re PR c++/88261 (ICE: verify_gimple failed (error: non-trivial conversion at assignment))

        PR c++/88261
        PR c++/69338
        PR c++/69696
        PR c++/69697
        * cp-tree.h (LOOKUP_ALLOW_FLEXARRAY_INIT): New flag value.
        * typeck2.c (digest_init_r): Raise an error for non-static
        initialization of a flexible array member.
        (process_init_constructor, massage_init_elt,
        process_init_constructor_array, process_init_constructor_record,
        process_init_constructor_union, process_init_constructor): Add the
        flags parameter and pass it thru.
        (store_init_value): Pass LOOKUP_ALLOW_FLEXARRAY_INIT parameter to
        digest_init_flags for static decls.

gcc/testsuite:
2019-01-07  Bernd Edlinger  <bernd.edlinger@hotmail.de>

        PR c++/88261
        PR c++/69338
        PR c++/69696
        PR c++/69697
        * gcc.dg/array-6.c: Move from here ...
        * c-c++-common/array-6.c: ... to here and add some more test coverage.
        * g++.dg/pr69338.C: New test.
        * g++.dg/pr69697.C: Likewise.
        * g++.dg/ext/flexary32.C: Likewise.
        * g++.dg/ext/flexary3.C: Adjust test.
        * g++.dg/ext/flexary12.C: Likewise.
        * g++.dg/ext/flexary13.C: Likewise.
        * g++.dg/ext/flexary15.C: Likewise.
        * g++.dg/warn/Wplacement-new-size-1.C: Likewise.
        * g++.dg/warn/Wplacement-new-size-2.C: Likewise.
        * g++.dg/warn/Wplacement-new-size-6.C: Likewise.

From-SVN: r267653

Fix ICE in get_initial_defs_for_reduction (PR 88567)

The use of "j" in:

      init = permute_results[number_of_vectors - j - 1];

was out-of-sync with the new flat loop structure.  Now that all that
reversing is gone, we can just use the result of duplicate_and_interleave
directly.

The other cases shouldn't be affected by postponing the insertion
of ctor_seq, since gimple_build* appends to the seq without clearing
it first (unlike some of the gimplify routines).

The ICE is already covered by gcc.dg/vect/pr63379.c.

2019-01-07  Richard Sandiford  <richard.sandiford@arm.com>

gcc/
PR middle-end/88567
* tree-vect-loop.c (get_initial_defs_for_reduction): Pass the
output vector directly to duplicate_and_interleave instead of
going through a temporary.  Postpone insertion of ctor_seq to
the end of the loop.

From-SVN: r267652

PR libstdc++/87787 avoid undefined null args to memcpy and memmove

The C++ char_traits and ctype APIs do not disallow null pointer
arguments, so we need explicit checks to ensure we don't forward null
pointers to memcpy or memmove.

PR libstdc++/87787
* include/bits/char_traits.h (char_traits::move): Do not pass null
pointers to memmove.
* include/bits/locale_facets.h
(ctype<char>::widen(const char*, const char*, char*)): Do not
pass null pointers to memcpy.
(ctype<char>::narrow(const char*, const char*, char, char*)):
Likewise.
(ctype<char>::do_widen(const char*, const char*, char*)):
Likewise.
(ctype<char>::do_narrow(const char*, const char*, char, char*)):
Likewise.

From-SVN: r267651

Investigating PR target/86891 revealed a number of issues with the way the...

Investigating PR target/86891 revealed a number of issues with the way
the AArch64 backend was handing overflow detection patterns.  Firstly,
expansion for signed and unsigned types is not the same as in one form
the overflow is detected via the C flag and in the other it is done
via the V flag in the PSR.  Secondly, particular care has to be taken
when describing overflow of signed types: the comparison has to be
performed conceptually on a value that cannot overflow and compared to
a value that might have overflowed.

It became apparent that some of the patterns were simply unmatchable
(they collapse to NEG in the RTL rather than subtracting from zero)
and a number of patterns were overly restrictive in terms of the
immediate constants that they supported.  I've tried to address all of
these issues as well.

gcc:

PR target/86891
* config/aarch64/aarch64.c (aarch64_expand_subvti): New parameter
unsigned_p.  Handle signed and unsigned overflow correction as
required.
* config/aarch64/aarch64-protos.h (aarch64_expand_subvti): Update
prototype.
* config/aarch64/aarch64.md (addv<mode>4): Use aarch64_plus_operand
for operand 2.
(add<mode>3_compareV_imm): Make this callable for expanding.
(subv<GPI:mode>4): Use register_operand for operand 1.  Use
aarch64_plus_operand for operand 2.
(subv<GPI:mode>_insn): New insn pattern.
(subv<GPI:mode>_imm): Likewise.
(negv<GPI:mode>3): New expand pattern.
(negv<GPI:mode>_insn): New insn pattern.
(negv<GPI:mode>_cmp_only): Likewise.
(cmpv<GPI:mode>_insn): Likewise.
(subvti4): Use register_operand for operand 1.  Update call to
aarch64_expand_subvti.
(usubvti4): Likewise.
(negvti3): New expand pattern.
(negdi_carryout): New insn pattern.
(negvdi_carryinV): New insn pattern.
(sub<mode3>_compare1_imm): Delete named insn pattern, make anonymous
version the named version.
(peepholes to convert to sub<mode3>_compare1_imm): Adjust order of
operands.
(usub<GPI:mode>3_carryinC, usub<GPI:mode>3_carryinC_z1): New insn
patterns.
(usub<GPI:mode>3_carryinC_z2, usub<GPI:mode>3_carryinC): New insn
patterns.
(sub<mode>3_carryinCV, sub<mode>3_carryinCV_z1_z2): Delete.
(sub<mode>3_carryinCV_z1, sub<mode>3_carryinCV_z2): Delete.
(sub<mode>3_carryinCV): Delete.
(sub<GPI:mode>3_carryinV): New expand pattern.
sub<mode>3_carryinV, sub<mode>3_carryinV_z2): New insn patterns.

testsuite:

* gcc.target/aarch64/subs_compare_2.c: Make '#' immediate prefix
optional in scan pattern.

From-SVN: r267650

tree-ssa-uncprop.c (ssa_equip_hash_traits): Remove in favor of tree_operand_hash.

2019-01-07  Richard Biener  <rguenther@suse.de>

* tree-ssa-uncprop.c (ssa_equip_hash_traits): Remove in favor
of tree_operand_hash.

From-SVN: r267649

Update documentation for C++17 filesystem library

* doc/xml/manual/spine.xml: Update copyright years.
* doc/xml/manual/status_cxx2017.xml: Adjust note about -lstdc++fs.
* doc/xml/manual/using.xml: Remove requirement to link with -lstdc++fs
for C++17 filesystem library.
* doc/html/*: Regenerate.

From-SVN: r267648

Fix build for systems without POSIX truncate

Older versions of newlib do not provide truncate so add a configure
check for it, and provide a fallback definition.

There were also some missing exports in the linker script, which went
unnoticed because there are no tests for some functions. A new link-only
test checks that every filesystem operation function is defined by the
library.

* acinclude.m4 (GLIBCXX_CHECK_FILESYSTEM_DEPS): Check for truncate.
* config.h.in: Regenerate.
* config/abi/pre/gnu.ver: Order patterns for filesystem operations
alphabetically and add missing entries for copy_symlink,
hard_link_count, rename, and resize_file.
* configure: Regenerate.
* src/c++17/fs_ops.cc (resize_file): Remove #if so posix::truncate is
used unconditionally.
* src/filesystem/ops-common.h (__gnu_posix::truncate)
[!_GLIBCXX_HAVE_TRUNCATE]: Provide fallback definition that only
supports truncating to zero length.
* testsuite/27_io/filesystem/operations/all.cc: New test.
* testsuite/27_io/filesystem/operations/resize_file.cc: New test.

From-SVN: r267647

[2/2] PR88598: Optimise reduc (bit_and)

This patch folds certain reductions of X & CST to X[I] & CST[I] if I is
the only nonzero element of CST.  This includes the motivating case in
which CST[I] is -1.

We could do the same for REDUC_MAX on unsigned types, but I wasn't sure
that that special case was worth it.

2019-01-07  Richard Sandiford  <richard.sandiford@arm.com>

gcc/
PR tree-optimization/88598
* tree.h (single_nonzero_element): Declare.
* tree.c (single_nonzero_element): New function.
* match.pd: Fold certain reductions of X & CST to X[I] & CST[I]
if I is the only nonzero element of CST.

gcc/testsuite/
PR tree-optimization/88598
* gcc.dg/vect/pr88598-1.c: New test.
* gcc.dg/vect/pr88598-2.c: Likewise.
* gcc.dg/vect/pr88598-3.c: Likewise.
* gcc.dg/vect/pr88598-4.c: Likewise.
* gcc.dg/vect/pr88598-5.c: Likewise.
* gcc.dg/vect/pr88598-6.c: Likewise.

From-SVN: r267646

[1/2] PR88598: Optimise x * { 0 or 1, 0 or 1, ... }

The PR has:

    vect__6.24_42 = vect__5.23_41 * { 0.0, 1.0e+0, 0.0, 0.0 };

which for -fno-signed-zeros -fno-signaling-nans can be simplified to:

    vect__6.24_42 = vect__5.23_41 & { 0, -1, 0, 0 };

I deliberately didn't handle COMPLEX_CST or CONSTRUCTOR in
initializer_each_zero_or_onep since there are no current use cases.

The patch also makes (un)signed_type_for handle floating-point types.
I tried to audit all callers and the few that handle null returns would
be unaffected.

2019-01-07  Richard Sandiford  <richard.sandiford@arm.com>

gcc/
PR tree-optimization/88598
* tree.h (initializer_each_zero_or_onep): Declare.
* tree.c (initializer_each_zero_or_onep): New function.
(signed_or_unsigned_type_for): Handle float types too.
(unsigned_type_for, signed_type_for): Update comments accordingly.
* match.pd: Fold x * { 0 or 1, 0 or 1, ...} to
x & { 0 or -1, 0 or -1, ... }.

gcc/testsuite/
PR tree-optimization/88598
* gcc.dg/pr88598-1.c: New test.
* gcc.dg/pr88598-2.c: Likewise.
* gcc.dg/pr88598-3.c: Likewise.
* gcc.dg/pr88598-4.c: Likewise.
* gcc.dg/pr88598-5.c: Likewise.

From-SVN: r267645

Replace outdated references to x86_64-unknown-linux-gnu in docs

* doc/install.texi: Replace references to x86_64-unknown-linux-gnu
with x86_64-pc-linux-gnu.

From-SVN: r267643

[nvptx] Force vl32 if calling vector-partitionable routines

With PTX_MAX_VECTOR_LENGTH set to larger than PTX_WARP_SIZE, routines can be
called from offloading regions with vector-size set to larger than warp size.
OTOH, vector-partitionable routines assume warp-sized vector length.

Detect if we're calling a vector-partitionable routine from an offloading
region, and if so, fall back to warp-sized vector length in that region.

2019-01-07  Tom de Vries  <tdevries@suse.de>

PR target/85486
* config/nvptx/nvptx.c (has_vector_partitionable_routine_calls_p): New
function.
(nvptx_goacc_validate_dims): Force vl32 if calling vector-partitionable
routines.

From-SVN: r267640

sse.md (vec_extract<mode><ssehalfvecmodelower>): Use V_256_512 iterator instead of V_512 and TARGET_AVX instead of...

* config/i386/sse.md (vec_extract<mode><ssehalfvecmodelower>): Use
V_256_512 iterator instead of V_512 and TARGET_AVX instead of
TARGET_AVX512F as condition.

From-SVN: r267639

re PR debug/88723 (PR debug/88635 patch breaks testsuite_shared.cc compilation)

PR debug/88723
* dwarf2out.c (const_ok_for_output_1): Remove redundant call to
const_not_ok_for_debug_p target hook.
(mem_loc_descriptor) <case UNSPEC>: Only call const_ok_for_output_1
on UNSPEC and subexpressions thereof if all subexpressions of the
UNSPEC are CONSTANT_P.

From-SVN: r267638

re PR tree-optimization/88676 (missed opportunity in integer conditional)

PR tree-optimization/88676
* tree-ssa-phiopt.c (two_value_replacement): New function.
(tree_ssa_phiopt_worker): Call it.

* gcc.dg/tree-ssa/pr88676.c: New test.
* gcc.dg/pr88676.c: New test.
* gcc.dg/tree-ssa/pr15826.c: Just verify there is no goto,
allow &.

From-SVN: r267634

re PR sanitizer/88619 (ICE in asan_emit_stack_protection, at asan.c:1574 since r266664)

PR sanitizer/88619
* cfgexpand.c (expand_stack_vars): Only align prev_offset to
ASAN_MIN_RED_ZONE_SIZE, not to maximum of that and alignb.

* c-c++-common/asan/pr88619.c: New test.

From-SVN: r267633

re PR c++/85052 (Implement support for clang's __builtin_convertvector)

PR c++/85052
* tree-vect-generic.c: Include insn-config.h and recog.h.
(expand_vector_piecewise): Add defaulted ret_type argument,
if non-NULL, use that in preference to type for the result type.
(expand_vector_parallel): Formatting fix.
(do_vec_conversion, do_vec_narrowing_conversion,
expand_vector_conversion): New functions.
(expand_vector_operations_1): Call expand_vector_conversion
for VEC_CONVERT ifn calls.
* internal-fn.def (VEC_CONVERT): New internal function.
* internal-fn.c (expand_VEC_CONVERT): New function.
* fold-const-call.c (fold_const_vec_convert): New function.
(fold_const_call): Use it for CFN_VEC_CONVERT.
* doc/extend.texi (__builtin_convertvector): Document.
c-family/
* c-common.h (enum rid): Add RID_BUILTIN_CONVERTVECTOR.
(c_build_vec_convert): Declare.
* c-common.c (c_build_vec_convert): New function.
c/
* c-parser.c (c_parser_postfix_expression): Parse
__builtin_convertvector.
cp/
* cp-tree.h (cp_build_vec_convert): Declare.
* parser.c (cp_parser_postfix_expression): Parse
__builtin_convertvector.
* constexpr.c: Include fold-const-call.h.
(cxx_eval_internal_function): Handle IFN_VEC_CONVERT.
(potential_constant_expression_1): Likewise.
* semantics.c (cp_build_vec_convert): New function.
* pt.c (tsubst_copy_and_build): Handle CALL_EXPR to
IFN_VEC_CONVERT.
testsuite/
* c-c++-common/builtin-convertvector-1.c: New test.
* c-c++-common/torture/builtin-convertvector-1.c: New test.
* g++.dg/ext/builtin-convertvector-1.C: New test.
* g++.dg/cpp0x/constexpr-builtin4.C: New test.

From-SVN: r267632

[nvptx] Handle large vector reductions

Add support for vector reductions with openacc vector_length larger than
warp-size.

2019-01-07  Tom de Vries  <tdevries@suse.de>

* config/nvptx/nvptx-protos.h (nvptx_output_red_partition): Declare.
* config/nvptx/nvptx.c (vector_red_size, vector_red_align,
vector_red_partition, vector_red_sym): New global variables.
(nvptx_option_override): Initialize vector_red_sym.
(nvptx_declare_function_name): Restore red_partition register.
(nvptx_file_end): Emit code to declare the vector reduction variables.
(nvptx_output_red_partition): New function.
(nvptx_expand_shared_addr): Add vector argument. Use it to handle
large vector reductions.
(enum nvptx_builtins): Add NVPTX_BUILTIN_VECTOR_ADDR.
(nvptx_init_builtins): Add VECTOR_ADDR.
(nvptx_expand_builtin): Update call to nvptx_expand_shared_addr.
Handle nvptx_expand_shared_addr.
(nvptx_get_shared_red_addr): Add vector argument and handle large
vectors.
(nvptx_goacc_reduction_setup): Add offload_attrs argument and handle
large vectors.
(nvptx_goacc_reduction_init): Likewise.
(nvptx_goacc_reduction_fini): Likewise.
(nvptx_goacc_reduction_teardown): Likewise.
(nvptx_goacc_reduction): Update calls to nvptx_goacc_reduction_{setup,
init,fini,teardown}.
(nvptx_init_axis_predicate): Initialize vector_red_partition.
(nvptx_set_current_function): Init vector_red_partition.
* config/nvptx/nvptx.md (UNSPECV_RED_PART): New unspecv.
(nvptx_red_partition): New insn.
* config/nvptx/nvptx.h (struct machine_function): Add red_partition.

From-SVN: r267631

[nvptx] Don't emit barriers for empty loops -- fix

When compiling an empty loop:
...
  long long v1;
  #pragma acc parallel num_gangs (640) num_workers(1) vector_length (128)
  #pragma acc loop
    for (v1 = 0; v1 < 20; v1 += 2)
        ;
...
the compiler emits two subsequent bar.syncs.  This triggers some bug on my
quadro m1200 (I'm assuming in the ptxas/JIT compiler) that hangs the testcase.

This patch works around the bug by doing an optimization: we detect that this is
an empty loop (a forked immediately followed by a joining), and don't emit the
barriers.

The patch does not include the test-case yet, since vector_length (128) is not
yet supported at this point.

2019-01-07  Tom de Vries  <tdevries@suse.de>

PR target/85381
* config/nvptx/nvptx.c (nvptx_process_pars): Don't emit barriers for
empty loops.

From-SVN: r267630

[nvptx] Add support for a per-worker broadcast buffer and barrier

Add support for a per-worker broadcast buffer and barrier, to be used for
openacc vector_length larger than warp-size.

2019-01-07  Tom de Vries  <tdevries@suse.de>

* config/nvptx/nvptx.c (oacc_bcast_partition): Declare.
(nvptx_option_override): Init oacc_bcast_partition.
(nvptx_init_oacc_workers): New function.
(nvptx_declare_function_name): Call nvptx_init_oacc_workers.
(nvptx_needs_shared_bcast): New function.
(nvptx_find_par): Generalize to enable vectors to use shared-memory
to propagate state.
(nvptx_shared_propagate): Initialize vector bcast partition and
synchronization state.
(nvptx_single):  Generalize to enable vectors to use shared-memory
to propagate state.
(nvptx_process_pars): Likewise.
(nvptx_set_current_function): Initialize oacc_broadcast_partition.
* config/nvptx/nvptx.h (struct machine_function): Add
bcast_partition and sync_bar members.

From-SVN: r267629

[nvptx] Allow larger PTX_MAX_VECTOR_LENGTH in nvptx_goacc_validate_dims_1

Allow PTX_MAX_VECTOR_LENGTH to be define as larger than PTX_WARP_SIZE in
nvptx_goacc_validate_dims_1.

2019-01-07  Tom de Vries  <tdevries@suse.de>

* config/nvptx/nvptx.c (nvptx_welformed_vector_length_p)
(nvptx_apply_dim_limits): New function.
(nvptx_goacc_validate_dims_1): Allow PTX_MAX_VECTOR_LENGTH larger than
PTX_WARP_SIZE.

From-SVN: r267628

[nvptx] Postpone warnings in nvptx_goacc_validate_dims_1

Move warnings in nvptx_goacc_validate_dims_1 to as late as possible.  This
allows us more flexibility in setting the dimensions.

2019-01-07  Tom de Vries  <tdevries@suse.de>

* config/nvptx/nvptx.c (nvptx_goacc_validate_dims_1): Move warnings to
as late as possible.

From-SVN: r267627

[nvptx] Eliminate PTX_VECTOR_LENGTH

Remove PTX_VECTOR_LENGTH and replace uses of it with PTX_DEFAULT_VECTOR_LENGTH,
PTX_MAX_VECTOR_LENGTH and PTX_WARP_SIZE.

2019-01-07  Tom de Vries  <tdevries@suse.de>

* config/nvptx/nvptx.c (PTX_VECTOR_LENGTH): Remove.
(PTX_DEFAULT_VECTOR_LENGTH, PTX_MAX_VECTOR_LENGTH): Define.
(nvptx_goacc_validate_dims_1, nvptx_dim_limit)
(nvptx_goacc_reduction_fini): Use PTX_DEFAULT_VECTOR_LENGTH,
PTX_MAX_VECTOR_LENGTH and PTX_WARP_SIZE instead of PTX_VECTOR_LENGTH.

From-SVN: r267626

[nvptx] Add asserts in nvptx_goacc_validate_dims

Add a few asserts to nvptx_goacc_validate_dims.

2019-01-07  Tom de Vries  <tdevries@suse.de>

* config/nvptx/nvptx.c (nvptx_goacc_validate_dims): Add asserts.

From-SVN: r267625

[nvptx] Fix libgomp.oacc-c-c++-common/vector-length-128-3.c

The vector-length-128-3.c test-case uses GOMP_OPENACC_DIM=-:-:128, but '-' is
not yet supported on trunk.  Use GOMP_OPENACC_DIM=::128 instead.

2019-01-07  Tom de Vries  <tdevries@suse.de>

* testsuite/libgomp.oacc-c-c++-common/vector-length-128-3.c: Fix
GOMP_OPENACC_DIM argument.

From-SVN: r267624

[openacc] Add oacc_get_min_dim

Expose oacc_min_dims to backends.

2019-01-07  Tom de Vries  <tdevries@suse.de>

* omp-offload.c (oacc_get_min_dim): New function.
* omp-offload.h (oacc_get_min_dim): Declare.

From-SVN: r267623

re PR target/88521 (GCC from r266355 miscompiles x265 for mingw-w64 target)

PR target/88521
* config/i386/i386.c (function_value_ms_64): Return small sturct in
AX_REG and float/double in FIRST_SSE_REG for 4 or 8 byte modes.

From-SVN: r267622

Make GFORTRAN_9 symbol node depend on GFORTRAN_8.

At some point when the GFORTRAN_9 node was added it was forgotten to
make it depend on GFORTRAN_8. This patch fixes this.

Committed as obvious.

2019-01-07  Janne Blomqvist  <jb@gcc.gnu.org>

       * gfortran.map (GFORTRAN_9): Make GFORTRAN_9 node depend on
       GFORTRAN_8.

From-SVN: r267621

Daily bump.

From-SVN: r267619

PR libstdc++/86756 Move rest of std::filesystem to libstdc++.so

Move std::filesystem directory iterators and operations from
libstdc++fs.a to main libstdc++ library. These components have many
dependencies on OS support, which is not available on all targets. Some
additional autoconf checks and conditional compilation is needed to
ensure the files will build for all targets. Previously this code was
not compiled without --enable-libstdcxx-filesystem-ts but the C++17
components should be available for all hosted builds.

The tests for these components no longer need to link to libstdc++fs.a,
but are not expected to pass on all targets. To avoid numerous failures
on targets which are not expected to pass the tests (due to missing OS
functionality) leave the dg-require-filesystem-ts directives in place
for now. This will ensure the tests only run for builds where the
filesystem-ts library is built, which presumably means some level of OS
support is present.

PR libstdc++/86756
* acinclude.m4 (GLIBCXX_CHECK_FILESYSTEM_DEPS): Check for utime and
lstat and define _GLIBCXX_USE_UTIME and _GLIBCXX_USE_LSTAT.
* config.h.in: Regenerate.
* config/abi/pre/gnu.ver (GLIBCXX_3.4.26): Export symbols for
remaining std::filesystem types and functions.
* configure: Regenerate.
* src/c++17/Makefile.am: Add C++17 filesystem sources.
* src/c++17/Makefile.in: Regenerate.
* src/c++17/cow-fs_dir.cc: Move src/filesystem/cow-std-dir.cc to
here, and change name of included file.
* src/c++17/cow-fs_ops.cc: Move src/filesystem/cow-std-ops.cc to
here, and change name of included file.
* src/c++17/fs_dir.cc: Move src/filesystem/std-dir.cc to here. Change
path to dir-common.h.
* src/c++17/fs_ops.cc: Move src/filesystem/std-ops.cc to here. Change
path to ops-common.h. Disable -Wunused-parameter warnings.
(internal_file_clock): Define unconditionally.
[!_GLIBCXX_HAVE_SYS_STAT_H] (internal_file_clock::from_stat): Do not
define.
(do_copy_file, do_space): Move definitions to ops.common.h.
(copy, file_size, hard_link_count, last_write_time, space): Only
perform operation when _GLIBCXX_HAVE_SYS_STAT_H is defined, otherwise
report an error.
(last_write_time, read_symlink): Remove unused attributes from
parameters.
* src/filesystem/Makefile.am: Remove C++17 filesystem sources.
* src/filesystem/Makefile.in: Regenerate.
* src/filesystem/cow-std-dir.cc: Move to src/c++17/cow-fs_dir.cc.
* src/filesystem/cow-std-ops.cc: Move to src/c++17/cow-fs_ops.cc.
* src/filesystem/std-dir.cc: Move to src/c++17/fs_dir.cc.
* src/filesystem/std-ops.cc: Move to src/c++17/fs_ops.cc.
* src/filesystem/dir-common.h [!_GLIBCXX_HAVE_DIRENT_H]: Define
dummy types and functions instead of using #error.
* src/filesystem/dir.cc [!_GLIBCXX_HAVE_DIRENT_H]: Use #error.
* src/filesystem/ops-common.h [!_GLIBCXX_USE_LSTAT] (lstat): Define
in terms of stat.
[!_GLIBCXX_HAVE_UNISTD_H]: Define dummy types and functions.
(do_copy_file, do_space): Move definitions here from std-ops.cc.
* src/filesystem/ops.cc: Adjust calls to do_copy_file and do_space
to account for new namespace.
* testsuite/27_io/filesystem/directory_entry/86597.cc: Remove
-lstdc++fs from dg-options.
* testsuite/27_io/filesystem/directory_entry/lwg3171.cc: Likewise.
* testsuite/27_io/filesystem/file_status/1.cc: Likewise.
* testsuite/27_io/filesystem/filesystem_error/cons.cc: Likewise.
* testsuite/27_io/filesystem/filesystem_error/copy.cc: Likewise.
* testsuite/27_io/filesystem/iterators/directory_iterator.cc:
Likewise.
* testsuite/27_io/filesystem/iterators/pop.cc: Likewise.
* testsuite/27_io/filesystem/iterators/recursive_directory_iterator.cc:
Likewise.
* testsuite/27_io/filesystem/operations/absolute.cc: Likewise.
* testsuite/27_io/filesystem/operations/canonical.cc: Likewise.
* testsuite/27_io/filesystem/operations/copy.cc: Likewise.
* testsuite/27_io/filesystem/operations/copy_file.cc: Likewise.
* testsuite/27_io/filesystem/operations/create_directories.cc:
Likewise.
* testsuite/27_io/filesystem/operations/create_directory.cc: Likewise.
* testsuite/27_io/filesystem/operations/create_symlink.cc: Likewise.
* testsuite/27_io/filesystem/operations/current_path.cc: Likewise.
* testsuite/27_io/filesystem/operations/equivalent.cc: Likewise.
* testsuite/27_io/filesystem/operations/exists.cc: Likewise.
* testsuite/27_io/filesystem/operations/file_size.cc: Likewise.
* testsuite/27_io/filesystem/operations/is_empty.cc: Likewise.
* testsuite/27_io/filesystem/operations/last_write_time.cc: Likewise.
* testsuite/27_io/filesystem/operations/permissions.cc: Likewise.
* testsuite/27_io/filesystem/operations/proximate.cc: Likewise.
* testsuite/27_io/filesystem/operations/read_symlink.cc: Likewise.
* testsuite/27_io/filesystem/operations/relative.cc: Likewise.
* testsuite/27_io/filesystem/operations/remove.cc: Likewise.
* testsuite/27_io/filesystem/operations/remove_all.cc: Likewise.
* testsuite/27_io/filesystem/operations/space.cc: Likewise.
* testsuite/27_io/filesystem/operations/status.cc: Likewise.
* testsuite/27_io/filesystem/operations/symlink_status.cc: Likewise.
* testsuite/27_io/filesystem/operations/temp_directory_path.cc:
Likewise.
* testsuite/27_io/filesystem/operations/weakly_canonical.cc: Likewise.

From-SVN: r267616

PR libstdc++/86756 add std::filesystem::path to libstdc++.so

Move the C++17 std::filesystem::path definitions from the libstdc++fs.a
archive to the main libstdc++ library. The path classes do not depend on
any OS functions, so can be defined unconditionally on all targets
(rather than depending on --enable-libstdcxx-filesystem-ts). The tests
should pass on all targets too.

PR libstdc++/86756
* config/abi/pre/gnu.ver (GLIBCXX_3.4): Make various patterns for
typeinfo and vtables less greedy.
(GLIBCXX_3.4.26): Export symbols for std::filesystem::path.
* src/c++17/Makefile.am: Add fs_path.cc and cow-fs_path.cc.
* src/c++17/Makefile.in: Regenerate.
* src/c++17/cow-fs_path.cc: Move src/filesystem/cow-std-path.cc to
here, and change name of included file.
* src/c++17/fs_path.cc: Move src/filesystem/std-path.cc to here.
* src/filesystem/Makefile.am: Remove std-path.cc and cow-std-path.cc
from sources.
* src/filesystem/Makefile.in: Regenerate.
* src/filesystem/cow-std-path.cc: Move to src/c++17/cow-fs_path.cc.
* src/filesystem/std-path.cc: Move to src/c++17/fs_path.cc.
* testsuite/27_io/filesystem/path/append/path.cc: Remove -lstdc++fs
from dg-options and remove dg-require-filesystem-ts.
* testsuite/27_io/filesystem/path/append/source.cc: Likewise.
* testsuite/27_io/filesystem/path/assign/assign.cc: Likewise.
* testsuite/27_io/filesystem/path/assign/copy.cc: Likewise.
* testsuite/27_io/filesystem/path/compare/compare.cc: Likewise.
* testsuite/27_io/filesystem/path/compare/lwg2936.cc: Likewise.
* testsuite/27_io/filesystem/path/compare/path.cc: Likewise.
* testsuite/27_io/filesystem/path/compare/strings.cc: Likewise.
* testsuite/27_io/filesystem/path/concat/path.cc: Likewise.
* testsuite/27_io/filesystem/path/concat/strings.cc: Likewise.
* testsuite/27_io/filesystem/path/construct/80762.cc: Likewise.
* testsuite/27_io/filesystem/path/construct/copy.cc: Likewise.
* testsuite/27_io/filesystem/path/construct/default.cc: Likewise.
* testsuite/27_io/filesystem/path/construct/format.cc: Likewise.
* testsuite/27_io/filesystem/path/construct/locale.cc: Likewise.
* testsuite/27_io/filesystem/path/construct/range.cc: Likewise.
* testsuite/27_io/filesystem/path/construct/string_view.cc: Likewise.
* testsuite/27_io/filesystem/path/decompose/extension.cc: Likewise.
* testsuite/27_io/filesystem/path/decompose/filename.cc: Likewise.
* testsuite/27_io/filesystem/path/decompose/parent_path.cc: Likewise.
* testsuite/27_io/filesystem/path/decompose/relative_path.cc: Likewise.
* testsuite/27_io/filesystem/path/decompose/root_directory.cc:
Likewise.
* testsuite/27_io/filesystem/path/decompose/root_name.cc: Likewise.
* testsuite/27_io/filesystem/path/decompose/root_path.cc: Likewise.
* testsuite/27_io/filesystem/path/decompose/stem.cc: Likewise.
* testsuite/27_io/filesystem/path/generation/normal.cc: Likewise.
* testsuite/27_io/filesystem/path/generation/normal2.cc: Likewise.
* testsuite/27_io/filesystem/path/generation/proximate.cc: Likewise.
* testsuite/27_io/filesystem/path/generation/relative.cc: Likewise.
* testsuite/27_io/filesystem/path/generic/generic_string.cc: Likewise.
* testsuite/27_io/filesystem/path/itr/components.cc: Likewise.
* testsuite/27_io/filesystem/path/itr/traversal.cc: Likewise.
* testsuite/27_io/filesystem/path/modifiers/clear.cc: Likewise.
* testsuite/27_io/filesystem/path/modifiers/make_preferred.cc:
Likewise.
* testsuite/27_io/filesystem/path/modifiers/remove_filename.cc:
Likewise.
* testsuite/27_io/filesystem/path/modifiers/replace_extension.cc:
Likewise.
* testsuite/27_io/filesystem/path/modifiers/replace_filename.cc:
Likewise.
* testsuite/27_io/filesystem/path/modifiers/swap.cc: Likewise.
* testsuite/27_io/filesystem/path/native/string.cc: Likewise.
* testsuite/27_io/filesystem/path/nonmember/append.cc: Likewise.
* testsuite/27_io/filesystem/path/nonmember/hash_value.cc: Likewise.
* testsuite/27_io/filesystem/path/query/empty.cc: Likewise.
* testsuite/27_io/filesystem/path/query/has_extension.cc: Likewise.
* testsuite/27_io/filesystem/path/query/has_filename.cc: Likewise.
* testsuite/27_io/filesystem/path/query/has_parent_path.cc: Likewise.
* testsuite/27_io/filesystem/path/query/has_relative_path.cc: Likewise.
* testsuite/27_io/filesystem/path/query/has_root_directory.cc:
Likewise.
* testsuite/27_io/filesystem/path/query/has_root_name.cc: Likewise.
* testsuite/27_io/filesystem/path/query/has_root_path.cc: Likewise.
* testsuite/27_io/filesystem/path/query/has_stem.cc: Likewise.
* testsuite/27_io/filesystem/path/query/is_absolute.cc: Likewise.
* testsuite/27_io/filesystem/path/query/is_relative.cc: Likewise.

From-SVN: r267615

PR libstdc++/87431 fix regression introduced by r264574

The previous patch for PR 87431 assumed that initialing a scalar type
could not throw, but it can obtain its value via a conversion operator,
which could throw. This meant the variant could get into a valueless
state, but the valueless_by_exception() member function would always
return false.

This patch fixes it by changing the emplace members to have strong
exception safety when initializing a contained value of trivially
copyable type. The _M_valid() member gets a corresponding change to
always return true for trivially copyable types, not just scalar types.

Strong exception safety (i.e. never becoming valueless) is achieved by
only replacing the current contained value once any potentially throwing
operations have completed. If constructing the new contained value can
throw then a new std::variant object is constructed to hold it, and then
move-assigned to *this (which won't throw).

PR libstdc++/87431
* include/std/variant (_Variant_storage<true, _Types...>::_M_valid):
Check is_trivially_copyable instead of is_scalar.
(variant::emplace<N, Args>(Args&&...)): If construction of the new
contained value can throw and its type is trivially copyable then
construct into a temporary variant and move from it, to provide the
strong exception safety guarantee.
(variant::emplace<N, U, Args>(initializer_list<U>, Args&&...)):
Likewise.
* testsuite/20_util/variant/87431.cc: New test.
* testsuite/20_util/variant/run.cc: Adjust test so that throwing
conversion causes valueless state.

From-SVN: r267614

re PR tree-optimization/86020 (Performance regression in Eigen geometry.cpp test starting with r248334)

PR tree-opt/86020
Revert:
2017-05-22  Jan Hubicka  <hubicka@ucw.cz>

        * ipa-inline.c (edge_badness): Use inlined_time instead of
        inline_summaries->get.

From-SVN: r267612

opts.c (enable_fdo_optimizations): Enable version-loops-for-strides...

* opts.c (enable_fdo_optimizations): Enable
version-loops-for-strides, loop-interchange, unrol-and-jam
and tree-loop-distribution.
* invoke.texi: Document newly enabled options.

From-SVN: r267611

re PR fortran/88658 (Intrinsic MAX1 returns a REAL result, should be INTEGER.)

2019-01-06  Thomas Koenig  <tkoenig@gcc.gnu.org>

PR fortran/88658
* gfortran.h: Add macro gfc_real_4_kind
* simplify.c (simplify_min_max): Special case for the types of
AMAX0, AMIN0, MAX1 and MIN1, which actually change the types of
their arguments.

2019-01-06  Thomas Koenig  <tkoenig@gcc.gnu.org>

PR fortran/88658
* gfortran.dg/min_max_type_2.f90: New test.

From-SVN: r267609

re PR c/88363 (alloc_align attribute doesn't accept enumerated arguments)

PR c/88363
* c-c++-common/attributes-4.c (falloc_align_int128,
falloc_size_int128): Guard with #ifdef __SIZEOF_INT128__.

From-SVN: r267608

PR libstdc++/88607 add tests using -finput-charset=ascii

This verifies that the <bits/extc++.h> header can be compiled with ASCII
as the input character set.

PR libstdc++/88607
* testsuite/17_intro/headers/c++1998/charset.cc: New test.
* testsuite/17_intro/headers/c++2011/charset.cc: New test.
* testsuite/17_intro/headers/c++2014/charset.cc: New test.
* testsuite/17_intro/headers/c++2017/charset.cc: New test.
* testsuite/17_intro/headers/c++2020/charset.cc: New test.

From-SVN: r267607

Daily bump.

From-SVN: r267606

invoke.texi (max-inline-insns-small): New parameters.

* doc/invoke.texi (max-inline-insns-small): New parameters.
* ipa-inline.c (want_early_inline_function_p): simplify.
(want_inline_small_function_p): Fix pasto from previous patch;
use max-inline-insns-small bound.
* params.def (max-inline-insns-small): New param.
* ipa-fnsummary.c (analyze_function_body): Initialize time/size
variables correctly.

From-SVN: r267603

Define new filesystem::__file_clock type

In C++17 the clock used for filesystem::file_time_type is unspecified,
allowing it to be chrono::system_clock. The C++2a draft requires it to
be a distinct type, with additional member functions to convert to/from
other clocks (either the system clock or UTC). In order to avoid an ABI
change later, this patch defines a new distinct type now, which will be
used for std::chrono::file_clock later.

* include/bits/fs_fwd.h (__file_clock): Define new clock.
(file_time_type): Redefine in terms of __file_clock.
* src/filesystem/ops-common.h (file_time): Add FIXME comment about
overflow.
* src/filesystem/std-ops.cc (is_set(perm_options, perm_options)): Give
internal linkage.
(internal_file_lock): New helper type for accessing __file_clock.
(do_copy_file): Use internal_file_lock to convert system time to
file_time_type.
(last_write_time(const path&, error_code&)): Likewise.
(last_write_time(const path&, file_time_type, error_code&)): Likewise.

From-SVN: r267602

invoke.texi: Document max-inline-insns-size...

* doc/invoke.texi: Document max-inline-insns-size,
uninlined-function-insns, uninlined-function-time,
uninlined-thunk-insns and uninlined-thunk-time.
* params.def: Add max-inline-insns-size,
uninlined-function-insns, uninlined-function-time,
uninlined-thunk-insns and uninlined-thunk-time.
* ipa-fnsummary.c (compute_fn_summary, analyze_function_body): Use
new parameters.
* ipa-inline.c (can_inline_edge_by_limits_p,
want_inline_small_function_p): Use new parameters.

From-SVN: r267601

* ipa-fnsummary.c (analyze_function_body): Fix accounting of time.

From-SVN: r267600

plugindir1.c: Adjust dg-prune-output for Darwin.

2019-01-05  Dominique d'Humieres  <dominiq@gcc.gnu.org>

* gcc.dg/plugin/plugindir1.c: Adjust dg-prune-output for Darwin.
* gcc.dg/plugin/plugindir2.c: Likewise.
* gcc.dg/plugin/plugindir3.c: Likewise.
* gcc.dg/plugin/plugindir4.c: Likewise.

From-SVN: r267599

re PR fortran/88009 (ICE in find_intrinsic_vtab, at fortran/class.c:2761)

2019-01-05  Janus Weil  <janus@gcc.gnu.org>

PR fortran/88009
* class.c (gfc_find_derived_vtab): Mark the _final component as
artificial.
(find_intrinsic_vtab): Ditto. Also add an extra check to avoid
dereferencing a null pointer and adjust indentation.
* resolve.c (resolve_fl_variable): Add extra check to avoid
dereferencing a null pointer. Move variable declarations to local scope.
(resolve_fl_procedure): Add extra check to avoid dereferencing a null
pointer.
* symbol.c (check_conflict): Suppress errors for artificial symbols.

2019-01-05  Janus Weil  <janus@gcc.gnu.org>

PR fortran/88009
* gfortran.dg/blockdata_10.f90: New test case.

From-SVN: r267598

re PR middle-end/82564 (ICE at -O1 and above: in assign_stack_temp_for_type, at function.c:783)

PR middle-end/82564
PR target/88620
* expr.c (expand_assignment): For calls returning VLA structures
if to_rtx is not a MEM, force it into a stack temporary.

* gcc.dg/nested-func-12.c: New test.
* gcc.c-torture/compile/pr82564.c: New test.

From-SVN: r267595

re PR debug/88635 (Assembler error when building with "-g -O2 -m32")

PR debug/88635
* dwarf2out.c (const_ok_for_output_1): Reject MINUS that contains
SYMBOL_REF, CODE_LABEL or UNSPEC in subexpressions of second argument.
Reject PLUS that contains SYMBOL_REF, CODE_LABEL or UNSPEC in
subexpressions of both operands.
(mem_loc_descriptor): Handle UNSPEC if target hook acks it and all the
subrtxes are CONSTANT_P.
* config/i386/i386.c (ix86_const_not_ok_for_debug_p): Revert
2018-11-09 changes.

* gcc.dg/debug/dwarf2/pr88635.c: New test.

From-SVN: r267594

re PR target/60563 (FAIL: g++.dg/ext/sync-4.C on *-apple-darwin*)

2019-01-05  Dominique d'Humieres  <dominiq@gcc.gnu.org>

PR target/60563
Missing PR entry in the previous commit.

From-SVN: r267593

sync-4.C: Add dg-xfail-run-if for darwin.

2019-01-05  Dominique d'Humieres  <dominiq@gcc.gnu.org>

* g++.dg/ext/sync-4.C: Add dg-xfail-run-if for darwin.

From-SVN: r267592

PR c/88546 - Copy attribute unusable for weakrefs

gcc/c-family/ChangeLog:

PR c/88546
* c-attribs.c (handle_copy_attribute): Avoid copying attribute leaf.
Handle C++ empty throw specification and C11 _Noreturn.
(has_attribute): Also handle C11 _Noreturn.

gcc/ChangeLog:

PR c/88546
* attribs.c (decls_mismatched_attributes): Avoid warning for attribute
leaf.

gcc/testsuite/ChangeLog:

PR c/88546
* g++.dg/ext/attr-copy.C: New test.
* gcc.dg/attr-copy-4.c: Disable macro expansion tracking.
* gcc.dg/attr-copy-6.c: New test.
* gcc.dg/attr-copy-7.c: New test.

From-SVN: r267591

runtime: prevent deadlock when profiling signal arrives during traceback

    Traceback routines, e.g. callers and funcentry, may call
    __go_get_backtrace_state. If a profiling signal arrives while we
    are in the critical section of __go_get_backtrace_state, it tries
    to do a traceback, which also calls __go_get_backtrace_state,
    which tries to enter the same critical section and will deadlock.
    Prevent this deadlock by setting up runtime_in_callers before
    calling __go_get_backtrace_state.

    Found while investigating golang/go#29448. Will add a test in the
    next CL.

    Updates golang/go#29448.

    Reviewed-on: https://go-review.googlesource.com/c/156037

From-SVN: r267590

* params.def (hot-bb-count-ws-permille): Set to 990.

From-SVN: r267589

Daily bump.

From-SVN: r267588

Export explicit instantiations for C++17 members of std::string

The C++17 standard added some new members to std::basic_string, which
were not previously instantiated in the library. This meant that the
extern template declarations had to be disabled for C++17 mode. With
this patch the new members are instantiated in the library and so the
explicit instantiation declarations can be used for C++17.

The new members added by C++2a are still not exported, and so the
explicit instantiation declarations are still disabled for C++2a.

* config/abi/pre/gnu.ver (GLIBCXX_3.4.21): Make patterns less greedy
for const member functions of std::basic_string.
(GLIBCXX_3.4.26): Export member functions of std::basic_string added
in C++17.
* include/bits/basic_string.h (basic_string(__sv_wrapper, const A&)):
Make non-standard constructor private.
[!_GLIBCXX_USE_CXX11_ABI] (basic_string(__sv_wrapper, const A&)):
Likewise.
* include/bits/basic_string.tcc (std::string, std::wstring): Declare
explicit instantiations for C++17 as well as earlier dialects.
* src/c++17/Makefile.am: Add new source files.
* src/c++17/Makefile.in: Regenerate.
* src/c++17/cow-string-inst.cc: New file defining explicit
instantiations for basic_string member functions added in C++17.
* src/c++17/string-inst.cc: Likewise.

From-SVN: r267585

Add allocator-extended copy/move ctors to COW string

Add these constructors from C++11 which were missing from the COW
basic_string. Additionally simplify the definitions of the
basic_string::reference and basic_string::const_reference types as
required by C++11.

This allows filesystem::path::string<Allocator>() to be simplified, so
that the same code is used for both basic_string implementations.

* config/abi/pre/gnu.ver (GLIBCXX_3.4.26): Export allocator-extended
copy/move constructors for old std::basic_string.
* include/bits/basic_string.h [!_GLIBCXX_USE_CXX11_ABI]
(basic_string::reference, basic_string::const_reference): Define
as plain references for C++11 and later.
(basic_string::basic_string()): Put constructor body outside
preprocessor conditional groups.
(basic_string::basic_string(basic_string&&)): Move _Alloc_hider
instead of copying it.
(basic_string::basic_string(const basic_string&, const _Alloc&)):
Define.
(basic_string::basic_string(basic_string&&, const _Alloc&)):
Define.
* include/bits/fs_path.h [!_GLIBCXX_USE_CXX11_ABI]: Remove special
cases for old basic_string.
* testsuite/21_strings/basic_string/cons/char/8.cc: Test
allocator-extended constructors unconditionally. Add extra members to
allocator type when using old string ABI.
* testsuite/21_strings/basic_string/allocator/71964.cc: Enable test
for old string ABI.
* testsuite/21_strings/basic_string/cons/wchar_t/8.cc: Likewise.

From-SVN: r267584

PR c/88363 - alloc_align attribute doesn't accept enumerated arguments

gcc/c-family/ChangeLog:

PR c/88363
* c-attribs.c (positional_argument): Also accept enumerated types.

gcc/testsuite/ChangeLog:

PR c/88363
* c-c++-common/attributes-4.c: New test.

gcc/ChangeLog:

PR c/88363
* doc/extend.texi (attribute alloc_align, alloc_size): Update.

From-SVN: r267583

Fix test failure when -fno-inline is used

This currently checks _GLIBCXX_USE_DUAL_ABI which is incorrect, as that
can be true when _GLIBCXX_USE_CXX11_ABI == 0. The correct check would be
_GLIBCXX_USE_CXX11_ABI == 1, but that's made redundant by the cxx11-abi
effective target that the test requires. However, the test will fail if
-fno-inline is used, so check __NO_INLINE__ instead.

* testsuite/23_containers/list/61347.cc: Avoid spurious failure when
-fno-inline added to test flags.

From-SVN: r267582

gdbinit.in: Turn off pagination for the skip commands, restore it to previous state afterwards.

* gdbinit.in: Turn off pagination for the skip commands, restore
it to previous state afterwards.

From-SVN: r267581

[PATCH][GCC][Aarch64] Change expected bfxil count in gcc.target/aarch64/combine_bfxil.c to 18 (PR/87763)

gcc/testsuite/Changelog:

2019-01-04  Sam Tebbs  <sam.tebbs@arm.com>

PR gcc/87763
* gcc.target/aarch64/combine_bfxil.c: Change scan-assembler-times bfxil
count to 18.

From-SVN: r267579

Remove XFAIL from test that no longer fails

This test started passing with the old ABI with r263808, so doesn't need
to be marked XFAIL now.

* testsuite/21_strings/basic_string/requirements/
explicit_instantiation/debug.cc: Remove XFAIL for old ABI.

From-SVN: r267578

Fix test failure with old Copy-On-Write std::string

* testsuite/27_io/filesystem/filesystem_error/copy.cc: Fix static
assertion failures with old std::string ABI.

From-SVN: r267577

Fix bugs in filesystem::path::lexically_normal()

Using path::_List::erase(const_iterator) to remove a non-final component
in path::lexically_normal() is a bug, because it leaves the following
component with an incorrect _M_pos value.

Instead of providing erase members that allow removing components from
the middle, replace them with pop_back() and
_M_erase_from(const_iterator) which only allow removing elements at the
end. Most uses of erase are unaffected, because they only remove
elements from the end anyway. The one use of erasure from the middle in
lexically_normal() is replaced by calls to pop_back() and/or clearing
the last component to leave it as an empty final filename.

Also replace the "???" comment in lexically_normal() to document when
that branch is taken.

* include/bits/fs_path.h (path::_List::erase): Replace both overloads
with ...
(path::pop_back(), path::_M_erase_from(const_iterator)): New member
functions that will only erase elements at the end.
* src/filesystem/std-path.cc (path::_List::_Impl::pop_back()): Define.
(path::_List::_Impl::_M_erase_from(const_iterator)): Define.
(path::_List::operator=(const _List&)): Use _M_erase_from(p) instead
of erase(p, end()).
(path::_List::pop_back()): Define.
(path::_List::_M_erase_from(const_iterator)): Define.
(path::operator/=(const path&)): Use pop_back to remove last component
and _M_erase_from to remove multiple components.
(path::_M_append(basic_string_view<value_type>)): Likewise.
(path::operator+=(const path&)): Likewise.
(path::_M_concat(basic_string_view<value_type>)): Likewise.
(path::remove_filename()): Likewise.
(path::lexically_normal()): Use _List::_Impl iterators instead of
path::iterator. Use pop_back to remove components from the end. Clear
trailing filename, instead of using erase(const_iterator) to remove
a non-final component.
* testsuite/27_io/filesystem/path/generation/normal.cc: Test
additional cases.
* testsuite/27_io/filesystem/path/generation/normal2.cc: New test.

From-SVN: r267576

* gnatvsn.ads: Bump copyright year.

From-SVN: r267575

Fix concatenation bug in filesystem::path

When erasing a trailing empty filename component, the output iterator
was not decremented, causing the next component to be created at the
wrong position.

* src/filesystem/std-path.cc (path::operator+=(const path&)): Fix
incorrect treatment of empty filename after trailing slash.
* testsuite/27_io/filesystem/path/concat/path.cc: Test problem case.

From-SVN: r267574

Avoid spurious test failures when -fno-inline in test flags

These tests rely on inlining, so if -fno-inline is added to the compiler
flags the tests fail. Use the predefined __NO_INLINE__ macro to detect
that situation, and don't bother testing the move assignment.

* testsuite/21_strings/basic_string/modifiers/assign/char/
move_assign_optim.cc: Avoid spurious failure when -fno-inline added
to test flags.
* testsuite/21_strings/basic_string/modifiers/assign/wchar_t/
move_assign_optim.cc: Likewise.

From-SVN: r267573

re PR fortran/48543 (Collapse identical strings)

2019-01-04  Thomas Koenig  <tkoenig@gcc.gnu.org>

    PR fortran/48543
    * gfortran.dg/const_chararacter_merge.f90: Actually remove.
    * gfortran.dg/merge_char_const.f90: Restore.

From-SVN: r267572

re PR target/88594 (ICE in int_mode_for_mode, at stor-layout.c:403)

PR target/88594
* config/i386/i386.c (ix86_expand_divmod_libfunc): Use mode instead
of GET_MODE (opN) as modes of the libcall arguments.

* gcc.dg/pr88594.c: New test.

From-SVN: r267571

x86: relax mask register constraints

While their use for masking is indeed restricted to %k1...%k7, use as
"normal" insn operands also permits %k0. Remove the unnecessary
limitations, requiring quite a few testsuite adjustments.

Oddly enough some AVX512{F,DQ} test cases already check for %k[0-7],
while others did permit {%k0} - where they get touched here anyway this
gets fixed at the same time.

From-SVN: r267570

PR tree-optimization/88659 - ICE in maybe_warn_nonstring_arg

gcc/ChangeLog:
* calls.c (maybe_warn_nonstring_arg): Avoid assuming maxlen is set.

gcc/testsuite/ChangeLog:
* gcc.dg/Wstringop-truncation-6.c: New test.

From-SVN: r267569

Daily bump.

From-SVN: r267568

PR libstdc++/88607 replace some more UTF-8 characters

2019-01-03  Jonathan Wakely  <jwakely@redhat.com>
    Jakub Jelinek  <jakub@redhat.com>

PR libstdc++/88607
* include/experimental/memory: Replace UTF-8 quote characters.
* include/std/future: Replace UTF-8 "em dash" characters.

Co-Authored-By: Jakub Jelinek <jakub@redhat.com>
From-SVN: r267565

PR libstdc++/88607 replace or remove unnecessary UTF-8 characters

There are a number of UTF-8 characters in comments which add no value
and can be replaced with ASCII equivalents, or removed entirely for the
section sign (U+00A7).

PR libstdc++/88607
* include/bits/forward_list.h: Replace UTF-8 "ligature fi" character.
* include/debug/forward_list: Likewise.
* include/experimental/bits/shared_ptr.h: Remove UTF-8 "section sign"
character.
* include/experimental/chrono: Likewise.
* include/experimental/functional: Likewise.
* include/experimental/ratio: Likewise.
* include/experimental/system_error: Likewise.
* include/experimental/tuple: Likewise.
* include/experimental/type_traits: Likewise.
* include/parallel/workstealing.h: Replace UTF-8 "en dash" character.
* include/parallel/multiseq_selection.h: Likewise.

From-SVN: r267564

PR libstdc++/88681 export missing symbols

These new facet functions were added to GCC 5.1 but the versions for the
old std::string ABI were never exported from the shared library.

PR libstdc++/88681
* config/abi/pre/gnu.ver: Add missing exports.
* testsuite/22_locale/collate_byname/88681.cc: New test.
* testsuite/22_locale/time_get/get/char/88681.cc: New test.
* testsuite/22_locale/time_get/get/wchar_t/88681.cc: New test.

From-SVN: r267563

rs6000-string.c (expand_block_move): Don't use unaligned vsx and avoid lxvd2x/stxvd2x.

2019-01-03  Aaron Sawdey  <acsawdey@linux.ibm.com>

* config/rs6000/rs6000-string.c (expand_block_move): Don't use
unaligned vsx and avoid lxvd2x/stxvd2x.
(gen_lvx_v4si_move): New function.

From-SVN: r267562

[nvptx] Add vector_length 128 testcases

Add a couple of test-cases using vector length 128, while checking that we
override to vector length 32.

2019-01-03  Tom de Vries  <tdevries@suse.de>

* testsuite/libgomp.oacc-c-c++-common/vector-length-128-1.c: New test.
* testsuite/libgomp.oacc-c-c++-common/vector-length-128-3.c: New test.

From-SVN: r267559

[nvptx] Add nvptx_mach_vector_length, nvptx_mach_max_workers

The vector length and maximum number of workers are known compile-time.  Make
these easily available during code generation via new functions.

2019-01-03  Tom de Vries  <tdevries@suse.de>

* config/nvptx/nvptx.c (MACH_VECTOR_LENGTH, MACH_MAX_WORKERS): Define.
(init_axis_dim, nvptx_mach_max_workers, nvptx_mach_vector_length): New
function.
* config/nvptx/nvptx.h (struct machine_function): Add axis_dims.

From-SVN: r267558

[nvptx] Factor out populate_offload_attrs

Factor out populate_offload_attrs from nvptx_reorg.

2019-01-03  Tom de Vries  <tdevries@suse.de>

* config/nvptx/nvptx.c (struct offload_attrs): New.
(populate_offload_attrs): New function.  Factor mask extraction out of
nvptx_reorg.  Add extraction of dimensions.
(nvptx_reorg): Use populate_offload_attrs.

From-SVN: r267557

[nvptx] Add early-out cases in nvptx_goacc_validate_dims

Add early-out cases for for oacc_min_dims_p and routine_p in
nvptx_goacc_validate_dims, allowing simplification of the rest of the
function.

2019-01-03  Tom de Vries  <tdevries@suse.de>

* config/nvptx/nvptx.c (nvptx_goacc_validate_dims_1): Add early-out
cases for oacc_min_dims_p and routine_p.  Add asserts for
oacc_default_dims_p and offload_region_p.

From-SVN: r267556

[nvptx] Eliminate changed local var in nvptx_goacc_validate_dims

The TARGET_GOACC_VALIDATE_DIMS hook requires an implementation to return a bool
indicating whether the dims parameter has changed.

Factor nvptx_goacc_validate_dims_1 out of nvptx_goacc_validate_dims, and
calculate the return value in nvptx_goacc_validate_dims.

2019-01-03  Tom de Vries  <tdevries@suse.de>

* config/nvptx/nvptx.c (nvptx_goacc_validate_dims_1): New function,
factored out of ...
(nvptx_goacc_validate_dims): ... here.

From-SVN: r267555

re PR fortran/48543 (Collapse identical strings)

2019-01-02  Thomas Koenig  <tkoenig@gcc.gnu.org>

    PR fortran/48543
    * gfortran.dg/const_chararacter_merge.f90: Remove.

From-SVN: r267553

re PR lto/85574 (LTO bootstapped binaries differ)

2019-01-03  Jan Hubicka  <hubicka@ucw.cz>

PR tree-optimization/85574
* tree-ssa-uncprop.c (struct equiv_hash_elt): Remove unused
structure.
(struct ssa_equip_hash_traits): Declare.
(val_ssa_equiv): Use custom hash traits using operand_equal_p.

From-SVN: r267552

Update config.guess, config.sub (PR target/88535)

PR target/88535
* config.guess: Import upstream version 2019-01-03.
* config.sub: Import upstream version 2019-01-01.

From-SVN: r267551

re PR debug/88644 (Unexpected pub type info eliminated after r246973 (causes pubtypes-*.c to regress).)

PR debug/88644
* dwarf2out.c (modified_type_die): If type is equal to sizetype,
change it to qualified_type.

* gcc.dg/debug/dwarf2/pr88644.c: New test.
* gcc.dg/debug/dwarf2/pr80263.c: Remove darwin hack.

2019-01-03  Iain Sandoe  <iain@sandoe.co.uk>

* gcc.dg/pubtypes-2.c: Adjust expected pubtypes length.
* gcc.dg/pubtypes-3.c: Likewise.
* gcc.dg/pubtypes-4.c: Likewise.

From-SVN: r267550

ipa-utils.c (scale_ipa_profile_for_fn): Break out from ...

* ipa-utils.c (scale_ipa_profile_for_fn): Break out from ...
(ipa_merge_profiles): ... here; do not ICE on thunks and aliases.

From-SVN: r267549

re PR c++/88636 (ICE: Segmentation fault (in c_tree_chain_next))

PR c++/88636
* decl.c (builtin_function_1): Return result of pushdecl_top_level
or pushdecl rather than decl.

* g++.target/i386/pr88636.C: New test.

From-SVN: r267548

tree.c (handle_nodiscard_attribute): Improve warning location.

/cp
2019-01-03  Paolo Carlini  <paolo.carlini@oracle.com>

* tree.c (handle_nodiscard_attribute): Improve warning location.

/testsuite
2019-01-03  Paolo Carlini  <paolo.carlini@oracle.com>

* g++.dg/cpp1z/nodiscard3.C: Test locations too.

From-SVN: r267547

Adjust a dump file in a test-case (PR testsuite/88436).

2019-01-03  Martin Liska  <mliska@suse.cz>

PR testsuite/88436
* gcc.target/powerpc/pr54240.c: Scan phiopt2.

From-SVN: r267541

Daily bump.

From-SVN: r267540

gimple-fold.c (strlen_range_kind): Remove SRK_LENRANGE_2.

* gimple-fold.c (strlen_range_kind): Remove SRK_LENRANGE_2.
(get_range_strlen_tree): Update appropriately.
(get_range_strlen)
* gimple-fold.h (get_range_strlen): Drop unused last argument.

Co-Authored-By: Jeff Law <law@redhat.com>
From-SVN: r267537

PR c++/81486 - CTAD failing with ().

* g++.dg/cpp1z/class-deduction60.C: New test.
* g++.dg/cpp1z/class-deduction61.C: New test.

From-SVN: r267535

PR c++/88631 - CTAD failing for value-initialization.

* typeck2.c (build_functional_cast): Try deducing the template
arguments even if there are no arguments to deduce from.

* g++.dg/cpp1z/class-deduction59.C: New test.

From-SVN: r267533

PR c++/88612 - ICE with -Waddress-of-packed-member.

* call.c (convert_for_arg_passing): Only give warnings with tf_warning.
* typeck.c (convert_for_assignment): Likewise.

* g++.dg/warn/Waddress-of-packed-member1.C: New test.

From-SVN: r267532

gimple-fold.c (gimple_fold_builtin_strlen): Use set_strlen_range rather than set_range_info.

* gimple-fold.c (gimple_fold_builtin_strlen): Use set_strlen_range
rather than set_range_info.
* tree-ssa-strlen.c (set_strlen_range): Extracted from
maybe_set_strlen_range.  Handle potentially boundary crossing
cases more conservatively.
(maybe_set_strlen_range): Parts refactored into set_strlen_range.
Call set_strlen_range.
* tree-ssa-strlen.h (set_strlen_range): Add prototype.

* gcc.dg/strlenopt-36.c: Update.
* gcc.dg/strlenopt-45.c: Update.
* gcc.c-torture/execute/strlen-5.c: New test.
* gcc.c-torture/execute/strlen-6.c: New test.
* gcc.c-torture/execute/strlen-7.c: New test.

Co-Authored-By: Jeff Law <law@redhat.com>
From-SVN: r267531

re PR testsuite/87304 (gcc.dg/vect/bb-slp-over-widen-1.c fails starting with r262371)

PR testsuite/87304
* gcc.dg/vect/bb-slp-over-widen-1.c: Expect basic block vectorized
messages only on vect_hw_misalign targets.

From-SVN: r267528

parse.c (decode_statement): Suppress "Unclassifiable statement" error if previous error messages were emittes.

2019-01-01  Steven G. Kargl  <kargl@gcc.gnu.org>

* parse.c (decode_statement):  Suppress "Unclassifiable statement"
error if previous error messages were emittes.

2019-01-01  Steven G. Kargl  <kargl@gcc.gnu.org>

* gfortran.dg/argument_checking_7.f90: Remove run-on error message.
* gfortran.dg/dec_d_lines_3.f: Ditto.
* gfortran.dg/dec_structure_24.f90: Ditto.
* gfortran.dg/dec_structure_26.f90: Ditto.
* gfortran.dg/dec_structure_27.f90: Ditto.
* gfortran.dg/dec_type_print_3.f90: Ditto.
* gfortran.dg/derived_name_1.f90: Ditto.
* gfortran.dg/error_recovery_1.f90: Ditto.
* gfortran.dg/gomp/pr29759.f90: Ditto.
* gfortran.dg/pr36192.f90: Ditto.
* gfortran.dg/pr56007.f90: Ditto.
* gfortran.dg/pr56520.f90: Ditto.
* gfortran.dg/pr78741.f90: Ditto.
* gfortran.dg/print_fmt_2.f90: Ditto.
* gfortran.dg/select_type_20.f90: Ditto.

From-SVN: r267526

* es.po: Update.

From-SVN: r267521

re PR middle-end/88663 (internal compiler error: in check, at tree-vrp.c:188)

PR middle-end/88663
* gimple-fold.c (get_range_strlen): Update prototype to no longer
need the flexp argument.
(get_range_strlen_tree): Drop flexp argument.  Drop flexp argument
from calls to get_range_strlen.  Update comments.  Just update
VAL for an unterminated const char array and let the reset of the
code handle it normally.  No longer try to set *flexp.  Adjust
return value.
(get_range_strlen): Update for the new get_range_strlen API.
(get_maxval_strlen): Similarly.
(gimple_fold_builtin_strlen): Handle update meaning of return value
from get_range_strlen.
* gimple-ssa-sprintf.c (get_string_length): Update for the new
get_range_strlen API.

Co-Authored-By: Jeff Law <law@redhat.com>
From-SVN: r267520

re PR c++/86875 (ICE in tsubst_copy, at cp/pt.c:15478)

PR c++/86875
* g++.dg/cpp1y/lambda-generic-86875.C: New test.

From-SVN: r267519

Add more testcases for class template argument deduction of maps

This adds additional tests for std::map and std::multimap CTAD. The
tests ensure that deduction works for braced-init-list of value_type
objects, and for pairs of input iterators (with both std::pair<Key, T>
and value_type as the iterator's value_type). This ensures deduction
from value_type still works, as well as the non-value_type cases in LWG
3025.

Similar tests for unordered maps do not work, apparently because the
constructor taking an initializer_list<value_type> is not usable for
deduction, and the deduction guide taking initializer_list<pair<Key, T>>
deduces key_type to be const. I am not addressing that.

* testsuite/23_containers/map/cons/deduction.cc: Test deduction from
initializer_list<value_type> and from input iterator ranges.
* testsuite/23_containers/multimap/cons/deduction.cc: Likewise.

From-SVN: r267518

re PR fortran/48543 (Collapse identical strings)

2019-01-02  Thomas Koenig  <tkoenig@gcc.gnu.org>

PR fortran/48543
* gfortran.dg/const_chararacter_merge.f90: New test.

From-SVN: r267517

re PR lto/88130 (ICE in copy_function_or_variable, at lto-streamer-out.c:2315 since r260963)

PR lto/88130
* varpool.c (varpool_node::ctor_useable_for_folding_p): Also return
false at WPA time when body was removed.
* g++.dg/torture/pr88130.C: New testcase.

From-SVN: r267515

Calculate prediction remainder at proper place (PR tree-optimization/88650).

2019-01-02  Martin Liska  <mliska@suse.cz>

PR tree-optimization/88650
* predict.c (set_even_probabilities): Calculate probability
remainer only when really used.
2019-01-02  Martin Liska  <mliska@suse.cz>

PR tree-optimization/88650
* gfortran.dg/predict-3.f90: New test.

From-SVN: r267513