Daily bump.

From-SVN: r259426

[PR c++/85039] no type definitions in builtin offsetof

Types defined within a __builtin_offsetof argument don't always get
properly recorded as members of their context types, so if they're
anonymous, we may fail to assign them an anon type index for mangling
and ICE.

We shouldn't allow types to be introduced in __builtin_offsetof, I
think, and Jason says the std committee agrees, so I've arranged for
us to reject them.

Even then, we still parse the definitions and attempt to assign
mangled names to its member functions, so the ICE remains.  Since
we've already reported an error, we might as well complete the name
assignment with an arbitrary index, thus avoiding the ICE.

We used to have a test that expected to be able to define types in
__builtin_offsetof; this patch removes that specific test.

for  gcc/cp/ChangeLog

PR c++/85039
* parser.c (cp_parser_builtin_offset): Reject type definitions.
* mangle.c (nested_anon_class_index): Avoid crash returning -1
if we've seen errors.

for  gcc/testsuite/ChangeLog

PR c++/85039
* g++.dg/pr85039-1.C: New.
* g++.dg/pr85039-2.C: New.
* g++.dg/parse/semicolon3.C: Remove test_offsetof.

From-SVN: r259423

re PR target/85080 (gcc.dg/vect/costmodel/ppc/costmodel-pr37194.c fails starting with r248678)

[gcc/testsuite]

2018-04-16  Bill Schmidt  <wschmidt@linux.ibm.com>

PR target/85080
* gcc.dg/vect/costmodel/ppc/costmodel-pr37194.c: Skip dump checks
if the target supports efficient unaligned storage accesses.

From-SVN: r259407

[openacc] Fix ICE when compiling tile loop containing infinite loop

2018-04-16  Cesar Philippidis  <cesar@codesourcery.com>
    Tom de Vries  <tom@codesourcery.com>

PR middle-end/84955
* omp-expand.c (expand_oacc_for): Add dummy false branch for
tiled basic blocks without omp continue statements.

* testsuite/libgomp.oacc-c-c++-common/pr84955.c: New test.
* testsuite/libgomp.oacc-fortran/pr84955.f90: New test.

Co-Authored-By: Tom de Vries <tom@codesourcery.com>
From-SVN: r259406

re PR target/83660 (ICE with vec_extract inside expression statement)

2018-04-16  Aaron Sawdey  <acsawdey@linux.ibm.com>

PR target/83660
* config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Mark
vec_extract expression as having side effects to make sure it gets
a cleanup point.

2018-04-16  Aaron Sawdey  <acsawdey@linux.ibm.com>

PR target/83660
* gcc.target/powerpc/pr83660.C: New test.

From-SVN: r259403

i386: Check error_mark_node in multiversioning

Since CET is applied to the whole program, it is correct to disallow
-fcf-protection=full without -mcet.  But compiler shouldn't crash.

gcc/

PR target/85403
* config/i386/i386.c (get_builtin_code_for_version): Check
error_mark_node.

gcc/testsuite/

PR target/85403
* gcc.target/i386/pr85403.c: New test.
---
 gcc/config/i386/i386.c                            |  2 ++
 gcc/testsuite/g++.dg/ext/mv1.C                    |  2 +-
 gcc/testsuite/g++.dg/ext/mv14.C                   |  2 +-
 gcc/testsuite/g++.dg/ext/mv15.C                   |  2 +-
 gcc/testsuite/g++.dg/ext/mv16.C                   |  2 +-
 gcc/testsuite/g++.dg/ext/mv17.C                   |  2 +-
 gcc/testsuite/g++.dg/ext/mv18.C                   |  2 +-
 gcc/testsuite/g++.dg/ext/mv19.C                   |  2 +-
 gcc/testsuite/g++.dg/ext/mv20.C                   |  2 +-
 gcc/testsuite/g++.dg/ext/mv21.C                   |  2 +-
 gcc/testsuite/g++.dg/ext/mv22.C                   |  2 +-
 gcc/testsuite/g++.dg/ext/mv23.C                   |  2 +-
 gcc/testsuite/g++.dg/ext/mv26.C                   |  1 +
 gcc/testsuite/g++.dg/ext/mv6.C                    |  2 +-
 gcc/testsuite/g++.dg/ext/mvc1.C                   |  1 +
 gcc/testsuite/gcc.target/i386/cet-notrack-icf-1.c |  2 +-
 gcc/testsuite/gcc.target/i386/cet-notrack-icf-3.c |  2 +-
 gcc/testsuite/gcc.target/i386/cet-property-2.c    |  2 +-
 gcc/testsuite/gcc.target/i386/mvc1.c              |  1 +
 gcc/testsuite/gcc.target/i386/mvc10.c             |  1 +
 gcc/testsuite/gcc.target/i386/mvc11.c             |  2 +-
 gcc/testsuite/gcc.target/i386/mvc6.c              |  2 +-
 gcc/testsuite/gcc.target/i386/mvc7.c              |  1 +
 gcc/testsuite/gcc.target/i386/mvc8.c              |  2 +-
 gcc/testsuite/gcc.target/i386/mvc9.c              |  2 +-
 gcc/testsuite/gcc.target/i386/pr81213.c           |  1 +
 gcc/testsuite/gcc.target/i386/pr81214.c           |  1 +
 gcc/testsuite/gcc.target/i386/pr85403.c           | 10 ++++++++++
 gcc/testsuite/gcc.target/i386/sse-26.c            |  2 +-
 29 files changed, 39 insertions(+), 20 deletions(-)
 create mode 100644 gcc/testsuite/gcc.target/i386/pr85403.c

diff --git a/gcc/config/i386/i386.c b/gcc/config/i386/i386.c
index 6fa5b0add02..8a73fc0d316 100644
--- a/gcc/config/i386/i386.c
+++ b/gcc/config/i386/i386.c
@@ -32344,6 +32344,8 @@ get_builtin_code_for_version (tree decl, tree *predicate_list)
        &global_options_set);

       gcc_assert (target_node);
+      if (target_node == error_mark_node)
+ return 0;
       new_target = TREE_TARGET_OPTION (target_node);
       gcc_assert (new_target);

diff --git a/gcc/testsuite/gcc.target/i386/pr85403.c b/gcc/testsuite/gcc.target/i386/pr85403.c
new file mode 100644
index 00000000000..f4fb12dd4e2
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/pr85403.c
@@ -0,0 +1,10 @@
+/* { dg-do compile } */
+/* { dg-options "-fcf-protection -mcet" } */
+/* { dg-require-ifunc "" } */
+
+__attribute__((target_clones("avx","arch=slm","arch=core-avx2","default")))
+int
+foo ()
+{
+  return -2;
+} /* { dg-error "requires Intel CET support" } */

From-SVN: r259400

Fixed g++.dg/ext/mv16.C with -march=native.

gcc/
PR target/84331
* gcc/config.gcc: Support "skylake".
* gcc/config/i386/i386-c.c (ix86_target_macros_internal): Handle
PROCESSOR_SKYLAKE.
* gcc/config/i386/i386.c (m_SKYLAKE): Define.
(processor_target_table): Add "skylake".
(ix86_option_override_internal): Add "skylake".
(get_builtin_code_for_version): Handle PROCESSOR_SKYLAKE,
PROCESSOR_CANNONLAKE.
(get_builtin_code_for_version): Fix priority for
PROCESSOR_ICELAKE_CLIENT, PROCESSOR_ICELAKE_SERVER,
PROCESSOR_SKYLAKE-AVX512.
* gcc/config/i386/i386.h (processor_costs): Define TARGET_SKYLAKE.
(processor_type): Add PROCESSOR_SKYLAKE.

gcc/testsuite/
PR target/84331
* gcc/testsuite/gcc.target/i386/funcspec-56.inc: Test arch=skylake.

From-SVN: r259399

re PR target/84945 (UBSAN: gcc/config/i386/i386.c:33312:22: runtime error: shift exponent 32 is too large for 32-bit type 'int')

PR target/84945
* config/i386/cpuinfo.c (set_feature): Wrap into do while (0) to avoid
-Wdangling-else warnings.  Mask shift counts to avoid
-Wshift-count-negative and -Wshift-count-overflow false positives.

From-SVN: r259398

re PR c++/85112 (ICE with invalid constexpr)

2018-04-16  Paolo Carlini  <paolo.carlini@oracle.com>
    Jason Merrill  <jason@redhat.com>

PR c++/85112
* convert.c (convert_to_integer_1): Use direct recursion for
enumeral types and types with a precision less than the number
of bits in their mode.

/testsuite
2018-04-16  Paolo Carlini  <paolo.carlini@oracle.com>
    Jason Merrill  <jason@redhat.com>

PR c++/85112
* g++.dg/cpp0x/pr85112.C: New.

Co-Authored-By: Jason Merrill <jason@redhat.com>
From-SVN: r259397

Make Filesystem TS tests pass in C++17 mode

The <testsuite_fs.h> header defaults to using std::filesystem in C++17
mode. The Filesystem TS tests need to define the macro that causes
std::experimental::filesystem to be used instead.

* testsuite/experimental/filesystem/file_status/1.cc: Add
-DUSE_FILESYSTEM_TS to dg-options.
* testsuite/experimental/filesystem/iterators/directory_iterator.cc:
Likewise.
* testsuite/experimental/filesystem/iterators/pop.cc: Likewise.
* testsuite/experimental/filesystem/iterators/
recursive_directory_iterator.cc: Likewise.
* testsuite/experimental/filesystem/operations/absolute.cc: Likewise.
* testsuite/experimental/filesystem/operations/canonical.cc: Likewise.
* testsuite/experimental/filesystem/operations/copy.cc: Likewise.
* testsuite/experimental/filesystem/operations/copy_file.cc: Likewise.
* testsuite/experimental/filesystem/operations/create_directories.cc:
Likewise.
* testsuite/experimental/filesystem/operations/create_directory.cc:
Likewise.
* testsuite/experimental/filesystem/operations/create_symlink.cc:
Likewise.
* testsuite/experimental/filesystem/operations/current_path.cc:
Likewise.
* testsuite/experimental/filesystem/operations/equivalent.cc: Likewise.
* testsuite/experimental/filesystem/operations/exists.cc: Likewise.
* testsuite/experimental/filesystem/operations/file_size.cc: Likewise.
* testsuite/experimental/filesystem/operations/is_empty.cc: Likewise.
* testsuite/experimental/filesystem/operations/last_write_time.cc:
Likewise.
* testsuite/experimental/filesystem/operations/permissions.cc:
Likewise.
* testsuite/experimental/filesystem/operations/read_symlink.cc:
Likewise.
* testsuite/experimental/filesystem/operations/remove.cc: Likewise.
* testsuite/experimental/filesystem/operations/remove_all.cc: Likewise.
* testsuite/experimental/filesystem/operations/status.cc: Likewise.
* testsuite/experimental/filesystem/operations/temp_directory_path.cc:
Likewise.
* testsuite/experimental/filesystem/path/append/path.cc: Likewise.
* testsuite/experimental/filesystem/path/assign/assign.cc: Likewise.
* testsuite/experimental/filesystem/path/assign/copy.cc: Likewise.
* testsuite/experimental/filesystem/path/compare/compare.cc: Likewise.
* testsuite/experimental/filesystem/path/compare/path.cc: Likewise.
* testsuite/experimental/filesystem/path/compare/strings.cc: Likewise.
* testsuite/experimental/filesystem/path/concat/path.cc: Likewise.
* testsuite/experimental/filesystem/path/concat/strings.cc: Likewise.
* testsuite/experimental/filesystem/path/construct/copy.cc: Likewise.
* testsuite/experimental/filesystem/path/construct/default.cc:
Likewise.
* testsuite/experimental/filesystem/path/construct/locale.cc: Likewise.
* testsuite/experimental/filesystem/path/construct/range.cc: Likewise.
* testsuite/experimental/filesystem/path/construct/string_view.cc:
Likewise.
* testsuite/experimental/filesystem/path/decompose/extension.cc:
Likewise.
* testsuite/experimental/filesystem/path/decompose/filename.cc:
Likewise.
* testsuite/experimental/filesystem/path/decompose/parent_path.cc:
Likewise.
* testsuite/experimental/filesystem/path/decompose/relative_path.cc:
Likewise.
* testsuite/experimental/filesystem/path/decompose/root_directory.cc:
Likewise.
* testsuite/experimental/filesystem/path/decompose/root_name.cc:
Likewise.
* testsuite/experimental/filesystem/path/decompose/root_path.cc:
Likewise.
* testsuite/experimental/filesystem/path/decompose/stem.cc: Likewise.
* testsuite/experimental/filesystem/path/generic/generic_string.cc:
Likewise.
* testsuite/experimental/filesystem/path/itr/traversal.cc: Likewise.
* testsuite/experimental/filesystem/path/modifiers/clear.cc: Likewise.
* testsuite/experimental/filesystem/path/modifiers/make_preferred.cc:
Likewise.
* testsuite/experimental/filesystem/path/modifiers/remove_filename.cc:
Likewise.
* testsuite/experimental/filesystem/path/modifiers/replace_extension.cc:
Likewise.
* testsuite/experimental/filesystem/path/modifiers/replace_filename.cc:
Likewise.
* testsuite/experimental/filesystem/path/modifiers/swap.cc: Likewise.
* testsuite/experimental/filesystem/path/native/string.cc: Likewise.
* testsuite/experimental/filesystem/path/nonmember/hash_value.cc:
Likewise.
* testsuite/experimental/filesystem/path/query/empty.cc: Likewise.
* testsuite/experimental/filesystem/path/query/has_extension.cc:
Likewise.
* testsuite/experimental/filesystem/path/query/has_filename.cc:
Likewise.
* testsuite/experimental/filesystem/path/query/has_parent_path.cc:
Likewise.
* testsuite/experimental/filesystem/path/query/has_relative_path.cc:
Likewise.
* testsuite/experimental/filesystem/path/query/has_root_directory.cc:
Likewise.
* testsuite/experimental/filesystem/path/query/has_root_name.cc:
Likewise.
* testsuite/experimental/filesystem/path/query/has_root_path.cc:
Likewise.
* testsuite/experimental/filesystem/path/query/has_stem.cc: Likewise.
* testsuite/experimental/filesystem/path/query/is_relative.cc:
Likewise.

From-SVN: r259396

Add sse_unaligned_load_optimal and sse_unaligned_store_optimal to Skylake.

gcc/
PR target/84413
* config/i386/x86-tune.def (X86_TUNE_SSE_UNALIGNED_LOAD_OPTIMAL,
X86_TUNE_SSE_UNALIGNED_STORE_OPTIMAL): Add m_SKYLAKE_AVX512

From-SVN: r259395

re PR testsuite/85326 (`make check` fails with `--disable-bootstrap` and `--enable-languages=c`)

[gcc/testsuite]

2018-04-15  Bill Schmidt  <wschmidt@linux.ibm.com>

PR testsuite/85326
* g++.dg/ext/undef-bool-1.C: New file.
* gcc.target/powerpc/powerpc.exp: Remove .C support.
* gcc.target/powerpc/undef-bool-1.C: Remove file.

From-SVN: r259393

Daily bump.

From-SVN: r259392

Daily bump.

From-SVN: r259389

rs6000: Disable -m[no-]direct-move (PR85293)

The -mno-direct-move option causes a lot of problems, since it forces
us to be able to generate code for p8 and up with some crucial
instructions missing.  This patch removes the -m[no-]direct-move
options so that the user cannot put us into this unexpected situation
anymore.  Internally we still have all the same flags, and they are
automatically set based on -mcpu; getting rid of that is a lot more
work and will have to wait for GCC 9 (in some places the flag is used
to see if we are compiling for a p8 _at all_).

PR target/85293
* config/rs6000/rs6000.opt (mdirect-move): Make deprecated.
* doc/invoke.texi (RS/6000 and PowerPC Options): Remove -mdirect-move
and -mno-direct-move.

gcc/testsuite/
PR target/85293
* gcc.target/powerpc/pr80098-2.c: Remove -mdirect-move.  Remove the
corresponding dg-error clause.
* gcc.target/powerpc/pr80098-3.c: Ditto.
* gcc.target/powerpc/pr80103-1.c: Delete.

From-SVN: r259386

re PR fortran/81773 ([Coarray] Get with vector index on lhs leads to incorrect caf_get_by_ref() call.)

gcc/fortran/ChangeLog:

2018-04-14  Andre Vehreschild  <vehre@gcc.gnu.org>

PR fortran/81773
PR fortran/83606
* dependency.c (gfc_dep_resolver): Coarray indexes are to be ignored
during dependency computation.  They define no data dependency.
* trans-array.c (conv_array_index_offset): The stride can not be set
here, prevent fail.
* trans-intrinsic.c (conv_caf_send): Add creation of temporary array
for caf_get's result and copying to the array with vectorial
indexing.

gcc/testsuite/ChangeLog:

2018-04-14  Andre Vehreschild  <vehre@gcc.gnu.org>

PR fortran/81773
PR fortran/83606
* gfortran.dg/coarray/get_to_indexed_array_1.f90: New test.
* gfortran.dg/coarray/get_to_indirect_array.f90: New test.

From-SVN: r259385

re PR fortran/85387 (incorrect output with optimization /= 0)

2018-04-14  Thomas Koenig  <tkoenig@gcc.gnu.org>

PR fortran/85387
* frontend-passes.c (traverse_io_block): Check for start, end or
stride being defined by an outer implied DO loop.

2018-04-14  Thomas Koenig  <tkoenig@gcc.gnu.org>

PR fortran/85387
* gfortran.dg/implied_do_io_5.f90: New test.

From-SVN: r259384

Daily bump.

From-SVN: r259383

rs6000: Fix _mm_slli_epi{32,64} for shift values 16 through 31 and negative (PR84302)

The powerpc versions of _mm_slli_epi32 and __mm_slli_epi64 in emmintrin.h
do not properly handle shift values between 16 and 31, inclusive.
These are setting up the shift with vec_splat_s32, which only accepts
*5 bit signed* shift values, or a range of -16 to 15.  Values above 15
produce an error:

  error: argument 1 must be a 5-bit signed literal

Fix is to effectively reduce the range for which vec_splat_s32 is used
to < 32 and use vec_splats otherwise.

Also, __mm_slli_epi{16,32,64}, when given a negative shift value,
should always return a vector of {0}.

PR target/83402
* config/rs6000/emmintrin.h (_mm_slli_epi{16,32,64}):
Ensure that vec_splat_s32 is only called with 0 <= shift < 16.
Ensure negative shifts result in {0}.

gcc/testsuite/
PR target/83402
* gcc.target/powerpc/sse2-psllw-1.c: Refactor and add tests for
several values:  positive, negative, and zero.
* gcc.target/powerpc/sse2-pslld-1.c: Same.
* gcc.target/powerpc/sse2-psllq-1.c: Same.

From-SVN: r259380

re PR rtl-optimization/79916 (ICE in Max. number of generated reload insns per insn is achieved (90))

2018-04-13  Vladimir Makarov  <vmakarov@redhat.com>

PR rtl-optimization/79916
* config/rs6000/rs6000.c (rs6000_emit_move): Use assigned hard
regs (if any) to define how to gnerate SD moves when LRA is in
progress.

2018-04-13  Vladimir Makarov  <vmakarov@redhat.com>

PR rtl-optimization/79916
* gcc.target/powerpc/pr79916.c: New.

From-SVN: r259379

re PR rtl-optimization/85393 (Miscompilation with hot/cold partitioning starting with r254832)

PR rtl-optimization/85393
* except.h (expand_dw2_landing_pad_for_region): Remove declaration.
* except.c (expand_dw2_landing_pad_for_region): Make static.
* bb-reorder.c (fix_up_crossing_landing_pad): In new_bb emit just
a label and unconditional jump to old_bb, rather than
expand_dw2_landing_pad_for_region insn(s) and jump to single_succ
basic block.

* g++.dg/opt/pr85393.C: New test.
* g++.dg/opt/pr85393-aux.cc: New file.

From-SVN: r259378

re PR rtl-optimization/85376 (wrong code with -Og -fno-dce -fgcse -fno-tree-ccp -fno-tree-copy-prop)

PR rtl-optimization/85376
* simplify-rtx.c (simplify_const_unary_operation): For CLZ and CTZ and
zero op0, if C?Z_DEFINED_VALUE_AT_ZERO is false, return NULL_RTX
instead of a specific value.

* gcc.dg/pr85376.c: New test.

From-SVN: r259377

Fix broken sed command from previous commit

* src/c++11/Makefile.am: Fix sed command.
* src/c++11/Makefile.in: Regenerate.

From-SVN: r259374

re PR rtl-optimization/83852 (ICE in sel_redirect_edge_and_branch, at sel-sched-ir.c:5644 on 32-bit BE powerpc targets)

        PR rtl-optimization/83852
        * gcc.dg/pr83852.c: New testcase.

From-SVN: r259373

Fix __iosfail_type_info hack to work on darwin

* src/c++11/Makefile.am: Rewrite sed rule to be less fragile and to
handle mangled names starting with double underscores on darwin.
* src/c++11/Makefile.in: Regenerate.

From-SVN: r259371

IBM Z: Get rid of target specific C++ testcase

gcc/testsuite/ChangeLog:

2018-04-13  Andreas Krebbel  <krebbel@linux.ibm.com>

        PR testsuite/85326
        * gcc.target/s390/pr77822-1.C: Rename to ...
        * gcc.target/s390/pr77822-1.c: ... this. Add asm scan check.
        * gcc.target/s390/pr77822-2.c: Add asm scan check.
        * gcc.target/s390/s390.exp: Remove C from testcase regexps.

From-SVN: r259369

re PR tree-optimization/82965 (gcc.dg/vect/pr79347.c starts failing after r254379)

PR tree-optimization/82965
PR tree-optimization/83991
* cfgloopanal.c (expected_loop_iterations_unbounded): Add
by_profile_only parameter.
* cfgloopmanip.c (scale_loop_profile): Further scale loop's profile
        information if the loop was predicted to iterate too many times.
* cfgloop.h (expected_loop_iterations_unbounded): Update prototype

Co-Authored-By: Bin Cheng <bin.cheng@arm.com>
From-SVN: r259368

re PR target/71991 (Inconsistency for __attribute__ ((__always_inline__)) among LTO and non-LTO compilation)

PR lto/71991
* config/i386/i386.c (ix86_can_inline_p): Allow safe transitions for
always inline.
* gcc.target/i386/pr71991.c: New testcase.

From-SVN: r259367

re PR middle-end/81657 (FAIL: gcc.dg/20050503-1.c scan-assembler-not call)

PR middle-end/81657
* expr.h (enum block_op_methods): Add BLOCK_OP_NO_LIBCALL_RET.
* expr.c (emit_block_move_hints): Handle BLOCK_OP_NO_LIBCALL_RET.
* builtins.c (expand_builtin_memory_copy_args): Use
BLOCK_OP_NO_LIBCALL_RET method for mempcpy with non-ignored target,
handle dest_addr == pc_rtx.

* gcc.dg/string-opt-1.c: Remove bogus comment.  Expect a mempcpy
call.

Co-Authored-By: Jakub Jelinek <jakub@redhat.com>
From-SVN: r259366

Fix gcc.dg/debug/pr41893-1.c with Solaris ld (PR lto/81968)

PR lto/81968
* simple-object.c (handle_lto_debug_sections): Keep .comment
section.

From-SVN: r259364

Daily bump.

From-SVN: r259363

Prevent erroneous "macro had not yet been defined" messages (PR c++/85385)

PR c++/85385 reports an issue where we emit bogus "macro had not yet been
defined" notes when a macro is mis-used:

  $ cat test.c
  #define MACRO(X,Y)

  void test ()
  {
    MACRO(42);
  }

  $ ./xg++ -B. -c test.c
  test.c:5:11: error: macro "MACRO" requires 2 arguments, but only 1 given
     MACRO(42);
             ^
  test.c: In function ‘void test()’:
  test.c:5:3: error: ‘MACRO’ was not declared in this scope
     MACRO(42);
     ^~~~~
  test.c:5:3: note:
  test.c:1: note: it was later defined here
   #define MACRO(X,Y)

The macro *had* been defined, it was merely misused.

This patch fixes the issue by only issuing the note if the use location
is before the definition location (using linemap_location_before_p).

gcc/cp/ChangeLog:
PR c++/85385
* name-lookup.c (macro_use_before_def::maybe_make): New function,
checking that the use is indeed before the definition.
(macro_use_before_def::macro_use_before_def): Make private.
(macro_use_before_def::~macro_use_before_def): Make private.  Move
check for UNKNOWN_LOCATION to macro_use_before_def::maybe_make.
(lookup_name_fuzzy): Call macro_use_before_def::maybe_make rather
than using new directly.

gcc/testsuite/ChangeLog:
PR c++/85385
* g++.dg/diagnostic/macro-arg-count.C: New test.

From-SVN: r259360

re PR fortran/83064 (DO CONCURRENT and auto-parallelization)

2018-04-12  Thomas Koenig  <tkoenig@gcc.gnu.org>

PR fortran/83064
PR testsuite/85346
* trans-stmt.c (gfc_trans_forall_loop): Use annot_expr_ivdep_kind
for annotation and remove dependence on -ftree-parallelize-loops.

2018-04-12  Thomas Koenig  <tkoenig@gcc.gnu.org>

PR fortran/83064
PR testsuite/85346
* gfortran.dg/do_concurrent_5.f90: Dynamically allocate main work
array and move test to libgomp/testsuite/libgomp.fortran.
* gfortran.dg/do_concurrent_6.f90: New test.

2018-04-12  Thomas Koenig  <tkoenig@gcc.gnu.org>

PR fortran/83064
PR testsuite/85346
* testsuite/libgomp.fortran/do_concurrent_5.f90: Move modified
test from gfortran.dg to here.

From-SVN: r259359

Fix comments that misspell names of files and classes

* src/c++11/Makefile.am: Fix comment.
* src/c++11/Makefile.in: Regenerate.
* src/c++11/cxx11-ios_failure.cc: Fix comment.
* src/c++98/ios_failure.cc: Likewise.

From-SVN: r259357

PR c++/85356 - ICE with pointer to member function.

* pt.c (maybe_instantiate_noexcept): Do instantiate in templates if
flag_noexcept_type.  Build the new spec within the function context.
* except.c (build_noexcept_spec): Do get constant value in templates
if flag_noexcept_type.
* decl.c (check_redeclaration_exception_specification): Don't
instantiate noexcept on a dependent declaration.

From-SVN: r259356

re PR c++/85258 (ICE with invalid range-based for-loop)

PR c++/85258
* constexpr.c (reduced_constant_expression_p): Return false for null
trees.

* g++.dg/parse/error61.C: New test.

From-SVN: r259355

rs6000: Fix an ICE with -mno-direct-move (PR85291)

PR target/85291
* config/rs6000/rs6000.md (fix_trunc<mode>si2): Use legacy code if
asked to not generate direct moves.
(fix_trunc<mode>si2_stfiwx): Similar.
(fix_trunc<mode>si2_internal): Similar.

From-SVN: r259354

re PR debug/83157 (gcc.dg/guality/pr41616-1.c fail, inline instances refer to concrete instance as abstract origin)

PR debug/83157
* var-tracking.c (add_stores): Handle STRICT_LOW_PART SET_DEST.
* cselib.c (cselib_record_sets): For STRICT_LOW_PART dest,
lookup if dest in some wider mode is known to be const0_rtx and
if so, record permanent equivalence for it to be ZERO_EXTEND of
the narrower mode destination.

From-SVN: r259353

Revert 259346.

gcc/
* lto-streamer-out.c (output_function): Revert 259346.
* omp-expand.c (expand_oacc_for): Likewise.

libgomp/
* testsuite/libgomp.oacc-c-c++-common/pr84955.c: Revert 259346.
* testsuite/libgomp.oacc-fortran/pr84955.f90:Likewise.

From-SVN: r259351

re PR target/85347 (New testcase vec-ldl-1.c FAILs on powerpc64-linux)

gcc/testsuite/ChangeLog:

2018-04-12  Kelvin Nilsen  <kelvin@gcc.gnu.org>

PR target/85347
* gcc.target/powerpc/vec-ldl-1.c: Change dejagnu directives to
specify -mvsx on gcc command line.

From-SVN: r259350

Remove #define made redundant by r259281

The definition of __throw_ios_failure is no longer in this file, so
setting the macro here is unnecessary.

* src/c++11/ios.cc: Remove redundant macro definition.

From-SVN: r259349

sel-sched: move cleanup_cfg before calculate_dominance_info (PR 85354)

PR rtl-optimization/85354
* sel-sched-ir.c (sel_init_pipelining): Move cfg_cleanup call...
* sel-sched.c (sel_global_init): ... here.

From-SVN: r259348

re PR target/85238 (lto-wrapper: fatal error:  simple_object_copy_lto_debug_sections not implemented: Invalid argument  on Cygwin)

PR target/85238
* lto-wrapper.c (debug_objcopy): Open the files in binary mode.
* dwarf2out.c (dwarf2out_early_finish): Do not generate assembly in LTO
mode for PE-COFF targets.
* config/i386/i386-protos.h (i386_pe_asm_lto_start): Declare.
(i386_pe_asm_lto_end): Likewise.
* config/i386/cygming.h (TARGET_ASM_LTO_START): Define.
(TARGET_ASM_LTO_END): Likewise.
* config/i386/winnt.c (saved_debug_info_level): New static variable.
(i386_pe_asm_lto_start): New function.
(i386_pe_asm_lto_end): Likewise.

From-SVN: r259347

re PR middle-end/84955 (Incorrect OpenACC tile expansion)

PR middle-end/84955

gcc/
* lto-streamer-out.c (output_function): Fix CFG loop state before
streaming out.
* omp-expand.c (expand_oacc_for): Handle calls to internal
functions like regular functions.

libgomp/
* testsuite/libgomp.oacc-c-c++-common/pr84955.c: New test.
* testsuite/libgomp.oacc-fortran/pr84955.f90: New test.

Co-Authored-By: Richard Biener <rguenther@suse.de>
From-SVN: r259346

re PR lto/85371 (Compiling code with -g -flto gives an ICE on darwin after revision r259317)

2018-04-12  Richard Biener  <rguenther@suse.de>

PR lto/85371
* dwarf2out.c (init_sections_and_labels): Use debug_line_section[_label]
for the early LTO debug to properly generate references to it
during DIE emission.  Do not re-use that for the skeleton for
split-dwarf.
(dwarf2out_early_finish): Likewise.

From-SVN: r259345

re PR target/85328 (accessing ymm16 with non-avx512 instruction form)

PR target/85328
* config/i386/sse.md
(<mask_codefor>avx512dq_vextract<shuffletype>64x2_1<mask_name> split,
<mask_codefor>avx512f_vextract<shuffletype>32x4_1<mask_name> split,
vec_extract_lo_<mode><mask_name> split, vec_extract_lo_v32hi,
vec_extract_lo_v64qi): For non-AVX512VL if input is xmm16+ reg
and output is a reg, avoid creating invalid lowpart subreg, but
instead split into a 512-bit move.  Don't split if not AVX512VL,
input is xmm16+ reg and output is a mem.
(vec_extract_lo_<mode><mask_name>, vec_extract_lo_v32hi,
vec_extract_lo_v64qi): Don't require split if not AVX512VL, input is
xmm16+ reg and output is a mem.

* gcc.target/i386/pr85328.c: New test.

From-SVN: r259344

IBM Z: Spectre: Prevent thunk cfi to be emitted with -fno-dwarf2-cfi-asm

The CFI magic we emit as part of the indirect branch thunks in order to
have somewhat sane unwind information must not be emitted with
-fno-dwarf2-cfi-asm.

gcc/ChangeLog:

2018-04-12  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>

* config/s390/s390.c (s390_output_indirect_thunk_function): Check
also for flag_dwarf2_cfi_asm.

gcc/testsuite/ChangeLog:

2018-04-12  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>

* gcc.target/s390/nobp-no-dwarf2-cfi.c: New test.

From-SVN: r259340

Wzero-as-null-pointer-constant-7.C: Move...

2018-04-11  Paolo Carlini  <paolo.carlini@oracle.com>

* g++.dg/warn/Wzero-as-null-pointer-constant-7.C: Move...
* g++.dg/cpp0x/Wzero-as-null-pointer-constant-3.C: ... here.

From-SVN: r259339

re PR rtl-optimization/85342 (ICE: SIGSEGV in copyprop_hardreg_forward_1 (regcprop.c:995) with -O2 -mavx512vl)

PR rtl-optimization/85342
* regcprop.c (copyprop_hardreg_forward_1): Remove replaced array, use
a bool scalar var inside of the loop instead.  Don't try to update
recog_data.operand after failed apply_change_group.

* gcc.target/i386/pr85342.c: New test.

From-SVN: r259338

[nvptx] Fix handling of extern var with flexible array member

2018-04-12  Tom de Vries  <tom@codesourcery.com>

PR target/85296
* config/nvptx/nvptx.c (flexible_array_member_type_p): New function.
(nvptx_assemble_decl_begin): Add undefined param.  Declare undefined
array with flexible array member as array without given dimension.
(nvptx_assemble_undefined_decl): Set nvptx_assemble_decl_begin call
argument for undefined param to true.

From-SVN: r259337

Daily bump.

From-SVN: r259336

* lib/asan-dg.exp: Don't run tests if ulimit -v is set.

From-SVN: r259333

Wzero-as-null-pointer-constant-7.C: Move...

2018-04-11  Paolo Carlini  <paolo.carlini@oracle.com>

* g++.dg/warn/Wzero-as-null-pointer-constant-7.C: Move...
* g++.dg/cpp0x/Wzero-as-null-pointer-constant-3.C: ... here.

From-SVN: r259327

re PR testsuite/85190 (gcc.dg/vect/pr81196.c FAILs)

gcc/testsuite
PR testsuite/85190
* gcc.dg/vect/pr81196.c: Remove function with undefined behavior.

From-SVN: r259326

re PR c++/70808 (Spurious -Wzero-as-null-pointer-constant for nullptr_t)

PR c++/70808
* g++.dg/warn/Wzero-as-null-pointer-constant-7.C: Require c++11
effective target.

From-SVN: r259325

re PR target/85321 (Missing documentation and option misc for ppc64le)

2018-04-11  Aaron Sawdey  <acsawdey@linux.ibm.com>

PR target/85321
* doc/invoke.texi (RS/6000 and PowerPC Options): Document options
-mcall- and -mtraceback=. Remove options -mabi=spe and -mabi=no-spe
from PowerPC section.
* config/rs6000/sysv4.opt (mcall-): Improve help text.
* config/rs6000/rs6000.opt (mblock-compare-inline-limit=): Trim
help text that is too long.
* config/rs6000/rs6000.opt (mblock-compare-inline-loop-limit=): Trim
help text that is too long.
* config/rs6000/rs6000.opt (mstring-compare-inline-limit=): Trim
help text that is too long.

From-SVN: r259324

alpha.md (stack_probe_internal): Rename from "probe_stack".

* config/alpha/alpha.md (stack_probe_internal): Rename
from "probe_stack".  Update all callers.

From-SVN: r259323

sched-deps: respect deps->readonly in macro-fusion (PR 84566)

PR rtl-optimization/84566
* sched-deps.c (sched_analyze_insn): Check deps->readonly when invoking
sched_macro_fuse_insns.

From-SVN: r259322

sched-rgn: run add_branch_dependencies for sel-sched (PR 84301)

PR target/84301
* sched-rgn.c (add_branch_dependences): Move sel_sched_p check here...
(compute_block_dependences): ... from here.

testsuite/
* gcc.target/i386/pr84301.c: New test.

From-SVN: r259321

re PR tree-optimization/85331 (during GIMPLE pass, internal compiler error: Segmentation fault (-O1 and above))

PR tree-optimization/85331
* vec-perm-indices.h (vec_perm_indices::clamp): Change input type
from int to HOST_WIDE_INT.

* gcc.c-torture/execute/pr85331.c: New test.

From-SVN: r259320

Improve IPA-CP handling of self-recursive calls

2018-04-11  Martin Jambor  <mjambor@suse.cz>

PR ipa/84149
* ipa-cp.c (propagate_vals_across_pass_through): Expand comment.
(cgraph_edge_brings_value_p): New parameter dest_val, check if it is
not the same as the source val.
(cgraph_edge_brings_value_p): New parameter.
(gather_edges_for_value): Pass destination value to
cgraph_edge_brings_value_p.
(perhaps_add_new_callers): Likewise.
(get_info_about_necessary_edges): Likewise and exclude values brought
only by self-recursive edges.
(create_specialized_node): Redirect only clones of self-calling edges.
(+self_recursive_pass_through_p): New function.
(find_more_scalar_values_for_callers_subset): Use it.
(find_aggregate_values_for_callers_subset): Likewise.
(known_aggs_to_agg_replacement_list): Removed.
(decide_whether_version_node): Re-calculate known constants for all
remaining context clones.

From-SVN: r259319

re PR c++/85032 (Wrong non-constant condition for static assertion)

PR c++/85032
* constexpr.c (potential_constant_expression_1): Consider conversions
from classes to literal types potentially constant.

* g++.dg/cpp0x/pr51225.C: Adjust error message.
* g++.dg/cpp1z/constexpr-if21.C: New test.

From-SVN: r259318

re PR lto/85339 (With early LTO debug the early DWARF misses line-info)

2018-04-11  Richard Biener  <rguenther@suse.de>

PR lto/85339
* dwarf2out.c (dwarf2out_finish): Remove DW_AT_stmt_list attribute
from early DWARF output.
(dwarf2out_early_finish): Output line info unconditionally into
early DWARF and add reference to it.

From-SVN: r259317

re PR target/85281 (Assembler messages: Error: operand size mismatch for `vpbroadcastb' with -mavx512bw -masm=intel)

PR target/85281
* config/i386/sse.md (iptr): Add V16SFmode and V8DFmode cases.
(<avx512>_vec_dup<mode><mask_name>): Use a single pattern for modes
other than V2DFmode using iptr mode attribute.
(<avx512>_vec_dup<mode><mask_name>): Use iptr mode attribute.

* gcc.target/i386/pr85281.c: New test.

From-SVN: r259316

fix PR 84659 references in ChangeLog files

From-SVN: r259314

sel-sched: run cleanup_cfg just before loop_optimizer_init (PR 84659)

PR rtl-optimization/85659
* sel-sched-ir.c (sel_init_pipelining): Invoke cleanup_cfg.

testsuite/
* gcc.dg/pr84659.c: New test.

From-SVN: r259313

Clean up attribute value comparison in lto-symtab.c.

2018-04-11  Martin Liska  <mliska@suse.cz>

* lto-symtab.c (lto_symtab_merge_p): Use attribute_value_equal
function.

From-SVN: r259312

re PR rtl-optimization/85302 (ICE in size_of_loc_descr, at dwarf2out.c:1771 on i686-linux-gnu)

PR debug/85302
* dwarf2out.c (skip_loc_list_entry): Don't call size_of_locs if
SIZEP is NULL.
(output_loc_list): Pass address of a dummy size variable even in the
locview handling loop.
(index_location_lists): Add comment on why skip_loc_list_entry can't
call size_of_locs.

* g++.dg/debug/dwarf2/pr85302.C: New test.

From-SVN: r259311

[ARM] Fix PR85261: ICE with FPSCR setter builtin

Instruction pattern for setting the FPSCR expects the input value to be
in a register. However, __builtin_arm_set_fpscr expander does not ensure
that this is the case and as a result GCC ICEs when the builtin is
called with a constant literal.

This commit fixes the builtin to force the input value into a register.
It also remove the unneeded volatile in the existing fpscr test and
fixes the function prototype.

2018-04-11  Thomas Preud'homme  <thomas.preudhomme@arm.com>

    gcc/
    PR target/85261
    * config/arm/arm-builtins.c (arm_expand_builtin): Force input operand
    into register.

    gcc/testsuite/
    PR target/85261
    * config/arm/arm-builtins.c (arm_expand_builtin): Force input operand
    into register.

From-SVN: r259310

Update libstdc++ manual in preparation for GCC 8 release

* doc/xml/manual/abi.xml: Document header locations in recent
releases.
* doc/xml/manual/evolution.xml: Add API changes since GCC 5.
* doc/xml/manual/spine.xml: Update copyright years.
* doc/xml/manual/strings.xml: Adjust tolower example to avoid
undefined behaviour.
* doc/xml/manual/test.xml: Update outdated notes on VERIFY in tests.
* doc/html/*: Regenerate.

From-SVN: r259308

Daily bump.

From-SVN: r259307

re PR c++/70808 (Spurious -Wzero-as-null-pointer-constant for nullptr_t)

/cp
2018-04-10  Paolo Carlini  <paolo.carlini@oracle.com>

PR c++/70808
* init.c (build_zero_init_1): Handle NULLPTR_TYPE_P being true of
the type like TYPE_PTR_OR_PTRMEM_P.

/testsuite
2018-04-10  Paolo Carlini  <paolo.carlini@oracle.com>

PR c++/70808
* g++.dg/warn/Wzero-as-null-pointer-constant-7.C: New.

From-SVN: r259303

re PR target/85321 (Missing documentation and option misc for ppc64le)

2018-04-10  Aaron Sawdey  <acsawdey@linux.ibm.com>

PR target/85321
* doc/invoke.texi (RS/6000 and PowerPC Options): Document options
-mblock-compare-inline-limit, -mblock-compare-inline-loop-limit,
and -mstring-compare-inline-limit.

From-SVN: r259302

MAINTAINERS: Update my email address.

2018-04-10  Aaron Sawdey  <acsawdey@linux.ibm.com>

* MAINTAINERS: Update my email address.

From-SVN: r259301

forgot the PR marker in the changelog

From-SVN: r259300

rs6000: Fix stack clash for big residuals (PR85287)

The stack clash protection code had a logic error in how it decided
whether to put the final update size in a register, or to emit it
directly in an insn.  This fixes it.  It also tidies some surrounding
code.

PR target/85287
* gcc/config/rs6000/rs6000.md (allocate_stack): Put the residual size
for stack clash protection in a register whenever we need it to be in
a register.

From-SVN: r259299

rs6000: Enable -fasynchronous-unwind-tables by default

To find out where on-entry register values live at any point in a
program, GDB currently tries to parse to parse the executable code.
This does not work very well, for example it gets confused if some
accesses to the stack use the frame pointer (r31) and some use the
stack pointer (r1).  A symptom is that backtraces can be cut short.

This patch enables -fasynchronous-unwind-tables by default for rs6000,
which causes us to emit DWARF unwind tables for all functions, solving
these problems.

This not do anything for sub-targets without DWARF, and only for ELF
sub-targets for now.

It increases executable size, but only modestly, and does not change
memory use, only the disk image.

* common/config/rs6000/rs6000-common.c (rs6000_option_init_struct):
Enable -fasynchronous-unwind-tables by default if OBJECT_FORMAT_ELF.

gcc/testsuite/
* gcc.target/powerpc/dfmode_off.c: Add -fno-asynchronous-unwind-tables.
* gcc.target/powerpc/dimode_off.c: Ditto.
* gcc.target/powerpc/tfmode_off.c: Ditto.
* gcc.target/powerpc/timode_off.c: Ditto.

From-SVN: r259298

rs6000: Improve --help=target (PR85321)

This updates the help text for some options to mention the allowed
values for -mXX=XX.

PR target/85321
* config/rs6000/rs6000.opt (mtraceback=): Show the allowed values in
the help text.
(mlong-double-): Ditto.
* config/rs6000/sysv4.opt (msdata=): Ditto.
(mtls-size=): Ditto.

From-SVN: r259296

rs6000-c.c (altivec_overloaded_builtins): Remove erroneous entries for "vector int vec_ldl (int...

gcc/ChangeLog:

2018-04-10  Kelvin Nilsen  <kelvin@gcc.gnu.org>

* config/rs6000/rs6000-c.c (altivec_overloaded_builtins): Remove
erroneous entries for
"vector int vec_ldl (int, long int *)", and
"vector unsigned int vec_ldl (int, unsigned long int *)".
Add comments and entries for
"vector bool char vec_ldl (int, bool char *)",
"vector bool short vec_ldl (int, bool short *)",
"vector bool int vec_ldl (int, bool int *)",
"vector bool long long vec_ldl (int, bool long long *)",
"vector pixel vec_ldl (int, pixel *)",
"vector long long vec_ldl (int, long long *)",
"vector unsigned long long vec_ldl (int, unsigned long long *)".
* config/rs6000/rs6000.c (rs6000_init_builtins): Initialize new
type tree bool_long_long_type_node and correct definition of
bool_V2DI_type_node to make reference to this new type tree.
(rs6000_mangle_type): Replace erroneous reference to
bool_long_type_node with bool_long_long_type_node.
* config/rs6000/rs6000.h (enum rs6000_builtin_type_index): Add
comments to emphasize sign distinctions for char and int types and
replace RS6000_BTI_bool_long constant with
RS6000_BTI_bool_long_long constant.  Also add comment to restrict
use of RS6000_BTI_pixel.
(bool_long_type_node): Remove this macro definition.
(bool_long_long_type_node): New macro definition

gcc/testsuite/ChangeLog:

2018-04-10  Kelvin Nilsen  <kelvin@gcc.gnu.org>

* gcc.target/powerpc/vec-ldl-1.c: New test.
* gcc.dg/vmx/ops-long-1.c: Correct test programs to reflect
corrections to ABI implementation.

From-SVN: r259294

PR debug/65821 - wrong location for main().

* call.c (clear_location_r, convert_default_arg): Revert.
* tree.c (break_out_target_exprs): Add clear_location parm.
(struct bot_data): New.
(bot_manip): Clear location if requested.
* init.c (get_nsdmi): Pass clear_location.

From-SVN: r259291

[PR target/85056] Address -Wmaybe-uninitialized diagnostic

gcc/testsuite/
* gcc.target/nvptx/pr85056.c (main): Initialize "sum".

From-SVN: r259288

cppopts.texi: Use "side effect" instead of side-effect.

gcc/ChangeLog:

* doc/cppopts.texi: Use "side effect" instead of side-effect.
* doc/extend.texi: Same.
* doc/generic.texi: Same.
* doc/implement-c.texi: Same.
* doc/invoke.texi: Same.
* doc/md.texi: Same.
* doc/rtl.texi: Same.
* doc/tree-ssa.texi: Same.

From-SVN: r259287

Update links to archived copy of SGI STL docs

* doc/xml/faq.xml: Update links to archived copy of SGI STL docs.
* doc/xml/manual/backwards_compatibility.xml: Likewise.
* doc/xml/manual/containers.xml: Likewise.
* doc/xml/manual/debug_mode.xml: Likewise.
* doc/xml/manual/extensions.xml: Likewise.
* doc/xml/manual/policy_data_structures_biblio.xml: Likewise.
* doc/xml/manual/using.xml: Likewise.
* doc/xml/manual/utilities.xml: Likewise.

From-SVN: r259286

re PR rtl-optimization/85300 (ICE in exact_int_to_float_conversion_p, at simplify-rtx.c:895)

PR rtl-optimization/85300
* combine.c (subst): Handle subst of CONST_SCALAR_INT_P new_rtx also
into FLOAT and UNSIGNED_FLOAT like ZERO_EXTEND, return a CLOBBER if
simplify_unary_operation fails.

* gcc.dg/pr85300.c: New test.

From-SVN: r259285

gdbhooks: Add new pretty-printers for: varpool_node, symtab_node, cgraph_edge and ipa_ref.

2018-04-10  Martin Liska  <mliska@suse.cz>

* gdbhooks.py: Add pretty-printers for varpool_node, symtab_node,
cgraph_edge and ipa_ref.

From-SVN: r259283

Show pertinent parameter (PR c++/85110)

gcc/cp/ChangeLog:
PR c++/85110
* call.c (get_fndecl_argument_location): Make non-static.
* cp-tree.h (get_fndecl_argument_location): New decl.
* typeck.c (convert_for_assignment): When complaining due to
conversions for an argument, show the location of the parameter
within the decl.

gcc/testsuite/ChangeLog:
PR c++/85110
* g++.dg/cpp1z/direct-enum-init1.C: Update for the cases
where we now show the pertinent parameter.
* g++.dg/diagnostic/aka2.C: Likewise.
* g++.dg/diagnostic/param-type-mismatch-2.C: Likewise.

From-SVN: r259282

PR libstdc++/85222 allow catching iostream errors as gcc4-compatible ios::failure

Define a new exception type derived from std::ios::failure[abi:cxx11]
which also aggregates an object of the gcc4-compatible ios::failure
type. Make __throw_ios_failure throw this new type for iostream errors
that raise exceptions. Provide custom type info for the new type so that
it can be caught by handlers for the gcc4-compatible ios::failure type
as well as handlers for ios::failure[abi:cxx11] and its bases.

PR libstdc++/85222
* src/c++11/Makefile.am [ENABLE_DUAL_ABI]: Add special rules for
cxx11-ios_failure.cc to rewrite type info for __ios_failure.
* src/c++11/Makefile.in: Regenerate.
* src/c++11/cxx11-ios_failure.cc (__ios_failure, __iosfail_type_info):
New types.
[_GLIBCXX_USE_DUAL_ABI] (__throw_ios_failure): Define here.
* src/c++11/ios.cc (__throw_ios_failure): Remove definition.
* src/c++98/ios_failure.cc (__construct_ios_failure)
(__destroy_ios_failure, is_ios_failure_handler): New functions.
[!_GLIBCXX_USE_DUAL_ABI] (__throw_ios_failure): Define here.
* testsuite/27_io/ios_base/failure/dual_abi.cc: New.
* testsuite/27_io/basic_ios/copyfmt/char/1.cc: Revert changes to
handler types, to always catch std::ios_base::failure.
* testsuite/27_io/basic_ios/exceptions/char/1.cc: Likewise.
* testsuite/27_io/basic_istream/extractors_arithmetic/char/
exceptions_failbit.cc: Likewise.
* testsuite/27_io/basic_istream/extractors_arithmetic/wchar_t/
exceptions_failbit.cc: Likewise.
* testsuite/27_io/basic_istream/extractors_other/char/
exceptions_null.cc: Likewise.
* testsuite/27_io/basic_istream/extractors_other/wchar_t/
exceptions_null.cc: Likewise.
* testsuite/27_io/basic_istream/sentry/char/12297.cc: Likewise.
* testsuite/27_io/basic_istream/sentry/wchar_t/12297.cc: Likewise.
* testsuite/27_io/basic_ostream/inserters_other/char/
exceptions_null.cc: Likewise.
* testsuite/27_io/basic_ostream/inserters_other/wchar_t/
exceptions_null.cc: Likewise.
* testsuite/27_io/ios_base/storage/2.cc: Likewise.

From-SVN: r259281

PR c++/85312 - P0962 cleanup

PR c++/85312 - P0962 cleanup
* parser.c (cp_parser_perform_range_for_lookup): Remove unreachable
diagnostics.

From-SVN: r259279

PR debug/65821 - wrong location for main().

* call.c (clear_location_r): New.
(convert_default_arg): Use it.
* tree.c (bot_manip): Remove builtin_LINE/FILE handling.

From-SVN: r259278

PR c++/85285 - ICE with flexible array after substitution.

* pt.c (instantiate_class_template_1): Check for flexible array in
union.

From-SVN: r259277

* parser.c (cp_parser_check_template_parameters): Improve comment.

From-SVN: r259276

re PR fortran/85313 (gcc/fortran/openmp.c: 2 * confused logic ?)

PR fortran/85313
* openmp.c (resolve_omp_do): Remove bogus if (j < i) break;.
(resolve_oacc_nested_loops): Likewise.  Formatting fix.

* gfortran.dg/gomp/pr85313.f90: New test.

From-SVN: r259275

Fix obvious error in handling of error attribute (PR lto/85248).

2018-04-10  Martin Liska  <mliska@suse.cz>

PR lto/85248
* lto-symtab.c (lto_symtab_merge_p): Do not check for
TREE_VALUES of error attributes.

From-SVN: r259274

re PR c++/24314 (Extra "template<>" in partial specialization is compiled successfuly.)

2018-04-10  Paolo Carlini  <paolo.carlini@oracle.com>

PR c++/24314
* g++.dg/parse/template29.C: New.

From-SVN: r259273

re PR testsuite/85190 (gcc.dg/vect/pr81196.c FAILs)

gcc/testsuite
PR testsuite/85190
* gcc.dg/vect/pr81196.c: Adjust pointer for aligned access.

From-SVN: r259272

re PR target/85177 (wrong code with -O -fno-tree-ccp -fno-tree-sra -mavx512f)

PR target/85177
PR target/85255
* config/i386/sse.md
(<extract_type>_vinsert<shuffletype><extract_suf>_mask): Fix
computation of the VEC_MERGE selector from mask.
(<extract_type>_vinsert<shuffletype><extract_suf>_1<mask_name>):
Fix decoding of the VEC_MERGE selector into mask.

* gcc.target/i386/avx512f-pr85177.c: New test.
* gcc.target/i386/avx512f-pr85255.c: New test.

From-SVN: r259269

Add missing cases to vect_get_smallest_scalar_type (PR 85286)

In this PR we used WIDEN_SUM_EXPR to vectorise:

  short i, y;
  int sum;
  [...]
  for (i = x; i > 0; i--)
    sum += y;

with 4 ints and 8 shorts per vector.  The problem was that we set
the VF based only on the ints, then calculated the number of vector
copies based on the shorts, giving 4/8.  Previously that led to
ncopies==0, but after r249897 we pick it up as an ICE.

In this particular case we could vectorise the reduction by setting
ncopies based on the output type rather than the input type, but it
doesn't seem worth adding a special "optimisation" for such a
pathological case.  I think it's really an instance of the more general
problem that we can't vectorise using combinations of (say) 64-bit and
128-bit vectors on targets that support both.

2018-04-10  Richard Sandiford  <richard.sandiford@linaro.org>

gcc/
PR tree-optimization/85286
* tree-vect-data-refs.c (vect_get_smallest_scalar_type):

gcc/testsuite/
PR tree-optimization/85286
* gcc.dg/vect/pr85286.c: New test.

From-SVN: r259268

Set insn_last_address in final_1

final_1 already sets insn_current_address for each instruction, making
it possible to use some of the address functions in final.c during
assembly generation.  This patch also sets insn_last_address, since
as the comment says, we can treat final as a shorten_branches pass that
does nothing.  It's then possible to use insn_current_reference_address
during final as well.

This is needed for the aarch64.md definitions of far_branch to work:

   (set (attr "far_branch")
(if_then_else (and (ge (minus (match_dup 2) (pc)) (const_int -1048576))
   (lt (minus (match_dup 2) (pc)) (const_int 1048572)))
      (const_int 0)
      (const_int 1)))]

This value (tested only during final) uses the difference between
the INSN_ADDRESSES of operand 2 and insn_current_reference_address
to calculate a conservatively-correct estimate of the branch distance.
It takes into account the worst-case gap due to alignment, whereas
a direct comparison of INSN_ADDRESSES would give an unreliable,
optimistic result.

2018-04-10  Richard Sandiford  <richard.sandiford@linaro.org>

gcc/
* final.c (final_1): Set insn_last_address as well as
insn_current_address.

From-SVN: r259267

[explow] PR target/85173: validize memory before passing it on to target probe_stack

In this PR the expansion code emits an invalid memory address for the stack probe, which the backend fails to recognise.
The address is created explicitly in anti_adjust_stack_and_probe_stack_clash in explow.c and passed down to gen_probe_stack
without any validation in emit_stack_probe.

This patch fixes the ICE by calling validize_mem on the memory location before passing it down to the target.
Jakub pointed out that we also want to create valid addresses for the probe_stack_address case, so this patch
creates an expand operand and legitimizes it before passing it down to the probe_stack_address expander.

This patch passes bootstrap and testing on arm-none-linux-gnueabihf and aarch64-none-linux-gnu
and ppc64le-redhat-linux on gcc112 in the compile farm.

PR target/85173
* explow.c (emit_stack_probe): Call validize_mem on memory location
before passing it to gen_probe_stack.  Create address operand and
legitimize it for the probe_stack_address case.

* gcc.target/arm/pr85173.c: New test.

From-SVN: r259266

Be more carefull about DECL merging in LTO (PR lto/85248).

2018-04-10  Richard Biener  <rguenther@suse.de>
    Martin Liska  <mliska@suse.cz>

PR lto/85248
* lto-symtab.c (lto_symtab_merge_p): Handle noreturn attribute.
2018-04-10  Jakub Jelinek  <jakub@redhat.com>

PR lto/85248
* gcc.dg/lto/pr85248_0.c: New test.
* gcc.dg/lto/pr85248_1.c: New test.

From-SVN: r259265