adacore.com>

* restrict.ads, restrict.adb (Restricted_Profile): Adjust
comment, use Restricted_Tasking to compare restrictions.
* s-rident.ads (Profile_Name): Add Restricted_Tasking and
reorder literals.
(Profile_Info): Set restrictions for Restricted_Tasking.

2016-10-12  Ed Schonberg  <schonberg@adacore.com>

* sem_ch3.adb (Analyze_Full_Type_Declaration): Set Ghost status
of type before elaborating inherited operations, so that the
Ghost status is set properly for them.
* ghost.adb (Check_Ghost_Overriding): A ghost subprogram can
override an abstract subprogram coming from an interface
operation.

From-SVN: r241026

re PR ada/64057 (possible issue in the shared implementation of Ada.Strings.Unbounded)

2016-10-12  Eric Botcazou  <ebotcazou@adacore.com>

PR ada/64057
* exp_ch5.adb (Is_Non_Local_Array): Return true for every array
that is not a component or slice of an entity in the current
scope.

From-SVN: r241025

DWARF: space-optimize loc. descr. for integer literals on 32-bit targets

This enhances location description generation so that the generated
opcodes for integer literals are as space-efficient when HOST_WIDE_INT
is 64-bits large than when it's 32-bits large. In particular, this
reduces the size of the opcodes generated to produce big unsigned
literals using small literal integers instead.

gcc/

* dwarf2out.c (int_loc_descriptor): Generate opcodes for another
equivalent 32-bit constant (modulo 2**32) when that yields
smaller instructions.
(size_of_int_loc_descriptor): Update accordingly.

gcc/testsuite/

* gnat.dg/debug8.adb: New testcase.

From-SVN: r241024

DWARF: fix scoping for descriptions of local types

In Ada, it is possible to have nested subprograms in the following
configuration:

    procedure Parent is
       type T;
       [...]
       procedure Child (Value : T) is
       begin
          [...]
       end Child;
    begin
       [...]
    end Parent;

As we currently generate debugging information for Child first before
Parent, the debug info for T appears in global scope since the DIE for
Parent does not exist yet.

This patch makes sure that when we generate early debug info for a
nested function, we trigger generation for the parent function first.

gcc/

* dwarf2out.c (dwarf2out_early_global_decl): For nested
functions, call dwarf2out_decl on the parent function first.

gcc/testsuite/

* gnat.dg/debug9.adb: New testcase.

From-SVN: r241023

match.pd ((X /[ex] A) * A -> X): Remove unnecessary constraint on the conversion.

2016-10-12  Richard Biener  <rguenther@suse.de>

* match.pd ((X /[ex] A) * A -> X): Remove unnecessary constraint
on the conversion.

From-SVN: r241022

2016-10-12  Richard Biener  <rguenther@suse.de>

* tree-ssa-propagate.c
(substitute_and_fold_dom_walker::before_dom_children): Do not
ignore ASSERT_EXPRs but only preserve them.
* tree-vrp.c (remove_range_assertions): Deal with ASSERT_EXPRs
that have been propagated into.
(vrp_finalize): Enable DCE for substitute_and_fold.

* gcc.dg/tree-ssa/vrp35.c: Adjust.
* gcc.dg/tree-ssa/vrp36.c: Likewise.
* gcc.dg/tree-ssa/vrp46.c: Likewise.

From-SVN: r241021

re PR middle-end/77920 (186.crafty doesn't compile)

2016-10-12  Richard Biener  <rguenther@suse.de>

PR tree-optimization/77920
* tree-vrp.c (simplify_div_or_mod_using_ranges): Simplify.
(simplify_min_or_max_using_ranges): Pass in gsi and use it.
(simplify_abs_using_ranges): Likewise.
(simplify_conversion_using_ranges): Likewise.
(simplify_stmt_using_ranges): Adjust.

* gcc.dg/torture/pr77920.c: New testcase.

From-SVN: r241020

re PR tree-optimization/77929 (ICE: verify_gimple failed (error: non-trivial conversion at assignment))

PR tree-optimization/77929
* tree-ssa-reassoc.c (optimize_range_tests_var_bound): Handle
(*ops)[ranges[i].idx]->op != ranges[i].exp case.

* gcc.c-torture/compile/pr77929.c: New test.

From-SVN: r241019

re PR target/77934 (pattern for mtvsrdd needs to use b constraint not r)

2016-10-12  Aaron Sawdey  <acsawdey@linux.vnet.ibm.com>

PR target/77934
* config/rs6000/vmx.md (vsx_concat_<mode>): The mtvsrdd instruction
needs a base register for arg 1.

From-SVN: r241017

Daily bump.

From-SVN: r241016

gcc/
* common.opt (Wimplicit-fallthrough) Turn into alias to
-Wimplicit-fallthrough=3.  Remove EnabledBy.
(Wimplicit-fallthrough=): New option.
* gimplify.c (warn_implicit_fallthrough_r): Use
OPT_Wimplicit_fallthrough_ instead of OPT_Wimplicit_fallthrough.
* doc/invoke.texi (-Wimplicit-fallthrough): Document as alias
to -Wimplicit-fallthrough=3.
(-Wimplicit-fallthrough=): Document.
gcc/c-family/
* c.opt (Wextra): Add as C/C++/ObjC/ObjC++ option.
(Wimplicit-fallthrough=): Enable for these languages by -Wextra.
* c-opts.c (sanitize_cpp_opts): Initialize
cpp_opts->cpp_warn_implicit_fallthrough.
gcc/testsuite/
* c-c++-common/Wimplicit-fallthrough-25.c: New test.
* c-c++-common/Wimplicit-fallthrough-26.c: New test.
* c-c++-common/Wimplicit-fallthrough-27.c: New test.
* c-c++-common/Wimplicit-fallthrough-28.c: New test.
* c-c++-common/Wimplicit-fallthrough-29.c: New test.
* c-c++-common/Wimplicit-fallthrough-30.c: New test.
* c-c++-common/Wimplicit-fallthrough-31.c: New test.
* c-c++-common/Wimplicit-fallthrough-32.c: New test.
* c-c++-common/Wimplicit-fallthrough-33.c: New test.
libcpp/
* include/cpplib.h (struct cpp_options): Add
cpp_warn_implicit_fallthrough.
* init.c (cpp_create_reader): Initialize it to 0.
* lex.c (fallthrough_comment_p): Handle different
cpp_warn_implicit_fallthrough levels.  Whitespace fixes.

From-SVN: r241013

sparc.c (emit_scc_insn): Remove direct support for EQ and GEU in DImode if TARGET_SUBXC.

* config/sparc/sparc.c (emit_scc_insn): Remove direct support for EQ
and GEU in DImode if TARGET_SUBXC.
* config/sparc/sparc.md (seqdi<W:mode>_zero): Remove TARGET_SUBXC.
(seqdi<W:mode>_zero_subxc): Delete.
(neg_seqdi<W:mode>_zero): Remove TARGET_VIS3.
(neg_seqdi<W:mode>_zero_vis3): Delete.
(plus_seqdi<W:mode>_zero): Likewise.
(minus_seqdi<W:mode>_zero): Likewise.
(plus_plus_sltu<W:mode>): Accept only register.
(addx<W:mode>): Likewise.
(plus_sltu<W:mode>_vis3): Likewise.
(plus_plus_sltu<W:mode>_vis3): Likewise.
(neg_sgeu<W:mode>_vis3): Delete.
(minus_sgeu<W:mode>_vis3): Likewise.
(addxc<W:mode>): Accept only registers.
(neg_sltu<W:mode>_subxc): Write %%g0 instead of 0.
(minus_neg_sltu<W:mode>_subxc): Accept only register.
(neg_plus_sltu<W:mode>_subxc): Likewise.
(minus_sltu<W:mode>_subxc): Write %%g0 instead of 0.
(minus_minus_sltu<W:mode>_subxc): Accept only register.
(sgeu<W:mode>_insn_subxc): Delete.
(plus_sgeu<W:mode>_subxc): Likewise.
(subxc<W:mode>): Accept only register.
(scc splitter): Split always GEU again.

From-SVN: r241012

iresolve.c (is_trig_resolved, [...]): Formatting fixes.

* iresolve.c (is_trig_resolved, resolve_trig_call): Formatting fixes.
* simplify.c (simplify_trig_call, degrees_f, radians_f,
gfc_simplify_atrigd, gfc_simplify_cotan): Likewise.

From-SVN: r241010

re PR tree-optimization/77424 (Identical statements in if-else branches)

PR tree-optimization/77424
* tree-ssa-threadupdate.c (thread_through_all_blocks): Remove
dead conditionals.  Assert that all e->aux fields are NULL.

From-SVN: r241009

re PR fortran/77942 (ICE: Floating point exception, in gfc_simplify_cshift, at fortran/simplify.c:1845)

2016-10-11  Steven G. Kargl  <kargl@gcc.gnu.org>

PR fortran/77942
* simplify.c (gfc_simplify_cshift): Check for zero.

2016-10-11  Steven G. Kargl  <kargl@gcc.gnu.org>

PR fortran/77942
* gfortran.dg/pr77942.f90

From-SVN: r241008

system-linux-armeb.ads (Backend_Overflow_Checks): Change to True.

* system-linux-armeb.ads (Backend_Overflow_Checks): Change to True.
* system-linux-mips.ads (Backend_Overflow_Checks): Likewise.
* system-linux-mips64el.ads (Backend_Overflow_Checks): Likewise.
* system-linux-mipsel.ads (Backend_Overflow_Checks): Likewise.
* system-linux-sparcv9.ads (Backend_Overflow_Checks): Likewise.
* system-rtems.ads: Likewise.

From-SVN: r241003

Split print_rtx into subroutines

gcc/ChangeLog:
* print-rtl.c (print_rtx): Rename "i" to "idx".  Split out the
operand-printing "switch" statement into...
(print_rtx_operand_code_0): ...this new function, ...
(print_rtx_operand_code_e): ...this new function, ...
(print_rtx_operand_codes_E_and_V): ...this new function, ...
(print_rtx_operand_code_i): ...this new function, ...
(print_rtx_operand_code_r): ...this new function, ...
(print_rtx_operand_code_u): ...this new function, ...
(print_rtx_operand): ...and this new function.

From-SVN: r241002

Fix bad commit for -fdec-math.

2016-10-11  Fritz Reese  <fritzoreese@gmail.com>

Fix bad commit for -fdec-math.

* iresolve.c (get_radians, get_degrees): Fix sloppy commit.
* simplify.c (degrees_f, radians_f): Ditto.

From-SVN: r241001

simplify.c (radians_f): Fix mpdr_mod.

2016-10-11  Jerry DeLisle  <jvdelisle@gcc.gnu.org>

* simplify.c (radians_f): Fix mpdr_mod.
* ireolce.c (get_degrees): Declare tmp.

From-SVN: r241000

adaint.c: Include process.h, signal.h, dir.h and utime.h for DJGPP.

2016-10-11  Andris Pavenis  <andris.pavenis@iki.fi>

* adaint.c: Include process.h, signal.h, dir.h and utime.h for DJGPP.
ISALPHA: include <ctype.h> and define to isalpha for DJGPP when IN_RTS is defined.
(DIR_SEPARATOR) define to '\\' for DJGPP.
(__gnat_get_file_names_case_sensitive): return 0 for DJGPP unless
overriden in environment
(__gnat_is_absolute_path): Support MS-DOS style absolute paths for DJGPP.
(__gnat_portable_spawn): Use spewnvp for DJGPP.
(__gnat_portable_no_block_spawn): Use spawnvp for DJGPP.
(__gnat_portable_wait): Return 0 for DJGPP.

From-SVN: r240999

Makefile.in (LIBGNAT_TARGET_PAIRS): Define for DJGPP target

2016-10-11  Andris Pavenis  <andris.pavenis@iki.fi>

* gcc-interface/Makefile.in (LIBGNAT_TARGET_PAIRS): Define for DJGPP target
(EH_MECHANISM): Define to -gcc for DJGPP
* system-djgpp.ads: New file

From-SVN: r240998

ctrl_c.c: Do not use macro SA_RESTART for DJGPP.

2016-10-11  Andris Pavenis  <andris.pavenis@iki.fi>

* ctrl_c.c: Do not use macro SA_RESTART for DJGPP.
* gsocket.h: Do not support sockets for DJGPP.
* init.c (__gnat_install_handler): Implememt for DJGPP
* sysdep.c: Include <io.h> for DJGPP
(_setmode): Define to setmode for DJGPP
(__gnat_set_mode): Add implementation for DJGPP
(__gnat_localtime_tzoff): Use localtime_r for DJGPP
* terminals.c: Add DJGPP to list of unsupported platforms.
* env.c (__gnat_clearenv): use _gnat_unsetenv on all entries for DJGPP

From-SVN: r240997

alpha-passes.def: New file.

* config/alpha/alpha-passes.def: New file.
* config/alpha/t-alpha: New file.
* config/alpha/alpha-protos.h (gcc::context, rtl_opt_pass): Declare.
(make_pass_handle_trap_shadows): New prototype.
(make_pass_align_insns): Ditto.
* config/alpha/alpha.c (alpha_option_override): Don't register
passes here.
* config.gcc (alpha*-*-*) Add alpha/t-alpha to tmake_file.

From-SVN: r240996

re PR target/77924 (-mfloat128-type change broke AIX)

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

PR target/77924
* config/rs6000/rs6000.c (rs6000_init_builtins): Only create the
distinct __ibm128 IBM extended double type if long doubles are
128-bits and the default format for long double is IEEE 128-bit.

From-SVN: r240994

configure: redirect -fsplit-stack compilation to dev/null

    Avoid an error message in the middle of the configure output.

    Patch by Eric Botcazou.

    Reviewed-on: https://go-review.googlesource.com/30813

From-SVN: r240993

dwarf2out.c (DEBUG_STR_OFFSETS_SECTION): Remove conditional.

2016-10-11  Richard Biener  <rguenther@suse.de>

* dwarf2out.c (DEBUG_STR_OFFSETS_SECTION): Remove conditional.
(init_sections_and_labels): Use DEBUG_DWO_STR_OFFSETS_SECTION.
(verify_die): New function.
(dwarf2out_finish): Call it.
(output_line_info): Handle case of -gsplit-dwarf without
DWARF2_ASM_LINE_DEBUG_INFO.

From-SVN: r240992

re PR debug/77931 (PASS->FAIL: gdb.cp/namespace.exp: print ina)

2016-10-11  Richard Biener  <rguenther@suse.de>

PR debug/77931
* gimple-low.c (lower_gimple_bind): Handle arbitrary common
sub-chains of BLOCK_VARS and gimple_bind_vars.

From-SVN: r240991

Fix integer load reservation for -march=znver1
2016-10-11  Venkataramanan Kumar  <Venkataramanan.kumar@amd.com>

* config/i386/znver1.md : Fix imov/imovx load type reservations.

From-SVN: r240990

New flag -fdec-math for COTAN and degree trig intrinsics.

2016-10-11  Fritz Reese  <fritzoreese@gmail.com>

New flag -fdec-math for COTAN and degree trig intrinsics.

gcc/fortran/
* lang.opt: New flag -fdec-math.
* options.c (set_dec_flags): Enable with -fdec.
* invoke.texi, gfortran.texi, intrinsic.texi: Update documentation.
* intrinsics.c (add_functions, do_simplify): New intrinsics
with -fdec-math.
* gfortran.h (gfc_isym_id): New isym GFC_ISYM_COTAN.
* gfortran.h (gfc_resolve_atan2d, gfc_resolve_cotan,
gfc_resolve_trigd, gfc_resolve_atrigd): New prototypes.
* iresolve.c (resolve_trig_call, get_degrees, get_radians,
is_trig_resolved, gfc_resolve_cotan, gfc_resolve_trigd,
gfc_resolve_atrigd, gfc_resolve_atan2d): New functions.
* intrinsics.h (gfc_simplify_atan2d, gfc_simplify_atrigd,
gfc_simplify_cotan, gfc_simplify_trigd): New prototypes.
* simplify.c (simplify_trig_call, degrees_f, radians_f,
gfc_simplify_cotan, gfc_simplify_trigd, gfc_simplify_atrigd,
gfc_simplify_atan2d): New functions.

gcc/testsuite/gfortran.dg/
* dec_math.f90: New testsuite.

From-SVN: r240989

Minor tweaks

From-SVN: r240988

* MAINTAINERS: Update email address.

From-SVN: r240987

Fix gcc.dg/tree-ssa/pr59597.c failure for avr

Declare loop index variable j as a 32 bit int instead of assuming ints
are 32 bits. The smaller int size on the avr makes prior passes optimize
away the loop exit check (j < 10000000), as the constant is outside
the range of a 16 bit int.

gcc/testsuite/ChangeLog

2016-10-11  Senthil Kumar Selvaraj  <senthil_kumar.selvaraj@atmel.com>

* gcc.dg/tree-ssa/pr59597.c: Typedef  __INT32_TYPE__ to i32.
(main): Declare j as i32 instead of int.

From-SVN: r240986

exp_dbug.adb (Debug_Renaming_Declaration): Process underlying types.

* exp_dbug.adb (Debug_Renaming_Declaration): Process underlying types.
Emit GNAT encodings for object renamings involving record components
whose normalized bit offset is not null.
* uintp.h (UI_No_Uint): Declare.
* gcc-interface/gigi.h (can_materialize_object_renaming_p): New.
* gcc-interface/utils.c (can_materialize_object_renaming_p): New
function.
* gcc-interface/trans.c (gnat_to_gnu) <N_Object_Renaming_Declaration>:
In code generation mode, materialize all renamings as long as they need
debug info and we are not optimizing.

From-SVN: r240985

Optimise std::list::assign(initializer_list<T>) slightly

* include/bits/stl_list.h (assign(initializer_list<value_type>)): Call
_M_assign_dispatch directly with correct tag.

From-SVN: r240984

Simplify std::rethrow_if_nested definition

* doc/xml/manual/intro.xml: Document LWG 2484 status.
* libsupc++/nested_exception.h (_Throw_with_nested_impl)
(_Throw_with_nested_helper): Remove.
(__throw_with_nested_impl): New overloaded functions to implement
throw_with_nested logic.
(throw_with_nested): Call __throw_with_nested_impl.
(_Rethrow_if_nested_impl): Remove
(__rethrow_if_nested_impl): New overloaded functions to implement
rethrow_if_nested logic.
(rethrow_if_nested): Call __rethrow_if_nested_impl.
* testsuite/18_support/nested_exception/rethrow_if_nested.cc: Test
problem cases from LWG 2484.

From-SVN: r240983

Enable printers and xmethods for std::__cxx11::list

* python/libstdcxx/v6/printers.py (build_libstdcxx_dictionary):
Register printer for std::__cxx11::list.
* python/libstdcxx/v6/xmethods.py (ListMethodsMatcher.match): Match
std::__cxx11::list as well as std::list.

From-SVN: r240982

Remove redundant const qualifliers from string_view

* include/experimental/string_view (basic_string_view::_S_compare):
Remove redundant const from return type.
* include/std/string_view (basic_string_view:_S_compare): Likewise.

From-SVN: r240981

More noexcept for standard mutex types and locks

* include/bits/std_mutex.h (mutex::native_handle)
(lock_guard(mutex_type&, adopt_lock_t)
(unique_lock(mutex_type&, adopt_lock_t): Add noexcept.
* include/std/mutex (recursive_mutex, timed_mutex)
(recursive_timed_mutex): Likewise.

From-SVN: r240980

PR77710: fix triplet in builtin-sprintf-warn-4.c

2016-10-11  Thomas Preud'homme  <thomas.preudhomme@arm.com>

    gcc/testsuite/
    PR testsuite/PR77710
    * gcc.dg/tree-ssa/builtin-sprintf-warn-4.c: Use *-*-* as catch-all
    target triplet instead of *-*-*-*.

From-SVN: r240979

utils2.c (build_binary_op): Add a NO_FOLD argument.

* gcc-interface/utils2.c (build_binary_op): Add a NO_FOLD
argument.  Disable folding when true.
* gcc-interface/gigi.h (choices_to_gnu): Remove declaration.
(build_binary_op): Update signature and comment.
* gcc-interface/decl.c (choices_to_gnu): Make static.  Disable
folding for all calls to build_binary_op.

From-SVN: r240978

fe.h (Constant_Value): Declare.

* fe.h (Constant_Value): Declare.
* gcc-interface/decl.c (compile_time_known_address_p): Also consider
references to constants.

From-SVN: r240977

trans.c (gnat_to_gnu): Adjust comment.

* gcc-interface/trans.c (gnat_to_gnu) <N_Op_Add>: Adjust comment.
<N_Op_Minus>: Add comment and missing guard.
* gcc-interface/trans.c (build_binary_op_trapv): Use an explicit test.

From-SVN: r240976

utils.c (type_unsigned_for_rm): New predicate.

* gcc-interface/utils.c (type_unsigned_for_rm): New predicate.
(make_type_from_size): Use it.
(unchecked_convert): Likewise.  Do not skip the extension step if the
source type is not integral.

From-SVN: r240975

c-common.c (warning_candidate_p): Change the return type to bool and return true/false instead of 1/0.

* c-common.c (warning_candidate_p): Change the return type to bool
and return true/false instead of 1/0.
(vector_mode_valid_p): Likewise.

From-SVN: r240974

system-linux-ppc64.ads: Delete.

* system-linux-ppc64.ads: Delete.
* system-linux-ppc.ads: Make 32-bit/64-bit neutral.
* gcc-interface/Makefile.in (PowerPC/Linux): Simplify.

Co-Authored-By: Tristan Gingold <gingold@adacore.com>
From-SVN: r240973

Add missing file

From-SVN: r240972

sparc.opt (msubxc): New option.

* config/sparc/sparc.opt (msubxc): New option.
* doc/invoke.texi (SPARC options): Document it and tidy up.
* doc/tm.texi.in (Condition Codes): Adjust SPARC example.
* doc/tm.texi: Regenerate.
* config/sparc/sparc-modes.def (CC_NOOV): Rename into...
(CCNZ): ...this.
(CCX_NOOV): Rename into...
(CCXNZ): ...this.
(CCC): New.
(CCXC): Likewise.
* config/sparc/predicates.m (fcc_register_operand): Simplify.
(fcc0_register_operand): Likewise.
(icc_register_operand): New.
(icc_or_fcc_register_operand): Simplify.
(nz_comparison_operator): New.
(c_comparison_operator): Likewise.
(noov_compare_operator): Rename into...
(icc_comparison_operator): ...this.  Use above predicates.
(noov_compare64_operator): Rename into...
(v9_comparison_operator): ...this and tidy up.
(fcc_comparison_operator): New.
(icc_or_fcc_comparison_operator): Likewise.
(v9_register_compare_operator): Rename info...
(v9_register_comparison_operator): ...this.
* config/sparc/sparc.c (TARGET_FIXED_CONDITION_CODE_REGS): Define.
(sparc_option_override): Remove redundant VIS masks and add MASK_SUBXC
for Niagara-7.
(sparc_fixed_condition_code_regs): New function.
(select_cc_mode): Remove ATTRIBUTE_UNUSED.  Adjust for CCNZ/CCXNZ
renaming and add support for CCC/CCXC.
(output_cbranch): Likewise.
(sparc_print_operand): Likewise.
(gen_v9_scc): Remove obsolete assertion.
(emit_scc_insn): Emit RTL directly for EQ and NE.  Add direct support
for EQ in DImode if TARGET_SUBXC.  Remove test on TARGET_VIS3 for GEU.
(output_cbcond): Remove bogus handling of CC modes.
(sparc_register_move_cost): Return 100 for NO_REGS.
* config/sparc/sparc.md (W): New mode iterator.
(length): Adjust for noov_compare64_operator renaming.
(cmpsi_sne): New instruction.
(cmpdi_sne): Likewise.
(seqdi_special): Delete.
(seqdi_special): Likewise.
(snesi<P:mode>_special): Likewise.
(snedi_special): Likewise.
(snedi_special_vis3): Likewise.
(snesi patterns): Use W iterator.
(snedi patterns): Likewise.  Add TARGET_SUBXC patterns.
(sltu patterns): Likewise.
(sgeu patterns): Likewise.
(scc splitter): Do not split GEU in DImode if TARGET_SUBXC.
(normal_branch): Use icc_comparison_operator predicate.
(inverted_branch): Likewise.
(cbcond_sp32): Use comparison_operator predicate.
(cbcond_sp64): Likewise.
(normal_int_branch_sp64): Adjust for renaming
(inverted_int_branch_sp64): Likewise.
(mov<I:mode>_cc_reg_sp64): Likewise.
(movsf_cc_reg_sp6): Likewise.
(movdf_cc_reg_sp64): Likewise.
(movtf_cc_reg_hq_sp64): Likewise.
(movtf_cc_reg_sp64): Likewise.
(mov<I:mode>_cc_v9): Use icc_or_fcc_comparison_operator predicate.
(movsf_cc_v9): Likewise.
(movdf_cc_v9): Likewise.
(movtf_cc_hq_v9): Likewise.
(movtf_cc_v9): Likewise.
(adddi3): Call gen_adddi3_sp32.
(adddi3_insn_sp32): Rename to...
(adddi3_sp32): ...this.  Accept only register_operand as operand #1
and use CCCmode for the carry.
(addx_extend_sp32): Use CCCmode for the carry.
(addx_extend_sp64): Delete.
(adddi3_extend_sp32): Use CCCmode for the carry.
(cmp_plus patterns): Use CCNZ/CCXNZ mode and add C variants.
(subdi3): Call gen_subdi3_sp32.
(subdi3_insn_sp32): Rename to...
(subdi3_sp32): ...this and use CCmode for the carry.
(subx_extend_sp32): Use CCCmode for the carry.
(subx_extend_sp64): Delete.
(subdi3_extend_sp32): Use CCmode for the carry.
(cmp_minus patterns): Use CCNZ/CCXNZ mode and add C variants.
(negdi3): Call gen_negdi3_sp32.
(negdi3_sp32): Use CCCmode for the carry.
(cmp_neg patterns): Use CCNZ/CCXNZ mode and add C variants.
(cmp_nz_ashift_1): Use CCNZ mode.
(cmp_nz_set_ashift_1): Likewise.
(ctrapsi4): Use comparison_operator predicate.
(ctrapdi4): Likewise.
(trapsi_insn): Use icc_comparison_operator predicate.
(trapdi_insn): Likewise.
(edge8 patterns): Use CCNZmode.
(edge16 patterns): Likewise.
(edge32 patterns): Likewise.

From-SVN: r240971

visium-modes.def (CC_NOOV): Rename into...

* config/visium/visium-modes.def (CC_NOOV): Rename into...
(CCNZ): ...this.
(CC_BTST): Rename into...
(CCC): ...this.
* config/visium/predicates.md (real_add_operand): New.
(visium_btst_operator): Rename into...
(visium_equality_comparison_operator): ...this.
(visium_noov_operator): Rename into...
(visium_nz_comparison_operator): ...this.
(visium_c_comparison_operator): New.
(visium_branch_operator): Adjust and deal with all CC modes.
* config/visium/visium.c (visium_adjust_cost): Adjust.
(visium_split_double_add): Use the *_set_carry patterns.
(visium_select_cc_mode): Add support for CCC mode and adjust.
(output_cbranch): Adjust and use the carry-based operators for
floating-point comparisons.
* config/visium/visium.md (flags_subst_arith): Adjust.
(addsi3_insn_set_carry): New instruction.
(subsi3_insn_set_carry): Likewise.
(negsi2_insn_set_carry): Likewise.
(btst): Adjust.
(cmp<mode>_sne): Likewise.
(cbranch<mode>4): Use ordered_comparison_operator.
(cbranch<mode>4_insn): Likewise.
(cbranchsi4_btst_insn): Adjust.

From-SVN: r240969

Remove RECORD_TYPE special-casing in std_canonical_va_list_type

2016-10-11  Tom de Vries  <tom@codesourcery.com>

PR middle-end/77558
* builtins.c (std_canonical_va_list_type): Remove RECORD_TYPE
special-casing.

From-SVN: r240968

tree.h (build_complex_type): Add second parameter with default.

* tree.h (build_complex_type): Add second parameter with default.
* tree.c (build_complex_type): Add NAMED second parameter and adjust
recursive call.  Create a TYPE_DECL only if NAMED is true.
(build_common_tree_nodes): Pass true in calls to build_complex_type.

From-SVN: r240967

New avr-passes.def to register AVR specific passes.

* config/avr/avr-passes.def: New file.
* config/avr/t-avr (PASSES_EXTRA): Add avr-passes.def.
* config/avr/avr-protos.h (gcc::context, rtl_opt_pass): Declare.
(make_avr_pass_recompute_note): New proto.
* config/avr/avr.c (make_avr_pass_recompute_notes): New function.
(avr_pass_recompute_notes): Use anonymous namespace.
(avr_register_passes): Remove function...
(avr_option_override): ...and its call.

From-SVN: r240966

[MIPS] Disable -mbranch-likely for -Os when targetting generic arch

gcc/
* config/mips/mips-cpus.def: Replace PTF_AVOID_BRANCHLIKELY with
PTF_AVOID_BRANCHLIKELY_ALWAYS for generic architecture and with
PTF_AVOID_BRANCHLIKELY_SPEED for others.
(mips2, mips3, mips4): Add PTF_AVOID_BRANCHLIKELY_SIZE to tune flags.
* config/mips/mips.c (mips_option_override): Enable the branch likely
depending on the tune flags and optimization level.
* config/mips/mips.h (PTF_AVOID_BRANCHLIKELY): Remove.
(PTF_AVOID_BRANCHLIKELY_SPEED): Define.
(PTF_AVOID_BRANCHLIKELY_SIZE): Likewise.
(PTF_AVOID_BRANCHLIKELY_ALWAYS): Likewise.

From-SVN: r240965

lto-streamer-out.c (collect_block_tree_leafs): New helper.

2016-10-11  Richard Biener  <rguenther@suse.de>

* lto-streamer-out.c (collect_block_tree_leafs): New helper.
(output_function): Properly stream the whole block tree.
* lto-streamer-in.c (input_function): Likewise.

From-SVN: r240964

Makefile.in (C_COMMON_OBJS): Add c-family/c-warn.o.

* Makefile.in (C_COMMON_OBJS): Add c-family/c-warn.o.

* c-common.c (fold_for_warn): No longer static.
(bool_promoted_to_int_p): Likewise.
(c_common_get_narrower): Likewise.
(constant_expression_warning): Move to c-warn.c.
(constant_expression_error): Likewise.
(overflow_warning): Likewise.
(warn_logical_operator): Likewise.
(find_array_ref_with_const_idx_r): Likewise.
(warn_tautological_cmp): Likewise.
(expr_has_boolean_operands_p): Likewise.
(warn_logical_not_parentheses): Likewise.
(warn_if_unused_value): Likewise.
(strict_aliasing_warning): Likewise.
(sizeof_pointer_memaccess_warning): Likewise.
(check_main_parameter_types): Likewise.
(conversion_warning): Likewise.
(warnings_for_convert_and_check): Likewise.
(match_case_to_enum_1): Likewise.
(match_case_to_enum): Likewise.
(c_do_switch_warnings): Likewise.
(warn_for_omitted_condop): Likewise.
(readonly_error): Likewise.
(lvalue_error): Likewise.
(invalid_indirection_error): Likewise.
(warn_array_subscript_with_type_char): Likewise.
(warn_about_parentheses): Likewise.
(warn_for_unused_label): Likewise.
(warn_for_div_by_zero): Likewise.
(warn_for_memset): Likewise.
(warn_for_sign_compare): Likewise.
(do_warn_double_promotion): Likewise.
(do_warn_unused_parameter): Likewise.
(record_locally_defined_typedef): Likewise.
(maybe_record_typedef_use): Likewise.
(maybe_warn_unused_local_typedefs): Likewise.
(maybe_warn_bool_compare): Likewise.
(maybe_warn_shift_overflow): Likewise.
(warn_duplicated_cond_add_or_warn): Likewise.
(diagnose_mismatched_attributes): Likewise.
* c-common.h: Move the declarations from c-warn.c to its own section.
* c-warn.c: New file.

From-SVN: r240963

Daily bump.

From-SVN: r240962

compiler: move Backend/Linemap creation out of front end.

    Push the calls to create Backend and Linemap object out
    of the front end into the back end, and instead pass pointers to
    these objects in the go_create_gogo_args struct. This allows
    for more flexibility in the interfaces used to create the
    objects.

    Reviewed-on: https://go-review.googlesource.com/30698

* go-gcc.h: New file.
* go-c.h (struct go_create_gogo_args): Add backend and linemap
fields.
* go-lang.c: Include "go-gcc.h".
(go_langhook_init): Set linemap and backend fields of args.
* go-gcc.cc: Include "go-gcc.h".
* go-linemap.cc: Include "go-gcc.h".

From-SVN: r240959

Accidentally failed to commit these earlier, as part of:

    Update the compiler to use the new names.  Add calls to printlock and
    printunlock around print statements.  Move expression evaluation before
    the call to printlock.  Update g's writebuf field to a slice, and adjust
    C code accordingly.

    Reviewed-on: https://go-review.googlesource.com/30717

From-SVN: r240958

Infer and push new value ranges for x in y < x.

gcc/ChangeLog:

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

* tree-vrp.c (evrp_dom_walker::try_add_new_range): New.
(evrp_dom_walker::before_dom_children): Infer and push new value
ranges for x in y < x.

From-SVN: r240957

runtime: copy print/println support from Go 1.7

    Update the compiler to use the new names.  Add calls to printlock and
    printunlock around print statements.  Move expression evaluation before
    the call to printlock.  Update g's writebuf field to a slice, and adjust
    C code accordingly.

    Reviewed-on: https://go-review.googlesource.com/30717

From-SVN: r240956

Always support float128 on ia64 (PR target/77586).

Bug 77586, and previously
<https://gcc.gnu.org/ml/gcc-bugs/2016-08/msg03233.html>, reports
ia64-elf failing to build because of float128_type_node being NULL,
but being used by the back end for __float128.

The global float128_type_node is only available conditionally, if
target hooks indicate TFmode is not only available as a scalar mode
and of the right format, but also supported in libgcc.  The back-end
support, however, expects the type always to be available for
__float128 even if the libgcc support is missing.

Although a target-specific node could be restored in the case where
libgcc support is missing, it seems better to address the missing
libgcc support.  Thus, this patch enables TFmode soft-fp in libgcc
globally for all ia64 targets.  Support for XFmode in libgcc (that is,
for libgcc2.c XFmode functions, not soft-fp) is also enabled for all
ia64 targets so that ia64 no longer needs to define the
TARGET_LIBGCC_FLOATING_MODE_SUPPORTED_P hook.

I've confirmed that ia64-elf builds cc1 with this patch and it passes
-fself-test.  I have not otherwise tested the patch.  It's plausible
that ia64-elf and ia64-freebsd might work as-is, but ia64-vms probably
needs further changes, by someone familiar with VMS shared libraries,
to implement an equivalent of ia64/t-softfp-compat in that case
(avoiding conflicts between __divtf3 from soft-fp and the old alias
for __divxf3).

PR target/77586
gcc:
* config/ia64/ia64.c (ia64_libgcc_floating_mode_supported_p)
(TARGET_LIBGCC_FLOATING_MODE_SUPPORTED_P): Remove.
* config/ia64/elf.h (IA64_NO_LIBGCC_TFMODE): Likewise.
* config/ia64/freebsd.h (IA64_NO_LIBGCC_TFMODE): Likewise.
* config/ia64/vms.h (IA64_NO_LIBGCC_XFMODE)
(IA64_NO_LIBGCC_TFMODE): Likewise.

libgcc:
* config.host (ia64*-*-elf*, ia64*-*-freebsd*, ia64-hp-*vms*): Use
soft-fp.

From-SVN: r240955