PR 80769: Incorrect strlen optimisation

In this testcase, we (correctly) record after:

  strcpy (p1, "abcde");
  char *p2 = strchr (p1, '\0');
  strcpy (p2, q);

that the length of p1 and p2 can be calculated by converting the
second strcpy to:

  tmp = stpcpy (p2, q)

and then doing tmp - p1 for p1 and tmp - p2 for p2.  This is delayed
until we know whether we actually need it.  Then:

  char *p3 = strchr (p2, '\0');

forces us to calculate the length of p2 in this way.  At this point
we had three related strinfos:

  p1: delayed length, calculated from tmp = stpcpy (p2, q)
  p2: known length, tmp - p2
  p3: known length, 0

After:

  memcpy (p3, "x", 2);

we use adjust_related_strinfos to add 1 to each length.  However,
that didn't do anything for delayed lengths because:

  else if (si->stmt != NULL)
    /* Delayed length computation is unaffected.  */
    ;

So after the memcpy we had:

  p1: delayed length, calculated from tmp = stpcpy (p2, q)
  p2: known length, tmp - p2 + 1
  p3: known length, 1

where the length of p1 was no longer correct.

2017-05-16  Richard Sandiford  <richard.sandiford@linaro.org>

gcc/
PR tree-optimization/80769
* tree-ssa-strlen.c (strinfo): Document that "stmt" is also used
for malloc and calloc.  Document the new invariant that all related
strinfos have delayed lengths or none do.
(verify_related_strinfos): Move earlier in file.
(set_endptr_and_length): New function, split out from...
(get_string_length): ...here.  Also set the lengths of related
strinfos.
(zero_length_string): Assert that chainsi has known (rather than
delayed) lengths.
(adjust_related_strinfos): Likewise.

gcc/testsuite/
PR tree-optimization/80769
* gcc.dg/strlenopt-31.c: New test.
* gcc.dg/strlenopt-31g.c: Likewise.

From-SVN: r249879

PR81136: ICE from inconsistent DR_MISALIGNMENTs

The test case triggered this assert in vect_update_misalignment_for_peel:

      gcc_assert (DR_MISALIGNMENT (dr) / dr_size ==
                  DR_MISALIGNMENT (dr_peel) / dr_peel_size);

The problem was that:

- one memory reference guaranteed a high base alignment, when considering
  that reference in isolation.  This meant that we could calculate the
  vector misalignment for its DR at compile time.

- the other memory reference only guaranteed a low base alignment,
  when considering that reference in isolation.  We therefore couldn't
  calculate the vector misalignment for its DR at compile time.

- when looking at the values of the two addresses as a pair (rather
  than the memory references), it was obvious that they had the same
  misalignment, whatever that misalignment happened to be.

This is working as designed, so the patch restricts the assert to cases
in which both addresses have a compile-time misalignment.

In the test case this looks like a missed opportunity.  Both references
are unconditional, so it should be possible to use the highest of the
available base alignment guarantees when analyzing each reference.
A later patch does this, but the problem would still remain for
conditional references.

2017-07-02  Richard Sandiford  <richard.sandiford@linaro.org>

gcc/
PR tree-optimization/81136
* tree-vect-data-refs.c (vect_update_misalignment_for_peel): Only
assert that two references with the same misalignment have the same
compile-time misalignment if those compile-time misalignments
are known.

gcc/testsuite/
PR tree-optimization/81136
* gcc.dg/vect/pr81136.c: New test.

From-SVN: r249878

Always print attributes when dumping tree

A tree type dump currently doesn't print the attributes. Since we have
so many now and they do many interesting things dumping them can be
useful. So dump them by default for tree type dumps.

gcc/:

2017-07-01  Andi Kleen  <ak@linux.intel.com>

        * print-tree.c (print_node): Print all attributes.

From-SVN: r249877

Daily bump.

From-SVN: r249876

cfg.c (scale_bbs_frequencies): New function.

* cfg.c (scale_bbs_frequencies): New function.
* cfg.h (scale_bbs_frequencies): Declare it.
* cfgloopanal.c (single_likely_exit): Cleanup.
* cfgloopmanip.c (scale_loop_frequencies): Take profile_probability
as parameter.
(scale_loop_profile): Likewise.
(loop_version): Likewise.
(create_empty_loop_on_edge): Update.
* cfgloopmanip.h (scale_loop_frequencies, scale_loop_profile,
scale_loop_frequencies, scale_loop_profile, loopify,
loop_version): Update prototypes.
* modulo-sched.c (sms_schedule): Update.
* predict.c (unlikely_executed_edge_p): Also check probability.
(probably_never_executed_edge_p): Fix typo.
* tree-if-conv.c (version_loop_for_if_conversion): Update.
* tree-parloops.c (gen_parallel_loop): Update.
* tree-ssa-loop-ivcanon.c (try_peel_loop): Update.
* tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Update.
* tree-ssa-loop-split.c (split_loop): Update.
* tree-ssa-loop-unswitch.c (tree_unswitch_loop): Update.
* tree-vect-loop-manip.c (vect_do_peeling): Update.
(vect_loop_versioning): Update.
* tree-vect-loop.c (scale_profile_for_vect_loop): Update.

From-SVN: r249872

* trans-mem.c (split_bb_make_tm_edge): Update profile.

From-SVN: r249871

tree-if-conv.c (combine_blocks): Use make_single_succ_edge to keep profile consistent.

* tree-if-conv.c (combine_blocks): Use make_single_succ_edge
to keep profile consistent.

From-SVN: r249870

cfgrtl.c (rtl_flow_call_edges_add): Update profile.

* cfgrtl.c (rtl_flow_call_edges_add): Update profile.
* tree-cfg.c (gimple_flow_call_edges_add): Likewise.
* profile-count.h (max_safe_multiplier): Make unsigned.
(profile_count::guessed_zero): New.

From-SVN: r249869

bb-reorder.c (fix_up_crossing_landing_pad, [...]): Use make_single_succ_edge to keep profile consistent.

* bb-reorder.c (fix_up_crossing_landing_pad,
fix_crossing_conditional_branches): Use make_single_succ_edge
to keep profile consistent.

From-SVN: r249868

tree-vect-loop.c (optimize_mask_stores): Use make_single_succ_edge to update profile.

* tree-vect-loop.c (optimize_mask_stores): Use make_single_succ_edge
to update profile.

From-SVN: r249867

re PR sanitizer/81262 (verify_flow_info failed for asmgoto test-case with -fsanitize=undefined)

PR sanitizer/81262
* bb-reorder.c (fix_up_fall_thru_edges): Move variable declarations to
the right scopes, make sure cond_jump isn't preserved between multiple
iterations.  Search for fallthru edge whenever there are 3+ edges and
use find_fallthru_edge for it.

From-SVN: r249866

re PR sanitizer/81262 (verify_flow_info failed for asmgoto test-case with -fsanitize=undefined)

PR sanitizer/81262
* bb-reorder.c (fix_up_fall_thru_edges): Move variable declarations to
the right scopes, make sure cond_jump isn't preserved between multiple
iterations.  Search for fallthru edge whenever there are 3+ edges and
use find_fallthru_edge for it.

* gcc.c-torture/compile/pr81262.c: New test.
* g++.dg/ubsan/pr81262.C: New test.

From-SVN: r249865

Patch by Alexander Monakov <amonakov@ispras.ru>

Patch by Alexander Monakov <amonakov@ispras.ru>
* sel-sched-ir.c (compute_succs_info): Handle uninitialized
probabilities consistently.

From-SVN: r249864

sel-sched-ir.c (compute_succs_info): Handle uninitialized probabilities consistently.

* sel-sched-ir.c (compute_succs_info): Handle uninitialized
probabilities consistently.

From-SVN: r249863

pa.c (pa_expand_compare_and_swap_loop): Update call of emit_cmp_and_jump_insns.

* pa.c (pa_expand_compare_and_swap_loop): Update call of
emit_cmp_and_jump_insns.

From-SVN: r249862

Daily bump.

From-SVN: r249861

re PR ipa/81261 (ICE with -fno-guess-branch-probability)

PR ipa/81261
* tree-inline.c (expand_call_inline): Combine profile statuses.

From-SVN: r249856

PR c++/81257 - ICE with invalid ::template.

PR c++/54769 - wrong lookup of dependent template-name.
* parser.c (cp_parser_template_name): Revert part of last change.

From-SVN: r249855

tree-if-conv.c (predicate_scalar_phi): Update new_stmt if fold_stmt returned true.

2017-06-30  Andrew Pinski  <apinski@cavium.com>

        * tree-if-conv.c (predicate_scalar_phi): Update new_stmt if
        fold_stmt returned true.
2017-06-30  Andrew Pinski  <apinski@cavium.com>

        * gcc.dg/torture/pr81245.c: New testcase.

From-SVN: r249854

config-lang.in (gtfiles): Add cp/lex.c.

* config-lang.in (gtfiles): Add cp/lex.c.
* cp-tree.h (mangle_convop_name_for_type): Rename ...
(make_conv_op_name): ... here.  Move to lex.
* lambda.c (maybe_add_lambda_conv_op): Update.
* parser.c (cp_parser_conversion_function_id): Update.
* pt.c (tsubst_decl, tsubst_baselink, tsubst_copy,
tsubst_copy_and_build): Update.
* semantics.c (apply_deduced_return_type): Update.
* mangle.c (conv_type_hasher, conv_type_names,
mangle_conv_op_name_for_type): Move to ...
* lex.c (conv_type_hasher, conv_type_names, make_convop_name):
... here.  Rename.

* libcp1plugin.cc (plugin_build_decl): Use make_conv_op_name.
(plugin_build_dependent_expr): Likewise.

From-SVN: r249852

ggc.h (empty_string): Delete.

* ggc.h (empty_string): Delete.
* cfgexpand.c (expand_asm_stmt): Use plain "".
* optabs.c (expand_asm_memory_barrier): Likewise.
* stringpool.c (empty_string): Delete.
(digit_vector, digit_string): Delete.
(ggc_alloc_string): Use plain "", don't optimize single digit
strings.  Use ggc_alloc_atomic.

From-SVN: r249851

[rtlanal] Do a better job of costing parallel sets containing flag-setting operations.

Many parallel set insns are of the form of a single set that also sets
the condition code flags.  In this case the cost of such an insn is
normally the cost of the part that doesn't set the flags, since
updating the condition flags is simply a side effect.

At present all such insns are treated as having unknown cost (ie 0)
and combine assumes that such insns are infinitely more expensive than
any other insn sequence with a non-zero cost.

This patch addresses this problem by allowing insn_rtx_cost to ignore
the condition setting part of a PARALLEL iff there is exactly one
comparison set and one non-comparison set.  If the only set operation
is a comparison we still use that as the basis of the insn cost.

* rtlanal.c (insn_rtx_cost): If a parallel contains exactly one
comparison set and one other set, use the cost of the non-comparison
set.

Bootstrapped on aarch64-none-linuxgnu

From-SVN: r249850

ggc.h: Replace all 'static inline' with plain 'inline'.

* ggc.h: Replace all 'static inline' with plain 'inline'.  Fix
some formatting.

From-SVN: r249848

tree-cfg.c (group_case_labels_stmt): Merge scanning and compressing loops.

* tree-cfg.c (group_case_labels_stmt): Merge scanning and compressing
loops.  Remove now unneeded calls to gimple_switch_set_label() that
just set removed labels to NULL_TREE.

From-SVN: r249847

tree-ssanames.c (set_range_info_raw): Abstract from ...

* tree-ssanames.c (set_range_info_raw): Abstract from ...
(set_range_info): ...here.  Only call set_range_info_raw if domain
is useful.
(set_nonzero_bits): Call set_range_info_raw.
* tree-ssanames.h (set_range_info_raw): New.
testsuite/
* gcc.dg/Walloca-14.c: Adapt test to recognize new complaint of
unbounded use.

From-SVN: r249846

Fix location of typeid() (PR c++/80014)

gcc/cp/ChangeLog:
PR c++/80014
* parser.c (cp_parser_postfix_expression): Construct a location
for typeid expressions.

gcc/testsuite/ChangeLog:
PR c++/80014
* g++.dg/plugin/diagnostic-test-expressions-1.C (std::type_info):
Add declaration.
(test_typeid): New test function.

From-SVN: r249845

re PR target/81225 (ICE with -mavx512ifma -O3 -ffloat-store)

PR target/81225
* config/i386/sse.md (vec_extract_lo_<mode><mask_name>): For
V8FI, V16FI and VI8F_256 iterators, use <store_mask_predicate> instead
of nonimmediate_operand and <store_mask_constraint> instead of m for
the input operand.  For V8FI iterator, always split if input is a MEM.
For V16FI and V8SF_256 iterators, don't test if both operands are MEM
if <mask_applied>.  For VI4F_256 iterator, use <store_mask_predicate>
instead of register_operand and <store_mask_constraint> instead of v for
        the input operand.  Make sure both operands aren't MEMs for if not
<mask_applied>.

* gcc.target/i386/pr81225.c: New test.

From-SVN: r249844

cp-tree.h (lookup_fnfields_1, [...]): Don't declare.

* cp-tree.h (lookup_fnfields_1, class_method_index_for_fn): Don't
declare.
(lookup_all_conversions): Declare.
* class.c (get_basefndecls): Use lookup_fnfields_slot.
* decl.c (register_dtor_fn): Use lookup_fnfields_slot.
* decl2.c (check_class_fn): Use lookup_fnfields_slot.  Rework
diagnostics.
* pt.c (retrieve_specialization): Use lookup_fnfields_slot.
(check_explicit_specialization): Use lookup_fnfields_slot_nolazy,
lookup_all_conversions.
* search.c (lookup_fnfields_1): Make static.
(lookup_all_conversions): New.
(class_method_index_for_fn): Delete.
* semantics.c (classtype_has_nothrow_assign_or_copy_p): Use
lookup_fnfields_slot.

* g++.dg/concepts/memfun-err.C: Adjust diagnostics.
* g++.dg/cpp0x/decltype9.C: Likewise.
* g++.dg/cpp0x/forw_enum9.C: Likewise.
* g++.dg/lookup/decl1.C: Likewise.
* g++.dg/lookup/extern-c-redecl.C: Likewise.
* g++.dg/other/pr28432.C: Likewise.
* g++.dg/parse/crash12.C: Likewise.
* g++.dg/parse/enum3.C: Likewise.
* g++.dg/parse/operator6.C: Likewise.
* g++.dg/template/crash69.C: Likewise.
* g++.dg/template/error27.C: Likewise.
* g++.dg/template/error28.C: Likewise.
* g++.dg/template/memfriend6.C: Likewise.
* g++.old-deja/g++.mike/err1.C: Likewise.
* g++.old-deja/g++.mike/p811.C: Likewise.
* g++.old-deja/g++.other/crash25.C: Likewise.
* g++.old-deja/g++.other/dtor4.C: Likewise.
* g++.old-deja/g++.pt/t37.C: Likewise.

From-SVN: r249843

* lto-wrapper.c (copy_file) Close both file descriptors before
exiting normally.

From-SVN: r249841

Fix removal of ifunc (PR ipa/81214).

2017-06-30  Martin Liska  <mliska@suse.cz>

PR ipa/81214
* gcc.target/i386/pr81214.c: New test.
2017-06-30  Martin Liska  <mliska@suse.cz>

PR ipa/81214
* multiple_target.c (create_dispatcher_calls): Make ifunc
also for function that don't have calls or are not referenced.

From-SVN: r249840

tree-vect-slp.c (vect_slp_analyze_node_operations): Only analyze the first scalar stmt.

2017-06-30  Richard Biener  <rguenther@suse.de>

* tree-vect-slp.c (vect_slp_analyze_node_operations): Only
analyze the first scalar stmt.  Move vector type computation
for the BB case here from ...
* tree-vect-stmts.c (vect_analyze_stmt): ... here.  Guard
live operation processing in the SLP case properly.

From-SVN: r249839

call.c (build_new_method_call_1): Use constructo_name to get ctor name.

* call.c (build_new_method_call_1): Use constructo_name to get
ctor name.  Move argument processing earlier to merge cdtor
handling blocks.
* decl.c (grokfndecl): Cdtors have special names.
* method.c (implicitly_declare_fn): Likewise. Simplify flag setting.
* pt.c (check_explicit_specialization): Cdtor name is already
special.
* search.c (class_method_index_for_fn): Likewise.

* g++.dg/plugin/decl-plugin-test.C: Expect special ctor name.

From-SVN: r249838

re PR c++/81229 (ICE in c_tree_chain_next on aarch64)

PR c++/81229
* name-lookup.c (do_pushdecl): Reset IDENTIFIER_TYPE when finding
a matching TYPE_DECL.

* g++.dg/lookup/pr81229.C: New.

From-SVN: r249835

graph.c (draw_cfg_node_succ_edges): Fix broken dot syntax.

2017-06-30  Richard Biener  <rguenther@suse.de>

* graph.c (draw_cfg_node_succ_edges): Fix broken dot syntax.

From-SVN: r249834

Call BUILT_IN_ASAN_HANDLE_NO_RETURN before BUILT_IN_UNWIND_RESUME (PR sanitizer/81021).

2017-06-30  Martin Liska  <mliska@suse.cz>

PR sanitizer/81021
* g++.dg/asan/pr81021.C: New test.
2017-06-30  Martin Liska  <mliska@suse.cz>

PR sanitizer/81021
* tree-eh.c (lower_resx): Call BUILT_IN_ASAN_HANDLE_NO_RETURN
before BUILT_IN_UNWIND_RESUME when ASAN is used.

From-SVN: r249833

Documentation: Clean up AArch64 options.

gcc/
2017-06-30  Yvan Roux  <yvan.roux@linaro.org>

* doc/invoke.texi (AArch64): Add missing options and remove redundant
ones.

From-SVN: r249832

re PR tree-optimization/81249 (ICE: error: incompatible types in PHI argument 0)

2017-06-30  Richard Biener  <rguenther@suse.de>

PR tree-optimization/81249
* tree-vect-loop.c (vect_create_epilog_for_reduction): Convert
condition reduction result to original scalar type.

* g++.dg/torture/pr81249.C: New testcase.

From-SVN: r249831

profile-count.h: Fix typos and whitespace issues.

gcc/ChangeLog:

2017-06-30  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>

* profile-count.h (enum profile_quality): Fix typos and whitespace
issues.

From-SVN: r249830

S/390: Adjust to the recent branch probability changes.

This fixes the bootstrap failure triggered by the recent changes wrt
branch probabilities aka emit_cmp_and_jump_insns does not accept
integers as branch probability anymore.

gcc/ChangeLog:

2017-06-30  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>

* config/s390/s390.c (s390_expand_setmem): Adjust to the new data
type for branch probabilities.

From-SVN: r249829

aarch64-fusion-pairs.def: Add ALU_BRANCH entry.

2017-06-29  Julian Brown  <julian@codesourcery.com>
    Naveen H.S  <Naveen.Hurugalawadi@cavium.com>

* config/aarch64/aarch64-fusion-pairs.def: Add ALU_BRANCH entry.
* config/aarch64/aarch64.c (AARCH64_FUSE_ALU_BRANCH): New fusion type.
(thunderx2t99_tunings): Set AARCH64_FUSE_ALU_BRANCH flag.
(aarch_macro_fusion_pair_p): Add support for AARCH64_FUSE_ALU_BRANCH.

Co-Authored-By: Naveen H.S <Naveen.Hurugalawadi@cavium.com>
From-SVN: r249828

aarch64.c (aarch_macro_fusion_pair_p): Push the check for CC usage into AARCH64_FUSE_CMP_BRANCH.

2017-06-29  Naveen H.S  <Naveen.Hurugalawadi@cavium.com>

* config/aarch64/aarch64.c (aarch_macro_fusion_pair_p): Push the
check for CC usage into AARCH64_FUSE_CMP_BRANCH.
* config/i386/i386.c (ix86_macro_fusion_pair_p): Push the check for
CC usage from generic code to here.
* sched-deps.c (sched_macro_fuse_insns): Move the condition for
CC usage into the target macros.

From-SVN: r249827

Daily bump.

From-SVN: r249826

netbsd.h (NETBSD_LIB_SPEC): Add -lc when creating shared objects.

2017-06-29  Maya Rashish  <coypu@sdf.org>

* config/netbsd.h (NETBSD_LIB_SPEC): Add -lc when creating shared
objects.

From-SVN: r249822

arm-builtins.c: Include profile-count.h

* arm/arm-builtins.c: Include profile-count.h
* except.c (sjlj_emit_function_enter): Use
profile_probability::unlikely.

From-SVN: r249821

class.c (add_method): Change pair of errors to error + inform.

/cp
2017-06-29  Paolo Carlini  <paolo.carlini@oracle.com>

* class.c (add_method): Change pair of errors to error + inform.
(handle_using_decl): Likewise.

/testsuite
2017-06-29  Paolo Carlini  <paolo.carlini@oracle.com>

* g++.dg/cpp0x/inh-ctor3.C: Adjust for dg-message vs dg-error.
* g++.dg/diagnostic/variadic1.C: Likewise.
* g++.dg/gomp/udr-3.C: Likewise.
* g++.dg/overload/error1.C: Likewise.
* g++.dg/overload/error2.C: Likewise.
* g++.dg/template/duplicate1.C: Likewise.
* g++.old-deja/g++.benjamin/warn02.C: Likewise.
* g++.old-deja/g++.brendan/arm2.C: Likewise.
* g++.old-deja/g++.other/redecl2.C: Likewise.
* g++.old-deja/g++.other/redecl4.C: Likewise.
* g++.old-deja/g++.pt/memtemp78.C: Likewise.

From-SVN: r249820

* constexpr.c, error.c, tree.c: Remove WITH_CLEANUP_EXPR handling.

From-SVN: r249818

PR c++/81180 - ICE with C++17 deduction of member class template.

* pt.c (build_deduction_guide): Correct member template handling.

From-SVN: r249816

rs6000.c (toc_relative_expr_p): Make tocrel_base and tocrel_offset be pointer args rather than implicitly using...

2017-06-29  Aaron Sawdey  <acsawdey@linux.vnet.ibm.com>

        * config/rs6000/rs6000.c (toc_relative_expr_p): Make tocrel_base
        and tocrel_offset be pointer args rather than implicitly using
        static versions.
        (legitimate_constant_pool_address_p, rs6000_emit_move,
        const_load_sequence_p, adjust_vperm): Add local tocrel_base and
        tocrel_offset and use in toc_relative_expr_p call.
        (print_operand, print_operand_address): Use static tocrel_base_oac
        and tocrel_offset_oac.
        (rs6000_output_addr_const_extra): Use static tocrel_base_oac and
        tocrel_offset_oac.

From-SVN: r249815

PR c++/81188 - matching decltype of member function call.

* tree.c (cp_tree_equal): Remove COMPONENT_REF special case.

From-SVN: r249813

builtins.md (ffssi2_internal): Correct constraint.

2017-06-29  Maya Rashish  <coypu@sdf.org>

* config/vax/builtins.md (ffssi2_internal): Correct constraint.

From-SVN: r249812

expr.c (expand_expr): When testing for unaligned objects...

* expr.c (expand_expr) <normal_inner_ref>: When testing for unaligned
objects, take into account only the alignment of 'op0' and 'mode1' if
'op0' is a MEM.

From-SVN: r249811

re PR fortran/77765 (ICE in gfc_match_oacc_routine, at fortran/openmp.c:1781)

PR fortran/77765

gcc/fortran/
* openmp.c (gfc_match_oacc_routine): Check if proc_name exist before
comparing the routine name against it.

gcc/testsuite/
* gfortran.dg/goacc/pr77765.f90: New test.

From-SVN: r249809

ccmp_2.c: New test.

2017-06-29  Steve Ellcey  <sellcey@cavium.com>

* gcc.target/aarch64/ccmp_2.c: New test.

From-SVN: r249806

ccmp.c (ccmp_tree_comparison_p): New function.

2017-06-29  Steve Ellcey  <sellcey@cavium.com>

* ccmp.c (ccmp_tree_comparison_p): New function.
(ccmp_candidate_p): Update to use above function.
(get_compare_parts): New function.
(expand_ccmp_next): Update to use new functions.
(expand_ccmp_expr_1): Take tree arg instead of gimple, update to use
new functions.
(expand_ccmp_expr): Pass tree instead of gimple to expand_ccmp_expr_1,
take mode as argument.
* ccmp.h (expand_ccmp_expr): Add mode as argument.
* expr.c (expand_expr_real_1): Pass mode as argument.

From-SVN: r249805

re PR c++/81247 (ICE on invalid C++ code with malformed namespace declaration: in do_push_nested_namespace, at cp/name-lookup.c:6002)

PR c++/81247
* parser.c (cp_parser_namespace_definition): Immediately close the
namespace if there's no open-brace.
* name-lookup.c (do_pushdecl): Reset OLD when pushing into new
namespace.

From-SVN: r249804

bid2dpd_dpd2bid.c (_bid_to_dpd32): Fix whitespace.

* bid/bid2dpd_dpd2bid.c (_bid_to_dpd32): Fix whitespace.
(_dpd_to_bid32): Simplify identical code on multiple branches.
Fix whitespace.
(_bid_to_dpd64, _dpd_to_bid64): Likewise.
(_bid_to_dpd128, _dpd_to_bid128): Likewise.

From-SVN: r249803

combine: Print insns with the cost dump

In the combine dump file, at the start there is a list of the RTL cost
of every insn.  The only thing listed about the insns is the UID though.
To make it more useful, this patch prints the insn itself as well (in
slim format).

* combine.c (combine_instructions): Print insns to dump_file, together
with their costs.

From-SVN: r249802

asan.c (asan_emit_stack_protection): Update.

* asan.c (asan_emit_stack_protection): Update.
(create_cond_insert_point): Update.
* auto-profile.c (afdo_propagate_circuit): Update.
* basic-block.h (struct edge_def): Turn probability to
profile_probability.
(EDGE_FREQUENCY): Update.
* bb-reorder.c (find_traces_1_round): Update.
(better_edge_p): Update.
(sanitize_hot_paths): Update.
* cfg.c (unchecked_make_edge): Initialize probability to uninitialized.
(make_single_succ_edge): Update.
(check_bb_profile): Update.
(dump_edge_info): Update.
(update_bb_profile_for_threading): Update.
* cfganal.c (connect_infinite_loops_to_exit): Initialize new edge
probabilitycount to 0.
* cfgbuild.c (compute_outgoing_frequencies): Update.
* cfgcleanup.c (try_forward_edges): Update.
(outgoing_edges_match): Update.
(try_crossjump_to_edge): Update.
* cfgexpand.c (expand_gimple_cond): Update make_single_succ_edge.
(expand_gimple_tailcall): Update.
(construct_init_block): Use make_single_succ_edge.
(construct_exit_block): Use make_single_succ_edge.
* cfghooks.c (verify_flow_info): Update.
(redirect_edge_succ_nodup): Update.
(split_edge): Update.
(account_profile_record): Update.
* cfgloopanal.c (single_likely_exit): Update.
* cfgloopmanip.c (scale_loop_profile): Update.
(set_zero_probability): Remove.
(duplicate_loop_to_header_edge): Update.
* cfgloopmanip.h (loop_version): Update prototype.
* cfgrtl.c (try_redirect_by_replacing_jump): Update.
(force_nonfallthru_and_redirect): Update.
(update_br_prob_note): Update.
(rtl_verify_edges): Update.
(purge_dead_edges): Update.
(rtl_lv_add_condition_to_bb): Update.
* cgraph.c: (cgraph_edge::redirect_call_stmt_to_calle): Update.
* cgraphunit.c (init_lowered_empty_function): Update.
(cgraph_node::expand_thunk): Update.
* cilk-common.c: Include profile-count.h
* dojump.c (inv): Remove.
(jumpifnot): Update.
(jumpifnot_1): Update.
(do_jump_1): Update.
(do_jump): Update.
(do_jump_by_parts_greater_rtx): Update.
(do_compare_rtx_and_jump): Update.
* dojump.h (jumpifnot, jumpifnot_1, jumpif_1, jumpif, do_jump,
do_jump_1. do_compare_rtx_and_jump): Update prototype.
* dwarf2cfi.c: Include profile-count.h
* except.c (dw2_build_landing_pads): Use make_single_succ_edge.
(sjlj_emit_dispatch_table): Likewise.
* explow.c: Include profile-count.h
* expmed.c (emit_store_flag_force): Update.
(do_cmp_and_jump): Update.
* expr.c (compare_by_pieces_d::generate): Update.
(compare_by_pieces_d::finish_mode): Update.
(emit_block_move_via_loop): Update.
(store_expr_with_bounds): Update.
(store_constructor): Update.
(expand_expr_real_2): Update.
(expand_expr_real_1): Update.
* expr.h (try_casesi, try_tablejump): Update prototypes.
* gimple-pretty-print.c (dump_probability): Update.
(dump_profile): New.
(dump_gimple_label): Update.
(dump_gimple_bb_header): Update.
* graph.c (draw_cfg_node_succ_edges): Update.
* hsa-gen.c (convert_switch_statements): Update.
* ifcvt.c (cheap_bb_rtx_cost_p): Update.
(find_if_case_1): Update.
(find_if_case_2): Update.
* internal-fn.c (expand_arith_overflow_result_store): Update.
(expand_addsub_overflow): Update.
(expand_neg_overflow): Update.
(expand_mul_overflow): Update.
(expand_vector_ubsan_overflow): Update.
* ipa-cp.c (good_cloning_opportunity_p): Update.
* ipa-split.c (split_function): Use make_single_succ_edge.
* ipa-utils.c (ipa_merge_profiles): Update.
* loop-doloop.c (add_test): Update.
(doloop_modify): Update.
* loop-unroll.c (compare_and_jump_seq): Update.
(unroll_loop_runtime_iterations): Update.
* lra-constraints.c (lra_inheritance): Update.
* lto-streamer-in.c (input_cfg): Update.
* lto-streamer-out.c (output_cfg): Update.
* mcf.c (adjust_cfg_counts): Update.
* modulo-sched.c (sms_schedule): Update.
* omp-expand.c (expand_omp_for_init_counts): Update.
(extract_omp_for_update_vars): Update.
(expand_omp_ordered_sink): Update.
(expand_omp_for_ordered_loops): Update.
(expand_omp_for_generic): Update.
(expand_omp_for_static_nochunk): Update.
(expand_omp_for_static_chunk): Update.
(expand_cilk_for): Update.
(expand_omp_simd): Update.
(expand_omp_taskloop_for_outer): Update.
(expand_omp_taskloop_for_inner): Update.
* omp-simd-clone.c (simd_clone_adjust): Update.
* optabs.c (expand_doubleword_shift): Update.
(expand_abs): Update.
(emit_cmp_and_jump_insn_1): Update.
(expand_compare_and_swap_loop): Update.
* optabs.h (emit_cmp_and_jump_insns): Update prototype.
* predict.c (predictable_edge_p): Update.
(edge_probability_reliable_p): Update.
(set_even_probabilities): Update.
(combine_predictions_for_insn): Update.
(combine_predictions_for_bb): Update.
(propagate_freq): Update.
(estimate_bb_frequencies): Update.
(force_edge_cold): Update.
* profile-count.c (profile_count::dump): Add missing space into dump.
(profile_count::debug): Add newline.
(profile_count::differs_from_p): Explicitly convert to unsigned.
(profile_count::stream_in): Update.
(profile_probability::dump): New member function.
(profile_probability::debug): New member function.
(profile_probability::differs_from_p): New member function.
(profile_probability::differs_lot_from_p): New member function.
(profile_probability::stream_in): New member function.
(profile_probability::stream_out): New member function.
* profile-count.h (profile_count_quality): Rename to ...
(profile_quality): ... this one.
(profile_probability): New.
(profile_count): Update.
* profile.c (compute_branch_probabilities): Update.
* recog.c (peep2_attempt): Update.
* sched-ebb.c (schedule_ebbs): Update.
* sched-rgn.c (find_single_block_region): Update.
(compute_dom_prob_ps): Update.
(schedule_region): Update.
* sel-sched-ir.c (compute_succs_info): Update.
* stmt.c (struct case_node): Update.
(do_jump_if_equal): Update.
(get_outgoing_edge_probs): Update.
(conditional_probability): Update.
(emit_case_dispatch_table): Update.
(expand_case): Update.
(expand_sjlj_dispatch_table): Update.
(emit_case_nodes): Update.
* targhooks.c: Update.
* tracer.c (better_p): Update.
(find_best_successor): Update.
* trans-mem.c (expand_transaction): Update.
* tree-call-cdce.c: Update.
* tree-cfg.c (gimple_split_edge): Upate.
(move_sese_region_to_fn): Upate.
* tree-cfgcleanup.c (cleanup_control_expr_graph): Upate.
* tree-eh.c (lower_resx): Upate.
(cleanup_empty_eh_move_lp): Upate.
* tree-if-conv.c (version_loop_for_if_conversion): Update.
* tree-inline.c (copy_edges_for_bb): Update.
(copy_cfg_body): Update.
* tree-parloops.c (gen_parallel_loop): Update.
* tree-profile.c (gimple_gen_ic_func_profiler): Update.
(gimple_gen_time_profiler): Update.
* tree-ssa-dce.c (remove_dead_stmt): Update.
* tree-ssa-ifcombine.c (update_profile_after_ifcombine): Update.
* tree-ssa-loop-im.c (execute_sm_if_changed): Update.
* tree-ssa-loop-ivcanon.c (remove_exits_and_undefined_stmts): Update.
(unloop_loops): Update.
(try_peel_loop): Update.
* tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Update.
* tree-ssa-loop-split.c (connect_loops): Update.
(split_loop): Update.
* tree-ssa-loop-unswitch.c (tree_unswitch_loop): Update.
(hoist_guard): Update.
* tree-ssa-phionlycprop.c (propagate_rhs_into_lhs): Update.
* tree-ssa-phiopt.c (replace_phi_edge_with_variable): Update.
(value_replacement): Update.
* tree-ssa-reassoc.c (branch_fixup): Update.
* tree-ssa-tail-merge.c (replace_block_by): Update.
* tree-ssa-threadupdate.c (remove_ctrl_stmt_and_useless_edges): Update.
(create_edge_and_update_destination_phis): Update.
(compute_path_counts): Update.
(recompute_probabilities): Update.
(update_joiner_offpath_counts): Update.
(freqs_to_counts_path): Update.
(duplicate_thread_path): Update.
* tree-switch-conversion.c (hoist_edge_and_branch_if_true): Update.
(struct switch_conv_info): Update.
(gen_inbound_check): Update.
* tree-vect-loop-manip.c (slpeel_add_loop_guard): Update.
(vect_do_peeling): Update.
(vect_loop_versioning): Update.
* tree-vect-loop.c (scale_profile_for_vect_loop): Update.
(optimize_mask_stores): Update.
* ubsan.c (ubsan_expand_null_ifn): Update.
* value-prof.c (gimple_divmod_fixed_value): Update.
(gimple_divmod_fixed_value_transform): Update.
(gimple_mod_pow2): Update.
(gimple_mod_pow2_value_transform): Update.
(gimple_mod_subtract): Update.
(gimple_mod_subtract_transform): Update.
(gimple_ic): Update.
(gimple_stringop_fixed_value): Update.
(gimple_stringops_transform): Update.
* value-prof.h: Update.

From-SVN: r249800

Makefile.am (MOSTLYCLEANFILES): Remove testing files and logs.

* Makefile.am (MOSTLYCLEANFILES): Remove testing files and logs.
(mostlyclean-local): Remove check-runtime-dir, cgo-test-dir,
carchive-test-dir.
(ECHO_ENV): Define.
(check-go-tool): Depend on cgo.  Write command to testlog.
(check-runtime): New target.
(check-cgo-test): New target.
(check-carchive-test): New target.
(check): Depend on check-runtime, check-cgo-test,
check-carchive-test.  Add @ to prettify output.
(.PHONY): Add check-runtime, check-cgo-test, check-carchive-test.
* Makefile.in: Rebuild.

From-SVN: r249799

rs6000-c.c: Add support for built-in functions vector signed int vec_signed (vector float)...

gcc/ChangeLog:

2017-06-29  Carl Love  <cel@us.ibm.com>

* config/rs6000/rs6000-c.c: Add support for built-in functions
vector signed int vec_signed (vector float);
vector signed long long vec_signed (vector double);
vector signed int vec_signed2 (vector double, vector double);
vector signed int vec_signede (vector double);
vector signed int vec_signedo (vector double);
* config/rs6000/rs6000.c (rs6000_generate_vsigned2_code): Add
instruction generator.
* config/rs6000/vsx.md (UNSPEC_VSX_XVCVSPSXWS, UNSPEC_VSX_XVCVSPSXDS,
UNSPEC_VSX_VSIGNED2): Add UNSPECS.
(vsx_xvcvspsxws, vsx_xvcvdpuxds_scale, vsx_xvcvspuxws, vsigned2_v2df):
Add define_insn.
(vsignedo_v2df, vsignede_v2df, vunsigned2_v2df, vunsignedo_v2df,
vunsignede_v2df): Add define_expands.
* config/rs6000/rs6000-builtin.def (VEC_SIGNED, VEC_UNSIGNED,
VEC_SIGNED2, VEC_UNSIGNED2, VEC_SIGNEDE, VEC_UNSIGNEDE, VEC_SIGNEDO,
VEC_UNSIGNEDO): Add definitions.
* config/vsx.md (UNSPEC_VSX_XVCVSPSXWS, UNSPEC_VSX_XVCVSPSXDS,
UNSPEC_VSX_VSIGNED2): Add UNSPECs.
(vsx_xvcvspsxws, vsx_xvcvspuxws): Add define_insn.
(vsigned2_v2df, vsigendo_v2df, vsignede_v2df,
vunsigned2_v2df, vunsignedo_v2df, vunsignede_v2df): Add define_expands.
* config/rs6000/altivec.h (vec_signed, vec_signed2,
vec_signede and vec_signedo, vec_unsigned, vec_unsigned2,
vec_unsignede, vec_unsignedo): Add builtin defines.
* config/rs6000-protos.h (rs6000_generate_vsigned2_code): Add extern
declaration.
* doc/extend.texi: Update the built-in documentation file for the
new built-in functions.

gcc/testsuite/ChangeLog:

2017-06-29  Carl Love  <cel@us.ibm.com>

* gcc.target/powerpc/builtins-3-runnable.c (test_int_result,
test_unsigned_int_result, test_ll_int_result,
test_ll_unsigned_int_result): Add result checking functions, add
debug support.
(main): Add builtin function tests.

From-SVN: r249798

PR c++/81164 - ICE with invalid inherited constructor.

* search.c (binfo_direct_p): New.
* name-lookup.c (do_class_using_decl): Use it.

From-SVN: r249797

misc/cgo/testcarchive: fix test to work for gccgo

    This test is not yet run, but it will be soon.

    Reviewed-on: https://go-review.googlesource.com/47038

From-SVN: r249795

cmd/go: fix -buildmode={c-archive,c-shared,pie} for gccgo

    The tests are misc/cgo tests that are not currently run but will be
    run soon.

    Reviewed-on: https://go-review.googlesource.com/47037

From-SVN: r249794

builtins-3-vec_reve-runnable.c (dg-options, dg-skip-if): Add -mvsx -mcpu=power7.

gcc/testsuite/ChangeLog:

2017-06-29  Carl Love  <cel@us.ibm.com>

* gcc.target/powerpc/builtins-3-vec_reve-runnable.c (dg-options,
dg-skip-if): Add -mvsx -mcpu=power7.

From-SVN: r249792

cp-tree.h (THIS_NAME, [...]): Delete.

* cp-tree.h (THIS_NAME, IN_CHARGE_NAME, VTBL_PTR_TYPE,
VTABLE_DELTA_NAME, VTABLE_PFN_NAME): Delete.
* decl.c (initialize_predefined_identifiers): Name cdtor special
names consistently.  Use literals for above deleted defines.
(cxx_init_decl_processing): Use literal for vtbl_ptr_type name,

From-SVN: r249791

lex.c (maybe_add_lang_type_raw): Exit early, rather than use a flag.

* lex.c (maybe_add_lang_type_raw): Exit early, rather than use a
flag.

From-SVN: r249790

call.c (check_dtor_name): Use constructor_name for enums too.

* call.c (check_dtor_name): Use constructor_name for enums too.
(build_new_method_call_1): Use constructor_name for cdtors and
show ~ for dtor.
* class.c (build_self_reference): Use TYPE_NAME to get name of
self reference.
* name-lookup (constructor_name): Use DECL_NAME directly.
(constructor_name_p): Reimplement.
(push_class_level_binding_1): Use TYPE_NAME directly.

From-SVN: r249789

class.c (finish_struct): Use OVL_P.

* class.c (finish_struct): Use OVL_P.
(get_vfield_name): Measure constructor_name length.
* cp-tree.h (SET_CLASS_TYPE_P): Add RECORD_OR_UNION_CHECK.
(NON_UNION_CLASS_TYPE_P): Check RECORD_TYPE up front.
* cxx-pretty-print.c (is_destructor_name): Delete.
(pp_cxx_unqualified_id): Remove bogus destructor name checking.
* decl.c (grokfndecl): Move cheap checks first when looking for
implicit extern cness.

From-SVN: r249788

parser.c (cp_parser_direct_declarator): Reorder if to avoid indentation.

* parser.c (cp_parser_direct_declarator): Reorder if to avoid
indentation. Remove unnecessary assignment of constructor name.

From-SVN: r249787

Whitespace cleanups.

* call.c (name_as_c_string): Move CONST_CAST to return.
(build_new_method_call_1): Remove unneeded bracing.
* class.c (include_empty_classes): Unbreak line.
* constraint.cc (tsubst_check_constraint): Add space.
* cp-tree.h (lang_decl_ns): Add comment.
(PTRMEM_CST_MEMBER): Break line.
* decl.c (grokfndecl): Add blank lines. Unbreak some others.
(grokdeclarator): Remove lines, move declaration to first use.
* decl2.c (decl_needed_p): Fix indentation.
(c_parse_final_cleanups): Remove blank line.
* method.c (implicitly_declare_fn): Move declaration to first use.
* search.c (current_scope): Add blank lines.

From-SVN: r249786

tree-vect-loop.c (vect_analyze_scalar_cycles_1): Do not add reduction chains to LOOP_VINFO_REDUCTIONS.

2017-06-29  Richard Biener  <rguenther@suse.de>

* tree-vect-loop.c (vect_analyze_scalar_cycles_1): Do not add
reduction chains to LOOP_VINFO_REDUCTIONS.
* tree-vect-slp.c (vect_analyze_slp): Continue looking for
SLP reductions after processing reduction chains.

* gcc.dg/vect/slp-reduc-8.c: New testcase.

From-SVN: r249785

builtins.c (fold_builtin_FUNCTION): Use lang_hooks.decl_printable_name.

gcc/
* builtins.c (fold_builtin_FUNCTION): Use
lang_hooks.decl_printable_name.

gcc/cp/
* g++.dg/cpp1y/builtin_FUNCTION.C: New.

From-SVN: r249784

re PR middle-end/81194 (ICE during RTL pass: expand)

gcc/
PR middle-end/81194
* cfgexpand.c (expand_gimple_stmt_1): Handle switch statements
with only one label.
* stmt.c (expand_case): Assert NCASES is greater than one.

gcc/testsuite/
PR middle-end/81194
* g++.dg/pr81194.C: New test.

From-SVN: r249783

Add a testcase for PR ipa/80565

The testcase in

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=80565

is passing now.  Check in the testcase and resolve it as fixed.

PR ipa/80565
* gcc.dg/torture/pr80565.c: New file.

From-SVN: r249782

tree-cfg.c (group_case_labels_stmt): Return whether we changed anything.

2017-06-29  Richard Biener  <rguenther@suse.de>

* tree-cfg.c (group_case_labels_stmt): Return whether we changed
anything.
(group_case_labels): Likewise.
(find_taken_edge): Push sanity checking on val to workers...
(find_taken_edge_cond_expr): ... here
(find_taken_edge_switch_expr): ... and here, handle cases
with just a default label.
* tree-cfg.h (group_case_labels_stmt): Adjust prototype.
(group_case_labels): Likewise.
* tree-cfgcleanup.c (execute_cleanup_cfg_post_optimizing): When
group_case_labels does anything cleanup the CFG again.

From-SVN: r249780

re PR tree-optimization/81196 (Number of iterations found for p!=q but not for p<q)

PR tree-optimization/81196
* tree-ssa-loop-niter.c (number_of_iterations_cond): Handle loop
exit condition comparing two IVs.

gcc/testsuite
* gcc.dg/vect/pr81196.c: New.

From-SVN: r249778

[arm] Fix bootstrap - missing initializer in tail entry of autogenerated code

My patch yesterday accidentally missed a hunk that added the
update to the tail entry of the autogenerated data structure
produced by parsecpu.awk.  This causes native bootstraps to
fail.

This patch adds back the missing hunk.

2017-06-29  Richard Earnshaw  <rearnsha@arm.com>

* config/arm/parsecpu.awk (gen_comm_data): Add initializer for
profile to the dummy entry at the end of the list of architectures.
* config/arm/arm-cpu-cdata.h: Regenerated.

From-SVN: r249777

2017-06-29  Michael Collison  <michael.collison@arm.com>

Fix date on prevfious ChangeLog entry.

From-SVN: r249775

re PR target/70119 (AArch64 should take advantage of implicit truncation of variable shift amount without defining SHIFT_COUNT_TRUNCATED)

2017-06-29  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
    Michael Collison <michael.collison@arm.com>

PR target/70119
* config/aarch64/aarch64.md (*aarch64_<optab>_reg_<mode>3_mask1):
New pattern.
(*aarch64_reg_<mode>3_neg_mask2): New pattern.
(*aarch64_reg_<mode>3_minus_mask): New pattern.
(*aarch64_<optab>_reg_di3_mask2): New pattern.
* config/aarch64/aarch64.c (aarch64_rtx_costs): Account for cost
of shift when the shift amount is masked with constant equal to
the size of the mode.
* config/aarch64/predicates.md (subreg_lowpart_operator): New
predicate.

2017-06-29  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
    Michael Collison <michael.collison@arm.com>

PR target/70119
* gcc.target/aarch64/var_shift_mask_1.c: New test.

Co-Authored-By: Michael Collison <michael.collison@arm.com>
From-SVN: r249774

Fix -mbranch-cost range.

2017-06-29  Martin Liska  <mliska@suse.cz>

* config/i386/i386.opt: Change range from [1,5] to [0,5].

From-SVN: r249773

re PR ipa/80565 (ICE at -O2 and -O3 in 32-bit mode (not 64-bit) on x86_64-linux-gnu (in edge_badness, at ipa-inline.c:1028))

2017-06-29  Yury Gribov  <tetra2005@gmail.com>

PR bootstrap/80565
        * ipa-cp.c (allocate_and_init_ipcp_value): Add initialization
code.
        * ipa-inline.h
(edge_growth_cache_entry::edge_growth_cache_entry): New
function.
        (reset_edge_growth_cache): Update to use constructor.

From-SVN: r249771

Daily bump.

From-SVN: r249769

[AArch64] Do not increase data alignment at -Os and with -fconserve-stack.

We unnecessarily align data to 8 byte alignments even when -Os is
specified. This brings the logic in the AArch64 backend more in line
with the ARM backend and helps gain some image size in a few
places. Caught by an internal report on the size of rodata sections
being high with aarch64 gcc.

* config/aarch64/aarch64.h (AARCH64_EXPAND_ALIGNMENT): New.
  (DATA_ALIGNMENT): Update to use AARCH64_EXPAND_ALIGNMENT.
  (LOCAL_ALIGNMENT): Update to use AARCH64_EXPAND_ALIGNMENT.

Bootstrapped and regression tested on aarch64-none-linux-gnu with no
regressions.

From-SVN: r249764

PR c++/72764 - ICE with invalid template typename.

* decl.c (build_typename_type): No longer static.
* tree.c (strip_typedefs): Use it instead of make_typename_type.

From-SVN: r249760

* config/i386/avx512vlintrin.h (_mm256_permutexvar_epi64)
(_mm256_permutexvar_epi32, _mm256_permutex_epi64): New intrinsics.

testsuite/ChangeLog:

* gcc.target/i386/avx512vl-vpermd-1.c (_mm256_permutexvar_epi32):
Test new intrinsic.
* gcc.target/i386/avx512vl-vpermq-imm-1.c (_mm256_permutex_epi64):
Ditto.
* gcc.target/i386/avx512vl-vpermq-var-1.c (_mm256_permutexvar_epi64):
Ditto.
* gcc.target/i386/avx512f-vpermd-2.c: Do not check for AVX512F_LEN.
* gcc.target/i386/avx512f-vpermq-imm-2.c: Ditto.
* gcc.target/i386/avx512f-vpermq-var-2.c: Ditto.

From-SVN: r249759

PR c++/69300 - ICE with self-referential noexcept

* pt.c (maybe_instantiate_noexcept): Check for recursion.

From-SVN: r249757

PR c++/61022 - error with variadic template template parm

* pt.c (convert_template_argument): Keep the TYPE_PACK_EXPANSION.

From-SVN: r249756

PR c++/72801 - ICE with variadic partial specialization

* pt.c (unify_pack_expansion): Use PACK_EXPANSION_EXTRA_ARGS.

From-SVN: r249755

PR c++/55639 - partial specialization with ::template

* parser.c (cp_parser_class_head): Handle ::template.

From-SVN: r249754

PR c++/45976 - error with ::template in declarator.

* pt.c (resolve_typename_type): Fix TEMPLATE_ID_EXPR handling.

From-SVN: r249753

PR c++/54769 - wrong lookup of dependent template-name.

* parser.c (cp_parser_template_name): Handle dependent object type.
(cp_parser_nested_name_specifier_opt): Make template_keyword_p a
parameter.
(cp_parser_id_expression): Pass it.
(cp_parser_diagnose_invalid_type_name): Handle TEMPLATE_ID_EXPR.

From-SVN: r249752

Make the TEMPLATE_ID_EXPR location a range.

* parser.c (cp_parser_template_id): Use the range location on the
TEMPLATE_ID_EXPR.

From-SVN: r249751

PR c++/81204 - parse error with dependent template-name

* parser.c (cp_parser_lookup_name): Disqualify function templates
after lookup.

From-SVN: r249750

multiarch support for non-glibc linux systems

Current multiarch directory name is always *-linux-gnu* on linux,
this patch configures different names for uclibc and musl targets.

2017-06-28  Szabolcs Nagy  <szabolcs.nagy@arm.com>

* config.gcc (*-linux-musl*): Add t-musl tmake_file.
(*-linux-uclibc*): Add t-uclibc tmake_file.
* config/t-musl: New.
* config/t-uclibc: New.

From-SVN: r249745

[ARM] Consistently check for neon in vect effective targets

Conditions checked for ARM targets in vector-related effective targets
are inconsistent:

* sometimes arm*-*-* is checked
* sometimes Neon is checked
* sometimes arm_neon_ok and sometimes arm_neon is used for neon check
* sometimes check_effective_target_* is used, sometimes
* is-effective-target

This patch consolidate all of these check into using is-effective-target
arm_neon and when little endian was checked, the check is kept.

2017-06-28  Thomas Preud'homme  <thomas.preudhomme@arm.com>

    gcc/testsuite/
    * lib/target-supports.exp (check_effective_target_vect_int): Replace
    current ARM check by ARM NEON's availability check.
    (check_effective_target_vect_intfloat_cvt): Likewise.
    (check_effective_target_vect_uintfloat_cvt): Likewise.
    (check_effective_target_vect_floatint_cvt): Likewise.
    (check_effective_target_vect_floatuint_cvt): Likewise.
    (check_effective_target_vect_shift): Likewise.
    (check_effective_target_whole_vector_shift): Likewise.
    (check_effective_target_vect_bswap): Likewise.
    (check_effective_target_vect_shift_char): Likewise.
    (check_effective_target_vect_long): Likewise.
    (check_effective_target_vect_float): Likewise.
    (check_effective_target_vect_perm): Likewise.
    (check_effective_target_vect_perm_byte): Likewise.
    (check_effective_target_vect_perm_short): Likewise.
    (check_effective_target_vect_widen_sum_hi_to_si_pattern): Likewise.
    (check_effective_target_vect_widen_sum_qi_to_hi): Likewise.
    (check_effective_target_vect_widen_mult_qi_to_hi): Likewise.
    (check_effective_target_vect_widen_mult_hi_to_si): Likewise.
    (check_effective_target_vect_widen_mult_qi_to_hi_pattern): Likewise.
    (check_effective_target_vect_widen_mult_hi_to_si_pattern): Likewise.
    (check_effective_target_vect_widen_shift): Likewise.
    (check_effective_target_vect_extract_even_odd): Likewise.
    (check_effective_target_vect_interleave): Likewise.
    (check_effective_target_vect_multiple_sizes): Likewise.
    (check_effective_target_vect64): Likewise.
    (check_effective_target_vect_max_reduc): Likewise.

From-SVN: r249744

[arm] Fix incorrect __ARM_ARCH_PROFILE for -march=armv7

ACLE explicitly states that when targetting the common subset of
ARMv7-A, ARMv7-R and ARMv7-M, the __ARM_ARCH_PROFILE macro should not
be set.  We currently set it to 'M' which is clearly erroneous.

The logic for creating this is very convoluted and also somewhat
fragile, so I've taken the opportunity to use the new CPU and
architecture definition infrastructure to record the profile for each
architecture explicitly rather than try to reconstruct it from other
data.  I think this results in a much more robust solution.

2017-06-28  Richard Earnshaw  <rearnsha@arm.com>

* config/arm/parsecpu.awk (profile): Parse new keyword in an arch
context.
(gen_comm_data): Emit architectural setting of arch_prof.
* config/arm/arm-cpus.in (armv6-m, armv6s-m, armv7-a, armv7ve): Set the
profile.
(armv7-r, armv7-m, armv7e-m, armv8-a, armv8.1-a, armv8.2-a): Likewise.
(armv8-m.base, armv8-m.main): Likewise.
* arm-protos.h (arm_build_target): Add profile field.
(arch_option): Likewise.
* config/arm/arm.c (arm_configure_build_target): Copy the profile to
the active target.
* config/arm/arm.h (TARGET_ARM_ARCH_PROFILE): Use
arm_active_target.profile.

From-SVN: r249743

re PR tree-optimization/81227 (ICE in get_single_symbol, at tree-vrp.c:799)

2017-06-28  Richard Biener  <rguenther@suse.de>

PR middle-end/81227
* fold-const.c (negate_expr_p): Use TYPE_UNSIGNED, not
TYPE_OVERFLOW_WRAPS.
* match.pd (negate_expr_p): Likewise.
* tree-ssa-reassoc.c (optimize_range_tests_diff): Use
fold_build2, not fold_binary.

* gcc.dg/pr81227.c: New testcase.

From-SVN: r249742

This patch fixes a failure in gcc.target/aarch64/reload-valid-spoff.c triggered...

This patch fixes a failure in gcc.target/aarch64/reload-valid-spoff.c
triggered by https://gcc.gnu.org/ml/gcc-patches/2017-06/msg01367.html.

In ILP32 all memory accesses must have Pmode as the base address, but
aarch64_expand_mov_immediate wasn't emitting a conversion in one case.
Besides fixing this add an assert that flags any MEM operands that are
not Pmode.

    gcc/
* config/aarch64/aarch64 (aarch64_expand_mov_immediate):
Convert memory address to Pmode.
(aarch64_print_operand): Assert MEM operands are always Pmode.

From-SVN: r249741

Improve Cortex-A53 shift bypass

The aarch_forward_to_shift_is_not_shifted_reg bypass always returns true
on AArch64 shifted instructions.  This causes the bypass to activate in
too many cases, resulting in slower execution on Cortex-A53 like reported
in PR79665.

This patch uses the arm_no_early_alu_shift_dep condition instead which
improves the example in PR79665 by ~7%.  Given it is no longer used,
remove aarch_forward_to_shift_is_not_shifted_reg.  Also remove an
unnecessary REG_P check.

    gcc/
PR target/79665
* config/arm/aarch-common.c (arm_no_early_alu_shift_dep):
Remove redundant if.
(aarch_forward_to_shift_is_not_shifted_reg): Remove.
* config/arm/aarch-common-protos.h
(aarch_forward_to_shift_is_not_shifted_re): Remove.
* config/arm/cortex-a53.md: Use arm_no_early_alu_shift_dep in bypass.

From-SVN: r249740

re PR ipa/81238 (Target clone support does not make default clone static.)

[gcc]
2017-06-28  Michael Meissner  <meissner@linux.vnet.ibm.com>

PR ipa/81238
* multiple_target.c (create_dispatcher_calls): Set the default
clone to be static, not public.

[gcc/testsuite]
2017-06-28  Michael Meissner  <meissner@linux.vnet.ibm.com>

PR target/81193
* lib/target-supports.exp
(check_ppc_cpu_supports_hw_available): New test to make sure
__builtin_cpu_supports works on power7 and newer.

From-SVN: r249737