re PR fortran/86110 (ICE in gfc_resolve_character_array_constructor, at fortran/array.c:2044)

2018-06-13  Steven G. Kargl  <kargl@gcc.gnu.org>

PR fortran/86110
* array.c (gfc_resolve_character_array_constructor): Avoid NULL
pointer dereference.

2018-06-13  Steven G. Kargl  <kargl@gcc.gnu.org>

PR fortran/86110
* gfortran.dg/pr86110.f90: New test.

From-SVN: r261561

compiler: include global variable preinit blocks in ast dumps

    Dump out the blocks corresponding to variable pre-inits when
    -fgo-dump-ast is in effect. Each preinit block is prefixed with a
    comment indicating the variable it is initializing.

    Reviewed-on: https://go-review.googlesource.com/118636

From-SVN: r261555

PR libstdc++/86127 avoid unnecessary allocator conversions

There is no need to use an allocator of the correct value_type when
calling allocator_traits::construct and allocator_traits::destroy. The
existing node allocator can be used, instead of constructing a new
allocator object every time.

There's also no benefit to using __gnu_cxx::__alloc_traits instead of
std::allocator_traits to get the pointer and const_pointer types.
std::forward_list is only available for C++11 and later, when
std::allocator_traits is available too.

PR libstdc++/86127
* include/bits/forward_list.h (_Fwd_list_base::_Tp_alloc_type): Remove
unused typedef.
(_Fwd_list_base::_Node_alloc_traits): Use allocator_traits instead of
__gnu_cxx::__alloc_traits.
(_Fwd_list_base::_M_create_node, _Fwd_list_base::_M_erase_after):
Use node allocator to create and destroy elements.
(forward_list::_Tp_alloc_type): Remove unused typedef.
(forward_list::_Alloc_traits): Use allocator_traits instead of
__gnu_cxx::__alloc_traits.

From-SVN: r261554

tree-vect-patterns.c (vect_recog_vector_vector_shift_pattern): Properly set vector type of the intermediate stmt.

2018-06-13  Richard Biener  <rguenther@suse.de>

* tree-vect-patterns.c (vect_recog_vector_vector_shift_pattern):
Properly set vector type of the intermediate stmt.
* tree-vect-stmts.c (vectorizable_operation): The destination
var always has vectype_out type.

From-SVN: r261553

rl78.c (move_elim_pass): Use TDF_NONE rather than integer 0 for argument to print_rtl_with_bb.

        * config/rl78/rl78.c (move_elim_pass): Use TDF_NONE rather than
        integer 0 for argument to print_rtl_with_bb.
        (rl78_reorg): Likewise.

From-SVN: r261552

re PR fortran/85703 ([openacc] ICE in resolve_fntype, at fortran/resolve.c:16313)

PR fortran/85703

gcc/fortran/
* parse.c (decode_oacc_directive): Set gfc_matching_function
to false.
(decode_omp_directive): Likewise.

gcc/testsuite/
* gfortran.dg/goacc/pr85703.f90: New test.
* gfortran.dg/gomp/pr85703.f90: New test.

From-SVN: r261551

re PR fortran/85702 ([openacc] ICE in gfc_format_decoder, at fortran/error.c:943)

PR fortran/85702

gcc/fortran/
* openmp.c (gfc_match_oacc_wait): Use %C to report error location.

gcc/testsuite/
* gfortran.dg/goacc/pr85702.f90: New test.

From-SVN: r261550

libgo: update to Go 1.10.3 release

    Reviewed-on: https://go-review.googlesource.com/118495

From-SVN: r261549

emit-rtl.c: more typesafety

This patch converts various rtx to rtx_insn * (or rtx_code_label *).
It also convert the various "_loc" params from int to location_t

gcc/ChangeLog:
* config/arc/arc.c (hwloop_optimize): Strengthen local "end_label"
from rtx to rtx_insn *.
* config/bfin/bfin.c (hwloop_optimize): Likewise for local
"label".
(add_sched_insns_for_speculation): Likewise for local "target",
converting usage of JUMP_LABEL to JUMP_LABEL_AS_INSN.
* config/c6x/c6x.c (reorg_split_calls): Strengthen param "call_labels"
from rtx_insn ** to rtx_code_label **.
(reorg_emit_nops): Likewise.
(c6x_reorg): Likewise for local "call_labels".
* config/sh/sh-protos.h (get_dest_uid): Strengthen 1st param from
rtx to rtx_insn *.
* config/sh/sh.c (dump_table): Strengthen local "lab" from rtx to
rtx_code_label *, adding safe_as_a <rtx_code_label *> casts to
the loops over LABEL_REFS.
(fixup_addr_diff_vecs): Add as_a <rtx_insn *> to usage of
braf_label.
(barrier_align): Convert usage of JUMP_LABEL to JUMP_LABEL_AS_INSN.
(get_dest_uid): Strengthen param "label" from rtx to rtx_insn *.
(split_branches): Strengthen local "olabel" from rtx to
rtx_insn *, adding a safe_as_a cast.
* emit-rtl.c (next_real_insn): Strengthen param from "rtx"
to "rtx_insn *".
(add_insn_after): Likewise for first two params.
(add_insn_before): Likewise.
(remove_insn): Likewise for param.
(emit_pattern_before_noloc): Likewise for second and third params.
(emit_jump_insn_before_noloc): Convert NULL_RTX to NULL.
(emit_call_insn_before_noloc): Likewise.
(emit_debug_insn_before_noloc): Strengthen "before" param from "rtx"
to "rtx_insn *".
(emit_barrier_before): Likewise.
(emit_label_before): Strengthen "label" param from "rtx" to
"rtx_code_label *".  Strengthen "before" param from "rtx" to
"rtx_insn *".
(emit_insn_after_1): Strengthen "after" param from "rtx" to
"rtx_insn *".
(emit_pattern_after_noloc): Likewise.
(emit_insn_after_noloc): Likewise.
(emit_jump_insn_after_noloc): Likewise.
(emit_call_insn_after_noloc): Likewise.
(emit_debug_insn_after_noloc): Likewise.
(emit_barrier_after): Likewise.
(emit_label_after): Likewise for both params.
(emit_pattern_after_setloc): Likewise for "after" param.  Convert
"loc" param from "int" to "location_t".
(emit_insn_after_setloc): Likewise.
(emit_jump_insn_after_setloc): Likewise.
(emit_call_insn_after_setloc): Likewise.
(emit_debug_insn_after_setloc): Likewise.
(emit_pattern_before_setloc): Likewise for "before" param.  Convert
"loc" param from "int" to "location_t".
(emit_pattern_before): Convert NULL_RTX to NULL.
(emit_insn_before_setloc): Convert "loc" param from "int" to
"location_t".
(emit_jump_insn_before_setloc): Likewise.
(emit_call_insn_before_setloc): Likewise.
(emit_debug_insn_before_setloc): Strengthen "before" param from rtx to
rtx_insn *.  Convert "loc" param from "int" to "location_t".
* rtl.h (emit_insn_before_setloc, emit_jump_insn_before_setloc,
emit_call_insn_before_setloc, emit_debug_insn_before_setloc):
Convert 3rd param from "int" to "location_t".
(emit_barrier_before, emit_barrier_after, next_real_insn):
Strengthen param from rtx to rtx_insn *.
(emit_label_before): Strengthen 1st param from "rtx" to
"rtx_code_label *".  Strengthen 2nd param from "rtx" to
"rtx_insn *".
(emit_insn_after_noloc, emit_jump_insn_after_noloc,
emit_call_insn_after_noloc, emit_debug_insn_after_noloc):
Strengthen 2nd param from "rtx" to "rtx_insn *".
(emit_insn_after_setloc, emit_jump_insn_after_setloc)
emit_call_insn_after_setloc, emit_debug_insn_after_setloc):
Likewise. Convert 3rd param from "int" to "location_t".
(emit_label_after): Strengthen 1st param from "rtx" to
"rtx_code_label *".
(next_real_insn, remove_insn): Strengthen param from "rtx" to
"rtx_insn *".
(add_insn_before, add_insn_after): Strengthen 1st and 2nd params
from "rtx" to "rtx_insn *".

From-SVN: r261547

cgraph.c (cgraph_node::get_untransformed_body): Dump function bodies streamed in with -Q.

* cgraph.c (cgraph_node::get_untransformed_body): Dump function
bodies streamed in with -Q.
* dumpfile.c (dump_files): Add lto-stream-out dump file.
* dumpfile.h (tree_dump_index): Add lto_stream_out.
* gimple-streamer-out.c: Include gimple-pretty-print.h
(output_bb): Dump stmts streamed.
* lto-section-out.c: Include print-tree.h
(lto_begin_section): Dump sections created.
(lto_output_decl_index): Dump decl encoded.
* lto-streamer-out.c: Include print-tree.h
(create_output_block): Dump output block created.
(DFS::DFS_write_tree_body): Dump DFS SCCs streamed.
(output_function): Dump function output.
(output_constructor): Dump constructor streamed.
(write_global_stream): Output indexes encoded.
(produce_asm_for_decls): Dump streams encoded.
* lto-streamer.c (streamer_dump_file): New global var.
* lto-streamer.h (streamer_dump_file): Declare.
* passes.c (ipa_write_summaries): Initialize streamer dump.
* varpool.c (varpool_node::get_constructor): Dump constructors streamed
in.

From-SVN: r261546

re PR target/86048 (.seh_savexmm offset is negative error when compiling libpng)

PR target/86048
* config/i386/winnt.c (i386_pe_seh_cold_init): Do not emit negative
offsets for register save directives.  Emit a second batch of save
directives, if need be, when the function accesses prior frames.

From-SVN: r261544

[ARC] Update fma expansions.

Accept at most a single constant for fma patterns.

gcc/
2018-03-21  Claudiu Zissulescu  <claziss@synopsys.com>

* config/arc/fpu.md (fmasf4): Force operand to register.
(fnmasf4): Likewise.

gcc/testsuite
2018-03-21  Claudiu Zissulescu  <claziss@synopsys.com>

* gcc.target/arc/fma-1.c: New test.

From-SVN: r261543

[ARC] Reimplement return padding operation for ARC700.

For ARC700, adding padding if necessary to avoid a mispredict.  A
return could happen immediately after the function start.  A
call/return and return/return must be 6 bytes apart to avoid
mispredict.

The old implementation was doing this operation very late in the
compilation process, and the additional nop instructions and/or
forcing some other instruction to take their long form was not taken
into account when generating brcc instructions. Thus, wrong code could
be generated.

gcc/
2017-03-24  Claudiu Zissulescu  <claziss@synopsys.com>

* config/arc/arc-protos.h (arc_pad_return): Remove.
* config/arc/arc.c (machine_function): Remove force_short_suffix
and size_reason.
(arc_print_operand): Adjust printing of '&'.
(arc_verify_short): Remove conditional printing of short suffix.
(arc_final_prescan_insn): Remove reference to size_reason.
(pad_return): New function.
(arc_reorg): Call pad_return.
(arc_pad_return): Remove.
(arc_init_machine_status): Remove reference to force_short_suffix.
* config/arc/arc.md (vunspec): Add VUNSPEC_ARC_BLOCKAGE.
(attr length): When attribute iscompact is true force to 2
regardless; in the case of maybe check if we want to force the
instruction to have 4 bytes length.
(nopv): Change it to generate 4 byte long nop as well.
(blockage): New pattern.
(simple_return): Remove call to arc_pad_return.
(p_return_i): Likewise.

gcc/testsuite/
2017-03-24  Claudiu Zissulescu  <claziss@synopsys.com>

* gcc.target/arc/pr9001107555.c: New file.

From-SVN: r261542

[ARC] Define LINK_GCC_C_SEQUENCE_SPEC.

If no specs file is provided, default to nosys library.

gcc/
2017-05-03  Claudiu Zissulescu  <claziss@synopsys.com>

* config/arc/elf.h (LINK_GCC_C_SEQUENCE_SPEC): Define.

From-SVN: r261541

[ARC] SYNC instruction is valid on all ARC cores.

gcc/
2017-05-03  Claudiu Zissulescu  <claziss@synopsys.com>

* config/arc/builtins.def (SYNC): SYNC instruction is valid on all
ARC cores.

From-SVN: r261540

[ARC] Allow EX instruction unconditionally

gcc/
2017-05-02  Claudiu Zissulescu  <claziss@synopsys.com>

* config/arc/arc.c (atomic_exchangesi): EX instruction is default
for ARC700 and ARCv2.

From-SVN: r261539

re PR tree-optimization/86076 (ICE: verify_gimple failed (error: location references block not in block tree))

2018-06-13  Chenghua Xu <paul.hua.gm@gmail.com>

PR target/86076
* config/mips/loongson.md (vec_setv4hi): Gen_lowpart for
operands[2] instead of operands[1].

From-SVN: r261538

debug.h (__glibcxx_requires_can_increment_range): New.

2018-06-13  François Dumont  <fdumont@gcc.gnu.org>

* include/debug/debug.h (__glibcxx_requires_can_increment_range): New.
(__glibcxx_requires_can_decrement_range): New.

From-SVN: r261537

PR c++/86098 - ICE with template placeholder for TTP.

* typeck.c (structural_comptypes) [TEMPLATE_TYPE_PARM]: Check
CLASS_PLACEHOLDER_TEMPLATE.

From-SVN: r261536

Daily bump.

From-SVN: r261535

Tighten LRA test for reloading the inner reg of a paradoxical subreg

This patch fixes an LRA cycling problem on the attached testcase.
The original insn was:

(insn 74 72 76 8 (set (reg:V2DI 287 [ _166 ])
        (subreg:V2DI (reg/v/f:DI 112 [ d ]) 0)) 1060 {*aarch64_simd_movv2di}
     (nil))

which IRA converted to:

(insn 74 72 580 8 (set (reg:V2DI 287 [ _166 ])
        (subreg:V2DI (reg/v/f:DI 517 [orig:112 d ] [112]) 0)) 1060 {*aarch64_simd_movv2di}
     (nil))

after creating loop allocnos.  It happens that the ALLOCNO_WMODEs for
both 112 and 517 were not set to V2DI due to another bug that I'll post
a separate patch for, but we nevertheless got a valid allocation of
register 1.

LRA's first try at constraining the instruction gave:

         Choosing alt 5 in insn 74:  (0) ?w  (1) r {*aarch64_simd_movv2di}

at which point all was good.  But LRA later decided it needed
to spill r517:

    Spill r517 after risky transformations

so the next constraint attempt gave:

         Choosing alt 0 in insn 74:  (0) =w  (1) m {*aarch64_simd_movv2di}

which was still good.  Then during inheritance we had:

      Creating newreg=672 from oldreg=517, assigning class GENERAL_REGS to inheritance r672
    Original reg change 517->672 (bb8):
   74: r287:V2DI=r672:DI#0
    Add inheritance<-original before:
  939: r672:DI=r517:DI

    Inheritance reuse change 517->672 (bb8):
  620: r572:DI=r672:DI
      REG_DEAD r672:DI

    Use smallest class of POINTER_REGS and GENERAL_REGS
      Creating newreg=673 from oldreg=517, assigning class POINTER_REGS to inheritance r673
    Original reg change 517->673 (bb8):
  936: r669:DI=r673:DI
    Add inheritance<-original before:
  940: r673:DI=r517:DI

("Use smallest class of POINTER_REGS and GENERAL_REGS" ought to
give GENERAL_REGS.  That might be a missed optimisation, and probably
due to both classes having the same number of allocatable registers.
I'll look at that as a follow-on.)

Thus LRA created two inheritance registers for r517, one (r673)
that included the unallocatable x31 and another (r672) that didn't.
The r672 references included the paradoxical subreg in insn 74 but the
r673 ones didn't.  LRA then allocated x30 to r673, which was a valid
choice.

Later LRA decided to "undo" the inheritance for insn 620, but because
of the double inheritance, it got confused as to what the original
situation was, and made insn 74 use the other inheritance register
instead of r517:

********** Undoing inheritance #2: **********

Inherit 11 out of 12 (91.67%)
   Insn after restoring regs:
  620: r572:DI=r517:DI
      REG_DEAD r517:DI
    Change reload insn:
   74: r287:V2DI=r673:DI#0       <-------------------
   Insn after restoring regs:
  939: r517:DI=r673:DI
      REG_DEAD r673:DI

This might be a bug in itself: we should probably look through sets
of other inheritance pseudos to find the "real" origin.

Either way, at this point we had a situation in which r673 was used in an
insn whose subreg was larger than the biggest_mode that r673 had when it
was allocated.  While x30 was valid for the original biggest_mode, it
wasn't valid for this subreg use.

The next attempt to constrain insn 74 was:

        Choosing alt 5 in insn 74:  (0) ?w  (1) r {*aarch64_simd_movv2di}
      Creating newreg=684, assigning class GENERAL_REGS to r684
   74: r287:V2DI=r684:V2DI
    Inserting insn reload before:
  951: r684:V2DI=r673:DI#0

where LRA reloaded the SUBREG rather than the SUBREG_REG.  And it
then cycled trying the same thing when reloading the reload (and the
reload of the reload, etc.).

What it should be doing here is reloading the SUBREG_REG instead.
There's already code to cope with this case when the paradoxical
subreg falls outside the class (which isn't true here, since r673
is POINTER_REGS and POINTER_REGS includes x31).  But I think we
should also test whether LRA is entitled to allocate the spanned
registers.  Not doing that seems like a bug regardless of the above
missed optimisation and the mix-up undoing inheritance.

2018-05-30  Richard Sandiford  <richard.sandiford@linaro.org>

gcc/
* lra-constraints.c (simplify_operand_subreg): In the paradoxical
case, check whether the outer register overlaps an unallocatable
register, not just whether it fits the required class.

gcc/testsuite/
* g++.dg/torture/aarch64-vect-init-1.C: New test.

From-SVN: r261531

Use poly_int rtx accessors instead of hwi accessors

This patch generalises various places that used hwi rtx accessors so
that they can handle poly_ints instead.  In many cases these changes
are by inspection rather than because something had shown them to be
necessary.

2018-06-12  Richard Sandiford  <richard.sandiford@linaro.org>

gcc/
* poly-int.h (can_div_trunc_p): Add new overload in which all values
are poly_ints.
* alias.c (get_addr): Extend CONST_INT handling to poly_int_rtx_p.
(memrefs_conflict_p): Likewise.
(init_alias_analysis): Likewise.
* cfgexpand.c (expand_debug_expr): Likewise.
* combine.c (combine_simplify_rtx, force_int_to_mode): Likewise.
* cse.c (fold_rtx): Likewise.
* explow.c (adjust_stack, anti_adjust_stack): Likewise.
* expr.c (emit_block_move_hints): Likewise.
(clear_storage_hints, push_block, emit_push_insn): Likewise.
(store_expr_with_bounds, reduce_to_bit_field_precision): Likewise.
(emit_group_load_1): Use rtx_to_poly_int64 for group offsets.
(emit_group_store): Likewise.
(find_args_size_adjust): Use strip_offset.  Use rtx_to_poly_int64
to read the PRE/POST_MODIFY increment.
* calls.c (store_one_arg): Use strip_offset.
* rtlanal.c (rtx_addr_can_trap_p_1): Extend CONST_INT handling to
poly_int_rtx_p.
(set_noop_p): Use rtx_to_poly_int64 for the elements selected
by a VEC_SELECT.
* simplify-rtx.c (avoid_constant_pool_reference): Use strip_offset.
(simplify_binary_operation_1): Extend CONST_INT handling to
poly_int_rtx_p.
* var-tracking.c (compute_cfa_pointer): Take a poly_int64 rather
than a HOST_WIDE_INT.
(hard_frame_pointer_adjustment): Change from HOST_WIDE_INT to
poly_int64.
(adjust_mems, add_stores): Update accodingly.
(vt_canonicalize_addr): Track polynomial offsets.
(emit_note_insn_var_location): Likewise.
(vt_add_function_parameter): Likewise.
(vt_initialize): Likewise.

From-SVN: r261530

config.gcc (alpha*-*-freebsd*): Remove.

* config.gcc (alpha*-*-freebsd*): Remove.
* config/alpha/freebsd.h: Remove.

contrib/

* config-list.mk (LIST): Remove alpha-freebsd6.

From-SVN: r261529

decl2.c (coerce_new_type, [...]): Add location_t parameter and adjust error_at calls.

/cp
2018-06-12  Paolo Carlini  <paolo.carlini@oracle.com>

* decl2.c (coerce_new_type, coerce_delete_type): Add location_t
parameter and adjust error_at calls.
* decl.c (grok_op_properties): Adjust calls.
* cp-tree.h (oerce_new_type, coerce_delete_type): Adjust decls.

/testsuite
2018-06-12  Paolo Carlini  <paolo.carlini@oracle.com>

* g++.dg/init/delete3.C: New.
* g++.dg/init/new49.C: Likewise.
* g++.dg/init/new25.C: Test locations too.
* g++.dg/template/new4.C: Likewise.
* g++.old-deja/g++.jason/operator.C: Likewise.

From-SVN: r261528

decl2.c (coerce_new_type, [...]): Add location_t parameter and adjust error_at calls.

/cp
2018-06-12  Paolo Carlini  <paolo.carlini@oracle.com>

* decl2.c (coerce_new_type, coerce_delete_type): Add location_t
parameter and adjust error_at calls.
* decl.c (grok_op_properties): Adjust calls.
* cp-tree.h (oerce_new_type, coerce_delete_type): Adjust decls.

/testsuite
2018-06-12  Paolo Carlini  <paolo.carlini@oracle.com>

* g++.dg/init/delete3.C: New.
* g++.dg/init/new49.C: Likewise.
* g++.dg/init/new25.C: Test locations too.
* g++.dg/template/new4.C: Likewise.
* g++.old-deja/g++.jason/operator.C: Likewise.

From-SVN: r261527

Core issue 1331 - const mismatch with defaulted copy constructor

Core issue 1331 - const mismatch with defaulted copy constructor
* class.c (check_bases_and_members): When checking a defaulted
function, mark it as deleted rather than giving an error.

* g++.dg/cpp0x/defaulted15.C (struct F): Remove dg-error.
* g++.dg/cpp0x/defaulted52.C: New test.
* g++.dg/cpp0x/defaulted53.C: New test.
* g++.dg/cpp0x/defaulted54.C: New test.
* g++.dg/cpp0x/defaulted55.C: New test.
* g++.dg/cpp0x/defaulted56.C: New test.
* g++.dg/cpp0x/defaulted57.C: New test.
* g++.dg/cpp0x/defaulted58.C: New test.
* g++.dg/cpp0x/defaulted59.C: New test.
* g++.dg/cpp0x/defaulted60.C: New test.

From-SVN: r261526

macros.h (__glibcxx_check_can_increment_range): New.

2018-06-12  François Dumont  <fdumont@gcc.gnu.org>

* include/debug/macros.h (__glibcxx_check_can_increment_range): New.
(__glibcxx_check_can_decrement_range): New.
* include/debug/debug.h (__glibcxx_requires_can_increment_range): New.
(__glibcxx_requires_can_decrement_range): New.
* include/bits/stl_algobase.h (std::copy(_II, _II, _OI)): Use
__glibcxx_requires_can_increment_range.
(std::move(_II, _II, _OI)): Likewise.
(std::copy_backward(_BI, _BI, _BI2)): Use
__glibcxx_requires_can_decrement_range.
(std::move_backward(_BI, _BI, _BI2)): Likewise.
* testsuite/25_algorithms/copy_backward/debug/1_neg.cc: New.
* testsuite/25_algorithms/copy_backward/debug/2_neg.cc: New.
* testsuite/25_algorithms/copy_backward/debug/3_neg.cc: New.
* testsuite/25_algorithms/equal/debug/1_neg.cc: New.
* testsuite/25_algorithms/equal/debug/2_neg.cc: New.
* testsuite/25_algorithms/equal/debug/3_neg.cc: New.

From-SVN: r261525

P0935R0 Eradicating unnecessarily explicit default constructors

Explicit default constructors are problematic, so this change removes
them from <random> and <ext/random>, as per P0935R0.

* include/bits/random.h (linear_congruential_engine)
(mersenne_twister_engine, subtract_with_carry_engine, random_device)
(uniform_real_distribution, normal_distribution)
(lognormal_distribution, gamma_distribution, chi_squared_distribution)
(cauchy_distribution, fisher_f_distribution, student_t_distribution)
(bernoulli_distribution, binomial_distribution,geometric_distribution)
(negative_binomial_distribution, exponential_distribution)
(weibull_distribution, extreme_value_distribution): Add non-explicit
default constructors. Remove default argument for first parameter of
explicit constructors.
(piecewise_constant_distribution, piecewise_linear_distribution):
Make default constructor non-explicit.
* include/bits/uniform_int_dist.h (uniform_int_distribution): Add
non-explicit default constructors. Remove default argument for first
parameter of explicit constructor.
(simd_fast_mersenne_twister_engine, beta_distribution)
(rice_distribution, nakagami_distribution, pareto_distribution)
(k_distribution, arcsine_distribution, hoyt_distribution)
(triangular_distribution, von_mises_distribution)
(hypergeometric_distribution, logistic_distribution)
(uniform_inside_sphere_distribution): Likewise.
(uniform_on_sphere_distribution): Make default constructor
non-explicit.
* testsuite/26_numerics/random/bernoulli_distribution/cons/default.cc:
Test for non-explicit default constructor. Fix references to standard.
* testsuite/26_numerics/random/binomial_distribution/cons/default.cc:
Likewise.
* testsuite/26_numerics/random/cauchy_distribution/cons/default.cc:
Likewise.
* testsuite/26_numerics/random/chi_squared_distribution/cons/default.cc:
Likewise.
* testsuite/26_numerics/random/discrete_distribution/cons/default.cc:
Likewise.
* testsuite/26_numerics/random/exponential_distribution/cons/default.cc:
Likewise.
* testsuite/26_numerics/random/extreme_value_distribution/cons/default.cc:
Likewise.
* testsuite/26_numerics/random/fisher_f_distribution/cons/default.cc:
Likewise.
* testsuite/26_numerics/random/gamma_distribution/cons/default.cc:
Likewise.
* testsuite/26_numerics/random/geometric_distribution/cons/default.cc:
Likewise.
* testsuite/26_numerics/random/lognormal_distribution/cons/default.cc:
Likewise.
* testsuite/26_numerics/random/negative_binomial_distribution/cons/default.cc:
Likewise.
* testsuite/26_numerics/random/normal_distribution/cons/default.cc:
Likewise.
* testsuite/26_numerics/random/piecewise_constant_distribution/cons/default.cc:
Likewise.
* testsuite/26_numerics/random/piecewise_linear_distribution/cons/default.cc:
Likewise.
* testsuite/26_numerics/random/poisson_distribution/cons/default.cc:
Likewise.
* testsuite/26_numerics/random/student_t_distribution/cons/default.cc:
Likewise.
* testsuite/26_numerics/random/uniform_int_distribution/cons/default.cc:
Likewise.
* testsuite/26_numerics/random/uniform_real_distribution/cons/default.cc:
Likewise.
* testsuite/26_numerics/random/weibull_distribution/cons/default.cc:
Likewise.
* testsuite/ext/random/arcsine_distribution/cons/default.cc: Likewise.
* testsuite/ext/random/beta_distribution/cons/default.cc: Likewise.
* testsuite/ext/random/hoyt_distribution/cons/default.cc: Likewise.
* testsuite/ext/random/hypergeometric_distribution/cons/default.cc:
Likewise.
* testsuite/ext/random/k_distribution/cons/default.cc: Likewise.
* testsuite/ext/random/logistic_distribution/cons/default.cc: Likewise.
* testsuite/ext/random/nakagami_distribution/cons/default.cc: Likewise.
* testsuite/ext/random/normal_mv_distribution/cons/default.cc:
Likewise.
* testsuite/ext/random/pareto_distribution/cons/default.cc: Likewise.
* testsuite/ext/random/rice_distribution/cons/default.cc: Likewise.
* testsuite/ext/random/triangular_distribution/cons/default.cc:
Likewise.
* testsuite/ext/random/uniform_inside_sphere_distribution/cons/default.cc:
Likewise.
* testsuite/ext/random/uniform_on_sphere_distribution/cons/default.cc:
Likewise.
* testsuite/ext/random/von_mises_distribution/cons/default.cc:
Likewise.
* testsuite/util/testsuite_common_types.h
(implicitly_default_constructible): New helper.

From-SVN: r261522

spellcheck: support transpositions aka Damerau-Levenshtein (PR other/69968)

gcc/fortran/ChangeLog:
PR other/69968
* misc.c (gfc_closest_fuzzy_match): Update for renaming of
levenshtein_distance to get_edit_distance.

gcc/ChangeLog:
PR other/69968
* spellcheck-tree.c (levenshtein_distance): Rename to...
(get_edit_distance): ...this, and update for underlying renaming.
* spellcheck-tree.h (levenshtein_distance): Rename to...
(get_edit_distance): ...this.
* spellcheck.c (levenshtein_distance): Rename to...
(get_edit_distance): ...this.  Convert from Levenshtein distance
to Damerau-Levenshtein distance by supporting transpositions of
adjacent characters.  Rename "v1" to "v_next" and "v0" to
"v_one_ago".
(selftest::levenshtein_distance_unit_test_oneway): Rename to...
(selftest::test_edit_distance_unit_test_oneway): ...this, and
update for underlying renaming.
(selftest::levenshtein_distance_unit_test): Rename to...
(selftest::test_get_edit_distance_unit): ...this, and update for
underlying renaming.
(selftest::test_find_closest_string): Add example from PR 69968
where transposition helps
(selftest::test_metric_conditions): Update for renaming.
(selftest::test_metric_conditions): Likewise.
(selftest::spellcheck_c_tests): Likewise.
* spellcheck.h (levenshtein_distance): Rename both overloads to...
(get_edit_distance): ...this.
(best_match::consider): Update for renaming.

gcc/testsuite/ChangeLog:
PR other/69968
* gcc.dg/spellcheck-transposition.c: New test.

From-SVN: r261521

PR tree-optimization/85259 - Missing -Wstringop-overflow= since r256683

gcc/ChangeLog:

PR tree-optimization/85259
* builtins.c (compute_objsize): Handle constant offsets.
* gimple-ssa-warn-restrict.c (maybe_diag_offset_bounds): Return
true iff a warning has been issued.
* gimple.h (gimple_nonartificial_location): New function.
* tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Call
gimple_nonartificial_location and handle -Wno-system-headers.
(handle_builtin_stxncpy): Same.

gcc/testsuite/ChangeLog:

PR tree-optimization/85259
* gcc.dg/Wstringop-overflow-5.c: New test.
* gcc.dg/Wstringop-overflow-6.c: New test.

From-SVN: r261518

re PR fortran/44491 (Diagnostic just shows "<During initialization>" instead of a locus)

2018-06-12  Steven G. Kargl  <kargl@gcc.gnu.org>

PR fortran/44491
* expr.c (gfc_check_assign): Select non-NULL locus.

2018-06-12  Steven G. Kargl  <kargl@gcc.gnu.org>

PR fortran/44491
* gfortran.dg/pr44491.f90: New testcase

From-SVN: r261516

PR c/85931 -  -Wsizeof-pointer-memaccess for strncpy with size of source

gcc/c-family/ChangeLog:

PR c/85931
* c-warn.c (sizeof_pointer_memaccess_warning): Avoid warning when
sizeof source and destination yields the same value.

gcc/ChangeLog:

PR c/85931
* fold-const.c (operand_equal_p): Handle SAVE_EXPR.

gcc/testsuite/ChangeLog:

PR c/85931
* gcc.dg/Wstringop-truncation-3.c: New test.

From-SVN: r261515

p8-vec-xl-xst-v2.c: New.(actually added this time).

[testsuite]

2018-05-31  Will Schmidt  <will_schmidt@vnet.ibm.com>

* gcc.target/powerpc/p8-vec-xl-xst-v2.c: New.(actually added this time).

From-SVN: r261510

p8-vec-xl-xst-v2.c: New.

[testsuite]

2018-05-31  Will Schmidt  <will_schmidt@vnet.ibm.com>

* gcc.target/powerpc/p8-vec-xl-xst-v2.c: New.
* gcc.target/powerpc/p8-vec-xl-xst.c: Disable gimple-folding.
* gcc.target/powerpc/swaps-p8-17.c: Same.

From-SVN: r261509

fold-vec-store-vec_vsx_st-char.c: New.

[testsuite]

2018-05-31  Will Schmidt  <will_schmidt@vnet.ibm.com>

* gcc.target/powerpc/fold-vec-store-vec_vsx_st-char.c: New.
* gcc.target/powerpc/fold-vec-store-vec_vsx_st-double.c: New.
* gcc.target/powerpc/fold-vec-store-vec_vsx_st-float.c: New.
* gcc.target/powerpc/fold-vec-store-vec_vsx_st-int.c: New.
* gcc.target/powerpc/fold-vec-store-vec_vsx_st-longlong.c: New.
* gcc.target/powerpc/fold-vec-store-vec_vsx_st-short.c: New.

From-SVN: r261508

fold-vec-store-builtin_vec_xst-char.c: New.

[testsuite]

2018-05-31  Will Schmidt  <will_schmidt@vnet.ibm.com>

* gcc.target/powerpc/fold-vec-store-builtin_vec_xst-char.c: New.
* gcc.target/powerpc/fold-vec-store-builtin_vec_xst-double.c: New.
* gcc.target/powerpc/fold-vec-store-builtin_vec_xst-float.c: New.
* gcc.target/powerpc/fold-vec-store-builtin_vec_xst-int.c: New.
* gcc.target/powerpc/fold-vec-store-builtin_vec_xst-longlong.c: New.
* gcc.target/powerpc/fold-vec-store-builtin_vec_xst-short.c: New.

From-SVN: r261507

fold-vec-store-vec_xst-char.c: New.

[testsuite]

    2018-05-31  Will Schmidt  <will_schmidt@vnet.ibm.com>

* gcc.target/powerpc/fold-vec-store-vec_xst-char.c: New.
* gcc.target/powerpc/fold-vec-store-vec_xst-double.c: New.
* gcc.target/powerpc/fold-vec-store-vec_xst-float.c: New.
* gcc.target/powerpc/fold-vec-store-vec_xst-int.c: New.
* gcc.target/powerpc/fold-vec-store-vec_xst-longlong.c: New.
* gcc.target/powerpc/fold-vec-store-vec_xst-short.c: New.

From-SVN: r261506

fold-vec-load-vec_vsx_ld-char.c: New.

[testsuite]

2018-05-31  Will Schmidt  <will_schmidt@vnet.ibm.com>

* gcc.target/powerpc/fold-vec-load-vec_vsx_ld-char.c: New.
* gcc.target/powerpc/fold-vec-load-vec_vsx_ld-double.c: New.
* gcc.target/powerpc/fold-vec-load-vec_vsx_ld-float.c: New.
* gcc.target/powerpc/fold-vec-load-vec_vsx_ld-int.c: New.
* gcc.target/powerpc/fold-vec-load-vec_vsx_ld-longlong.c: New.
* gcc.target/powerpc/fold-vec-load-vec_vsx_ld-short.c: New.

From-SVN: r261505

fold-vec-load-builtin_vec_xl-char.c: New.

[testsuite]

2018-05-31  Will Schmidt  <will_schmidt@vnet.ibm.com>

* gcc.target/powerpc/fold-vec-load-builtin_vec_xl-char.c: New.
* gcc.target/powerpc/fold-vec-load-builtin_vec_xl-double.c: New.
* gcc.target/powerpc/fold-vec-load-builtin_vec_xl-float.c: New.
* gcc.target/powerpc/fold-vec-load-builtin_vec_xl-int.c: New.
* gcc.target/powerpc/fold-vec-load-builtin_vec_xl-longlong.c: New.
* gcc.target/powerpc/fold-vec-load-builtin_vec_xl-short.c: New.

From-SVN: r261504

fold-vec-load-vec_xl-char.c: New testcase.

[testsuite]

2018-06-12  Will Schmidt  <will_schmidt@vnet.ibm.com>

* gcc.target/powerpc/fold-vec-load-vec_xl-char.c: New testcase.
* gcc.target/powerpc/fold-vec-load-vec_xl-double.c: New testcase.
* gcc.target/powerpc/fold-vec-load-vec_xl-float.c: New testcase.
* gcc.target/powerpc/fold-vec-load-vec_xl-int.c: New testcase.
* gcc.target/powerpc/fold-vec-load-vec_xl-longlong.c: New testcase.
* gcc.target/powerpc/fold-vec-load-vec_xl-short.c: New testcase.

From-SVN: r261503

rs6000-c.c (altivec_overloaded_builtins): Add BUILTIN_VEC_XST entries for pointer to double and long long.

[gcc]

2018-06-12  Will Schmidt  <will_schmidt@vnet.ibm.com>

* config/rs6000/rs6000-c.c (altivec_overloaded_builtins): Add
BUILTIN_VEC_XST entries for pointer to double and long long.

From-SVN: r261502

Linux/x86: Update TARGET_THREAD_SPLIT_STACK_OFFSET comments

Glibc 2.18 was changed by

commit ecbf434213c0333d81706074e4d107ac45011635
Author: Andreas Jaeger <aj@suse.de>
Date:   Wed May 15 20:20:54 2013 +0200

    Reserve new TLS field for x86 and x86_64

        [BZ #10686]
        * sysdeps/x86_64/tls.h (struct tcbhead_t): Add __private_ss
        field.
        * sysdeps/i386/tls.h (struct tcbhead_t): Likewise.

to reduce the size of __private_tm to make room for __private_ss, which
was supposed to be used for TARGET_THREAD_SPLIT_STACK_OFFSET:

typedef struct
{
  void *tcb;            /* Pointer to the TCB.  Not necessarily the
                           thread descriptor used by libpthread.  */
  dtv_t *dtv;
  void *self;           /* Pointer to the thread descriptor.  */
  int multiple_threads;
  uintptr_t sysinfo;
  uintptr_t stack_guard;
  uintptr_t pointer_guard;
  int gscope_flag;
  int __glibc_reserved1;
  /* Reservation of some values for the TM ABI.  */
  void *__private_tm[4];
  /* GCC split stack support.  */
  void *__private_ss;
} tcbhead_t;

But the offset of __private_ss for i386 was mistakenly set to 0x30,
instead of 0x34 and libgcc/config/i386/morestack.S has:

cmpl %gs:0x30,%eax # See if we have enough space.
movl %eax,%gs:0x30 # Save the new stack boundary.
movl %eax,%gs:0x30 # Save the new stack boundary.
movl %ecx,%gs:0x30 # Save new stack boundary.
movl %eax,%gs:0x30
movl %gs:0x30,%eax
movl %eax,%gs:0x30

Since update TARGET_THREAD_SPLIT_STACK_OFFSET changes split stack ABI,
glibc 2.28 has been changed by

commit 0221ce2a90be2d40fc90f0b5dcec77a1ec013f53
Author: H.J. Lu <hjl.tools@gmail.com>
Date:   Tue Jun 12 06:23:28 2018 -0700

    i386: Change offset of __private_ss to 0x30 [BZ #23250]

to match GCC:

typedef struct
{
  void *tcb; /* Pointer to the TCB.  Not necessarily the
   thread descriptor used by libpthread.  */
  dtv_t *dtv;
  void *self; /* Pointer to the thread descriptor.  */
  int multiple_threads;
  uintptr_t sysinfo;
  uintptr_t stack_guard;
  uintptr_t pointer_guard;
  int gscope_flag;
  int __glibc_reserved1;
  /* Reservation of some values for the TM ABI.  */
  void *__private_tm[3];
  /* GCC split stack support.  */
  void *__private_ss;
  void *__glibc_reserved2;
} tcbhead_t;

PR target/85990
* config/i386/gnu-user.h (TARGET_THREAD_SPLIT_STACK_OFFSET):
Update comments.
* config/i386/gnu-user64.h (TARGET_THREAD_SPLIT_STACK_OFFSET):
Likewise.

From-SVN: r261501

genfixes: exit 1 when autogen not found.

* genfixes: exit 1 when autogen not found.
* genfixes: Remove some redundant code.
* genfixes: Update URL to autogen source code.

From-SVN: r261500

inclhack.def: Fix fixup for assert.h on vxworks.

* inclhack.def: Fix fixup for assert.h on vxworks.
* fixincl.x: Regenerate.

From-SVN: r261499

Document IntegerRange in internal documentation.

2018-06-12  Martin Liska  <mliska@suse.cz>

* doc/options.texi: Document IntegerRange.

From-SVN: r261498

Come up with Deprecated option flag.

2018-06-12  Martin Liska  <mliska@suse.cz>

* doc/options.texi: Document Deprecated option flag.
2018-06-12  Martin Liska  <mliska@suse.cz>

* config/i386/i386.opt: Make MPX-related options as Deprecated.
* opt-functions.awk: Handle Deprecated flag.
* opts-common.c (decode_cmdline_option): Handle cl_deprecated
        and report error.
(read_cmdline_option): Report warning for a deprecated option.
* opts.h (struct cl_option): Add new field cl_deprecated.
(CL_ERR_DEPRECATED): New.
2018-06-12  Martin Liska  <mliska@suse.cz>

* c.opt: Make MPX-related options as Deprecated.
2018-06-12  Martin Liska  <mliska@suse.cz>

* g++.dg/opt/mpx.C: New test.
* gcc.target/i386/mpx.c: New test.

From-SVN: r261497

[PATCH] [ARC][QuarkSE] Update

QuarkSE has lp_count width set to 16 bits. Update the compiler to
consider it.

gcc/
2018-06-12  Claudiu Zissulescu  <claziss@synopsys.com>

* config/arc/arc-arch.h (arc_extras): New enum.
(arc_cpu_t):Add field extra.
(arc_cpu_types): Consider the extras.
* config/arc/arc-cpus.def: Add extras info.
* config/arc/arc-opts.h (processor_type): Consider extra field.
* config/arc/arc.c (arc_override_options): Handle extra field.

From-SVN: r261496

[ARC] Update warning messges.

When we pass an mcpu to the compiler we have two types of (hardware
configuration) flags that are set:

1. Architecture specific, for example code-density is always enabled
for ARCHS architectures. These options are overwriting whatever the
corresponding user options with the preset ones.

2. CPU specific, for example archs is using LL64 option by
default. These options can be freely enabled or disabled.

Because of the above complexity, we need to throw some errors for the
user to know when he/she does something which goes against the above
rules. Thus, I came up with the following set of rules:

1. Overwriting default architecture specific hardware option: it is
ignored, a warning is thrown;

2. Overwriting default CPU specific hardware option: it is taken into
account, a warning is thrown.

gcc/
2018-06-12  Claudiu Zissulescu  <claziss@synopsys.com>

* config/arc/arc-arch.h: Update ARC_OPTX macro.
* config/arc/arc-options.def (ARC_OPTX): Introduce a new doc
field.
* config/arc/arc.c (arc_init): Update pic warning.
(irq_range): Update irq range parsing warnings.
(arc_override_options): Update various warning messages.
(arc_handle_aux_attribute): Likewise.

gcc/testsuite
2018-06-12  Claudiu Zissulescu  <claziss@synopsys.com>

* gcc.target/arc/builtin_simdarc.c: Update test.
* gcc.target/arc/mulsi3_highpart-2.c: Likewise.
* gcc.target/arc/tumaddsidi4.c: Likewise.

From-SVN: r261495

MIPS: Fix build after r261489.

gcc/
* config/mips/i6400.md (i6400_fpu_fadd): Remove frint.

From-SVN: r261494

sourcebuild.texi: Document usage of line number 0 in verify compiler messages directives.

* doc/sourcebuild.texi: Document usage of line number 0 in verify
compiler messages directives.

From-SVN: r261493

libitm/x86: Correct offsets of __private_tm and pointer_guard

In glibc, sysdeps/i386/nptl/tls.h has

typedef struct
{
  void *tcb;            /* Pointer to the TCB.  Not necessarily the
                           thread descriptor used by libpthread.  */
  dtv_t *dtv;
  void *self;           /* Pointer to the thread descriptor.  */
  int multiple_threads;
  uintptr_t sysinfo;
  uintptr_t stack_guard;
  uintptr_t pointer_guard;
  int gscope_flag;
  int __glibc_reserved1;
  /* Reservation of some values for the TM ABI.  */
  void *__private_tm[4];
  /* GCC split stack support.  */
  void *__private_ss;
} tcbhead_t;

and sysdeps/x86_64/nptl/tls.h has

typedef struct
{
  void *tcb;            /* Pointer to the TCB.  Not necessarily the
                           thread descriptor used by libpthread.  */
  dtv_t *dtv;
  void *self;           /* Pointer to the thread descriptor.  */
  int multiple_threads;
  int gscope_flag;
  uintptr_t sysinfo;
  uintptr_t stack_guard;
  uintptr_t pointer_guard;
  unsigned long int vgetcpu_cache[2];
  int __glibc_reserved1;
  int __glibc_unused1;
  /* Reservation of some values for the TM ABI.  */
  void *__private_tm[4];
  /* GCC split stack support.  */
  void *__private_ss;
  long int __glibc_reserved2;
  /* Must be kept even if it is no longer used by glibc since programs,
     like AddressSanitizer, depend on the size of tcbhead_t.  */
  __128bits __glibc_unused2[8][4] __attribute__ ((aligned (32)));

  void *__padding[8];
} tcbhead_t;

The offsets of __private_tm are

i386:   36 bytes
x32:    48 bytes
x86_64: 80 bytes

and the offsets of pointer_guard are:

i386:   24 bytes
x32:    28 bytes
x86_64: 48 bytes

But config/linux/x86/tls.h had

 #ifdef __x86_64__
 #ifdef __LP64__
 # define SEG_READ(OFS)          "movq\t%%fs:(" #OFS "*8),%0"
 # define SEG_WRITE(OFS)         "movq\t%0,%%fs:(" #OFS "*8)"
 # define SEG_DECODE_READ(OFS)   SEG_READ(OFS) "\n\t" \
                                 "rorq\t$17,%0\n\t" \
                                 "xorq\t%%fs:48,%0"
 # define SEG_ENCODE_WRITE(OFS)  "xorq\t%%fs:48,%0\n\t" \
                                 "rolq\t$17,%0\n\t" \
                                 SEG_WRITE(OFS)
 #else
 // For X32.
 # define SEG_READ(OFS)          "movl\t%%fs:(" #OFS "*4),%0"
 # define SEG_WRITE(OFS)         "movl\t%0,%%fs:(" #OFS "*4)"
 # define SEG_DECODE_READ(OFS)   SEG_READ(OFS) "\n\t" \
                                 "rorl\t$9,%0\n\t" \
                                 "xorl\t%%fs:24,%0"
 # define SEG_ENCODE_WRITE(OFS)  "xorl\t%%fs:24,%0\n\t" \
                                 "roll\t$9,%0\n\t" \
                                 SEG_WRITE(OFS)
 #endif
 #else
 # define SEG_READ(OFS)  "movl\t%%gs:(" #OFS "*4),%0"
 # define SEG_WRITE(OFS) "movl\t%0,%%gs:(" #OFS "*4)"
 # define SEG_DECODE_READ(OFS)   SEG_READ(OFS) "\n\t" \
                                 "rorl\t$9,%0\n\t" \
                                 "xorl\t%%gs:24,%0"
 # define SEG_ENCODE_WRITE(OFS)  "xorl\t%%gs:24,%0\n\t" \
                                 "roll\t$9,%0\n\t" \
                                 SEG_WRITE(OFS)
 #endif

static inline struct gtm_thread *gtm_thr(void)
{
  struct gtm_thread *r;
  asm volatile (SEG_READ(10) : "=r"(r));
  return r;
}

static inline void set_gtm_thr(struct gtm_thread *x)
{
  asm volatile (SEG_WRITE(10) : : "r"(x));
}

static inline struct abi_dispatch *abi_disp(void)
{
  struct abi_dispatch *r;
  asm volatile (SEG_DECODE_READ(11) : "=r"(r));
  return r;
}

static inline void set_abi_disp(struct abi_dispatch *x)
{
  void *scratch;
  asm volatile (SEG_ENCODE_WRITE(11) : "=r"(scratch) : "0"(x));
}

SEG_READ, SEG_WRITE, SEG_DECODE_READ and SEG_ENCODE_WRITE were correct
only for x86-64.

Update SEG_READ and SEG_WRITE to use the offset of __private_tm as base
and correct the offset of pointer_guard for x32.  This patch doesn't
change ABI of libitm.

PR libitm/85988
* config/linux/x86/tls.h (SEG_READ): Use the offset of
__private_tm as base.
(SEG_WRITE): Likewise.
(SEG_ENCODE_WRITE): Correct the offset of pointer_guard for x32.
(gtm_thr): Replace SEG_READ(10) with SEG_READ(0).
(set_gtm_thr): Replace SEG_WRITE(10) with SEG_WRITE(0).
(abi_disp): Replace SEG_DECODE_READ(11) with SEG_DECODE_READ(1).
(set_abi_disp): Replace SEG_ENCODE_WRITE(11) with
SEG_ENCODE_WRITE(1).

From-SVN: r261491

MIPS: Add i6500 processor as an alias for i6400.

gcc/ChangeLog:

2018-06-12  Matthew Fortune  <mfortune@gmail.com>

* config/mips/mips-cpus.def: New MIPS_CPU for i6500.
* config/mips/mips-tables.opt: Regenerate.
* config/mips/mips.h (MIPS_ISA_LEVEL_SPEC): Mark i6500 as
mips64r6.
* doc/invoke.texi: Document -march=i6500.

From-SVN: r261490

MIPS: Update I6400 scheduler.

gcc/ChangeLog:

2018-06-12  Prachi Godbole  <prachi.godbole@imgtec.com>

* config/mips/i6400.md (i6400_gpmuldiv): Remove cpu_unit.
(i6400_gpmul): Add cpu_unit.
(i6400_gpdiv): Likewise.
(i6400_msa_add_d): Update reservations.
(i6400_msa_int_add) Likewise.
(i6400_msa_short_logic3) Likewise.
(i6400_msa_short_logic2) Likewise.
(i6400_msa_short_logic) Likewise.
(i6400_msa_move) Likewise.
(i6400_msa_cmp) Likewise.
(i6400_msa_short_float2) Likewise.
(i6400_msa_div_d) Likewise.
(i6400_msa_long_logic1) Likewise.
(i6400_msa_long_logic2) Likewise.
(i6400_msa_mult) Likewise.
(i6400_msa_long_float2) Likewise.
(i6400_msa_long_float4) Likewise.
(i6400_msa_long_float5) Likewise.
(i6400_msa_long_float8) Likewise.
(i6400_fpu_fadd): Include frint type.
(i6400_fpu_store): New define_insn_reservation.
(i6400_fpu_load): Likewise.
(i6400_fpu_move): Likewise.
(i6400_fpu_fcmp): Likewise.
(i6400_fpu_fmadd): Likewise.
(i6400_int_mult): Include imul3nc type and update reservation.
(i6400_int_div): Include idiv3 type and update reservation.
(i6400_int_load): Update to check type not move_type.
(i6400_int_store): Likewise.
(i6400_int_prefetch): Set zero latency.

From-SVN: r261489

ada-tree.h (TYPE_RETURN_BY_DIRECT_REF_P): Change from using TYPE_LANG_FLAG_4 to using TYPE_LANG_FLAG_0.

* gcc-interface/ada-tree.h (TYPE_RETURN_BY_DIRECT_REF_P): Change from
using TYPE_LANG_FLAG_4 to using TYPE_LANG_FLAG_0.
(TYPE_ALIGN_OK): Move around.
(TYPE_PADDING_FOR_COMPONENT): Remove superfluous parentheses.
* gcc-interface/decl.c (change_qualified_type): Move to...
(gnat_to_gnu_entity): Adjust comment.
* gcc-interface/gigi.h (change_qualified_type): ...here; make inline.
(ceil_pow2): Use ceil_log2.
* gcc-interface/utils.c (finish_subprog_decl): Add couple of comments
and do not set TREE_SIDE_EFFECTS.
(handle_noreturn_attribute): Use change_qualified_type.

From-SVN: r261486

decl.c (gnat_to_gnu_entity): Do not get the expression of a dispatch table that is not being defined.

* gcc-interface/decl.c (gnat_to_gnu_entity) <E_Constant>: Do not get
the expression of a dispatch table that is not being defined.
<E_Record_Subtype>: Remove obsolete kludge.

From-SVN: r261483

decl.c (warn_on_field_placement): Use specific wording for discriminants.

* gcc-interface/decl.c (warn_on_field_placement): Use specific wording
for discriminants.
(warn_on_list_placement): New static function.
(components_to_record): Use it to warn on multiple fields in list.

From-SVN: r261480

decl.c (variant_desc): Add AUX field.

* gcc-interface/decl.c (variant_desc): Add AUX field.
(gnat_to_gnu_entity) <discrete_type>: Do not call compute_record_mode
directly.
(reverse_sort_field_list): New static function.
(components_to_record): Place the variant part at the beginning of the
field list when there is an obvious order of increasing position.
(build_variant_list): Initialize it.
(create_variant_part_from): Do not call compute_record_mode directly.
(copy_and_substitute_in_layout): Likewise.  Always sort the fields with
fixed position in order of increasing position, in the record and all
the variants, in any.  Call reverse_sort_field_list.
* gcc-interface/utils.c (make_packable_type): Compute the sizes before
calling finish_record_type.  Do not call compute_record_mode directly.
(finish_record_type): Overhaul final processing depending on REP_LEVEL
and call finish_bitfield_layout if it is equal to one or two.

From-SVN: r261479

[testsuite] Add gcc.dg-selftests/dg-final.exp

2018-06-12  Tom de Vries  <tdevries@suse.de>

* gcc.dg-selftests/dg-final.exp: New file.

From-SVN: r261476

Remove me from Write After Approval.

From-SVN: r261475

gcc.c: Document new %@{...} sequence.

* gcc.c: Document new %@{...} sequence.
(LINK_COMMAND_SPEC): Use it for the -L switches.
(cpp_unique_options): Use it for the -I switches.
(at_file_argbuf): New global variable.
(in_at_file): Likewise.
(alloc_args): Create at_file_argbuf.
(clear_args): Truncate at_file_argbuf.
(store_arg): If in_at_file, push the argument onto at_file_argbuf.
(open_at_file): New function.
(close_at_file): Likewise.
(create_at_file): Delete.
(do_spec_1) <'i'>: Use open_at_file/close_at_file.
<'o'>: Likewise.
<'@'>: New case.
(validate_switches_from_spec): Deal with %@{...} sequence.
(validate_switches): Likewise.
(driver::finalize): Call clear_args.

From-SVN: r261474

Daily bump.

From-SVN: r261472

PR c++/85792 -Wctor-dtor-privacy and inherited constructor.

* class.c (maybe_warn_about_overly_private_class): Handle inherited
constructors.

From-SVN: r261459

PR c++/85963 - -Wunused-but-set with ?: in template.

* pt.c (tsubst_copy_and_build) [COND_EXPR]: Call mark_rvalue_use.

From-SVN: r261458

vx-common.h (USE_TM_CLONE_REGISTRY): #define to 0.

2018-06-11  Rasmus Villemoes  <rasmus.villemoes@prevas.dk>

* config/vx-common.h (USE_TM_CLONE_REGISTRY): #define to 0.

From-SVN: r261457

decl.c (grok_op_properties): Consistently use the location of the decl...

/cp
2018-06-11  Paolo Carlini  <paolo.carlini@oracle.com>

* decl.c (grok_op_properties): Consistently use the location
of the decl; remove special casing of POSTINCREMENT_EXPR and
POSTDECREMENT_EXPR wrt default arguments.

/testsuite
2018-06-11  Paolo Carlini  <paolo.carlini@oracle.com>

* g++.dg/lookup/new2.C: Test locations too.
* g++.dg/other/ellipsis1.C: Likewise.
* g++.dg/other/operator1.C: Likewise.
* g++.dg/other/operator2.C: Likewise.
* g++.dg/overload/operator2.C: Likewise.
* g++.dg/parse/defarg11.C: Likewise.
* g++.dg/parse/operator4.C: Likewise.
* g++.dg/template/error30.C: Likewise.
* g++.dg/template/explicit-args3.C: Likewise.
* g++.dg/warn/effc1.C: Likewise.
* g++.old-deja/g++.brendan/prepost2.C: Likewise.
* g++.old-deja/g++.brendan/prepost3.C: Likewise.
* g++.old-deja/g++.bugs/900215_01.C: Likewise.
* g++.old-deja/g++.jason/conversion5.C: Likewise.
* g++.old-deja/g++.jason/operator.C: Likewise.

From-SVN: r261456

decl.c (grok_op_properties): Consistently use the location of the decl...

/cp
2018-06-11  Paolo Carlini  <paolo.carlini@oracle.com>

* decl.c (grok_op_properties): Consistently use the location
of the decl; remove special casing of POSTINCREMENT_EXPR and
POSTDECREMENT_EXPR wrt default arguments.

/testsuite
2018-06-11  Paolo Carlini  <paolo.carlini@oracle.com>

* g++.dg/lookup/new2.C: Test locations too.
* g++.dg/other/ellipsis1.C: Likewise.
* g++.dg/other/operator1.C: Likewise.
* g++.dg/other/operator2.C: Likewise.
* g++.dg/overload/operator2.C: Likewise.
* g++.dg/parse/defarg11.C: Likewise.
* g++.dg/parse/operator4.C: Likewise.
* g++.dg/template/error30.C: Likewise.
* g++.dg/template/explicit-args3.C: Likewise.
* g++.dg/warn/effc1.C: Likewise.
* g++.old-deja/g++.brendan/prepost2.C: Likewise.
* g++.old-deja/g++.brendan/prepost3.C: Likewise.
* g++.old-deja/g++.bugs/900215_01.C: Likewise.
* g++.old-deja/g++.jason/conversion5.C: Likewise.
* g++.old-deja/g++.jason/operator.C: Likewise.

From-SVN: r261455

Fix ptrmem comparison for unions.

* constexpr.c (cxx_eval_binary_expression): Special case comparison
of pointers to members of the same union.

From-SVN: r261454

invoke.texi (-Wall): List -Wc++17-compat.

gcc/ChangeLog:

* doc/invoke.texi (-Wall): List -Wc++17-compat.
(Wno-class-memaccess): Add @opindex.
(Wno-templates, Wno-multiple-inheritance): Same.
(Wno-virtual-inheritance, Wno-namespaces): Same.
(Wno-pedantic, Wno-chkp, Wcoverage-mismatch): Same.
(Wno-format-overflow, Wno-format-truncation): Same.
(Wno-unused-local-typedefs, Walloc-size-larger-than): Same.
(Wno-alloc-size-larger-than, Wframe-larger-than): Same
(Wno-stack-usage, Wno-c++-compat, Wno-c++11-compat): Same.
(Wno-c++14-compat, Wno-c++17-compat, Wno-openmp-simd): Same.
(Wno-unsuffixed-float-constants, Wno-addr-space-convert): Same.
(Wno-misspelled-isr): Same.

From-SVN: r261453

PR tree-optimization/86083 - handle non-constant assignments in strlen

gcc/ChangeLog:

PR tree-optimization/86083
* tree-ssa-strlen.c (handle_char_store): Use tree_expr_nonzero_p.

gcc/testsuite/ChangeLog:

PR tree-optimization/86083
* gcc.dg/strlenopt-44.c: New test.

From-SVN: r261452

libgo: add riscv and js/wasm as known targets

    Incorporates cut down versions of https://golang.org/cl/102835 and
    https://golang.org/cl/106256 from the master sources.

    This will tell go/build to skip files with those tags.

    Reviewed-on: https://go-review.googlesource.com/117996

From-SVN: r261451

tree-eh.c (lower_eh_constructs_2): Add a comma to comment.

2018-06-11 Zhouyi Zhou <zhouzhouyi@gmail.com>

* tree-eh.c (lower_eh_constructs_2): Add a comma to comment.

From-SVN: r261450

re PR fortran/45521 ([F08] GENERIC resolution with ALLOCATABLE/POINTER and PROCEDURE)

2018-06-11  Janus Weil  <janus@gcc.gnu.org>

PR fortran/45521
* interface.c (compare_ptr_alloc): New function.
(generic_correspondence): Call it.

2018-06-11  Janus Weil  <janus@gcc.gnu.org>

PR fortran/45521
* gfortran.dg/generic_32.f90: New test.
* gfortran.dg/generic_33.f90: New test.

From-SVN: r261448

PR c++/86094 - wrong code with defaulted move ctor.

* tree.c (type_has_nontrivial_copy_init): Fix move ctor handling.

From-SVN: r261444

* doc/invoke.texi: Document -fabi-version=12.

From-SVN: r261443

* fr.po: Update.

From-SVN: r261439

altivec-7.c (main): Remove tests vec_unpackh(vecubi[0]) and vec_unpackl(vecubi[0]) returning long...

gcc/testsuite/ChangeLog:

2018-06-11  Carl Love  <cel@us.ibm.com>
* gcc.target/powerpc/altivec-7.c (main): Remove tests
vec_unpackh(vecubi[0]) and vec_unpackl(vecubi[0]) returning
long long bool.  Remove duplicate dg-final for xxlxor.  Update
dg-final instruction counts.
* gcc.target/powerpc/altivec-37.c (main): New file for
tests vec_unpackh and vec_unpackl returning long long bool and
long long int.

From-SVN: r261438

rs6000: Put constraints on the correct operand in movdi (PR85755)

Some of the mov* patterns use ^ and $ constraint modifiers, which mean
give a penalty to this alternative if this operand needs a reload.  They
are meant here to give a penalty if a register operand needs reloading
(because it needs to be in a different kind of register), not when a
memory operand needs reloading (which is easy and cheap to do).

This patch fixes the movdi patterns.  This fixes PR85755.

The following are changed (name, old constraints, new constraints):
FPR store   ^m := d     m := ^d
FPR move    ^d := d     ^d := ^d
AVX store   ^wY := wb   wY := ^wb
AVX store   $Z := wv    Z := $wv
VSX move    ^wi := wi   ^wi := ^wi

PR target/85755
* config/rs6000/rs6000.md (*movdi_internal32): Put constraint modifiers
on the correct operand.
(*movdi_internal64): Ditto.

---
 gcc/config/rs6000/rs6000.md | 24 ++++++++++++------------
 1 file changed, 12 insertions(+), 12 deletions(-)

diff --git a/gcc/config/rs6000/rs6000.md b/gcc/config/rs6000/rs6000.md
index a2605a0..f06591f 100644
--- a/gcc/config/rs6000/rs6000.md
+++ b/gcc/config/rs6000/rs6000.md
@@ -8567,16 +8567,16 @@ (define_insn_and_split "reload_gpr_from_vsxsf"

 (define_insn "*movdi_internal32"
   [(set (match_operand:DI 0 "nonimmediate_operand"
-         "=Y,        r,         r,         ^m,        ^d,         ^d,
-          r,         ^wY,       $Z,        ^wb,       $wv,        ^wi,
+         "=Y,        r,         r,         m,         ^d,         ^d,
+          r,         wY,        Z,         ^wb,       $wv,        ^wi,
           *wo,       *wo,       *wv,       *wi,       *wi,        *wv,
           *wv")

  (match_operand:DI 1 "input_operand"
-          "r,        Y,         r,         d,         m,          d,
-           IJKnGHF,  wb,        wv,        wY,        Z,          wi,
-           Oj,       wM,        OjwM,      Oj,        wM,         wS,
-           wB"))]
+         "r,         Y,         r,         ^d,        m,          ^d,
+          IJKnGHF,   ^wb,       $wv,       wY,        Z,          ^wi,
+          Oj,        wM,        OjwM,      Oj,        wM,         wS,
+          wB"))]

   "! TARGET_POWERPC64
    && (gpc_reg_operand (operands[0], DImode)
@@ -8643,17 +8643,17 @@ (define_split
 (define_insn "*movdi_internal64"
   [(set (match_operand:DI 0 "nonimmediate_operand"
                "=YZ,       r,         r,         r,         r,          r,
-                ^m,        ^d,        ^d,        ^wY,       $Z,         $wb,
+                m,         ^d,        ^d,        wY,        Z,          $wb,
                 $wv,       ^wi,       *wo,       *wo,       *wv,        *wi,
                 *wi,       *wv,       *wv,       r,         *h,         *h,
                 ?*r,       ?*wg,      ?*r,       ?*wj")

  (match_operand:DI 1 "input_operand"
-                "r,        YZ,        r,         I,         L,          nF,
-                 d,        m,         d,         wb,        wv,         wY,
-                 Z,        wi,        Oj,        wM,        OjwM,       Oj,
-                 wM,       wS,        wB,        *h,        r,          0,
-                 wg,       r,         wj,        r"))]
+               "r,         YZ,        r,         I,         L,          nF,
+                ^d,        m,         ^d,        ^wb,       $wv,        wY,
+                Z,         ^wi,       Oj,        wM,        OjwM,       Oj,
+                wM,        wS,        wB,        *h,        r,          0,
+                wg,        r,         wj,        r"))]

   "TARGET_POWERPC64
    && (gpc_reg_operand (operands[0], DImode)
--
1.8.3.1

From-SVN: r261435

Fix typo done in MPX removal (PR tree-optimization/86089).

2018-06-11  Martin Liska  <mliska@suse.cz>

        PR tree-optimization/86089
* tree-ssa-strlen.c (get_string_length): Move back removed hunk.
2018-06-11  Martin Liska  <mliska@suse.cz>

        PR tree-optimization/86089
* gcc.dg/tree-ssa/pr86089.c: New test.

From-SVN: r261434

Remove redundant intrinsics

gcc/
* config/i386/vaesintrin.h (_mm_aesdec_epi128, _mm_aesdeclast_epi128)
_mm_aesenc_epi128, _mm_aesenclast_epi128: Remove.
* config/i386/vpclmulqdqintrin.h (_mm_clmulepi64_epi128): Remove.
gcc/testsuite/
* gcc.target/i386/avx512fvl-vaes-1.c: Remove 128bit versions from test.
* gcc.target/i386/vpclmulqdq.c: Ditto.

From-SVN: r261433

dwarf2out.c (gen_compile_unit_die): Fallout to DW_LANG_Ada83 for Ada with strict dwarf2.

2018-06-11  Olivier Hainque  <hainque@adacore.com>

        * dwarf2out.c (gen_compile_unit_die): Fallout to DW_LANG_Ada83
        for Ada with strict dwarf2.

testsuite/

        * gnat.dg/lang-dw2.adb: New test.
        * gnat.dg/lang-dw3.adb: New test.

From-SVN: r261432

[Ada] Fix wrong code for initialization of fat pointer with -Og

2018-06-11  Eric Botcazou  <ebotcazou@adacore.com>

gcc/ada/

* gcc-interface/decl.c (gnat_to_gnu_entity) <E_Array_Type>: Reuse the
existing fields of a dummy fat pointer type, if any.  Clear the
TYPE_DECL_SUPPRESS_DEBUG on the fat pointer type after completing it.

From-SVN: r261431

[Ada] Suppress the expansion of ignored assertion pragmas

This patch suppresses the expansion of ignored assertion pragmas.

2018-06-11  Hristian Kirtchev  <kirtchev@adacore.com>

gcc/ada/

* contracts.adb (Process_Body_Postconditions): Expand only checked
postconditions.
(Process_Contract_Cases_For): Expand only checked contract cases.
(Process_Inherited_Preconditions): Ignored class-wide preconditions are
partially expanded because some of their semantic checks are tied to
the expansion.
(Process_Preconditions_For): Expand only checked preconditions.
(Process_Spec_Postconditions): Expand only checked preconditions.
Ignored class-wide preconditions are partially expanded because some of
their semantic checks are tied to the expansion.
* exp_prag.adb (Expand_N_Pragma): Suppress the expansion of ignored
assertion pragmas.
* exp_util.adb (Add_Inherited_Invariants): Code clean up.
* sem_util.adb (Propagate_Invariant_Attributes): Code clean up.

gcc/testsuite/

* gnat.dg/assertion_policy1.adb, gnat.dg/assertion_policy1_pkg.adb,
gnat.dg/assertion_policy1_pkg.ads: New testcase.

From-SVN: r261430

[Ada] Minor reformatting

2018-06-11  Hristian Kirtchev  <kirtchev@adacore.com>

gcc/ada/

* exp_ch9.adb, exp_unst.adb, inline.adb, libgnat/a-ciorma.adb,
libgnat/a-ciormu.adb, libgnat/a-ciorse.adb, libgnat/a-coorma.adb,
libgnat/a-coormu.adb, libgnat/a-coorse.adb, sem_prag.adb: Minor
reformatting.

From-SVN: r261429

[Ada] Typo fixes and minor reformatting

2018-06-11  Gary Dismukes  <dismukes@adacore.com>

gcc/ada/

* exp_unst.ads, exp_unst.adb: Typo fixes and minor reformatting.

From-SVN: r261428

[Ada] Double finalization of limited controlled result

This patch disables a build-in-place optimization when a function returns a
limited controlled result because the optimization may violate the semantics of
finalizable types by performing illegal calls to Finalize.

In general, the optimization causes the result object of a build-in-place
function to be allocated at the caller site, with a pointer to the object
passed to the function. The function then simply initializes the caller-
allocated object.

This mode of operation however violates semantics of finalizable types when
the context of the call is allocation. The act of allocating the controlled
object at the caller site will place it on the associated access type's
finalization master. If the function fails the initialization of the object,
the malformed object will still be finalized when the finalization master
goes out of scope. This is dangerous, and must not happen.

------------
-- Source --
------------

--  pack.ads

with Ada.Finalization; use Ada.Finalization;

package Pack is
   type Lim_Ctrl is new Limited_Controlled with null record;
   procedure Finalize (Obj : in out Lim_Ctrl);

   type Lim_Ctrl_Ptr is access all Lim_Ctrl;

   function Make_Lim_Ctrl_Bad_Init return Lim_Ctrl;
   function Make_Lim_Ctrl_OK_Init return Lim_Ctrl;
end Pack;

--  pack.adb

with Ada.Text_IO; use Ada.Text_IO;

package body Pack is
   procedure Finalize (Obj : in out Lim_Ctrl) is
   begin
      Put_Line ("     Finalize");
   end Finalize;

   function Make_Lim_Ctrl_Bad_Init return Lim_Ctrl is
   begin
      return Result : Lim_Ctrl := raise Program_Error do
         null;
      end return;
   end Make_Lim_Ctrl_Bad_Init;

   function Make_Lim_Ctrl_OK_Init return Lim_Ctrl is
   begin
      return Result : Lim_Ctrl do
         raise Program_Error;
      end return;
   end Make_Lim_Ctrl_OK_Init;
end Pack;

--  main.adb

with Ada.Text_IO; use Ada.Text_IO;
with Pack;        use Pack;

procedure Main is
begin
   begin
      Put_Line ("1) Heap-allocated bad init");

      declare
         Obj : Lim_Ctrl_Ptr := new Lim_Ctrl'(Make_Lim_Ctrl_Bad_Init);
      begin
         Put_Line ("1) ERROR: Heap-allocated bad init: exception not raised");
      end;

   exception
      when Program_Error =>
         Put_Line ("1) Heap-allocated bad init: Program_Error raised");
      when others =>
         Put_Line ("1) ERROR: Heap-allocatd bad init: unexpected exception");
   end;

   begin
      Put_Line ("2) Stack-allocated bad init");

      declare
         Obj : Lim_Ctrl := Make_Lim_Ctrl_Bad_Init;
      begin
         Put_Line ("2) ERROR: Stack-allocated bad init: exception not raised");
      end;

   exception
      when Program_Error =>
         Put_Line ("2) Stack-allocated bad init: Program_Error raised");
      when others =>
         Put_Line ("2) ERROR: Stack-allocated bad init: unexpected exception");
   end;

   begin
      Put_Line ("3) Heap-allocated OK init");

      declare
         Obj : Lim_Ctrl_Ptr := new Lim_Ctrl'(Make_Lim_Ctrl_OK_Init);
      begin
         Put_Line ("3) ERROR: Heap-allocated OK init: exception not raised");
      end;

   exception
      when Program_Error =>
         Put_Line ("3) Heap-allocated OK init: Program_Error raised");
      when others =>
         Put_Line ("3) ERROR: Heap-allocatd OK init: unexpected exception");
   end;

   begin
      Put_Line ("4) Stack-allocated OK init");

      declare
         Obj : Lim_Ctrl := Make_Lim_Ctrl_OK_Init;
      begin
         Put_Line ("4) ERROR: Stack-allocated OK init: exception not raised");
      end;

   exception
      when Program_Error =>
         Put_Line ("4) Stack-allocated OK init: Program_Error raised");
      when others =>
         Put_Line ("4) ERROR: Stack-allocated OK init: unexpected exception");
   end;
end Main;

----------------------------
-- Compilation and output --
----------------------------

$ gnatmake -q main.adb
$ ./main
1) Heap-allocated bad init
1) Heap-allocated bad init: Program_Error raised
2) Stack-allocated bad init
2) Stack-allocated bad init: Program_Error raised
3) Heap-allocated OK init
     Finalize
3) Heap-allocated OK init: Program_Error raised
4) Stack-allocated OK init
     Finalize
4) Stack-allocated OK init: Program_Error raised

2018-06-11  Hristian Kirtchev  <kirtchev@adacore.com>

gcc/ada/

* exp_ch6.adb (Add_Unconstrained_Actuals_To_Build_In_Place_Call): Do
not add any actuals when the size of the object is known, and the
caller will allocate it.
(Build_Heap_Allocator): Rename to Build_Heap_Or_Pool_Allocator to
better illustrate its functionality. Update the comment on the
generated code.  Generate a branch for the heap and pool cases where
the object is not necessarity controlled.
(Expand_N_Extended_Return_Statement): Expand the extended return
statement into four branches depending the requested mode if the caller
will not allocate the object on its side.
(Make_Build_In_Place_Call_In_Allocator): Do not allocate a controlled
object on the caller side because this will violate the semantics of
finalizable types. Instead notify the function to allocate the object
on the heap or a user-defined storage pool.
(Needs_BIP_Alloc_Form): A build-in-place function needs to be notified
which of the four modes to employ when returning a limited controlled
result.
* exp_util.adb (Build_Allocate_Deallocate_Proc): Remove a redundant
guard which is already covered in Needs_Finalization.

From-SVN: r261427

[Ada] Improve last exception info availability from C++ handlers

The Most_Recent_Exception service failed to provide accurate information on an
Ada exception caught by a C++ handler for foreign exceptions. The service
relies on updates of a "current exception buffer" from live exception objects
at various points of the propagation process and this update was not performed
early enough for the case of foreign exception handlers in non-Ada handlers.

The correction applied here consists in moving one of the updates earlier in
the raise process, just before unwinding starts, then refine the update API to
prevent a redundant copy during the unwinding search phase for the same
exception.

The example below, compiled with

gcc -c b.cc
gnatmake -g main.adb -largs b.o --LINK=g++

is expected to run and display

ada info:
Checking Most_Recent_Exception for CONSTRAINT_ERROR ... OK!

// b.cc

extern "C" {
 void foo ();
 extern void _ada_trigger ();
 extern void _ada_occurrence_info ();
}

void foo ()
{
 try {
   _ada_trigger ();
 } catch (const abi::__foreign_exception &e) {
   printf ("ada info:\n");
   _ada_occurrence_info();
 }
}

-- main.adb

with EH;
procedure Main is
begin
  EH.Foo;
end;

-- eh.adb

with Gnat.Most_Recent_Exception;
with Ada.Text_IO; use Ada.Text_IO;

package body EH is

  procedure Ada_Trigger is
  begin
     raise Constraint_Error;
  end;

  procedure Ada_Occurrence_Info is
  begin
     Check_MRE ("CONSTRAINT_ERROR");
  end;

  function Pre_Check_MRE (Ename : String) return Exception_Id is
     MROA : Exception_Occurrence_Access :=
       GNAT.Most_Recent_Exception.Occurrence_Access;
  begin
     Put ("Checking Most_Recent_Exception for " & Ename & " ... ");

     if MROA = null then
        Put_Line ("Most recent exception occurrence access is NULL");
        return Null_Id;
     else
        return Exception_Identity (MROA.all);
     end if;
  end;

  procedure Diagnose_MRE (MRID : Exception_Id; Ok : Boolean) is
  begin
     if Ok then
        Put_Line ("OK!");
     else
        Put_Line ("Err, Most_Recent_Exception was " & Exception_Name (MRID));
     end if;
  end;

  procedure Check_MRE (Eid : Exception_Id) is
     MRID : Exception_Id := Pre_Check_MRE (Ename => Exception_Name (Eid));
  begin
     Diagnose_MRE (MRID, Ok => Eid = MRID);
  end;

  procedure Check_MRE (Ename : String) is
     MRID : Exception_Id := Pre_Check_MRE (Ename => Ename);
  begin
     Diagnose_MRE (MRID, Ok => Ename = Exception_Name (MRID));
  end;

end;

-- eh.ads

with Ada.Exceptions; use Ada.Exceptions;
package EH is
  procedure Ada_Trigger with
    Export, Convention => C, External_Name => "_ada_trigger";

  procedure Ada_Occurrence_Info with
    Export, Convention => C, External_Name => "_ada_occurrence_info";

  procedure Foo with Import, Convention => C, External_Name => "foo";

  procedure Check_MRE (Eid : Exception_Id);
  procedure Check_MRE (Ename : String);

end;

2018-06-11  Olivier Hainque  <hainque@adacore.com>

gcc/ada/

* libgnat/s-excmac*.ads: Factorize Unwind_Action definitions ...
* libgnat/a-exexpr.adb: ... Here, then add comments describing the
major datastructures associated with the current exception raised.
(Setup_Current_Excep): Accept a "Phase" argument conveying the
unwinding phase during which this subprogram is called.  For an Ada
exception, don't update the current exception buffer from the raised
exception object during SEARCH_PHASE, as this is redundant with the
call now issued just before propagation starts.
(Propagate_GCC_Exception): Move call to Setup_Current_Excep ahead of
the unwinding start, conveying Phase 0.
(Unhandled_Except_Handler): Pass UA_CLEANUP_PHASE as the Phase value on
the call to Setup_Current_Excep.
* raise-gcc.c (personality_body): Pass uw_phases as the Phase argument
on calls to Setup_Current_Excep.

From-SVN: r261426

[Ada] Unnesting: refactor handling of uplevel refs. for unconstrained arrays

2018-06-11  Ed Schonberg  <schonberg@adacore.com>

gcc/ada/

* exp_unst.ads, exp_unst.adb (Needs_Fat_Pointer,
Build_Access_Type_Decl): New subprograms to handle uplevel references
to formals of an unconstrained array type. The activation record
component for these is an access type, and the reference is rewritten
as an explicit derefenrence of that component.

From-SVN: r261425

[Ada] Dangling cursor checks in Element function

In Ada.Containers.Ordered_Maps, if a dangling cursor is passed to the Element
function, execution is erroneous. Therefore, the compiler is not obligated to
detect this error. However, this patch inserts code that will detect this error
in some cases, and raise Program_Error. The same applies to Ordered_Sets,
Ordered_Multisets, Indefinite_Ordered_Maps, Indefinite_Ordered_Sets, and
Indefinite_Ordered_Multisets. No test available for erroneous execution.

2018-06-11  Bob Duff  <duff@adacore.com>

gcc/ada/

* libgnat/a-ciorma.adb, libgnat/a-ciormu.adb, libgnat/a-ciorse.adb,
libgnat/a-coorma.adb, libgnat/a-coormu.adb, libgnat/a-coorse.adb:
(Element): Add code to detect dangling cursors in some cases.

From-SVN: r261424

[Ada] Mark parameters as coming from source for GNATprove

When building a separate subprogram declaration for possible inlining of
local subprograms in GNATprove mode, correctly mark subprogram parameters
as coming from source.

This has no impact on compilation.

2018-06-11  Yannick Moy  <moy@adacore.com>

gcc/ada/

* sem_ch6.adb (Build_Subprogram_Declaration): Mark parameters as coming
from source.

From-SVN: r261423

[Ada] Missing predicate function body for derived type in nested package

This patch fixes a bug in the construction of predicate functions.  For a
derived type, we must ensure that the parent type is already frozen so that its
predicate function has been constructed already. This is necessary if the
parent is declared in a nested package and its own freeze point has not been
reached when the derived type is frozen by a local object declaration.

2018-06-11  Ed Schonberg  <schonberg@adacore.com>

gcc/ada/

* sem_ch13.adb (Build_Predicate_Functions): For a derived type, ensure
that its parent is already frozen so that its predicate function, if
any, has already been constructed.

gcc/testsuite/

* gnat.dg/predicate1.adb: New testcase.

From-SVN: r261422

[Ada] Reject violation of SPARK 6.1.4(12) with enclosing task unit

SPARK 6.1.4(12) applies both to enclosing subprograms and enclosing task
units, but the latter was not correctly rejected.

2018-06-11  Yannick Moy  <moy@adacore.com>

gcc/ada/

* sem_prag.adb (Check_Mode_Restriction_In_Enclosing_Context): Adapt for
possible task unit as the enclosing context.

gcc/testsuite/

* gnat.dg/spark1.adb, gnat.dg/spark1.ads: New testcase.

From-SVN: r261421

[Ada] Minor tweaks for Repinfo

2018-06-11  Eric Botcazou  <ebotcazou@adacore.com>

gcc/ada/

* gnat1drv.adb: Remove use clause for Repinfo.
(Gnat1drv): Beef up comment about the interplay between -gnatc and
back-end annotations.  Use full qualified name for List_Rep_Info.

From-SVN: r261420

[Ada] Make GNAT.Array_Split a preelaborable unit

This patch makes GNAT.Array_Split a preelaborable unit. As a result, it can be
withed by other preelaborated untis.

2018-06-11  Hristian Kirtchev  <kirtchev@adacore.com>

gcc/ada/

* libgnat/g-arrspl.ads: Add pragma Preelaborate.

gcc/testsuite/

* gnat.dg/gnat_array_split1.adb, gnat.dg/gnat_array_split1.ads: New
testcase.

From-SVN: r261419

[Ada] Simplify expansion of "and then" in CodePeer mode

2018-06-11  Arnaud Charlet  <charlet@adacore.com>

gcc/ada/

* exp_ch4.adb (Expand_Record_Equality): Remove extraneous "True and
then" and general logical "ada" in codepeer mode.

From-SVN: r261418

[Ada] Crash on protected type entry family

The compiler may blow up compiling a the body of a protected type that has a
family entry whose entry index specification contains a call to a function.

2018-06-11  Javier Miranda  <miranda@adacore.com>

gcc/ada/

* exp_ch9.adb (Expand_N_Protected_Body): Add missing handling of
N_Call_Marker nodes.

gcc/testsuite/

* gnat.dg/prot4.adb: New testcase.

From-SVN: r261417

[Ada] Minor reformatting

2018-06-11  Arnaud Charlet  <charlet@adacore.com>

gcc/ada/

* exp_ch3.adb, exp_unst.adb, inline.adb, sem_prag.adb: Minor
reformatting.

From-SVN: r261416

[Ada] Add Suppressible argument to Assertion_Policy in documentation

2018-06-11  Yannick Moy  <moy@adacore.com>

gcc/ada/

* doc/gnat_rm/implementation_defined_pragmas.rst: Add Suppressible
argument to Assertion_Policy
* gnat_rm.texi: Regenerate.

From-SVN: r261415

[Ada] Do not query the representation information in CodePeer/GNATprove

Representation information generated when user calls the compiler with -gnatR
switch is not available when running the frontend inside CodePeer or GNATprove.
Do not query such information in that case, as this leads to spurious messages
that it is not available.

There is no impact on compilation.

2018-06-11  Yannick Moy  <moy@adacore.com>

gcc/ada/

* gnat1drv.adb: Do not check representation information in CodePeer and
GNATprove modes, as these modes call a special backend instead of gigi,
so do not have the information.

From-SVN: r261414

[Ada] Mark extended return of unconstrained type as never inlined

Calls to subprograms whose body was an extended return of an unconstrained
type were marked as not inlined, while the subprogram itself was marked as
always inlined. This was inconsistent and could lead to crash in GNATprove.
Now such subprograms are marked as not candidates for inlining.

This mostly impacts GNATprove, as it relates to frontend inlining which is
not used anymore in normal compilation.

2018-06-11  Yannick Moy  <moy@adacore.com>

gcc/ada/

* inline.adb (Build_Body_To_Inline): Consider case of extended return
of unconstrained type as one case where inlining is not supported.
(Expand_Inlined_Call): Remove special case for body as extended return
of unconstrained type.

From-SVN: r261413

[Ada] Do not force Part_Of on generic units

This fixes the code checking SPARK RM 7.2.6(3) so that generic child units
are not forced to use Part_Of to relate their abstract state to the state
of their parent.

2018-06-11  Yannick Moy  <moy@adacore.com>

gcc/ada/

* sem_prag.adb (Analyze_Part_Of): Only allow Part_Of on non-generic
unit.
(Check_Missing_Part_Of): Do not force Part_Of on generic unit.

gcc/testsuite/

* gnat.dg/part_of1-instantiation.adb,
gnat.dg/part_of1-instantiation.ads,
gnat.dg/part_of1-private_generic.adb,
gnat.dg/part_of1-private_generic.ads, gnat.dg/part_of1.ads: New
testcase.

From-SVN: r261412