diagnostics: fix bad interaction between line spans and line numbers

Without this patch, the "line span" markers and the line numbering
interacted badly, leading to stray copies of the line-span markers
appearing as prefixes on the first source line in a span:

missing-header-fixit-3.c: In function 'test':
missing-header-fixit-3.c:9:3: warning: implicit declaration of function 'printf' [-Wimplicit-function-declaration]
9 |   printf ("%i of %i\n", i, j);
  |   ^~~~~~
missing-header-fixit-3.c:9:3: warning: incompatible implicit declaration of built-in function 'printf'
missing-header-fixit-3.c:9:3: note: include '<stdio.h>' or provide a declaration of 'printf'
missing-header-fixit-3.c:1:1:
  |+#include <stdio.h>
missing-header-fixit-3.c:1:1:1 | /* Example of a fix-it hint that adds a #include directive,
missing-header-fixit-3.c:9:3:
missing-header-fixit-3.c:9:3:9 |   printf ("%i of %i\n", i, j);
  |   ^~~~~~

With this patch, we now correctly print:

missing-header-fixit-3.c: In function 'test':
missing-header-fixit-3.c:9:3: warning: implicit declaration of function 'printf' [-Wimplicit-function-declaration]
9 |   printf ("%i of %i\n", i, j);
  |   ^~~~~~
missing-header-fixit-3.c:9:3: warning: incompatible implicit declaration of built-in function 'printf'
missing-header-fixit-3.c:9:3: note: include '<stdio.h>' or provide a declaration of 'printf'
missing-header-fixit-3.c:1:1:
+ |+#include <stdio.h>
1 | /* Example of a fix-it hint that adds a #include directive,
missing-header-fixit-3.c:9:3:
9 |   printf ("%i of %i\n", i, j);
  |   ^~~~~~

gcc/ChangeLog:
* diagnostic.c (default_diagnostic_start_span_fn): Call pp_string
to emit the span, rather than setting it as the prefix.

gcc/testsuite/ChangeLog:
* gcc.dg/missing-header-fixit-3.c: New test.

From-SVN: r263606

diagnostics: tweak to line-insertion fix-it hints with line-numbering

This commit slightly tweaks line-insertion fix-it hints, so that
with line-numbering, rather than e.g.:

 99 |       x = a;
    |+      break;
110 |     case 'b':
    |     ^~~~~~~~

we fill the margin with "+":

 99 |       x = a;
+++ |+      break;
110 |     case 'b':
    |     ^~~~~~~~

to emphasize that this is a suggested new line, rather than the user's
source.

gcc/ChangeLog:
* diagnostic-show-locus.c (layout::start_annotation_line): Add
"margin_char" parameter, defaulting to space.  Use it in place
of pp_space for the initial part of the margin.
(layout::print_leading_fixits): Use '+' when filling the margin
of line-insertion fix-it hints.

gcc/testsuite/ChangeLog:
* gcc.dg/plugin/diagnostic-test-show-locus-bw-line-numbers.c
(test_fixit_insert_newline): Update expected output to show '+'
characters in margin of line-insertion fix-it hint.

From-SVN: r263605

rs6000: Delete old add+cmp patterns

There are some patterns that recognise the parallel of an add and a
compare, and split it back to the same two insns.  This apparently
helped RIOS machines before RTL scheduling existed?  Either way, it
isn't helpful anymore, and even hurts a tiny bit.  So, delete it.

* config/rs6000/rs6000.md (two unnamed define_insn and define_split):
Delete.

From-SVN: r263604

rs6000: Remove "length 4" from other insns

There were many insns that set "length 4" explicitly while that does
not make anything clearer to the reader.  So, simplify the code.

* config/rs6000/altivec.md: Don't set length attribute to the default
value.
* config/rs6000/darwin.md: Ditto.
* config/rs6000/dfp.md: Ditto.
* config/rs6000/htm.md: Ditto.
* config/rs6000/rs6000.md: Ditto.
* config/rs6000/sync.md: Ditto.
* config/rs6000/vsx.md: Ditto.

From-SVN: r263603

rs6000: Remove "length 4" from branch insns

Now that it is the default for branch insns like for all other insns,
we don't need to set it explicitly so often anymore.

* config/rs6000/altivec.md: Don't set length attribute to the default
value, for branch instructions.
* config/rs6000/darwin.md: Ditto.
* config/rs6000/rs6000.md: Ditto.

From-SVN: r263602

rs6000: Change the length attribute default

This moves what is currently the default of the length attribute to
the only branch instruction patterns where it applies, namely, the
B-form instructions.  It was used for the "jump" instruction as well
before, but that is an I-form instruction and always has length 4.

* config/rs6000/rs6000.md (length): Always define as const_int 4.
(unnamed conditional branch define_insn): Set length to 4 or 8
depending on offset.
(<bd>_<mode>): Similar, for alternative 0.
(<bd>tf_<mode>): Ditto.

From-SVN: r263601

[PATCH] Macro definition parameter parsing

https://gcc.gnu.org/ml/gcc-patches/2018-08/msg00977.html
libcpp/
* internal.h (_cpp_save_parameter): Take parmno, not macro.
(_cpp_unsave_parameters): Declare.
* macro.c (_cpp_save_parameter): Take parm number, not macro.
Return true on success.
(_cpp_unsave_parameters): New.
(parse_params): Take parm_no and variadic pointers, not macro.
Reimplement parsing logic.
(create_iso_definition): Adjust parse_params changes.  Call
_cpp_unsave_parameters here.
(_cpp_create_definition): Don't unsave params here.
* traditional.c (scan_parameters): Take n_param pointer, adjust.
(_cpp_create_trad_definition): Ajust scan_parameters change.  Call
_cpp_unsave_parameters.
gcc/testsuite/
* gcc.dg/cpp/macsyntx.c: Adjust expected errors.
* gcc.dg/cpp/macsyntx2.c: likewise.

From-SVN: r263600

re PR testsuite/86745 (gcc.target/i386/avx-cvt-2.c etc. FAIL on 64-bit x86)

PR testsuite/86745
* gcc.target/i386/avx-cvt-2.c: Loosen scan-assembler strings.
* gcc.target/i386/avx-cvt-2.c: Ditto.

From-SVN: r263598

Fix warning with -Wsign-compare -Wsystem-headers

* include/tr1/legendre_function.tcc (__sph_legendre): Avoid warning
about signed/unsigned comparison.

From-SVN: r263596

-Wmisleading-indentation: fix ICE in get_visual_column (PR c++/70693)

PR c++/70693 reports a crash within -Wmisleading-indentation in
get_visual_column, reading past the end of a source line.

The issue occurs due to a stray carriage return aka '\r' aka ^M, occurring
towards the end of line 35 of attachment 38289 - but not at the end itself.

This carriage return confuses our line numbering: from that point in the
file, the lexer (and thus location_t values) use line numbers that are
one larger than those seen by input.c, "cat -n" and emacs.

This discrepancy between the lexer's line numbering and input.c's line
numbering leads to an out-of-range read in get_visual_column (trying to
read column 8, to locate the first non-whitespace on the line containing
"break;", but finding the next line, which is only 4 characters long).

This patch fixes the ICE by adding a range check to get_visual_column
before accessing the input.c line buffer.  This is arguably papering
over the root cause, but there are presumably other ways of triggering
such an out-of-range read by writing to the source file after the lexer
but before -Wmisleading-indentation, and we ought to be not ICE in the
face of that.

gcc/c-family/ChangeLog:
PR c++/70693
* c-common.c (selftest::c_family_tests): Call
selftest::c_indentation_c_tests.
* c-common.h (selftest::c_indentation_c_tests): New decl.
* c-indentation.c: Include "selftest.h".
(next_tab_stop): Add "tab_width" param, rather than accessing
cpp_opts.
(get_visual_column): Likewise.  Clarify comment.  Bulletproof
against reading past the end of the line.
(get_first_nws_vis_column): Add "tab_width" param.
(detect_intervening_unindent): Likewise.
(should_warn_for_misleading_indentation): Read tab width from
cpp_opts and pass around.
(selftest::test_next_tab_stop): New test.
(selftest::assert_get_visual_column_succeeds): New function.
(ASSERT_GET_VISUAL_COLUMN_SUCCEEDS): New macro.
(selftest::assert_get_visual_column_fails): New function.
(ASSERT_GET_VISUAL_COLUMN_FAILS): New macro.
(selftest::test_get_visual_column): New test.
(selftest::c_indentation_c_tests): New function.

gcc/testsuite/ChangeLog:
PR c++/70693
* c-c++-common/Wmisleading-indentation-pr70693.c: New test.

From-SVN: r263595

Allow larger copies when not slow_unaligned_access and no padding.

This allows copy_blkmode_to_reg to perform larger copies when it is safe to do so by calculating
the bitsize per iteration doing the maximum copy allowed that does not read more
than the amount of bits left to copy.

Strictly speaking, this copying is only done if:

  1. the target supports fast unaligned access
  2. no padding is being used.

This should avoid the issues of the first patch (PR85123) but still work for targets that are safe
to do so.

Original patch https://gcc.gnu.org/ml/gcc-patches/2017-11/msg01088.html
Previous respin https://gcc.gnu.org/ml/gcc-patches/2018-04/msg00239.html

gcc/
2018-08-16  Tamar Christina  <tamar.christina@arm.com>

* expr.c (copy_blkmode_to_reg): Perform larger copies when safe.

From-SVN: r263594

Avoid deprecation warning with -Wsystem-headers

C++17 says to use std::uncaught_exceptions() here instead of
std::uncaught_exception() but since we only care whether the result is
non-zero (and we aren't planning to remove the deprecated version) we
can just keep using std::uncaught_exception() and suppress the warning.

* include/std/ostream (basic_ostream::sentry::~sentry): Suppress
deprecation warnings for using uncaught_exception().

From-SVN: r263593

Remove rtl.texi references to old RTX code class names

Committed on behalf of Matthew Malcomson.

2018-08-16  Matthew Malcomson  <matthew.malcomson@arm.com>

gcc/
* doc/rtl.texi: Replace old RTX class names with new names.

From-SVN: r263592

expmed.h (canonicalize_comparison): New declaration.

* expmed.h (canonicalize_comparison): New declaration.
* expmed.c (canonicalize_comparison, equivalent_cmp_code): New function.
* expmed.c (emit_store_flag_1): Add call to canonicalize_comparison.
* optabs.c (prepare_cmp_insn): Likewise.
* rtl.h (unsigned_condition_p): New function which checks if a
comparison operator is unsigned.

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

From-SVN: r263591

Fix bootstrap with --enable-fully-dynamic-string

PR libstdc++/86447
* src/c++11/cow-stdexcept.cc [_GLIBCXX_FULLY_DYNAMIC_STRING]
(logic_error::logic_error(logic_error&&))
(logic_error::operator=(logic_error&&))
(runtime_error::runtime_error(runtime_error&&))
(runtime_error::operator=(runtime_error&&)): Copy strings instead of
moving, to avoid allocating empty reps for moved-from strings.

From-SVN: r263590

[PATCH] CPP Macro predicates

https://gcc.gnu.org/ml/gcc-patches/2018-08/msg00897.html
libcpp/
* include/cpplib.h (cpp_user_macro_p, cpp_builtin_macro_p)
(cpp_macro_p): New inlines.
* directives.c (do_pragma_poison): Use cpp_macro_p.
(do_ifdef, do_ifndef): Likewise.  Use _cpp_maybe_notify_macro_use.
(cpp_pop_definition): Use cpp_macro_p.  Move _cpp_free_definition
earlier.  Don't zap node directly.
* expr.c (parse_defined): Use _cpp_maybe_notify_macro_use &
cpp_macro_p.
* files.c (should_stack_file): Use cpp_macro_p.
* identifiers.c (cpp_defined): Likewise.
* internal.h (_cpp_mark_macro): Use cpp_user_macro_p.
(_cpp_notify_macro_use): Declare.
(_cpp_maybe_notify_macro_use): New inline.
* lex.c (is_macro): Use cpp_macro_p.
* macro.c (_cpp_warn_if_unused_macro): Use cpp_user_macro_p.
(enter_macro_context): Likewise.
(_cpp_create_definition): Use cpp_builtin_macro_p,
cpp_user_macro_p.  Move _cpp_free_definition earlier.
(_cpp_notify_macro_use): New, broken out of multiple call sites.
* traditional.c (fun_like_macro_p): Use cpp_builtin_macro_p.
(maybe_start_funlike, _cpp_scan_out_logical_line)
(push_replacement_text): Likewise.
gcc/c-family/
* c-ada-spec.c (count_ada_macro): Use cpp_user_macro_p.
(store_ada_macro): Likewise.
* c-ppoutput.c (cb_used_define, dump_macro): Likewise.
* c-spellcheck.cc (should-suggest_as_macro_p): Likewise,
gcc/
* config/rs6000/rs6000-c.c (rs6000_macro_to_expend): Use cpp_macro_p.
* config/powerpcspc/powerpcspe-c.c (rs6000_macro_to_expend): Likewise.
gcc/cp/
* name-lookup.c (lookup_name_fuzzy): Likewise.
gcc/fortran/
* cpp.c (dump_macro): Use cpp_user_macro_p.

From-SVN: r263587

Add __USER_LABEL_PREFIX__ to the asm names for targets that need it.

2018-08-16  Iain Sandoe <iain@sandoe.co.uk>

gcc/testsuite

        * gcc.dg/memcmp-1.c (lib_memcmp): Apply __USER_LABEL_PREFIX__.
        (lib_strncmp): Likewise.

From-SVN: r263586

Darwin doesn't have pthread_barrier, at least up to Darwin18
so don't try to do tests with it.

2018-08-16  Iain Sandoe <iain@sandoe.co.uk>

gcc/testsuite/
* c-c++-common/asan/pointer-subtract-3.c: Skip for Darwin.
* c-c++-common/asan/pointer-subtract-4.c: Likewise.
* g++.dg/torture/pr44295.C : Likewise.

From-SVN: r263585

Update fall through pattern for FP16 patterns in ARM.

The original issue comes from the fact that the code does

... foo (... bar)
{
  return bar;
}

The expansion of the return statement causes GCC to try to return the value in
a register.  GCC will try to emit the move then, from MEM to REG (due to the SSA
temporary.).  It checks for a mov optab for this which isn't available and
then tries to do the move in bits using emit_move_multi_word.

emit_move_multi_word will split the move into sub parts, but then needs to get
the sub parts and does this using subregs, but it's told it can't do subregs!

The compiler is now stuck in an infinite loop.

The way this is worked around in the back-end is that we have move patterns in
neon.md that usually just force the register instead of checking with the
back-end. This prevents emit_move_multi_word from being needed.  However the
pattern for V4HF and V8HF were guarded by TARGET_NEON && TARGET_FP16.

I don't believe the TARGET_FP16 guard to be needed, because the pattern doesn't
actually generate code and requires another pattern for that, and a reg to reg move
should always be possible anyway. So allowing the force to register here is safe
and it allows the compiler to generate a correct error instead of ICEing in an
infinite loop.

gcc/
2018-08-16  Tamar Christina  <tamar.christina@arm.com>

PR target/84711
* config/arm/arm.c (arm_can_change_mode_class): Disallow subreg.
* config/arm/neon.md (movv4hf, movv8hf): Refactored to..
(mov<mov>): ..this and enable unconditionally.

From-SVN: r263584

fix section naming convention for Darwin in three tests

2018-08-15  Iain Sandoe <iain@sandoe.co.uk>

gcc/testsuite:

PR testsuite/78544
* gcc.dg/tree-prof/section-attr-1.c: Add Darwin-specific section attributes and matching.
* gcc.dg/tree-prof/section-attr-2.c: Likewise.
* gcc.dg/tree-prof/section-attr-3.c: Likewise.
6

From-SVN: r263583

Remove dead pattern on ARM.

About 13 years ago the reg-to-reg patterns were split up, before that time
output_move_double could actually handle this case.

After the split was done most patterns were updated except for *neon_mov
which incorrectly retained reg,reg as a valid alternative.

However output_move_double can not handle this and simply returns ""
and asserts.

This pattern is essentially dead and I'm removing it for clarity.

gcc/
2018-08-16  Tamar Christina  <tamar.christina@arm.com>

* config/arm/neon.md (*neon_mov<mode>): Remove reg-to-reg alternative.

From-SVN: r263582

[aarch64] Make common aarch64 options target-dependent

This patch replaces the "Common" attribute in some aarch64 options with "Target"
to make them target-dependent. This doesn't affect their behaviour or how they
are documented with "--help", but does mean that they follow the option
specification that target-specific options are given the "Target" attribute.

Tested by ensuring the changed options are still accepted as before and that
they aren't invokable from an invalid target. Bootstrapped and regression tested
on aarch64-none-elf with ... regressions.

2018-08-16  Sam Tebbs  <sam.tebbs@arm.com>

* config/aarch64/aarch64.opt (mlow-precision-recip-sqrt)
(mlow-precision-sqrt, mlow-precision-div, mverbose-cost-dump): Replace
"Common" with "Target".

From-SVN: r263578

Fix pr81923 test for targets using __USER_LABEL_PREFIX__

2018-08-16  Iain Sandoe <iain@sandoe.co.uk>

gcc/testsuite

* gcc.dg/asan/pr81923.c: Stringify __USER_LABEL_PREFIX__.

From-SVN: r263577

Daily bump.

From-SVN: r263576

i386.opt (mmitigate-rop): Mark as deprecated.

* config/i386/i386.opt (mmitigate-rop): Mark as deprecated.
* doc/invoke.texi (mmitigate-rop): Remove.
* config/i386/i386.c: Do not include "regrename.h".
(ix86_rop_should_change_byte_p, reg_encoded_number)
(ix86_get_modrm_for_rop, set_rop_modrm_reg_bits, ix86_mitigate_rop):
Remove.
(ix86_reorg): Remove call to ix86_mitigate_rop.
* config/i386/i386.md (attr "modrm_class"): Remove.
(cmp<mode>_ccno_1, mov<mode>_xor, movstrict<mode>_xor)
(x86_mov<mode>cc_0_m1. x86_mov<mode>cc_0_m1_se)
(x86_mov<mode>cc_0_m1_neg): Remove modrm_class attribute override.

testsuite/Changelog:

* gcc.target/i386/rop1.c: Remove.
* gcc.target/i386/pr83554 (dg-options): Remove -mmitigate-rop.

From-SVN: r263572

ChangeLog: Fix whitespace.

* ChangeLog: Fix whitespace.
* testsuite/ChangeLog: Ditto.

From-SVN: r263571

fold-vec-splat-char.c: New.

[testsuite]

2018-08-13  Will Schmidt  <will_schmidt@vnet.ibm.com>

* gcc.target/powerpc/fold-vec-splat-char.c: New.
* gcc.target/powerpc/fold-vec-splat-floatdouble.c: New.
* gcc.target/powerpc/fold-vec-splat-int.c: New.
* gcc.target/powerpc/fold-vec-splat-longlong.c: New.
* gcc.target/powerpc/fold-vec-splat-pixel.c: New.
* gcc.target/powerpc/fold-vec-splat-short.c: New.

From-SVN: r263570

rs600.c (rs6000_gimple_fold_builtin): Add entries to allow folding of mergeh() and mergel() for the float and double...

[gcc]

2018-08-07 Will Schmidt  <will_schmidt@vnet.ibm.com>

* config/rs6000/rs600.c (rs6000_gimple_fold_builtin): Add entries to
allow folding of mergeh() and mergel() for the float and double types.
(fold_mergehl_helper): Rework to handle building a permute tree
for float vectors.

From-SVN: r263569

Fix experimental::pmr typedefs and add tests

The typedefs in <experimental/regex> and <experimental/string> don't
need to be in the __cxx11 namespace, because they are only aliases and
so will have the same mangled name as the underlying types.

* include/experimental/regex: Remove begin/end macros for namespace.
* include/experimental/string: Likewise.
* testsuite/experimental/polymorphic_allocator/pmr_typedefs_deque.cc:
New test.
* testsuite/experimental/polymorphic_allocator/
pmr_typedefs_forward_list.cc: New test.
* testsuite/experimental/polymorphic_allocator/pmr_typedefs_list.cc:
New test.
* testsuite/experimental/polymorphic_allocator/pmr_typedefs_map.cc:
New test.
* testsuite/experimental/polymorphic_allocator/pmr_typedefs_match.cc:
New test.
* testsuite/experimental/polymorphic_allocator/
pmr_typedefs_multimap.cc: New test.
* testsuite/experimental/polymorphic_allocator/
pmr_typedefs_multiset.cc: New test.
* testsuite/experimental/polymorphic_allocator/pmr_typedefs_set.cc:
New test.
* testsuite/experimental/polymorphic_allocator/pmr_typedefs_string.cc:
New test.
* testsuite/experimental/polymorphic_allocator/
pmr_typedefs_unordered_map.cc: New test.
* testsuite/experimental/polymorphic_allocator/
pmr_typedefs_unordered_multimap.cc: New test.
* testsuite/experimental/polymorphic_allocator/
pmr_typedefs_unordered_multiset.cc: New test.
* testsuite/experimental/polymorphic_allocator/
pmr_typedefs_unordered_set.cc: New test.
* testsuite/experimental/polymorphic_allocator/pmr_typedefs_vector.cc:
New test.

From-SVN: r263568

i386.c (expand_vec_perm_movs): Enable V4SFmode for TARGET_SSE.

* config/i386/i386.c (expand_vec_perm_movs): Enable V4SFmode
for TARGET_SSE.

From-SVN: r263567

Refactor uses-allocator construction

Remove duplicated logic in experimental::pmr::polymorphic_allocator by
calling the __uses_allocator_construct helper.

Fix bugs in std::pmr::polymorphic_allocator with incorrect SFINAE
constraint and incorrect argument order.

* include/bits/uses_allocator.h (__uses_allocator_construct): Qualify
calls to __uses_allocator_construct_impl and __use_alloc.
* include/experimental/memory_resource
(polymorphic_allocator::_M_construct): Remove.
(polymorphic_allocator::construct): Call __uses_allocator_construct.
Qualify calls to __use_alloc.
* include/std/memory_resource (polymorphic_allocator::construct): Fix
type in SFINAE constraint. Use constexpr if instead of tag dispatching
to _S_construct overloads.
(polymorphic_allocator::construct(pair<T1, T2>*, ...)): Fix order of
arguments to _S_construct_p.
(polymorphic_allocator::_S_construct): Remove.
(polymorphic_allocator::_S_construct_p): Return allocators by value
not by reference.
* include/std/scoped_allocator (scoped_allocator_adaptor::construct):
Qualify calls to __use_alloc.
* testsuite/20_util/polymorphic_allocator/construct_pair.cc: New test,
copied from testsuite/20_util/scoped_allocator/construct_pair.cc.
* testsuite/experimental/polymorphic_allocator/1.cc: New test.
* testsuite/experimental/polymorphic_allocator/construct_pair.cc:
New test.

From-SVN: r263566

diagnostics: add labeling of source ranges

This patch adds the ability to label source ranges within a rich_location,
to be printed by diagnostic_show_locus.

For example:

pr69554-1.c:11:18: error: invalid operands to binary + (have 'const char *' and 'const char *')
11 |   return (p + 1) + (q + 1);
   |          ~~~~~~~ ^ ~~~~~~~
   |             |         |
   |             |         const char *
   |             const char *

The patch implements labels for various type mismatch errors in the C and
C++ frontends, and in -Wformat.  I implemented it wherever accurate location
information was guaranteed (there are other places that could benefit, but
we need better location information in those places).

The labels can be disabled via -fno-diagnostics-show-labels.

Similarly:

param-type-mismatch.C: In function 'int test_1(int, int, float)':
param-type-mismatch.C:11:27: error: invalid conversion from 'int' to 'const char*' [-fpermissive]
11 |   return callee_1 (first, second, third);
   |                           ^~~~~~
   |                           |
   |                           int
param-type-mismatch.C:7:43: note:   initializing argument 2 of 'int callee_1(int, const char*, float)'
7 | extern int callee_1 (int one, const char *two, float three);
  |                               ~~~~~~~~~~~~^~~

where the first "error" describing the bad argument gets a label
describing the type inline (since it's non-obvious from "second").
The "note" describing the type of the param of the callee *doesn't*
get a label, since that information is explicit there in the
source ("const char *two").

The idea is that in any diagnostic where two aspects of the source aren't
in sync it ought to be easier for the user if we directly show them the
mismatching aspects inline (e.g. types).

As well as type mismatch errors, perhaps labels could also be used for
buffer overflow warnings, for describing the capacity of the destination
buffer vs the size of what's being written:

  sprintf (buf, "filename: %s\n", file);
           ^~~   ~~~~~~~~~~~^~~
           |                |
           capacity: 32     10 + strlen(file) + 2

or somesuch.  Another idea might be for macro expansion warnings:

warning: repeated side effects in macro expansion...
   x = MIN (p++, q++);
       ~~~~^~~~~~~~~~
note: ...expanded here as
 #define MIN(X,Y) (X<Y?X:Y)
         ^~~ ~ ~   ~ ~ ~ ~
             | |   | | | |
             | |   | | | q++
             | |   | | p++
             | |   | q++
             | q++ p++
             p++

The patch removes some logic from multiline.exp which special-cased
lines ending with a '|' character (thus complicating testing of this
patch).  I believe that this was a vestige from experiments I did to
support strippng dg directives from the output; it was present in the
earliest version of multiline.exp I posted:
  "[RFC, stage1] Richer source location information for gcc 6 (location ranges etc)"
    https://gcc.gnu.org/ml/gcc-patches/2015-03/msg00837.html
and I believe was neved used.

gcc/c-family/ChangeLog:
* c-format.c: Include "selftest-diagnostic.h" and
"gcc-rich-location.h".
(format_warning_at_char): Pass NULL for new label params of
format_warning_va.
(class indirection_suffix): New class.
(class range_label_for_format_type_mismatch): New class.
(format_type_warning): Move logic for generating "*" suffix to
class indirection_suffix.  Create "fmt_label" and "param_label"
to show their types, and pass them to the
format_warning_at_substring calls.
(selftest::test_type_mismatch_range_labels): New test.
(selftest::c_format_c_tests): Call it.

gcc/c/ChangeLog:
* c-objc-common.c: Include "gcc-rich-location.h".
(c_tree_printer): Move implemenation of '%T' to...
(print_type): ...this new function.
(range_label_for_type_mismatch::get_text): New function.
* c-typeck.c (convert_for_assignment): Add type labels to the rhs
range for the various ic_argpass cases.
(class maybe_range_label_for_tree_type_mismatch): New class.
(build_binary_op): Use it when calling binary_op_error.

gcc/cp/ChangeLog:
* call.c: Include "gcc-rich-location.h".
(convert_like_real): Add range label for "invalid conversion"
diagnostic.
(perform_implicit_conversion_flags): Add type label to the
"could not convert" error.
* error.c: Include "gcc-rich-location.h".
(range_label_for_type_mismatch::get_text): New function.
* typeck.c (convert_for_assignment): Add type label to
the "cannot convert" error if a location is available.

gcc/ChangeLog:
* common.opt (fdiagnostics-show-labels): New option.
* diagnostic-show-locus.c (class layout_range): Add field
"m_label".
(class layout): Add field "m_show_labels_p".
(layout_range::layout_range): Add param "label" and use it to
initialize m_label.
(make_range): Pass in NULL for new "label" param of layout_range's
ctor.
(layout::layout): Initialize m_show_labels_p.
(layout::maybe_add_location_range): Pass in loc_range->m_label
when constructing layout_range instances.
(struct line_label): New struct.
(layout::print_any_labels): New member function.
(layout::print_line): Call it if label-printing is enabled.
(selftest::test_one_liner_labels): New test.
(selftest::test_diagnostic_show_locus_one_liner): Call it.
* diagnostic.c (diagnostic_initialize): Initialize
context->show_labels_p.
* diagnostic.h (struct diagnostic_context): Add field
"show_labels_p".
* doc/invoke.texi (Diagnostic Message Formatting Options): Add
-fno-diagnostics-show-labels.
* dwarf2out.c (gen_producer_string): Add
OPT_fdiagnostics_show_labels to the ignored options.
* gcc-rich-location.c (gcc_rich_location::add_expr): Add "label"
param.
(gcc_rich_location::maybe_add_expr): Likewise.
* gcc-rich-location.h (gcc_rich_location::gcc_rich_location): Add
label" param, defaulting to NULL.
(gcc_rich_location::add_expr): Add "label" param.
(gcc_rich_location::maybe_add_expr): Likewise.
(class text_range_label): New class.
(class range_label_for_type_mismatch): New class.
* gimple-ssa-sprintf.c (fmtwarn): Pass NULL for new label params
of format_warning_va.
(fmtwarn_n): Likewise for new params of format_warning_n_va.
* lto-wrapper.c (merge_and_complain): Add
OPT_fdiagnostics_show_labels to the "pick one setting" options.
(append_compiler_options): Likewise to the dropped options.
(append_diag_options): Likewise to the passed-on options.
* opts.c (common_handle_option): Handle the new option.
* selftest-diagnostic.c
(test_diagnostic_context::test_diagnostic_context): Enable
show_labels_p.
* substring-locations.c: Include "gcc-rich-location.h".
(format_warning_n_va): Add "fmt_label" and "param_label" params
and use them as appropriate.
(format_warning_va): Add "fmt_label" and "param_label" params,
passing them on to format_warning_n_va.
(format_warning_at_substring): Likewise.
(format_warning_at_substring_n): Likewise.
* substring-locations.h (format_warning_va): Add "fmt_label" and
"param_label" params.
(format_warning_n_va): Likewise.
(format_warning_at_substring): Likewise.
(format_warning_at_substring_n): Likewise.
* toplev.c (general_init): Initialize global_dc->show_labels_p.

gcc/testsuite/ChangeLog:
* g++.dg/diagnostic/aka3.C: New test.
* g++.dg/diagnostic/param-type-mismatch-2.C: Update expected
output to show range labels.
* g++.dg/diagnostic/param-type-mismatch.C: Likewise.
* g++.dg/plugin/plugin.exp (plugin_test_list): Add...
* g++.dg/plugin/show-template-tree-color-labels.C: New test.
* gcc.dg/bad-binary-ops.c: Update expected output to show range
labels.  Add an "aka" example.
* gcc.dg/cpp/pr66415-1.c: Update expected output to show range
labels.
* gcc.dg/format/diagnostic-ranges.c: Likewise.
* gcc.dg/format/pr72858.c: Likewise.
* gcc.dg/format/pr78498.c: Likewise.
* gcc.dg/param-type-mismatch.c: Add "-Wpointer-sign" to options.
Update expected output to show range labels.  Add examples of
-Wincompatible-pointer-types and -Wpointer-sign for parameters.
* gcc.dg/plugin/diagnostic-test-show-locus-bw-line-numbers.c:
Update expected output to show range labels.
* gcc.dg/plugin/diagnostic-test-show-locus-bw.c: Likewise.
(test_very_wide_line): Adjust so that label is at left-clipping
boundary.
(test_very_wide_line_2): New test.
* gcc.dg/plugin/diagnostic-test-show-locus-color-line-numbers.c:
Update expected output to show range labels.
* gcc.dg/plugin/diagnostic-test-show-locus-color.c: Likewise.
* gcc.dg/plugin/diagnostic-test-show-locus-no-labels.c: New test.
* gcc.dg/plugin/diagnostic_plugin_show_trees.c (show_tree): Update
for new param to gcc_rich_location::add_expr.
* gcc.dg/plugin/diagnostic_plugin_test_show_locus.c (add_range):
Add "label" param.
(test_show_locus): Add examples of labels to various tests.  Tweak
the "very wide_line" test case and duplicate it, to cover the
boundary values for clipping of labels against the left-margin.
* gcc.dg/plugin/plugin.exp (plugin_test_list): Add
diagnostic-test-show-locus-no-labels.c.
* gcc.dg/pr69554-1.c: Update expected output to show range labels.
Update line numbers of dg-locus directives.
* gcc.dg/pr69627.c:  Update expected output to show range labels.
* lib/multiline.exp (proc _build_multiline_regex): Remove
special-case handling of lines with trailing '|'.

libcpp/ChangeLog:
* include/line-map.h (struct location_range): Add "m_label" field.
(class rich_location): Add description of labels to leading
comment.
(rich_location::rich_location): Add "label" param, defaulting to
NULL.
(rich_location::add_range): Likewise.
(struct label_text): New struct.
(class range_label): New abstract base class.
* line-map.c (rich_location::rich_location): Add "label" param;
use it.
(rich_location::add_range): Likewise.

From-SVN: r263564

Do not expand the call to memcmp at all when overflow is detected.

From-SVN: r263563

c-decl.c (start_decl): Do not warn if variables is named as main and is a local variable.

2018-08-15  Prathamesh Kulkarni  <prathamesh.kulkarni@linaro.org>

c/
* c-decl.c (start_decl): Do not warn if variables is named as main
and is a local variable.

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

From-SVN: r263562

PR tree-optimization/71625 - missing strlen optimization on different array

PR tree-optimization/71625 - missing strlen optimization on different array
initialization style (avoid compilation errors on aarch64)

gcc/ChangeLog:
* config/aarch64/aarch64-builtins.c
(aarch64_init_simd_builtin_types): Clear Poly8_t's TYPE_STRING_FLAG.

From-SVN: r263561

S/390: Remove branch splitting loop

Since there is no branch splitting anymore, the loop is no longer
necessary: pool chunkification can be done in one step.

gcc/ChangeLog:

2018-08-15  Ilya Leoshkevich  <iii@linux.ibm.com>

* config/s390/s390.c (s390_reorg): Remove loop.

From-SVN: r263557

Don't make unsized objects into extern.

2018-08-15  Iain Sandoe <iain@sandoe.co.uk>

gcc/c:

        PR c/19315
        * c-decl.c (finish_decl): Don't add the 'extern' storage class to
        objects of unknown size.

gcc/testsuite:

        PR c/19315
        gcc.dg/graphite/pr82451.c: Make array 'a' an extern.
        gcc.dg/redecl-10.c: Expect warnings for the static vars with unknown
        size.

From-SVN: r263556

Darwin - remove unnecessary target hook

gcc/

2018-08-15  Iain Sandoe  <iain@sandoe.co.uk>

* config/darwin.c
 (darwin_function_switched_text_sections): Delete.
* gcc/config/darwin.h
 (TARGET_ASM_FUNCTION_SWITCHED_TEXT_SECTIONS): Likewise.

From-SVN: r263555

Fix single-threaded build for targets without atomics

* src/c++17/memory_resource.cc [!_GLIBCXX_HAS_GTHREADS]
(atomic_mem_res): Add unsynchronized definition for single-threaded.

From-SVN: r263554

Update Darwin section names for DWARF5

gcc/

2018-08-15  Iain Sandoe  <iain@sandoe.co.uk>

PR target/81685
* config/darwin.h: (DEBUG_STR_OFFSETS_SECTION, DEBUG_LOCLISTS_SECTION,
DEBUG_RNGLISTS_SECTION) new macros.  (DEBUG_PUBNAMES_SECTION,
DEBUG_PUBTYPES_SECTION) update to include GNU variant.

From-SVN: r263553

Fix merging of 2 predictors (PR tree-optimization/86925).

2018-08-15  Martin Liska  <mliska@suse.cz>

        PR tree-optimization/86925
* predict.c (expr_expected_value_1): When taking
        later predictor, assign also probability.
        Use fold_build2_initializer_loc in order to fold
        the expression in -frounding-math.
2018-08-15  Martin Liska  <mliska@suse.cz>

        PR tree-optimization/86925
* gcc.dg/predict-20.c: New test.
* gcc.dg/predict-21.c: New test.

From-SVN: r263552

decl.c (check_previous_goto_1): When decl_jump_unsafe returns 2 emit an error instead of a permerror.

/cp
2018-08-15  Paolo Carlini  <paolo.carlini@oracle.com>

* decl.c (check_previous_goto_1): When decl_jump_unsafe returns 2
emit an error instead of a permerror.

/testsuite
2018-08-15  Paolo Carlini  <paolo.carlini@oracle.com>

* g++.dg/init/goto3.C: Adjust for error instead of permerror.

From-SVN: r263551

backport: ffi.c (ffi_call_go, [...]): New functions.

Backport of RISC-V support for libffi go closures
* src/riscv/ffi.c (ffi_call_go, ffi_prep_go_closure): New
functions.
(ffi_call_int): Renamed from ffi_call.
(ffi_call_asm, ffi_closure_inner): Adjust interface.
* src/riscv/ffitarget.h (FFI_GO_CLOSURES): Define.
* src/riscv/sysv.S (ffi_go_closure_asm): New function.
(ffi_closure_asm, ffi_call_asm): Update for adjusted interfaces.

From-SVN: r263550

i386.c (expand_vec_perm_movs): New method matching movs patterns.

* config/i386/i386.c (expand_vec_perm_movs): New method matching movs
patterns.
(expand_vec_perm_1): Try the new method.

* gcc.target/i386/sse2-movs.c: New test.

From-SVN: r263549

re PR target/86547 (s390x: Maximum number of LRA assignment passes is achieved (30) when compiling a small inline assembler snippet)

        PR target/86547
* lra-lives.c (remove_some_program_points_and_update_live_ranges):
        Check whether lra_live_max_point is 0 before dividing.

From-SVN: r263548

Daily bump.

From-SVN: r263547

Include <sys/random.h> for getentropy on Solaris

* configure.ac: Check for <sys/random.h>.
* configure, config.h.in: Regenerate.
* intrinsics/random.c [HAVE_SYS_RANDOM_H]: Include <sys/random.h>.

From-SVN: r263543

PR libstdc++/86954 use non-placement operator delete

As explained in the PR, there's no reason to call the nothrow delete,
we can just use the normal one.

PR libstdc++/86954
* include/bits/stl_tempbuf.h (return_temporary_buffer): Use
non-placement delete.

From-SVN: r263542

PR tree-optimization/86650 - -Warray-bounds missing inlining context

gcc/ChangeLog:

PR tree-optimization/86650
* tree-vrp.c (vrp_prop::check_array_ref): Print an inform message.
(vrp_prop::check_mem_ref): Same.

gcc/testsuite/ChangeLog:

PR tree-optimization/86650
* gcc.dg/Warray-bounds-34.c: New test.

From-SVN: r263541

re PR fortran/86116 (Ambiguous generic interface not recognised)

2018-08-14  Janus Weil  <janus@gcc.gnu.org>

PR fortran/86116
* interface.c (compare_type): Remove a CLASS/TYPE check.
(compare_type_characteristics): New function that behaves like the old
'compare_type'.
(gfc_check_dummy_characteristics, gfc_check_result_characteristics):
Call 'compare_type_characteristics' instead of 'compare_type'.

2018-08-14  Janus Weil  <janus@gcc.gnu.org>

PR fortran/86116
* gfortran.dg/generic_34.f90: New test case.

From-SVN: r263540

Simplify overflow checks in duration literals

* include/std/chrono (__check_overflow): Simplify definition.
(_Checked_integral_constant): Remove.

From-SVN: r263537

PR libstdc++/86846 Alternative to pointer-width atomics

Define a class using std::mutex for when std::atomic<memory_resource*>
cannot be used to implement the default memory resource.

When std::mutex constructor is not constexpr the constant_init trick
won't work, so just define a global and use init_priority for it. The
compiler warns about using reserved priority, so put the definition in a
header file using #pragma GCC system_header to suppress the warning.

PR libstdc++/86846
* src/c++17/default_resource.h: New file, defining default_res.
* src/c++17/memory_resource.cc [ATOMIC_POINTER_LOCK_FREE != 2]
(atomic_mem_res): Define alternative for atomic<memory_resource*>
using a mutex instead of atomics.

From-SVN: r263536

PR libstdc++/85343 overload __throw_ios_failure to take errno

[ios::failure] p2: "When throwing ios_base::failure exceptions,
implementations should provide values of ec that identify the specific
reason for the failure."

This adds a new overload of __throw_ios_failure that can be passed
errno, to store error_code(errno, system_category()) in the exception
object.

PR libstdc++/85343
* acinclude.m4 (libtool_VERSION): Bump version.
* config/abi/pre/gnu.ver (GLIBCXX_3.4.26): Add new symbol version.
Export new symbol.
* configure: Regenerate.
* doc/xml/manual/abi.xml: Document new versions.
* include/bits/fstream.tcc (basic_filebuf<C, T>::underflow)
(basic_filebuf<C, T>::xsgetn): Pass errno to __throw_ios_failure.
* include/bits/functexcept.h (__throw_ios_failure(const char*, int)):
Declare new overload.
* src/c++11/cxx11-ios_failure.cc (__ios_failure): Add new constructor
and static member function.
(__throw_ios_failure(const char*, int)): Define.
* src/c++98/ios_failure.cc [!_GLIBCXX_USE_DUAL_ABI]
(__throw_ios_failure(const char*, int)): Define.
* testsuite/util/testsuite_abi.cc: Update known and latest versions.

From-SVN: r263535

Rope iterators: don't retain pointers when copied

Rope iterators sometimes contain pointers to an internal buffer
inside the iterator itself.  When such an iterator is copied, the
copy incorrectly retains pointers to the original.

This patch takes the simple approach of not copying the cached
information when the internal buffer is being used, instead
requiring it to be recomputed when the copied iterator is
dereferenced.  An alternative would be to adjust the pointers so
they refer to the buffer in the copy.

2018-08-14  Jeremy Sawicki  <jeremy-gcc@sawicki.us>

* include/ext/rope (_Rope_iterator_base(const _Rope_iterator_base&))
(_Rope_const_iterator::operator=(const _Rope_const_iterator&))
(_Rope_iterator::operator=(const _Rope_iterator&)): Ensure
copied/assigned rope iterators don't retain pointers to the iterator
they were copied/assigned from.
* testsuite/ext/rope/7.cc: New.

From-SVN: r263534

pretty-print.c (eat_esc_sequence): Swap the foreground and background colors if...

* gcc/pretty-print.c (eat_esc_sequence): Swap the foreground and background
colors if the COMMON_LVB_REVERSE_VIDEO flag is set, and clear it thereafter,
as it only works for DBCS.

From-SVN: r263531

pretty-print.c (mingw_ansi_fputs): Do not call _close() on the handle returned by _get_osf_handle().

* gcc/pretty-print.c (mingw_ansi_fputs): Do not call _close() on the handle
returned by _get_osf_handle().

From-SVN: r263530

Daily bump.

From-SVN: r263527

re PR c++/57891 (No diagnostic of narrowing conversion in non-type template argument)

PR c++/57891
* call.c (struct conversion): Add check_narrowing_const_only.
(build_converted_constant_expr): Set check_narrowing and
check_narrowing_const_only.  Give error if expr is error node.
(convert_like_real): Pass it to check_narrowing.
* cp-tree.h (check_narrowing): Add a default parameter.
* decl.c (compute_array_index_type): Use input_location instead of
location_of.
* pt.c (convert_nontype_argument): Return NULL_TREE if tf_error.
* typeck2.c (check_narrowing): Don't warn for instantiation-dependent
expressions.  Call maybe_constant_value instead of
fold_non_dependent_expr.  Don't mention { } in diagnostic.  Only check
narrowing for constants if CONST_ONLY.

* g++.dg/cpp0x/Wnarrowing6.C: New test.
* g++.dg/cpp0x/Wnarrowing7.C: New test.
* g++.dg/cpp0x/Wnarrowing8.C: New test.
* g++.dg/cpp0x/Wnarrowing9.C: New test.
* g++.dg/cpp0x/Wnarrowing10.C: New test.
* g++.dg/cpp0x/constexpr-47969.C: Adjust dg-error.
* g++.dg/cpp0x/constexpr-ex2.C: Likewise.
* g++.dg/cpp0x/constexpr-targ.C: Likewise.
* g++.dg/cpp0x/scoped_enum2.C: Likewise.
* g++.dg/ext/stmtexpr15.C: Likewise.
* g++.dg/gomp/pr47963.C: Likewise.
* g++.dg/init/new37.C: Likewise.
* g++.dg/init/new43.C: Likewise.
* g++.dg/other/fold1.C: Likewise.
* g++.dg/parse/array-size2.C: Likewise.
* g++.dg/template/dependent-name3.C: Likewise.
* g++.dg/cpp0x/constexpr-data2.C: Add dg-error.
* g++.dg/other/vrp1.C: Likewise.
* g++.dg/template/char1.C: Likewise.

From-SVN: r263523

Use getentropy() for seeding PRNG

The getentropy function, found on Linux, OpenBSD, and recently also
FreeBSD, can be used to get random bytes to initialize the PRNG.  It
is similar to the traditional way of reading from /dev/urandom, but
being a system call rather than a special file, it doesn't suffer from
problems like running out of file descriptors, or failure when running
in a container where /dev/urandom may not be available.

Regtested on x86_64-pc-linux-gnu, Ok for trunk?

2018-08-13  Janne Blomqvist  <jb@gcc.gnu.org>

* configure.ac: Check for getentropy.
* intrinsics/random.c (getosrandom): Use getentropy if available.
* config.h.in: Regenerated.
* configure: Regenerated.

From-SVN: r263522

rs6000: Fix pr56605.c

After the combine 2-2 changes, this testcase does not have a ZERO_EXTEND
in the intermediate code, but an AND instead.

gcc/testsuite/
* gcc.target/powerpc/pr56605.c: The generated code can have an AND
instead of a ZERO_EXTEND.

From-SVN: r263521

rs6000.c (rs6000_gimple_fold_builtin): Add support for folding vec_perm.

[gcc]

2018-07-06  Will Schmidt  <will_schmidt@vnet.ibm.com>

* gcc/config/rs6000/rs6000.c (rs6000_gimple_fold_builtin): Add support
for folding vec_perm.

From-SVN: r263520

rs6000.c (rs6000_gimple_fold_builtin): Add support for gimple-folding of vec_pack() and vec_unpack() intrinsics.

[gcc]

2018-08-13 Will Schmidt  <will_schmidt@vnet.ibm.com>

* config/rs6000/rs6000.c (rs6000_gimple_fold_builtin):
Add support for gimple-folding of vec_pack() and vec_unpack()
intrinsics.

From-SVN: r263519

PR libstdc++/45093 avoid warnings for _M_destroy_node

PR libstdc++/45093
* include/bits/stl_tree.h (_Rb_tree::_M_destroy_node(_Link_type)):
Combine definitions to avoid --detect-odr-violations warning.

From-SVN: r263516

Minor optimisations in operator new(size_t, align_val_t)

* libsupc++/new_opa.cc (operator new(size_t, align_val_t)): Use
__is_pow2 to check for valid alignment. Avoid branching when rounding
size to multiple of alignment.

From-SVN: r263515

Add <bit> and <version> to freestanding headers

* include/Makefile.am: Install <bit> and <version> for freestanding.
* include/Makefile.in: Regenerate.
* testsuite/17_intro/freestanding.cc: Check for <bit> and <version>.

From-SVN: r263514

Revert "libstdc++-v3: Have aligned_alloc() on Newlib"

This reverts commit r263461 / 2e920cd849b3cf0a72df4f172e27676a3e70b73f
because aligned_alloc is not defined for baremetal newlib targets, see
https://gcc.gnu.org/ml/libstdc++/2018-08/msg00065.html

Revert
2018-08-10  Sebastian Huber  <sebastian.huber@embedded-brains.de>

PR target/85904
* configure.ac: Define HAVE_ALIGNED_ALLOC if building for
Newlib.
* configure: Regenerate.

From-SVN: r263513

rs6000.c (rs6000_builtin_valid_without_lhs): Add vec_xst variants to the list.

[gcc]

2018-08-13 Will Schmidt <will_schmidt@vnet.ibm.com>

* config/rs6000/rs6000.c (rs6000_builtin_valid_without_lhs): Add
vec_xst variants to the list.
(rs6000_gimple_fold_builtin): Add support for folding unaligned
vector loads and stores.

From-SVN: r263512

PR tree-optimization/71625 - missing strlen optimization on different array initialization style

gcc/c/ChangeLog:

PR tree-optimization/71625
* c-parser.c (c_parser_declaration_or_fndef): Call
braced_list_to_string.

gcc/c-family/ChangeLog:

PR tree-optimization/71625
* c-common.c (braced_list_to_string): New function.
* c-common.h (braced_list_to_string): Declare it.

gcc/cp/ChangeLog:

PR tree-optimization/71625
* decl.c (check_initializer):  Call braced_list_to_string.
(eval_check_narrowing): New function.
* gcc/cp/typeck2.c (digest_init_r): Accept strings literals
as initilizers for all narrow character types.

gcc/testsuite/ChangeLog:

PR tree-optimization/71625

* g++.dg/init/string2.C: New test.
* g++.dg/init/string3.C: New test.
* g++.dg/init/string4.C: New test.
* gcc.dg/init-string-3.c: New test.
* gcc.dg/strlenopt-55.c: New test.
* gcc.dg/strlenopt-56.c: New test.

From-SVN: r263511

P0806R2 - Deprecate implicit capture of this via [=]

P0806R2 - Deprecate implicit capture of this via [=]
* lambda.c (add_default_capture): Formatting fixes.  Warn about
deprecated implicit capture of this via [=].

* g++.dg/cpp2a/lambda-this1.C: New test.
* g++.dg/cpp2a/lambda-this2.C: New test.
* g++.dg/cpp2a/lambda-this3.C: New test.

From-SVN: r263508

re PR c++/86915 (Segmentation fault for an array of auto in template parameter)

PR c++/86915
* decl.c (create_array_type_for_decl): Handle null name.

* g++.dg/diagnostic/auto1.C: New test.

From-SVN: r263507

config.gcc (rs6000-ibm-aix4.x): Delete.

        * config.gcc (rs6000-ibm-aix4.x): Delete.
        (rs6000-ibm-aix5.1): Delete.
        (rs6000-ibm-aix5.2): Delete.
        (rs6000-ibm-aix5.3): Delete.
        * config/rs6000/aix43.h: Delete.
        * config/rs6000/aix51.h: Delete.
        * config/rs6000/aix52.h: Delete.
        * config/rs6000/t-aix43: Delete.

From-SVN: r263506

[nvptx] Use CUDA driver API to select default runtime launch geometry

The CUDA driver API starting version 6.5 offers a set of runtime functions to
calculate several occupancy-related measures, as a replacement for the occupancy
calculator spreadsheet.

This patch adds a heuristic for default runtime launch geometry, based on the
new runtime function cuOccupancyMaxPotentialBlockSize.

Build on x86_64 with nvptx accelerator and ran libgomp testsuite.

2018-08-13  Cesar Philippidis  <cesar@codesourcery.com>
    Tom de Vries  <tdevries@suse.de>

PR target/85590
* plugin/cuda/cuda.h (CUoccupancyB2DSize): New typedef.
(cuOccupancyMaxPotentialBlockSize): Declare.
* plugin/cuda-lib.def (cuOccupancyMaxPotentialBlockSize): New
CUDA_ONE_CALL_MAYBE_NULL.
* plugin/plugin-nvptx.c (CUDA_VERSION < 6050): Define
CUoccupancyB2DSize and declare
cuOccupancyMaxPotentialBlockSize.
(nvptx_exec): Use cuOccupancyMaxPotentialBlockSize to set the
default num_gangs and num_workers when the driver supports it.

Co-Authored-By: Tom de Vries <tdevries@suse.de>
From-SVN: r263505

S/390: Factor out constant pool ref decomposition

gcc/ChangeLog:

2018-08-13  Ilya Leoshkevich  <iii@linux.ibm.com>

* config/s390/s390.c (s390_decompose_constant_pool_ref):
        New function.
(s390_decompose_address): Factor out constant pool ref
        decomposition.

From-SVN: r263504

Daily bump.

From-SVN: r263503

re PR fortran/66679 ([OOP] ICE with class(*) and transfer)

2018-08-12  Paul Thomas  <pault@gcc.gnu.org>

PR fortran/66679
* trans-intrinsic.c (gfc_conv_intrinsic_transfer): Class array
elements are returned as references to the data element. Get
the class expression by stripping back the references. Use this
for the element size.

2018-08-12  Paul Thomas  <pault@gcc.gnu.org>

PR fortran/66679
* gfortran.dg/transfer_class_3.f90: New test.

From-SVN: r263499

[NDS32] Change return type of predicator to bool.

gcc/
* config/nds32/nds32-predicates.c
(nds32_can_use_bclr_p): Change return type as bool.
(nds32_can_use_bset_p): Ditto.
(nds32_can_use_btgl_p): Ditto.
(nds32_can_use_bitci_p): Ditto.
* config/nds32/nds32-protos.h
(nds32_can_use_bclr_p): Change declaration.
(nds32_can_use_bset_p): Ditto.
(nds32_can_use_btgl_p): Ditto.
(nds32_can_use_bitci_p): Ditto.

From-SVN: r263497

[NDS32] Add new option: -msched-prolog-epilog

gcc/
* config/nds32/nds32.c (nds32_expand_prologue, nds32_expand_epilogue):
Support -msched-prolog-epilog option.
* config/nds32/nds32.opt (msched-prolog-epilog): New option.

From-SVN: r263496

[NDS32] Enable -malways-align by default at -O1 and above.

gcc/
* common/config/nds32/nds32-common.c
(nds32_option_optimization_table): Enalbe -malways-align.

From-SVN: r263495

re PR fortran/86906 (erroneous name clash with renaming in use statement)

2018-08-12  Paul Thomas  <pault@gcc.gnu.org>

PR fortran/86906
* resolve.c (resolve_fl_variable_derived): Check if the derived
type is use associated before checking for the host association
error.

2018-08-12  Paul Thomas  <pault@gcc.gnu.org>

PR fortran/86906
* gfortran.dg/use_rename_9.f90: New test.

From-SVN: r263494

[NDS32] Implement more C ISR extension.

gcc/
* config.gcc (nds32*): Add nds32_isr.h and nds32_init.inc in
extra_headers.
* common/config/nds32/nds32-common.c (nds32_handle_option): Handle
OPT_misr_secure_ case.
* config/nds32/nds32-isr.c: Implementation of backward compatibility.
* config/nds32/nds32-protos.h (nds32_isr_function_critical_p): New.
* config/nds32/nds32.c (nds32_attribute_table): Add critical and
secure attribute.
* config/nds32/nds32.h (nds32_isr_nested_type): Add NDS32_CRITICAL.
(nds32_isr_info): New field security_level.
(TARGET_ISR_VECTOR_SIZE_4_BYTE): New macro.
* config/nds32/nds32.md (return_internal): Consider critical attribute.
* config/nds32/nds32.opt (misr-secure): New option.
* config/nds32/nds32_init.inc: New file.
* config/nds32/nds32_isr.h: New file.

libgcc/
* config/nds32/t-nds32-isr: Rearrange object dependency.
* config/nds32/initfini.c: Add dwarf2 unwinding support.
* config/nds32/isr-library/adj_intr_lvl.inc: Consider new extensions
and registers usage.
* config/nds32/isr-library/excp_isr.S: Ditto.
* config/nds32/isr-library/intr_isr.S: Ditto.
* config/nds32/isr-library/reset.S: Ditto.
* config/nds32/isr-library/restore_all.inc: Ditto.
* config/nds32/isr-library/restore_mac_regs.inc: Ditto.
* config/nds32/isr-library/restore_partial.inc: Ditto.
* config/nds32/isr-library/restore_usr_regs.inc: Ditto.
* config/nds32/isr-library/save_all.inc: Ditto.
* config/nds32/isr-library/save_mac_regs.inc: Ditto.
* config/nds32/isr-library/save_partial.inc: Ditto.
* config/nds32/isr-library/save_usr_regs.inc: Ditto.
* config/nds32/isr-library/vec_vid*.S: Consider 4-byte vector size.

From-SVN: r263493

Daily bump.

From-SVN: r263492

pa.md (UNSPEC_MEMORY_BARRIER): New unspec enum.

gcc
* config/pa/pa.md (UNSPEC_MEMORY_BARRIER): New unspec enum.
Update comment for atomic instructions.
(atomic_storeqi, atomic_storehi, atomic_storesi, atomic_storesf,
atomic_loaddf, atomic_loaddf_1, atomic_storedf, atomic_storedf_1):
Remove.
(atomic_loaddi): Revise fence expansion to only emit fence prior to
load for __ATOMIC_SEQ_CST model.
(atomic_loaddi_1): Remove float register target.
(atomic_storedi): Handle CONST_INT values.
(atomic_storedi_1): Remove float register source.  Add special case
for zero value.
(memory_barrier): New expander and insn.

libgcc
* config/pa/linux-atomic.c: Update comment.
(FETCH_AND_OP_2, OP_AND_FETCH_2, FETCH_AND_OP_WORD, OP_AND_FETCH_WORD,
COMPARE_AND_SWAP_2, __sync_val_compare_and_swap_4,
SYNC_LOCK_TEST_AND_SET_2, __sync_lock_test_and_set_4): Use
__ATOMIC_RELAXED for atomic loads.
(SYNC_LOCK_RELEASE_1): New define.  Use __sync_synchronize() and
unordered store to release lock.
(__sync_lock_release_8): Likewise.
(SYNC_LOCK_RELEASE_2): Remove define.

From-SVN: r263488

re PR tree-optimization/86835 (Bogus "is used uninitialized" warning with -ffast-math)

PR tree-optimization/86835
* tree-ssa-math-opts.c (insert_reciprocals): Even when inserting
new_stmt after def_gsi, make sure to insert new_square_stmt after
that stmt, not 2 stmts before it.

* gcc.dg/pr86835.c: New test.

From-SVN: r263487

PR c++/86728 - C variadic generic lambda.

* parser.c (cp_parser_parameter_declaration): Don't turn 'auto' into
a pack if it's followed by a declarator-id.

From-SVN: r263484

Daily bump.

From-SVN: r263483

PR libstdc++/68210 adjust operator new and delete for LWG 206

Ensure that nothrow versions of new and delete call the ordinary
versions of new or delete, instead of calling malloc or free directly.

These files are all compiled with -std=gnu++14 so can use noexcept and
nullptr to make the code more readable.

PR libstdc++/68210
* doc/xml/manual/intro.xml: Document LWG 206 change.
* libsupc++/del_op.cc: Replace _GLIBCXX_USE_NOEXCEPT with noexcept.
* libsupc++/del_opa.cc: Likewise.
* libsupc++/del_opant.cc: Likewise.
* libsupc++/del_opnt.cc: Likewise. Call operator delete(ptr) instead
of free(ptr).
* libsupc++/del_ops.cc: Replace _GLIBCXX_USE_NOEXCEPT with noexcept.
* libsupc++/del_opsa.cc: Likewise.
* libsupc++/del_opva.cc: Likewise.
* libsupc++/del_opvant.cc: Likewise.
* libsupc++/del_opvnt.cc: Likewise. Call operator delete[](ptr)
instead of operator delete(ptr).
* libsupc++/del_opvs.cc: Replace _GLIBCXX_USE_NOEXCEPT with noexcept.
* libsupc++/del_opvsa.cc: Likewise.
* libsupc++/new_op.cc: Use __builtin_expect in check for zero size.
* libsupc++/new_opa.cc: Use nullptr instead of literal 0.
* libsupc++/new_opant.cc: Likewise. Replace _GLIBCXX_USE_NOEXCEPT
with noexcept.
* libsupc++/new_opnt.cc: Likewise. Call operator new(sz) instead of
malloc(sz).
* libsupc++/new_opvant.cc: Use nullptr and noexcept.
* libsupc++/new_opvnt.cc: Likewise. Call operator new[](sz) instead of
operator new(sz, nothrow).
* testsuite/18_support/new_nothrow.cc: New test.

From-SVN: r263478

Fix formatting in ChangeLog

From-SVN: r263476

re PR fortran/57160 (short-circuit IF only with -ffrontend-optimize)

2018-08-10  Janus Weil  <janus@gcc.gnu.org>

PR fortran/57160
* invoke.texi (frontend-optimize): Mention short-circuiting.
* options.c (gfc_post_options): Disable -ffrontend-optimize with -Og.
* resolve.c (resolve_operator): Warn about short-circuiting only with
-ffrontend-optimize.
* trans-expr.c (gfc_conv_expr_op): Use short-circuiting operators only
with -ffrontend-optimize. Without that flag, make sure that both
operands are evaluated.

2018-08-10  Janus Weil  <janus@gcc.gnu.org>

PR fortran/57160
* gfortran.dg/actual_pointer_function_1.f90: Fix invalid test case.
* gfortran.dg/inline_matmul_23.f90: Add option "-ffrontend-optimize".
* gfortran.dg/short_circuiting_2.f90: New test case.
* gfortran.dg/short_circuiting_3.f90: New test case.

From-SVN: r263471

Remove not needed __builtin_expect due to malloc predictor.

2018-08-10  Martin Liska  <mliska@suse.cz>

* libsupc++/new_op.cc (new): Remove __builtin_expect as malloc
        predictor can handle that.
* libsupc++/new_opa.cc: Likewise.
* libsupc++/new_opnt.cc (new): Likewise.

From-SVN: r263470

Fix wrongly removed line.

2018-08-10  Martin Liska  <mliska@suse.cz>

* MAINTAINERS: Revert change in previous commit and
        join lines.

From-SVN: r263469

Fix extra line in MAINTAINERS.

2018-08-10  Martin Liska  <mliska@suse.cz>

* MAINTAINERS: Remove extra line.

From-SVN: r263468

i386: do not use SImode mul-highpart on 64-bit (PR 82418)

PR target/82418
* config/i386/i386.md (<s>mul<mode>3_highpart): Use DWIH mode iterator
instead of SWI48.

testsuite/
* gcc.target/i386/pr82418.c: New test.

From-SVN: r263467

Introduce __builtin_expect_with_probability (PR target/83610).

2018-08-10  Martin Liska  <mliska@suse.cz>

        PR target/83610
* builtin-types.def (BT_FN_LONG_LONG_LONG_DOUBLE): Add new
        function type.
* builtins.c (expand_builtin_expect_with_probability):
        New function.
(expand_builtin_expect_with_probability): New function.
(build_builtin_expect_predicate): Add new argumnet probability
        for BUILT_IN_EXPECT_WITH_PROBABILITY.
(fold_builtin_expect):
(fold_builtin_2):
(fold_builtin_3):
* builtins.def (BUILT_IN_EXPECT_WITH_PROBABILITY):
* builtins.h (fold_builtin_expect): Set new argument.
* doc/extend.texi: Document __builtin_expect_with_probability.
* doc/invoke.texi: Likewise.
* gimple-fold.c (gimple_fold_call): Pass new argument.
* ipa-fnsummary.c (find_foldable_builtin_expect): Handle
        also BUILT_IN_EXPECT_WITH_PROBABILITY.
* predict.c (get_predictor_value): New function.
(expr_expected_value): Add new argument probability. Assume
        that predictor and probability are always non-null.
(expr_expected_value_1): Likewise.  For __builtin_expect and
        __builtin_expect_with_probability set probability.  Handle
        combination in binary expressions.
(tree_predict_by_opcode): Simplify code by simply calling
        get_predictor_value.
(pass_strip_predict_hints::execute): Add handling of
        BUILT_IN_EXPECT_WITH_PROBABILITY.
* predict.def (PRED_BUILTIN_EXPECT_WITH_PROBABILITY): Add
        new predictor.
* tree.h (DECL_BUILT_IN_P): New function.
2018-08-10  Martin Liska  <mliska@suse.cz>

        PR target/83610
* gcc.dg/predict-17.c: New test.
* gcc.dg/predict-18.c: New test.
* gcc.dg/predict-19.c: New test.

From-SVN: r263466

Strip only selected predictors after early tree passes (PR tree-optimization/85799).

2018-08-10  Martin Liska  <mliska@suse.cz>

        PR tree-optimization/85799
* passes.def: Add argument for pass_strip_predict_hints.
* predict.c (class pass_strip_predict_hints): Add new argument
        early_p.
(strip_predictor_early): New function.
(pass_strip_predict_hints::execute): Call the function to
        strip predictors.
(strip_predict_hints): New function.
* predict.def: Fix comment.
2018-08-10  Martin Liska  <mliska@suse.cz>

        PR tree-optimization/85799
* gcc.dg/pr85799.c: New test.

From-SVN: r263465

Clarify source of tm.texi to copy for GFDL grant

When tm.texi.in is updated in the source tree, the following message
gets displayed:

Verify that you have permission to grant a GFDL license for all
new text in tm.texi, then copy it to <gcc src dir>/gcc/doc/tm.texi.

Having been myself and some colleagues confused several time by that
message as to what tm.texi to copy, I think it would be clearer to
indicate the absolute path for the source as well. This patch achieves
that.

2018-08-10  Thomas Preud'homme  <thomas.preudhomme@linaro.org>

    gcc/
    * Makefile.in: Clarify which tm.texi to copy over to assert the
    right to grant a GFDL license for all.

From-SVN: r263464

libstdc++-v3: Have aligned_alloc() on Newlib

While building for Newlib, some configure checks must be hard coded.
The aligned_alloc() is supported since 2015 in Newlib.

libstdc++-v3/

PR target/85904
* configure.ac: Define HAVE_ALIGNED_ALLOC if building for
Newlib.
* configure: Regenerate.

From-SVN: r263461

Daily bump.

From-SVN: r263460

Define aliases for containers using polymorphic_allocator

These aliases are placed in the top-level header, e.g. <vector> not
<bits/stl_vector.h>. This ensures that they refer to whichever of
std::vector or __debug::vector or __profile::vector is in use when the
header is included.

* include/std/deque (std::pmr::deque): Declare alias.
* include/std/forward_list (std::pmr::forward_list): Likewise.
* include/std/list (std::pmr::list): Likewise.
* include/std/map (std::pmr::map, std::pmr::multimap): Likewise.
* include/std/regex (std::pmr::match_results, std::pmr::cmatch)
(std::pmr::smatch, std::pmr::wcmatch, std::pmr::wsmatch): Likewise.
* include/std/set (std::pmr::set, std::pmr::multiset): Likewise.
* include/std/string (std::pmr::basic_string, std::pmr::string)
(std::pmr::u16string, std::pmr::u32string, std::pmr::wstring):
Likewise.
* include/std/unordered_map (std::pmr::unordered_map)
(std::pmr::unordered_multimap): Likewise.
* include/std/unordered_set (std::pmr::unordered_set)
(std::pmr::unordered_multiset): Likewise.
* include/std/vector (std::pmr::vector): Likewise.
* testsuite/21_strings/basic_string/types/pmr_typedefs.cc: New test.
* testsuite/23_containers/deque/types/pmr_typedefs.cc: New test.
* testsuite/23_containers/forward_list/pmr_typedefs.cc: New test.
* testsuite/23_containers/list/pmr_typedefs.cc: New test.
* testsuite/23_containers/map/pmr_typedefs.cc: New test.
* testsuite/23_containers/multimap/pmr_typedefs.cc: New test.
* testsuite/23_containers/multiset/pmr_typedefs.cc: New test.
* testsuite/23_containers/set/pmr_typedefs.cc: New test.
* testsuite/23_containers/unordered_map/pmr_typedefs.cc: New test.
* testsuite/23_containers/unordered_multimap/pmr_typedefs.cc: New
test.
* testsuite/23_containers/unordered_multiset/pmr_typedefs.cc: New
test.
* testsuite/23_containers/unordered_set/pmr_typedefs.cc: New test.
* testsuite/23_containers/vector/pmr_typedefs.cc: New test.
* testsuite/28_regex/match_results/pmr_typedefs.cc: New test.

From-SVN: r263456

m68k.c (m68k_adjust_decorated_operand): Remove unused variable.

* config/m68k/m68k.c (m68k_adjust_decorated_operand): Remove
unused variable.

From-SVN: r263455

re PR middle-end/86897 (gcc.dg/uninit-suppress_2.c FAILs)

PR middle-end/86897
* gcc.dg/uninit-suppress_2.c: Disable DOM.

From-SVN: r263454