sse.md (<avx512>_fmadd_<mode>_mask<round_name>, [...]): Use <round_nimm_predicate> instead of nonimmediate_operand.

* config/i386/sse.md (<avx512>_fmadd_<mode>_mask<round_name>,
<avx512>_fmadd_<mode>_mask3<round_name>,
<avx512>_fmsub_<mode>_mask<round_name>,
<avx512>_fmsub_<mode>_mask3<round_name>,
<avx512>_fnmadd_<mode>_mask<round_name>,
<avx512>_fnmadd_<mode>_mask3<round_name>,
<avx512>_fnmsub_<mode>_mask<round_name>,
<avx512>_fnmsub_<mode>_mask3<round_name>,
<avx512>_fmaddsub_<mode>_mask<round_name>,
<avx512>_fmaddsub_<mode>_mask3<round_name>,
<avx512>_fmsubadd_<mode>_mask<round_name>,
<avx512>_fmsubadd_<mode>_mask3<round_name>): Use
<round_nimm_predicate> instead of nonimmediate_operand.
(fmai_vmfmadd_<mode><round_name>, fmai_vmfmsub_<mode><round_name>,
fmai_vmfnmadd_<mode><round_name>, fmai_vmfnmsub_<mode><round_name>):
Use register_operand instead of <round_nimm_predicate> for the
operand that needs to match output.
(*fmai_fmadd_<mode>, *fmai_fmsub_<mode>,
*fmai_fnmadd_<mode><round_name>, *fmai_fnmsub_<mode><round_name>):
Likewise.  Formatting fixes.

From-SVN: r269869

re PR target/89784 (Missing AVX512 intrinsics)

PR target/89784
* config/i386/i386.c (enum ix86_builtins): Remove
IX86_BUILTIN_VFMSUBSD3_MASK3 and IX86_BUILTIN_VFMSUBSS3_MASK3.
* config/i386/i386-builtin.def (__builtin_ia32_vfmaddsd3_mask,
__builtin_ia32_vfmaddsd3_mask3, __builtin_ia32_vfmaddsd3_maskz,
__builtin_ia32_vfmsubsd3_mask3, __builtin_ia32_vfmaddss3_mask,
__builtin_ia32_vfmaddss3_mask3, __builtin_ia32_vfmaddss3_maskz,
__builtin_ia32_vfmsubss3_mask3): New builtins.
* config/i386/sse.md (avx512f_vmfmadd_<mode>_mask<round_name>,
avx512f_vmfmadd_<mode>_mask3<round_name>,
avx512f_vmfmadd_<mode>_maskz_1<round_name>,
*avx512f_vmfmsub_<mode>_mask<round_name>,
avx512f_vmfmsub_<mode>_mask3<round_name>,
*avx512f_vmfmasub_<mode>_maskz_1<round_name>,
*avx512f_vmfnmadd_<mode>_mask<round_name>,
*avx512f_vmfnmadd_<mode>_mask3<round_name>,
*avx512f_vmfnmadd_<mode>_maskz_1<round_name>,
*avx512f_vmfnmsub_<mode>_mask<round_name>,
*avx512f_vmfnmsub_<mode>_mask3<round_name>,
*avx512f_vmfnmasub_<mode>_maskz_1<round_name>): New define_insns.
(avx512f_vmfmadd_<mode>_maskz<round_expand_name>): New define_expand.
* config/i386/avx512fintrin.h (_mm_mask_fmadd_sd, _mm_mask_fmadd_ss,
_mm_mask3_fmadd_sd, _mm_mask3_fmadd_ss, _mm_maskz_fmadd_sd,
_mm_maskz_fmadd_ss, _mm_mask_fmsub_sd, _mm_mask_fmsub_ss,
_mm_mask3_fmsub_sd, _mm_mask3_fmsub_ss, _mm_maskz_fmsub_sd,
_mm_maskz_fmsub_ss, _mm_mask_fnmadd_sd, _mm_mask_fnmadd_ss,
_mm_mask3_fnmadd_sd, _mm_mask3_fnmadd_ss, _mm_maskz_fnmadd_sd,
_mm_maskz_fnmadd_ss, _mm_mask_fnmsub_sd, _mm_mask_fnmsub_ss,
_mm_mask3_fnmsub_sd, _mm_mask3_fnmsub_ss, _mm_maskz_fnmsub_sd,
_mm_maskz_fnmsub_ss, _mm_mask_fmadd_round_sd, _mm_mask_fmadd_round_ss,
_mm_mask3_fmadd_round_sd, _mm_mask3_fmadd_round_ss,
_mm_maskz_fmadd_round_sd, _mm_maskz_fmadd_round_ss,
_mm_mask_fmsub_round_sd, _mm_mask_fmsub_round_ss,
_mm_mask3_fmsub_round_sd, _mm_mask3_fmsub_round_ss,
_mm_maskz_fmsub_round_sd, _mm_maskz_fmsub_round_ss,
_mm_mask_fnmadd_round_sd, _mm_mask_fnmadd_round_ss,
_mm_mask3_fnmadd_round_sd, _mm_mask3_fnmadd_round_ss,
_mm_maskz_fnmadd_round_sd, _mm_maskz_fnmadd_round_ss,
_mm_mask_fnmsub_round_sd, _mm_mask_fnmsub_round_ss,
_mm_mask3_fnmsub_round_sd, _mm_mask3_fnmsub_round_ss,
_mm_maskz_fnmsub_round_sd, _mm_maskz_fnmsub_round_ss): New intrinsics.

* gcc.target/i386/sse-13.c (__builtin_ia32_vfmaddsd3_mask,
__builtin_ia32_vfmaddsd3_mask3, __builtin_ia32_vfmaddsd3_maskz,
__builtin_ia32_vfmsubsd3_mask3, __builtin_ia32_vfmaddss3_mask,
__builtin_ia32_vfmaddss3_mask3, __builtin_ia32_vfmaddss3_maskz,
__builtin_ia32_vfmsubss3_mask3): Define.
* gcc.target/i386/sse-23.c (__builtin_ia32_vfmaddsd3_mask,
__builtin_ia32_vfmaddsd3_mask3, __builtin_ia32_vfmaddsd3_maskz,
__builtin_ia32_vfmsubsd3_mask3, __builtin_ia32_vfmaddss3_mask,
__builtin_ia32_vfmaddss3_mask3, __builtin_ia32_vfmaddss3_maskz,
__builtin_ia32_vfmsubss3_mask3): Define.
* gcc.target/i386/avx-1.c (__builtin_ia32_vfmaddsd3_mask,
__builtin_ia32_vfmaddsd3_mask3, __builtin_ia32_vfmaddsd3_maskz,
__builtin_ia32_vfmsubsd3_mask3, __builtin_ia32_vfmaddss3_mask,
__builtin_ia32_vfmaddss3_mask3, __builtin_ia32_vfmaddss3_maskz,
__builtin_ia32_vfmsubss3_mask3): Define.
* gcc.target/i386/sse-14.c: Add tests for
_mm_mask{,3,z}_f{,n}m{add,sub}_round_s{s,d} builtins.
* gcc.target/i386/sse-22.c: Likewise.

2019-03-22  Hongtao Liu  <hongtao.liu@intel.com>

* gcc.target/i386/avx512f-vfmaddXXXsd-1.c (avx512f_test): Add tests
for _mm_mask{,3,z}_*.
* gcc.target/i386/avx512f-vfmaddXXXss-1.c (avx512f_test): Likewise.
* gcc.target/i386/avx512f-vfmsubXXXsd-1.c (avx512f_test): Likewise.
* gcc.target/i386/avx512f-vfmsubXXXss-1.c (avx512f_test): Likewise.
* gcc.target/i386/avx512f-vfnmaddXXXsd-1.c (avx512f_test): Likewise.
* gcc.target/i386/avx512f-vfnmaddXXXss-1.c (avx512f_test): Likewise.
* gcc.target/i386/avx512f-vfnmsubXXXsd-1.c (avx512f_test): Likewise.
* gcc.target/i386/avx512f-vfnmsubXXXss-1.c (avx512f_test): Likewise.
* gcc.target/i386/avx512f-vfmaddXXXsd-2.c: New test.
* gcc.target/i386/avx512f-vfmaddXXXss-2.c: New test.
* gcc.target/i386/avx512f-vfmsubXXXsd-2.c: New test.
* gcc.target/i386/avx512f-vfmsubXXXss-2.c: New test.
* gcc.target/i386/avx512f-vfnmaddXXXsd-2.c: New test.
* gcc.target/i386/avx512f-vfnmaddXXXss-2.c: New test.
* gcc.target/i386/avx512f-vfnmsubXXXsd-2.c: New test.
* gcc.target/i386/avx512f-vfnmsubXXXss-2.c: New test.

From-SVN: r269868

PR tree-optimization/89350 - Wrong -Wstringop-overflow= warning since r261518

gcc/ChangeLog:

PR tree-optimization/89350
* builtins.c (compute_objsize): Also ignore offsets whose upper
bound is negative.
* gimple-ssa-warn-restrict.c (builtin_memref): Add new member.
(builtin_memref::builtin_memref): Initialize new member.
Allow EXPR to be null.
(builtin_memref::extend_offset_range): Replace local with a member.
Avoid assuming pointer offsets are unsigned.
(builtin_memref::set_base_and_offset): Determine base object
before computing offset range.
(builtin_access::builtin_access): Handle memset.
(builtin_access::generic_overlap): Replace local with a member.
(builtin_access::strcat_overlap): Same.
(builtin_access::overlap): Same.
(maybe_diag_overlap): Same.
(maybe_diag_access_bounds): Same.
(wrestrict_dom_walker::check_call): Handle memset.
(check_bounds_or_overlap): Same.

gcc/testsuite/ChangeLog:

PR tree-optimization/89350
* gcc.dg/Wstringop-overflow.c: Xfail overly ambitious tests.
* gcc.dg/Wstringop-overflow-11.c: New test.
* gcc.dg/Wstringop-overflow-12.c: New test.
* gcc.dg/pr89350.c: New test.
* gcc.dg/pr40340-1.c: Adjust expected warning.
* gcc.dg/pr40340-2.c: Same.
* gcc.dg/pr40340-4.c: Same.
* gcc.dg/pr40340-5.c: Same.

From-SVN: r269867

Daily bump.

From-SVN: r269866

Integrate C++17 parallel algorithms

This is the Intel implementation of the C++17 parallel
algorithms, which has been donated to both GCC and LLVM. The upstream
project is at -

https://reviews.llvm.org/source/pstl/

The new files in the include/pstl sub-directory are covered by the
LICENSE.txt in that sub-directory, as are the tests in
testsuite/**/pstl/*

* include/Makefile.am (std_header): Add ${std_srcdir}/execution.
(pstl_srcdir, pstl_builddir, pstl_headers): New variables.
(allstamped): Add stamp-pstl.
(install-headers): Add ptsl_builddir.
* include/Makefile.in: Regenerate.
* include/bits/c++config: Add pstl configuration.
* include/pstl/LICENSE.txt: New file.
* include/pstl/algorithm_fwd.h: New file.
* include/pstl/algorithm_impl.h: New file.
* include/pstl/execution_defs.h: New file.
* include/pstl/execution_impl.h: New file.
* include/pstl/glue_algorithm_defs.h: New file.
* include/pstl/glue_algorithm_impl.h: New file.
* include/pstl/glue_execution_defs.h: New file.
* include/pstl/glue_memory_defs.h: New file.
* include/pstl/glue_memory_impl.h: New file.
* include/pstl/glue_numeric_defs.h: New file.
* include/pstl/glue_numeric_impl.h: New file.
* include/pstl/memory_impl.h: New file.
* include/pstl/numeric_fwd.h: New file.
* include/pstl/numeric_impl.h: New file.
* include/pstl/parallel_backend.h: New file.
* include/pstl/parallel_backend_tbb.h: New file.
* include/pstl/parallel_backend_utils.h: New file.
* include/pstl/parallel_impl.h: New file.
* include/pstl/pstl_config.h: New file.
* include/pstl/unseq_backend_simd.h: New file.
* include/pstl/utils.h: New file.
* include/std/algorithm: Include parallel algorithm implementations.
* include/std/execution: New file.
* include/std/memory: Include parallel algorithm implementations.
* include/std/numeric: Include parallel algorithm implementations.
* include/std/version: Add parallel algorithms feature test macro.
* testsuite/util/pstl/pstl_test_config.h: New file.
* testsuite/util/pstl/test_utils.h: New file.
* testsuite/20_util/specialized_algorithms/pstl/uninitialized_construct.cc: New file.
* testsuite/20_util/specialized_algorithms/pstl/uninitialized_copy_move.cc: New file.
* testsuite/20_util/specialized_algorithms/pstl/uninitialized_fill_destroy.cc: New file.
* testsuite/25_algorithms/pstl/alg_merge/inplace_merge.cc: New file.
* testsuite/25_algorithms/pstl/alg_merge/merge.cc: New file.
* testsuite/25_algorithms/pstl/alg_modifying_operations/copy_if.cc: New file.
* testsuite/25_algorithms/pstl/alg_modifying_operations/copy_move.cc: New file.
* testsuite/25_algorithms/pstl/alg_modifying_operations/fill.cc: New file.
* testsuite/25_algorithms/pstl/alg_modifying_operations/generate.cc: New file.
* testsuite/25_algorithms/pstl/alg_modifying_operations/is_partitioned.cc: New file.
* testsuite/25_algorithms/pstl/alg_modifying_operations/partition.cc: New file.
* testsuite/25_algorithms/pstl/alg_modifying_operations/partition_copy.cc: New file.
* testsuite/25_algorithms/pstl/alg_modifying_operations/remove.cc: New file.
* testsuite/25_algorithms/pstl/alg_modifying_operations/remove_copy.cc: New file.
* testsuite/25_algorithms/pstl/alg_modifying_operations/replace.cc: New file.
* testsuite/25_algorithms/pstl/alg_modifying_operations/replace_copy.cc: New file.
* testsuite/25_algorithms/pstl/alg_modifying_operations/rotate.cc: New file.
* testsuite/25_algorithms/pstl/alg_modifying_operations/rotate_copy.cc: New file.
* testsuite/25_algorithms/pstl/alg_modifying_operations/swap_ranges.cc: New file.
* testsuite/25_algorithms/pstl/alg_modifying_operations/transform_binary.cc: New file.
* testsuite/25_algorithms/pstl/alg_modifying_operations/transform_unary.cc: New file.
* testsuite/25_algorithms/pstl/alg_modifying_operations/unique.cc: New file.
* testsuite/25_algorithms/pstl/alg_modifying_operations/unique_copy_equal.cc: New file.
* testsuite/25_algorithms/pstl/alg_nonmodifying/adjacent_find.cc: New file.
* testsuite/25_algorithms/pstl/alg_nonmodifying/all_of.cc: New file.
* testsuite/25_algorithms/pstl/alg_nonmodifying/any_of.cc: New file.
* testsuite/25_algorithms/pstl/alg_nonmodifying/count.cc: New file.
* testsuite/25_algorithms/pstl/alg_nonmodifying/equal.cc: New file.
* testsuite/25_algorithms/pstl/alg_nonmodifying/find.cc: New file.
* testsuite/25_algorithms/pstl/alg_nonmodifying/find_end.cc: New file.
* testsuite/25_algorithms/pstl/alg_nonmodifying/find_first_of.cc: New file.
* testsuite/25_algorithms/pstl/alg_nonmodifying/find_if.cc: New file.
* testsuite/25_algorithms/pstl/alg_nonmodifying/for_each.cc: New file.
* testsuite/25_algorithms/pstl/alg_nonmodifying/mismatch.cc: New file.
* testsuite/25_algorithms/pstl/alg_nonmodifying/none_of.cc: New file.
* testsuite/25_algorithms/pstl/alg_nonmodifying/nth_element.cc: New file.
* testsuite/25_algorithms/pstl/alg_nonmodifying/reverse.cc: New file.
* testsuite/25_algorithms/pstl/alg_nonmodifying/reverse_copy.cc: New file.
* testsuite/25_algorithms/pstl/alg_nonmodifying/search_n.cc: New file.
* testsuite/25_algorithms/pstl/alg_sorting/includes.cc: New file.
* testsuite/25_algorithms/pstl/alg_sorting/is_heap.cc: New file.
* testsuite/25_algorithms/pstl/alg_sorting/is_sorted.cc: New file.
* testsuite/25_algorithms/pstl/alg_sorting/lexicographical_compare.cc: New file.
* testsuite/25_algorithms/pstl/alg_sorting/minmax_element.cc: New file.
* testsuite/25_algorithms/pstl/alg_sorting/partial_sort.cc: New file.
* testsuite/25_algorithms/pstl/alg_sorting/partial_sort_copy.cc: New file.
* testsuite/25_algorithms/pstl/alg_sorting/set.cc: New file.
* testsuite/25_algorithms/pstl/alg_sorting/sort.cc: New file.
* testsuite/26_numerics/pstl/numeric_ops/adjacent_difference.cc: New file.
* testsuite/26_numerics/pstl/numeric_ops/reduce.cc: New file.
* testsuite/26_numerics/pstl/numeric_ops/scan.cc: New file.
* testsuite/26_numerics/pstl/numeric_ops/transform_reduce.cc: New file.
* testsuite/26_numerics/pstl/numeric_ops/transform_scan.cc: New file.
* testsuite/testsuite/20_util/specialized_algorithms/pstl/uninitialized_construct.cc: New file.
* testsuite/testsuite/20_util/specialized_algorithms/pstl/uninitialized_copy_move.cc: New file.
* testsuite/testsuite/20_util/specialized_algorithms/pstl/uninitialized_fill_destroy.cc: New file.
* testsuite/testsuite/25_algorithms/pstl/alg_merge/inplace_merge.cc: New file.
* testsuite/testsuite/25_algorithms/pstl/alg_merge/merge.cc: New file.
* testsuite/testsuite/25_algorithms/pstl/alg_modifying_operations/copy_if.cc: New file.
* testsuite/testsuite/25_algorithms/pstl/alg_modifying_operations/copy_move.cc: New file.
* testsuite/testsuite/25_algorithms/pstl/alg_modifying_operations/fill.cc: New file.
* testsuite/testsuite/25_algorithms/pstl/alg_modifying_operations/generate.cc: New file.
* testsuite/testsuite/25_algorithms/pstl/alg_modifying_operations/is_partitioned.cc: New file.
* testsuite/testsuite/25_algorithms/pstl/alg_modifying_operations/partition.cc: New file.
* testsuite/testsuite/25_algorithms/pstl/alg_modifying_operations/partition_copy.cc: New file.
* testsuite/testsuite/25_algorithms/pstl/alg_modifying_operations/remove.cc: New file.
* testsuite/testsuite/25_algorithms/pstl/alg_modifying_operations/remove_copy.cc: New file.
* testsuite/testsuite/25_algorithms/pstl/alg_modifying_operations/replace.cc: New file.
* testsuite/testsuite/25_algorithms/pstl/alg_modifying_operations/replace_copy.cc: New file.
* testsuite/testsuite/25_algorithms/pstl/alg_modifying_operations/rotate.cc: New file.
* testsuite/testsuite/25_algorithms/pstl/alg_modifying_operations/rotate_copy.cc: New file.
* testsuite/testsuite/25_algorithms/pstl/alg_modifying_operations/swap_ranges.cc: New file.
* testsuite/testsuite/25_algorithms/pstl/alg_modifying_operations/transform_binary.cc: New file.
* testsuite/testsuite/25_algorithms/pstl/alg_modifying_operations/transform_unary.cc: New file.
* testsuite/testsuite/25_algorithms/pstl/alg_modifying_operations/unique.cc: New file.
* testsuite/testsuite/25_algorithms/pstl/alg_modifying_operations/unique_copy_equal.cc: New file.
* testsuite/testsuite/25_algorithms/pstl/alg_nonmodifying/adjacent_find.cc: New file.
* testsuite/testsuite/25_algorithms/pstl/alg_nonmodifying/all_of.cc: New file.
* testsuite/testsuite/25_algorithms/pstl/alg_nonmodifying/any_of.cc: New file.
* testsuite/testsuite/25_algorithms/pstl/alg_nonmodifying/count.cc: New file.
* testsuite/testsuite/25_algorithms/pstl/alg_nonmodifying/equal.cc: New file.
* testsuite/testsuite/25_algorithms/pstl/alg_nonmodifying/find.cc: New file.
* testsuite/testsuite/25_algorithms/pstl/alg_nonmodifying/find_end.cc: New file.
* testsuite/testsuite/25_algorithms/pstl/alg_nonmodifying/find_first_of.cc: New file.
* testsuite/testsuite/25_algorithms/pstl/alg_nonmodifying/find_if.cc: New file.
* testsuite/testsuite/25_algorithms/pstl/alg_nonmodifying/for_each.cc: New file.
* testsuite/testsuite/25_algorithms/pstl/alg_nonmodifying/mismatch.cc: New file.
* testsuite/testsuite/25_algorithms/pstl/alg_nonmodifying/none_of.cc: New file.
* testsuite/testsuite/25_algorithms/pstl/alg_nonmodifying/nth_element.cc: New file.
* testsuite/testsuite/25_algorithms/pstl/alg_nonmodifying/reverse.cc: New file.
* testsuite/testsuite/25_algorithms/pstl/alg_nonmodifying/reverse_copy.cc: New file.
* testsuite/testsuite/25_algorithms/pstl/alg_nonmodifying/search_n.cc: New file.
* testsuite/testsuite/25_algorithms/pstl/alg_sorting/includes.cc: New file.
* testsuite/testsuite/25_algorithms/pstl/alg_sorting/is_heap.cc: New file.
* testsuite/testsuite/25_algorithms/pstl/alg_sorting/is_sorted.cc: New file.
* testsuite/testsuite/25_algorithms/pstl/alg_sorting/lexicographical_compare.cc: New file.
* testsuite/testsuite/25_algorithms/pstl/alg_sorting/minmax_element.cc: New file.
* testsuite/testsuite/25_algorithms/pstl/alg_sorting/partial_sort.cc: New file.
* testsuite/testsuite/25_algorithms/pstl/alg_sorting/partial_sort_copy.cc: New file.
* testsuite/testsuite/25_algorithms/pstl/alg_sorting/set.cc: New file.
* testsuite/testsuite/25_algorithms/pstl/alg_sorting/sort.cc: New file.
* testsuite/testsuite/26_numerics/pstl/numeric_ops/adjacent_difference.cc: New file.
* testsuite/testsuite/26_numerics/pstl/numeric_ops/reduce.cc: New file.
* testsuite/testsuite/26_numerics/pstl/numeric_ops/scan.cc: New file.
* testsuite/testsuite/26_numerics/pstl/numeric_ops/transform_reduce.cc: New file.
* testsuite/testsuite/26_numerics/pstl/numeric_ops/transform_scan.cc: New file.

From-SVN: r269863

re PR lto/89692 (ICE in streamer_write_chain, at tree-streamer-out.c:506)

PR lto/89692
* tree.c (fld_type_variant, fld_incomplete_type_of,
fld_process_array_type): Call fld->pset.add and don't call
add_tree_to_fld_list if it returns true.
(free_lang_data_in_type): Similarly with self-recursive call.  Purge
non-marked types from TYPE_NEXT_VARIANT list.
(find_decls_types_r): Call fld_worklist_push for TYPE_CANONICAL (t).

* g++.dg/other/pr89692.C: New test.

From-SVN: r269862

re PR c++/71446 (Incorrect overload resolution when using designated initializers)

PR c++/71446
* call.c (filed_in_pset): Change pset from hash_set<tree> * to
hash_set<tree, true> &, adjust uses accordingly.
(build_aggr_conv): Change pset from hash_set<tree> *
to hash_set<tree, true>.  Replace goto fail; with return NULL;,
adjust pset uses.

From-SVN: r269861

re PR c++/89767 (ICE with tuple and optimization)

PR c++/89767
* parser.c (cp_parser_lambda_introducer): Add ids and first_capture_id
variables, check for duplicates in this function.
* lambda.c (add_capture): Don't check for duplicates nor use
IDENTIFIER_MARKED.
(register_capture_members): Don't clear IDENTIFIER_MARKED here.

* g++.dg/cpp1y/lambda-init18.C: New test.
* g++.dg/cpp1y/lambda-init19.C: New test.
* g++.dg/cpp1y/pr89767.C: New test.

From-SVN: r269860

hash-table.h (hash_table): Add Lazy template parameter defaulted to false...

* hash-table.h (hash_table): Add Lazy template parameter defaulted
to false, if true, don't alloc_entries during construction, but defer
it to the first method that needs m_entries allocated.
(hash_table::hash_table, hash_table::~hash_table,
hash_table::alloc_entries, hash_table::find_empty_slot_for_expand,
hash_table::too_empty_p, hash_table::expand, hash_table::empty_slow,
hash_table::clear_slot, hash_table::traverse_noresize,
hash_table::traverse, hash_table::iterator::slide): Adjust all methods.
* hash-set.h (hash_set): Add Lazy template parameter defaulted to
false.
(hash_set::contains): If Lazy is true, use find_slot_with_hash with
NO_INSERT instead of find_with_hash.
(hash_set::traverse, hash_set::iterator, hash_set::iterator::m_iter,
hash_set::m_table): Add Lazy to template params of hash_table.
(gt_ggc_mx, gt_pch_nx): Use false as Lazy in hash_set template param.
* attribs.c (test_attribute_exclusions): Likewise.
* hash-set-tests.c (test_set_of_strings): Add iterator tests for
hash_set.  Add tests for hash_set with Lazy = true.
c-family/
* c-common.c (per_file_includes_t): Use false as Lazy in hash_set
template param.
jit/
* jit-recording.c (reproducer::m_set_identifiers): Use false as Lazy
in hash_set template param.

From-SVN: r269859

[PR72741] Properly handle clauses specifying the level of parallelism for 'external' Fortran OpenACC routines

..., so as to also for these enable the generic middle end OMP code to verify
proper nesting of loops/routines regarding their levels of parallelism.

gcc/fortran/
PR fortran/72741
* openmp.c (gfc_match_oacc_routine): Set the level of parallelism
for all variants.
(gfc_resolve_oacc_routines): Call gfc_add_omp_declare_target.
gcc/testsuite/
PR fortran/72741
* c-c++-common/goacc/routine-3-extern.c: New file.
* c-c++-common/goacc/routine-3.c: Adjust.
* c-c++-common/goacc/routine-4-extern.c: New file.
* c-c++-common/goacc/routine-4.c: Adjust.
* gfortran.dg/goacc/routine-module-3.f90: New file.
* gfortran.dg/goacc/routine-external-level-of-parallelism-1.f: New
file.
* gfortran.dg/goacc/routine-external-level-of-parallelism-2.f:
Likewise.

Co-Authored-By: Cesar Philippidis <cesar@codesourcery.com>
From-SVN: r269858

[PR89773] Fortran OpenACC 'routine' directive refuses procedures with implicit EXTERNAL attribute

gcc/fortran/
PR fortran/89773
* gfortran.h (gfc_oacc_routine_name): Add loc member.
(gfc_resolve_oacc_routines): Declare.
* openmp.c (gfc_match_oacc_routine): Move some error checking
into...
(gfc_resolve_oacc_routines): ... this new function.
* resolve.c (resolve_codes): Call it.
gcc/testsuite/
PR fortran/89773
* gfortran.dg/goacc/pr89773.f90: New file.
* gfortran.dg/goacc/pr77765.f90: Adjust.
* gfortran.dg/goacc/routine-6.f90: Adjust, and extend.

From-SVN: r269857

[PR72741] The name in a Fortran OpenACC 'routine' directive refers to the containing subroutine or function

gcc/fortran/
PR fortran/72741
* openmp.c (gfc_match_oacc_routine): Clarify.
gcc/testsuite/
PR fortran/72741
* gfortran.dg/goacc/routine-module-mod-1.f90: Update.

From-SVN: r269856

[PR72741] Encode OpenACC 'routine' directive's level of parallelism inside Fortran module files

If 'use'ing with an old GCC a new module file (with OpenACC 'routine'
directive's level of parallelism encoded), then that expectedly fails as
follows:

    f951: Fatal Error: Reading module 'routine_module_mod_1' at line 27 column 21: find_enum(): Enum not found

If 'use'ing with a new GCC an old module file (without OpenACC 'routine'
directive's level of parallelism encoded), then that (silently) continues to
accept the module file, and will proceed with the previous, erroneous behavior.

These seem to be acceptable compromises, instead of incrementing 'MOD_VERSION'.

gcc/fortran/
PR fortran/72741
* module.c (verify_OACC_ROUTINE_LOP_NONE): New function.
(enum ab_attribute): Add AB_OACC_ROUTINE_LOP_GANG,
AB_OACC_ROUTINE_LOP_WORKER, AB_OACC_ROUTINE_LOP_VECTOR,
AB_OACC_ROUTINE_LOP_SEQ.
(attr_bits): Add these.
(mio_symbol_attribute): Handle these.
gcc/testsuite/
PR fortran/72741
* gfortran.dg/goacc/routine-module-1.f90: New file.
* gfortran.dg/goacc/routine-module-2.f90: Likewise.
* gfortran.dg/goacc/routine-module-mod-1.f90: Likewise.

From-SVN: r269855

[testsuite, Fortran] Provide 'dg-compile-aux-modules' in 'gfortran.dg/goacc/goacc.exp'

..., as yet another copy from 'gfortran.dg/dg.exp', which there are a few
already.

gcc/testsuite/
* gfortran.dg/goacc/goacc.exp (dg-compile-aux-modules): New proc.

From-SVN: r269854

[testsuite] Fix 'dg-compile-aux-modules' diagnostic

gcc/testsuite/
PR fortran/56408
* gcc.target/powerpc/ppc-fortran/ppc-fortran.exp
(dg-compile-aux-modules): Fix diagnostic.
* gfortran.dg/coarray/caf.exp (dg-compile-aux-modules): Likewise.
* gfortran.dg/dg.exp (dg-compile-aux-modules): Likewise.

From-SVN: r269851

[testsuite, Fortran] Apply DejaGnu 1.4.4 work-around also to 'gfortran.dg/coarray/caf.exp:dg-compile-aux-modules'

See trunk r215293.  This unifies all 'dg-compile-aux-modules' instances.

gcc/testsuite/
PR fortran/56408
* gfortran.dg/coarray/caf.exp (dg-compile-aux-modules): Workaround
missing nexted dg-test call support in dejaGNU 1.4.4.

From-SVN: r269848

[testsuite, Fortran] Consistently set 'DEFAULT_FFLAGS'

In the same 'runtest' instance, 'global' variables persist from one '*.exp'
file to another.

All other '*.exp' files are using " -pedantic-errors" instead of the empty
string as the default for 'DEFAULT_FFLAGS'.  Thus this setting of
'DEFAULT_FFLAGS' is not idempotent, depends on whether
'gfortran.dg/ieee/ieee.exp', or an other defining '*.exp' file is executed
first.

gcc/testsuite/
PR fortran/29383
* gfortran.dg/ieee/ieee.exp (DEFAULT_FFLAGS): Set the same as in
other '*.exp' files.

From-SVN: r269845

compiler: add a newline to function receiver type's debug dump

    Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/168408

From-SVN: r269841

re PR middle-end/89779 (internal compiler error: tree check: expected class ‘type’, have ‘exceptional’ (error_mark) in tree_nop_conversion_p, at tree.c:12798)

2019-03-21  Richard Biener  <rguenther@suse.de>

PR tree-optimization/89779
* tree.c (tree_nop_conversion): Consolidate and fix defensive
checks with respect to released SSA names now having error_mark_node
type.
* fold-const.c (operand_equal_p): Likewise.

* gcc.dg/torture/pr89779.c: New testcase.

From-SVN: r269838

In C++17 <math.h> should not put special functions in global namespace

IS 29124 8.2 [sf.mathh] says that <math.h> should add the names of the
special functions to the global namespace.  However, C++17 Annex D
[depr.c.headers] excludes those functions explicitly, so they should not
be placed in the global namespace unconditionally for C++17.

Only add them to the global namespace when IS 29124 is explicitly
requested via the __STDCPP_WANT_MATH_SPEC_FUNCS__ macro.

* include/c_compatibility/math.h [!__STDCPP_WANT_MATH_SPEC_FUNCS__]
(assoc_laguerre, assoc_laguerref, assoc_laguerrel, assoc_legendre)
(assoc_legendref, assoc_legendrel, beta, betaf, betal, comp_ellint_1)
(comp_ellint_1f, comp_ellint_1l, comp_ellint_2, comp_ellint_2f)
(comp_ellint_2l, comp_ellint_3, comp_ellint_3f, comp_ellint_3l)
(cyl_bessel_i, cyl_bessel_if, cyl_bessel_il, cyl_bessel_j)
(cyl_bessel_jf, cyl_bessel_jl, cyl_bessel_k, cyl_bessel_kf)
(cyl_bessel_kl, cyl_neumann, cyl_neumannf, cyl_neumannl, ellint_1)
(ellint_1f, ellint_1l, ellint_2, ellint_2f, ellint_2l, ellint_3)
(ellint_3f, ellint_3l, expint, expintf, expintl, hermite, hermitef)
(hermitel, laguerre, laguerref, laguerrel, legendre, legendref)
(legendrel, riemann_zeta, riemann_zetaf, riemann_zetal, sph_bessel)
(sph_besself, sph_bessell, sph_legendre, sph_legendref, sph_legendrel)
(sph_neumann, sph_neumannf, sph_neumannl): Only add using-declarations
when the special functions IS is enabled, not for C++17.
* testsuite/26_numerics/headers/cmath/functions_global_c++17.cc:
Replace with ...
* testsuite/26_numerics/headers/cmath/functions_global.cc: New test,
without checks for special functions in C++17.
* testsuite/26_numerics/headers/cmath/special_functions_global.cc:
New test.

From-SVN: r269837

PR libstdc++/88066 Use <> for includes not ""

These headers were missed in the previous commit for this bug.

There are also several "" includes in the profile mode headers, but
because they're deprecated I'm not fixing them.

* include/backward/hash_map: Use <> for includes not "".
* include/backward/hash_set: Likewise.
* include/backward/strstream: Likewise.
* include/tr1/bessel_function.tcc: Likewise.
* include/tr1/exp_integral.tcc: Likewise.
* include/tr1/legendre_function.tcc: Likewise.
* include/tr1/modified_bessel_func.tcc: Likewise.
* include/tr1/riemann_zeta.tcc: Likewise.

From-SVN: r269835

re PR c++/78645 (ICE on invalid code (Segmentation fault, cxx_eval_call_expression))

2019-03-21  Paolo Carlini  <paolo.carlini@oracle.com>

PR c++/78645
* g++.dg/cpp0x/constexpr-ice20.C: New.

From-SVN: r269834

re PR c++/89571 (ICE in nothrow_spec_p, at cp/except.c:1238)

/cp
2019-03-21  Paolo Carlini  <paolo.carlini@oracle.com>

PR c++/89571
* method.c (after_nsdmi_defaulted_late_checks): Avoid passing
error_mark_node to comp_except_specs.

/testsuite
2019-03-21  Paolo Carlini  <paolo.carlini@oracle.com>

PR c++/89571
* g++.dg/cpp0x/noexcept37.C: New.

From-SVN: r269832

Daily bump.

From-SVN: r269831

d: Fix ICE force_type_die, at dwarf2out.c using nested types

In functions whose return type is instantiated from a nested template,
make sure that all members of the instance are emitted before finishing
the outer function, otherwise they will be removed during the
prune_unused_types pass.

gcc/d/ChangeLog:

2019-03-21  Iain Buclaw  <ibuclaw@gdcproject.org>

PR d/89017
* d-codegen.cc (d_decl_context): Skip over template instances when
finding the context.
* decl.cc (DeclVisitor::visit(TemplateDeclaration)): New override.
(build_type_decl): Include parameters in name of template types.

gcc/testsuite/ChangeLog:

2019-03-21  Iain Buclaw  <ibuclaw@gdcproject.org>

PR d/89017
* gdc.dg/pr89017.d: New test.

From-SVN: r269828

re PR fortran/71861 ([F03] ICE in write_symbol(): bad module symbol)

fix PR 71861

2019-03-20  Janus Weil  <janus@gcc.gnu.org>

PR fortran/71861
* symbol.c (check_conflict): ABSTRACT attribute conflicts with
INTRINSIC attribute.

2019-03-20  Janus Weil  <janus@gcc.gnu.org>

PR fortran/71861
* gfortran.dg/interface_abstract_5.f90: New test case.

From-SVN: r269827

PR c++/87480 - decltype of member access in default template arg

The issue here is that declval<T>().d is considered instantiation-dependent
within a template, as the access to 'd' might depend on the particular
specialization.  But when we're deducing template arguments for a call, we
know that the call and the arguments are non-dependent, so we can do the
substitution as though we aren't in a template.  Which strictly speaking we
aren't, since the default argument is considered a separate definition.

* pt.c (type_unification_real): Accept a dependent result in
template context.

From-SVN: r269826

S/390: Fix PR89775. Stackpointer save/restore instructions removed

Even if a global register is being clobbered in a function we usually
do not save and restore it. However, we still have to do this if it is
a special register. Most of the places in the backend handle this
correctly but not the prologue/epilogue optimization.

gcc/ChangeLog:

2019-03-20  Andreas Krebbel  <krebbel@linux.ibm.com>

PR target/89775
* config/s390/s390.c (global_not_special_regno_p): Move to make it
available to ...
(s390_optimize_register_info): Use global_not_special_regno_p to
check for global regs.

2019-03-20  Jakub Jelinek  <jakub@redhat.com>

PR target/89775
* gcc.target/s390/pr89775-1.c: New test.
* gcc.target/s390/pr89775-2.c: New test.

From-SVN: r269823

re PR target/89752 (ICE in emit_move_insn, at expr.c:3723)

PR target/89752
* lra-constraints.c (process_alt_operands) <reg>: For BLKmode, don't
update this_alternative nor this_alternative_set.

* g++.target/aarch64/aarch64.exp: New file.
* g++.target/aarch64/pr89752.C: New test.

From-SVN: r269819

Daily bump.

From-SVN: r269817

PR tree-optimization/89688 - -Wstringop-overflow confused by const 2D array of char

gcc/c/ChangeLog:

PR tree-optimization/89688
* c-decl.c (finish_decl): Call braced_lists_to_string for more
kinds of initializers.

gcc/c-family/ChangeLog:

PR tree-optimization/89688
* c-common.c (braced_list_to_string): Make static.
(braced_lists_to_strings): Define new function.
* c-common.h (braced_list_to_string): Remove.
(braced_lists_to_strings): Declare.

gcc/cp/ChangeLog:

PR tree-optimization/89688
* typeck2.c (store_init_value): Call braced_lists_to_string for more
kinds of initializers.

gcc/testsuite/ChangeLog:

PR tree-optimization/89688
* gcc.dg/strlenopt-61.c: New test.
* g++.dg/warn/Wstringop-overflow-2.C: New test.

From-SVN: r269814

RISC-V: Fix %lo overflow with BLKmode references.

gcc/
PR target/89411
* config/riscv/riscv.c (riscv_valid_lo_sum_p): New arg x.  New locals
align, size, offset.  Use them to handle a BLKmode reference.  Update
comment.
(riscv_classify_address): Pass info->offset to riscv_valid_lo_sum_p.

gcc/testsuite/
PR target/89411
* gcc.target/riscv/losum-overflow.c: New test.

From-SVN: r269813

re PR rtl-optimization/89768 (ICE in compare_and_jump_seq at loop-unroll.c:838)

PR rtl-optimization/89768
* loop-unroll.c (unroll_loop_constant_iterations): Use gen_int_mode
instead of GEN_INT.
(unroll_loop_runtime_iterations): Likewise.

From-SVN: r269812

compiler,runtime: pass old slice's ptr/len/cap by value to growslice

    In the C calling convention, on AMD64, and probably a number of
    other architectures, a 3-word struct argument is passed on stack.
    This is less efficient than passing in three registers. Further,
    this may affect the code generation in other part of the program,
    even if the function is not actually called.

    Slices are common in Go and append is a common slice operation,
    which calls growslice in the growing path. To improve the code
    generation, pass the slice header's three fields as separate
    values, instead of a struct, to growslice.

    The drawback is that this makes the runtime implementation
    slightly diverges from the gc runtime.

    Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/168277

From-SVN: r269811

PR tree-optimization/89644 - false-positive -Warray-bounds on strncpy with unterminated array

gcc/ChangeLog:
* tree-ssa-strlen.c (handle_builtin_stxncpy): Use full_string_p
rather than endptr as an indicator of nul-termination.

From-SVN: r269809

PR tree-optimization/89644 - False-positive -Warray-bounds diagnostic on strncpy

gcc/ChangeLog:

PR tree-optimization/89644
* tree-ssa-strlen.c (handle_builtin_stxncpy): Consider unterminated
arrays in determining sequence sizes in strncpy and stpncpy.

gcc/testsuite/ChangeLog:

PR tree-optimization/89644
* gcc.dg/Wstringop-truncation-8.c: New test.

From-SVN: r269807

Fix set of even probabilities (PR middle-end/89737).

2019-03-19  Martin Liska  <mliska@suse.cz>

PR middle-end/89737
* predict.c (combine_predictions_for_bb): Empty likely_edges and
unlikely_edges if there's an edge that belongs to both these sets.
2019-03-19  Martin Liska  <mliska@suse.cz>

PR middle-end/89737
* gcc.dg/pr89737.c: New test.

From-SVN: r269804

rs6000: Unaligned stfiwx on older CPUs (PR89746)

The "classic" PowerPCs (6xx/7xx) are not STRICT_ALIGNMENT, but their
floating point units are.  This is not normally a problem, the ABIs
make everything FP aligned.  The RTL patterns converting FP to integer
however get a potentially unaligned destination, and we do not want to
do an stfiwx on that on such older CPUs.

This fixes it.  It does not change anything for TARGET_MFCRF targets
(POWER4 and later).  It also won't change anything for strict-alignment
targets, or CPUs without hardware FP of course, or CPUs that do not
implement stfiwx (older 4xx/5xx/8xx).

It does not change the corresponding fixuns* pattern, because that can
not be enabled on any CPU that cannot handle unaligned FP well.

PR target/89746
* config/rs6000/rs6000.md (fix_trunc<mode>si2_stfiwx): If we have a
non-TARGET_MFCRF target, and the dest is memory but not 32-bit aligned,
go via a stack temporary.

From-SVN: r269802

re PR target/89378 ([MIPS] FAIL: gcc.dg/vect/pr88598-3.c  -mmsa (internal compiler error))

PR target/89378
* config/mips/mips.c (mips_expand_vec_cond_expr): Use gen_lowpart
instead of gen_rtx_SUBREG.
* config/mips/mips-msa.md (vec_extract<mode><unitmode>): Likewise.

From-SVN: r269801

re PR debug/88389 (-flto -g -gsplit-dwarf is broken)

2019-03-19  Richard Biener  <rguenther@suse.de>

PR debug/88389
* opts.c (finish_options): Disable -gsplit-dwarf when doing LTO.

From-SVN: r269800

re PR lto/87089 (tree check: expected class 'type', have 'declaration' (namespace_decl) in type_with_linkage_p, at ipa-utils.h)

PR lto/87809
PR lto/89335
* tree.c (free_lang_data_in_decl): Do not free context of C++
destrutors.

* g++.dg/lto/pr87089_0.C: New testcase.
* g++.dg/lto/pr87089_1.C: New testcase.
* g++.dg/lto/pr89335_0.C: New testcase.

From-SVN: r269799

libgo: fix build on AIX

    Since aix/ppc64 has been added to GC toolchain, a mix between new and
    old files were created in gcc toolchain.
    This commit corrects this merge for aix/ppc64 and aix/ppc.

    Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/167658

From-SVN: r269797

re PR target/89736 (New test pr87532-mc.c fails on compiler not defaulting to VSX)

gcc/testsuite/ChangeLog:

2019-03-19  Kelvin Nilsen  <kelvin@gcc.gnu.org>

PR target/89736
* gcc.target/powerpc/pr87532-mc.c: Modify dejagnu directives to
restrict this test to vsx targets.

From-SVN: r269796

re PR target/89506 (ICE: in decompose, at rtl.h:2266 with -Og -g)

PR target/89506
* config/arm/arm.md (cmpsi2_addneg): Swap the alternatives and use
subs for the first alternative except when operands[3] is 1.

From-SVN: r269795

Update libstdc++ API Evolution documentation

* doc/xml/manual/allocator.xml: Link to table documenting evolution
of extension allocators.
* doc/xml/manual/evolution.xml: Use angle brackets for header names.
Document new headers in 7.2, 8.1 and 9.1 releases.
* doc/xml/manual/using.xml: Adjust link target for new_allocator.
* doc/html/*: Regenerate.

From-SVN: r269794

re PR target/89752 (ICE in emit_move_insn, at expr.c:3723)

PR target/89752
* gimplify.c (gimplify_asm_expr): For output argument with
TREE_ADDRESSABLE type, clear allows_reg if it allows memory, otherwise
diagnose error.

* g++.dg/ext/asm15.C: Check for particular diagnostic wording.
* g++.dg/ext/asm16.C: Likewise.
* g++.dg/ext/asm17.C: New test.

From-SVN: r269793

re PR rtl-optimization/89753 (ICE in unroll_loop_constant_iterations, at loop-unroll.c:498)

PR rtl-optimization/89753
* loop-unroll.c (decide_unroll_constant_iterations): Make guard for
explicit unrolling factor even more robust.

From-SVN: r269791

re PR target/89726 (Incorrect inlined version of 'ceil' for 32bit)

PR target/89726
* config/i386/i386.c (ix86_expand_floorceildf_32): In ceil
compensation use x2 += 1 instead of x2 -= -1 and when honoring
signed zeros, do another copysign after the compensation.

* gcc.target/i386/fpprec-1.c (x): Add 6 new constants.
(expect_round, expect_rint, expect_floor, expect_ceil, expect_trunc):
Add expected results for them.

From-SVN: r269790

re PR c/89734 (const qualifier on return type not erased inside __typeof__)

PR c/89734
* c-decl.c (grokdeclarator): Call c_build_qualified_type on function
return type even if quals_used is 0.  Formatting fixes.

* gcc.dg/pr89734.c: New test.

From-SVN: r269789

Daily bump.

From-SVN: r269788

PR tree-optimization/89720 - Spurious -Warray-bounds warning on a range with max < min

gcc/ChangeLog:

PR tree-optimization/89720
* tree-vrp.c (vrp_prop::check_mem_ref): Treat range with max < min
more conservatively, the same as anti-range.

gcc/testsuite/ChangeLog:

PR tree-optimization/89720
* gcc.dg/Warray-bounds-42.c: New test.

From-SVN: r269785

mh-mingw: Also set __USE_MINGW_ACCESS flag for C++ code.

From-SVN: r269784

* sv.po: Update.

From-SVN: r269783

Add a test for PR c++/89630

PR c++/89630
* g++.target/i386/pr89630.C: New test.

From-SVN: r269781

libgo: update to Go 1.12.1

    Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/167749

From-SVN: r269780

RISC-V: Fix testsuite regression caused by quotes around options.

2019-03-18  Kito Cheng  <kito.cheng@gmail.com>
* gcc.target/riscv/arch-1.c: Add quotes around march in dg-error.

From-SVN: r269779

PR c++/89630 - ICE with dependent using-decl as template arg.

Even though these two using-declarations have the same effect, they are not
the same declaration, and we don't need to work to treat them as the same
like we do for typedefs.  If we did need to, we would need to handle them
specially in iterative_hash_template_arg as well as here.

* tree.c (cp_tree_equal): Always return false for USING_DECL.

From-SVN: r269777

PR c++/89761 - ICE with sizeof... in pack expansion.

In this testcase we get confused when looking at the sizeof... because the
argument pack for 'args' has been wrapped in an ARGUMENT_PACK_SELECT as part
of expanding the fold-expression.  We handle this situation a bit lower down
in tsubst_pack_expansion, but that doesn't help the call to
argument_pack_element_is_expansion_p, which happens earlier.

* pt.c (argument_pack_element_is_expansion_p): Handle
ARGUMENT_PACK_SELECT.

From-SVN: r269776

PR c++/89640 - GNU attributes on lambda.

My patch for PR 60503 to fix C++11 attribute parsing on lambdas accidentally
removed support for GNU attributes.

* parser.c (cp_parser_lambda_declarator_opt): Allow GNU attributes.

From-SVN: r269775

rs6000: Fix altivec-7.c testcase

It currently wants to see lvx insns on AIX, and no lvx insns on Linux.
What is really wanted is lvx insns when no VSX, and lxv* insns if VSX.
This fixes it.

* gcc.target/powerpc/altivec-7.c: Look for lxv* if generating VSX
instructions, and lvx if not.

From-SVN: r269772

rs6000: Use pointers in bswap testcases

Currently these bswap testcases use global variables, which causes
problems with -m32: the memory access is a D-form access, and when
combine tries to combine that with the bswap it tries a D-form store
with byte reverse.  That instruction does not exist, and since combine
started with only two insns here it will not try splitting this.

This should be improved, but it is not what this test is testing, and
the "load" case already uses a pointer, so let's do that for the store
case as well.

* gcc.target/powerpc/bswap16.c: Use a pointer instead of a global for
the "store" test as well.
* gcc.target/powerpc/bswap32.c: Ditto.

From-SVN: r269771

rs6000: Fix pr18096-1.c test

For the big stack frame in the test GCC used to say

pr18096-1.c:7:6: error: total size of local objects too large

but now it says

pr18096-1.c:7:6: error: total size of local objects 2147483647 exceeds maximum 2147483392

Let's just allow both in the test.

gcc/testsuite/
* gcc.target/powerpc/pr18096-1.c: Allow an error message that says
"exceeds" instead of just one that talks about "too large".

From-SVN: r269770

re PR fortran/68009 (prototype for gfortran_runtime_error with inline matmul)

2019-03-18  Thomas Koenig  <tkoeng@gcc.gnu.org>

PR fortran/68009
* iresolve.c: Include trans.h.
(gfc_resolve_fe_runtine_error): Set backend_decl on
resolved_sym.

From-SVN: r269769

PR c++/89682 - wrong access error in default argument.

Here we were pushing into the right access context, but we were called from
a deferred checking context, so didn't end up doing the checks until after
we left the access context.

* pt.c (tsubst_default_argument): Don't defer access checks.

From-SVN: r269766

re PR tree-optimization/88945 (ICE in fold_convert_loc in FRE when using -fdump-tree-fre-details)

2019-03-18  Richard Biener  <rguenther@suse.de>

PR middle-end/88945
* tree-ssanames.c (release_ssa_name_fn): For released SSA names
use a TREE_TYPE of error_mark_node to avoid ICEs when dumping
basic-blocks that are removed.  Remove restoring SSA_NAME_VAR.
* tree-outof-ssa.c (eliminate_useless_phis): Remove redundant checking.

From-SVN: r269765

Implement circular print buffer.

2019-03-18  Andrew Stubbs  <ams@codesourcery.com>

gcc/
* config/gcn/gcn-run.c (struct output): Make next_output unsigned.
Extend queue to 1024 entries.
Add "consumed" field.
(gomp_print_output): Remove print_index parameter.
Add final parameter.
Change limit to unsigned.
Use consumed field to implement circular buffer.
Detect interrupted print in final pass.
Flush output at the end.
(run): Update gomp_print_output usage.
(main): Initialize kernargs->output_data.consumed.

From-SVN: r269764

Fix a case in which the vector cost model was ignored

This patch fixes a case in which we vectorised something with a
fully-predicated loop even after the cost model had rejected it.
E.g. the loop in the testcase has the costs:

  Vector inside of loop cost: 27
  Vector prologue cost: 0
  Vector epilogue cost: 0
  Scalar iteration cost: 7
  Scalar outside cost: 6
  Vector outside cost: 0
  prologue iterations: 0
  epilogue iterations: 0

and we can see that the loop executes at most three times, but we
decided to vectorise it anyway.

(The costs here are equal for three iterations, but the same thing
happens even when the vector code is strictly more expensive.)

The problem is the handling of "/VF" in:

  /* Calculate number of iterations required to make the vector version
     profitable, relative to the loop bodies only.  The following condition
     must hold true:
     SIC * niters + SOC > VIC * ((niters-PL_ITERS-EP_ITERS)/VF) + VOC
     where
     SIC = scalar iteration cost, VIC = vector iteration cost,
     VOC = vector outside cost, VF = vectorization factor,
     PL_ITERS = prologue iterations, EP_ITERS= epilogue iterations
     SOC = scalar outside cost for run time cost model check.  */

We treat the "/VF" as truncating, but for fully-predicated loops, it's
closer to a ceil division, since fractional iterations are handled by a
full iteration with some predicate bits set to false.

The easiest fix seemed to be to calculate the minimum number of vector
iterations first, then use that to calculate the minimum number of scalar
iterations.

Calculating the minimum number of vector iterations might make sense for
unpredicated loops too, since calculating the scalar niters directly
doesn't take into account the fact that the VIC multiple has to be an
integer.  But the handling of PL_ITERS and EP_ITERS for unpredicated
loops is a bit hand-wavy anyway, so maybe vagueness here cancels out
vagueness there?

Either way, changing this for unpredicated loops would be much too
invasive for stage 4, so the patch keeps it specific to fully-predicated
loops (i.e. SVE) for now.  There's no functional change for other targets.

2019-03-18  Richard Sandiford  <richard.sandiford@arm.com>

gcc/
* tree-vect-loop.c (vect_estimate_min_profitable_iters): Fix the
calculation of the minimum number of scalar iterations for
fully-predicated loops.

gcc/testsuite/
* gcc.target/aarch64/sve/cost_model_1.c: New test.

From-SVN: r269763

Add forgotten requeing in propagate_subaccesses_across_link

2019-03-18  Martin Jambor  <mjambor@suse.cz>

PR tree-optimization/89546
* tree-sra.c (propagate_subaccesses_across_link): Requeue new_acc if
any propagation to its children took place.

testsuite/
* gcc.dg/tree-ssa/pr89546.c: New test.

From-SVN: r269761

gcc/riscv: Correctly ignore empty C++ structs when flattening for ABI

This fixes PR target/89627.

The RISC-V ABI document[1] says:

   For the purposes of this section, "struct" refers to a C struct
   with its hierarchy flattened, including any array fields. That is,
   struct { struct { float f[1]; } g[2]; } and struct { float f; float
   g; } are treated the same. Fields containing empty structs or
   unions are ignored while flattening, even in C++, unless they have
   nontrivial copy constructors or destructors.

However, this flattening only applies when one of the fields of the
flattened structure can be placed into a floating point register,
otherwise no flattening occurs.

Currently GCC fails to correctly consider that empty C++ structures
have a non-zero size when constructing the arguments from a flattened
structure, and as a result, trying to pass a C++ structure like this:

  struct sf { struct {} e; float f; };

Doesn't work correctly, GCC fails to take the offset of 'f' within
'sf' into account and will actually pass the space backing 'e' as the
contents of 'f'.

This patch fixes this so that 'f' will be passed correctly.  A couple
of new tests are added to cover this functionality.

[1] https://github.com/riscv/riscv-elf-psabi-doc/blob/master/riscv-elf.md

gcc/ChangeLog:

PR target/89627
* config/riscv/riscv.c (riscv_pass_fpr_single): Add offset
parameter, and make use of it.
(riscv_get_arg_info): Pass offset to riscv_pass_fpr_single.

gcc/testsuite/ChangeLog:

PR target/89627
* g++.target/riscv/call-with-empty-struct-float.C: New file.
* g++.target/riscv/call-with-empty-struct-int.C: New file.
* g++.target/riscv/call-with-empty-struct.H: New file.
* g++.target/riscv/riscv.exp: New file.

From-SVN: r269760

re PR c++/85014 (internal compiler error: in lookup_base, at cp/search.c:185)

/cp
2019-03-18  Paolo Carlini  <paolo.carlini@oracle.com>

PR c++/85014
* semantics.c (finish_non_static_data_member): Check return value
of context_for_name_lookup and immediately return error_mark_node
if isn't a type.

/testsuite
2019-03-18  Paolo Carlini  <paolo.carlini@oracle.com>

PR c++/85014
* g++.dg/cpp0x/pr85014.C: New.

From-SVN: r269759

[ARC] Enable code density frame option for elf targets.

xxxx-xx-xx  Claudiu Zissulescu  <claziss@synopsys.com>

* config/arc/arc.opt (mcode-density-frame): Get the inital value
from TARGET_CODE_DENSITY_FRAME_DEFAULT.
* config/arc/elf.h (TARGET_CODE_DENSITY_FRAME_DEFAULT): Define.
* config/arc/linux.h (TARGET_CODE_DENSITY_FRAME_DEFAULT): Define.
* config/arc/arc.md (pop_multi_fp_blink): Adjust constraints to
match what the ops is doing.
(push_multi_fp_blink): Likewise.
* config/arc/arc.c (arc_override_options): Enable enter/leave when
compiling for size and elf target.
(arc_save_callee_enter): Adjust note to match what enter/leave
operation does.

From-SVN: r269758

[ARC] Fix tst_movb pattern.

xxxx-xx-xx  Claudiu Zissulescu  <claziss@synopsys.com>

* config/arc/arc.md (tst_movb): Fix constraint.

From-SVN: r269757

[ARC] Define TARGET_HAVE_SPECULATION_SAFE_VALUE.

xxxx-xx-xx  Claudiu Zissulescu  <claziss@synopsys.com>

* config/arc/arc.c (TARGET_HAVE_SPECULATION_SAFE_VALUE): Define.

From-SVN: r269756

[ARC] Introduce ADJUST_REG_ALLOC_ORDER.

The ARC port is changing the allocation order in the
arc_conditional_register_usage function, but this is not the proper
way. Thus, we employ ADJUST_REG_ALLOC_ORDER hook for this task.

gcc/
xxxx-xx-xx  Claudiu Zissulescu  <claziss@synopsys.com>

* config/arc/arc-protos.h (arc_adjust_reg_alloc_order): Declare.
* config/arc/arc.c (arc_conditional_register_usage): Remove all
reg_alloc_order references.
(size_alloc_order): Define.
(arc_adjust_reg_alloc_order): New function.
* config/arc/arc.h (REG_ALLOC_ORDER): Proper define the register
order.
(ADJUST_REG_ALLOC_ORDER): Define.
(HONOR_REG_ALLOC_ORDER): Likewise.

From-SVN: r269755

re PR target/87561 (416.gamess is slower by ~10% starting from r264866 with -Ofast)

2019-03-18  Richard Biener  <rguenther@suse.de>

PR target/87561
* config/i386/i386.c (ix86_add_stmt_cost): Pessimize strided
loads and stores a bit more.

From-SVN: r269754

re PR target/87561 (416.gamess is slower by ~10% starting from r264866 with -Ofast)

2019-03-18  Richard Biener  <rguenther@suse.de>

PR target/87561
* config/i386/i386.c (ix86_add_stmt_cost): Apply strided
load pessimization to stores as well.

From-SVN: r269753

re PR middle-end/86979 (ICE: in maybe_record_trace_start, at dwarf2cfi.c:2348 with -m32 on darwin)

PR middle-end/86979
* gcc.dg/pr86979.c: New test.

From-SVN: r269752

re PR rtl-optimization/89676 (Redundant moves for long long shift on 32bit x86)

        PR middle-end/89676
        * sel-sched.c (compute_av_set_at_bb_end): When we have an ineligible
        successor, use NULL as its av set.

From-SVN: r269751

re PR fortran/88008 (ICE in check_typebound_baseobject, at fortran/resolve.c:6058)

2019-03-17  Thomas Koenig  <tkoenig@gcc.gnu.org>

PR fortran/88008
* gfortran.h (expr_t): Add EXPR_UNKNOWN.
* expr.c (gfc_copy_expr): Add EXPR_UNKNOWN to switch statement.
(gfc_simplify_expr): Likewise.
* module.c (mio_expr): Likewise.
* resovle.c (extract_compcall_passed_object): Issue error on
unknown type.
(check_typebound_baseobject): Issue error on wrong type.
* trans-expr.c (gfc_apply_interface_mapping_to_expr): Add
EXPR_UNKNOWN to switch statement.

2019-03-17  Thomas Koenig  <tkoenig@gcc.gnu.org>

PR fortran/88008
 * gfortran.dg/typebound_call_31.f90: New test.

From-SVN: r269750

Daily bump.

From-SVN: r269749

PR c++/89571 - ICE with ill-formed noexcept on constructor.

Earlier changes to defer instantiating a defaulted noexcept-specifier that
depends on yet-unparsed default member initializers broke this testcase,
where instantiation fails for another reason.  In this case there's no
reason to defer and try again later, so let's not.

* pt.c (maybe_instantiate_noexcept): Only return false if defaulted.
(regenerate_decl_from_template): Use it for noexcept-specs.

From-SVN: r269746

pr83487-1_x.c: Use -fno-common option on hppa*-*-hpux*.

* gcc.dg/compat/pr83487-1_x.c: Use -fno-common option on hppa*-*-hpux*.
* gcc.dg/compat/pr83487-1_y.c: Likewise.

From-SVN: r269745

pointer_init_10.f90: Require visibility support.

* gfortran.dg/pointer_init_10.f90: Require visibility support.
* gfortran.dg/temporary_3.f90: Likewise.

From-SVN: r269744

pr88074.c: Require c99_runtime.

* gcc.dg/pr88074.c: Require c99_runtime.
* gcc.dg/warn-abs-1.c: Likewise.
* gfortran.dg/ISO_Fortran_binding_1.f90: Likewise.

From-SVN: r269743

re PR testsuite/89666 (FAIL: gcc.dg/ipa/ipa-icf-39.c scan-ipa-dump-times icf "Unified;" 2)

PR testsuite/89666
* c-c++-common/builtin-has-attribute-3.c: Define SKIP_ALIAS on
hppa*-*-hpux*.
* gcc.dg/attr-copy.c: Require alias support.
* gcc.dg/ipa/ipa-icf-39.c: Likewise.

From-SVN: r269742

Daily bump.

From-SVN: r269737

re PR fortran/89724 (Fortran diagnostics give wrong line number because of math-vector-fortran.h header file)

PR fortran/89724
* scanner.c (load_line): Remove linenum and current_line static
variables, add warned_tabs automatic variable.  Use current_file->line
instead of current_line and warned_tabs boolean to avoid diagnosing
tabs multiple times on the same line.

* gfortran.dg/continuation_15.f90: New test.
* gfortran.dg/continuation_16.f90: New test.

From-SVN: r269734

Fix type in last change.

From-SVN: r269733

re PR testsuite/89393 (FAIL: g++.dg/abi/ref-temp1.C  -std=c++14  scan-assembler .weak(_definition)?[ \t]_?_ZGR1bIvE)

PR testsuite/89393
* g++.dg/abi/ref-temp1.C: Skip on 32-bit hppa*-*-hpux*.
* g++.dg/cpp0x/pr84497.C: Likewise.

From-SVN: r269732

builtin-has-attribute-4.c: Skip on 32-bit hppa*-*-hpux*.

* c-c++-common/builtin-has-attribute-4.c: Skip on 32-bit hppa*-*-hpux*.
* gcc.dg/attr-copy-6.c: Likewise.
* gcc.dg/pr87793.c: Likewise.

From-SVN: r269731

re PR inline-asm/84941 (internal compiler error: in reg_overlap_mentioned_p, at rtlanal.c:1870 (reg_overlap_mentioned_p()/match_asm_constraints_1()))

PR testsuite/84941
* gcc.dg/pr84941.c: Skip on hppa*-*-*.

From-SVN: r269730

Wattributes-10.c: Add -fno-common option on hppa*-*-hpux*.

* gcc.dg/Wattributes-10.c: Add -fno-common option on hppa*-*-hpux*.
* gcc.dg/gimplefe-34.c: Likewise.
* gcc.dg/compat/pr83487-2_x.c: Use -fno-common option on hppa*-*-hpux*.
* gcc.dg/compat/pr83487-2_y.c: Likewise.

From-SVN: r269729

re PR testsuite/84174 (FAIL: gcc.dg/Wattributes-6.c  (test for warnings, line 404))

PR testsuite/84174
* gcc.dg/Wattributes-6.c: Skip warning check at line 404 on
hppa*64*-*-*.

From-SVN: r269728

re PR testsuite/83453 (FAIL: c-c++-common/Wattributes.c  -std=gnu++98  (test for warnings, line 404))

PR testsuite/83453
* c-c++-common/Wattributes.c: Skip a warning check on hppa*64*-*-*.

From-SVN: r269727

target-supports.exp (check_ascii_locale_available): Remove hppa*-*-hpux*.

* lib/target-supports.exp (check_ascii_locale_available): Remove
hppa*-*-hpux*.

From-SVN: r269726

re PR fortran/84394 (compiler error when using modules with derived types in block data subprograms)

2019-03-16  Thomas Koenig  <tkoenig@gcc.gnu.org>

PR fortran/84394
* symbol.c (gfc_add_subroutine): If we are encountering a
subrtoutine within a BLOCK DATA and the name starts with an
underscore, do not check.

2019-03-16  Thomas Koenig  <tkoenig@gcc.gnu.org>

PR fortran/84394
* gfortran.dg/blockdata_11.f90: New test.

From-SVN: r269721

Daily bump.

From-SVN: r269720

re PR fortran/60091 (Misleading error messages in rank-2 pointer assignment to rank-1 target)

2019-03-15  Harald Anlauf  <anlauf@gmx.de>

PR fortran/60091
* expr.c (gfc_check_pointer_assign): Correct and improve error
messages for invalid pointer assignments.

PR fortran/60091
* gfortran.dg/pointer_remapping_3.f08: Adjust error messages.
* gfortran.dg/pointer_remapping_7.f90: Adjust error message.

From-SVN: r269717

LRA: side_effects_p stmts' output is not invariant (PR89721)

PR89721 shows LRA treating an unspec_volatile's result as invariant,
which of course isn't correct.  This patch fixes it.

PR rtl-optimization/89721
* lra-constraints (invariant_p): Return false if side_effects_p holds.

From-SVN: r269716

re PR target/87532 (bad results from vec_extract(unsigned char, foo) dependent upon function inline)

gcc/ChangeLog:

2019-03-15  Kelvin Nilsen  <kelvin@gcc.gnu.org>

PR target/87532
* config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin):
When handling vec_extract, use modular arithmetic to allow
constant selectors greater than vector length.
* config/rs6000/rs6000.c (rs6000_expand_vector_extract): Allow
V1TImode vectors to have constant selector values greater than 0.
Use modular arithmetic to compute vector index.
(rs6000_split_vec_extract_var): Use modular arithmetic to compute
index for in-memory vectors.  Correct code generation for
in-register vectors.
(altivec_expand_vec_ext_builtin): Use modular arithmetic to
compute index.

gcc/testsuite/ChangeLog:

2019-03-15  Kelvin Nilsen  <kelvin@gcc.gnu.org>

PR target/87532
* gcc.target/powerpc/fold-vec-extract-char.p8.c: Modify expected
instruction selection.
* gcc.target/powerpc/fold-vec-extract-int.p8.c: Likewise.
* gcc.target/powerpc/fold-vec-extract-short.p8.c: Likewise.
* gcc.target/powerpc/pr87532-mc.c: New test.
* gcc.target/powerpc/pr87532.c: New test.
* gcc.target/powerpc/vec-extract-v16qiu-v2.h: New test.
* gcc.target/powerpc/vec-extract-v16qiu-v2a.c: New test.
* gcc.target/powerpc/vec-extract-v16qiu-v2b.c: New test.
* gcc.target/powerpc/vsx-builtin-10a.c: New test.
* gcc.target/powerpc/vsx-builtin-10b.c: New test.
* gcc.target/powerpc/vsx-builtin-11a.c: New test.
* gcc.target/powerpc/vsx-builtin-11b.c: New test.
* gcc.target/powerpc/vsx-builtin-12a.c: New test.
* gcc.target/powerpc/vsx-builtin-12b.c: New test.
* gcc.target/powerpc/vsx-builtin-13a.c: New test.
* gcc.target/powerpc/vsx-builtin-13b.c: New test.
* gcc.target/powerpc/vsx-builtin-14a.c: New test.
* gcc.target/powerpc/vsx-builtin-14b.c: New test.
* gcc.target/powerpc/vsx-builtin-15a.c: New test.
* gcc.target/powerpc/vsx-builtin-15b.c: New test.
* gcc.target/powerpc/vsx-builtin-16a.c: New test.
* gcc.target/powerpc/vsx-builtin-16b.c: New test.
* gcc.target/powerpc/vsx-builtin-17a.c: New test.
* gcc.target/powerpc/vsx-builtin-17b.c: New test.
* gcc.target/powerpc/vsx-builtin-18a.c: New test.
* gcc.target/powerpc/vsx-builtin-18b.c: New test.
* gcc.target/powerpc/vsx-builtin-19a.c: New test.
* gcc.target/powerpc/vsx-builtin-19b.c: New test.
* gcc.target/powerpc/vsx-builtin-20a.c: New test.
* gcc.target/powerpc/vsx-builtin-20b.c: New test.
* gcc.target/powerpc/vsx-builtin-9a.c: New test.
* gcc.target/powerpc/vsx-builtin-9b.c: New test.

From-SVN: r269715