[AArch64] Extend SVE reverse permutes to predicates

This is tested by the main SVE ACLE patches, but since it affects
the evpc routines, it seemed worth splitting out.

2019-10-29  Richard Sandiford  <richard.sandiford@arm.com>

gcc/
* config/aarch64/aarch64-sve.md (@aarch64_sve_rev<PRED_ALL:mode>):
New pattern.
* config/aarch64/aarch64.c (aarch64_evpc_rev_global): Handle all
SVE modes.

From-SVN: r277562

[AArch64] Add FFR and FFRT registers

This patch adds the First Fault Register to the AArch64 port, as well
as a fake register known as the FFR Token or FFRT.  The main ACLE
patch explains what the FFRT does and how it works.

2019-10-29  Richard Sandiford  <richard.sandiford@arm.com>

gcc/
* config/aarch64/aarch64.md (FFR_REGNUM, FFRT_REGNUM): New constants.
* config/aarch64/aarch64.h (FIRST_PSEUDO_REGISTER): Bump to
FFRT_REGNUM + 1.
(FFR_REGS, PR_AND_FFR_REGS): New register classes.
(REG_CLASS_NAMES, REG_CLASS_CONTENTS): Add entries for them.
* config/aarch64/aarch64.c (pr_or_ffr_regnum_p): New function.
(aarch64_hard_regno_nregs): Handle the new register classes.
(aarch64_hard_regno_mode_ok): Likewise.
(aarch64_regno_regclass): Likewise.
(aarch64_class_max_nregs): Likewise.
(aarch64_register_move_cost): Likewise.
(aarch64_conditional_register_usage): Don't treat FFR and FFRT
as general register_operands.

From-SVN: r277561

Fix unsigned type overflow in memory report.

2019-10-29  Martin Liska  <mliska@suse.cz>

* ggc-common.c: One can't subtract unsigned types
in compare function.

From-SVN: r277560

Print header in dump_memory_report.

2019-10-29  Martin Liska  <mliska@suse.cz>

* cgraphunit.c (symbol_table::compile): Pass
title as dump_memory_report argument.
* toplev.c (dump_memory_report):  New argument.
(finalize): Pass new argument.
* toplev.h (dump_memory_report): Add argument.
2019-10-29  Martin Liska  <mliska@suse.cz>

* lto.c (do_whole_program_analysis): Pass
title as dump_memory_report argument.

From-SVN: r277559

Move Leak in GCC memory report to the first column.

2019-10-29  Martin Liska  <mliska@suse.cz>

* ggc-common.c: Move Leak to the first column.

From-SVN: r277558

Remove misleading sorting function in ggc memory report.

2019-10-29  Martin Liska  <mliska@suse.cz>

* cgraphunit.c (symbol_table::compile): Remove argument
for dump_memory_report.
* ggc-common.c (dump_ggc_loc_statistics): Likewise.
(compare_final): Remove in order to make report
better readable.
* ggc.h (dump_ggc_loc_statistics):  Remove argument.
* mem-stats.h (mem_alloc_description::get_list):
Do not pass cmp.
(mem_alloc_description::dump): Likewise here.
* toplev.c (dump_memory_report): Remove final
argument.
(finalize): Likewise.
* toplev.h (dump_memory_report): Remove argument.
2019-10-29  Martin Liska  <mliska@suse.cz>

* lto.c (do_whole_program_analysis): Remove argument.

From-SVN: r277557

[AArch64] Handle scalars in cmp and shift immediate queries

The SVE ACLE has convenience functions that take scalar arguments
instead of vectors.  This patch makes it easier to implement the shift
and compare functions by making the associated immediate queries work
for scalar immediates as well as vector duplicates of them.

The "const" codes in the predicates were a holdover from an early
version of the SVE port in which we used (const ...) wrappers for
variable-length vector constants.  I'll remove other instances
of them in a separate patch.

2019-10-29  Richard Sandiford  <richard.sandiford@arm.com>

gcc/
* config/aarch64/aarch64.c (aarch64_sve_cmp_immediate_p)
(aarch64_simd_shift_imm_p): Accept scalars as well as vectors.
* config/aarch64/predicates.md (aarch64_sve_cmp_vsc_immediate)
(aarch64_sve_cmp_vsd_immediate): Accept "const_int", but don't
accept "const".

From-SVN: r277556

Add a simulate_enum_decl langhook

Similarly to the simulate_builtin_function_decl patch, this one
adds a hook for simulating an enum declaration in the source
language.  Again, the main SVE ACLE patch has tests for various
error conditions.

2019-10-29  Richard Sandiford  <richard.sandiford@arm.com>

gcc/
* coretypes.h (string_int_pair): New typedef.
* langhooks-def.h (LANG_HOOKS_SIMULATE_ENUM_DECL): Define.
(LANG_HOOKS_FOR_TYPES_INITIALIZER): Include it.
* langhooks.h (lang_hooks_for_types::simulate_enum_decl): New hook.

gcc/c/
* c-tree.h (c_simulate_enum_decl): Declare.
* c-decl.c (c_simulate_enum_decl): New function.
* c-objc-common.h (LANG_HOOKS_SIMULATE_ENUM_DECL): Define to the above.

gcc/cp/
* cp-objcp-common.h (cxx_simulate_enum_decl): Declare.
(LANG_HOOKS_SIMULATE_ENUM_DECL): Define to the above.
* decl.c (cxx_simulate_enum_decl): New function.

From-SVN: r277555

Add a simulate_builin_function_decl langhook

Although it's possible to define the SVE intrinsics in a normal header
file, it's much more convenient to define them directly in the compiler.
This also speeds up compilation and gives better error messages.

The idea is therefore for arm_sve.h (the main intrinsics header file)
to have the pragma:

    #pragma GCC aarch64 "arm_sve.h"

telling GCC to define (almost) everything arm_sve.h needs to define.
The target then needs a way of injecting new built-in function
declarations during compilation.

The main hook for defining built-in functions is add_builtin_function.
This is designed for use at start-up, and so has various features that
are correct in that context but not for the pragma above:

  (1) the location is always BUILTINS_LOCATION, whereas for arm_sve.h
      it ought to be the location of the pragma.

  (2) the function is only immediately visible if it's in the implementation
      namespace, whereas the pragma is deliberately injecting functions
      into the general namespace.

  (3) there's no attempt to emulate a normal function declaration in
      C or C++, whereas functions declared by the pragma should be
      checked in the same way as an open-coded declaration would be.
      E.g. we should get an error if there was a previous incompatible
      declaration.

  (4) in C++, the function is treated as extern "C" and so can't be
      overloaded, whereas SVE intrinsics do use function overloading.

This patch therefore adds a hook that targets can use to inject
the equivalent of a source-level function declaration, but bound
to a BUILT_IN_MD function.

The main SVE intrinsic patch has tests to make sure that we report an
error for conflicting definitions that appear either before or after
including arm_sve.h.

2019-10-29  Richard Sandiford  <richard.sandiford@arm.com>

gcc/
* langhooks.h (lang_hooks::simulate_builtin_function_decl): New hook.
(simulate_builtin_function_decl): Declare.
* langhooks-def.h (LANG_HOOKS_SIMULATE_BUILTIN_FUNCTION_DECL): Define.
(LANG_HOOKS_INITIALIZER): Include it.
* langhooks.c (add_builtin_function_common): Rename to...
(build_builtin_function): ...this.  Add a location parameter and use
it instead of BUILTINS_LOCATION.  Remove the hook parameter and return
the decl instead.
(add_builtin_function): Update accordingly, passing the returned
decl to the lang hook.
(add_builtin_function_ext_scope): Likewise
(simulate_builtin_function_decl): New function.

gcc/c/
* c-tree.h (c_simulate_builtin_function_decl): Declare.
* c-decl.c (c_simulate_builtin_function_decl): New function.
* c-objc-common.h (LANG_HOOKS_SIMULATE_BUILTIN_FUNCTION_DECL): Define
to the above.

gcc/cp/
* cp-tree.h (cxx_simulate_builtin_function_decl): Declare.
* decl.c (cxx_simulate_builtin_function_decl): New function.
* cp-objcp-common.h (LANG_HOOKS_SIMULATE_BUILTIN_FUNCTION_DECL): Define
to the above.

From-SVN: r277554

re PR tree-optimization/92241 (ice in vect_mark_pattern_st mts, at tree-vect-patterns.c:5175)

2019-10-29  Richard Biener  <rguenther@suse.de>

PR tree-optimization/92241
* gcc.dg/torture/pr92241-2.c: New testcase.

From-SVN: r277553

install.texi (--enable-offload-targets): Fix up a typo in the example, use actual names of supported offload targets.

* doc/install.texi (--enable-offload-targets): Fix up a typo in the
example, use actual names of supported offload targets.

From-SVN: r277552

re PR target/92258 (ICE: output_operand: invalid %-code)

PR target/92258
* config/i386/sse.md (iptr): Revert 2019-10-27 change.

* gcc.target/i386/pr92258.c: New test.

From-SVN: r277551

Daily bump.

From-SVN: r277550

tree-ssa-strlen.c (get_addr_stridx): Add argument and use it.

gcc/ChangeLog:

* tree-ssa-strlen.c (get_addr_stridx): Add argument and use it.
(handle_store): Pass argument to get_addr_stridx.

gcc/testsuite/ChangeLog:

* gcc.dg/strlenopt-89.c: New test.
* gcc.dg/strlenopt-90.c: New test.
* gcc.dg/Wstringop-overflow-20.c: New test.

From-SVN: r277546

PR tree-optimization/92226 - live nul char store to array eliminated

gcc/testsuite/ChangeLog:

PR tree-optimization/92226
* gcc.dg/strlenopt-88.c: New test.

gcc/ChangeLog:

PR tree-optimization/92226
* tree-ssa-strlen.c (compare_nonzero_chars): Return -1 also when
the offset is in the open range outlined by SI's length.

From-SVN: r277545

PR c/66970 - Add __has_builtin() macro

gcc/ChangeLog:

PR c/66970
* doc/cpp.texi (__has_builtin): Document.
* doc/extend.texi (__builtin_frob_return_addr): Correct spelling.

gcc/c/ChangeLog:

PR c/66970
* c-decl.c (names_builtin_p): Define a new function.

gcc/c-family/ChangeLog:

PR c/66970
* c-common.c (c_common_nodes_and_builtins): Call c_define_builtins
even when only preprocessing.
* c-common.h (names_builtin_p): Declare new function.
* c-lex.c (init_c_lex): Set has_builtin.
(c_common_has_builtin): Define a new function.
* c-ppoutput.c (init_pp_output): Set has_builtin.

gcc/cp/ChangeLog:

PR c/66970
* cp-objcp-common.c (names_builtin_p): Define new function.

gcc/testsuite/ChangeLog:

PR c/66970
* c-c++-common/cpp/has-builtin-2.c: New test.
* c-c++-common/cpp/has-builtin-3.c: New test.
* c-c++-common/cpp/has-builtin.c: New test.

From-SVN: r277544

re PR target/82981 (unnecessary __multi3 call for mips64r6 linux kernel)

PR target/82981
        * config/mips/mips.md (<u>mulditi3): Generate patterns for high
        doubleword and low doubleword result of multiplication on
        MIPS64R6.

        * gcc.target/mips/mips64r6-ti-mult.c: New test.

From-SVN: r277537

cp-demangle.c (d_print_mod): Add a space before printing `complex` and `imaginary`, as opposed to after.

* cp-demangle.c (d_print_mod): Add a space before printing `complex`
and `imaginary`, as opposed to after.
* testsuite/demangle-expected: Adjust test.

From-SVN: r277535

mips.c (DIRECT_BUILTIN_PURE): New macro.

        * config/mips/mips.c (DIRECT_BUILTIN_PURE): New macro. Add a
        pure qualifier to the built-in.
        (MSA_BUILTIN_PURE): New macro. Add a pure qualifier to the MSA
        built-ins.
        (struct mips_builtin_description): Add is_pure flag.
        (mips_init_builtins): Mark built-in as pure if the flag in the
        corresponding mips_builtin_description struct is set.

        * gcc.target/mips/mips-builtins-pure.c: New test.

From-SVN: r277534

mips-msa.md (msa_insert_<msaftm_f>): Add an alternative which covers the floating-point input value.

        * config/mips/mips-msa.md (msa_insert_<msaftm_f>): Add an
        alternative which covers the floating-point input value. Also
        forbid the split of insert.d pattern for floating-point values.

        * gcc.target/mips/msa-insert-split.c: New test.

From-SVN: r277533

gcc/riscv: Add a mechanism to remove some calls to _riscv_save_0

When using the -msave-restore flag we end up with calls to
_riscv_save_0 and _riscv_restore_0.  These functions adjust the stack
and save or restore the return address.  Due to grouping multiple
save/restore stub functions together the save/restore 0 calls actually
save s0, s1, s2, and the return address, but only the return address
actually matters.  Leaf functions don't call the save/restore stubs,
so whenever we do see a call to the save/restore stubs, the store of
the return address is required.

If we look in gcc/config/riscv/riscv.c at the function
riscv_expand_prologue and riscv_expand_epilogue we can see that it
would be reasonably easy to adjust these functions to avoid the calls
to the save/restore stubs for those cases where we are about to call
_riscv_save_0 and _riscv_restore_0, however, the actual code size
saving this would give is debatable, with linker relaxation, the calls
to save/restore are often just 4-bytes, and can sometimes even be
2-bytes, while leaving the stack adjust and return address save inline
is always going to be 4-bytes.

The interesting case is when we call _riscv_save_0 and
_riscv_restore_0, and also have a frame that would (without
save/restore) have resulted in a tail call.  In this case if we could
remove the save/restore calls, and restore the tail call then we would
get a real size saving.

The problem is that the choice of generating a tail call or not is
done during the gimple expand pass, at which point we don't know how
many registers we need to save (or restore).

The solution presented in this patch offers a partial solution to this
problem.  By using the TARGET_MACHINE_DEPENDENT_REORG pass to
implement a very limited pattern matching we identify functions that
call _riscv_save_0 and _riscv_restore_0, and which could be converted
to make use of a tail call.  These functions are then converted to the
non save/restore tail call form.

This should result in a code size reduction when compiling with -Os
and with the -msave-restore flag.

gcc/ChangeLog:

        * config.gcc: Add riscv-sr.o to extra_objs for riscv.
        * config/riscv/riscv-sr.c: New file.
        * config/riscv/riscv.c (riscv_reorg): New function.
        (TARGET_MACHINE_DEPENDENT_REORG): Define.
        * config/riscv/riscv.h (SIBCALL_REG_P): Define.
        (riscv_remove_unneeded_save_restore_calls): Declare.
        * config/riscv/t-riscv (riscv-sr.o): New build rule.

gcc/testsuite/ChangeLog:

        * gcc.target/riscv/save-restore-2.c: New file.
        * gcc.target/riscv/save-restore-3.c: New file.
        * gcc.target/riscv/save-restore-4.c: New file.
        * gcc.target/riscv/save-restore-5.c: New file.
        * gcc.target/riscv/save-restore-6.c: New file.
        * gcc.target/riscv/save-restore-7.c: New file.
        * gcc.target/riscv/save-restore-8.c: New file.

From-SVN: r277527

re PR tree-optimization/92163 (ICE: Segmentation fault (in bitmap_set_bit))

2019-10-28  Prathamesh Kulkarni  <prathamesh.kulkarni@linaro.org>

PR tree-optimization/92163
* tree-ssa-dse.c (delete_dead_or_redundant_assignment): New param
need_eh_cleanup with default value NULL. Gate on need_eh_cleanup
before calling bitmap_set_bit.
(dse_optimize_redundant_stores): Pass global need_eh_cleanup to
delete_dead_or_redundant_assignment.
(dse_dom_walker::dse_optimize_stmt): Likewise.
* tree-ssa-dse.h (delete_dead_or_redundant_assignment): Adjust prototype.

testsuite/
* gcc.dg/tree-ssa/pr92163.c: New test.

From-SVN: r277525

re PR middle-end/91272 ([SVE] Use fully-masked loops for CLASTB reductions)

2019-10-28  Prathamesh Kulkarni  <prathamesh.kulkarni@linaro.org>

PR middle-end/91272
* tree-vect-stmts.c (vectorizable_condition): Support
EXTRACT_LAST_REDUCTION with fully-masked loops.

testsuite/
* gcc.target/aarch64/sve/clastb_1.c: Add dg-scan.
* gcc.target/aarch64/sve/clastb_2.c: Likewise.
* gcc.target/aarch64/sve/clastb_3.c: Likewise.
* gcc.target/aarch64/sve/clastb_4.c: Likewise.
* gcc.target/aarch64/sve/clastb_5.c: Likewise.
* gcc.target/aarch64/sve/clastb_6.c: Likewise.
* gcc.target/aarch64/sve/clastb_7.c: Likewise.
* gcc.target/aarch64/sve/clastb_8.c: Likewise.

From-SVN: r277524

re PR tree-optimization/92252 (ICE: Segmentation fault (in vect_stmt_to_vectorize))

2019-10-28  Richard Biener  <rguenther@suse.de>

PR tree-optimization/92252
* tree-vect-slp.c (vect_get_and_check_slp_defs): Adjust
STMT_VINFO_REDUC_IDX when swapping operands.

* gcc.dg/torture/pr92252.c: New testcase.

From-SVN: r277517

re PR tree-optimization/92241 (ice in vect_mark_pattern_st mts, at tree-vect-patterns.c:5175)

2019-10-28  Richard Biener  <rguenther@suse.de>

PR tree-optimization/92241
* tree-vect-loop.c (vect_fixup_scalar_cycles_with_patterns): When
we failed to update the reduction index do not use the pattern
stmts for the reduction chain.
(vectorizable_reduction): When the reduction chain is corrupt,
fail.
* tree-vect-patterns.c (vect_mark_pattern_stmts): Stop when we
fail to update the reduction chain.

* gcc.dg/torture/pr92241.c: New testcase.

From-SVN: r277516

[C++ PATCH] simplify deferred parsing lexer

https://gcc.gnu.org/ml/gcc-patches/2019-10/msg01962.html

We use an eof_token global variable as a sentinel on a deferred parse
(such as in-class function definitions, or default args). This
complicates retrieving the next token in certain places.

As such deferred parses always nest properly and completely before
resuming the outer lexer, we can simply morph the token after the
deferred buffer into a CPP_EOF token and restore it afterwards. I
finally got around to implementing it with this patch.

One complication is that we have to change the discriminator for when
the token's value is a tree. We can't look at the token's type because
it might have been overwritten. I add a bool flag to the token
(there's several spare bits), and use that. This does simplify the
discriminator because we just check a single bit, rather than a set of
token types.

* parser.h (struct cp_token): Drop {ENUM,BOOL}_BITFIELD C-ism.
Add tree_check_p flag, use as nested union discriminator.
(struct cp_lexer): Add saved_type & saved_keyword fields.
* parser.c (eof_token): Delete.
(cp_lexer_new_main): Always init last_token to last token of
buffer.
(cp_lexer_new_from_tokens): Overlay EOF token at end of range.
(cp_lexer_destroy): Restore token under the EOF.
(cp_lexer_previous_token_position): No check for eof_token here.
(cp_lexer_get_preprocessor_token): Clear tree_check_p.
(cp_lexer_peek_nth_token): Check CPP_EOF not eof_token.
(cp_lexer_consume_token): Assert not CPP_EOF, no check for
eof_token.
(cp_lexer_purge_token): Likewise.
(cp_lexer_purge_tokens_after): No check for EOF token.
(cp_parser_nested_name_specifier, cp_parser_decltype)
(cp_parser_template_id): Set tree_check_p.

From-SVN: r277514

tree-vect-loop.c (vect_create_epilog_for_reduction): Use STMT_VINFO_REDUC_IDX from the actual stmt.

2019-10-28  Richard Biener  <rguenther@suse.de>

* tree-vect-loop.c (vect_create_epilog_for_reduction): Use
STMT_VINFO_REDUC_IDX from the actual stmt.
(vect_transform_reduction): Likewise.
(vectorizable_reduction): Compute the reduction chain length,
do not recompute the reduction operand index.  Remove no longer
necessary restriction for condition reduction chains.

From-SVN: r277513

re PR c/92249 (ICE in c_parser_gimple_compound_statement w/ GIMPLE testcases)

2019-10-28  Richard Biener  <rguenther@suse.de>

PR c/92249
* gimple-parser.c (c_parser_parse_gimple_body): Make
current_bb the entry block initially to easier recover
from errors.
(c_parser_gimple_compound_statement): Adjust.

From-SVN: r277512

re PR target/92225 (ice in gen_smaxv2di3, at config/i386/sse.md:12225)

PR target/92225
* config/i386/sse.md (REDUC_SSE_SMINMAX_MODE): Use TARGET_SSE4_2
condition for V2DImode.

testsuite/ChangeLog:

PR target/92225
* gcc.target/i386/pr92225.c: New test.

From-SVN: r277510

sse.md (sse_cvtss2si<rex64namesuffix>_2): Remove %k operand modifier.

* config/i386/sse.md (sse_cvtss2si<rex64namesuffix>_2):
Remove %k operand modifier.
(*vec_extractv2df_1_sse): Remove %q operand modifier.

From-SVN: r277509

Fix unroll-and-jam.c on 32bit

where LIM interacts with foo10.  On 64bit LIM doesn't do the problematic
change for whatever reason, but it seems better to disable LIM
alltogether, which requires a minor change in the testcase.

From-SVN: r277508

Move jump threading before reload

r266734 has introduced a new instance of jump threading pass in order to
take advantage of opportunities that combine opens up.  It was perceived
back then that it was beneficial to delay it after reload, since that
might produce even more such opportunities.

Unfortunately jump threading interferes with hot/cold partitioning.  In
the code from PR92007, it converts the following

  +-------------------------- 2/HOT ------------------------+
  |                                                         |
  v                                                         v
3/HOT --> 5/HOT --> 8/HOT --> 11/COLD --> 6/HOT --EH--> 16/HOT
            |                               ^
            |                               |
            +-------------------------------+

into the following:

  +---------------------- 2/HOT ------------------+
  |                                               |
  v                                               v
3/HOT --> 8/HOT --> 11/COLD --> 6/COLD --EH--> 16/HOT

This makes hot bb 6 dominated by cold bb 11, and because of this
fixup_partitions makes bb 6 cold as well, which in turn makes EH edge
6->16 a crossing one.  Not only can't we have crossing EH edges, we are
also not allowed to introduce new crossing edges after reload in
general, since it might require extra registers on some targets.

Therefore, move the jump threading pass between combine and hot/cold
partitioning.  Building SPEC 2006 and SPEC 2017 with the old and the new
code indicates that:

* When doing jump threading right after reload, 3889 edges are threaded.
* When doing jump threading right after combine, 3918 edges are
  threaded.

This means this change will not introduce performance regressions.

gcc/ChangeLog:

2019-10-28  Ilya Leoshkevich  <iii@linux.ibm.com>

PR rtl-optimization/92007
* cfgcleanup.c (thread_jump): Add an assertion that we don't
call it after reload if hot/cold partitioning has been done.
(class pass_postreload_jump): Rename to
pass_jump_after_combine.
(make_pass_postreload_jump): Rename to
make_pass_jump_after_combine.
* passes.def(pass_postreload_jump): Move before reload, rename
to pass_jump_after_combine.
* tree-pass.h (make_pass_postreload_jump): Rename to
make_pass_jump_after_combine.

gcc/testsuite/ChangeLog:

2019-10-28  Ilya Leoshkevich  <iii@linux.ibm.com>

PR rtl-optimization/92007
* g++.dg/opt/pr92007.C: New test (from Arseny Solokha).

From-SVN: r277507

re PR ipa/92242 (LTO ICE in ipa_get_cs_argument_count ipa-prop.h:598)

PR ipa/92242
* ipa-fnsummary.c (ipa_merge_fn_summary_after_inlining): Check
for missing EDGE_REF
* ipa-prop.c (update_jump_functions_after_inlining): Likewise.

From-SVN: r277504

Fortran] OpenACC – libgomp/testsuite – use 'stop' and 'dg-do run'

        * testsuite/libgomp.oacc-fortran/abort-1.f90: Add 'dg-do run'.
        * testsuite/libgomp.oacc-fortran/abort-2.f90: Ditto.
        * testsuite/libgomp.oacc-fortran/acc_on_device-1-1.f90: Ditto.
        * testsuite/libgomp.oacc-fortran/acc_on_device-1-2.f90: Ditto.
        * testsuite/libgomp.oacc-fortran/acc_on_device-1-3.f90: Ditto.
        * testsuite/libgomp.oacc-fortran/lib-1.f90: Ditto.
        * testsuite/libgomp.oacc-fortran/common-block-1.f90:
        Use 'stop' not abort().
        * testsuite/libgomp.oacc-fortran/common-block-2.f90: Ditto.
        * testsuite/libgomp.oacc-fortran/common-block-3.f90: Ditto.
        * testsuite/libgomp.oacc-fortran/data-1.f90: Ditto.
        * testsuite/libgomp.oacc-fortran/data-2.f90: Ditto.
        * testsuite/libgomp.oacc-fortran/data-5.f90: Ditto.
        * testsuite/libgomp.oacc-fortran/dummy-array.f90: Ditto.
        * testsuite/libgomp.oacc-fortran/gemm-2.f90: Ditto.
        * testsuite/libgomp.oacc-fortran/gemm.f90: Ditto.
        * testsuite/libgomp.oacc-fortran/host_data-2.f90: Ditto.
        * testsuite/libgomp.oacc-fortran/host_data-3.f90: Ditto.
        * testsuite/libgomp.oacc-fortran/host_data-4.f90: Ditto.
        * testsuite/libgomp.oacc-fortran/kernels-collapse-3.f90: Ditto.
        * testsuite/libgomp.oacc-fortran/kernels-collapse-4.f90: Ditto.
        * testsuite/libgomp.oacc-fortran/kernels-independent.f90: Ditto.
        * testsuite/libgomp.oacc-fortran/kernels-loop-1.f90: Ditto.
        * testsuite/libgomp.oacc-fortran/kernels-map-1.f90: Ditto.
        * testsuite/libgomp.oacc-fortran/kernels-parallel-loop-data-enter-exit.f95:
        Ditto.
        * testsuite/libgomp.oacc-fortran/kernels-private-vars-loop-gang-1.f90:
        Ditto.
        * testsuite/libgomp.oacc-fortran/kernels-private-vars-loop-gang-2.f90:
        Ditto.
        * testsuite/libgomp.oacc-fortran/kernels-private-vars-loop-gang-3.f90:
        Ditto.
        * testsuite/libgomp.oacc-fortran/kernels-private-vars-loop-gang-6.f90:
        Ditto.
        * testsuite/libgomp.oacc-fortran/kernels-private-vars-vector-1.f90:
        Ditto.
        * testsuite/libgomp.oacc-fortran/kernels-private-vars-vector-2.f90:
        Ditto.
        * testsuite/libgomp.oacc-fortran/kernels-private-vars-worker-1.f90:
        Ditto.
        * testsuite/libgomp.oacc-fortran/kernels-private-vars-worker-2.f90:
        Ditto.
        * testsuite/libgomp.oacc-fortran/kernels-private-vars-worker-3.f90:
        Ditto.
        * testsuite/libgomp.oacc-fortran/kernels-private-vars-worker-4.f90:
        Ditto.
        * testsuite/libgomp.oacc-fortran/kernels-private-vars-worker-5.f90:
        Ditto.
        * testsuite/libgomp.oacc-fortran/kernels-private-vars-worker-6.f90:
        Ditto.
        * testsuite/libgomp.oacc-fortran/kernels-private-vars-worker-7.f90:
        Ditto.
        * testsuite/libgomp.oacc-fortran/kernels-reduction-1.f90: Ditto.
        * testsuite/libgomp.oacc-fortran/lib-12.f90: Ditto.
        * testsuite/libgomp.oacc-fortran/lib-13.f90: Ditto.
        * testsuite/libgomp.oacc-fortran/lib-14.f90: Ditto.
        * testsuite/libgomp.oacc-fortran/kernels-acc-loop-reduction-2.f90:
        Likewise and also add 'dg-do run'.
        * testsuite/libgomp.oacc-fortran/kernels-acc-loop-reduction.f90:
        Ditto.

From-SVN: r277503

Fortran] PR91863 - fix call to bind(C) with array descriptor

        PR fortran/91863
        * trans-expr.c (gfc_conv_gfc_desc_to_cfi_desc): Don't free data
        memory as that's done on the Fortran side.
        (gfc_conv_procedure_call): Handle void* pointers from
        gfc_conv_gfc_desc_to_cfi_desc.

        PR fortran/91863
        * gfortran.dg/bind-c-intent-out.f90: New.

From-SVN: r277502

rs6000: Enable limited unrolling at -O2

In PR88760, there are a few disscussion about improve or tune unroller for
targets. And we would agree to enable unroller for small loops at O2 first.
And we could see performance improvement(~10%) for below code:
```
  subroutine foo (i, i1, block)
    integer :: i, i1
    integer :: block(9, 9, 9)
    block(i:9,1,i1) = block(i:9,1,i1) - 10
  end subroutine foo

```
This kind of code occurs a few times in exchange2 benchmark.

Similar C code:
```
  for (i = 0; i < n; i++)
    arr[i] = arr[i] - 10;
```

On powerpcle, for O2 , enable -funroll-loops and limit
PARAM_MAX_UNROLL_TIMES=2 and PARAM_MAX_UNROLLED_INSNS=20, we can see >2%
overall improvement for SPEC2017.

This patch is only for rs6000 in which we see visible performance improvement.

gcc/
2019-10-25  Jiufu Guo  <guojiufu@linux.ibm.com>

PR tree-optimization/88760
* config/rs6000/rs6000-common.c (rs6000_option_optimization_table):
Enable -funroll-loops for -O2 and above.
* config/rs6000/rs6000.c (rs6000_option_override_internal): Set
PARAM_MAX_UNROLL_TIMES to 2 and PARAM_MAX_UNROLLED_INSNS to 20, and
do not turn on web and rngreg implicitly, if the unroller is not
explicitly enabled.

gcc.testsuite/
2019-10-25  Jiufu Guo  <guojiufu@linux.ibm.com>

PR tree-optimization/88760
* gcc.target/powerpc/small-loop-unroll.c: New test.
* c-c++-common/tsan/thread_leak2.c: Update test.
* gcc.dg/pr59643.c: Update test.
* gcc.target/powerpc/loop_align.c: Update test.
* gcc.target/powerpc/ppc-fma-1.c: Update test.
* gcc.target/powerpc/ppc-fma-2.c: Update test.
* gcc.target/powerpc/ppc-fma-3.c: Update test.
* gcc.target/powerpc/ppc-fma-4.c: Update test.
* gcc.target/powerpc/pr78604.c: Update test.

From-SVN: r277501

Daily bump.

From-SVN: r277499

* locales.c (iso_3166): Add missing comma after "United-States".

From-SVN: r277492

fprintf-2.c: Silence a Free/NetBSD libc warning.

2019-10-27  Andreas Tobler  <andreast@gcc.gnu.org>

* gcc.c-torture/execute/fprintf-2.c: Silence a Free/NetBSD libc warning.
* gcc.c-torture/execute/printf-2.c: Likewise.
* gcc.c-torture/execute/user-printf.c: Likewise.

From-SVN: r277491

re PR fortran/86248 (LEN_TRIM in specification expression causes link failure)

2019-10-27  Paul Thomas  <pault@gcc.gnu.org>

PR fortran/86248
* resolve.c (flag_fn_result_spec): Correct a typo before the
function declaration.
* trans-decl.c (gfc_sym_identifier): Boost the length of 'name'
to allow for all variants. Simplify the code by using a pointer
to the symbol's proc_name and taking the return out of each of
the conditional branches. Allow symbols with fn_result_spec set
that do not come from a procedure namespace and have a module
name to go through the non-fn_result_spec branch.

2019-10-27  Paul Thomas  <pault@gcc.gnu.org>

PR fortran/86248
* gfortran.dg/char_result_19.f90 : New test.
* gfortran.dg/char_result_mod_19.f90 : Module for the new test.

From-SVN: r277487

ipa-prop.c (ipa_propagate_indirect_call_infos): Do not remove jump functions.

* ipa-prop.c (ipa_propagate_indirect_call_infos): Do not remove
jump functions.

From-SVN: r277486

fix cgraph comment

This comment cut&pasto fix was split out of another patch I'm about to
contribute, as the current version of the patch no longer touches cgraph
data structures.

for  gcc/ChangeLog

* cgraph.c (cgraph_node::rtl_info): Fix cut&pasto in comment.
* cgraph.h (cgraph_node::rtl_info): Likewise.

From-SVN: r277485

ipa-cp.c (propagate_constants_across_call): If args are not available just drop everything to varying.

* ipa-cp.c (propagate_constants_across_call): If args are not available
just drop everything to varying.
(find_aggregate_values_for_callers_subset): Watch for missing
edge summary.
(find_more_scalar_values_for_callers_subs): Likewise.
* ipa-prop.c (ipa_compute_jump_functions_for_edge,
update_jump_functions_after_inlining, propagate_controlled_uses):
Watch for missing summaries.
(ipa_propagate_indirect_call_infos): Remove summary after propagation
is finished.
(ipa_write_node_info): Watch for missing summaries.
(ipa_read_edge_info): Create new ref.
(ipa_edge_args_sum_t): Add remove.
(IPA_EDGE_REF_GET_CREATE): New macro.
* ipa-fnsummary.c (evaluate_properties_for_edge): Watch for missing
edge summary.
(remap_edge_change_prob): Likewise.

From-SVN: r277484

ipa-inline-transform.c (inline_call): update function summaries after expanidng thunk.

* ipa-inline-transform.c (inline_call): update function summaries
after expanidng thunk.

From-SVN: r277483

ipa-icf.c (sem_function::merge): Update function summaries.

* ipa-icf.c (sem_function::merge): Update function summaries.
* ipa-prop.h (ipa_get_param): Do not sanity check for WPA.

From-SVN: r277482

Remove redudant <iptr> when operand already has scalar mode.

gcc/
* config/i386/sse.md (*<sse>_vm<plusminus_insn><mode>3,
<sse>_vm<multdiv_mnemonic><mode>3): Remove <iptr> since
operand is already scalar mode.
(iptr): Remove SF/DF.

From-SVN: r277481

Daily bump.

From-SVN: r277480

codecvt.xml: Switch pubs.opengroup.org to https.

* doc/xml/manual/codecvt.xml: Switch pubs.opengroup.org to https.
* doc/xml/manual/locale.xml (LC_ALL): Ditto.
* doc/xml/manual/messages.xml: Ditto.

From-SVN: r277476

baseline_symbols.txt: Update.

* config/abi/post/hppa-linux-gnu/baseline_symbols.txt: Update.

From-SVN: r277475

rs6000: Fix allocate_stack in a corner case (PR91289)

When we have -fstack-limit-symbol with sysv we can end up with a non-
existing instruction (you cannot add an immediate to register 0).  Fix
this by using register 11 instead.  It might be used for something else
already though, so save and restore its value around this.  In
optimizing compiles these extra moves are usually removed again: the
restore by cprop_hardreg, and then the save by rtl_dce.

PR target/91289
* config/rs6000/rs6000-logue.c (rs6000_emit_allocate_stack): Don't add
an immediate to r0; use r11 instead.  Save and restore r11 to r0 around
this.

From-SVN: r277472

Adjust predicates and constraints of scalar insns.

Changelog

gcc/
* config/i386/sse.md
(<sse>_vm<plusminus_insn><mode>3<mask_scalar_name><round_scalar_name>,
<sse>_vm<multdiv_mnemonic><mode>3<mask_scalar_name><round_scalar_name>,
<sse>_vmsqrt<mode>2<mask_scalar_name><round_scalar_name>,
<sse>_vm<code><mode>3<mask_scalar_name><round_saeonly_scalar_name>,
<sse>_vmmaskcmp<mode>3):
Change predicates from vector_operand to nonimmediate_operand,
constraints xBm to xm, since scalar operations don't need
memory address alignment.
(avx512f_vmcmp<mode>3<round_saeonly_name>,
avx512f_vmcmp<mode>3_mask<round_saeonly_name>): Replace
round_saeonly_nimm_predicate with
round_saeonly_nimm_scalar_predicate.
(fmai_vmfmadd_<mode><round_name>, fmai_vmfmsub_<mode><round_name>,
fmai_vmfnmadd_<mode><round_name>,fmai_vmfnmsub_<mode><round_name>,
*fmai_fmadd_<mode>, *fmai_fmsub_<mode>,
*fmai_fnmadd_<mode><round_name>, *fmai_fnmsub_<mode><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_vmfmsub_<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_vmfnmsub_<mode>_maskz_1<round_name>,
cvtusi2<ssescalarmodesuffix>32<round_name>,
cvtusi2<ssescalarmodesuffix>64<round_name>, ): Replace
round_nimm_predicate with round_nimm_scalr_predicate.
(avx512f_sfixupimm<mode><sd_maskz_name><round_saeonly_name>,
avx512f_sfixupimm<mode>_mask<round_saeonly_name>,
avx512er_vmrcp28<mode><round_saeonly_name>,
avx512er_vmrsqrt28<mode><round_saeonly_name>,
): Replace round_saeonly_nimm_predicate with
round_saeonly_nimm_scalar_predicate.
(avx512dq_vmfpclass<mode><mask_scalar_merge_name>): Replace
vector_operand with nonimmediate_operand.
* config/i386/subst.md (round_scalar_nimm_predicate,
round_saeonly_scalar_nimm_predicate): Replace
vector_operand with nonimmediate_operand.

From-SVN: r277470

Fix false dependence of scalar operation vrcp/vsqrt/vrsqrt/vrndscale
For instructions with xmm operand:

op %xmmN,%xmmQ,%xmmQ ----> op %xmmN, %xmmN, %xmmQ

for instruction with mem operand or gpr operand:

op mem/gpr, %xmmQ, %xmmQ

--->  using pass rpad ---->

xorps %xmmN, %xmmN, %xxN
op mem/gpr, %xmmN, %xmmQ

Performance influence of SPEC2017 fprate which is tested on SKX
----
503.bwaves_r -0.03%
507.cactuBSSN_r -0.22%
508.namd_r -0.02%
510.parest_r 0.37%
511.povray_r 0.74%
519.lbm_r 0.24%
521.wrf_r 2.35%
526.blender_r 0.71%
527.cam4_r 0.65%
538.imagick_r 0.95%
544.nab_r -0.37
549.fotonik3d_r 0.24%
554.roms_r 0.90%
fprate geomean 0.50%
-----

Changelog
gcc/
* config/i386/i386.md (*rcpsf2_sse): Add
avx_partial_xmm_update, prefer m constraint for TARGET_AVX.
(*rsqrtsf2_sse): Ditto.
(*sqrt<mode>2_sse): Ditto.
(sse4_1_round<mode>2): separate constraint vm, add
avx_partail_xmm_update, prefer m constraint for TARGET_AVX.
* config/i386/sse.md (*sse_vmrcpv4sf2"): New define_insn used
by pass rpad.
(*<sse>_vmsqrt<mode>2<mask_scalar_name><round_scalar_name>*):
Ditto.
(*sse_vmrsqrtv4sf2): Ditto.
(*avx512f_rndscale<mode><round_saeonly_name>): Ditto.
(*sse4_1_round<ssescalarmodesuffix>): Ditto.
(sse4_1_round<ssescalarmodesuffix>): Add m constraint and
<iptr> pointer size modifier since vround support memory operand.

gcc/testsuite
* gcc.target/i386/pr87007-4.c: New test.
* gcc.target/i386/pr87007-5.c: Ditto.

From-SVN: r277469

Daily bump.

From-SVN: r277468

PR c++/91581 - ICE in exception-specification of defaulted ctor.

* g++.dg/cpp0x/noexcept55.C: New test.

From-SVN: r277462

Use implicitly-defined copy operations for test iterators

All of these special member functions do exactly what the compiler would
do anyway. By defining them as defaulted for C++11 and later we prevent
move constructors and move assignment operators being defined (which is
consistent with the previous semantics).

Also move default init of the input_iterator_wrapper members from the
derived constructor to the protected base constructor.

* testsuite/util/testsuite_iterators.h (output_iterator_wrapper)
(input_iterator_wrapper, forward_iterator_wrapper)
bidirectional_iterator_wrapper, random_access_iterator_wrapper): Remove
user-provided copy constructors and copy assignment operators so they
are defined implicitly.
(input_iterator_wrapper): Initialize members in default constructor.
(forward_iterator_wrapper): Remove assignments to members of base.

From-SVN: r277459

Fix compilation with Clang

The new constexpr destructor on std::allocator breaks compilation with
Clang in C++2a mode. This only makes it constexpr if the compiler
supports the P0784R7 features.

* include/bits/allocator.h: Check __cpp_constexpr_dynamic_alloc
before making the std::allocator destructor constexpr.
* testsuite/20_util/allocator/requirements/constexpr.cc: New test.

From-SVN: r277458

re PR target/85969 (avr/gen-avr-mmcu-specs.c:56: unused function ?)

PR target/85969
* config/avr/gen-avr-mmcu-specs.c (str_prefix_p): Remove unused
static function.

From-SVN: r277455

[Fortran] OpenACC – permit common blocks in some clauses

2019-10-25  Cesar Philippidis <cesar@codesourcery.com>
    Tobias Burnus  <tobias@codesourcery.com>

gcc/fortran/
* openmp.c (gfc_match_omp_map_clause): Add and pass allow_commons
argument.
(gfc_match_omp_clauses): Update calls to permit common blocks for
OpenACC's copy/copyin/copyout, create/delete, host,
pcopy/pcopy_in/pcopy_out, present_or_copy, present_or_copy_in,
present_or_copy_out, present_or_create and self.

gcc/
* gimplify.c (oacc_default_clause): Privatize fortran common blocks.
(omp_notice_variable): Defer the expansion of DECL_VALUE_EXPR for
common block decls.

gcc/testsuite/
* gfortran.dg/goacc/common-block-1.f90: New test.
* gfortran.dg/goacc/common-block-2.f90: New test.
* gfortran.dg/goacc/common-block-3.f90: New test.

libgomp/
* testsuite/libgomp.oacc-fortran/common-block-1.f90: New test.
* testsuite/libgomp.oacc-fortran/common-block-2.f90: New test.
* testsuite/libgomp.oacc-fortran/common-block-3.f90: New test.

Reviewed-by: Thomas Schwinge <thomas@codesourcery.com>
Co-Authored-By: Tobias Burnus <tobias@codesourcery.com>
From-SVN: r277451

pr70100.c: Add -mvsx.

        * gcc.target/powerpc/pr70100.c: Add -mvsx.
        Allow AIX ABI function name.

From-SVN: r277450

Guard use of concepts with feature test macro

This fixes a regression when using Clang.

* include/bits/range_cmp.h: Check __cpp_lib_concepts before defining
concepts. Fix comment.

From-SVN: r277449

re PR tree-optimization/92222 (ice in useless_type_conversion_p, at gimple-expr.c:86)

2019-10-25  Richard Biener  <rguenther@suse.de>

PR tree-optimization/92222
* tree-vect-slp.c (_slp_oprnd_info::first_pattern): Remove.
(_slp_oprnd_info::second_pattern): Likewise.
(_slp_oprnd_info::any_pattern): New.
(vect_create_oprnd_info): Adjust.
(vect_get_and_check_slp_defs): Compute whether any stmt is
in a pattern.
(vect_build_slp_tree_2): Avoid building up a node from scalars
if any of the operand defs, not just the first, is in a pattern.

* gcc.dg/torture/pr92222.c: New testcase.

From-SVN: r277448

tree-vect-slp.c (vect_get_and_check_slp_defs): Only fail swapping if we actually have to modify the IL on a shared stmt.

2019-10-25  Richard Biener  <rguenther@suse.de>

* tree-vect-slp.c (vect_get_and_check_slp_defs): Only fail
swapping if we actually have to modify the IL on a shared stmt.
(vect_build_slp_tree_2): Never fail swapping on shared stmts
because we no longer modify the IL.

From-SVN: r277446

Fix failure in gcc.target/sve/reduc_strict_3.c

Unwanted unrolling meant that we had more single-precision FADDAs
than expected.

2019-10-25  Richard Sandiford  <richard.sandiford@arm.com>

gcc/testsuite/
* gcc.target/aarch64/sve/reduc_strict_3.c (double_reduc1): Prevent
the loop from being unrolled.

From-SVN: r277442

Update SVE tests for recent XPASSes

Recent target-independent patches mean that several SVE tests
now produce the code that we'd originally wanted them to produce.
Really nice to see :-)

This patch therefore updates the expected baseline, so that hopefully
we don't regress from this point in future.

2019-10-25  Richard Sandiford  <richard.sandiford@arm.com>

gcc/testsuite/
* gcc.target/aarch64/sve/loop_add_5.c: Remove XFAILs for tests
that now pass.
* gcc.target/aarch64/sve/reduc_1.c: Likewise.
* gcc.target/aarch64/sve/reduc_2.c: Likewise.
* gcc.target/aarch64/sve/reduc_5.c: Likewise.
* gcc.target/aarch64/sve/reduc_8.c: Likewise.
* gcc.target/aarch64/sve/slp_13.c: Likewise.
* gcc.target/aarch64/sve/slp_5.c: Likewise.  Update expected
WHILELO counts.
* gcc.target/aarch64/sve/slp_7.c: Likewise.

From-SVN: r277441

Fix typo in dump_tree_statistics.

2019-10-25  Martin Liska  <mliska@suse.cz>

* tree.c (dump_tree_statistics): Use sorted index 'j' and not 'i'.

From-SVN: r277440

Fix reductions for fully-masked loops

Now that vectorizable_operation vectorises most loop stmts involved
in a reduction, it needs to be aware of reductions in fully-masked loops.
The LOOP_VINFO_CAN_FULLY_MASK_P parts of vectorizable_reduction now only
apply to cases that use vect_transform_reduction.

This new way of doing things is definitely an improvement for SVE though,
since it means we can lift the old restriction of not using fully-masked
loops for reduction chains.

2019-10-25  Richard Sandiford  <richard.sandiford@arm.com>

gcc/
* tree-vect-loop.c (vectorizable_reduction): Restrict the
LOOP_VINFO_CAN_FULLY_MASK_P handling to cases that will be
handled by vect_transform_reduction.  Allow fully-masked loops
to be used with reduction chains.
* tree-vect-stmts.c (vectorizable_operation): Handle reduction
operations in fully-masked loops.
(vectorizable_condition): Reject EXTRACT_LAST_REDUCTION
operations in fully-masked loops.

gcc/testsuite/
* gcc.dg/vect/pr65947-1.c: No longer expect doubled dump lines
for FOLD_EXTRACT_LAST reductions.
* gcc.dg/vect/pr65947-2.c: Likewise.
* gcc.dg/vect/pr65947-3.c: Likewise.
* gcc.dg/vect/pr65947-4.c: Likewise.
* gcc.dg/vect/pr65947-5.c: Likewise.
* gcc.dg/vect/pr65947-6.c: Likewise.
* gcc.dg/vect/pr65947-9.c: Likewise.
* gcc.dg/vect/pr65947-10.c: Likewise.
* gcc.dg/vect/pr65947-12.c: Likewise.
* gcc.dg/vect/pr65947-13.c: Likewise.
* gcc.dg/vect/pr65947-14.c: Likewise.
* gcc.dg/vect/pr80631-1.c: Likewise.
* gcc.dg/vect/pr80631-2.c: Likewise.
* gcc.dg/vect/vect-cond-reduc-3.c: Likewise.
* gcc.dg/vect/vect-cond-reduc-4.c: Likewise.

From-SVN: r277438

tree-vect-loop.c (vectorizable_reduction): Verify STMT_VINFO_REDUC_IDX on the to be vectorized stmts is set up correctly.

2019-10-25  Richard Biener  <rguenther@suse.de>

* tree-vect-loop.c (vectorizable_reduction): Verify
STMT_VINFO_REDUC_IDX on the to be vectorized stmts is set up
correctly.
* tree-vect-patterns.c (vect_mark_pattern_stmts): Transfer
STMT_VINFO_REDUC_IDX from the original stmts to the pattern
stmts.

From-SVN: r277437

policy_data_structures_biblio.xml: Switch pubs.opengroup.org to https.

* doc/xml/manual/policy_data_structures_biblio.xml: Switch
pubs.opengroup.org to https.

From-SVN: r277436

* doc/xml/gnu/gpl-3.0.xml: Switch gnu.org to https.

From-SVN: r277435

status_cxx2020.xml: Add rows and update status.

2019-09-09  Edward Smith-Rowland  <3dw4rd@verizon.net>

* doc/xml/manual/status_cxx2020.xml: Add rows and update status.

From-SVN: r277434

Daily bump.

From-SVN: r277433

gimplify.h (omp_construct_selector_matches): Declare.

* gimplify.h (omp_construct_selector_matches): Declare.
* gimplify.c (struct gimplify_omp_ctx): Add code member.
(gimplify_call_expr): Call omp_resolve_declare_variant and remap
called function if needed for flag_openmp.
(gimplify_scan_omp_clauses): Set ctx->code.
(omp_construct_selector_matches): New function.
* omp-general.h (omp_constructor_traits_to_codes,
omp_context_selector_matches, omp_resolve_declare_variant): Declare.
* omp-general.c (omp_constructor_traits_to_codes,
omp_context_selector_matches, omp_resolve_declare_variant): New
functions.
c-family/
* c-common.h (c_omp_context_selector_matches): Remove.
* c-omp.c (c_omp_context_selector_matches): Remove.
* c-attribs.c (c_common_attribute_table): Add
"omp declare target {host,nohost,block}" attributes.
c/
* c-parser.c (c_finish_omp_declare_variant): Use
omp_context_selector_matches instead of
c_omp_context_selector_matches.
* c-decl.c (c_decl_attributes): Add "omp declare target block"
attribute in between declare target and end declare target
pragmas.
cp/
* decl2.c (cplus_decl_attributes): Add "omp declare target block"
attribute in between declare target and end declare target
pragmas.
testsuite/
* c-c++-common/gomp/declare-variant-8.c: New test.

From-SVN: r277427

arc.c (hwloop_optimize): Add missing space in string literal.

* config/arc/arc.c (hwloop_optimize): Add missing space in string
literal.
* config/rx/rx.c (rx_print_operand): Likewise.
* tree-vect-data-refs.c (vect_analyze_data_refs): Likewise.
* tree-ssa-loop-ch.c (should_duplicate_loop_header_p): Likewise.
* ipa-sra.c (create_parameter_descriptors, process_scan_results):
Likewise.
* genemit.c (emit_c_code): Likewise.
* plugin.c (try_init_one_plugin): Likewise.  Formatting fix.
cp/
* call.c (convert_arg_to_ellipsis): Add missing space in string
literal.

From-SVN: r277426

symbols-summary.h (fast_function_summary<T *, [...]): Free m_vector.

* symbols-summary.h (fast_function_summary<T *, V>::release,
fast_call_summary<T *, V>::release): Free m_vector.

From-SVN: r277425

cgraphunit.c (symbol_table::process_new_functions): Call ipa_free_size_summary.

* cgraphunit.c (symbol_table::process_new_functions): Call
ipa_free_size_summary.
* ipa-cp.c (ipcp_cloning_candidate_p): Update.
(devirtualization_time_bonus): Update.
(ipcp_propagate_stage): Update.
* ipa-fnsummary.c (ipa_size_summaries): New.
(ipa_fn_summary_alloc): Alloc size summary.
(dump_ipa_call_summary): Update.
(ipa_dump_fn_summary): Update.
(analyze_function_body): Update.
(compute_fn_summary): Likewise.
(ipa_get_stack_frame_offset): New function.
(inline_update_callee_summaries): Do not update frame offsets.
(ipa_merge_fn_summary_after_inlining): Update frame offsets here;
remove call and function summary.
(ipa_update_overall_fn_summary): Update.
(inline_read_section): Update.
(ipa_fn_summary_write): Update.
(ipa_free_fn_summary): Do not remove summaries.
(ipa_free_size_summary): New.
(release summary pass): Also run at WPA.
* ipa-fnsummary.h (ipa_size_summary): Declare.
(ipa_fn_summary): Remove size, self_size, stack_frame_offset,
estimated_self_stack_size.
(ipa_size_summary_t): New type.
(ipa_size_summaries): Declare.
(ipa_free_size_summary): Declare.
(ipa_get_stack_frame_offset): Declare.
* ipa-icf.c (sem_function::merge): Update.
* ipa-inline-analysis.c (estimate_size_after_inlining): Update.
(estimate_growth): Update.
(growth_likely_positive): Update.
(clone_inlined_nodes): Update.
(inline_call): Update.
* ipa-inline.c (caller_growth_limits): Update.
(edge_badness): Update.
(recursive_inlining): Update.
(inline_small_functions): Update.
(inline_to_all_callers_1): Update.
* ipa-prop.h (ipa_edge_args_sum_t): Update comment.
* lto-partition.c (add_symbol_to_partition_1): Update.
(undo_parittion): Update.

From-SVN: r277424

rs6000: Implement [u]avg<mode>3_ceil

We already had those in fact, just under other names.  Use the standard
names so that the vectorizer can use it.

* config/rs6000/altivec.md (altivec_vavgu<VI_char>): Rename to...
(uavg<mode>3_ceil): ... This.
(altivec_vavgs<VI_char>): Rename to...
(avg<mode>3_ceil): ... This.
* config/rs6000/rs6000-builtin.def (VAVGUB, VAVGSB, VAVGUH, VAVGSH,
VAVGUW, VAVGSW): Adjust.

From-SVN: r277421

Add missing space to diagnostic in reshape_init_r.

* decl.c (reshape_init_r): Add missing space.

From-SVN: r277419

[C++ PATCH] Template parm index fix

https://gcc.gnu.org/ml/gcc-patches/2019-10/msg01761.html
* pt.c (reduce_template_parm_level): Attach the new TPI to the new
DECL.
(convert_generic_types_to_packs): Pass the copied type to
reduce_template_parm_level.

From-SVN: r277416

[dump] small source cleanup

https://gcc.gnu.org/ml/gcc-patches/2019-10/msg01760.html
* dumpfile.c (dump_begin): Reorder decls to use RAII.

From-SVN: r277415

Revert ABI changes to std::allocator in C++20

The recent C++20 changes to remove the std::allocator<void> explicit
specialization and the destructor in the std::allocator primary template
change the result of some is_trivially_xxx type traits. To avoid those
changes, this patch restores the explicit specialization and the
destructor.

In order to meet the C++20 requirements the std::allocator<void>
explicit specialization must provide the same interface as the primary
template (except for the unusable allocate and deallocate member
functions) and the destructor in the primary template must be constexpr.

* include/bits/allocator.h (allocator<void>): Restore the explicit
specialization for C++20, but make its API consistent with the primary
template.
(allocator::~allocator()): Restore the destructor for C++20, but make
it constexpr.
* testsuite/20_util/allocator/rebind_c++20.cc: Check allocator<void>.
* testsuite/20_util/allocator/requirements/typedefs_c++20.cc: Likewise.
* testsuite/20_util/allocator/void.cc: Check that constructors and
destructors are trivial. Check for converting constructor in C++20.
* testsuite/ext/malloc_allocator/variadic_construct.cc: Simplify
dejagnu target selector.
* testsuite/ext/new_allocator/variadic_construct.cc: Likewise.

From-SVN: r277410

ipa-sra-19.c: Avoid unprototyped function

Power and IBM Z require a function prototype if a vector argument is
passed.  Complete the prototype of k to prevent errors from being
triggered on these platforms

gcc/testsuite/ChangeLog:

2019-10-24  Andreas Krebbel  <krebbel@linux.ibm.com>

* gcc.dg/ipa/ipa-sra-19.c: Remove dg-skip-if. Add argument type to
prototype of k.

From-SVN: r277409

Make gt_pch_nx unreachable in symbol-summary classes.

2019-10-24  Martin Liska  <mliska@suse.cz>

* symbol-summary.h (gt_pch_nx): Mark all functions
with gcc_unreachable as we do not expect to be called.

From-SVN: r277408

Finish moving constraint and logic functionality of out pt.c.

Also, reimplement and re-enable subsumption caching.

gcc/cp/
* config-lang.in (gtfiles): Add logic.cc.
* constraint.cc (atomic_constraints_identical_p): Add assertions.
(hash_atomic_constraint): Likewise.
(constraints_equivalent_p): New.
(inchash::add_constraint): New.
(iterative_hash_constraint): New.
(decl_constraints): Moved from pt.c.
(get_constraints): Likewise.
(set_constraints): Likewise.
(remove_constraints): Likewise.
* cp-tree.h (CONSTR_P): New.
(init_constraint_processing): Remove.
(constraints_equivalent_p, iterative_hash_constraint): Declare.
* decl.c (cxx_init_decl_processing): Don't initialize constraints.
* logic.cc (subsumption_entry): Moved from pt.c.
(subsumption_hasher): Likewise.
(subsumption_cache): Likewise.
(lookup_subsumption): Likewise.
(save_subsumption): Likewise.
(subsumes_constraints_nonnull): Use subsumption cache.
* pt.c: Move aforementioned declarations out of this file.
(init_constraint_processing): Remove.

From-SVN: r277407

tree-vect-slp.c (vect_get_and_check_slp_defs): For reduction chains try harder with operand swapping and instead of putting a...

2019-10-24  Richard Biener  <rguenther@suse.de>

* tree-vect-slp.c (vect_get_and_check_slp_defs): For reduction
chains try harder with operand swapping and instead of
putting a shifted chain into the reduction operands put
a repetition of the final reduction op there as if we'd
reassociate the expression.

* gcc.dg/vect/slp-reduc-10a.c: New testcase.
* gcc.dg/vect/slp-reduc-10b.c: Likewise.
* gcc.dg/vect/slp-reduc-10c.c: Likewise.
* gcc.dg/vect/slp-reduc-10d.c: Likewise.
* gcc.dg/vect/slp-reduc-10e.c: Likewise.

From-SVN: r277406

Simplify common case of use_future_t that uses std::allocator

There is no need to store and pass around the allocator object when it's
an instance of std::allocator. Define a partial specialization of
std::use_future_t and the corresponding completion token so that no
allocator is stored. Overload the completion handler constructor to not
expect an allocator to be stored.

* include/experimental/executor (__use_future_ct, use_future_t):
Define partial specializations for std::allocator.
(__use_future_ch): Overload constructor for completion tokens using
std::allocator.

From-SVN: r277404

ipa-reference.c (ipa_reference_optimization_summary_d): Rename statics_not_read and statics_not_written to statics_read and...

* ipa-reference.c (ipa_reference_optimization_summary_d): Rename
statics_not_read and statics_not_written to statics_read and
statics_written respectively.
(no_module_statics): New static var.
(ipa_reference_get_not_read_global): Rename to ...
(ipa_reference_get_read_global): ... this.
(ipa_reference_get_not_written_global): Rename to ...
(ipa_reference_get_written_global): ... this.
(dump_static_vars_set_to_file): Dump no_module_statics.
(copy_static_var_set): Add for propagation parameter.
(ipa_init): Initialize no_module_statics.
(ipa_ref_opt_summary_t::duplicate): Update.
(ipa_ref_opt_summary_t::remove): Update.
(propagate): Update.
(write_node_summary_p): Look correctly for bitmap differences.
(ipa_reference_write_optimization_summary): Update.
(ipa_reference_read_optimization_summary): Update.
* ipa-reference.h
(ipa_reference_get_not_read_global): Rename to ...
(ipa_reference_get_read_global): ... this.
(ipa_reference_get_not_written_global): Rename to ...
(ipa_reference_get_written_global): ... this.
* tree-ssa-alias.c (ref_maybe_used_by_call_p_1): Update.
(call_may_clobber_ref_p_1): Update.

From-SVN: r277403

MSP430: Remove unused msp430_hard_regno_nregs_*_padding functions

2019-10-24  Jozef Lawrynowicz  <jozef.l@mittosystems.com>

* config/msp430/msp430.c (msp430_hard_regno_nregs_has_padding): Remove
and add comment.
(msp430_hard_regno_nregs_with_padding): Remove.

From-SVN: r277395

MSP430: Tweaks to generation of 430X instructions

gcc/ChangeLog:

2019-10-24  Jozef Lawrynowicz  <jozef.l@mittosystems.com>

* config/msp430/constraints.md: Allow post_inc for "Ya" constraint.
* config/msp430/msp430.md (430x_shift_left): Use RLAM when the constant
shift amount is between 1 and 4.
(430x_arithmetic_shift_right): Use RRAM when the constant shift amount
is between 1 and 4.

gcc/testsuite/ChangeLog:

2019-10-24  Jozef Lawrynowicz  <jozef.l@mittosystems.com>

* gcc.target/msp430/emulate-slli.c: Skip for -mcpu=msp430.
Add shift by a constant 5 bits.
Update scan-assembler directives.
* gcc.target/msp430/emulate-srai.c: Likewise.
* gcc.target/msp430/emulate-srli.c: Skip for -mcpu=msp430.

From-SVN: r277394

re PR tree-optimization/92205 (ICE in vect_get_vec_def_for_stmt_copy, at tree-vect-stmts.c:1688 since r277322)

2019-10-24  Richard Biener  <rguenther@suse.de>

PR tree-optimization/92205
* tree-vect-loop.c (vectorizable_reduction): Restrict
search for alternate vectype_in to lane-reducing patterns
we support.

* gcc.dg/vect/pr92205.c: New testcase.

From-SVN: r277375

re PR tree-optimization/92203 (ICE in eliminate_stmt, at tree-ssa-sccvn.c:5492)

2019-10-24  Richard Biener  <rguenther@suse.de>

PR tree-optimization/92203
* treee-ssa-sccvn.c (eliminate_dom_walker::eliminate_stmt):
Skip eliminating conversion stmts inserted by insertion.

* gcc.dg/torture/pr92203.c: New testcase.

From-SVN: r277374

re PR tree-optimization/65930 (Reduction with sign-change not handled)

2019-10-24  Richard Biener  <rguenther@suse.de>

PR tree-optimization/65930
* gcc.dg/vect/vect-reduc-2short.c: Fix typo.

From-SVN: r277373

re PR tree-optimization/65930 (Reduction with sign-change not handled)

2019-10-24  Richard Biener  <rguenther@suse.de>

PR tree-optimization/65930
* gcc.dg/vect/vect-reduc-2char-big-array.c: Adjust again.
* gcc.dg/vect/vect-reduc-2char.c: Likewise.
* gcc.dg/vect/vect-reduc-2short.c: Likewise.
* gcc.dg/vect/vect-reduc-dot-s8b.c: Likewise.
* gcc.dg/vect/vect-reduc-pattern-2c.c: Likewise.

From-SVN: r277372

PR libstdc++/88338 Implement P0898R3, C++20 concepts library

The implementation is already complete but this updates the docs and
adds tests for the feature test macro.

* doc/xml/manual/status_cxx2020.xml: Update status.
* doc/html/*: Regenerate.
* testsuite/std/concepts/1.cc: New test.
* testsuite/std/concepts/2.cc: New test.

From-SVN: r277371

Define std::uniform_random_bit_generator concept for C++20

* include/bits/random.h (uniform_random_bit_generator): Define for
C++20.
* testsuite/26_numerics/random/concept.cc: New test.
* testsuite/26_numerics/random/pr60037-neg.cc: Adjust dg-error line.

From-SVN: r277369

S/390: Use UNSPEC_GET_TP for thread pointer loads

gcc/ChangeLog:

2019-10-24  Ilya Leoshkevich  <iii@linux.ibm.com>

* config/s390/s390.c (s390_get_thread_pointer): Use
gen_get_thread_pointer.
(s390_expand_split_stack_prologue): Likewise.
* config/s390/s390.md (UNSPEC_GET_TP): New UNSPEC.
(*get_tp_31): New 31-bit splitter for UNSPEC_GET_TP.
(*get_tp_64): New 64-bit splitter for UNSPEC_GET_TP.
(get_thread_pointer<mode>): Use UNSPEC_GET_TP, use
parameterized name.

gcc/testsuite/ChangeLog:

2019-10-24  Ilya Leoshkevich  <iii@linux.ibm.com>

* gcc.target/s390/load-thread-pointer-once-2.c: New test.

From-SVN: r277368

Fix another UBSAN in Fortran coarray.

2019-10-24  Martin Liska  <mliska@suse.cz>

PR fortran/92174
* array.c (gfc_resolve_array_spec): Break the loop
for out of bounds index.
* resolve.c (is_non_constant_shape_array): Likewise.

From-SVN: r277367

tree-vect-slp.c (vect_analyze_slp): When reduction group SLP discovery fails try to handle the reduction as part of...

2019-10-24  Richard Biener  <rguenther@suse.de>

* tree-vect-slp.c (vect_analyze_slp): When reduction group
SLP discovery fails try to handle the reduction as part
of SLP reduction discovery.

* gcc.dg/vect/slp-reduc-9.c: New testcase.

From-SVN: r277366

[C++ PATCH] 'std' identifier not needed

https://gcc.gnu.org/ml/gcc-patches/2019-10/msg01707.html
* cp-tree.c (CPTI_STD_IDENTIFIER): Delete.
(std_identifier): Delete.
(DECL_NAME_SPACE_STD_P): Compare against std_node.
* decl.c (initialize_predefined_identifiers): 'std' is not needed.
(cxx_init_decl_processing): Adjust creation of ::std.  Use
{push,pop}_nested_namespace.
(cxx_builtin_function): Use {push,pop}_nested_namespace.
* except.c (init_exception_processing): Likewise.
* rtti.c (init_rtti_processing): Likewise.

From-SVN: r277365

pr70010.c: Add -Wno-psabi.

        * gcc.target/powerpc/pr70010.c: Add -Wno-psabi.
        * gcc.target/powerpc/pr70010-[12].c: Require LTO.
        * gcc.target/powerpc/pr91275.c: Add -mcpu=power8.

From-SVN: r277363

Daily bump.

From-SVN: r277362