PR libstdc++/85930 fix misaligned reference

PR libstdc++/85930
* include/bits/shared_ptr_base.h (_Sp_make_shared_tag::_S_ti): Align
the static variable correctly.

From-SVN: r261155

re PR fortran/85981 (ICE in gfc_trans_string_copy, at fortran/trans-expr.c:6539)

2018-06-04  Steven G. Kargl  <kargl@gcc.gnu.org>

PR fortran/85981
* resolve.c (resolve_allocate_deallocate): Check errmsg is default
character kind.

2018-06-04  Steven G. Kargl  <kargl@gcc.gnu.org>

PR fortran/85981
* gfortran.dg/allocate_alloc_opt_14.f90: New test.
* gfortran.dg/allocate_alloc_opt_1.f90: Update error string.
* gfortran.dg/allocate_stat_2.f90: Ditto.
* gfortran.dg/deallocate_alloc_opt_1.f90: Ditto.

From-SVN: r261154

[MAINTAINERS] Update email address

* MAINTAINERS: Update my email address.

From-SVN: r261153

PR c++/61806 - missed SFINAE with partial specialization.

* cp-tree.h (deferring_access_check_sentinel): Add deferring_kind
parameter to constructor.
* pt.c (instantiate_class_template_1): Enable access checking
before call to most_specialized_partial_spec.

From-SVN: r261151

Fix expand_expr_real_1 handling of BLKmode bitfield references

The handling of bitfield references in expand_expr_real_1 includes:

            machine_mode ext_mode = mode;

            if (ext_mode == BLKmode
                && ! (target != 0 && MEM_P (op0)
                      && MEM_P (target)
                      && multiple_p (bitpos, BITS_PER_UNIT)))
              ext_mode = int_mode_for_size (bitsize, 1).else_blk ();

            if (ext_mode == BLKmode)
              {
                [...]
                gcc_assert (MEM_P (op0)

Here "mode" is the TYPE_MODE of the result, so when mode == BLKmode,
the target must be a MEM if nonnull, since no other rtl objects can
have BLKmode.  But there's no guarantee that the source value op0 is also
BLKmode and thus also a MEM: we can reach the assert for any source if
the bitsize being extracted is larger than the largest integer mode
(or larger than MAX_FIXED_MODE_SIZE).

This triggered for SVE with -msve-vector-bits=512, where we could
sometimes try to extract a BLKmode value from a 512-bit vector,
and where int_mode_for_size would rightly fail for large bitsizes.

The patch reuses the existing:

/* Otherwise, if this is a constant or the object is not in memory
   and need be, put it there.  */
else if (CONSTANT_P (op0) || (!MEM_P (op0) && must_force_mem))
  {
    memloc = assign_temp (TREE_TYPE (tem), 1, 1);
    emit_move_insn (memloc, op0);
    op0 = memloc;
    clear_mem_expr = true;
  }

to handle this case.

2018-05-29  Richard Sandiford  <richard.sandiford@linaro.org>

gcc/
* expr.c (expand_expr_real_1): Force the operand into memory if
its TYPE_MODE is BLKmode and if there is no integer mode for
the number of bits being extracted.

gcc/testsuite/
* gcc.target/aarch64/sve/extract_5.c: New test.

From-SVN: r261150

re PR target/85832 ([AVX512] possible shorter code when comparing with vector of zeros)

PR target/85832
PR target/86036
* config/i386/sse.md (<avx512>_eq<mode>3<mask_scalar_merge_name>_1):
Use vptestnm rather than vptestm in (=Yc,v,C) variant.

* gcc.target/i386/avx512f-pr85832.c: Expect vptestnm rather than
vptestm.
* gcc.target/i386/avx512vl-pr85832.c: Likewise.
* gcc.target/i386/avx512vlbw-pr85832.c: Likewise.
* gcc.target/i386/avx512bw-pr85832.c: Likewise.
* gcc.target/i386/avx512bw-pr86036.c: New test.

From-SVN: r261148

PR c++/85765 - SFINAE and non-type default template arg.

* pt.c (type_unification_real): Do full semantic processing if
substituting a partial args list replaces all template parms.

From-SVN: r261146

tree-cfgcleanup.c (cleanup_tree_cfg_1): Fold into...

2018-06-04  Richard Biener  <rguenther@suse.de>

* tree-cfgcleanup.c (cleanup_tree_cfg_1): Fold into...
(cleanup_tree_cfg_noloop): ... single caller.  Do
start_recording_case_labels later.

From-SVN: r261145

cldemoteintrin.h: Change define from _X86INTRIN_H_INCLUDED to _IMMINTRIN_H_INCLUDED.

2018-06-04  Sebastian Peryt  <sebastian.peryt@intel.com>

        * config/i386/cldemoteintrin.h: Change define from _X86INTRIN_H_INCLUDED
        to _IMMINTRIN_H_INCLUDED.
        * config/i386/pconfigintrin.h: Ditto.
        * config/i386/waitpkgintrin.h: Ditto.
        * config/i386/immintrin.h: Add includes for sgxintrin.h,
        pconfigintrin.h, waitpkgintrin.h and cldemoteintrin.h.
        * config/i386/x86intrin.h: Remove includes for mintrin.h, xmmintrin.h,
        emmintrin.h, pmmintrin.h, tmmintrin.h, smmintrin.h, wmmintrin.h,
        bmiintrin.h, bmi2intrin.h, lzcntintrin.h, sgxintrin.h, pconfigintrin.h,
        waitpkgintrin.h and cldemoteintrin.h.

From-SVN: r261144

[MAINTAINERS] Remove redundant write-after-approval entries

2018-06-04  Tom de Vries  <tdevries@suse.de>

* MAINTAINERS: Remove write-after-approval entries for component
maintainers.

From-SVN: r261143

re PR tree-optimization/86038 (ICE in to_reg_br_prob_base, at profile-count.h:242)

2018-06-04  Richard Biener  <rguenther@suse.de>

PR tree-optimization/86038
* tracer.c (find_best_successor): Check probability for
being initialized, bail out if not.

* gcc.dg/pr86038.c: New testcase.

From-SVN: r261142

[arm] PR target/86003 build failures with --with-cpu=xscale

The XScale cpu configuration in GCC has always been somewhat
non-conforming.  Although XScale isn't an architecture (it's simply an
implementation of ARMv5te), we do by tradition emit a specific
pre-define for it.  We achieve this effect by adding an additional
feature bit to the xscale CPU definition that isn't part of the base
architecture.

When I restructured the options last year I overlooked this oddity and
the result, of course, is that this configuration now fails to build
as intended.

What happens is that the driver (correctly) constructs an architecture
for the xscale cpu name (as armv5te) and passes it in addition to the
CPU name.  The backend code, on finding both a cpu and an architecture
specifies attempts to correlate the two and finds a difference due to
the additional feature bit and reports an inconsistency (fatally if
-werror is specified).

I think the best fix to this is to treat the xscale feature bit using
the same mechanism that we use for other 'quirks' in CPU
implementations and simply filter it out before comparing the
capabilities.  It has the additional benefit that it's also the
simplest fix.

PR target/86003
* config/arm/arm-cpus.in (ALL_QUIRKS): Add xscale feature to the list
of bits to ignore when comparing architectures.

From-SVN: r261140

re PR tree-optimization/69615 (0 to limit signed range checks don't always use unsigned compare)

PR tree-optimization/69615
* fold-const.c (merge_ranges): If range1 is - [x, x] and x is the
maximum or minimum of the type, try to merge it also as if
range1 is + [-, x - 1] or + [x + 1, -].

* gcc.dg/pr69615.c: New test.

From-SVN: r261139

re PR c++/86025 (ICE with -Wduplicated-branches and OpenMP critical)

PR c++/86025
* tree.c (inchash::add_expr): Handle IDENTIFIER_NODE.

* c-c++-common/gomp/pr86025.c: New test.

From-SVN: r261137

Daily bump.

From-SVN: r261136

* gcc.target/powerpc/altivec-7.c: Remove extra brace.

From-SVN: r261130

PR c++/85739 - ICE with pointer to member template parm.

* cvt.c (perform_qualification_conversions): Use cp_fold_convert.

From-SVN: r261129

re PR tree-optimization/86034 (wrong code for bit-field manipulation at -Os)

PR tree-optimization/86034
* gimple-ssa-store-merging.c (output_merged_store): Convert the RHS to
the unsigned bitfield type in a bit insertion sequence if it does not
have a larger precision than the bitfield size.
(process_store): Also bypass widening conversions for BIT_INSERT_EXPR.

From-SVN: r261128

re PR fortran/36497 (USE association, cray pointers and error checking)

2018-06-03  Paul Thomas  <pault@gcc.gnu.org>

PR fortran/36497
* decl.c (variable_decl): Use gfc_add_type for cray pointees.

2018-06-03  Paul Thomas  <pault@gcc.gnu.org>

PR fortran/36497
* gfortran.dg/cray_pointer_12.f90: New test.

From-SVN: r261127

re PR fortran/85938 (Spurious assert failure for matmul with reshaped array)

2018-06-02  Steven G. Kargl  <kargl@gcc.gnu.org>

PR fortran/85938
* gfortran.dg/pr85938.f90: Fixed by revision r261081

From-SVN: r261125

[NDS32] Implement peephole2 patterns for tuning code size.

gcc/
* config/nds32/nds32-peephole2.md: Add new patterns for code size.

From-SVN: r261124

[NDS32] Add -march=v3j.

gcc/
* config/nds32/nds32-opts.h (nds32_arch_type): Add ARCH_V3J.
* config/nds32/nds32.c (nds32_option_override): Consider ARCH_V3J.
* config/nds32/nds32.h (TARGET_ISA_V3): Add ARCH_V3J checking.
* config/nds32/nds32.opt (march): Add enum value Value(ARCH_V3J).

From-SVN: r261123

[NDS32] Disable -fdelete-null-pointer-checks by default for nds32*-*-elf.

gcc/
* common/config/nds32/nds32-common.c (nds32_option_optimization_table):
Disable -fdelete-null-pointer-checks for ELF toolchain.

From-SVN: r261122

PR c++/85761 - ICE with ill-formed use of const outer variable.

* expr.c (mark_use): Handle location wrappers.

From-SVN: r261121

Daily bump.

From-SVN: r261120

[NDS32] Support Linux target for nds32.

gcc/
* config.gcc (nds32*): Use nds32-linux.opt and nds32-elf.opt.
(nds32le-*-*, nds32be-*-*): Integrate checking process.
(nds32*-*-*): Add glibc and uclibc conditions.
* common/config/nds32/nds32-common.c (nds32_except_unwind_info): New.
(TARGET_EXCEPT_UNWIND_INFO): Define.
* config/nds32/elf.h: New file.
* config/nds32/linux.h: New file.
* config/nds32/nds32-elf.opt: New file.
* config/nds32/nds32-linux.opt: New file.
* config/nds32/nds32-fp-as-gp.c
(pass_nds32_fp_as_gp::gate): Consider TARGET_LINUX_ABI.
* config/nds32/nds32.c (nds32_conditional_register_usage): Consider
TARGET_LINUX_ABI.
(nds32_asm_file_end): Ditto.
(nds32_print_operand): Ditto.
(nds32_insert_attributes): Ditto.
(nds32_init_libfuncs): New function.
(TARGET_HAVE_TLS): Define.
(TARGET_INIT_LIBFUNCS): Define.
* config/nds32/nds32.h (TARGET_DEFAULT_RELAX): Apply different relax
spec content.
(TARGET_ELF): Apply different mcmodel setting.
(LINK_SPEC, LIB_SPEC, STARTFILE_SPEC, ENDFILE_SPEC): The content has
been migrated into elf.h and linux.h files.
* config/nds32/nds32.md (add_pc): Consider TARGET_LINUX_ABI.
* config/nds32/nds32.opt (mvh): Consider TARGET_LINUX_ABI.
(mcmodel): The content has been migrated into nds32-elf.opt and
nds32-linux.opt files.
* config/nds32/t-elf: New file.
* config/nds32/t-linux: New file.

libgcc/
* config.host (nds32*-linux*): New.
* config/nds32/linux-atomic.c: New file.
* config/nds32/linux-unwind.h: New file.

Co-Authored-By: Kito Cheng <kito.cheng@gmail.com>
Co-Authored-By: Monk Chiang <sh.chiang04@gmail.com>
From-SVN: r261116

[NDS32] Implement fp-as-gp optimization.

gcc/
* config/nds32/constants.md (unspec_volatile_element): Add
UNSPEC_VOLATILE_OMIT_FP_BEGIN and UNSPEC_VOLATILE_OMIT_FP_END.
* config/nds32/nds32-fp-as-gp.c: New implementation of fp_as_gp
optimization.
* config/nds32/nds32-protos.h (nds32_naked_function_p): Declare.
(make_pass_nds32_fp_as_gp): Declare.
* config/nds32/nds32.c (nds32_register_passes): Add fp_as_gp as one
optmization pass.
(nds32_asm_function_end_prologue): Remove unused asm output.
(nds32_asm_function_begin_epilogue): Remove unused asm output.
(nds32_asm_file_start): Output necessary fp_as_gp information.
(nds32_option_override): Adjust register usage.
(nds32_expand_prologue): Consider fp_as_gp situation.
(nds32_expand_prologue_v3push): Consider fp_as_gp situation.
* config/nds32/nds32.md (prologue): Check fp_as_gp_p and naked_p.
(epilogue): Ditto.
(return): Ditto.
(simple_return): Ditto.
(omit_fp_begin): Output special directive for fp_as_gp.
(omit_fp_end): Output special directive for fp_as_gp.
* config/nds32/nds32.opt (mfp-as-gp, mno-fp-as-gp, mforce-fp-as-gp,
mforbid-fp-as-gp): New options.

Co-Authored-By: Shiva Chen <shiva0217@gmail.com>
From-SVN: r261115

decl.c (gnat_to_gnu_entity): If this is not a definition...

* gcc-interface/decl.c (gnat_to_gnu_entity) <E_Constant>: If this is
not a definition, retrieve the expression only if it's a compile-time
known value if we are just annotating types.

* gcc-interface/utils.c (convert): Do not try to upcast properly for a
conversion between tagged types in type_annotate_only mode.

From-SVN: r261113

Add missing directive

From-SVN: r261109

ada-tree.h (TYPE_PADDING_FOR_COMPONENT): New macro.

* gcc-interface/ada-tree.h (TYPE_PADDING_FOR_COMPONENT): New macro.
* gcc-interface/decl.c (gnat_to_gnu_component_type): Cache the padding
type built for an aliased component with variable size.

From-SVN: r261106

discr53.ad[sb]: New test.

* gnat.dg/discr53.ad[sb]: New test.
* gnat.dg/discr53_pkg.ads: New helper.

From-SVN: r261102

PR c++/85764 - bogus 'this' not captured error.

* lambda.c (resolvable_dummy_lambda): Use nonlambda_method_basetype.
(nonlambda_method_basetype): Handle NSDMI.

From-SVN: r261101

* g++.dg/cpp1z/feat-cxx1z.C: Update __cpp_deduction_guides.

From-SVN: r261100

Commit 260294 on 2018-05-16 by Carl Love was supposed to add the following files.

gcc/testsuite/ChangeLog:

2018-06-01  Carl Love  <cel@us.ibm.com>

Commit 260294 on 2018-05-16 by Carl Love was supposed to add the
following files.

* gcc.target/powerpc/vsx-vector-6-be.p7.c: New test file.
* gcc.target/powerpc/vsx-vector-6-be.p8.c: New test file.

From-SVN: r261098

altivec-12.c (main): Fix declaration of ucz to make it consistent with the naming convention in the file.

gcc/testsuite/ChangeLog:

2018-06-01  Carl Love  <cel@us.ibm.com>

* gcc.target/powerpc/altivec-12.c (main): Fix declaration of ucz
to make it consistent with the naming convention in the file.
* gcc.target/powerpc/altivec-7-be.c: Move BE specific checks
to altivec-7.c.  Delete file.
* gcc.target/powerpc/altivec-7-le.c: Move LE specific checks
to altivec-7.c.  Delete file.
* gcc.target/powerpc/altivec-7.h: Move to altivec-7.c.
* gcc.target/powerpc/altivec-7.c (main): Add vec_unpackh and
vec_unpackl tests.  Update instruction counts.
* gcc.target/powerpc/builtins-1-le.c: Move LE specific checks to
tests to builtins-1.c.
* gcc.target/powerpc/builtins-1-be.c: Move BE specific
tests to builtins-1.c.
* gcc.target/powerpc/builtins-1.h: Move to file builtins-1.c.
* gcc.target/powerpc/builtins-1.c (main): Add test case for vec_and.
vec_round, vec_rsqrt, vec_rsqrte, vec_mergee, vec_mergh, vec_mergo.
Remove vec_ctf tests returning double.  Remove vec_cts with
double args. Remove vec_sel with invalid arguments. Add tests for
vec_splat.
Add instruction counts for new tests.
* gcc.target/powerpc/builtins-3-runnable.c (main): Add test for
vec_doublee, vec_doubleo, vec_doublel, vec_doubleh, vec_signed,
vec_unsigned.
* gcc.target/powerpc/builtins-3.c: Add tests test_sll_vuill_vuill_vuc,
test_sll_vsill_vsill_vuc.
* gcc.target/powerpc/p9-xxbr-2.c (rev_bool_long_long): Added test for
vec_revb.
* gcc.target/powerpc/vsx-7.h: Rename to vsx-7.c. Remove redundant
tests from altivec-7.h.
* gcc.target/powerpc/vsx-7-be.c: Remove file.
* gcc.target/powerpc/vsx-builtin-7.c: Add test functions splat_sc_s8,
splat_uc_u8, splat_ssi_s16, splat_usi_s16, splat_si_s32, splat_ui_u32,
splat_sll, splat_uc, splat_int128, splat_uint128.
Make second argument of vec_extract and vec_insert a signed int.
* gcc.target/powerpc/vsx-vector-5.c (vrint): Add vrint test for float
argument.

From-SVN: r261097

Daily bump.

From-SVN: r261096

DWARF5: Don't generate DW_AT_loclists_base for split compile unit DIEs.

The loclists_base attribute is used to point to the beginning of the
loclists index of a DWARF5 loclists table when using DW_FORM_loclistsx.
For split compile units the base is not given by the attribute, but is
either the first (and only) index in the .debug_loclists.dwo section,
or (when placed in a .dwp file) given by the DW_SECT_LOCLISTS row in
the .debug_cu_index section.

The loclists_base attribute is only valid for the full (or skeleton)
compile unit DIE in the main (relocatable) object. But GCC only ever
generates a loclists table index for the .debug_loclists section put
into the split DWARF .dwo file.

For split compile unit DIEs it is confusing (and not according to spec)
to also have a DW_AT_loclists_base attribute (which might be wrong,
since its relocatable offset won't actually be relocated).

gcc/ChangeLog

* dwarf2out.c (dwarf2out_finish): Remove generation of
DW_AT_loclists_base.

From-SVN: r261092

* c-cppbuiltin.c (c_cpp_builtins): Bump __cpp_deduction_guides to 201703.

From-SVN: r261090

gimple-ssa-store-merging.c: Include gimple-fold.h.

* gimple-ssa-store-merging.c: Include gimple-fold.h.
(struct store_immediate_info): Document BIT_INSERT_EXPR stores.
(struct merged_store_group): Add bit_insertion field.
(dump_char_array): Use standard hexadecimal format.
(merged_store_group::merged_store_group): Set bit_insertion to false.
(merged_store_group::apply_stores): Use optimal buffer size.  Deal
with BIT_INSERT_EXPR stores.  Move up code updating the mask and
also print the mask in the dump file.
(pass_store_merging::gate): Minor tweak.
(imm_store_chain_info::coalesce_immediate): Fix wrong association
of stores with groups in dump.  Allow coalescing of BIT_INSERT_EXPR
stores with INTEGER_CST stores.
(count_multiple_uses) <BIT_INSERT_EXPR>: New case.
(imm_store_chain_info::output_merged_store): Add try_bitpos variable
and use it throughout.  Generate bit insertion sequences if need be.
(pass_store_merging::process_store): Remove redundant condition.
Record stores from a SSA name to a bit-field with BIT_INSERT_EXPR.

From-SVN: r261089

altivec-35.c (foo): Add builtin test vec_madds.

gcc/testsuite/ChangeLog:

2018-06-01  Carl Love  <cel@us.ibm.com>
* gcc.target/powerpc/altivec-35.c (foo): Add builtin test vec_madds.
* gcc.target/powerpc/builtins-6-runnable.c (main): Fix typo for output.
Add vec_xst_be for signed and unsigned arguments.

From-SVN: r261088

ChangeLog tidying

From-SVN: r261087

CWG 1581: When are constexpr member functions defined?

* constexpr.c (instantiate_cx_fn_r, instantiate_constexpr_fns): New.
(cxx_eval_outermost_constant_expr): Call instantiate_constexpr_fns.

From-SVN: r261086

PR c++/58281 - explicit instantiation of constexpr

* pt.c (mark_decl_instantiated): Clear DECL_EXTERNAL.

From-SVN: r261085

* pt.c (instantiate_decl): Any defaulted function is defined.

From-SVN: r261084

MAINTAINERS (write after approval): Add myself.

  * MAINTAINERS (write after approval): Add myself.

From-SVN: r261082

re PR fortran/85816 (nested spread fails with "Integer overflow in xmallocarray")

2018-06-01  Steven G. Kargl  <kargl@gcc.gnu.org>

PR fortran/85816
PR fortran/85975
* libgfortran.h: Remove the GFC_DTYPE_COPY_SETRANK macro.
* intrinsics/reshape_generic.c: Directly assign rank.
* intrinsics/spread_generic.c: Ditto.
* m4/iforeach-s.m4: Ditto.
* m4/iforeach.m4: Ditto.
* m4/ifunction-s.m4: Ditto.
* m4/ifunction-s2.m4: Ditto.
* m4/ifunction.m4: Ditto.
* m4/ifunction_logical.m4: Ditto.
* m4/reshape.m4: Ditto.
* m4/spread.m4: Ditto.
* generated/all_l1.c: Regenerated from m4 files.
* generated/all_l16.c: Ditto.
* generated/all_l2.c: Ditto.
* generated/all_l4.c: Ditto.
* generated/all_l8.c: Ditto.
* generated/any_l1.c: Ditto.
* generated/any_l16.c: Ditto.
* generated/any_l2.c: Ditto.
* generated/any_l4.c: Ditto.
* generated/any_l8.c: Ditto.
* generated/count_16_l.c: Ditto.
* generated/count_1_l.c: Ditto.
* generated/count_2_l.c: Ditto.
* generated/count_4_l.c: Ditto.
* generated/count_8_l.c: Ditto.
* generated/iall_i1.c: Ditto.
* generated/iall_i16.c: Ditto.
* generated/iall_i2.c: Ditto.
* generated/iall_i4.c: Ditto.
* generated/iall_i8.c: Ditto.
* generated/iany_i1.c: Ditto.
* generated/iany_i16.c: Ditto.
* generated/iany_i2.c: Ditto.
* generated/iany_i4.c: Ditto.
* generated/iany_i8.c: Ditto.
* generated/iparity_i1.c: Ditto.
* generated/iparity_i16.c: Ditto.
* generated/iparity_i2.c: Ditto.
* generated/iparity_i4.c: Ditto.
* generated/iparity_i8.c: Ditto.
* generated/maxloc0_16_i1.c: Ditto.
* generated/maxloc0_16_i16.c: Ditto.
* generated/maxloc0_16_i2.c: Ditto.
* generated/maxloc0_16_i4.c: Ditto.
* generated/maxloc0_16_i8.c: Ditto.
* generated/maxloc0_16_r10.c: Ditto.
* generated/maxloc0_16_r16.c: Ditto.
* generated/maxloc0_16_r4.c: Ditto.
* generated/maxloc0_16_r8.c: Ditto.
* generated/maxloc0_16_s1.c: Ditto.
* generated/maxloc0_16_s4.c: Ditto.
* generated/maxloc0_4_i1.c: Ditto.
* generated/maxloc0_4_i16.c: Ditto.
* generated/maxloc0_4_i2.c: Ditto.
* generated/maxloc0_4_i4.c: Ditto.
* generated/maxloc0_4_i8.c: Ditto.
* generated/maxloc0_4_r10.c: Ditto.
* generated/maxloc0_4_r16.c: Ditto.
* generated/maxloc0_4_r4.c: Ditto.
* generated/maxloc0_4_r8.c: Ditto.
* generated/maxloc0_4_s1.c: Ditto.
* generated/maxloc0_4_s4.c: Ditto.
* generated/maxloc0_8_i1.c: Ditto.
* generated/maxloc0_8_i16.c: Ditto.
* generated/maxloc0_8_i2.c: Ditto.
* generated/maxloc0_8_i4.c: Ditto.
* generated/maxloc0_8_i8.c: Ditto.
* generated/maxloc0_8_r10.c: Ditto.
* generated/maxloc0_8_r16.c: Ditto.
* generated/maxloc0_8_r4.c: Ditto.
* generated/maxloc0_8_r8.c: Ditto.
* generated/maxloc0_8_s1.c: Ditto.
* generated/maxloc0_8_s4.c: Ditto.
* generated/maxloc1_16_i1.c: Ditto.
* generated/maxloc1_16_i16.c: Ditto.
* generated/maxloc1_16_i2.c: Ditto.
* generated/maxloc1_16_i4.c: Ditto.
* generated/maxloc1_16_i8.c: Ditto.
* generated/maxloc1_16_r10.c: Ditto.
* generated/maxloc1_16_r16.c: Ditto.
* generated/maxloc1_16_r4.c: Ditto.
* generated/maxloc1_16_r8.c: Ditto.
* generated/maxloc1_16_s1.c: Ditto.
* generated/maxloc1_16_s4.c: Ditto.
* generated/maxloc1_4_i1.c: Ditto.
* generated/maxloc1_4_i16.c: Ditto.
* generated/maxloc1_4_i2.c: Ditto.
* generated/maxloc1_4_i4.c: Ditto.
* generated/maxloc1_4_i8.c: Ditto.
* generated/maxloc1_4_r10.c: Ditto.
* generated/maxloc1_4_r16.c: Ditto.
* generated/maxloc1_4_r4.c: Ditto.
* generated/maxloc1_4_r8.c: Ditto.
* generated/maxloc1_4_s1.c: Ditto.
* generated/maxloc1_4_s4.c: Ditto.
* generated/maxloc1_8_i1.c: Ditto.
* generated/maxloc1_8_i16.c: Ditto.
* generated/maxloc1_8_i2.c: Ditto.
* generated/maxloc1_8_i4.c: Ditto.
* generated/maxloc1_8_i8.c: Ditto.
* generated/maxloc1_8_r10.c: Ditto.
* generated/maxloc1_8_r16.c: Ditto.
* generated/maxloc1_8_r4.c: Ditto.
* generated/maxloc1_8_r8.c: Ditto.
* generated/maxloc1_8_s1.c: Ditto.
* generated/maxloc1_8_s4.c: Ditto.
* generated/maxval1_s1.c: Ditto.
* generated/maxval1_s4.c: Ditto.
* generated/maxval_i1.c: Ditto.
* generated/maxval_i16.c: Ditto.
* generated/maxval_i2.c: Ditto.
* generated/maxval_i4.c: Ditto.
* generated/maxval_i8.c: Ditto.
* generated/maxval_r10.c: Ditto.
* generated/maxval_r16.c: Ditto.
* generated/maxval_r4.c: Ditto.
* generated/maxval_r8.c: Ditto.
* generated/minloc0_16_i1.c: Ditto.
* generated/minloc0_16_i16.c: Ditto.
* generated/minloc0_16_i2.c: Ditto.
* generated/minloc0_16_i4.c: Ditto.
* generated/minloc0_16_i8.c: Ditto.
* generated/minloc0_16_r10.c: Ditto.
* generated/minloc0_16_r16.c: Ditto.
* generated/minloc0_16_r4.c: Ditto.
* generated/minloc0_16_r8.c: Ditto.
* generated/minloc0_16_s1.c: Ditto.
* generated/minloc0_16_s4.c: Ditto.
* generated/minloc0_4_i1.c: Ditto.
* generated/minloc0_4_i16.c: Ditto.
* generated/minloc0_4_i2.c: Ditto.
* generated/minloc0_4_i4.c: Ditto.
* generated/minloc0_4_i8.c: Ditto.
* generated/minloc0_4_r10.c: Ditto.
* generated/minloc0_4_r16.c: Ditto.
* generated/minloc0_4_r4.c: Ditto.
* generated/minloc0_4_r8.c: Ditto.
* generated/minloc0_4_s1.c: Ditto.
* generated/minloc0_4_s4.c: Ditto.
* generated/minloc0_8_i1.c: Ditto.
* generated/minloc0_8_i16.c: Ditto.
* generated/minloc0_8_i2.c: Ditto.
* generated/minloc0_8_i4.c: Ditto.
* generated/minloc0_8_i8.c: Ditto.
* generated/minloc0_8_r10.c: Ditto.
* generated/minloc0_8_r16.c: Ditto.
* generated/minloc0_8_r4.c: Ditto.
* generated/minloc0_8_r8.c: Ditto.
* generated/minloc0_8_s1.c: Ditto.
* generated/minloc0_8_s4.c: Ditto.
* generated/minloc1_16_i1.c: Ditto.
* generated/minloc1_16_i16.c: Ditto.
* generated/minloc1_16_i2.c: Ditto.
* generated/minloc1_16_i4.c: Ditto.
* generated/minloc1_16_i8.c: Ditto.
* generated/minloc1_16_r10.c: Ditto.
* generated/minloc1_16_r16.c: Ditto.
* generated/minloc1_16_r4.c: Ditto.
* generated/minloc1_16_r8.c: Ditto.
* generated/minloc1_16_s1.c: Ditto.
* generated/minloc1_16_s4.c: Ditto.
* generated/minloc1_4_i1.c: Ditto.
* generated/minloc1_4_i16.c: Ditto.
* generated/minloc1_4_i2.c: Ditto.
* generated/minloc1_4_i4.c: Ditto.
* generated/minloc1_4_i8.c: Ditto.
* generated/minloc1_4_r10.c: Ditto.
* generated/minloc1_4_r16.c: Ditto.
* generated/minloc1_4_r4.c: Ditto.
* generated/minloc1_4_r8.c: Ditto.
* generated/minloc1_4_s1.c: Ditto.
* generated/minloc1_4_s4.c: Ditto.
* generated/minloc1_8_i1.c: Ditto.
* generated/minloc1_8_i16.c: Ditto.
* generated/minloc1_8_i2.c: Ditto.
* generated/minloc1_8_i4.c: Ditto.
* generated/minloc1_8_i8.c: Ditto.
* generated/minloc1_8_r10.c: Ditto.
* generated/minloc1_8_r16.c: Ditto.
* generated/minloc1_8_r4.c: Ditto.
* generated/minloc1_8_r8.c: Ditto.
* generated/minloc1_8_s1.c: Ditto.
* generated/minloc1_8_s4.c: Ditto.
* generated/minval1_s1.c: Ditto.
* generated/minval1_s4.c: Ditto.
* generated/minval_i1.c: Ditto.
* generated/minval_i16.c: Ditto.
* generated/minval_i2.c: Ditto.
* generated/minval_i4.c: Ditto.
* generated/minval_i8.c: Ditto.
* generated/minval_r10.c: Ditto.
* generated/minval_r16.c: Ditto.
* generated/minval_r4.c: Ditto.
* generated/minval_r8.c: Ditto.
* generated/norm2_r10.c: Ditto.
* generated/norm2_r16.c: Ditto.
* generated/norm2_r4.c: Ditto.
* generated/norm2_r8.c: Ditto.
* generated/parity_l1.c: Ditto.
* generated/parity_l16.c: Ditto.
* generated/parity_l2.c: Ditto.
* generated/parity_l4.c: Ditto.
* generated/parity_l8.c: Ditto.
* generated/product_c10.c: Ditto.
* generated/product_c16.c: Ditto.
* generated/product_c4.c: Ditto.
* generated/product_c8.c: Ditto.
* generated/product_i1.c: Ditto.
* generated/product_i16.c: Ditto.
* generated/product_i2.c: Ditto.
* generated/product_i4.c: Ditto.
* generated/product_i8.c: Ditto.
* generated/product_r10.c: Ditto.
* generated/product_r16.c: Ditto.
* generated/product_r4.c: Ditto.
* generated/product_r8.c: Ditto.
* generated/reshape_c10.c: Ditto.
* generated/reshape_c16.c: Ditto.
* generated/reshape_c4.c: Ditto.
* generated/reshape_c8.c: Ditto.
* generated/reshape_i16.c: Ditto.
* generated/reshape_i4.c: Ditto.
* generated/reshape_i8.c: Ditto.
* generated/reshape_r10.c: Ditto.
* generated/reshape_r16.c: Ditto.
* generated/reshape_r4.c: Ditto.
* generated/reshape_r8.c: Ditto.
* generated/spread_c10.c: Ditto.
* generated/spread_c16.c: Ditto.
* generated/spread_c4.c: Ditto.
* generated/spread_c8.c: Ditto.
* generated/spread_i1.c: Ditto.
* generated/spread_i16.c: Ditto.
* generated/spread_i2.c: Ditto.
* generated/spread_i4.c: Ditto.
* generated/spread_i8.c: Ditto.
* generated/spread_r10.c: Ditto.
* generated/spread_r16.c: Ditto.
* generated/spread_r4.c: Ditto.
* generated/spread_r8.c: Ditto.
* generated/sum_c10.c: Ditto.
* generated/sum_c16.c: Ditto.
* generated/sum_c4.c: Ditto.
* generated/sum_c8.c: Ditto.
* generated/sum_i1.c: Ditto.
* generated/sum_i16.c: Ditto.
* generated/sum_i2.c: Ditto.
* generated/sum_i4.c: Ditto.
* generated/sum_i8.c: Ditto.
* generated/sum_r10.c: Ditto.
* generated/sum_r16.c: Ditto.
* generated/sum_r4.c: Ditto.
* generated/sum_r8.c: Ditto.

2018-06-01  Steven G. Kargl  <kargl@gcc.gnu.org>

PR fortran/85816
PR fortran/85975
* gfortran.dg/pr85816.f90: New test.
* gfortran.dg/pr85975.f90: New test.

From-SVN: r261081

* Fix email address in ChangeLog.

From-SVN: r261080

* MAINTAINERS: Update my email address.

From-SVN: r261079

rs6000: Fix mangling for 128-bit float

This patch changes the (C++) mangling of the 128-bit float types.
IBM long double ("double-double") is mangled as "g", as before, and
IEEE 128-bit long double is mangled as "u9__ieee128".

* config/rs6000/rs6000.c (rs6000_mangle_type): Change the mangling of
the 128-bit floating point types.  Fix function comment.

From-SVN: r261078

re PR fortran/63570 ([F2018] Implement 13.7.137 RANDOM INIT (REPEATABLE, IMAGE DISTINCT))

2018-06-01  Steven G. Kargl  <kargl@gcc.gnu.org>

PR fortran/63570
* check.c (gfc_check_random_init): New function. Check arguments of
RANDOM_INIT.
* gfortran.h (GFC_ISYM_RANDOM_INIT): New enum token.
* intrinsic.c (add_subroutines): Add RANDOM_INIT to list of
subroutines.
(gfc_check_intrinsic_standard): Introduce Fortran 2018 check.
* intrinsic.h: Add prototypes for gfc_check_random_init and
gfc_resolve_random_init
* intrinsic.texi: Document new intrinsic subprogram.
* iresolve.c (gfc_resolve_random_init): Resolve routine name.
* trans-decl.c: Declare gfor_fndecl_random_init
* trans-intrinsic.c (conv_intrinsic_random_init): New function.
Translate call to RANDOM_INIT.
(gfc_conv_intrinsic_subroutine): Call it.
  * trans.h: Declare gfor_fndecl_random_init

2018-06-01  Steven G. Kargl  <kargl@gcc.gnu.org>

PR fortran/63570
* gfortran.dg/random_init_1.f90: New test.
* gfortran.dg/random_init_2.f90: New test.
* gfortran.dg/random_init_3.f90: New test.
* gfortran.dg/random_init_4.f90: New test.
* gfortran.dg/random_init_5.f90: New test.
* gfortran.dg/random_init_6.f90: New test.

2018-06-01  Steven G. Kargl  <kargl@gcc.gnu.org>

PR fortran/63570
* libgfortran/Makefile.am: Add random_init.f90 to build.
* libgfortran/Makefile.in: Regenerated.
* libgfortran/gfortran.map: Expose symbol for _gfortran_random_init.
* libgfortran/intrinsics/random_init.f90: Implementation.

From-SVN: r261075

[AArch64] Used prefer aliases SXTL(2) and UXTL(2)

The SSHL and USHL instructions are used with a shift operand of zero to
sign and zero-extend integer vectors into wider modes.
GCC makes extensive use of them to "unpack" vectors.

AArch64 defines a shorthand alias for that case. Instead of writing:
SSHLL <Vd>.<Ta>, <Vn>.<Tb>, 0

we can write

SXTL <Vd>.<Ta>, <Vn>.<Tb>

Similar for the unsigned versions and the high-part versions (SSHL2 -> SXTL2).
This makes the assembly of vectorised functions a bit more readable.

* config/aarch64/aarch64-simd.md
(aarch64_simd_vec_unpack<su>_lo_<mode>): Use UXTL and SXTL assembler
mnemonics.
(aarch64_simd_vec_unpack<su>_hi_<mode>): Use UXTL2 and SXTL2 assembler
mnemonics.

From-SVN: r261073

[MAINTAINERS] Update email address

2018-06-01  Tom de Vries  <tdevries@suse.de>

* MAINTAINERS: Update my email address.

From-SVN: r261068

Fix phi backedge detection in backprop (PR85989)

This PR is a nasty wrong code bug due to my fluffing a test for a
backedge in gimple-ssa-backprop.c.  Backedges are supposed to be
from definitions in the statement we're visiting to uses in statements
that we haven't visited yet.  However, the check failed to account for
PHIs in the current block that had already been processed, so if two
PHIs in the same block referenced each other, we'd treat both
references as backedges.

In more detail:

The first phase of the pass goes through all statements in an arbitrary
order, making optimistic assumptions about any statements that haven't
been visited yet.  The second phase then calculates a correct
(supposedly maximal) fixed point.

Although the first phase order is arbitrary in principle, we still use
the CFG rpo to cut down on the backedges.  This means that the only
thing that's truly arbitrary is the order that we process the PHIs
in a block.  Any order should be OK and should eventually give the
same results.  But we have to follow whatever order we pick,
and the pass wasn't doing that.

2018-06-01  Richard Sandiford  <richard.sandiford@linaro.org>

gcc/
PR tree-optimization/85989
* gimple-ssa-backprop.c (backprop::m_visited_phis): New member
variable.
(backprop::intersect_uses): Check it when deciding whether this
is a backedge reference.
(backprop::process_block): Add each phi to m_visited_phis
after visiting it, then clear it at the end.

gcc/testsuite/
PR tree-optimization/85989
* gcc.dg/torture/pr85989.c: New test.

From-SVN: r261064

tree-vectorizer.h (vect_dr_stmt): New function.

2018-06-01  Richard Biener  <rguenther@suse.de>

* tree-vectorizer.h (vect_dr_stmt): New function.
(vect_get_load_cost): Adjust.
(vect_get_store_cost): Likewise.
* tree-vect-data-refs.c (vect_analyze_data_ref_dependence):
Use vect_dr_stmt instead of DR_SMTT.
(vect_record_base_alignments): Likewise.
(vect_calculate_target_alignment): Likewise.
(vect_compute_data_ref_alignment): Likewise and make static.
(vect_update_misalignment_for_peel): Likewise.
(vect_verify_datarefs_alignment): Likewise.
(vector_alignment_reachable_p): Likewise.
(vect_get_data_access_cost): Likewise.  Pass down
vinfo to vect_get_load_cost/vect_get_store_cost instead of DR.
(vect_get_peeling_costs_all_drs): Likewise.
(vect_peeling_hash_get_lowest_cost): Likewise.
(vect_enhance_data_refs_alignment): Likewise.
(vect_find_same_alignment_drs): Likewise.
(vect_analyze_data_refs_alignment): Likewise.
(vect_analyze_group_access_1): Likewise.
(vect_analyze_group_access): Likewise.
(vect_analyze_data_ref_access): Likewise.
(vect_analyze_data_ref_accesses): Likewise.
(vect_vfa_segment_size): Likewise.
(vect_small_gap_p): Likewise.
(vectorizable_with_step_bound_p): Likewise.
(vect_prune_runtime_alias_test_list): Likewise.
(vect_analyze_data_refs): Likewise.
(vect_supportable_dr_alignment): Likewise.
* tree-vect-loop-manip.c (get_misalign_in_elems): Likewise.
(vect_gen_prolog_loop_niters): Likewise.
* tree-vect-loop.c (vect_analyze_loop_2): Likewise.
* tree-vect-patterns.c (vect_recog_bool_pattern): Do not
modify DR_STMT.
(vect_recog_mask_conversion_pattern): Likewise.
(vect_try_gather_scatter_pattern): Likewise.
* tree-vect-stmts.c (vect_model_store_cost): Pass stmt_info
to vect_get_store_cost.
(vect_get_store_cost): Get stmt_info instead of DR.
(vect_model_load_cost): Pass stmt_info to vect_get_load_cost.
(vect_get_load_cost): Get stmt_info instead of DR.

From-SVN: r261062

re PR tree-optimization/86017 (multiple consecutive calls to bzero/memset not merged)

2018-06-01  Richard Biener  <rguenther@suse.de>

PR middle-end/86017
* gimple-fold.c (var_decl_component_p): Also allow offsetted
vars wrapped in MEM_REFs.

* gcc.dg/tree-ssa/pr86017.c: New testcase.

From-SVN: r261061

c-ada-spec.c (dump_ada_declaration): Generate a forward declaration for a typedef independently of whether the...

* c-ada-spec.c (dump_ada_declaration) <TYPE_DECL>: Generate a forward
declaration for a typedef independently of whether the declaration of
the subtype is generated.

From-SVN: r261058

Fix SVE fallout from r260951

2018-06-01  Richard Sandiford  <richard.sandiford@linaro.org>

gcc/
* config/aarch64/aarch64.c (aarch64_ira_change_pseudo_allocno_class):
Fix subreg tests so that we only return a choice between
GENERAL_REGS and FP_REGS if the original classes included both.

From-SVN: r261057

re PR ipa/85960 (-fipa-pta and ifunc are incompatible)

2018-06-01  Richard Biener  <rguenther@suse.de>

PR ipa/85960
* tree-ssa-structalias.c (get_function_part_constraint):
Handle NULL fi->decl.
(find_func_aliases_for_call): Properly handle indirect
fi from direct call.
(find_func_clobbers): Likewise.
(ipa_pta_execute): Likewise.
(create_variable_info_for): For functions that are ifunc_resolver
resolve to a varinfo that contains the result of the resolver
call.
(associate_varinfo_to_alias): Do not treat ifunc resolvers as
aliases.

* gcc.dg/ipa/ipa-pta-19.c: New testcase.

From-SVN: r261056

2018-05-15  Michael Collison  <michael.collison@arm.com>

* config/aarch64/aarch64.md:
(*fix_to_zero_extenddfdi2): New pattern.
* gcc.target/aarch64/fix_extend1.c: New testcase.

From-SVN: r261051

Daily bump.

From-SVN: r261050

libgo: update to Go 1.10.2 release

    Reviewed-on: https://go-review.googlesource.com/115196

From-SVN: r261041

2nd Patch for PR78009
Patch for PR83026

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=78809
Inline strcmp with small constant strings

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=83026
missing strlen optimization for strcmp of unequal strings

The design doc for PR78809 is at:
https://www.mail-archive.com/gcc@gcc.gnu.org/msg83822.html

this patch is for the second part of change of PR78809 and PR83026:

B. for strncmp (s1, s2, n) (!)= 0 or strcmp (s1, s2) (!)= 0

   B.1. (PR83026) When the lengths of both arguments are constant and
        it's a strcmp:
      * if the lengths are NOT equal, we can safely fold the call
        to a non-zero value.
      * otherwise, do nothing now.

   B.2. (PR78809) When the length of one argument is constant, try to replace
   the call with a __builtin_str(n)cmp_eq call where possible, i.e:

   strncmp (s, STR, C) (!)= 0 in which, s is a pointer to a string, STR is a
   string with constant length, C is a constant.
     if (C <= strlen(STR) && sizeof_array(s) > C)
       {
         replace this call with
         __builtin_strncmp_eq (s, STR, C) (!)= 0
       }
     if (C > strlen(STR)
       {
         it can be safely treated as a call to strcmp (s, STR) (!)= 0
         can handled by the following strcmp.
       }

   strcmp (s, STR) (!)= 0 in which, s is a pointer to a string, STR is a
   string with constant length.
     if  (sizeof_array(s) > strlen(STR))
       {
         replace this call with
         __builtin_strcmp_eq (s, STR, strlen(STR)+1) (!)= 0
       }

   later when expanding the new __builtin_str(n)cmp_eq calls, first expand them
   as __builtin_memcmp_eq, if the expansion does not succeed, change them back
   to call to __builtin_str(n)cmp.

adding test case strcmpopt_2.c and strcmpopt_4.c into gcc.dg for part B of
PR78809 adding test case strcmpopt_3.c into gcc.dg for PR83026

From-SVN: r261039

re PR target/85984 (ICE in create_pseudo_cfg, at dwarf2cfi.c:2874)

PR target/85984
* bb-reorder.c (pass_partition_blocks::gate): Return false for
functions with naked attribute.

* gcc.target/i386/pr85984.c: New test.

From-SVN: r261037

re PR target/85591 (__builtin_cpu_is() is not detecting bdver2 with Model = 0x02)

PR target/85591
* config/i386/cpuinfo.c (get_amd_cpu): Return
AMDFAM15H_BDVER2 for AMDFAM15H model 0x2.

From-SVN: r261036

PR libstdc++/78870 support std::filesystem on Windows

PR libstdc++/78870 support std::filesystem on Windows
* config.h.in: Regenerate.
* configure: Regenerate.
* configure.ac: Check for link, readlink and symlink.
* include/bits/fs_path.h (path::operator/=(const path&)): Move
definition out of class body.
(path::is_absolute(), path::_M_append(path)): Likewise.
(operator<<(basic_ostream, const path&)): Use std::quoted directly.
(operator>>(basic_istream, path&)): Likewise.
(u8path): Reorder definitions and fix Windows implementation.
(path::is_absolute()): Define inline and fix for Windows.
[!_GLIBCXX_FILESYSTEM_IS_WINDOWS] (path::operator/=(const path&)):
Define POSIX version inline.
(path::_M_append(path)): Define inline.
* include/experimental/bits/fs_path.h (path::is_absolute()): Move
definition out of class body.
(operator<<(basic_ostream, const path&)): Fix type of delimiter and
escape characters.
(operator>>(basic_istream, path&)): Likewise.
(path::is_absolute()): Define inline and fix for Windows.
* src/filesystem/dir-common.h (__gnu_posix): New namespace.
(__gnu_posix::char_type, __gnu_posix::DIR, __gnu_posix::dirent)
(__gnu_posix::opendir, __gnu_posix::readdir, __gnu_posix::closedir):
Define as adaptors for Windows functions/types or as
using-declarations for POSIX functions/types.
(_Dir_base, get_file_type): Qualify names to use declarations from
__gnu_posix namespace.
(_Dir_base::is_dor_or_dotdot): New helper functions.
* src/filesystem/dir.cc (_Dir, recursive_directory_iterator): Qualify
names to use declarations from __gnu_posix namespace.
* src/filesystem/ops-common.h (__gnu_posix): New nested namespace.
(__gnu_posix::open, __gnu_posix::close, __gnu_posix::stat_type)
(__gnu_posix::stat, __gnu_posix::lstat, __gnu_posix::mode_t)
(__gnu_posix::chmod, __gnu_posix::mkdir, __gnu_posix::getcwd)
(__gnu_posix::chdir, __gnu_posix::utimbuf, __gnu_posix::utime)
(__gnu_posix::rename, __gnu_posix::truncate, __gnu_posix::char_type):
Define as adaptors for Windows functions/types or as
using-declarations for POSIX functions/types.
(stat_type, do_copy_file): Qualify names to use declarations from
__gnu_posix namespace.
(do_space): Declare new function.
(make_file_type): Only use S_ISLNK if defined.
* src/filesystem/ops.cc (char_ptr, filesystem::canonical): Use
path::value_type not char.
(filesystem::copy, create_dir, filesystem::create_directory): Qualify
names to use declarations from __gnu_posix namespace.
(filesystem::create_hard_link): Check HAVE_LINK autoconf macro and
add implementation for Windows.
(filesystem::create_symlink): Check HAVE_SYMLINK autoconf macro.
(filesystem::current_path(error_code&)): Use __gnu_posix::getcwd.
[!_PC_PATH_MAX]: Don't use pathconf.
[PATH_MAX]: Use if defined.
(filesystem::current_path(const path&, error_code&))
(filesystem::equivalent, do_stat, filesystem::hard_link_count)
(filesystem::last_write_time, filesystem::permissions): Use names
from __gnu_posix.
(filesystem::read_symlink): Check HAVE_READLINK autoconf macro.
(filesystem::remove) [_GLIBCXX_FILESYSTEM_IS_WINDOWS]: Add
implementation for Windows.
(filesystem::rename, filesystem::resize_file): Use names from
__gnu_posix.
(filesystem::space): Use do_space.
[_GLIBCXX_FILESYSTEM_IS_WINDOWS]: Get absolute path to directory.
(filesystem::status, filesystem::symlink_status): Use names from
__gnu_posix.
(filesystem::temp_directory_path): Add implementation for Windows.
* src/filesystem/path.cc (dot): Define constant.
(path::replace_extension): Use dot.
(path::_M_find_extension): Likewise. Use path::string_type not
std::string.
(path::_M_split_cmpts): Use dot.
(filesystem_error::_M_get_what): Use u8string() not native().
* src/filesystem/std-dir.cc (_Dir, recursive_directory_iterator):
Qualify names to use declarations from __gnu_posix namespace.
* src/filesystem/std-ops.cc (filesystem::absolute(const path&)): Use
correct error_code.
(filesystem::absolute(const path&, error_code&)): Add implementation
for Windows.
(char_ptr, filesystem::canonical): Use path::value_type not char.
(do_copy_file): Use names from __gnu_posix.
[_GLIBCXX_FILESYSTEM_IS_WINDOWS]: Do not use fchmod, fchmodat or
sendfile.
(filesystem::copy, create_dir, filesystem::create_directory): Qualify
names to use declarations from __gnu_posix namespace.
(filesystem::create_hard_link): Check HAVE_LINK autoconf macro and
add implementation for Windows.
(filesystem::create_symlink): Check HAVE_SYMLINK autoconf macro.
(filesystem::current_path(error_code&)): Use __gnu_posix::getcwd.
[!_PC_PATH_MAX]: Don't use pathconf.
[PATH_MAX]: Use if defined.
(filesystem::current_path(const path&, error_code&))
(filesystem::equivalent, do_stat, filesystem::hard_link_count)
(filesystem::last_write_time, filesystem::permissions): Use names
from __gnu_posix.
(filesystem::read_symlink): Check HAVE_READLINK autoconf macro.
(filesystem::remove) [_GLIBCXX_FILESYSTEM_IS_WINDOWS]: Add
implementation for Windows.
(filesystem::rename, filesystem::resize_file): Use names from
__gnu_posix.
(do_space): Define.
(filesystem::space): Use do_space.
(filesystem::status, filesystem::symlink_status): Use names from
__gnu_posix.
(filesystem::temp_directory_path): Add implementation for Windows.
* src/filesystem/std-path.cc
[_GLIBCXX_FILESYSTEM_IS_WINDOWS] (path::operator/=(const path&)):
Define for Windows.
(dot): Define constant.
(path::replace_extension, is_dot): Use dot.
(path::lexically_normal): Check _M_type instead of calling
non-existent function.
(path::_M_find_extension): Use dot. Use path::string_type not
std::string.
(path::_M_split_cmpts): Use dot.
(filesystem_error::_M_get_what): Use u8string() not native().
* testsuite/27_io/filesystem/iterators/directory_iterator.cc: Do not
use symlinks.
* testsuite/27_io/filesystem/iterators/recursive_directory_iterator.cc:
Likewise.
* testsuite/27_io/filesystem/operations/absolute.cc: Use
__gnu_test::root_path() instead of "/" and add Windows-specific tests.
* testsuite/27_io/filesystem/operations/canonical.cc: Use
path::string() to get narrow string, not path::native().
* testsuite/27_io/filesystem/operations/copy.cc: Construct fstreams
with std::filesystem::path not std::basic_string.
* testsuite/27_io/filesystem/operations/copy_file.cc: Likewise.
* testsuite/27_io/filesystem/operations/exists.cc: Use
__gnu_test::root_path() instead of "/".
* testsuite/27_io/filesystem/operations/is_empty.cc: Construct
fstreams with std::filesystem::path not std::basic_string.
* testsuite/27_io/filesystem/operations/last_write_time.cc: Use
path::string() to get narrow string.
* testsuite/27_io/filesystem/operations/space.cc: Check results for
errors, expect sensible values otherwise.
* testsuite/27_io/filesystem/operations/temp_directory_path.cc: Add
helpers for adjusting the environment on Windows.
* testsuite/27_io/filesystem/path/append/path.cc: Test
Windows-specific behaviour.
* testsuite/27_io/filesystem/path/construct/format.cc: Fix creation
of path::string_type objects.
* testsuite/27_io/filesystem/path/construct/locale.cc: Compare native
string to wide string on Windows.
* testsuite/27_io/filesystem/path/decompose/root_directory.cc: Allow
for backslash as root-directory.
* testsuite/27_io/filesystem/path/decompose/stem.cc: Use
path::string() to get narrow string.
* testsuite/27_io/filesystem/path/itr/traversal.cc: Test Windows-style
paths.
* testsuite/27_io/filesystem/path/native/string.cc: Use string_type
not std::string.
* testsuite/27_io/filesystem/path/query/is_absolute.cc: Adjust for
different definintion of absolute paths on Windows.
* testsuite/experimental/filesystem/iterators/directory_iterator.cc:
Do not use symlinks.
* testsuite/experimental/filesystem/operations/absolute.cc: Test
Windows behaviour.
* testsuite/experimental/filesystem/operations/copy.cc: Construct
fstreams with NTCTS not std::basic_string.
* testsuite/experimental/filesystem/operations/copy_file.cc: Likewise.
* testsuite/experimental/filesystem/operations/exists.cc: Use
__gnu_test::root_path() instead of "/".
* testsuite/experimental/filesystem/operations/is_empty.cc: Construct
fstreams with NTCTS not std::basic_string.
* testsuite/experimental/filesystem/operations/last_write_time.cc:
Use path::string() to get narrow string.
* testsuite/experimental/filesystem/operations/space.cc: Use
__gnu_test::root_path() instead of "/".
* testsuite/experimental/filesystem/operations/temp_directory_path.cc:
Add helpers for adjusting the environment on Windows.
* testsuite/experimental/filesystem/path/append/path.cc: Use
path::string() to get narrow strings for comparisons.
* testsuite/experimental/filesystem/path/concat/path.cc: Likewise.
* testsuite/experimental/filesystem/path/decompose/root_directory.cc:
Likewise.
* testsuite/experimental/filesystem/path/decompose/stem.cc: Likewise.
* testsuite/experimental/filesystem/path/native/string.cc: Use
string_type not std::string.
* testsuite/experimental/filesystem/path/query/is_absolute.cc:
Adjust for different definintion of absolute paths on Windows.
* testsuite/util/testsuite_fs.h (__gnu_test::root_path()): New
function.
(__gnu_test::scoped_file): Construct fstreams with NTCTS not
std::basic_string.

From-SVN: r261034

sse.md (avx_vec_concat<mode>): Substitute concat_tg_mode mode attribute with xtg_mode.

* config/i386/sse.md (avx_vec_concat<mode>):
Substitute concat_tg_mode mode attribute with xtg_mode.
(<mask_codefor>avx512dq_broadcast<mode><mask_name>_1): Ditto.
(concat_tg_mode): Remove mode attribute.

From-SVN: r261031

PR c/82063 - issues with arguments enabled by -Wall

gcc/c-family/ChangeLog:

PR c/82063
* c.opt (-Wno-alloc-size-larger-than): New option.
* doc/invoke.texi (-Walloc-size-larger-than): Update.

gcc/ChangeLog:

PR c/82063
* calls.c (alloc_max_size): Correct a logic error/typo.
Treat excessive arguments as infinite.  Warn for invalid arguments.

gcc/testsuite/ChangeLog:

PR c/82063
* gcc.dg/Walloc-size-larger-than-1.c: New test.
* gcc.dg/Walloc-size-larger-than-10.c: New test.
* gcc.dg/Walloc-size-larger-than-11.c: New test.
* gcc.dg/Walloc-size-larger-than-12.c: New test.
* gcc.dg/Walloc-size-larger-than-13.c: New test.
* gcc.dg/Walloc-size-larger-than-14.c: New test.
* gcc.dg/Walloc-size-larger-than-15.c: New test.
* gcc.dg/Walloc-size-larger-than-16.c: New test.
* gcc.dg/Walloc-size-larger-than-17.c: New test.
* gcc.dg/Walloc-size-larger-than-2.c: New test.
* gcc.dg/Walloc-size-larger-than-3.c: New test.
* gcc.dg/Walloc-size-larger-than-4.c: New test.
* gcc.dg/Walloc-size-larger-than-5.c: New test.
* gcc.dg/Walloc-size-larger-than-6.c: New test.
* gcc.dg/Walloc-size-larger-than-7.c: New test.
* gcc.dg/Walloc-size-larger-than-8.c: New test.
* gcc.dg/Walloc-size-larger-than-9.c: New test.
* gcc.dg/Walloc-size-larger-than.c: New test.

From-SVN: r261030

x86: Re-enable partial_reg_dependency and movx for Haswell

r254152 disabled partial_reg_dependency and movx for Haswell and newer
Intel processors.  r258972 restored them for skylake-avx512.  For Haswell,
movx improves performance.  But partial_reg_stall may be better than
partial_reg_dependency in theory.  We will investigate performance impact
of partial_reg_stall vs partial_reg_dependency on Haswell for GCC 9.  In
the meantime, this patch restores both partial_reg_dependency and mox for
Haswell in GCC 8.

On Haswell, improvements for EEMBC benchmarks with

-mtune-ctrl=movx,partial_reg_dependency -Ofast -march=haswell

vs

-Ofast -mtune=haswell

are

automotive
=========
  aifftr01 (default) - goodperf: Runtime improvement of   2.6% (time).
  aiifft01 (default) - goodperf: Runtime improvement of   2.2% (time).

networking
=========
  ip_pktcheckb1m (default) - goodperf: Runtime improvement of   3.8% (time).
  ip_pktcheckb2m (default) - goodperf: Runtime improvement of   5.2% (time).
  ip_pktcheckb4m (default) - goodperf: Runtime improvement of   4.4% (time).
  ip_pktcheckb512k (default) - goodperf: Runtime improvement of   4.2% (time).

telecom
=========
  fft00data_1 (default) - goodperf: Runtime improvement of   8.4% (time).
  fft00data_2 (default) - goodperf: Runtime improvement of   8.6% (time).
  fft00data_3 (default) - goodperf: Runtime improvement of   9.0% (time).

PR target/85829
* config/i386/x86-tune.def: Re-enable partial_reg_dependency
and movx for Haswell.

From-SVN: r261028

re PR middle-end/85879 (ICE in expand_debug_locations, at cfgexpand.c:5405)

Fix PR middle-end/85879

gcc/
* gimplify.c (gimplify_adjust_omp_clauses): Add 'remove = true'
when emitting error on private/firstprivate reductions.
* omp-low.c (lower_omp_target): Avoid reference-type processing
on pointers for firstprivate clause.

gcc/testsuite/
* gfortran.dg/goacc/pr77371-1.f90: New test.
* gfortran.dg/goacc/pr77371-2.f90: New test.
* gfortran.dg/goacc/pr85879.f90: New test.

Co-Authored-By: Cesar Philippidis <cesar@codesourcery.com>
From-SVN: r261025

PR libstdc++/85951 for make_signed/make_unsigned for character types

Because the wide character types are neither signed integer types nor
unsigned integer types they need to be transformed to an integral type
of the correct size and the lowest rank (which is not necessarily the
underlying type). Reuse the helpers for enumeration types to select the
correct integer.

The refactoring of __make_unsigned_selector and __make_signed_selector
slightly reduces the number of template instantiations and so reduces
memory usage.

PR libstdc++/85951
* include/std/type_traits [_GLIBCXX_USE_C99_STDINT_TR1]: Do not define
uint_least16_t and uint_least32_t.
(__make_unsigned<wchar_t>): Define unconditionally.
(__make_unsigned_selector<_Tp, true, false>): Remove intermediate
typedefs.
(__make_unsigned_selector_base): New type to provide helper templates.
(__make_unsigned_selector<_Tp, false, true>): Reimplement using
__make_unsigned_selector_base helpers.
(__make_unsigned<char16_t>, __make_unsigned<char32_t>): Define.
(__make_signed_selector<_Tp, true, false>): Remove intermediate
typedefs.
(__make_signed<wchar_t>, __make_signed<char16_t>)
(__make_signed<char32_t>)): Define unconditionally.
* testsuite/20_util/make_signed/requirements/typedefs-3.cc: Check
wchar_t, char16_t and char32_t are transformed correctly.
* testsuite/20_util/make_signed/requirements/typedefs_neg.cc: Adjust
dg-error lineno.
* testsuite/20_util/make_unsigned/requirements/typedefs-3.cc: Check
wchar_t, char16_t and char32_t are transformed correctly.
* testsuite/20_util/make_unsigned/requirements/typedefs_neg.cc: Adjust
dg-error lineno.

From-SVN: r261023

stack_usage1.adb: Replace -fstack-usage with -Wstack-usage.

* gnat.dg/stack_usage1.adb: Replace -fstack-usage with -Wstack-usage.
* gnat.dg/stack_usage1b.adb: Likewise.
* gnat.dg/stack_usage1c.adb: Likewise.
* gnat.dg/stack_usage3.adb: Likewise.
* gnat.dg/stack_usage1_pkg.adb: Delete.

From-SVN: r261021

store_merging_10.c: Turn dg-do compile into dg-do run.

* gcc.dg/store_merging_10.c: Turn dg-do compile into dg-do run.
* gcc.dg/store_merging_11.c: Likewise.
* gcc.dg/store_merging_13.c: Likewise.
* gcc.dg/store_merging_14.c: Likewise.
* gcc.dg/store_merging_15.c: Likewise.
* gcc.dg/store_merging_16.c: Likewise.  Remove local variable.

From-SVN: r261019

[Ada] Remove obsolete figure in documentation

2018-05-31  Pierre-Marie de Rodat  <derodat@adacore.com>

gcc/ada/

* doc/gnat_ugn/project-manager-figure.png: Delete.

From-SVN: r261018

[Ada] Add missing spec files for VxWorks runtimes

2018-05-31  Pierre-Marie de Rodat  <derodat@adacore.com>

gcc/ada/

* vxworks-arm-link.spec, vxworks-e500-link.spec,
vxworks-gnat-crtbe-link.spec, vxworks-smp-arm-link.spec,
vxworks-smp-e500-link.spec, vxworks-smp-ppc-link.spec,
vxworks-smp-x86-link.spec: New files.

From-SVN: r261017

[Ada] Minor copyright header adjustments

2018-05-31  Pierre-Marie de Rodat  <derodat@adacore.com>

gcc/ada/

* gnatvsn.adb: Re-center the copyright header.
* indepsw-darwin.adb: Adjust the starting copyright year to 2011.

From-SVN: r261016

[Ada] Document Stubs_Subdir in gnattest section on user guide

2018-05-31  Fedor Rybin  <frybin@adacore.com>

gcc/ada/

* doc/gnat_ugn/gnat_utility_programs.rst: Document Stubs_Subdir in
gnattest section on user guide.

From-SVN: r261015

[Ada] Static predicate check on characters of a string literal

This patch implements the rule given in R< 4.2 (11): if the component type of
a string literal is a character type with a static predicate, that predicate
must be applied to each character in the string.

Compiling the example below must yield:

   gcc -c -gnata main.adb

  main.adb:4:23: warning: static expression fails static predicate check on "C"
  main.adb:4:23: warning: expression is no longer considered static
  main.adb:4:24: warning: static expression fails static predicate check on "C"
  main.adb:4:24: warning: expression is no longer considered static
  main.adb:4:25: warning: static expression fails static predicate check on "C"
  main.adb:4:25: warning: expression is no longer considered static

Execution must yield:

  raised SYSTEM.ASSERTIONS.ASSERT_FAILURE :
                                        Static_Predicate failed at main.adb:4
----

procedure Main is
   subtype C is Character with Static_Predicate => C in 'A' | 'B' | 'C';
   type S is array (Positive range <>) of C;
   X : constant S := "abc";
begin
   null;
end;

2018-05-31  Ed Schonberg  <schonberg@adacore.com>

gcc/ada/

* sem_res.adb (Resolve_String_Literal): If the type is a string type
whose component subtype has a static predicate, ensure that the
predicate is applied to each character by expanding the string into the
equivalent aggregate. This is also done if the component subtype is
constrained.

From-SVN: r261014

[Ada] Remove obsolete code in trans.c

2018-05-31  Eric Botcazou  <ebotcazou@adacore.com>

gcc/ada/

* gcc-interface/trans.c (Call_to_gnu): Remove obsolete code.

From-SVN: r261013

[Ada] Detect returning procedures annotated with No_Return

GNAT was emitting a warning about procedures with No_Return aspect on the
spec and a returning body, but failed to handle similar procedures with
no explicit spec. Now fixed.

This was also affecting GNATprove, where an undetected mismatch between
No_Return aspect and the body was a soundness bug, i.e. GNATprove was
silently accept code that raise a runtime exception.

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

procedure P (X : Boolean) with No_Return is
begin
   if X then
      raise Program_Error;
   end if;
end;

-----------------
-- Compilation --
-----------------

$ gcc -c p.adb
p.adb:3:04: warning: implied return after this statement will raise
                     Program_Error
p.adb:3:04: warning: procedure "P" is marked as No_Return

2018-05-31  Piotr Trojanek  <trojanek@adacore.com>

gcc/ada/

* sem_ch6.adb (Check_Missing_Return): Handle procedures with no
explicit spec.

From-SVN: r261012

[Ada] Fix wrong value returned for unconstrained packed array

2018-05-31  Eric Botcazou  <ebotcazou@adacore.com>

gcc/ada/

* gcc-interface/trans.c (Call_to_gnu): In the by-reference case, if
the type of the parameter is an unconstrained array type, convert
to the type of the actual before the type of the formal only if the
conversion was suppressed earlier.  Use in_param and gnu_actual_type
local variables throughout, and uniform spelling for In Out or Out.
Also remove dead code in the component-by-reference case.

From-SVN: r261011

[Ada] Fix __gnat_backtrace for VxWorks7 on x86

A STORAGE ERROR is raised in __gnat_backtrace:

adainit: 0x00400DBC

Execution of ce.vxe terminated by unhandled exception
raised STORAGE_ERROR : SIGSEGV: possible stack overflow
Call stack traceback locations:
0x4082f1 0x408323 0x4080c9

It was passing with vxsim because the WRS_RTP_BASE is set to a different
place hence the (CURRENT) < (TOP_STACK) was stopping the backtrace at the
right time. So let's stop at the main symbol when RTS=rtp.

2018-05-31  Frederic Konrad  <konrad@adacore.com>

gcc/ada/

* tracebak.c (STOP_FRAME): Harden condition.
(is_return_from, EXTRA_STOP_CONDITION): New helpers for VxWorks in RTP
mode.

From-SVN: r261010

[Ada] Illegal copy of limited object

This patch fixes a spurious copy of a limited object, when that object
is a discriminated record component of a limited type LT, and the enclosing
record is initialized by means of an aggregate, one of whose components is a
call to a build-in-place function that returns an unconstrained object of
type T.

2018-05-31  Ed Schonberg  <schonberg@adacore.com>

gcc/ada/

* checks.adb (Apply_Discriminant_Check): Do not apply discriminant
check to a call to a build-in-place function, given that the return
object is limited and cannot be copied.

gcc/testsuite/

* gnat.dg/limited1.adb, gnat.dg/limited1_inner.adb,
gnat.dg/limited1_inner.ads, gnat.dg/limited1_outer.adb,
gnat.dg/limited1_outer.ads: New testcase.

From-SVN: r261009

[Ada] Update comment on __atomic_compare_exchange in s-atomic_primitives

Remove mention of unavailability, long obsolete, and reword suggestion of use
to indicate that we might want to switch to an internal interface using them.
The current wording suggests just that we should bind the current
Sync_Compare_And_Swap Ada subprograms to __atomic_compare builtins instead of
__sync_compare, which would be highly confusing.

2018-05-31  Olivier Hainque  <hainque@adacore.com>

gcc/ada/

* libgnat/s-atopri.ads: Update comment on __atomic_compare_exchange
builtins.

From-SVN: r261008

[Ada] Fix internal error on allocator with function call

2018-05-31  Eric Botcazou  <ebotcazou@adacore.com>

gcc/ada/

* gcc-interface/trans.c (Call_to_gnu): If this is a function call and
there is no target, also create a temporary for the return value for
an allocator if the type is an unconstrained record type with default
discriminant.

From-SVN: r261007

[Ada] Spurious tampering check failure

This patch modifies the transient scope mechanism to create a scope when the
condition of an iteration scheme returns a controlled result or involves the
secondary stack. As a result, a while loop which iterates over a container
properly manages the tampering bit at each iteration of the loop.

2018-05-31  Hristian Kirtchev  <kirtchev@adacore.com>

gcc/ada/

* exp_ch7.adb (Find_Transient_Context): An iteration scheme is a valid
boudary for a transient scope.

gcc/testsuite/

* gnat.dg/tampering_check1.adb, gnat.dg/tampering_check1_ivectors.ads,
gnat.dg/tampering_check1_trim.adb, gnat.dg/tampering_check1_trim.ads:
New testcase.

From-SVN: r261006

[Ada] Rename "GPL" version to "Community"

2018-05-31  Valentine Reboul  <reboul@adacore.com>

gcc/ada/

* gnatvsn.ads: Rename "GPL" version to "Community".

From-SVN: r261005

[Ada] Minor comment improvement

2018-05-31  Ed Schonberg  <schonberg@adacore.com>

gcc/ada/

* einfo.ads: Improve documentation for the Is_Unsigned_Type entity
flag.

From-SVN: r261004

[Ada] Simplify call to Unique_Defining_Entity on protected entry declarations

Calling Unique_Defining_Entity on protectected entry declarations is
equivalent to calling a simpler Defining_Entity; use the simpler routine.

Simplification only; semantics unaffected, so no test provided.

2018-05-31  Piotr Trojanek  <trojanek@adacore.com>

gcc/ada/

* sem_prag.adb (Analyze_Pragma): Replace call to Unique_Defining_Entity
with a semantically equivalent call to Defining_Entity.

From-SVN: r261003

[Ada] Set Etype on rewriteen Max_Queue_Length expressions

Rewriting of Max_Queue_Length expression into N_Integer_Literal should probably
be done in expansion and not in analysis, but anyway it should not strip the
expression from its Etype because backends (e.g. GNATprove) expect that Etype
to be present.

No frontend test is provided, because GNAT doesn't care about the missing
Etype decoration. This patch allows to simplify AST processing in GNATprove.

2018-05-31  Piotr Trojanek  <trojanek@adacore.com>

gcc/ada/

* sem_prag.adb (Analyze_Pragma): Set Etype on the rewritten
Max_Queue_Length expression.

From-SVN: r261002

[Ada] Update gnatelim documentation wrt. compilation of main unit

2018-05-31  Sergey Rybin  <rybin@adacore.com>

gcc/ada/

* doc/gnat_ugn/gnat_and_program_execution.rst: gnatelim does not need
that the main unit to be built before the call, now it computes the
closure of the main unit itself.
* gnat_ugn.texi: Regenerate.

From-SVN: r261001

[Ada] Fix tracking of the first item in Repinfo

2018-05-31  Eric Botcazou  <ebotcazou@adacore.com>

gcc/ada/

* repinfo.adb (List_Structural_Record_Layout): Set First to false
after having listed the fields of the parent type, if any.

From-SVN: r261000

[Ada] Fix strange behavior of Object_Size for packed array subtype

2018-05-31  Eric Botcazou  <ebotcazou@adacore.com>

gcc/ada/

* gcc-interface/decl.c (gnat_to_gnu_entity) <discrete_type>: Do not
deal with the ___XP suffix for packed array types here...
<E_Array_Subtype>: ...or here when processing the implementation type
but when processing the original type instead.  Do not reuse the DECL
of the implementation type for the original type.  Tidy up.

From-SVN: r260999

[Ada] Post warning on object size clause for subtype

This ensures that a warning for an object size clause present on a subtype
is posted on the clause and not on a size clause present on the type.

2018-05-31  Eric Botcazou  <ebotcazou@adacore.com>

gcc/ada/

* einfo.ads (Object_Size_Clause): Declare.
* einfo.adb (Object_Size_Clause): New function.
* gcc-interface/utils.c (maybe_pad_type): Test Has_Size_Clause before
retrieving Size_Clause and post the warning on the object size clause
if Has_Object_Size_Clause is true.

gcc/testsuite/

* gnat.dg/size_clause1.adb: New testcase.

From-SVN: r260998

[Ada] Fix compiler crash for tagged private types

2018-05-31  Javier Miranda  <miranda@adacore.com>

gcc/ada/

* sem_util.ads, sem_util.adb (Find_Primitive_Eq): New subprogram.
* exp_ch4.adb (Expand_Composite_Equality): Use the new subprogram
Find_Primitive_Eq to search for the primitive of types whose underlying
type is a tagged type.

gcc/testsuite/

* gnat.dg/tagged1.adb, gnat.dg/tagged1.ads: New testcase.

From-SVN: r260997

[Ada] Fix check on placement of multiple loop (in)variant pragmas

Loop (in)variants should appear next to each other, which is checked by GNAT
frontend. As statements inserted during expansion may break this contiguity,
GNAT recognizes specially such statements which originate in loop pragmas. In
some cases, this special treatment was not properly put in place, which lead to
spurious errors being issued.

2018-05-31  Yannick Moy  <moy@adacore.com>

gcc/ada/

* sem_prag.adb (Analyze_Pragma.Check_Loop_Pragma_Placement): Inverse
order of treatment between nodes recognized as loop pragmas (or
generated from one) and block statements.

From-SVN: r260996

[Ada] Posix 2008: reimplement System.OS_Primitives.Clock using clock_gettime

gettimeofday is deprecated in Posix 2008, clock_gettime is the recommended
replacement.

2018-05-31  Doug Rupp  <rupp@adacore.com>

gcc/ada/

* libgnat/s-osprim__posix2008.adb (Clock): Implement using
clock_gettime.

From-SVN: r260995

[Ada] Unnesting: add a predicate to help expansion of task and protected types

2018-05-31  Ed Schonberg  <schonberg@adacore.com>

gcc/ada/

* exp_unst.ads, exp_unst.adb (In_Synchronized_Unit): New predicate to
mark subprograms that cannot be eliminated because they must be treated
as reachable from outside the current unit. This includes entry bodies
and protected operations.

From-SVN: r260994

[Ada] Convert operands of generated code to Standard.Unsigned for modular ops.

2018-05-31  Ed Schonberg  <schonberg@adacore.com>

gcc/ada/

* exp_ch4.adb (Expand_Modular_Addition, Expand_Modular_Subtraction):
Convert all operands of the generated code to Standard.Unsigned, to
prevent spurious visibility errors. This is needed when the generated
expansion involves a modular index type of an array type, and the index
type is not immediately visible.

From-SVN: r260993

re PR sanitizer/86012 (libsanitizer build failure on sparc64-linux-gnu)

2018-05-31  Matthias Klose  <doko@ubuntu.com>

        PR sanitizer/86012
        * sanitizer_common/sanitizer_platform_limits_posix.cc: Define
        SIZEOF_STRUCT_USTAT for 32bit sparc.

From-SVN: r260990

Patch implementing vld1_*_x3, vst1_*_x2 and vst1_*_x3 intrinsics for AARCH64 for all types.

From-SVN: r260989