S/390: Fix vec_nabs and vec_vfi builtin flags

gcc/ChangeLog:

2017-10-09  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>

* config/s390/s390-builtins.def (vec_nabs, vec_vfi): Fix builtin
flags.

From-SVN: r253544

S/390: PR82463: Fix vec_madd header file definition

The builtin was not correctly defined in the vecintrin.h header file.

gcc/testsuite/ChangeLog:

2017-10-09  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>

PR target/82463
* gcc.target/s390/zvector/pr82463.c: New test.

gcc/ChangeLog:

2017-10-09  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>

PR target/82463
* config/s390/vecintrin.h (vec_madd, vec_msub): Fix macro
definitions.

From-SVN: r253543

S/390: PR82465: Fix vec_sqrt builtin flags

The vector double variant is available with z13.  A wrong flag in the
s390-builtins.def file triggered an error when compiling for z13.

gcc/testsuite/ChangeLog:

2017-10-09  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>

PR target/82465
* gcc.target/s390/zvector/pr82465.c: New test.

gcc/ChangeLog:

2017-10-09  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>

PR target/82465
* config/s390/s390-builtins.def (vec_sqrt): Fix builtin flags.

From-SVN: r253542

re PR target/82464 (s390x z14: vector float: invalid parameter combination for intrinsic '__builtin_s390_vec_xor')

PR target/82464
* config/s390/s390-builtins.def (s390_vec_xor_flt_a,
s390_vec_xor_flt_b, s390_vec_xor_flt_c): New.

From-SVN: r253541

Fix ssa-dse-26.c

Add missing -fno-short-enums to ensure this test passes on arm-none-eabi
which uses short enums.

    gcc/testsuite/
* gcc.dg/tree-ssa/ssa-dse-26.c (dg-options): Add -fno-short-enums.

From-SVN: r253540

Allow non-wi <op> wi

This patch uses global rather than member operators for wide-int.h,
so that the first operand can be a non-wide-int type.

The patch also removes the and_not and or_not member functions.
It was already inconsistent to have member functions for these
two operations (one of which was never used) and not other wi::
ones like udiv.  After the operator change, we'd have the additional
inconsistency that "non-wi & wi" would work but "non-wi.and_not (wi)"
wouldn't.

2017-10-09  Richard Sandiford  <richard.sandiford@linaro.org>

gcc/
* wide-int.h (WI_BINARY_OPERATOR_RESULT): New macro.
(WI_BINARY_PREDICATE_RESULT): Likewise.
(wi::binary_traits::operator_result): New type.
(wi::binary_traits::predicate_result): Likewise.
(generic_wide_int::operator~, unary generic_wide_int::operator-)
(generic_wide_int::operator==, generic_wide_int::operator!=)
(generic_wide_int::operator&, generic_wide_int::and_not)
(generic_wide_int::operator|, generic_wide_int::or_not)
(generic_wide_int::operator^, generic_wide_int::operator+
(binary generic_wide_int::operator-, generic_wide_int::operator*):
Delete.
(operator~, unary operator-, operator==, operator!=, operator&)
(operator|, operator^, operator+, binary operator-, operator*): New
functions.
* expr.c (get_inner_reference): Use wi::bit_and_not.
* fold-const.c (fold_binary_loc): Likewise.
* ipa-prop.c (ipa_compute_jump_functions_for_edge): Likewise.
* tree-ssa-ccp.c (get_value_from_alignment): Likewise.
(bit_value_binop): Likewise.
* tree-ssa-math-opts.c (find_bswap_or_nop_load): Likewise.
* tree-vrp.c (zero_nonzero_bits_from_vr): Likewise.
(extract_range_from_binary_expr_1): Likewise.
(masked_increment): Likewise.
(simplify_bit_ops_using_ranges): Likewise.

From-SVN: r253539

[PR 82416] Do not extend operands to at least 32 bits

2017-10-09  Martin Jambor  <mjambor@suse.cz>

PR hsa/82416
gcc/
* hsa-common.h (hsa_op_with_type): New method extend_int_to_32bit.
* hsa-gen.c (hsa_extend_inttype_to_32bit): New function.
(hsa_type_for_scalar_tree_type): Use it.  Always force min32int for
COMPLEX types.
(hsa_fixup_mov_insn_type): New function.
(hsa_op_with_type::get_in_type): Use it.
(hsa_build_append_simple_mov): Likewise.  Allow sub-32bit
immediates in an assert.
(hsa_op_with_type::extend_int_to_32bit): New method.
(gen_hsa_insns_for_bitfield): Fixup instruction and intermediary
types.  Convert to dest type if necessary.
(gen_hsa_insns_for_bitfield_load): Fixup load type if necessary.
(reg_for_gimple_ssa): Pass false as min32int to
hsa_type_for_scalar_tree_type.
(gen_hsa_addr): Fixup type when creating addresable temporary.
(gen_hsa_cmp_insn_from_gimple): Extend operands if necessary.
(gen_hsa_unary_operation): Extend operands and convert to dest type if
necessary.  Call hsa_fixup_mov_insn_type.
(gen_hsa_binary_operation): Changed operand types to hsa_op_with_type,
extend operands and convert to dest type if necessary.
(gen_hsa_insns_for_operation_assignment): Extend operands and convert
to dest type if necessary.
(set_output_in_type): Call hsa_fixup_mov_insn_type.  Just ude dest
if conversion nt necessary and size matches.
(gen_hsa_insns_for_load): Call hsa_fixup_mov_insn_type, convert
to dest type if necessary.
(gen_hsa_insns_for_store): Call hsa_fixup_mov_insn_type.
(gen_hsa_insns_for_switch_stmt): Likewise. Also extend operands if
necessary.
(gen_hsa_clrsb): Likewise.
(gen_hsa_ffs): Likewise.
(gen_hsa_divmod): Extend operands and convert to dest type if
necessary.
(gen_hsa_atomic_for_builtin): Change type of op to hsa_op_with_type.

libgomp/
* testsuite/libgomp.hsa.c/pr82416.c: New test.

From-SVN: r253538

Fix gcc.dg/cold-1.c

2017-10-09  Tom de Vries  <tom@codesourcery.com>

* gcc.dg/cold-1.c (foo1): Fix warning line number.  Make warning line
number relative.
(abort): Declare.

From-SVN: r253537

Daily bump.

From-SVN: r253536

rs6000: Cleanup of various jump patterns

This cleans up formatting of many jump patterns.  Mostly it is writing
output statements as brace blocks, and deleting default arguments.  It
also makes indentation a little bit more consistent, and uses b%T0
(instead of bctr/blr alternatives) in two more places.

* config/rs6000/rs6000.md (conditional branch): Clean up formatting.
Remove empty default arguments.  Use a brace block as output
statement.
(conditional return): Ditto.
(jump): Ditto.
(indirect_jump): Ditto.  Use b%T0 instead of bctr/blr.
(tablejump, tablejumpsi, tablejumpdi, *tablejump<mode>_internal1):
Ditto.
(group_ending_nop): Ditto.
(doloop_end): Ditto.
(ctr<mode>, ctr<mode>_internal1, ctr<mode>_internal2): Ditto.
(splitters for those): Ditto.

From-SVN: r253532

rs6000: Update conditional jump patterns, no more (pc) (label_ref)

Currently we have two patterns for every conditional jump: one jumping
if some condition is met, and the other jumping if it is not met.  We
don't need that second form because all our conditions have an opposite
already, and the generic code can deal with that just fine.

This patch deletes that second form.  We generated it directly in one
case; adjusted here.

* config/rs6000/rs6000-string.c (expand_strncmp_align_check): Invert
a conditional jump (and the compare for it) so that pc_rtx is the
last operand.
* config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Adjust
for the deleted and renamed ctr<mode>_internal[234] patterns.
* config/rs6000/rs6000.md: Delete second conditional branch pattern.
Delete second conditional return pattern.
(ctr<mode>_internal2): Delete this second bdnz pattern.
(ctr<mode>_internal3): Rename to ctr<mode>_internal2.
(ctr<mode>_internal4): Delete this second bdz pattern.

From-SVN: r253531

tree-outof-ssa.h (ssaexpand): Add partitions_for_undefined_values.

* tree-outof-ssa.h (ssaexpand): Add partitions_for_undefined_values.
(always_initialized_rtx_for_ssa_name_p): New predicate.
* tree-outof-ssa.c (remove_ssa_form): Initialize new field of SA.
(finish_out_of_ssa): Free new field of SA.
* tree-ssa-coalesce.h (get_undefined_value_partitions): Declare.
* tree-ssa-coalesce.c: Include tree-ssa.h.
(get_parm_default_def_partitions): Remove extern keyword.
(get_undefined_value_partitions): New function.
* expr.c (expand_expr_real_1) <expand_decl_rtl>: For a SSA_NAME, do
not set SUBREG_PROMOTED_VAR_P on the sub-register if it may contain
uninitialized bits.
* loop-iv.c (iv_get_reaching_def): Disqualify all subregs.

From-SVN: r253530

* builtins.def (BUILT_IN_SETJMP): Revert latest change.

From-SVN: r253529

check.c (gfc_check_x): Remove function.

2017-10-08  Steven G. Kargl  <kargl@gcc.gnu.org>

* check.c (gfc_check_x): Remove function.
* intrinsic.c (add_functions): Use gfc_check_fn_r.

From-SVN: r253528

stack-check-5.c: Skip with -fstack-protector.

* gcc.dg/stack-check-5.c: Skip with -fstack-protector.
* gcc.dg/stack-check-6.c: Likewise.
* gcc.dg/stack-check-6a.c: Likewise.

From-SVN: r253527

re PR fortran/82375 (PDT components in PDT declarations fail to compile)

2017-10-08  Paul Thomas  <pault@gcc.gnu.org>

PR fortran/82375
* module.c : Bump up MOD_VERSION to 15.
(mio_component): Edit comment about PDT specification list.
(mio_expr, mio_symbol): Include the expression and symbol PDT
specification lists in the same way as in mio_component.

From-SVN: r253526

dump_prase_tree (show_symbol): Output list of variables in NAMELIST.

2017-10-08  Thomas Koenig  <tkoenig@gcc.gnu.org>

* dump_prase_tree (show_symbol): Output list of variables in
NAMELIST.
(show_code_node): Add new line for ELSE and END DO for DO
CONCURRENT.
* invoke.texi: Document that the output of
-fdump-fortran-original, -fdump-fortran-optimized and
-fdump-parse-tree is unsable and may lead to ICEs.

From-SVN: r253525

i386.c (ix86_expand_set_or_movmem): Disable 512bit loops for targets that preffer 128bit.

* i386.c (ix86_expand_set_or_movmem): Disable 512bit loops for targets
that preffer 128bit.

From-SVN: r253524

* i386.c (has_dispatch): Disable for Ryzen.

From-SVN: r253523

arm.c (arm_set_return_address): Use MEM_VOLATILE_P on the target mem instead of RTX_FRAME_RELATED_P on the insn...

2017-10-08  Olivier Hainque  <hainque@adacore.com>

        * config/arm/arm.c (arm_set_return_address): Use MEM_VOLATILE_P
        on the target mem instead of RTX_FRAME_RELATED_P on the insn to
        prevent DSE.
        (thumb_set_return_address): Likewise.

From-SVN: r253522

arm-common.c (arm_except_unwind_info): Handle DWARF2_UNWIND_INFO.

2017-10-08  Olivier Hainque  <hainque@adacore.com>

        * common/config/arm/arm-common.c (arm_except_unwind_info):
        Handle DWARF2_UNWIND_INFO.

From-SVN: r253521

aarch64.md (*aarch64_reg_<optab>_minus<mode>3): New pattern.

2017-10-07  Michael Collison <michael.collison@arm.com>

* config/aarch64/aarch64.md (*aarch64_reg_<optab>_minus<mode>3):
New pattern.

2017-10-07  Michael Collison <michael.collison@arm.com>

* gcc.target/aarch64/var_shift_mask_2.c: New test.

From-SVN: r253520

Daily bump.

From-SVN: r253519

builtins.def (BUILT_IN_SETJMP): Declare as library builtin instead of GCC builtin if...

* builtins.def (BUILT_IN_SETJMP): Declare as library builtin instead
of GCC builtin if DONT_USE_BUILTIN_SETJMP is defined.
* except.c (sjlj_emit_function_enter): If DONT_USE_BUILTIN_SETJMP is
defined, force the creation of a new block for a dispatch label.

From-SVN: r253515

re PR fortran/82375 (PDT components in PDT declarations fail to compile)

2017-10-07  Paul Thomas  <pault@gcc.gnu.org>

PR fortran/82375
* class.c (gfc_find_derived_vtab): Return NULL for a passed
pdt template to prevent bad procedures from being written.
* decl.c (gfc_get_pdt_instance): Do not use the default
initializer for pointer and allocatable pdt type components. If
the component is allocatbale, set the 'alloc_comp' attribute of
'instance'.
* module.c : Add a prototype for 'mio_actual_arglist'. Add a
boolean argument 'pdt'.
(mio_component): Call it for the parameter list of pdt type
components with 'pdt' set to true.
(mio_actual_arg): Add the boolean 'pdt' and, if it is set, call
mio_integer for the 'spec_type'.
(mio_actual_arglist): Add the boolean 'pdt' and use it in the
call to mio_actual_arg.
(mio_expr, mio_omp_udr_expr): Call mio_actual_arglist with
'pdt' set false.
* resolve.c (get_pdt_spec_expr): Add the parameter name to the
KIND parameter error.
(get_pdt_constructor): Check that cons->expr is non-null.
* trans-array.c (structure_alloc_comps): For deallocation of
allocatable components, ensure that parameterized components
are deallocated first. Likewise, when parameterized components
are allocated, nullify allocatable components first. Do not
recurse into pointer or allocatable pdt components while
allocating or deallocating parameterized components. Test that
parameterized arrays or strings are allocated before freeing
them.
(gfc_trans_pointer_assignment): Call the new function. Tidy up
a minor whitespace issue.
trans-decl.c (gfc_trans_deferred_vars): Set 'tmp' to NULL_TREE
to prevent the expression from being used a second time.

2017-10-07  Paul Thomas  <pault@gcc.gnu.org>

PR fortran/82375
* gfortran.dg/pdt_13.f03 : New test.
* gfortran.dg/pdt_14.f03 : New test.
* gfortran.dg/pdt_15.f03 : New test.

From-SVN: r253514

invoke.texi (Wsuggest-attribute=cold): Document.

* invoke.texi (Wsuggest-attribute=cold): Document.
* common.opt (Wsuggest-attribute=cold): New
* ipa-pure-const.c (warn_function_cold): New function.
* predict.c (compute_function_frequency): Use it.
* predict.h (warn_function_cold): Declare.

* gcc.dg/cold-1.c: New testcase.

From-SVN: r253513

tree-switch-conversion.c (do_jump_if_equal, [...]): Update profile.

* tree-switch-conversion.c (do_jump_if_equal, emit_cmp_and_jump_insns):
Update profile.

From-SVN: r253512

re PR c++/80805 (ICE in sufficiently complex code with -g (dump_aggr_type))

2017-10-07  Paolo Carlini  <paolo.carlini@oracle.com>

PR c++/80805
* g++.dg/cpp0x/pr80805.C: New.

From-SVN: r253511

re PR c++/80805 (ICE in sufficiently complex code with -g (dump_aggr_type))

2017-10-07  Paolo Carlini  <paolo.carlini@oracle.com>

PR c++/80805
* g++.dg/cpp0x/pr80805.C: New.

From-SVN: r253510

re PR fortran/49232 (Pointer assignment of stride to CONTIGUOUS pointer not diagnosed as invalid)

2017-10-07  Thomas Koenig  <tkoenig@gcc.gnu.org>

PR fortran/49232
* expr.c (gfc_check_pointer_assign): Error
for non-contiguous rhs.

2017-10-07  Thomas Koenig  <tkoenig@gcc.gnu.org>

PR fortran/49232
* gfortran.dg/contiguous_4.f90: New test.

From-SVN: r253509

gfortran.h (async_io_dt): Add external reference.

2017-10-07  Thomas Koenig  <tkoenig@gcc.gnu.org>

* gfortran.h (async_io_dt): Add external reference.
* io.c (async_io_dt): Add variable.
(compare_to_allowed_values): Add prototyte. Add optional argument
num. If present, set it to the number of the entry that was
matched.
(check_io_constraints): If this is for an asynchronous I/O
statement, set async_io_dt and set the asynchronous flag for
a SIZE tag.
* resolve.c (resolve_transfer): If async_io_dt is set, set
the asynchronous flag on the variable.
(resolve_fl_namelist): If async_io_dt is set, set the asynchronous
flag on all elements of the namelist.

From-SVN: r253508

Fix libgomp.oacc-fortran/{firstprivate-1,parallel-reduction}.f90 for non-nvidia devices

2017-10-07  Tom de Vries  <tom@codesourcery.com>

* testsuite/libgomp.oacc-fortran/firstprivate-1.f90 (firstprivate):
Remove acc_device_nvidia references.
* testsuite/libgomp.oacc-fortran/parallel-reduction.f90 (reduction):
Same.

From-SVN: r253507

Daily bump.

From-SVN: r253506

re PR c++/66690 (error: use of ‘X’ before deduction of ‘auto’)

2017-10-06  Paolo Carlini  <paolo.carlini@oracle.com>

PR c++/66690
* g++.dg/cpp1y/pr66690.C: New.

From-SVN: r253502

2017-10-06  Paolo Carlini  <paolo.carlini@oracle.com>

* Fix last ChangeLog entry.

From-SVN: r253498

re PR c++/47791 (finish function is using literal value instead of a #defined one)

2017-10-06  Paolo Carlini  <paolo.carlini@oracle>

PR c++/47791
* decl.c (finish_function): Take a bool intead of an int; adjust.
* cp-tree.h (finish_function): Adjust declaration.
* decl2.c (generate_tls_wrapper, finish_objects,
finish_static_storage_duration_function): Adjust calls.
* lambda.c (maybe_add_lambda_conv_op, finish_lambda_function):
Likewise.
* method.c (synthesize_method): Likewise.
* optimize.c (maybe_thunk_body, maybe_clone_body): Likewise.
* pt.c (instantiate_decl): Likewise.
* parser.c (cp_parser_function_definition_after_declarator,
cp_parser_late_parsing_for_member, cp_parser_omp_declare_reduction):
Likewise.
(cp_parser_ctor_initializer_opt,
cp_parser_ctor_initializer_opt_and_function_body,
cp_parser_function_try_block,
cp_parser_function_definition_after_declarator,
cp_parser_function_transaction): Return void; adjust declarations.

From-SVN: r253497

[PR c++/82424] Dont convert dependent types

https://gcc.gnu.org/ml/gcc-patches/2017-10/msg00385.html
cp/
PR c++/82424
* name-lookup.c (check_local_shadow): Don't try and convert
dependent types.

testsuite/
PR c++/82424
* g++.dg/warn/pr82424.C: New.

From-SVN: r253496

re PR c++/82299 (-Wuseless-cast errors on typed enums used in member data initializers in c++1z)

PR c++/82299
* decl.c (reshape_init): Suppress warn_useless_cast for direct enum
init.
* typeck.c (convert_for_assignment): Likewise.

* g++.dg/cpp0x/pr82299.C: New test.

From-SVN: r253495

P0704R1 - fixing const-qualified pointers to members

P0704R1 - fixing const-qualified pointers to members
* typeck2.c (build_m_component_ref): For -std=c++2a allow
pointer to const & qualified method on rvalue.

* g++.dg/cpp2a/ptrmem1.C: New test.

From-SVN: r253494

[C++ PATCH] hash-table for extern-c fns.

https://gcc.gnu.org/ml/gcc-patches/2017-10/msg00376.html
Use hash_table for extern "C" names
* name-lookup.c (extern_c_fns): Use hash_table.
(check_extern_c_conflict): Adjust.
(c_linkage_bindings): Adjust.

From-SVN: r253493

Add sanopt support for UBSAN_PTR.

2017-10-06  Martin Liska  <mliska@suse.cz>

* sanopt.c (struct sanopt_tree_triplet_hash): Remove inline
keyword for member functions.
(struct sanopt_tree_couple): New struct.
(struct sanopt_tree_couple_hash): New function.
(struct sanopt_ctx): Add new hash_map.
(has_dominating_ubsan_ptr_check): New function.
(record_ubsan_ptr_check_stmt): Likewise.
(maybe_optimize_ubsan_ptr_ifn): Likewise.
(sanopt_optimize_walker): Handle IFN_UBSAN_PTR.
(pass_sanopt::execute): Handle also SANITIZE_POINTER_OVERFLOW.
2017-10-06  Martin Liska  <mliska@suse.cz>

* c-c++-common/ubsan/ptr-overflow-sanitization-1.c: New test.

From-SVN: r253492

ztest.c (test_large): Pass unsigned long *, not size_t *, to zlib uncompress function.

* ztest.c (test_large): Pass unsigned long *, not size_t *, to
zlib uncompress function.

From-SVN: r253491

Committed on behalf of Sudi Das

2017-10-06  Sudakshina Das  <sudi.das@arm.com>

PR target/82440
* config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Only call
aarch64_simd_valid_immediate on CONST_VECTORs.
(aarch64_reg_or_bic_imm): Likewise.

*** gcc/testsuite/ChangeLog ***

2017-10-06  Sudakshina Das  <sudi.das@arm.com>

* gcc.target/aarch64/bic_imm_1.c: New test.
* gcc.target/aarch64/orr_imm_1.c: Likewise.

From-SVN: r253490

[C++ PATCH] use hash-table for namespace contents

https://gcc.gnu.org/ml/gcc-patches/2017-10/msg00365.html
Use hash_table for namespace bindings
* cp-tree.h (struct named_decl_hash): New.
(lang_decl_ns): Change type of bindings field.
* lex.c (maybe_add_lang_decl_raw): Adjust.
* name-lookup.c (find_namespace_slot): Adjust.
(do_pushdecl): Push NULL-named namespace.
(do_push_nested_namespace): Adjust.
(push_namespace): Push anonymous namespace as NULL name.

From-SVN: r253489

PR82396 workaround

r253236 broke AArch64 bootstrap.  This is a temporary workaround that
disables qsort checking in the scheduler to enable continued development
and testing on AArch64.  This will be removed once the autopref scheduling
code has been fixed.

    gcc/
PR rtl-optimization/82396
* haifa-sched.c (ready_sort_real): Disable qsort checking.

From-SVN: r253487

graphite-dependences.c (scop_get_reads): Move code to...

2017-10-06  Sebastian Pop  <sebpop@gmail.com>

* graphite-dependences.c (scop_get_reads): Move code to...
        (scop_get_must_writes): Move code to...
        (scop_get_may_writes): Move code to...
        (scop_get_reads_and_writes): ... here.
        (scop_get_dependences): Call scop_get_reads_and_writes.

From-SVN: r253486

re PR c++/60153 (internal compiler error: in dependent_type_p, at cp/pt.c:21951)

2017-10-06  Paolo Carlini  <paolo.carlini@oracle.com>

PR c++/60153
* g++.dg/cpp0x/variadic-crash3.C: New.

From-SVN: r253484

re PR tree-optimization/82434 (-fstore-merging does not work reliably.)

PR tree-optimization/82434
* fold-const.h (can_native_encode_type_p,
can_native_encode_string_p): Remove.
* fold-const.c (native_encode_int): Formatting fixes.  If ptr is NULL,
don't encode anything, just return what would be otherwise returned.
(native_encode_fixed, native_encode_complex, native_encode_vector):
Likewise.
(native_encode_string): Likewise.  Inline by hand
can_native_encode_string_p.
(can_native_encode_type_p): Remove.
(can_native_encode_string_p): Remove.
* tree-vect-stmts.c (vectorizable_store): Instead of testing just
STRING_CSTs using can_native_encode_string_p, test all
CONSTANT_CLASS_P values using native_encode_expr with NULL ptr.
* gimple-ssa-store-merging.c (encode_tree_to_bitpos): Remove last
argument from native_encode_expr.
(rhs_valid_for_store_merging_p): Use native_encode_expr with NULL ptr.
(pass_store_merging::execute): Don't unnecessarily look for 3 stmts,
but just 2.

* gcc.dg/store_merging_9.c: New test.

From-SVN: r253483

re PR tree-optimization/82397 (qsort comparator non-negative on sorted output: 1 in vect_analyze_data_ref_accesses)

2017-10-06  Richard Biener  <rguenther@suse.de>

PR tree-optimization/82397
* tree-vect-data-refs.c (dr_group_sort_cmp): Do not use
operand_equal_p but rely on data_ref_compare_tree for detecting
equalities.
(vect_analyze_data_ref_accesses): Use data_ref_compare_tree
to match up with dr_group_sort_cmp.

* gfortran.dg/pr82397.f: New testcase.

From-SVN: r253482

PR82322: S/390: Fix vec_ceil and friends

vec_ceil and friends are expanded by vecintrin.h to __builtin_s390_vfi
which is an overloaded builtin being replaced by either
__builtin_s390_vfisb or __builtin_s390_vfidb depending on its argument
types.

The problem in this PR was that the overloaded builtin definition of
__builtin_s390_vfi was missing in s390-builtins.def.

gcc/ChangeLog:

2017-10-06  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>

PR target/82322
* config/s390/s390-builtins.def (s390_vfi): Define new overloaded
builtin.
* config/s390/s390-builtin-types.def: Regenerate.

gcc/testsuite/ChangeLog:

2017-10-06  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>

PR target/82322
* gcc.target/s390/zvector/pr82322.c: New test.

From-SVN: r253481

PR82317: S/390: Fix vec_min/vec_max builtins for IBM z13.

With IBM z14 a hardware instruction for floating point min and max has
been added while for IBM z13 we emulated min/max for vector double with
compare and select.  This testcase makes sure that we fall back to the
emulated variant when compiling for z13.

gcc/ChangeLog:

2017-10-06  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>

PR target/82317
* config/s390/s390-builtin-types.def: Regenerate.
* config/s390/s390-builtins.def (s390_vfmaxdb_4, s390_vfmindb_4):
Change flag from B_VXE to B_VX.
(s390_vec_min_dbl): Remove B_VXE flag.

gcc/testsuite/ChangeLog:

2017-10-06  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>

PR target/82317
* gcc.target/s390/zvector/pr82317.c: New test.

From-SVN: r253480

re PR tree-optimization/82422 ([graphite] ICE in set_codegen_error, at graphite-isl-ast-to-gimple.c:248)

2017-10-06  Richard Biener  <rguenther@suse.de>

PR tree-optimization/82422
* gcc.dg/graphite/pr82422.c: New testcase.

From-SVN: r253479

re PR tree-optimization/82421 ([graphite] ICE: tree check: expected ssa_name, have integer_cst in get_rename, at graphite-isl-ast-to-gimple.c:1127)

2017-10-06  Richard Biener  <rguenther@suse.de>

PR tree-optimization/82421
* gcc.dg/graphite/pr82421.c: New testcase.

From-SVN: r253478

re PR c/82437 (false-positive -Wtautological-compare warning with -std=gnu89)

PR c/82437
* c-warn.c (warn_tautological_bitwise_comparison): Instead of
using to_widest use wide_int with the larger of the two precisions.

* c-c++-common/Wtautological-compare-6.c: New test.

From-SVN: r253476

graphite-isl-ast-to-gimple.c: Include ssa.h and tree-ssa.h.

2017-10-06  Richard Biener  <rguenther@suse.de>

* graphite-isl-ast-to-gimple.c: Include ssa.h and tree-ssa.h.
(translate_isl_ast_to_gimple::translate_pending_phi_nodes,
translate_isl_ast_to_gimple::is_valid_rename,
translate_isl_ast_to_gimple::get_rename,
translate_isl_ast_to_gimple::get_def_bb_for_const,
translate_isl_ast_to_gimple::get_new_name,
translate_isl_ast_to_gimple::collect_all_ssa_names,
translate_isl_ast_to_gimple::copy_loop_phi_args,
translate_isl_ast_to_gimple::collect_all_ssa_names,
translate_isl_ast_to_gimple::copy_loop_phi_args,
translate_isl_ast_to_gimple::copy_loop_phi_nodes,
translate_isl_ast_to_gimple::add_close_phis_to_merge_points,
translate_isl_ast_to_gimple::add_close_phis_to_outer_loops,
translate_isl_ast_to_gimple::copy_loop_close_phi_args,
translate_isl_ast_to_gimple::copy_loop_close_phi_nodes,
translate_isl_ast_to_gimple::copy_cond_phi_args,
translate_isl_ast_to_gimple::copy_cond_phi_nodes,
translate_isl_ast_to_gimple::edge_for_new_close_phis,
translate_isl_ast_to_gimple::add_phi_arg_for_new_expr,
translate_isl_ast_to_gimple::rename_uses,
translate_isl_ast_to_gimple::rename_all_uses): Remove.
(translate_isl_ast_to_gimple::get_rename_from_scev): Simplify.
(set_rename_for_each_def): Likewise.
(graphite_copy_stmts_from_block): Handle debug stmt resetting
here.  Handle rewriting SCEV analyzable uses here.
(copy_bb_and_scalar_dependences): Generate code for PHI
copy-in/outs.
(graphite_regenerate_ast_isl): Adjust.
* graphite-scop-detection.c (trivially_empty_bb_p): Move to sese.[ch].
(add_write, add_read): New functions.
(build_cross_bb_scalars_def): Use it and simplify.
(build_cross_bb_scalars_use): Likewise.
(graphite_find_cross_bb_scalar_vars): Inline into...
(try_generate_gimple_bb): ...here.  Add dependences for PHIs,
simulating out-of-SSA.  Compute liveout and add dependencies.
(build_scops): Force an empty entry block.
* sese.h (sese_info_t::liveout, sese_info_t::debug_liveout): New
members.
(sese_build_liveouts): Declare.
(sese_trivially_empty_bb_p): Likewise.
* sese.c (sese_build_liveouts_bb): Properly handle PHIs,
compute liveout and debug_liveout.
(sese_bad_liveouts_use): Remove.
(sese_reset_debug_liveouts_bb): Likewise.
(sese_reset_debug_liveouts): Rewrite in terms of debug_liveout.
(sese_build_liveouts): Build liveout and debug_liveout and store
it in region.
(new_sese_info): Adjust.
(free_sese_info): Likewise.
(sese_insert_phis_for_liveouts): Reset debug stmts from here,
do not build liveout here.
(move_sese_in_condition): Adjust region entry.
(scev_analyzable_p): Match up with chrec_apply requirements.
(sese_trivially_empty_bb_p): New.
* tree-into-ssa.c (get_reaching_def): Properly support generating
default-defs for incremental rewrite of anonymous names.

* gcc.dg/graphite/id-15.c: No longer expect a code generation error.
* gcc.dg/graphite/id-16.c: Likewise.
* gcc.dg/graphite/pr46168.c: Likewise.
* gcc.dg/graphite/pr68756.c: Likewise.
* gcc.dg/graphite/pr69728.c: Likewise.
* gcc.dg/graphite/pr71575-2.c: Likewise.
* gcc.dg/graphite/pr77362.c: Likewise.
* gcc.dg/graphite/pr81373.c: Likewise.
* gcc.dg/graphite/run-id-pr67700-1.c: Likewise.
* gfortran.dg/graphite/interchange-1.f: Likewise.
* gfortran.dg/graphite/pr42334-1.f: Likewise.
* gfortran.dg/graphite/pr42393-1.f90: Likewise.
* gfortran.dg/graphite/pr42393.f90: Likewise.
* gfortran.dg/graphite/pr47019.f: Likewise.
* gfortran.dg/graphite/id-17.f: Likewise.
* gfortran.dg/graphite/id-19.f: Likewise.
* gfortran.dg/graphite/run-id-2.f90: Likewise.
* gfortran.dg/graphite/pr42326-1.f90: Likewise.
* gfortran.dg/graphite/pr42326.f90: Likewise.
* gfortran.dg/graphite/pr68550-2.f90: Likewise.
* gfortran.dg/graphite/pr29581.f90: Likewise.  No longer expect
a code generation error.
* gfortran.dg/graphite/run-id-3.f90: Likewise.
* gfortran.dg/graphite/pr29832.f90: Likewise.

From-SVN: r253475

graphite-sese-to-poly.c (extract_affine): For casts increasing precision do not perform modulo reduction.

2017-10-06  Richard Biener  <rguenther@suse.de>

* graphite-sese-to-poly.c (extract_affine): For casts increasing
precision do not perform modulo reduction.

From-SVN: r253474

re PR tree-optimization/82436 (465.tonto ICE in vect_get_slp_vect_defs, at tree-vect-slp.c:3410)

2017-10-06  Richard Biener  <rguenther@suse.de>

PR tree-optimization/82436
* tree-vect-slp.c (vect_supported_load_permutation_p): More
conservatively choose the vectorization factor when checking
whether we can perform the required load permutation.
(vect_transform_slp_perm_load): Assert when we may not fail.

* gcc.dg/vect/pr82436.c: New testcase.

From-SVN: r253473

Daily bump.

From-SVN: r253472

compiler: drop special handling of unexported func/var assembler names

    For example, for the package math/big, we used to generate unexported
    names as `big.trim` and exported names as `math_big.NewInt`.  After
    this change we will use `math_big` consistently.

    Reviewed-on: https://go-review.googlesource.com/68651

From-SVN: r253468

rs6000: Fix an error message in sysv4.h

While looking at PR82411, I got the error message

error: '-msdata=-mcall' and 'eabi-linux' are incompatible

which caused some head-scratching.

* config/rs6000/sysv4.h (SUBTARGET_OVERRIDE_OPTIONS): Correct error
message for incompatible -msdata=* and -mcall-* options.

From-SVN: r253467

* call.c (convert_arg_to_ellipsis): Use the result of force_rvalue.

From-SVN: r253465

Fix new testsuite failure on newlib targets

* testsuite/20_util/to_chars/1.cc: Add dg-require-string-conversions.

From-SVN: r253463

c-pretty-print.c (pp_c_parameter_type_list): Print ...

2017-10-05  Bernd Edlinger  <bernd.edlinger@hotmail.de>

* c-pretty-print.c (pp_c_parameter_type_list): Print ... for variadic
functions.

testsuite:
2017-10-05  Bernd Edlinger  <bernd.edlinger@hotmail.de>

* gcc.dg/Wincompatible-pointer-types-1.c: New test.

From-SVN: r253460

compiler: centralize all symbol name handling

    Consolidate all symbol name handling into the new file names.cc.  This
    is intended to define all the names that will appear in the generated
    assembly code.  Names that will not appear in the assembly code, such
    as local variable names or label names, remain where they are.

    This consolidation is not intended to change any of the existing
    symbol names.  Tested by building without and without this patch and
    comparing the libgo symbol table.

    Reviewed-on: https://go-review.googlesource.com/68310

* Make-lang.in (GO_OBJS): Add go/names.o.

From-SVN: r253458

elf.c (elf_zlib_fetch): Change pval argument to uint64_t *.

* elf.c (elf_zlib_fetch): Change pval argument to uint64_t *.
Read a four byte integer.
(elf_zlib_inflate): Change val to uint64_t.  Align pin to a 32-bit
boundary before ever calling elf_zlib_fetch.
* ztest.c (test_large): Simplify print statements a bit.

From-SVN: r253456

re PR ada/82393 (Compilation error on cygwin64)

PR ada/82393
* mingw32.h (_O_U8TEXT, _O_U16TEXT, _O_WTEXT): Delete.
* sysdep.c (__gnat_set_mode ): Use DJGPP version for Cygwin.

From-SVN: r253454

i386.c (ia32_multipass_dfa_lookahead): Default to issue rate for post-reload scheduling.

* i386.c (ia32_multipass_dfa_lookahead): Default to issue rate
for post-reload scheduling.

From-SVN: r253453

slp-perm-9.c: Use vect_sizes_16B_8B.

gcc/testsuite/
2017-10-05  Tamar Christina  <tamar.christina@arm.com>

* gcc.dg/vect/slp-perm-9.c: Use vect_sizes_16B_8B.
* lib/target-supports.exp (vect_sizes_16B_8B): New.

gcc/
2017-10-05  Tamar Christina  <tamar.christina@arm.com>

* doc/sourcebuild.texi (vect_sizes_16B_8B, vect_sizes_32B_16B): New.

From-SVN: r253452

vect-align-1.c: Fix vect_hw_misalign condition.

2017-09-28  Tamar Christina  <tamar.christina@arm.com>

* gcc.dg/vect/vect-align-1.c: Fix vect_hw_misalign condition.
* gcc.dg/vect/vect-align-2.c: Likewise.
* gcc.dg/vect/vect-multitypes-1.c: Likewise

From-SVN: r253451

i386.c (znver1_cost): Set branch_cost to 3 (instead of 2) to improve monte carlo in scimark.

* i386.c (znver1_cost): Set branch_cost to 3 (instead of 2)
to improve monte carlo in scimark.

From-SVN: r253449

i386.c (ix86_size_cost, [...]): Set reassociation width to 1.

* i386.c (ix86_size_cost, i386_cost, i486_cost, pentium_cost,
lakemont_cost, pentiumpro_cost, geode_cost, k6_cost,
athlon_cost, k8_cost, amdfam10_cost, btver1_cost, btver2_cost,
pentium4_cost, nocona_cost): Set reassociation width to 1.
(bdver1_cost, bdver2_cost, bdver3_cost, bdver4_cost): Set reassociation
width to 2 for fp operations and 1 otherwise.
(znver1_cost): Set scalar reassoc width to 4 and vector to 3 and 6
for int and fp.
(atom_cost): Set reassociation width to 2.
(slm_cost, generic_cost): Set fp reassociation width to 2 and 1 otherwise.
(intel_cost): Set fp reassociation width to 4 and 1 otherwise.
(core_cost): Set fp reassociation width to 4 and vector to 2.
(ix86_reassociation_width): Rewrite using cost table; special case
plus/minus on Zen; honor X86_TUNE_SSE_SPLIT_REGS
and TARGET_AVX128_OPTIMAL.
* i386.h (processor_costs): Add
reassoc_int, reassoc_fp, reassoc_vec_int, reassoc_vec_fp.
(TARGET_VECTOR_PARALLEL_EXECUTION, TARGET_REASSOC_INT_TO_PARALLEL,
TARGET_REASSOC_FP_TO_PARALLEL): Remove.
* x86-tune.def (X86_TUNE_REASSOC_INT_TO_PARALLEL): Remove.
(X86_TUNE_REASSOC_FP_TO_PARALLEL): Remove.
(X86_TUNE_VECTOR_PARALLEL_EXECUTION):  Remove.

From-SVN: r253448

Re: [PATCH] C++ warning on vexing parse

https://gcc.gnu.org/ml/gcc-patches/2017-10/msg00249.html
gcc/cp/
Warn on MVP declarations
* cp-tree.h (struct cp_declarator): Add parenthesized field.
* decl.c (grokdeclarator): Warn about unnecessary parens.
* parser.c (make_declarator): Init parenthesized field.
(cp_parser_direct_declarator): Set parenthesized field.

gcc/
* doc/invoke.texi (Wparentheses): Document C++ MVP behaviour.

gcc/testsuite/
* g++.dg/warn/mvp.C: New.

From-SVN: r253446

[C++ PATCH] Kill IDENTIFIER_GLOBAL_VALUE

https://gcc.gnu.org/ml/gcc-patches/2017-10/msg00239.html
Kill IDENTIFIER_GLOBAL_VALUE, SET_IDENTIFIER_GLOBAL_VALUE
* cp-tree.h (IDENTIFIER_GLOBAL_VALUE,
SET_IDENTIFIER_GLOBAL_VALUE): Delete.
* name-lookup.h (set_global_binding): Remove NAME parm.
(get_global_binding): New inline fn.
* name-lookup.c (set_global_binding): Remove NAME parm. Adjust.
(identifier_global_value): Move to ...
* cp-objcp-common.c (identifier_global_value): ... here.
* class.c (build_ctor_vtbl_group, build_vtbl_initializer): Adjust.
* decl.c (record_builtin_type, expand_static_init,
grokdeclarator): Adjust.
* decl2.c (get_guard, get_local_tls_init_fn, get_tls_init_fn,
get_tls_wrapper_fn, maybe_warn_sized_delete): Adjust.
* except.c (declare_library_fn, build_throw): Adjust.
* init.c (throw_bad_array_length): Adjust.
* rtti.c (throw_bad_cast, throw_bad_typeid, get_tinfo_decl): Adjust.

From-SVN: r253444

arm.c (arm_test_fpu_data): New.

2017-10-05  Tamar Christina  <tamar.christina@arm.com>

* config/arm/arm.c (arm_test_fpu_data): New.
(arm_run_selftests): Call arm_test_fpu_data.

From-SVN: r253443

Add sanitizer_linux_x86_64.lo if __x86_64__ is defined by $CC

Since size of "void *" is 4 bytes for x32, check if __x86_64__ is defined
by $CC, instead of

if test x$ac_cv_sizeof_void_p = x8; then

to decide wether sanitizer_linux_x86_64.lo should be used.

PR sanitizer/82379
* configure.tgt (SANITIZER_COMMON_TARGET_DEPENDENT_OBJECTS): Set
to sanitizer_linux_x86_64.lo if __x86_64__ is defined by $CC.

From-SVN: r253441

Re: [C++ PATCH] Move mangling alias out of ::

https://gcc.gnu.org/ml/gcc-patches/2017-10/msg00230.html
* decl2.c (record_mangling): Fix spello and formatting from
previous patch.

From-SVN: r253440

Fix libgomp.oacc-c-c++-common/{loop-red-g-1,routine-g-1}.c for non-nvidia devices

2017-10-05  Tom de Vries  <tom@codesourcery.com>

* testsuite/libgomp.oacc-c-c++-common/loop-red-g-1.c (main): Remove
vector_length(32) clause from acc parallel directive.
* testsuite/libgomp.oacc-c-c++-common/routine-g-1.c (main): Same.

From-SVN: r253439

pr82374.c (SIZE): Change from 1G to 1M to make it ilp32 friendly.

* gcc.dg/gomp/pr82374.c (SIZE): Change from 1G to 1M to make it ilp32
friendly.

From-SVN: r253438

id-17.f: Fix typo.

2017-10-05  Richard Biener  <rguenther@suse.de>

* gfortran.dg/graphite/id-17.f: Fix typo.
* gfortran.dg/graphite/id-19.f: Likewise.
* gfortran.dg/graphite/pr29832.f90: Likewise.
* gfortran.dg/graphite/pr42326-1.f90: Likewise.
* gfortran.dg/graphite/pr42326.f90: Likewise.
* gfortran.dg/graphite/pr68550-2.f90: Likewise.
* gfortran.dg/graphite/run-id-2.f90: Likewise.
* gfortran.dg/graphite/run-id-3.f90: Likewise.

From-SVN: r253437

Daily bump.

From-SVN: r253436

re PR c++/80471 ((gcc extension) Forwarding-reference `auto` function parameters do not follow the same rules as template functions or generic lambdas)

2017-10-04  Paolo Carlini  <paolo.carlini@oracle.com>

PR c++/80471
* g++.dg/cpp1y/pr80471.C: New.

From-SVN: r253432

re PR c++/78131 (Inconsistent evaluation for `constexpr` lambdas in templates between `static_assert`, `if constexpr(…)` and `constexpr` variables)

2017-10-04  Paolo Carlini  <paolo.carlini@oracle.com>

PR c++/78131
* g++.dg/cpp1z/constexpr-lambda17.C: New.

From-SVN: r253431

re PR c++/78018 ([C++14] "internal compiler error: Segmentation fault" with templates and lambdas)

2017-10-04  Paolo Carlini  <paolo.carlini@oracle.com>

PR c++/78018
* g++.dg/cpp1y/lambda-generic-78018.C: New.

From-SVN: r253430

[PATCH] remove some unneeded parens

https://gcc.gnu.org/ml/gcc-patches/2017-10/msg00200.html
* toplev.c (toplev::main): Remove excess parens on pretty_printer
decl.
* caller-save.c (insert_save): Remove excess parens on TO_SAVE parm.

From-SVN: r253427

[C++ PATCH] give builtin types consistent name

https://gcc.gnu.org/ml/gcc-patches/2017-10/msg00201.html
Give builtin types the correct name.
* name-lookup.c (set_global_binding): Assert name is DECL_NAME.
* decl.c (record_builtin_type): Reimplement, use new TYPE_DECL for
rname.

From-SVN: r253426

PR c++/82406 - C++17 error with noexcept function type

* g++.dg/ext/attrib54.C: New.

From-SVN: r253425

re PR c++/71946 (asm in toplevel lambda function rejected)

/cp
2017-10-04  Paolo Carlini  <paolo.carlini@oracle.com>
    Andrew Pinski  <apinski@cavium.com>

PR c++/71946
* parser.c (cp_parser_lambda_body): Set parser->in_function_body.

/testsuite
2017-10-04  Paolo Carlini  <paolo.carlini@oracle.com>
    Andrew Pinski  <apinski@cavium.com>

PR c++/71946
* g++.dg/cpp0x/lambda/lambda-asm1.C: New.
* g++.dg/cpp0x/lambda/lambda-stmtexpr1.C: Likewise.

Co-Authored-By: Andrew Pinski <apinski@cavium.com>
From-SVN: r253423

[PATCH][AArch64] Add BIC-imm and ORR-imm SIMD pattern

This patch adds the support for BIC (vector, immediate) and
ORR (vector, immediate) SIMD patterns to the AArch64 backend.

Committed on behalf of Sudi Das.

Reviewed-by: Richard Earnshaw <Richard.Earnshaw@arm.com>
Reviewed-by: James Greenhalgh <james.greenhalgh@arm.com>
gcc/

* config/aarch64/aarch64-protos.h (enum simd_immediate_check): New
check type for aarch64_simd_valid_immediate.
(aarch64_output_simd_mov_immediate): Update prototype.
(aarch64_simd_valid_immediate): Update prototype.
* config/aarch64/aarch64-simd.md (orr<mode>3): modified pattern to add
support for ORR-immediate.
(and<mode>3): modified pattern to add support for BIC-immediate.
* config/aarch64/aarch64.c (aarch64_simd_valid_immediate): Function
now checks for valid immediate for BIC and ORR based on new enum
argument.
(aarch64_output_simd_mov_immediate): Function now used to output
BIC/ORR imm as well based on new enum argument.
* config/aarch64/constraints.md (Do): New vector immediate constraint.
(Db) : Likewise.
* config/aarch64/predicates.md (aarch64_reg_or_orr_imm): New predicate.
(aarch64_reg_or_bic_imm): Likewise.

gcc/testsuite/

* gcc.target/aarch64/bic_imm_1.c: New test.
* gcc.target/aarch64/orr_imm_1.c: Likewise.

From-SVN: r253422

[C++ PATCH] Move mangling alias out of ::

https://gcc.gnu.org/ml/gcc-patches/2017-10/msg00199.html
gcc/cp/
Move mangling aliases out of global namespace.
* cp-tree.h (record_mangling): New.
(maybe_remove_implicit_alias): Delete.
* decl2.c (mangled_decls): New hash map.
(generate_mangling_alias): Reimplement using mangled_decls.
(record_mangling): New.
* mangle.c (decl_implicit_alias_p,
maybe_remove_implicit_alias): Delete.
(mangle_decl): Use record_mangling.
* name-lookup.c (supplement_binding_1): Remove
maybe_remove_implicit_alias check.

* call.c (convert_arg_to_ellipsis): Correct comment about passing
by reference.

gcc/testsuite/
* g++.dg/abi/mangle41.C: Adjust diagnostics.

libcc1/
* libcp1plugin.cc (supplement_binding): Don't use
maybe_remove_implicit_alias.

From-SVN: r253421

S/390: Fix mode in vector merge pattern.

vec_unpacks_hi_v4sf/vec_unpacks_lo_v4sf expand vec_mergeh and vec_mergel
patterns also for z13 with V4SF modes so the patterns should better
accept this.  Fixed by changing the mode iterator to V_128_NOSINGLE
which accepts V4SF unconditionally.

gcc/ChangeLog:

2017-10-04  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>

* config/s390/vx-builtins.md ("vec_mergeh<mode>")
("vec_mergel<mode>"): Change mode iterator to V_128_NOSINGLE.

From-SVN: r253420

Revert r253399:

PR rtl-optimization/82396
* haifa-sched.c (autopref_multipass_init): Simplify
initialization.
(autopref_rank_data): Simplify sort order.
* sched-int.h (autopref_multipass_data_): Remove
multi_mem_insn_p, min_offset and max_offset.

From-SVN: r253419

re PR c++/82373 (syntax error in error message)

PR c++/82373
* error.c (dump_function_decl): If show_return, call dump_type_suffix
on the same return type dump_type_prefix has been called on.

* g++.dg/cpp1y/pr82373.C: New test.

From-SVN: r253418

2017-10-04  Petr Ovtchenkov  <ptr@void-ptr.info>
    François Dumont  <fdumont@gcc.gnu.org>

* include/bits/streambuf_iterator.h
(istreambuf_iterator<>::operator*()): Do not capture iterator state
in Debug assertion.
(istreambuf_iterator<>::operator++()): Likewise and remove _M_sbuf check.
(istreambuf_iterator<>::operator++(int)): Likewise.
(istreambuf_iterator<>::_M_get()): Remove _M_c assignment.
(istreambuf_iterator<>::_S_is_eof()): New.
(istreambuf_iterator<>::_M_at_eof()): Adapt, use latter.
(find(istreambuf_iterator<>, istreambuf_iterator<>, _CharT)):
Return an iterator with _M_c set to eof to capture streambuf state
on evaluation.
(testsuite/24_iterators/istreambuf_iterator/2.cc): Add checks.

Co-Authored-By: François Dumont <fdumont@gcc.gnu.org>
From-SVN: r253417

PR c++/81525 - broken handling of auto in generic lambda.

* pt.c (tsubst_decl) [VAR_DECL]: Use strip_innermost_template_args.

From-SVN: r253414

C: underline parameters in mismatching function calls

In r253096
  ("C++: underline parameters in mismatching function calls"
  aka 5d78d423a5f7a1d135c7bb678e82007678d1313c
    https://gcc.gnu.org/ml/gcc-patches/2017-09/msg01546.html )
I updated the C++ FE's handling of mismatched types in function calls
so that it underlines the pertinent param of the callee, rather than
just the function name.

The following patch does the same for the C frontend.

Given e.g. this type mismatch:

  extern int callee (int one, const char *two, float three);

  int caller (int first, int second, float third)
  {
    return callee (first, second, third);
  }

the C FE currently emits (trunk):

  test.c: In function 'caller':
  test.c:5:25: warning: passing argument 2 of 'callee' makes pointer from
  integer without a cast [-Wint-conversion]
     return callee (first, second, third);
                           ^~~~~~
  test.c:1:12: note: expected 'const char *' but argument is of type 'int'
   extern int callee (int one, const char *two, float three);
              ^~~~~~

whereas with this patch the note underlines the pertinent param of
the callee:

  test.c: In function 'caller':
  test.c:5:25: warning: passing argument 2 of 'callee' makes pointer from
  integer without a cast [-Wint-conversion]
     return callee (first, second, third);
                           ^~~~~~
  test.c:1:41: note: expected 'const char *' but argument is of type 'int'
   extern int callee (int one, const char *two, float three);
                               ~~~~~~~~~~~~^~~

making the problem more obvious to the user.

As with the C++ patch, the patch:

(a) updates the locations of the params to cover the range of all
of their tokens, putting the caret on the first character of the
param name (if present), otherwise at the start of the first token
(doing so requires adding a last_token_location to the c_parser, so
we can determine the location of the last consumed token).

(b) updates the "note" to use the param location, rather than the
fndecl location

gcc/c/ChangeLog:
* c-decl.c (push_parm_decl): Store c_parm's location into the
PARAM_DECL.
(build_c_parm): Add "loc" param and store it within the c_parm.
* c-parser.c (struct c_parser): Add "last_token_location" field.
(c_parser_consume_token): Store location of the token into the
new field.
(c_parser_declaration_or_fndef): Store params into DECL_ARGUMENTS
when handling a FUNCTION_DECL, if it doesn't already have them.
(c_parser_parameter_declaration): Generate a location for the
parameter, and pass it to the call to build_c_parm.
* c-tree.h (struct c_parm): Add field "loc".
(build_c_parm): Add location_t param.
* c-typeck.c (get_fndecl_argument_location): New function.
(inform_for_arg): New function.
(convert_for_assignment): Use inform_for_arg when dealing with
ic_argpass.

gcc/testsuite/ChangeLog:
* gcc.dg/diagnostic-range-bad-called-object.c: Update expected
underlining for param.
* gcc.dg/param-type-mismatch.c: Update expected results to reflect
highlighting of parameters; add test coverage for trivial
parameter decls, and for callback parameters.
* gcc.dg/pr68533.c: Update location of two errors to reflect
location of params.

From-SVN: r253411

call.c (convert_arg_to_ellipsis): Correct comment about passing by reference.

* call.c (convert_arg_to_ellipsis): Correct comment about passing
by reference.

From-SVN: r253410

jit: implement gcc_jit_context_new_rvalue_from_vector

This patch implements a new API entrypoint:

/* Build a vector rvalue from an array of elements.

   "vec_type" should be a vector type, created using gcc_jit_type_get_vector.

   This API entrypoint was added in LIBGCCJIT_ABI_10; you can test for its
   presence using
     #ifdef LIBGCCJIT_HAVE_gcc_jit_context_new_rvalue_from_vector
*/
extern gcc_jit_rvalue *
gcc_jit_context_new_rvalue_from_vector (gcc_jit_context *ctxt,
                                        gcc_jit_location *loc,
                                        gcc_jit_type *vec_type,
                                        size_t num_elements,
                                        gcc_jit_rvalue **elements);

gcc/jit/ChangeLog:
* docs/cp/topics/expressions.rst (Vector expressions): New
section.
* docs/topics/compatibility.rst (LIBGCCJIT_ABI_10): New ABI tag.
* docs/topics/expressions.rst (Vector expressions): New section.
* docs/topics/types.rst (gcc_jit_type_get_vector): Add link to
gcc_jit_context_new_rvalue_from_vector.
* jit-common.h (gcc::jit:recording::vector_type): New forward
decl.
* jit-playback.c
(gcc::jit::playback::context::new_rvalue_from_vector): New method.
* jit-playback.h
(gcc::jit::playback::context::new_rvalue_from_vector): New method.
* jit-recording.c: In namespace gcc::jit::
(class comma_separated_string): New class.
(comma_separated_string::comma_separated_string): New ctor,
adapted from recording::call::make_debug_string.
(comma_separated_string::~comma_separated_string): New dtor.
In namespace gcc::jit::recording::
(context::new_rvalue_from_vector): New method.
(type::get_vector): Update for renaming of memento_of_get_vector.
(class memento_of_get_vector): Rename to...
(class vector_type): ..this.
(memento_of_new_rvalue_from_vector::memento_of_new_rvalue_from_vector):
New ctor.
(memento_of_new_rvalue_from_vector::replay_into): New method.
(memento_of_new_rvalue_from_vector::visit_children): New method.
(memento_of_new_rvalue_from_vector::make_debug_string): New
method.
(memento_of_new_rvalue_from_vector::write_reproducer): New method.
(call::make_debug_string): Split out arg-printing code into ctor
for comma_separated_string.
* jit-recording.h: In namespace gcc::jit::recording::
(context::new_rvalue_from_vector): New method.
(type::dyn_cast_vector_type): New virtual function.
(class memento_of_get_vector): Rename to...
(class vector_type): ...this.
(vector_type::unqualified): Remove this vfunc override in favor
of...
(vector_type::get_element_type): ...this new method.
(vector_type::get_num_units): New method.
(vector_type::dyn_cast_vector_type): New vfunc override.
(class memento_of_new_rvalue_from_vector): New class.
* libgccjit++.h (gccjit::context::new_rvalue): Add overload for
vector of rvalue.
* libgccjit.c (gcc_jit_context_new_binary_op): Strip off type
qualifications when checking that both operands have same type.
(gcc_jit_context_new_rvalue_from_vector): New API entrypoint.
* libgccjit.h
(LIBGCCJIT_HAVE_gcc_jit_context_new_rvalue_from_vector): New
macro.
(gcc_jit_context_new_rvalue_from_vector): New API entrypoint.
* libgccjit.map (LIBGCCJIT_ABI_10): New ABI tag.

gcc/testsuite/ChangeLog:
* jit.dg/test-expressions.c (make_test_of_vectors): New function.
(create_code): Call it.
* jit.dg/test-vector-rvalues.cc: New test case.

From-SVN: r253409

New target check: vect_peeling_profitable

gcc/ChangeLog:

2017-10-04  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>

* doc/sourcebuild.texi: Document vect_peeling_profitable.

gcc/testsuite/ChangeLog:

2017-10-04  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>

* g++.dg/vect/slp-pr56812.cc: xfail for targets which don't want
vector loop peeling.
* lib/target-supports.exp (check_effective_target_vect_peeling_profitable):
New proc.

From-SVN: r253407

New target check for double<->int conversions

gcc/ChangeLog:

2017-10-04  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>

* doc/sourcebuild.texi: Document vect_intdouble_cvt and
vect_doubleint_cvt.

gcc/testsuite/ChangeLog:

2017-10-04  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>

* gcc.dg/vect/pr66251.c: Replace vect_floatint_cvt with
vect_doubleint_cvt and vect_intfloat_cvt with vect_intdouble_cvt.
* gcc.dg/vect/vect-floatint-conversion-2.c: Replace
vect_floatint_cvt with vect_doubleint_cvt.
* gcc.dg/vect/vect-intfloat-conversion-3.c: Replace
vect_intfloat_cvt with vect_intdouble_cvt.
* gfortran.dg/vect/pr60510.f: Require vect_intdouble_cvt.
* gfortran.dg/vect/vect-8.f90: Make number of vectorized loops
depend on vect_intdouble_cvt.
* lib/target-supports.exp
(check_effective_target_vect_doubleint_cvt)
(check_effective_target_vect_intdouble_cvt): New procs.

From-SVN: r253406

New target check: vect_long_mult

We don't have a 64 bit vector integer multiply on z.  Add a specific
check for that.

gcc/ChangeLog:

2017-10-04  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>

* doc/sourcebuild.texi: Document vect_long_mult.

gcc/testsuite/ChangeLog:

2017-10-04  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>

* gcc.dg/vect/pr60656.c: Check vect_long_mult.
* lib/target-supports.exp (check_effective_target_vect_long_mult):
New proc.

From-SVN: r253405