libstdc++: Improve testing for path::operator+=(const string&)

* testsuite/27_io/filesystem/path/concat/strings.cc: Test more cases.

From-SVN: r279111

libstdc++: fix buffer overflow in path::operator+= (PR92853)

When concatenating a path ending in a root-directory onto another path,
we added an empty filename to the end of the path twice, but only
reserved space for one. That meant the second write went past the end of
the allocated buffer.

PR libstdc++/92853
* src/c++17/fs_path.cc (filesystem::path::operator+=(const path&)):
Do not process a trailing directory separator twice.
* testsuite/27_io/filesystem/path/concat/92853.cc: New test.
* testsuite/27_io/filesystem/path/concat/path.cc: Test more cases.

From-SVN: r279110

Come up with constructors of symtab_node, cgraph_node and varpool_node.

2019-12-09  Martin Liska  <mliska@suse.cz>

PR ipa/92737
* cgraph.c (symbol_table_test::symbol_table_test): Fix
coding style.
* cgraph.h (symtab_node::symtab_node): New constructor.
(cgraph_node::cgraph_node): Likewise.
(varpool_node::varpool_node): Likewise.
(symbol_table::allocate_cgraph_symbol): Use newly
created constructor.
(symbol_table::allocate_cgraph_symbol): Remove.
* cgraphunit.c (symtab_terminator): Likewise.
* varpool.c (varpool_node::create_empty): Use newly
created constructor.

From-SVN: r279109

libstdc++: Add C++20 P1032 constexpr to _GLIBCXX_DEBUG array

* testsuite/23_containers/array/tuple_interface/get_debug_neg.cc:
Fix static_assert line number.
* testsuite/23_containers/array/tuple_interface/
tuple_element_debug_neg.cc: Likewise.

From-SVN: r279108

Enable mask movement for VCOND_EXPR under avx512f for
128/256-bit vector when integer mask is available.

Changelog
gcc/
PR target/92686
* config/i386/sse.md
(*<avx512>_cmp<mode>3<mask_scalar_merge_name><round_saeonly_name>,
*<avx512>_cmp<mode>3<mask_scalar_merge_name>,
*<avx512>_ucmp<mode>3<mask_scalar_merge_name>,
*<avx512>_ucmp<mode>3<mask_scalar_merge_name>): New.
* config/i386/i386.c (ix86_print_operand): New operand substitution.
* config/i386/i386-expand.c (ix86_valid_mask_cmp_mode):
New function.
(ix86_expand_sse_cmp): Relax condition for integer mask from
512-bit vector to all 128/256/512-bit vector. Delete code gen
for avx512f compare patterns since we have generic pattern now.
(ix86_expand_sse_movcc): Adjust condition and codegen for
maskcmp.
(ix86_expand_int_sse_cmp): Don't canonicalize the comparison
when corresponding vector compare is available.

gcc/testsuite/
* gcc.target/i386/pr92686.inc: New file.
* gcc.target/i386/avx512bw-pr92686-vpcmp-1.c: New test.
* gcc.target/i386/avx512bw-pr92686-vpcmp-intelasm-1.c: Ditto.
* gcc.target/i386/avx512bw-pr92686-vpcmp-2.c: Ditto.
* gcc.target/i386/avx512vl-pr92686-vpcmp-1.c: Ditto.
* gcc.target/i386/avx512vl-pr92686-vpcmp-intelasm-1.c: Ditto.
* gcc.target/i386/avx512vl-pr92686-vpcmp-2.c: Ditto.
* gcc.target/i386/avx512bw-pr92686-movcc-1.c: Ditto.
* gcc.target/i386/avx512bw-pr92686-movcc-2.c: Ditto.
* gcc.target/i386/avx512vl-pr92686-movcc-1.c: Ditto.
* gcc.target/i386/avx512vl-pr92686-movcc-2.c: Ditto.
* gcc.target/i386/avx512vl-pr88547-1.c: Adjust testcase.
* gcc.target/i386/pr88547-1.c: Ditto.

From-SVN: r279107

re PR go/92861 (Passes relative time to sem_timedwait on GNU/Hurd)

PR go/92861
    runtime: use absolute time for sem_timedwait

    Patch by Samuel Thibault.

    Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/210457

From-SVN: r279106

Daily bump.

From-SVN: r279104

cvt.c (maybe_warn_nodiscard): Add workaround for GCC 3.4-4.4 - cast msg to (const char *) in conditional...

* cvt.c (maybe_warn_nodiscard): Add workaround for GCC 3.4-4.4 - cast
msg to (const char *) in conditional expressions.  Formatting fixes.

From-SVN: r279101

New bswap testcase.

2019-12-08  Andrew Pinski  <apinski@marvell.com>

        * gcc.c-torture/execute/bswap-3.c: New test.

From-SVN: r279100

Revert commit r279023.

2019-12-08  Sandra Loosemore  <sandra@codesourcery.com>

Revert commit r279023.

2019-12-05  Sandra Loosemore  <sandra@codesourcery.com>

gcc/
* config/nios2/linux.h (LINK_EH_SPEC): Define.

gcc/testsuite/
* g++.target/nios2/hello-pie.C: New.
* g++.target/nios2/nios2.exp: New.

From-SVN: r279094

Fix overflows in -fprofile-reorder-functions

This patch fixes three sissues with -fprofile-reorder-functions:
1) First is that tp_first_run is stored as 32bit integer while it can easily
   overflow (and does so during Firefox profiling).
2) Second problem is that flag_profile_functions can
   not be tested w/o function context.
   The changes to expand_all_functions makes it to work on mixed units by
   first outputting all functions w/o -fprofile-reorder-function (or with no
   profile info) and then outputting in first_run order
3) LTO partitioner was mixing up order by tp_first_run and by order.
   for no_reorder we definitly want to order via first, while for everything
   else we want to roder by second.

I have also merged duplicated comparators since they are bit fragile into
tp_first_run_node_cmp.

I originaly started to look into this because of undefined symbols with
Firefox PGO builds.  These symbols went away with fixing these bug but I am not
quite sure how. it is possible that there is another problem in lto_blanced_map
but even after reading the noreorder code few times carefuly I did not find it.
Other explanation would be that our new qsort with broken comparator due to
overflow can actualy remove some entries in the array, but that sounds bit
crazy.

Bootstrapped/regested x86_64-linux.

* cgraph.c (cgraph_node::dump): Make tp_first_run 64bit.
* cgraph.h (cgrpah_node): Likewise.
(tp_first_run_node_cmp): Deeclare.
* cgraphunit.c (node_cmp): Rename to ...
(tp_first_run_node_cmp): ... this; export; watch for 64bit overflows;
clear tp_first_run for no_reorder and !flag_profile_reorder_functions.
(expand_all_functions): Collect tp_first_run and normal functions to
two vectors so the other functions remain sorted. Do not check for
flag_profile_reorder_functions it is function local flag.
* profile.c (compute_value_histograms): Update tp_first_run printing.

* lto-partition.c (node_cmp): Turn into simple order comparsions.
(varpool_node_cmp): Remove.
(add_sorted_nodes): Use node_cmp.
(lto_balanced_map): Use tp_first_run_node_cmp.

From-SVN: r279093

opts.c (common_handle_option): Do not clear ipa_reference for -fprofile-use.

* opts.c (common_handle_option): Do not clear ipa_reference for
-fprofile-use.

From-SVN: r279092

Fix tp_first_run update in split_function.

The value 0 in tp_first_run is special meaing that profile is unknown.  We
should not set it to 1.

* ipa-split.c (split_function): Preserve 0 tp_first_run.

From-SVN: r279091

Silence overactive sanity check with -fpartial-profile-training

do_estimate_edge_time tests that cached and real values matches. This test
is not working precisely for global profiles because of roundoff issues when
profile of clones is subtracted from profile of offline body.  This is checked
by presence of ipa counter. This breaks with partial profile training because
we turn IPA profiles to local when they drop to 0.

* ipa-inline-analysis.c (do_estimate_edge_time): Silence overactive
sanity check.

From-SVN: r279090

re PR tree-optimization/92860 (Global flags affected by -O settings are clobbered by optimize attribute)

PR tree-optimization/92860
* common.opt (fprofile-reorder-functions, ftoplevel-reorder): Add
Optimization flag.

From-SVN: r279089

Error on Associate with a program.

2018-12-08  Thomas Koenig  <tkoenig@gcc.gnu.org>

    PR fortran/92780
    * resolve.c (resolve_assoc_var): Issue error if the associating
    entity is a program.

2018-12-08  Thomas Koenig  <tkoenig@gcc.gnu.org>

    PR fortran/92780
    * gfortran.dg/associate_50.f90: New test.

From-SVN: r279088

re PR fortran/92764 (ICE in gfc_procedure_use)

2018-12-08  Thomas Koenig  <tkoenig@gcc.gnu.org>

    PR fortran/92764
    * interface.c (gfc_procedure_use): Check for existence of derived
    component before using (twice).

2018-12-08  Thomas Koenig  <tkoenig@gcc.gnu.org>

    PR fortran/92764
    * gfortran.dg/interface_44.f90: New test.

From-SVN: r279087

re PR fortran/92755 (ICE in gfc_dep_resolver, at fortran/dependency.c:2123)

2019-12-08  Thomas Koenig  <tkoenig@gcc.gnu.org>

PR fortran/92755
* dependency.c (gfc_dep_resolver):  Move skipping of _data ref
into the loop.

2019-12-08  Thomas Koenig  <tkoenig@gcc.gnu.org>

PR fortran/92755
* gfortran.dg/dependency_57.f90: New test.

From-SVN: r279086

Fix ia32 testsuite failures from C2x attributes on types

* g++.dg/cpp0x/gen-attrs-36.C: Update expected diagnostics.
* g++.dg/cpp0x/gen-attrs-37.C: Likewise.
* g++.dg/cpp0x/gen-attrs-8.C: Likewise.

From-SVN: r279085

Daily bump.

From-SVN: r279084

Fix libstdc++ compiling for an aarch64 multilib with big-endian.

2019-12-07  Andrew Pinski  <apinski@marvell.com>

        * config/cpu/aarch64/opt/ext/opt_random.h: Wrap around with check
        for little-endian like ext/random is done.

From-SVN: r279081

PR c++/91678 - wrong error with decltype and location wrapper.

Compiling this testcase results in a bogus "invalid cast" error; this occurs
since the introduction of location wrappers in finish_id_expression.

Here we are parsing the decltype expression via cp_parser_decltype_expr which
can lead to calling various fold_* and c-family routines.  They use
non_lvalue_loc, but that won't create a NON_LVALUE_EXPR wrapper around a location
wrapper.

So before the location wrappers addition cp_parser_decltype_expr would return
NON_LVALUE_EXPR <c>.  Now it returns VIEW_CONVERT_EXPR<float *>(c), but the
STRIP_ANY_LOCATION_WRAPPER immediately following it strips the location wrapper,
and suddenly we don't know whether we have an lvalue anymore.  And that's sad
because then decltype produces the wrong type, causing nonsense errors.

* fold-const.c (maybe_lvalue_p): Handle VIEW_CONVERT_EXPR.

* g++.dg/cpp0x/decltype73.C: New test.

From-SVN: r279077

re PR middle-end/90840 (ICE in simplify_subreg, at simplify-rtx.c:6441)

PR middle-end/90840
* expr.c (expand_assignment): In the case of a CONCAT on the LHS, make
sure to pass a valid inner mode in calls to simplify_gen_subreg.

From-SVN: r279076

Fortran] PR 92793 - fix column used for error diagnostic

        PR fortran/92793
        * trans.c (gfc_get_location): Declare.
        * trans.c (gfc_get_location): Define; returns column-corrected location.
        (trans_runtime_error_vararg, gfc_trans_runtime_check,
        gfc_generate_module_code): Use new function.
        * trans-array.c (gfc_trans_auto_array_allocation): Likewise.
        * trans-common.c (build_field, get_init_field, create_common): Likewise.
        * trans-decl.c (gfc_build_label_decl, gfc_get_symbol_decl): Likewise.
        * trans-openmp.c (gfc_trans_omp_reduction_list, gfc_trans_omp_clauses):
        Likewise.
        * trans-stmt.c (gfc_trans_if_1): Likewise.

From-SVN: r279075

Fix @multitable handling in texi2pod.pl

While trying out Dennis's Armv8.6-A patch, I noticed that texi2pod.pl
didn't handle the new @multitable correctly.  There were two problems:

(1) @multitables nested in other @tables inherited the @item type from
    the enclosing @table.  Since the new @multitable is in a @table @samp,
    we applied @samp markup to the @multitable @items.  This in turn
    meant that it captured the @tab separator in the @item markup.

    Fixed by pushing an empty item code onto the stack.

(2) We didn't handle @headitem.  Fixed by enclosing it in italics,
    like we do for section headings.  This causes it to be underlined
    in the man output.

2019-12-07  Richard Sandiford  <richard.sandiford@arm.com>

contrib/
* texi2pod.pl: Handle @headitems in @multitables, printing them
in italics.  Push an empty item code onto the stack.

From-SVN: r279074

re PR c/87488 (hyperlink filenames in diagnostics)

PR c/87488
* pretty-print.c (pp_begin_url, pp_end_url, test_urls): Use BEL
instead of ST sequence to terminate OSC 8 strings.

Co-Authored-By: David Malcolm <dmalcolm@redhat.com>
Co-Authored-By: Jakub Jelinek <jakub@redhat.com>
From-SVN: r279073

Daily bump.

From-SVN: r279072

re PR c++/92831 (CWG1299 extend_ref_init_temps_1 punts on COND_EXPRs)

PR c++/92831
* call.c (build_conditional_expr_1): For ?: with omitted middle
operand use cp_stabilize_reference if arg1 is glvalue_p rather than
just if it is lvalue_p.

* g++.dg/ext/temp-extend1.C: New test.

From-SVN: r279069

Fix read buffer overflow in split_directories

        * make-relative-prefix.c (split_directories):
        Return early on empty 'name'

From-SVN: r279068

Add test for c++/92451.

This was ICEing from r277865 to r278786.

* g++.dg/overload/error4.C: New test.

From-SVN: r279067

parser.c (cp_parser_diagnose_invalid_type_name): Mention that concept is also available with -std=c++2a.

* parser.c (cp_parser_diagnose_invalid_type_name): Mention
that concept is also available with -std=c++2a.

From-SVN: r279066

c-cppbuiltin.c (c_cpp_builtins): Don't define __cpp_consteval for now.

* c-cppbuiltin.c (c_cpp_builtins): Don't define __cpp_consteval for
now.

* g++.dg/cpp2a/feat-cxx2a.C: Don't test __cpp_consteval for now.

From-SVN: r279065

PR c++/92831 - CWG 1299, not extending temporary lifetime for ?:

* cp-tree.h (extend_ref_init_temps): Add a new argument with NULL
default arg.
* call.c (set_up_extended_ref_temp): Add COND_GUARD argument, pass it
down to extend_ref_init_temps.  Before pushing cleanup, if COND_GUARD
is non-NULL, create a bool temporary if needed, initialize to false
and guard the cleanup with the temporary being true.
(extend_ref_init_temps_1): Add COND_GUARD argument, pass it down
to recursive calls and set_up_extended_ref_temp.  Handle COND_EXPR.
(extend_ref_init_temps): Add COND_GUARD argument, pass it down to
recursive calls and to extend_ref_init_temps_1.

* g++.dg/cpp0x/temp-extend2.C: New test.

From-SVN: r279064

re PR go/92820 (libgo.so.15 has executable stack)

PR go/92820
    runtime: only build go-context for x86 GNU/Linux

    Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/210258

From-SVN: r279063

re PR other/29842 ([meta-bug] outstanding patches / issues from STMicroelectronics)

PR go/29842
    runtime: update HURD support for mOS now being embedded

    Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/210285

From-SVN: r279062

re PR rtl-optimization/92176 (LRA problem with reloads for subreg operands)

2019-12-06  Andreas Krebbel  <krebbel@linux.ibm.com>
    Vladimir Makarov  <vmakarov@redhat.com>

PR rtl-optimization/92176
* lra.c (simplify_subreg_regno): Don't permit unconditional
changing mode for LRA too.

2019-12-06  Andreas Krebbel  <krebbel@linux.ibm.com>
    Vladimir Makarov  <vmakarov@redhat.com>

PR rtl-optimization/92176
* gcc.target/s390/pr92176.c: New test.

Co-Authored-By: Vladimir Makarov <vmakarov@redhat.com>
From-SVN: r279061

Fix Fortran linker errors on AMD GCN

2019-12-06  Kwok Cheung Yeung  <kcy@codesourcery.com>

libgomp/
* config/accel/proc.c (omp_get_num_procs): Apply ialias macro.

From-SVN: r279060

Avoid -Wincompatible-pointer-types on targets like arm-eabi.

gcc/testsuite/ChangeLog:

* gcc.dg/Wstringop-overflow-23.c: Use the correct argument type.
* gcc.dg/Wstringop-overflow-24.c: Same.

From-SVN: r279059

[C++] Pass type uses through the verify_type_context hook

This patch makes the C++ frontend work with the verify_type_context hook.
We need some new type contexts for features that don't exist in C, but
otherwise the patch is very similar to the C one.

TCTX_CAPTURE_BY_COPY could really be treated as an instance of
TCTX_FIELD, but the error message is better if we split it out.

2019-12-06  Richard Sandiford  <richard.sandiford@arm.com>

gcc/
* target.h (TCTX_ALLOCATION, TCTX_DEALLOCATION, TCTX_EXCEPTIONS)
(TCTX_CAPTURE_BY_COPY): New type_context_kinds.
* config/aarch64/aarch64-sve-builtins.cc (verify_type_context):
Handle them.

gcc/cp/
* decl.c (start_decl_1): Use verify_type_context to check whether
the target allows variables of a particular type to have static
or thread-local storage duration.
(check_array_initializer): Use verify_type_context to check whether
the target allows a particular type to be used as an array element.
(create_array_type_for_decl): Likewise.
(cp_finish_decl): Use verify_type_context to check whether
the target allows static member variables of a particular type.
(grokdeclarator): Likewise.  Also use verify_type_context to check
whether the target allows non-static member variables of a particular
type.
* except.c: Include target.h.
(is_admissible_throw_operand_or_catch_parameter): Use
verify_type_context to check whether the target allows particular
types to be thrown and caught.
* typeck2.c (add_exception_specifier): Likewise.
* init.c (build_new_1): Use verify_type_context to check whether
the target allows particular types to be dynamically allocated.
(build_vec_delete_1, build_delete): Use verify_type_context to check
whether the target allows particular types to be deleted.
* lambda.c (add_capture): Use verify_type_context to check
whether the target allows particular types to be captured by copy.
* pt.c: Include target.h.
(instantiate_class_template_1): Use verify_type_context to check
whether the target allows non-static member variables of a particular
type.
* typeck.c (cxx_alignof_expr): Use verify_type_context to check
whether the target allows the alignment of a particular type
to be measured.
(pointer_diff, cp_build_unary_op): Use verify_type_context to check
whether the target allows arithmetic involving pointers to particular
types.

gcc/testsuite/
* g++.dg/ext/sve-sizeless-1.C: New test.
* g++.dg/ext/sve-sizeless-2.C: Likewise.

From-SVN: r279058

Fix unrecognised GCN instruction.

2019-12-06  Andrew Stubbs  <ams@codesourcery.com>

gcc/
* config/gcn/gcn-valu.md (gather<mode>_insn_1offset<exec>): Use %o
in the asm output.

From-SVN: r279055

Enable QI/HImode vector moves

2019-12-06  Andrew Stubbs  <ams@codesourcery.com>

gcc/
* config/gcn/gcn-valu.md (VEC_1REG_MODE): Remove V64QI and V64HI.
(VEC_1REG_ALT): Likewise.
(VEC_ALL1REG_MODE): New mode iterator.
(VEC_1REG_INT_MODE): Remove V64QI and V64HI.
(VEC_1REG_INT_ALT): Likewise.
(VEC_ALL1REG_INT_MODE): New mode interator.
(VEC_ALL1REG_INT_ALT): Likewise.
(VEC_REG_MODE): Remove V64QI and V64HI.
(VEC_ALLREG_MODE): New mode interator.
(vec_merge): Change to VEC_ALLREG_MODE.
(vec_merge_with_clobber): Likewise.
(vec_merge_with_vcc): Likewise.
(mov<mode>): Likewise.
(mov<mode>_unspec): Likewise.
(*mov<mode>): Change to VEC_ALL1REG_MODE.
(mov<mode>_exec): Likewise.
(*mov<mode>_exec_match): Likewise.
(mov<mode>_sgprbase): Likewise.
(reload_in<mode>): Change to VEC_ALLREG_MODE.
(reload_out<mode>): Likewise.
(scalar address splits): Likewise.
(*vec_set<mode>): Change to VEC_ALL1REG_MODE.
(vec_set<mode>): Change to VEC_ALLREG_MODE.
(*vec_set<mode>_1): Change to VEC_ALL1REG_MODE.
(vec_duplicate<mode><exec>): Likewise.
(vec_extract<mode><scalar_mode>): Likewise.
(vec_init<mode><scalar_mode>): Change to VEC_ALLREG_MODE.
(gather_load<mode>): Likewise.
(gather<mode>_exec): Likewise.
(gather<mode>_expr<exec>): Likewise.
(gather<mode>_insn_1offset<exec>): Likewise.
(gather<mode>_insn_1offset_ds<exec>): Likewise.
(gather<mode>_insn_2offsets<exec>): Likewise.
(ds_bpermute<mode>): Change to VEC_ALL1REG_MODE.
(VEC_INT_MODE): Remove V64QI and V64HI.
(vcond_mask_<mode>di): Change to VEC_ALLREG_MODE.
(vcond<VEC_1REG_MODE:mode><VEC_1REG_ALT:mode>): Change to
VEC_ALL1REG_MODE.
(vcond<VEC_1REG_MODE:mode><VEC_1REG_ALT:mode>_exec): Likewise.
(vcondu<VEC_1REG_MODE:mode><VEC_1REG_INT_ALT:mode>): Likewise.
(vcondu<VEC_1REG_MODE:mode><VEC_1REG_INT_ALT:mode>_exec): Likewise.
(maskload<mode>di): Change to VEC_ALL1REG_MODE.
(maskstore<mode>di): Likewise.
(mask_gather_load<mode>): Likewise.
(mov_from_lane63_<mode>): Likewise.
* config/gcn/gcn.c (gcn_vector_mode_supported_p): Renable V64QImode
and V64HImode vectorization.
(gcn_related_vector_mode): New function.
(TARGET_VECTORIZE_RELATED_MODE): New define.

From-SVN: r279053

re PR fortran/92775 (Incorrect expression in DW_AT_byte_stride on an array)

PR fortran/92775
* trans.h (struct lang_type, struct lang_decl): Remove span member.
(GFC_DECL_SPAN, GFC_TYPE_ARRAY_SPAN): Remove macros.
* trans-array.h (gfc_get_descriptor_offsets_for_info): Add another
argument.
* trans-array.c (gfc_get_descriptor_offsets_for_info): Add SPAN_OFF
argument and initialize *SPAN_OFF to the offset of span field.
* trans-types.c (gfc_get_array_descr_info): Adjust
gfc_get_descriptor_offsets_for_info caller.  Compute elem_size
as base->span instead of TYPE_SIZE_UNIT (etype) constant.

From-SVN: r279045

[OpenMP/OpenACC/Fortran] Fix mapping of optional (present|absent) arguments

2019-12-06  Tobias Burnus  <tobias@codesourcery.com>
            Kwok Cheung Yeung <kcy@codesourcery.com>

        gcc/fortran/
        * trans-openmp.c (gfc_build_conditional_assign,
        gfc_build_conditional_assign_expr): New static functions.
        (gfc_omp_finish_clause, gfc_trans_omp_clauses): Handle mapping of
        absent optional arguments and fix mapping of present optional args.

        gcc/
        * omp-low.c (lower_omp_target): For optional arguments, deref once
        more to obtain the type.

        libgomp/
        * oacc-mem.c (update_dev_host, gomp_acc_insert_pointer): Just return
        if input it a NULL pointer.
        * testsuite/libgomp.oacc-c-c++-common/lib-43.c: Remove; dependent on
        diagnostic of NULL pointer.
        * testsuite/libgomp.oacc-c-c++-common/lib-47.c: Ditto.
        * testsuite/libgomp.fortran/optional-map.f90: New.
        * testsuite/libgomp.fortran/use_device_addr-1.f90
        (test_dummy_opt_callee_1_absent): New.
        (test_dummy_opt_call_1): Call it.
        * testsuite/libgomp.fortran/use_device_addr-2.f90: Likewise.
        * testsuite/libgomp.fortran/use_device_addr-3.f90: Likewise.
        * testsuite/libgomp.fortran/use_device_addr-4.f90: Likewise.
        * testsuite/libgomp.oacc-fortran/optional-cache.f95: New.
        * testsuite/libgomp.oacc-fortran/optional-data-copyin-by-value.f90: New.
        * testsuite/libgomp.oacc-fortran/optional-data-copyin.f90: New.
        * testsuite/libgomp.oacc-fortran/optional-data-copyout.f90: New.
        * testsuite/libgomp.oacc-fortran/optional-data-enter-exit.f90: New.
        * testsuite/libgomp.oacc-fortran/optional-declare.f90: New.
        * testsuite/libgomp.oacc-fortran/optional-firstprivate.f90: New.
        * testsuite/libgomp.oacc-fortran/optional-host_data.f90: New.
        * testsuite/libgomp.oacc-fortran/optional-nested-calls.f90: New.
        * testsuite/libgomp.oacc-fortran/optional-private.f90: New.
        * testsuite/libgomp.oacc-fortran/optional-reduction.f90: New.
        * testsuite/libgomp.oacc-fortran/optional-update-device.f90: New.
        * testsuite/libgomp.oacc-fortran/optional-update-host.f90: New.

Co-Authored-By: Kwok Cheung Yeung <kcy@codesourcery.com>
From-SVN: r279043

match.pd (nop_convert): Remove empty match.

2019-12-06  Richard Biener  <rguenther@suse.de>

* match.pd (nop_convert): Remove empty match.  Use nop_convert?
everywhere.

From-SVN: r279040

[testsuite][aarch64] type_redef_11.c: Update expected diagnostics.

After the fix for PR c/36941 and PR c/88827 (r278976), this test emits
a different error message and needs an update.

2019-12-06  Christophe Lyon  <christophe.lyon@linaro.org>

PR c/36941
PR c/88827
* gcc.target/aarch64/sve/acle/general-c/type_redef_11.c: Update
expected diagnostics.

From-SVN: r279039

Avoid quadratic behaviour in prune_runtime_alias_test_list

prune_runtime_alias_test_list used ordered_remove to remove a merged
alias pair, which made the function quadratic when many aliases could
be removed.

I had a testcase in which these memmoves accounted for an impressive
85% of compile time.  The fact that we had so many probably shows
a deeper problem, but still, it's easy to remove as we go.

2019-12-06  Richard Sandiford  <richard.sandiford@arm.com>

gcc/
* tree-data-ref.c (prune_runtime_alias_test_list): Exit early
for empty vectors.  Avoid using ordered_remove and instead
shuffle the vector as we go.

From-SVN: r279038

genmatch.c (enum tree_code): Remove CONVERT{0,1,2} and VIEW_CONVERT{0,1,2}.

2019-12-06  Richard Biener  <rguenther@suse.de>

* genmatch.c (enum tree_code): Remove CONVERT{0,1,2} and
VIEW_CONVERT{0,1,2}.
(expr::opt_grp): Add and initialize.
(lower_opt_convert): Rename to ...
(lower_opt): ... and work on opt_grp, simply switching operations
from being optional to being present or not.
(has_opt_convert): Rename to ...
(has_opt): ... and adjust.
(parser::parse_operation): Return the optional opt_grp,
remove special-casing of conditional operations and more generally
parse [digit]'?'.
(parser::parse_expr): Stick on the parsed opt_grp and perform
rough verification.
(parser::parse_for): Remove now unnecessary code.
(main): Likewise.
* doc/match-and-simplify.texi: Mention ? now works on all
unary operations and also match predicates.

From-SVN: r279037

cgraphclones.c (cgraph_node::create_clone): Only localize toplevel functions, not inline clones.

* cgraphclones.c (cgraph_node::create_clone): Only localize toplevel
functions, not inline clones.

From-SVN: r279036

* cfgexpand.c (expand_debug_expr): Do not ICE on SAD_EXPR.

From-SVN: r279035

genmatch.c (c_expr::gen_transform): Emit newlines from line number changes rather than after every semicolon.

2019-12-06  Richard Biener  <rguenther@suse.de>

* genmatch.c (c_expr::gen_transform): Emit newlines from line
number changes rather than after every semicolon.

From-SVN: r279034

re PR tree-optimization/92819 (Worse code generated on avx2 due to simplify_vector_constructor)

2019-12-06  Richard Biener  <rguenther@suse.de>

PR tree-optimization/92819
* match.pd (VEC_PERM_EXPR -> BIT_INSERT_EXPR): Handle inserts
into the last lane.  For two-element vectors try inserting
into the last lane when inserting into the first fails.

* gcc.target/i386/pr92819-1.c: New testcase.
* gcc.target/i386/pr92803.c: Adjust.

From-SVN: r279033

re PR go/92810 (Compiling GCC go for aarch64_be-marvell-linux-gnu fails)

PR go/92810
    libgo: recognize aarch64_be as arm64be

    Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/210038

From-SVN: r279032

PR testsuite/92829 - several test case failures starting with r278983

gcc/testsuite/ChangeLog:

* gcc.dg/Warray-bounds-56.c: Rearrange code to avoid powerpc64*
failures.

From-SVN: r279031

common.opt (fprofile-partial-training): Terminate description with full stop.

* common.opt (fprofile-partial-training): Terminate description with
full stop.

From-SVN: r279030

PR middle-end/92622 - FAIL: gcc.dg/Warray-bounds-22.c on ILP32: missing warnings for VLA on lines 67 and 69

gcc/ChangeLog:

PR middle-end/92622
* tree-vrp.c (vrp_prop::check_array_ref): Avoid using a variable
left uninitialized by get_addr_base_and_unit_offset ofn failure.

From-SVN: r279029

Daily bump.

From-SVN: r279028

Do not clear m_vr and bits in ipa_set_jf_unknown.

* ipa-prop.c (ipa_set_jf_unknown): Do not clear bits and m_vr.
(detect_type_change_from_memory_writes): Remoe parameter JFUNC.
(detect_type_change): Likewise.
(detect_type_change_ssa): Likewise.
(ipa_analyze_virtual_call_uses): Update.

From-SVN: r279025

re PR tree-optimization/92768 (Maybe a wrong code for vector constants)

PR tree-optimization/92768
* gcc.dg/pr92768.c: Add -w -Wno-psabi to dg-options.

From-SVN: r279024

Disable --eh-frame-hdr with -pie or -shared on nios2-linux-gnu.

2019-12-05  Sandra Loosemore  <sandra@codesourcery.com>

Disable --eh-frame-hdr with -pie or -shared on nios2-linux-gnu.

The nios2 ABI doesn't have appropriate relocations to support
--eh-frame-hdr with -pie or -shared, so we need to override the
default specs that add that link option.  Otherwise we get linker
warnings about invalid FDE encodings.

gcc/
* config/nios2/linux.h (LINK_EH_SPEC): Define.

gcc/testsuite/
* g++.target/nios2/hello-pie.C: New.
* g++.target/nios2/nios2.exp: New.

From-SVN: r279023

Fix g++.dg/torture/pr59226.C

this patch fixes ICE in g++.dg/torture/pr59226.C which was triggered by
new comdat_local sanity check.  What happens here is that function gets
inlined into its own thunk which makes it !comdat_local_p but the updating
code does not notice since thunk calls comdat local alias of the function
itself and we look at alias target rather than original callee.

This also shows that we miss optimization here.  Currently we will not inline
thunk out of its comdat local group w/o inlining function it is associated with
into it.

We should teach inline_call to reoslve edges to aliases while inlining and
relax calls_comdat_local flag.  But this needs bit more work, so I fix the
ICE first.

* ipa-inline-transform.c (inline_call): Fix maintenatnce of comdat_local

From-SVN: r279021

PR c++/91353 - P1331R2: Allow trivial default init in constexpr contexts.

This patch implements C++20 P1331, allowing trivial default initialization in
constexpr contexts.

* c-cppbuiltin.c (c_cpp_builtins): Adjust the value of __cpp_constexpr.

* class.c (trivial_default_constructor_is_constexpr): Return true in
C++20.
* constexpr.c (cx_check_missing_mem_inits): Allow missing field
initializers in C++20.
(cxx_eval_call_expression): Don't clear CONSTRUCTOR_NO_CLEARING for
constexpr constructors in C++20.
(reduced_constant_expression_p): Don't set FIELD for union and array
types.  Skip empty class fields without initializers.
* decl.c (check_for_uninitialized_const_var): Permit trivial default
initialization in constexpr.
(next_initializable_field): Don't skip vptr fields.
* method.c (walk_field_subobs): Still consider a constructor that
doesn't initialize all the members constexpr.

* g++.dg/cpp0x/constexpr-array6.C: Adjust dg-error.
* g++.dg/cpp0x/constexpr-ctor.C: Likewise.
* g++.dg/cpp0x/constexpr-diag3.C: Likewise.
* g++.dg/cpp0x/constexpr-diag4.C: Likewise.
* g++.dg/cpp0x/constexpr-ex3.C: Likewise.
* g++.dg/cpp0x/constexpr-template2.C: Likewise.
* g++.dg/cpp0x/constexpr-union2.C: Likewise.
* g++.dg/cpp0x/lambda/lambda-mangle.C: Rip out a piece of code ...
* g++.dg/cpp0x/lambda/lambda-mangle6.C: ... and put it here.
* g++.dg/cpp0x/pr79118.C: Adjust dg-error.
* g++.dg/cpp1y/constexpr-83921-3.C: Likewise.
* g++.dg/cpp1y/constexpr-neg1.C: Likewise.
* g++.dg/cpp1z/constexpr-lambda12.C: Likewise.
* g++.dg/cpp1z/feat-cxx1z.C: Use -std=c++17.
* g++.dg/cpp2a/constexpr-init1.C: New test.
* g++.dg/cpp2a/constexpr-init2.C: New test.
* g++.dg/cpp2a/constexpr-init3.C: New test.
* g++.dg/cpp2a/constexpr-init4.C: New test.
* g++.dg/cpp2a/constexpr-init5.C: New test.
* g++.dg/cpp2a/constexpr-init6.C: New test.
* g++.dg/cpp2a/constexpr-init7.C: New test.
* g++.dg/cpp2a/constexpr-init8.C: New test.
* g++.dg/cpp2a/constexpr-init9.C: New test.
* g++.dg/cpp2a/constexpr-init10.C: New test.
* g++.dg/cpp2a/constexpr-init11.C: New test.
* g++.dg/cpp2a/constexpr-init12.C: New test.
* g++.dg/cpp2a/constexpr-init13.C: New test.
* g++.dg/cpp2a/constexpr-init14.C: New test.
* g++.dg/cpp2a/constexpr-init15.C: New test.
* g++.dg/cpp2a/constexpr-try5.C: Adjust dg-error.
* g++.dg/cpp2a/feat-cxx2a.C: Test __cpp_constexpr.
* g++.dg/cpp2a/lambda-mangle.C: New test.
* g++.dg/debug/dwarf2/pr44641.C: Skip for c++2a.
* g++.dg/ext/stmtexpr21.C: Adjust dg-error.

Co-Authored-By: Jakub Jelinek <jakub@redhat.com>
From-SVN: r279019

PR c++/92271 - make __is_same alias for __is_same_as.

Richard Smith proposed adding a synonym for __is_same_as, to accomodate the
convention of exposing std::SOME_TRAIT<A, B>::value as __SOME_TRAIT(A, B).

So add that alias, and also adjust the C++ printer.  I didn't bother changing
the RID_ identifier.

* c-common.c: Add __is_same, an alias for __is_same_as.

* cxx-pretty-print.c (pp_cxx_trait_expression) <case CPTK_IS_SAME_AS>:
Print "__is_same".

* g++.dg/ext/is_same.C: New test.

From-SVN: r279018

cgraphclones.c (localize_profile): New function.

* cgraphclones.c (localize_profile): New function.
(cgraph_node::create_clone): Use it for partial profiles.
* common.opt (fprofile-partial-training): New flag.
* doc/invoke.texi (-fprofile-partial-training): Document.
* ipa-cp.c (update_profiling_info): For partial profiles do not
set function profile to zero.
* profile.c (compute_branch_probabilities): With partial profile
watch if edge count is zero and turn all probabilities to guessed.
(compute_branch_probabilities): For partial profiles do not apply
profile when entry count is zero.
* tree-profile.c (tree_profiling): Only do value_profile_transformations
when profile is read.

From-SVN: r279013

[Patch, GCC] Fix a condition post r278611

gcc/ChangeLog

2019-12-05  Sudakshina Das  <sudi.das@arm.com>

* tree-vect-loop.c (vect_model_reduction_cost): Remove reduction_type
check from if condition.

From-SVN: r279012

re PR go/92820 (libgo.so.15 has executable stack)

PR go/92820
    runtime: always mark assembly file as non-executable stack

    Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/210037

From-SVN: r279010

OpenMP] Fix use_device_… with absent optional arg

        gcc/fortran/
        * trans-openmp.c (gfc_omp_is_optional_argument,
        gfc_omp_check_optional_argument): Handle type(c_ptr),value which uses a
        hidden argument for the is-present check.

        gcc/
        * omp-low.c (lower_omp_target): For use_device_ptr/use_derice_addr
        and Fortran's optional arguments, unconditionally add the is-present
        condition before the libgomp call.

        libgomp/
        * testsuite/libgomp.fortran/use_device_ptr-optional-2.f90: Add
        'type(c_ptr), value' test case. Conditionally map the per-value
        passed arguments.

From-SVN: r279004

* cp-gimplify.c: Include memmodel.h.

From-SVN: r279003

Check for bitwise identity when encoding VECTOR_CSTs (PR 92768)

This PR shows that we weren't checking for bitwise-identical values
when trying to encode a VECTOR_CST, so -0.0 was treated the same as
0.0 for -fno-signed-zeros.  The patch adds a new OEP flag to select
that behaviour.

2019-12-05  Richard Sandiford  <richard.sandiford@arm.com>

gcc/
PR middle-end/92768
* tree-core.h (OEP_BITWISE): New flag.
* fold-const.c (operand_compare::operand_equal_p): Handle it.
* tree-vector-builder.h (tree_vector_builder::equal_p): Pass it.

gcc/testsuite/
PR middle-end/92768
* gcc.dg/pr92768.c: New test.

From-SVN: r279002

libstdc++: Fix bug in std::span test

The previous commit fixed the std::span constructors from const arrays,
revealing a bug in this test.

* testsuite/23_containers/span/lwg3255.cc: Fix test. Constructing a
span of non-const elements should not be possible from a const array
or an array of const elements.

From-SVN: r279001

libstdc++: Implement P1872R0 and P1394R0 for std::span

This also fixes a bug in the implementation of LWG 3255, which causes:
FAIL: 23_containers/span/lwg3255.cc (test for excess errors)
That's because the test was wrong and verified the buggy behaviour. That
will be fixed in the following commit.

2019-12-05  JeanHeyd "ThePhD" Meneide  <phdofthehouse@gmail.com>

Implement P1872R0 and P1394R0 for std::span
* include/bits/range_access.h (__adl_begin, __adl_end): Remove.
(sentinel_t, range_value_t, range_reference_t)
(range_rvalue_reference_t, __forwarding_range, disable_sized_range)
(output_range, input_range, forward_range, bidirectional_range)
(random_access_range, contiguous_range, common_range): Move here from
<ranges>, to make this the "ranges lite" internal header.
* include/std/ranges: Move basic aliases and concepts to
<bits/range_access.h>.
* include/std/span: Use concepts and ranges:: calls instead of
enable_if and friends.
* include/std/type_traits: Add __is_array_convertible trait.

From-SVN: r279000

re PR tree-optimization/92818 (Typo in vec_perm -> bit_insert pattern)

2019-12-05  Richard Biener  <rguenther@suse.de>

PR middle-end/92818
* tree-ssa-forwprop.c (simplify_vector_constructor): Improve
heuristics on what don't care element to choose.
* match.pd (VEC_PERM_EXPR -> BIT_INSERT_EXPR): Fix typo.

* gcc.target/i386/pr92818.c: New testcase.

From-SVN: r278998

libstdc++: Define std::lexicographical_compare_three_way for C++20

* include/bits/stl_algobase.h (lexicographical_compare_three_way):
Define for C++20.
* testsuite/25_algorithms/lexicographical_compare_three_way/1.cc: New
test.
* testsuite/25_algorithms/lexicographical_compare_three_way/
constexpr.cc: New test.

From-SVN: r278996

Fix __gcov_exit fn prototype.

2019-12-05  Martin Liska  <mliska@suse.cz>

PR gcov-profile/92817
* coverage.c (build_gcov_exit_decl): Remove superfluous
void_type_node.

From-SVN: r278995

MSP430: Fix postinc addressing mode being used for dst op of CMP insns

2019-12-05  Jozef Lawrynowicz  <jozef.l@mittosystems.com>

* config/msp430/msp430.md (cbranch<mode>4): Use
msp430_general_dst_nonv_operand instead of nonimmediate_operand for
dest operand of CMP instruction.
(cbranchpsi4_real): Likewise.
(cbranchqi4_real): Likewise.
(cbranchhi4_real): Likewise.
(cbranchpsi4_reversed): Likewise.
(cbranchqi4_reversed): Likewise.
(cbranchhi4_reversed): Likewise.

From-SVN: r278994

Skip test gcc/testsuite/gcc.dg/asm-4.c for amdgcn target

2019-12-05  Frederik Harwath  <frederik@codesourcery.com>

gcc/testsuite/
* gcc.dg/asm-4.c: Skip on target amdgcn-*-*.

From-SVN: r278993

re PR target/92055 ([avr] Support 64-bit double)

PR target/92055
* config/avr/t-avrlibc (MULTISUBDIR): Search for double, not double64.

From-SVN: r278992

re PR tree-optimization/92803 (error: type mismatch in 'vec_perm_expr' since r278764)

2019-12-05  Richard Biener  <rguenther@suse.de>

PR tree-optimization/92803
* tree-ssa-forwprop.c (simplify_vector_constructor): Fix
invariant vector construction.

* gcc.target/i386/pr92803.c: New testcase.

From-SVN: r278991

Fix profile name files without -fprofile-dir.

2019-12-05  Martin Liska  <mliska@suse.cz>

PR gcov-profile/91971
* coverage.c (coverage_init): Mangle full path
only when -fprofile-dir is used.

From-SVN: r278990

re PR target/92791 (ICE in extract_insn, at recog.c:2311 since r278645)

PR target/92791
* config/i386/i386.md (movstrict<mode>): Move test for
TARGET_PARTIAL_REG_STALL and not optimizing for size from
expander's condition to the body - FAIL; in that case.

From-SVN: r278989

re PR fortran/92781 (ICE in convert_nonlocal_reference_op, at tree-nested.c:1065)

PR fortran/92781
* trans-decl.c (gfc_get_symbol_decl): If sym->backend_decl is
current_function_decl, add length to current rather than parent
function and expect DECL_CONTEXT (length) to be current_function_decl.

* gfortran.dg/pr92781.f90: New test.

From-SVN: r278988

typeck2.c (build_functional_cast): Add location_t parameter and use it.

/gcc/cp
2019-12-05  Paolo Carlini  <paolo.carlini@oracle.com>

* typeck2.c (build_functional_cast): Add location_t parameter
and use it.
* cp-tree.h: Update declaration.
* parser.c (cp_parser_functional_cast): Adjust call.
* call.c (build_op_delete_call): Likewise.
(build_new_method_call_1): Likewise.
* decl.c (check_initializer): Likewise.
* pt.c (tsubst_copy_and_build): Likewise.
* semantics.c (finish_compound_literal): Likewise.

/libcc1
2019-12-05  Paolo Carlini  <paolo.carlini@oracle.com>

* libcp1plugin.cc (plugin_build_expression_list_expr): Adjust
build_functional_cast call.

/testsuite
2019-12-05  Paolo Carlini  <paolo.carlini@oracle.com>

* g++.dg/diagnostic/functional-cast-to-array-type-1.C: New.
* g++.dg/cpp0x/auto25.C: Check location(s) too.
* g++.dg/cpp0x/auto28.C: Likewise.
* g++.dg/init/reference2.C: Likewise.
* g++.dg/parse/template2.C: Likewise.
* g++.dg/template/error8.C: Likewise.
* g++.old-deja/g++.ns/crash3.C: Likewise.
* g++.old-deja/g++.ns/template7.C: Likewise.
* g++.old-deja/g++.pt/crash8.C: Likewise.

From-SVN: r278987

testsuite/libgomp.oacc-fortran:  Update dg-output to fix GCN

        * testsuite/libgomp.oacc-fortran/error_stop-1.f: Also don't
        expect dg-output of 'Error termination.' for GCN.
        * testsuite/libgomp.oacc-fortran/error_stop-2.f: Likewise.
        * testsuite/libgomp.oacc-fortran/error_stop-3.f: Likewise.

Plus: Fix date of a previous commit in:
        * gcc/fortran/ChangeLog
        * gcc/testsuite/ChangeLog

From-SVN: r278986

* edtest.c (test1): Add noclone attribute.

From-SVN: r278985

libbacktrace: simplify DWARF section handling

This is in preparation for adding DWARF 5 support.

* internal.h (enum dwarf_section): Define.
(struct dwarf_sections): Define.
(backtrace_dwarf_add): Update declaration to replace specific
section parameters with dwarf_sections parameter.
* dwarf.c (struct dwarf_data): Replace specific section fields
with dwarf_sections field.
(read_attribute): Use dwarf_sections with altlink.
(build_address_map): Replace specific section parameters with
dwarf_sections parameter.  Change all callers.
(read_line_info): Use dwarf_sections with ddata.
(read_referenced_name): Likewise.
(add_function_ranges): Likewise.
(read_function_entry): Likewise.
(read_function_info): Likewise.
(build_dwarf_data): Replace specific section parameters with
dwarf_sections parameter.  Change all callers.
(backtrace_dwarf_add): Likewise.
* elf.c (enum debug_section): Remove.
(dwarf_section_names): Remove .zdebug names.
(elf_add): Track zsections separately.  Build dwarf_sections.
* pecoff.c (enum debug_section): Remove.
(struct debug_section_info): Remove data field.
(coff_add): Build dwarf_sections.
* xcoff.c (enum dwarf_section): Remove.  Replace DWSECT_xxx
references with DEBUG_xxx references.
(xcoff_add): Build dwarf_sections.

From-SVN: r278984

PR middle-end/91582 - missing heap overflow detection for strcpy

gcc/ChangeLog:

PR middle-end/91582
* builtins.c (gimple_call_alloc_size): New function.
(compute_objsize): Add argument.  Call gimple_call_alloc_size.
Handle variable offsets and indices.
* builtins.h (gimple_call_alloc_size): Declare.
(compute_objsize): Add argument.
* gcc/gimple-ssa-warn-restrict.c: Remove assertions.
* tree-ssa-strlen.c (handle_store): Handle calls to allocated objects.

gcc/testsuite/ChangeLog:

PR middle-end/91582
* c-c++-common/Wstringop-truncation.c: Remove xfails.
* g++.dg/warn/Wstringop-overflow-4.C: New test.
* g++.dg/ext/attr-alloc_size.C: Suppress -Warray-bounds.
* gcc.dg/Warray-bounds-56.c: New test.
* gcc.dg/Wstringop-overflow-22.c: New test.
* gcc.dg/attr-alloc_size.c: Suppress -Warray-bounds.
* gcc.dg/attr-copy-2.c: Same.
* gcc.dg/builtin-stringop-chk-5.c: Remove xfails.
* gcc.dg/builtin-stringop-chk-8.c: Same.  Correct the text of expected
warnings.
* gcc.target/i386/pr82002-2a.c: Prune expected warning.
* gcc.target/i386/pr82002-2b.c: Same.

From-SVN: r278983

libstdc++: Define pretty printer for comparison categories

* python/libstdcxx/v6/printers.py (StdCmpCatPrinter): New printer.
* testsuite/libstdc++-prettyprinters/cxx20.cc: New test.

From-SVN: r278982

libstdc++: Implement spaceship for std::array (P1614R2)

As done for std::pair, this defines operator<=> as a non-member function
template and does not alter operator==, as expected to be proposed as
the resolution to an unpublished LWG issue.

Instead of calling std::lexicographical_compare_three_way the <=>
overload is implemented by hand to take advantage of the fact the
element types and array sizes are known to be the same.

* include/bits/cpp_type_traits.h (__is_byte<char8_t>): Add
specialization.
* include/std/array (operator<=>): Likewise.
* testsuite/23_containers/array/comparison_operators/constexpr.cc:
Test three-way comparisons and arrays of unsigned char.
* testsuite/23_containers/array/tuple_interface/get_neg.cc: Adjust
dg-error line numbers.

From-SVN: r278981

Daily bump.

From-SVN: r278980

Fix C handling of use of lvalues of incomplete types (PR c/36941, PR c/88827).

Bug 88827 points out that GCC should not be rejecting C code that
dereferences a pointer to an incomplete type in the case that uses &*
to take the address of the resulting lvalue, because no constraint is
violated in that case (other than for C90 when the incomplete type is
unqualified void, which we already handle correctly) and as the lvalue
never gets converted to an rvalue there is no undefined behavior
either.

This means that the diagnostic for such a dereference is bogus and
should be removed; if the lvalue gets converted to an rvalue, there
should be an appropriate error later for the use of the incomplete
type.  In most cases, there is, but bug 36941 points out the lack of a
diagnostic when the incomplete (non-void) type gets cast to void
(where a diagnostic seems appropriate for this undefined behavior as a
matter of quality of implementation).

This patch removes the bogus diagnostic (and C_TYPE_ERROR_REPORTED
which was only used in the code that is removed - only that one, bogus
diagnostic had this duplicate suppression, not any of the other, more
legitimate diagnostics for use of incomplete types) and makes
convert_lvalue_to_rvalue call require_complete_type for arguments not
of void types, so that all relevant code paths (possibly except some
for ObjC) get incomplete types diagnosed.  It's possible that this
makes some other checks for incomplete types obsolete, but no attempt
is made to remove any such checks.

Bootstrapped with no regressions for x86_64-pc-linux-gnu.

PR c/36941
PR c/88827
gcc/c:
* c-typeck.c (convert_lvalue_to_rvalue): Call
require_complete_type for arguments not of void types.
(build_indirect_ref): Do not diagnose dereferencing pointers to
incomplete types.
* c-tree.h (C_TYPE_ERROR_REPORTED): Remove.

gcc/testsuite:
* gcc.dg/lvalue-9.c, gcc.dg/lvalue-10.c: New tests.
* gcc.dg/array-8.c, gcc.dg/enum-incomplete-1.c,
gcc.dg/enum-incomplete-3.c, gcc.dg/noncompile/incomplete-3.c,
gcc.dg/pr48552-1.c, gcc.dg/pr48552-2.c, gcc.dg/pr63543.c,
gcc.dg/pr69796.c: Update expected diagnostics.

From-SVN: r278976

* cp-gimplify.c: Include tm_p.h.

From-SVN: r278975

Use fixed registers for queue ptr sgpr pair

gcc/
* config/gcn/gcn.h (FIXED_REGISTERS): Make s6/s7 fixed registers.

From-SVN: r278974

Do not define builtins that overload disabled builtins.

PR bootstrap/92661
* config/rs6000/rs6000-c.c (struct altivec_builtin_types): Move to
rs6000.h.
(altivec_overloaded_builtins): Move to rs6000-call.c.
* config/rs6000/rs6000.h (struct altivec_builtin_types): Moved from
rs6000-c.c.
* config/rs6000/rs6000-call.c (rs6000_builtin_info): Make static.
(altivec_overloaded_builtins): Moved from rs6000-c.c.
(rs6000_common_init_builtins): Do no define builtins that overload
builtins that have been disabled.

From-SVN: r278973

[ARM] Improve max_cond_insns setting for Cortex cores

To enable cores to use the correct max_cond_insns setting, use the core-specific
tuning when a CPU/tune is selected unless -mrestrict-it is explicitly set.

On Cortex-A57 this gives 1.1% performance gain on SPECINT2006 as well as a
0.4% codesize reduction.

    gcc/
* config/arm/arm.c (arm_option_override_internal):
Use max_cond_insns from CPU tuning unless -mrestrict-it is used.

From-SVN: r278968

Remove stray comma in primary_constraint_error.

* parser.c (enum primary_constraint_error): Remove stray comma.

From-SVN: r278967

[AArch64] Add support for fused compare and branch

Add support for fused compare with branch.  Rename the existing
AARCH64_FUSE_CMP_BRANCH to ALU_BRANCH, and AARCH64_FUSE_ALU_BRANCH
to ALU_CBZ to make it clear what is being fused.

    gcc/
* config/aarch64/aarch64.c
(thunderxt88_tunings): Use AARCH64_FUSE_ALU_BRANCH.
(thunderx_tunings): Likewise.
(tsv110_tunings): Use AARCH64_FUSE_ALU_BRANCH and AARCH64_FUSE_ALU_CBZ.
(thunderx2t99_tunings): Likewise.
(aarch_macro_fusion_pair_p): Add support for AARCH64_FUSE_CMP_BRANCH.
* config/aarch64/aarch64-fusion-pairs.def: Add ALU_CBZ fusion.

From-SVN: r278966

tree-ssa-sccvn.c (vn_reference_lookup_3): Properly guard empty CTOR and memset partial-def registering.

2019-12-04  Richard Biener  <rguenther@suse.de>

* tree-ssa-sccvn.c (vn_reference_lookup_3): Properly guard
empty CTOR and memset partial-def registering.  Take advantage
of fancy offset analysis in memset handling.

From-SVN: r278965

Fix VIEW_CONVERT_EXPRs for VECTOR_BOOLEAN_TYPE_Ps

In r278410 I added code to handle VIEW_CONVERT_EXPRs between
variable-length vectors.  This included support for decoding
a VECTOR_BOOLEAN_TYPE_P with subbyte elements.

However, it turns out that we were already mishandling such bool vectors
for fixed-length vectors: we treated each element as a stand-alone byte
instead of putting multiple elements into the same byte.  I think in
principle this could have been an issue for AVX512 as well.

This patch adds encoding support for boolean vectors and reuses
a version of the new decode support for fixed-length vectors.

2019-12-04  Richard Sandiford  <richard.sandiford@arm.com>

gcc/
* fold-const.c (native_encode_vector_part): Handle
VECTOR_BOOLEAN_TYPE_Ps that have subbyte precision.
(native_decode_vector_tree): Delete, moving the bulk of the code to...
(native_interpret_vector_part): ...this new function.  Use a pointer
and length instead of a vec<> and start index.
(native_interpret_vector): Use native_interpret_vector_part.
(fold_view_convert_vector_encoding): Likewise.

gcc/testsuite/
* gcc.target/aarch64/sve/acle/general/whilelt_5.c: New test.

From-SVN: r278964

tree-ssa-sccvn.c (vn_walk_cb_data::push_partial_def): Handle non-constant defs in the most trivial way.

2019-12-04  Richard Biener  <rguenther@suse.de>

* tree-ssa-sccvn.c (vn_walk_cb_data::push_partial_def): Handle
non-constant defs in the most trivial way.
(vn_reference_lookup_3): Also push down SSA partial defs.

* gcc.dg/tree-ssa/ssa-fre-84.c: New testcase.

From-SVN: r278963

Fortran] PR92754 - fix an issue with resolving intrinsic functions

        gcc/fortran/
        PR fortran/92754
        * intrinsic.c (gfc_intrinsic_func_interface): Set
        sym's flavor, intrinsic and function attribute if
        unset.

        gcc/testsuite/
        PR fortran/92754
        gfortran.dg/intrinsic_9.f90: New.

From-SVN: r278961

Initialize a BB count in switch lowering.

2019-12-04  Martin Liska  <mliska@suse.cz>

* tree-switch-conversion.c (switch_decision_tree::try_switch_expansion):
Initialize count of newly created BB.

From-SVN: r278959