compiler: revise exportdata fix for processing constant types

    This patch is an addendum to the fix for issue 34577, which was not
    sufficiently general. During export data processing, when looking at
    the types of constants mentioned in inlinable function bodies, include
    both locally defined constants and constant imported from other
    packages.

    Testcase for this bug is in CL 201017.

    Fixes golang/go#34852.

    Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/201018

From-SVN: r276976

re PR fortran/92004 (Rejection of different ranks for dummy array argument where actual argument is an element)

2019-10-14  Thomas Koenig  <tkoenig@gcc.gnu.org>

PR fortran/92004
* array.c (expand_constructor): Set from_constructor on
expression.
* gfortran.h (gfc_symbol): Add maybe_array.
(gfc_expr): Add from_constructor.
* interface.c (maybe_dummy_array_arg): New function.
(compare_parameter): If the formal argument is generated from a
call, check the conditions where an array element could be
passed to an array.  Adjust error message for assumed-shape
or pointer array.  Use correct language for assumed shaped arrays.
(gfc_get_formal_from_actual_arglist): Set maybe_array on the
symbol if the actual argument is an array element fulfilling
the conditions of 15.5.2.4.

2019-10-14  Thomas Koenig  <tkoenig@gcc.gnu.org>

PR fortran/92004
* gfortran.dg/argument_checking_24.f90: New test.
* gfortran.dg/abstract_type_6.f90: Add error message.
* gfortran.dg/argument_checking_11.f90: Correct wording
in error message.
* gfortran.dg/argumeent_checking_13.f90: Likewise.
* gfortran.dg/interface_40.f90: Add error message.

From-SVN: r276972

* gcc.c-torture/compile/pr85401: New test.

From-SVN: r276971

* doc/tree-ssa.texi: Update renamed macro name.

From-SVN: r276970

mips.c (mips_cannot_force_const_mem): Reject vector constants.

* config/mips/mips.c (mips_cannot_force_const_mem): Reject
vector constants.

* gcc.target/mips/constant-spill.c: New test.

From-SVN: r276969

PR c++/91930 - ICE with constrained inherited default ctor.

The testcase was crashing because lazily_declare_fn was failing to add a
defaulted constructor, because the implicit declaration was less constrained
than the inherited default constructor.  But when we have an inherited
constructor, we shouldn't be trying to declare a default constructor in the
first place, because it counts as "a user-declared constructor".  With that
fixed I needed to adjust a couple of inherited constructor testcases that
previously had been diagnosing the default constructor as deleted rather
than not declared.

* name-lookup.c (do_class_using_decl): Set TYPE_HAS_USER_CONSTRUCTOR
for inherited constructor.

From-SVN: r276968

[Darwin, machopic 9/n] Minor code clean-ups.

Improve some comments, replace some asserts that have been in the code
base for years with checking-asserts.

gcc/ChangeLog:

2019-10-14  Iain Sandoe  <iain@sandoe.co.uk>

* config/darwin.c: Use unsigned ints for the picbase label
counters, initialise the vars explicitly.
(update_pic_label_number_if_needed): Move a variable declaration
to where it's needed.
(machopic_output_function_base_name): Use a more strict checking
assert, and and unsigned int for the picbase label counter.
(machopic_get_function_picbase): Likewise.

From-SVN: r276967

runtime: correct facilities names in s390 CPU support

    Patch from Andreas Krebbel.

    Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/201038

From-SVN: r276964

re PR middle-end/92046 (Command line options (that are per-functions) are affecting --params which are global.)

2019-10-14  Richard Biener  <rguenther@suse.de>

PR middle-end/92046
* dse.c (scan_insn): Use param max_active_local_stores.
(dse_step1): Get PARAM_MAX_DSE_ACTIVE_LOCAL_STORES and adjust
based on optimization level.
* loop-invariant.c (move_loop_invariants): Adjust
LOOP_INVARIANT_MAX_BBS_IN_LOOP based on optimization level.
* opts.c (default_options_optimization): Do not adjust
PARAM_MAX_DSE_ACTIVE_LOCAL_STORES and
LOOP_INVARIANT_MAX_BBS_IN_LOOP here.

From-SVN: r276963

internal/cpu: define kdsaQuery for s390

    Patch from Andreas Krebbel.

    Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/201037

From-SVN: r276962

[ARM] Enable arm_legitimize_address for Thumb-2

Currently arm_legitimize_address doesn't handle Thumb-2 at all, resulting in
inefficient code.  Since Thumb-2 supports similar address offsets use the Arm
legitimization code for Thumb-2 to get significant codesize and performance
gains.  SPECINT2006 shows 0.4% gain on Cortex-A57, while SPECFP improves 0.2%.

    gcc/
* config/arm/arm.c (arm_legitimize_address): Remove Thumb-2 bailout.

From-SVN: r276961

[ARM] Switch to default sched pressure algorithm

Currently the Arm backend selects the alternative sched pressure algorithm.
The issue is that this doesn't take register pressure into account, and so
it causes significant additional spilling on Arm where there are only 14
allocatable registers.  Building SPEC2006 showed significant codesize gains
with the default pressure algorithm, so switch back to that.  PR77308 shows
~800 fewer instructions.

SPECINT2006 is ~0.6% faster on Cortex-A57 together with the other DImode
patches. Overall SPEC codesize is 1.1% smaller.

    gcc/
* config/arm/arm.c (arm_option_override): Don't override sched
pressure algorithm.

From-SVN: r276960

re PR tree-optimization/92069 (ice in vect_analyze_scalar_cycles_1, at tree-vect-loop.c:560)

2019-10-14  Richard Biener  <rguenther@suse.de>

PR tree-optimization/92069
* tree-vect-loop.c (vect_analyze_scalar_cycles_1): For nested
cycles do not set vect_nested_cycle on the latch definition.

* gcc.dg/torture/pr92069.c: New testcase.

From-SVN: r276959

decl.c (check_tag_decl): Use DECL_SOURCE_LOCATION.

/cp
2019-10-14  Paolo Carlini  <paolo.carlini@oracle.com>

* decl.c (check_tag_decl): Use DECL_SOURCE_LOCATION.

/testsuite
2019-10-14  Paolo Carlini  <paolo.carlini@oracle.com>

* g++.dg/cpp0x/constexpr-union5.C: Test location(s) too.
* g++.dg/diagnostic/bitfld2.C: Likewise.
* g++.dg/ext/anon-struct1.C: Likewise.
* g++.dg/ext/anon-struct6.C: Likewise.
* g++.dg/ext/flexary19.C: Likewise.
* g++.dg/ext/flexary9.C: Likewise.
* g++.dg/template/error17.C: Likewise.

From-SVN: r276958

re PR libgomp/92081 (FAIL: libgomp.fortran/target-simd.f90  execution test)

PR libgomp/92081
* testsuite/libgomp.fortran/target-simd.f90: Iterate from 1 rather
than 0.

From-SVN: r276956

Fix previous commit

        * error.c: Remove debug pragma added in previous commit.

From-SVN: r276955

re PR c++/92084 (ICE: tree check: expected tree that contains 'decl minimal' structure, have 'compound_expr' in gimplify_scan_omp_clauses, at gimplify.c:9039)

PR c++/92084
* semantics.c (handle_omp_array_sections_1): Temporarily disable
-fstrong-eval-order also for in_reduction and task_reduction clauses.

* g++.dg/gomp/pr92084.C: New test.

From-SVN: r276954

[Fortran] PR 92072 – fix %C corner case

        PR fortran/92072
        * error.c (error_print, gfc_format_decoder): Fix %C column-
        offset handling.

From-SVN: r276953

Add expr_callee_abi

This turned out to be useful for the SVE PCS support, and is a natural
tree-level analogue of insn_callee_abi.

2019-10-14  Richard Sandiford  <richard.sandiford@arm.com>

gcc/
* function-abi.h (expr_callee_abi): Declare.
* function-abi.cc (expr_callee_abi): New function.

From-SVN: r276952

[C] Avoid exposing internal details in aka types

The current aka diagnostics can sometimes leak internal details that
seem more likely to be distracting than useful.  E.g. on aarch64:

  void f (va_list *va) { *va = 1; }

gives:

  incompatible types when assigning to type ‘va_list’ {aka ‘__va_list’} from type ‘int’

where __va_list isn't something the user is expected to know about.
A similar thing happens for C++ on the arm_neon.h-based:

  float x;
  int8x8_t y = x;

which gives:

  cannot convert ‘float’ to ‘int8x8_t’ {aka ‘__Int8x8_t’} in initialization

This is accurate -- and __Int8x8_t is defined by the AArch64 PCS --
but it's not going to be meaningful to most users.

This patch stops the aka code looking through typedefs if all of
the following are true:

(1) the typedef is built into the compiler or comes from a system header

(2) the target of the typedef is anonymous or has a name in the
    implementation namespace

(3) the target type is a tag type or vector type, which have in common that:
    (a) we print their type names if they have one
    (b) what we print for anonymous types isn't all that useful
        ("struct <anonymous>" etc. for tag types, pseudo-C "__vector(N) T"
        for vector types)

The patch does this by recursively looking for the aka type, like the
C++ frontend already does.  This in turn makes "aka" work for distinct type
copies like __Int8x8_t on aarch64, fixing the ??? in aarch64/diag_aka_1.c.

2019-10-14  Richard Sandiford  <richard.sandiford@arm.com>

gcc/c-family/
* c-common.h (user_facing_original_type_p): Declare.
* c-common.c: Include c-spellcheck.h.
(user_facing_original_type_p): New function.

gcc/c/
* c-objc-common.c (useful_aka_type_p): Replace with...
(get_aka_type): ...this new function.  Given the original type,
decide which aka type to print (if any).  Only look through typedefs
if user_facing_original_type_p.
(print_type): Update accordingly.

gcc/testsuite/
* gcc.dg/diag-aka-5.h: New test.
* gcc.dg/diag-aka-5a.c: Likewise.
* gcc.dg/diag-aka-5b.c: Likewise.
* gcc.target/aarch64/diag_aka_1.c (f): Expect an aka to be printed
for myvec.

From-SVN: r276951

c-parser.c (c_parser_omp_all_clauses): Change bool NESTED_P argument into int NESTED...

c/
* c-parser.c (c_parser_omp_all_clauses): Change bool NESTED_P argument
into int NESTED, if it is 2, diagnose missing commas in between
clauses.
(c_parser_omp_context_selector): Pass 2 as last argument to
c_parser_omp_all_clauses.
cp/
* parser.c (cp_parser_omp_all_clauses): Change bool NESTED_P argument
into int NESTED, if it is 2, diagnose missing commas in between
clauses.
(cp_parser_omp_context_selector): Pass 2 as last argument to
cp_parser_omp_all_clauses.
testsuite/
* c-c++-common/gomp/declare-variant-7.c: Add tests for clauses not
separated by commas in simd selector trait properties.

From-SVN: r276950

Normalize unsigned ~[0,0] into [1,MAX].

From-SVN: r276949

Fix dump message issue

'}' is missed at the end.

gcc/ChangeLog:

2019-10-14  Xiong Hu Luo  <luoxhu@linux.ibm.com>

* tree-sra.c (dump_access): Add missing braces.

From-SVN: r276948

Daily bump.

From-SVN: r276947

[Darwin, machopic 8/n] Back out part of PR71767 fix.

We applied a conservative, but fairly large, hammer to fix PR71767.
However, ideally, we want minimise the number of symbols visible to
ld64 and to match the cases emitted by clang (since that's what ld64
is expecting).  Now we've improved the handling of indirections, we
can make the indirection symbols local when they are in the regular
non-lazy symbol pointers section.  We will continue to make any
indirections in the data section visible (since right now we have no
way to track if a given symbol follows a weak global).
This change makes no difference to handling of labels for constants
(to be revised in a future patch).

There's a mechanical change to a number of tests (allowing 'l' or 'L'
as the indirection symbol prefix).

gcc/ChangeLog:

2019-10-13  Iain Sandoe  <iain@sandoe.co.uk>

* config/darwin.c (machopic_indirection_name): Rework the
function to emit linker-visible symbols only for indirections
in the data section.  Clean up the code and update comments.

gcc/testsuite/ChangeLog:

2019-10-13  Iain Sandoe  <iain@sandoe.co.uk>

* gcc.target/i386/indirect-thunk-1.c: Allow 'l' or 'L' in
indirection label prefix, for Darwin.
* gcc.target/i386/indirect-thunk-2.c: Likewise.
* gcc.target/i386/indirect-thunk-3.c: Likewise.
* gcc.target/i386/indirect-thunk-4.c: Likewise.
* gcc.target/i386/indirect-thunk-attr-1.c: Likewise.
* gcc.target/i386/indirect-thunk-attr-2.c: Likewise.
* gcc.target/i386/indirect-thunk-attr-3.c: Likewise.
* gcc.target/i386/indirect-thunk-attr-4.c: Likewise.
* gcc.target/i386/indirect-thunk-attr-5.c: Likewise.
* gcc.target/i386/indirect-thunk-attr-6.c: Likewise.
* gcc.target/i386/indirect-thunk-extern-1.c: Likewise.
* gcc.target/i386/indirect-thunk-extern-2.c: Likewise.
* gcc.target/i386/indirect-thunk-extern-3.c: Likewise.
* gcc.target/i386/indirect-thunk-extern-4.c: Likewise.
* gcc.target/i386/indirect-thunk-inline-1.c: Likewise.
* gcc.target/i386/indirect-thunk-inline-2.c: Likewise.
* gcc.target/i386/indirect-thunk-inline-3.c: Likewise.
* gcc.target/i386/indirect-thunk-inline-4.c: Likewise.
* gcc.target/i386/pr32219-2.c: Likewise.
* gcc.target/i386/pr32219-3.c: Likewise.
* gcc.target/i386/pr32219-4.c: Likewise.
* gcc.target/i386/pr32219-7.c: Likewise.
* gcc.target/i386/pr32219-8.c: Likewise.
* gcc.target/i386/ret-thunk-14.c: Likewise.
* gcc.target/i386/ret-thunk-15.c: Likewise.
* gcc.target/i386/ret-thunk-9.c: Likewise.

From-SVN: r276943

[Darwin, machopic 7/n] Remove code that should be dead.

This code fragment was imported from the Apple branch (it was never
applied to mainline).  It is stated to fix up a problem sometimes
created by reload (before that had been extended to have greater
flexibility in assigning the pic registers).  In any event, reload
is no longer in use for the port.

gcc/ChangeLog:

2019-10-13  Iain Sandoe  <iain@sandoe.co.uk>

* config/darwin.c (machopic_indirect_data_reference): Remove
redundant code.

From-SVN: r276942

re PR fortran/91513 (Non-standard terminology in error message for pointer component assignment in pure procedure)

2019-10-13  Damian Rouson  <damain@sourceryinstitue.org>

PR fortran/91513
* resolve.c (resolve_ordinary_assign): Improved error message.

2019-10-13  Damian Rouson  <damain@sourceryinstitue.org>

PR fortran/91513
* gfortran.dg/impure_assignment_2.f90: Update dg-error regex.

From-SVN: r276941

re PR fortran/90297 (gcc/fortran/resolve.c: 2 * possibly redundant code ?)

2019-10-13  Steven G. Kargl  <kargl@gcc.gnu.org>

PR fortran/90297
* resolve.c (resolve_typebound_function): Remove code with no
functional effect.

From-SVN: r276940

[PATCH] teach gengtype about 'mutable'

https://gcc.gnu.org/ml/gcc-patches/2019-10/msg00942.html

In constifying some more of line-map I discovered gengtype didn't know mutable. Added thusly.
* gengtype-lex.l (CXX_KEYWORD): Add 'mutable'.

From-SVN: r276939

re PR fortran/92017 (ICE in gfc_expr_attr, at fortran/primary.c:2674)

2019-10-13  Thomas Koenig  <tkoenig@gcc.gnu.org>

PR fortran/92017
* expr.c (simplify_parameter_variable): Set the character length
of the result expression from the original expression if
necessary.

2019-10-13  Thomas Koenig  <tkoenig@gcc.gnu.org>

PR fortran/92017
* gfortran.dg/minmaxloc_14.f90: New test.

From-SVN: r276938

Remove dg-add-options c99_runtime

gcc:
* doc/sourcebuild.texi (Test Directives, Add Options): Remove
c99_runtime.

gcc/testsuite:
* lib/target-supports.exp (add_options_for_c99_runtime): Remove.
(check_effective_target_c99_runtime): Remove call to
add_options_for_c99_runtime.

* gcc.dg/builtins-18.c: Remove dg-add-options c99_runtime.
* gcc.dg/builtins-20.c: Likewise.
* gcc.dg/builtins-53.c: Likewise.
* gcc.dg/builtins-55.c: Likewise.
* gcc.dg/builtins-67.c: Likewise.
* gcc.dg/c99-tgmath-1.c: Likewise.
* gcc.dg/c99-tgmath-2.c: Likewise.
* gcc.dg/c99-tgmath-3.c: Likewise.
* gcc.dg/c99-tgmath-4.c: Likewise.
* gcc.dg/ipa/inline-8.c: Likewise.
* gcc.dg/ipa/ipa-icf-5.c: Likewise.
* gcc.dg/ipa/ipa-icf-7.c: Likewise.
* gcc.dg/nextafter-2.c: Likewise.
* gcc.dg/pr42427.c: Likewise.
* gcc.dg/pr78965.c: Likewise.
* gcc.dg/single-precision-constant.c: Likewise.
* gcc.dg/torture/builtin-convert-1.c: Likewise.
* gcc.dg/torture/builtin-convert-2.c: Likewise.
* gcc.dg/torture/builtin-convert-3.c: Likewise.
* gcc.dg/torture/builtin-convert-4.c: Likewise.
* gcc.dg/torture/builtin-fp-int-inexact.c: Likewise.
* gcc.dg/torture/builtin-fp-int-inexact-c2x.c: Likewise.
* gcc.dg/torture/builtin-integral-1.c: Likewise.
* gcc.dg/torture/builtin-power-1.c: Likewise.
* gcc.dg/tree-ssa/copy-sign-1.c: Likewise.
* gcc.dg/tree-ssa/minmax-2.c: Likewise.
* gcc.dg/tree-ssa/mult-abs-2.c: Likewise.
* gcc.target/i386/387-builtin-fp-int-inexact.c: Likewise.
* gcc.target/i386/387-rint-inline-1.c: Likewise.
* gcc.target/i386/387-rint-inline-2.c: Likewise.
* gcc.target/i386/conversion.c: Likewise.
* gcc.target/i386/pr47312.c: Likewise.
* gcc.target/i386/sse2-builtin-fp-int-inexact.c: Likewise.
* gcc.target/i386/sse2-rint-inline-1.c: Likewise.
* gcc.target/i386/sse2-rint-inline-2.c: Likewise.
* gcc.target/i386/sse4_1-builtin-fp-int-inexact.c: Likewise.
* gcc.target/i386/sse4_1-rint-inline.c: Likewise.

From-SVN: r276937

lto-common.c (read_cgraph_and_symbols): Grow ggc memory use after summary streaming.

* lto-common.c (read_cgraph_and_symbols): Grow ggc memory use after
summary streaming.

From-SVN: r276936

* lto.c (lto_wpa_write_files): Do not update bodies of clones.

From-SVN: r276935

lto-streamer-out.c (collect_block_tree_leafs): Renumber statements so non-virutal are before virutals.

* lto-streamer-out.c (collect_block_tree_leafs): Renumber statements
so non-virutal are before virutals.
(output_function): Avoid body modifications.

From-SVN: r276934

Daily bump.

From-SVN: r276933

pa.c (pa_output_call): Load descriptor address to register %r22.

* config/pa/pa.c (pa_output_call): Load descriptor address to register
%r22.  Load function address before global pointer.
(pa_attr_length_indirect_call): Adjust length of inline versions of
$$dyncall.
(pa_output_indirect_call): Remove fast inline version of $$dyncall
before normal cases.  Update inline $$dyncall sequences to preserve
function descriptor address in register %r22.
(TRAMPOLINE_CODE_SIZE): Adjust.
(pa_asm_trampoline_template): Revise 32-bit trampoline.  Don't assume
register %r22 contains trampoline address.
(pa_trampoline_init): Adjust offsets.
(pa_trampoline_adjust_address): Likewise.
* config/pa/pa.h (TRAMPOLINE_SIZE): Adjust 32-bit size.

From-SVN: r276928

[Darwin, machopic 6/n] Fix for 67183

When we're using the LLVM-based assembler (the default on modern Darwin)
the ordering of stubs and non-lazy symbol pointers is important.

Interleaving the output (current GCC behaviour) leads to crashes which
prevents us from building code with symbol stubs.

To resolve this, we order the output of stubs and symbol indirections:

 1. Any indirections in the data section
 2. Symbol stubs.
 3. Non-lazy symbol pointers.

At present, we still emit LTO sections after these.

gcc/ChangeLog:

2019-10-12  Iain Sandoe  <iain@sandoe.co.uk>

PR target/67183
* config/darwin.c (machopic_indirection): New field to flag
non-lazy-symbol-pointers in the data section.
(machopic_indirection_name): Compute if an indirection should
appear in the data section.
(machopic_output_data_section_indirection): New callback split
from machopic_output_indirection.
(machopic_output_stub_indirection): Likewise.
(machopic_output_indirection): Retain the code for non-lazy
symbol pointers in their regular section.
(machopic_finish): Use the new callbacks to order the indirection
output.

From-SVN: r276926

lib2funcs.S (__gcc_plt_call): Load branch target to %r21.

* config/pa/lib2funcs.S (__gcc_plt_call): Load branch target to %r21.
Load PIC register after branch target.  Fix white space.
* config/pa/milli64.S ($$dyncall): Separate LINUX and non LINUX
implementations.  Load PIC register after branch target.  Don't
clobber function pointer when it points to function descriptor.
Use nullification instead of branch in LINUX implementation.

From-SVN: r276925

[Darwin, machopic 5/n] Make machopic_finish() static.

It's only called from darwin.c.

gcc/ChangeLog:

2019-10-12  Iain Sandoe  <iain@sandoe.co.uk>

* config/darwin-protos.h (machopic_finish): Delete.
* config/darwin.c (machopic_finish): Make static.

From-SVN: r276924

[Darwin] Suppress emitting empty ctor/dtor sections.

Older versions of GCC emit empty .constructor/.destructor sections
whenever building for C++. In fact, these sections are only used for
kernel mode code - so don't emit them unless that's what we're
building.

gcc/ChangeLog:

2019-10-12  Iain Sandoe  <iain@sandoe.co.uk>

* config/darwin.c (darwin_file_end): Only emit empty CTOR/DTOR
sections when building kernel extension code.

From-SVN: r276923

Fix the ChangeLog for my previous commit

Fixes: 66ce32bb2a8 ("Fix the ChangeLog for my previous commit")

From-SVN: r276922

The inline keyword is supported in all new C standards

The documentation used to indicate that the inline keyword was only
supported by c99 and c11, whereas in fact it is supported by c99 and all
newer standards.

gcc/ChangeLog

2019-10-12  Palmer Dabbelt  <palmer@sifive.com>

        * doc/extend.texi (Alternate Keywords): Change "-std=c11" to "a
        later standard."

From-SVN: r276921

pa.c (pa_option_override): Remove trailing comma from warning.

* gcc/config/pa/pa.c (pa_option_override): Remove trailing comma
from warning.

From-SVN: r276920

re PR ada/91995 (gnat miscompilation and bootstrap failure on m68k-linux)

PR ada/91995
* sem_ch8.adb (Chain_Use_Clause): Remove second argument in calls
to Defining_Entity.
* sem_elab.adb (Find_Unit_Entity): Likewise.  Deal with N_Subunit
here in lieu of in Defining_Entity.
* sem_util.ads (Defining_Entity): Remove 2nd and 3th parameters.
* sem_util.adb (Defining_Entity): Remove 2nd and 3th parameters,
and adjust accordingly.  Deal with N_Compilation_Unit.

From-SVN: r276916

re PR middle-end/92063 (ICE in operation_could_trap_p, at tree-eh.c:2528 when compiling Python's Python/_warnings.c)

PR middle-end/92063
* tree-eh.c (operation_could_trap_helper_p) <case COND_EXPR>
<case VEC_COND_EXPR>: Return false with *handled = false.
(tree_could_trap_p): For {,VEC_}COND_EXPR return false instead of
recursing on the first operand.
* fold-const.c (simple_operand_p_2): Use generic_expr_could_trap_p
instead of tree_could_trap_p.
* tree-ssa-sccvn.c (vn_nary_may_trap): Formatting fixes.

* gcc.c-torture/compile/pr92063.c: New test.

From-SVN: r276915

c-common.h (c_omp_mark_declare_variant, [...]): Declare.

c-family/
* c-common.h (c_omp_mark_declare_variant,
c_omp_context_selector_matches): Declare.
* c-omp.c: Include attribs.h, gimplify.h, cgraph.h, symbol-summary.h
and hsa-common.h.
(c_omp_get_context_selector): Support second argument NULL.
(c_omp_mark_declare_variant, c_omp_context_selector_matches): New
functions.
* c-attribs.c (c_common_attribute_table): Remove "omp declare variant"
attribute, add "omp declare variant base" and
"omp declare variant variant" attributes.
c/
* c-parser.c (c_parser_omp_context_selector): Improve error recovery.
For simd properties, put them directly into TREE_VALUE.
(c_finish_omp_declare_variant): Call c_omp_mark_declare_variant.
If c_omp_context_selector_matches is 0, don't add attribute, otherwise
add "omp declare variant base" attribute rather than
"omp declare variant".
cp/
* parser.c (cp_parser_omp_context_selector): Improve error recovery.
For simd properties, put them directly into TREE_VALUE.
(cp_finish_omp_declare_variant): Add "omp declare variant base"
attribute rather than "omp declare variant".
testsuite/
* c-c++-common/gomp/declare-variant-2.c: Adjust for error recovery
improvements.  Add new tests.
* c-c++-common/gomp/declare-variant-4.c: New test.
* c-c++-common/gomp/declare-variant-5.c: New test.
* c-c++-common/gomp/declare-variant-6.c: New test.
* c-c++-common/gomp/declare-variant-7.c: New test.

From-SVN: r276914

compiler: mangle dots in pkgpath

    We need to mangle dots to avoid problems with -fgo-pkgpath=a.0.
    That will confuse the name mangling, which assumes that names
    entering the mangling cannot contain arbitrary dot characters.
    We don't need to mangle other characters; go_encode_id will handle them.

    Fixes golang/go#33871

    Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/200838

From-SVN: r276913

Daily bump.

From-SVN: r276912

Support decimal floating-point constants in C2x.

ISO C2x adds decimal floating point as an optional standard feature.
This patch accordingly makes GCC accept DFP constants (DF, DD, DL, df,
dd, dl suffixes) in strict C2X mode, with a pedwarn-if-pedantic for
older standards and a warning with -Wc11-c2x-compat even in C2x mode
(which in turn requires -Wc11-c2x-compat to be newly passed through to
libcpp).

Bootstrapped with no regressions on x86_64-pc-linux-gnu.

gcc/c-family:
* c.opt (Wc11-c2x-compat): Add CPP(cpp_warn_c11_c2x_compat)
CppReason(CPP_W_C11_C2X_COMPAT).

gcc/testsuite:
* gcc.dg/dfp/c11-constants-1.c, gcc.dg/dfp/c11-constants-2.c,
gcc.dg/dfp/c2x-constants-1.c, gcc.dg/dfp/c2x-constants-2.c: New
tests.
* gcc.dg/dfp/constants-pedantic.c: Use -std=gnu17 explicitly.
Update expected diagnostics.

libcpp:
* include/cpplib.h (struct cpp_options): Add dfp_constants and
cpp_warn_c11_c2x_compat.
(enum cpp_warning_reason): Add CPP_W_C11_C2X_COMPAT.
* init.c (struct lang_flags): Add dfp_constants.
(lang_defaults): Set dfp_constants to 1 for GNUC2X and STDC2X and
0 for other languages.
(cpp_set_lang): Set dfp_constants from language.
(cpp_create_reader): Set cpp_warn_c11_c2x_compat to -1.
* expr.c (interpret_float_suffix): Mention DFP constants as C2X in
comment.
(cpp_classify_number): Do not diagnose DFP constants for languages
setting dfp_constants, unless cpp_warn_c11_c2x_compat.

From-SVN: r276908

PR c++/92070 - bogus error with -fchecking=2.

* g++.dg/expr/cond17.C: New test.

From-SVN: r276907

PR c++/92049 - extra error with -fchecking=2.

The concepts merge brought this bit

@@ -26326,9 +26559,9 @@ build_non_dependent_expr (tree expr)
     unexpected recursive instantiations.  */
       && !parsing_nsdmi ()
       /* Don't do this during concept expansion either and for
-         the same reason.  */
-      && !expanding_concept ())
-    fold_non_dependent_expr (expr, tf_none);
+    the same reason.  */
+      && !parsing_constraint_expression_p ())
+    fold_non_dependent_expr (expr);

   STRIP_ANY_LOCATION_WRAPPER (expr);

(which I'm not finding in the ChangeLog).  Dropping tf_none means that
fold_non_dependent_expr will use tf_warning_or_error by default, and in
this test that causes an error:

  template<bool> struct cond;

  template<int> struct S {
    void f(int i) {
      cond<__builtin_constant_p(i)>();
    }
  };

  S<1> s;

where it complains that cond<false> is incomplete.  Which it is, but we're
not actually instantiating the function f, so issuing an error seems
overzealous (though not wrong), and it breaks a bunch of tests.  This patch
brings that tf_none back.  We will still complain if we do instantiate f.

* pt.c (build_non_dependent_expr): Call fold_non_dependent_expr
with tf_none.

* g++.dg/template/builtin2.C: New test.

From-SVN: r276906

typeck.c (cp_build_binary_op): Do not handle RROTATE_EXPR and LROTATE_EXPR.

2019-10-10  Paolo Carlini  <paolo.carlini@oracle.com>

* typeck.c (cp_build_binary_op): Do not handle RROTATE_EXPR and
LROTATE_EXPR.
* constexpr.c (cxx_eval_constant_expression): Likewise.
(potential_constant_expression_1): Likewise.
* cp-gimplify.c (cp_fold): Likewise.
* pt.c (tsubst_copy): Likewise.

From-SVN: r276903

Preserve the location of explicitly defaulted functions.

* decl2.c (mark_used): Don't clobber DECL_SOURCE_LOCATION on
explicitly defaulted functions.
* method.c (synthesize_method): Likewise.

From-SVN: r276902

Extend subst to simplify CONST_INT inside SIGN_EXTEND.

This addresses PR 91860 which has four testcases triggering internal errors.
The problem here is that in combine when handling debug insns, we are trying
to substitute
(sign_extend:DI (const_int 8160 [0x1fe0]))
as the value for
(reg:DI 78 [ _9 ])
in the debug insn
(debug_insn 29 28 30 2 (var_location:QI d (subreg:QI (reg:DI 78 [ _9 ]) 0)) "tmp4.c":11:5 -1
     (nil))
This eventually triggers an abort because 8160 is not a sign-extended
QImode value.

In subst there is already code check for a CONST_INT inside a ZERO_EXTEND and
simplify it.  This needs to be extended to also handle a SIGN_EXTEND the same
way.

gcc/
PR rtl-optimization/91860
* combine.c (subst): If new_rtx is a constant, also check for
SIGN_EXTEND when deciding whether to call simplify_unary_operation.

gcc/testsuite/
PR rtl-optimization/91860
* gcc.dg/pr91860-1.c: New testcase.
* gcc.dg/pr91860-2.c: New testcase.
* gcc.dg/pr91860-3.c: New testcase.
* gcc.dg/pr91860-4.c: New testcase.

From-SVN: r276901

re PR fortran/91649 (ICE in gfc_resolve_findloc, at fortran/iresolve.c:1827)

2019-10-11  Steven G. Kargl  <kargl@gcc.gnu.org>

PR fortran/91649
check.c (gfc_check_findloc): Additional checking for valid arguments

2019-10-11  Steven G. Kargl  <kargl@gcc.gnu.org>

PR fortran/91649
* gfortran.dg/pr91649.f90: New test.

From-SVN: r276900

re PR fortran/91715 (ICE in resolve_fntype, at fortran/resolve.c:16884)

2019-10-11  Steven G. Kargl  <kargl@gcc.gnu.org>

PR fortran/91715
* decl.c (gfc_match_prefix): If matching a type-spec returns an error,
it's an error so re-act correctly.

2019-10-11  Steven G. Kargl  <kargl@gcc.gnu.org>

PR fortran/91715
* gfortran.dg/function_kinds_5.f90: Prune run-on error.
* gfortran.dg/pr85543.f90: Ditto.
* gfortran.dg/pr91715.f90: New test.

From-SVN: r276899

re PR fortran/92018 (ICE in gfc_conv_constant_to_tree, at fortran/trans-const.c:370)

2019-10-11  Steven G. Kargl  <kargl@gcc.gnu.org>

PR fortran/92018
* check.c (reset_boz): New function.
(illegal_boz_arg, boz_args_check, gfc_check_complex, gfc_check_float,
gfc_check_transfer): Use it.
(gfc_check_dshift): Use reset_boz, and re-arrange the checking to
help suppress possible run-on errors.
(gfc_check_and): Restore checks for valid argument types.  Use
reset_boz, and re-arrange the checking to help suppress possible
  un-on errors.
* resolve.c (resolve_function): Actual arguments cannot be BOZ in
a function reference.

2019-10-11  Steven G. Kargl  <kargl@gcc.gnu.org>

PR fortran/92018
* gfortran.dg/gnu_logical_2.f90: Update dg-error regex.
* gfortran.dg/pr81509_2.f90: Ditto.
* gfortran.dg/pr92018.f90: New test.

From-SVN: r276898

re PR fortran/92019 (ICE in find_inquiry_ref, at expr.c:1790)

2019-10-11  Steven G. Kargl  <kargl@gcc.gnu.org>

PR fortran/92019
* array.c (match_subscript): BOZ cannot be an array subscript.

2019-10-11  Steven G. Kargl  <kargl@gcc.gnu.org>

PR fortran/92019
* gfortran.dg/pr92019.f90: New test.

From-SVN: r276897

Support _Decimal* keywords for C2x.

ISO C2x adds decimal floating point as an optional standard feature.
This patch accordingly makes GCC accept the _Decimal* keywords in
strict C2x mode, using pedwarn_c11 to get a warning for
-Wc11-c2x-compat.  (Constants, where the pedwarn is in libcpp, will be
dealt with separately.)

The _Decimal* keywords are marked with D_EXT, meaning they are not
considered keywords at all in strict conformance modes.  This is
contrary to the normal practice for most implementation-namespace
keywords, which are accepted in all standards modes but with
appropriate pedwarns for older standards.  This patch removes D_EXT
from those keywords so they are accepted as keywords for all
standards, consequently removing the gcc.dg/dfp/keywords-ignored-c99.c
test that is no longer valid.

(A new D_C2X for keywords will still be needed if any new keywords get
added that aren't in the implementation namespace for older standards;
there are proposals for such keywords, albeit as predefined macros
that might not actually need new keywords in the compiler in all
cases.  If the DFP keywords end up as decimal32 etc., of course
appropriate compiler and testcase changes will be needed, and a
version of keywords-ignored-c99.c would make sense again with such
spellings.)

Bootstrapped with no regressions on x86_64-pc-linux-gnu.

gcc/c:
* c-decl.c (declspecs_add_type): Use pedwarn_c11 for DFP types.

gcc/c-family:
* c-common.c (c_common_reswords): Do not use D_EXT for _Decimal32,
_Decimal64 and _Decimal128.

gcc/testsuite:
* gcc.dg/dfp/c11-keywords-1.c, gcc.dg/dfp/c11-keywords-2.c,
gcc.dg/dfp/c2x-keywords-1.c, gcc.dg/dfp/c2x-keywords-2.c: New
tests.
* gcc.dg/dfp/keywords-ignored-c99.c: Remove test.
* gcc.dg/dfp/constants-c99.c, gcc.dg/dfp/keywords-c89.c,
gcc.dg/dfp/keywords-c99.c: Use -pedantic-errors.

From-SVN: r276896

* fi.po: Update.

From-SVN: r276894

Implement <concepts> header for C++20

There are currently no tests for [concepts.compare], but they will be
added ASAP.

* include/Makefile.am: Add new header.
* include/Makefile.in: Regenerate.
* include/precompiled/stdc++.h: Include <concepts>.
* include/std/concepts: New header for C++20.
* include/std/version (__cpp_lib_concepts): Define.
* scripts/create_testsuite_files: Look for test files in new std
directory.
* testsuite/libstdc++-dg/conformance.exp: Likewise.
* testsuite/std/concepts/concepts.callable/invocable.cc: New test.
* testsuite/std/concepts/concepts.callable/regular_invocable.cc: New
test.
* testsuite/std/concepts/concepts.callable/relation.cc: New test.
* testsuite/std/concepts/concepts.callable/strictweakorder.cc: New
test.
* testsuite/std/concepts/concepts.lang/concept.arithmetic/
floating_point.cc: New test.
* testsuite/std/concepts/concepts.lang/concept.arithmetic/integral.cc:
New test.
* testsuite/std/concepts/concepts.lang/concept.arithmetic/
signed_integral.cc: New test.
* testsuite/std/concepts/concepts.lang/concept.arithmetic/
unsigned_integral.cc: New test.
* testsuite/std/concepts/concepts.lang/concept.assignable/1.cc: New
test.
* testsuite/std/concepts/concepts.lang/concept.common/1.cc: New test.
* testsuite/std/concepts/concepts.lang/concept.commonref/1.cc: New
test.
* testsuite/std/concepts/concepts.lang/concept.constructible/1.cc:
New test.
* testsuite/std/concepts/concepts.lang/concept.convertible/1.cc:
New test.
* testsuite/std/concepts/concepts.lang/concept.copyconstructible/1.cc:
New test.
* testsuite/std/concepts/concepts.lang/concept.defaultconstructible/
1.cc: New test.
* testsuite/std/concepts/concepts.lang/concept.derived/1.cc:
New test.
* testsuite/std/concepts/concepts.lang/concept.destructible/1.cc:
New test.
* testsuite/std/concepts/concepts.lang/concept.moveconstructible/1.cc:
New test.
* testsuite/std/concepts/concepts.lang/concept.same/1.cc:
New test.
* testsuite/std/concepts/concepts.lang/concept.swappable/swap.cc:
New test.
* testsuite/std/concepts/concepts.lang/concept.swappable/swappable.cc:
New test.
* testsuite/std/concepts/concepts.lang/concept.swappable/
swappable_with.cc: New test.
* testsuite/std/concepts/concepts.object/copyable.cc: New test.
* testsuite/std/concepts/concepts.object/movable.cc: New test.
* testsuite/std/concepts/concepts.object/regular.cc: New test.
* testsuite/std/concepts/concepts.object/semiregular.cc: New test.

From-SVN: r276892

Use __is_same_as for std::is_same and std::is_same_v

By using the built-in we don't need to match a partial specialization
for std::is_same and don't need to instantiate std::is_same at all for
uses of std::is_same_v.

* include/std/type_traits (is_same): Replace partial specialization
by using __is_same_as built-in in primary template.
(is_same_v): Use __is_same_as built-in instead of instantiating the
is_same trait.

From-SVN: r276891

PR libstdc++/92059 fix several bugs in tr2::dynamic_bitset

PR libstdc++/92059
* include/tr2/dynamic_bitset (__dynamic_bitset_base): Define all
special member functions as defaulted. Add noexcept to most members.
(__dynamic_bitset_base(size_t, unsigned long long, const _Alloc&)):
Mask off unwanted bits in the __val parameter. Avoid undefined left
shifts.
(__dynamic_bitset_base::_M_assign): Remove.
(__dynamic_bitset_base::_M_do_reset): Use std::fill.
(__dynamic_bitset_base::_M_are_all_aux): Avoid integer promotion when
block_type has lower rank than int.
(dynamic_bitset): Add noexcept to most members. Use injected-class-name
in return types and parameter types.
(dynamic_bitset::_M_Nb): Add default member initializer.
(dynamic_bitset(), dynamic_bitset(const dynamic_bitset&)): Define as
defaulted.
(dynamic_bitset(dynamic_bitset&&)): Clear source object after move.
(dynamic_bitset::operator=(const dynamic_bitset&)): Define as
defaulted.
(dynamic_bitset::operator=(dynamic_bitset&&)): Add noexcept-specifier.
Define without using swap, to propagate allocator correctly.
(dynamic_bitset(const char*, const _Alloc&)): Use strlen.
(dynamic_bitset::_M_do_sanitize, dynamic_bitset::_M_do_fill): Use
casts to avoid unwanted integer promotions.
(dynamic_bitset::_M_copy_from_ptr): Rearrange template parameters and
add default template arguments and default argument to simplify usage.
(dynamic_bitset::_M_copy_from_string): Adjust call to _M_copy_from_ptr.
(operator==(const dynamic_bitset&, const dynamic_bitset&))
(operator<(const dynamic_bitset&, const dynamic_bitset&)): Use _M_Nb.
* include/tr2/dynamic_bitset.tcc (dynamic_bitset::_M_copy_from_ptr):
Adjust template parameters to match declaration.
* testsuite/tr2/dynamic_bitset/cmp.cc: New test.
* testsuite/tr2/dynamic_bitset/cons.cc: New test.
* testsuite/tr2/dynamic_bitset/copy.cc: New test.
* testsuite/tr2/dynamic_bitset/move.cc: New test.
* testsuite/tr2/dynamic_bitset/pr92059.cc: New test.

From-SVN: r276890

Avoid warnings in <charconv>

* include/bits/charconv.h (__to_chars_len): Avoid -Wsign-compare
warnings.

From-SVN: r276889

Relax store_bit_field call in store_expr

store_bit_field already takes a poly_uint64 size, so we can relax the
INTVAL to rtx_to_poly_int64.  This is tested by the SVE ACLE patches.

2019-10-11  Richard Sandiford  <richard.sandiford@arm.com>

gcc/
* expr.c (store_expr): Use rtx_to_poly_int64 rather than
INTVAL when calling store_bit_field.

From-SVN: r276888

[ARM] Tweak HONOR_REG_ALLOC_ORDER

Setting HONOR_REG_ALLOC_ORDER improves codesize with -Os, however it generates
slower and larger code with -O2 and higher.  So only set it when optimizing for
size.  On Cortex-A57 this improves SPECINT2006 by 0.15% and SPECFP2006 by 0.25%
while reducing codesize.

    gcc/
* config/arm/arm.h (HONOR_REG_ALLOC_ORDER): Set when optimizing for
size.

From-SVN: r276887

tree-vect-loop.c (vect_analyze_loop_operations): Adjust call to vectorizable_live_operation.

2019-10-11  Bernd Edlinger  <bernd.edlinger@hotmail.de>

        * tree-vect-loop.c (vect_analyze_loop_operations): Adjust call to
        vectorizable_live_operation.
        (vectorizable_live_operation): Adjust parameters.
        * tree-vect-stmts.c (vect_init_vector,
        vect_gen_widened_results_half): Fix typo in function comment.
        (can_vectorize_live_stmts): Adjust function comment.
        Adjust parameters.  Adjust call to vectorizable_live_operation.
        (vect_analyze_stmt): Adjust call to can_vectorize_live_stmts.
        (vect_transform_stmt): Adjust function comment.  Adjust call to
        can_vectorize_live_stmts.
        * tree-vectorizer.h (vectorizable_live_operation): Adjust parameters.

From-SVN: r276886

Fortran] PR 92050 - fix ICE with -fcheck=all

        gcc/fortran/
        PR fortran/92050
        * trans-expr.c (gfc_conv_procedure_call): Handle code generated
        by -fcheck=all.

        gcc/testsuite/
        PR fortran/92050
        * gfortran.dg/pr92050.f90: New.

From-SVN: r276885

re PR tree-optimization/90883 (Generated code is worse if returned struct is unnamed)

2019-10-11  Richard Biener  <rguenther@suse.de>

PR tree-optimization/90883
PR tree-optimization/91091
* tree-ssa-sccvn.c (vn_reference_lookup_3): Use correct
alias-sets both for recording VN table entries and continuing
walking after translating through copies.  Handle same-sized
reads from SSA names by returning the plain SSA name.
(eliminate_dom_walker::eliminate_stmt): Properly handle
non-size precision stores in redundant store elimination.

* gcc.dg/torture/20191011-1.c: New testcase.
* gcc.dg/tree-ssa/ssa-fre-82.c: Likewise.
* gcc.dg/tree-ssa/ssa-fre-83.c: Likewise.
* gcc.dg/tree-ssa/redundant-assign-zero-1.c: Disable FRE.
* gcc.dg/tree-ssa/redundant-assign-zero-2.c: Likewise.

From-SVN: r276882

ggc-page.c (release_pages): Output statistics when !quiet_flag.

* ggc-page.c (release_pages): Output statistics when !quiet_flag.
(ggc_collect): Dump later to not interfere with release_page dump.
(ggc_trim): New function.
* ggc-none.c (ggc_trim): New.
* ggc.h (ggc_trim): Declare.

* lto.c (lto_wpa_write_files): Call ggc_trim.

From-SVN: r276878

re PR tree-optimization/92066 (Many vectorization tests FAIL)

2019-10-11  Richard Biener  <rguenther@suse.de>

PR tree-optimization/92066
PR tree-optimization/92046
* tree-vect-data-refs.c (vect_enhance_data_refs_alignment):
Fix bogus cost model check.

From-SVN: r276876

[OpenMP,Fortran] Fix several OpenMP use_device_addr/map/update errors

gcc/fortran/
* f95-lang.c (LANG_HOOKS_OMP_IS_ALLOCATABLE_OR_PTR): Re-define to
gfc_omp_is_allocatable_or_ptr.
* trans-decl.c (create_function_arglist): Set GFC_DECL_OPTIONAL_ARGUMENT
only if not passed by value.
* trans-openmp.c (gfc_omp_is_allocatable_or_ptr): New.
(gfc_trans_omp_clauses): For MAP, handle (present) optional arguments;
for target update, handle allocatable/pointer scalars.
* trans.h (gfc_omp_is_allocatable_or_ptr): Declare.

gcc/
* langhooks-def.h (LANG_HOOKS_OMP_IS_ALLOCATABLE_OR_PTR): Define.
(LANG_HOOKS_DECLS): Add it.
* langhooks.h (lang_hooks_for_decls): Add omp_is_allocatable_or_ptr;
update comment for omp_is_optional_argument.
* omp-general.c (omp_is_allocatable_or_ptr): New.
* omp-general.h (omp_is_allocatable_or_ptr): Declare.
* omp-low.c (scan_sharing_clauses, lower_omp_target): Handle
Fortran's optional arguments and allocatable/pointer scalars
with use_device_addr.

libgomp/
* testsuite/libgomp.fortran/use_device_addr-1.f90: New.
* testsuite/libgomp.fortran/use_device_addr-2.f90: New.

From-SVN: r276875

decl.c (elaborate_reference_1): Specifically deal with pointer displacement.

* gcc-interface/decl.c (elaborate_reference_1): Specifically deal with
pointer displacement.

* gcc-interface/decl.c (components_to_record): Use proper name.

* gcc-interface/trans.c (Sloc_to_locus): Use standard types.

From-SVN: r276874

decl.c (gnat_to_gnu_field): Adjust again the packing for a field without strict alignment and with an...

* gcc-interface/decl.c (gnat_to_gnu_field): Adjust again the packing
for a field without strict alignment and with an oversized clause.

From-SVN: r276873

S/390: Test signaling FP comparison instructions

gcc/testsuite/ChangeLog:

2019-10-11  Ilya Leoshkevich  <iii@linux.ibm.com>

PR target/77918
* gcc.target/s390/s390.exp: Enable Fortran tests.
* gcc.target/s390/zvector/autovec-double-quiet-eq.c: New test.
* gcc.target/s390/zvector/autovec-double-quiet-ge.c: New test.
* gcc.target/s390/zvector/autovec-double-quiet-gt.c: New test.
* gcc.target/s390/zvector/autovec-double-quiet-le.c: New test.
* gcc.target/s390/zvector/autovec-double-quiet-lt.c: New test.
* gcc.target/s390/zvector/autovec-double-quiet-ordered.c: New test.
* gcc.target/s390/zvector/autovec-double-quiet-uneq.c: New test.
* gcc.target/s390/zvector/autovec-double-quiet-unordered.c: New test.
* gcc.target/s390/zvector/autovec-double-signaling-eq-z13-finite.c: New test.
* gcc.target/s390/zvector/autovec-double-signaling-eq-z13.c: New test.
* gcc.target/s390/zvector/autovec-double-signaling-eq.c: New test.
* gcc.target/s390/zvector/autovec-double-signaling-ge-z13-finite.c: New test.
* gcc.target/s390/zvector/autovec-double-signaling-ge-z13.c: New test.
* gcc.target/s390/zvector/autovec-double-signaling-ge.c: New test.
* gcc.target/s390/zvector/autovec-double-signaling-gt-z13-finite.c: New test.
* gcc.target/s390/zvector/autovec-double-signaling-gt-z13.c: New test.
* gcc.target/s390/zvector/autovec-double-signaling-gt.c: New test.
* gcc.target/s390/zvector/autovec-double-signaling-le-z13-finite.c: New test.
* gcc.target/s390/zvector/autovec-double-signaling-le-z13.c: New test.
* gcc.target/s390/zvector/autovec-double-signaling-le.c: New test.
* gcc.target/s390/zvector/autovec-double-signaling-lt-z13-finite.c: New test.
* gcc.target/s390/zvector/autovec-double-signaling-lt-z13.c: New test.
* gcc.target/s390/zvector/autovec-double-signaling-lt.c: New test.
* gcc.target/s390/zvector/autovec-double-signaling-ltgt-z13-finite.c: New test.
* gcc.target/s390/zvector/autovec-double-signaling-ltgt-z13.c: New test.
* gcc.target/s390/zvector/autovec-double-signaling-ltgt.c: New test.
* gcc.target/s390/zvector/autovec-double-smax-z13.F90: New test.
* gcc.target/s390/zvector/autovec-double-smax.F90: New test.
* gcc.target/s390/zvector/autovec-double-smin-z13.F90: New test.
* gcc.target/s390/zvector/autovec-double-smin.F90: New test.
* gcc.target/s390/zvector/autovec-float-quiet-eq.c: New test.
* gcc.target/s390/zvector/autovec-float-quiet-ge.c: New test.
* gcc.target/s390/zvector/autovec-float-quiet-gt.c: New test.
* gcc.target/s390/zvector/autovec-float-quiet-le.c: New test.
* gcc.target/s390/zvector/autovec-float-quiet-lt.c: New test.
* gcc.target/s390/zvector/autovec-float-quiet-ordered.c: New test.
* gcc.target/s390/zvector/autovec-float-quiet-uneq.c: New test.
* gcc.target/s390/zvector/autovec-float-quiet-unordered.c: New test.
* gcc.target/s390/zvector/autovec-float-signaling-eq.c: New test.
* gcc.target/s390/zvector/autovec-float-signaling-ge.c: New test.
* gcc.target/s390/zvector/autovec-float-signaling-gt.c: New test.
* gcc.target/s390/zvector/autovec-float-signaling-le.c: New test.
* gcc.target/s390/zvector/autovec-float-signaling-lt.c: New test.
* gcc.target/s390/zvector/autovec-float-signaling-ltgt.c: New test.
* gcc.target/s390/zvector/autovec-fortran.h: New test.
* gcc.target/s390/zvector/autovec-long-double-signaling-ge.c: New test.
* gcc.target/s390/zvector/autovec-long-double-signaling-gt.c: New test.
* gcc.target/s390/zvector/autovec-long-double-signaling-le.c: New test.
* gcc.target/s390/zvector/autovec-long-double-signaling-lt.c: New test.
* gcc.target/s390/zvector/autovec.h: New test.

From-SVN: r276872

S/390: Use signaling FP comparison instructions

dg-torture.exp=inf-compare-1.c is failing, because (qNaN > +Inf)
comparison is compiled to CDB instruction, which does not signal an
invalid operation exception. KDB should have been used instead.

This patch introduces a new CCmode and a new pattern in order to
generate signaling instructions in this and similar cases.

gcc/ChangeLog:

2019-10-11  Ilya Leoshkevich  <iii@linux.ibm.com>

PR target/77918
* config/s390/2827.md: Add new opcodes.
* config/s390/2964.md: Likewise.
* config/s390/3906.md: Likewise.
* config/s390/8561.md: Likewise.
* config/s390/s390-builtins.def (s390_vfchesb): Use
the new vec_cmpgev4sf_quiet_nocc.
(s390_vfchedb): Use the new vec_cmpgev2df_quiet_nocc.
(s390_vfchsb): Use the new vec_cmpgtv4sf_quiet_nocc.
(s390_vfchdb): Use the new vec_cmpgtv2df_quiet_nocc.
(vec_cmplev4sf): Use the new vec_cmplev4sf_quiet_nocc.
(vec_cmplev2df): Use the new vec_cmplev2df_quiet_nocc.
(vec_cmpltv4sf): Use the new vec_cmpltv4sf_quiet_nocc.
(vec_cmpltv2df): Use the new vec_cmpltv2df_quiet_nocc.
* config/s390/s390-modes.def (CCSFPS): New mode.
* config/s390/s390.c (s390_match_ccmode_set): Support CCSFPS.
(s390_select_ccmode): Return CCSFPS for LT, LE, GT, GE and LTGT.
(s390_branch_condition_mask): Reuse CCS for CCSFPS.
(s390_expand_vec_compare): Use non-signaling patterns where
necessary.
(s390_reverse_condition): Support CCSFPS.
* config/s390/s390.md (*cmp<mode>_ccsfps): New pattern.
* config/s390/vector.md: (VFCMP_HW_OP): Remove.
(asm_fcmp_op): Likewise.
(*smaxv2df3_vx): Use pattern for quiet comparison.
(*sminv2df3_vx): Likewise.
(*vec_cmp<VFCMP_HW_OP:code><mode>_nocc): Remove.
(*vec_cmpeq<mode>_quiet_nocc): New pattern.
(vec_cmpgt<mode>_quiet_nocc): Likewise.
(vec_cmplt<mode>_quiet_nocc): New expander.
(vec_cmpge<mode>_quiet_nocc): New pattern.
(vec_cmple<mode>_quiet_nocc): New expander.
(*vec_cmpeq<mode>_signaling_nocc): New pattern.
(*vec_cmpgt<mode>_signaling_nocc): Likewise.
(*vec_cmpgt<mode>_signaling_finite_nocc): Likewise.
(*vec_cmpge<mode>_signaling_nocc): Likewise.
(*vec_cmpge<mode>_signaling_finite_nocc): Likewise.
(vec_cmpungt<mode>): New expander.
(vec_cmpunge<mode>): Likewise.
(vec_cmpuneq<mode>): Use quiet patterns.
(vec_cmpltgt<mode>): Allow only on z14+.
(vec_cmpordered<mode>): Use quiet patterns.
(vec_cmpunordered<mode>): Likewise.
(VEC_CMP_EXPAND): Add ungt and unge.

gcc/testsuite/ChangeLog:

2019-10-11  Ilya Leoshkevich  <iii@linux.ibm.com>

* gcc.target/s390/vector/vec-scalar-cmp-1.c: Adjust
expectations.

From-SVN: r276871

Do not allocate ggc during streaming.

* gimple-streamer-out.c (output_gimple_stmt): Add explicit function
parameter.
* lto-streamer-out.c: Include tree-dfa.h.
(output_cfg): Do not use cfun.
(lto_prepare_function_for_streaming): New.
(output_function): Do not push cfun; do not initialize loop optimizer.
* lto-streamer.h (lto_prepare_function_for_streaming): Declare.
* passes.c (ipa_write_summaries): Use it.
(ipa_write_optimization_summaries): Do not modify bodies.
* tree-dfa.c (renumber_gimple_stmt_uids): Add function parameter.
* tree.dfa.h (renumber_gimple_stmt_uids): Update prototype.
* tree-ssa-dse.c (pass_dse::execute): Update use of
renumber_gimple_stmt_uids.
* tree-ssa-math-opts.c (pass_optimize_widening_mul::execute): Likewise.

* lto.c (lto_wpa_write_files): Prepare all bodies for streaming.

From-SVN: r276870

decl.c (annotate_value): Really test the sign of the value when deciding to build a NEGATE_EXPR.

* gcc-interface/decl.c (annotate_value) <INTEGER_CST>: Really test the
sign of the value when deciding to build a NEGATE_EXPR.
<PLUS_EXPR>: Remove redundant line.
<BIT_AND_EXPR>: Do the negation here.

From-SVN: r276866

decl.c (Gigi_Equivalent_Type): New case.

* gcc-interface/decl.c (Gigi_Equivalent_Type) <E_Array_Subtype>: New
case.  Return the base type if the subtype is not constrained.

From-SVN: r276865

decl.c (gnat_to_gnu_entity): New case to deal with the definition of named numbers.

* gcc-interface/decl.c (gnat_to_gnu_entity) <E_Named_{Integer|Real}>:
New case to deal with the definition of named numbers.
<E_Variable>: Minor tweaks.  Set DECL_IGNORED_P on the CONST_DECL
if a corresponding variable is built.
* gcc-interface/trans.c (gnat_to_gnu) <N_Integer_Literal>: Return
error_mark_node instead of aborting on overflow for named numbers.
<N_Number_Declaration>: Reuse the <N_Object_Declaration> case and
deal with error_mark_node specifically.
* gcc-interface/utils.c (create_var_decl): Do not set DECL_IGNORED_P
on CONST_DECLs.
(gnat_write_global_declarations): Output global constants.

From-SVN: r276864

[Ada] Repair ChangeLog entries

My scripts and I unexpectedly erased gcc/ada/ChangeLog yesterday. This
commit restores the ChangeLog file to its previous state and adds the
entries for my commits yesterday.

From-SVN: r276861

re PR c++/91987 (-fstrict-eval-order issues)

PR c++/91987
cp/
* decl2.c (grok_array_decl): For -fstrong-eval-order, when array ref
operands have been swapped and at least one operand has side-effects,
revert the swapping before calling build_array_ref.
* typeck.c (cp_build_array_ref): For non-ARRAY_TYPE array ref with
side-effects on the index operand, if -fstrong-eval-order use
save_expr around the array operand.
(cp_build_binary_op): For shifts with side-effects in the second
operand, wrap first operand into SAVE_EXPR and evaluate it before
the shift.
* semantics.c (handle_omp_array_sections_1): Temporarily disable
flag_strong_eval_order during OMP_CLAUSE_REDUCTION array section
processing.
* cp-gimplify.c (gimplify_to_rvalue): New function.
(cp_gimplify_expr): Use it.
testsuite/
* g++.dg/cpp1z/eval-order6.C: New test.
* g++.dg/cpp1z/eval-order7.C: New test.
* g++.dg/cpp1z/eval-order8.C: New test.
* c-c++-common/gomp/pr91987.c: New test.

From-SVN: r276860

vec_promote_demote cost to 1 for non-Power7 VSX architectures

2019-10-11  Kewen Lin  <linkw@gcc.gnu.org>

    * config/rs6000/rs6000.c (rs6000_builtin_vectorization_cost): Lower
    vec_promote_demote cost to 1 for non-Power7 VSX architectures.

From-SVN: r276859

Daily bump.

From-SVN: r276858

Update DFP macros in float.h for C2x and TS 18661-2.

GCC's <float.h> includes various macros for decimal floating-point
types, defined if __STDC_WANT_DEC_FP__, as defined in TS 24732.

TS 18661-2 and C2X make some changes to those macros.  In TS 18661-2,
they are to be defined if __STDC_WANT_IEC_60559_DFP_EXT__, and the
*_SUBNORMAL_MIN macros are renamed to *_TRUE_MIN.  In C2X, the macros
are to be defined whenever DFP is supported, without needing any
feature test macro to be defined.

This patch updates the header accordingly.  There is no existing
predefined macro for whether the target supports DFP; rather than
adding a new macro for that purpose, this patch changes the predefined
macros used in <float.h> so that they are only defined when DFP is
supported and thus can be used as conditionals; this is the same way
predefined macros for _FloatN and _FloatNx types work.  Although
formally TR 24732 and TS 18661-2 say nothing about implementations
lacking DFP support, it seems appropriate to avoid defining the macros
in <float.h>, in the absence of DFP support, even if the respective
feature test macros are defined.

Bootstrapped with no regressions on x86_64-pc-linux-gnu.  Also tested
the c2x-float-no-dfp-* tests for aarch64-linux-gnu to make sure they
do pass in a no-DFP configuration.

gcc:
* ginclude/float.h [!__DEC32_MANT_DIG__]: Do not define DFP
macros.
[__STDC_WANT_IEC_60559_DFP_EXT__ || __STDC_VERSION__ > 201710L]:
Also define DFP macros for these conditions.
[!__STDC_WANT_DEC_FP__] (DEC32_SUBNORMAL_MIN, DEC64_SUBNORMAL_MIN,
DEC128_SUBNORMAL_MIN): Do not define.
[__STDC_WANT_IEC_60559_DFP_EXT__ || __STDC_VERSION__ > 201710L]
(DEC32_TRUE_MIN, DEC64_TRUE_MIN, DEC128_TRUE_MIN): New macros.

gcc/c-family:
* c-cppbuiltin.c (c_cpp_builtins): Do not define macros for DFP
types if DFP not supported.

gcc/testsuite:
* gcc.dg/c11-float-dfp-1.c, gcc.dg/c2x-float-no-dfp-1.c,
gcc.dg/c2x-float-no-dfp-2.c, gcc.dg/dfp/c2x-float-dfp-1.c,
gcc.dg/dfp/c2x-float-dfp-2.c, gcc.dg/dfp/c2x-float-dfp-3.c,
gcc.dg/dfp/tr24732-float-dfp-1.c,
gcc.dg/dfp/ts18661-2-float-dfp-1.c: New tests.

From-SVN: r276854

re PR middle-end/26241 (None of the IPA passes are documented in passes.texi)

2019-10-10  Xiong Hu Luo  <luoxhu@linux.ibm.com>
    Sandra Loosemore  <sandra@codesourcery.com>

gcc/

PR middle-end/26241
* doc/lto.texi (IPA): Reference to the IPA passes.
* doc/passes.texi (Pass manager): Add node IPA passes and
  description for each IPA pass.

Co-Authored-By: Sandra Loosemore <sandra@codesourcery.com>
From-SVN: r276851

ipa-reference.c: Do not include splay-tree.h

* ipa-reference.c: Do not include splay-tree.h
(reference_vars_to_consider): Turn to hash map.
(get_static_name, ipa_init, analyze_function, propagate,
stream_out_bitmap, ipa_reference_write_optimization_summary,
ipa_reference_write_optimization_summary): Update.

From-SVN: r276850

* ipa-reference.c (propagate): Fix releasing of IPA summaries.

From-SVN: r276849

[Darwin, machopic 4/n] Arrange to indirect IVARs when needed.

Objective C V2 (m64) IVAR offset refs from Apple GCC-4.x have an indirection
for m64 code on PPC (which is the only 64b user for Mach-O PIC).  Apple GCC
4.x places the indirections in the .data section, however this seems to have
been unintentional - and we are placing the indirections in the non-lazy
symbol pointers section.

gcc/ChangeLog:

2019-10-10  Iain Sandoe  <iain@sandoe.co.uk>

* config/darwin.c: Lookup Objective C metadata and force indirection
for IVAR refs.

From-SVN: r276848

[Objective-C, NeXT ABI]  Identify V2 IVAR refs by metadata.

For the NeXT 64b ABI, IVAR refs are supposed to be indirected for
Mach-O PIC.  Identify them so that we can act as needed.

gcc/objc/ChangeLog:

2019-10-10  Iain Sandoe  <iain@sandoe.co.uk>

* objc-next-metadata-tags.h (OCTI_RT_META_IVAR_REF): New.
(meta_ivar_ref): New.
* objc-next-runtime-abi-02.c
(next_runtime_abi_02_init_metadata_attributes): Create the
IVAR ref metadata identifier.
(ivar_offset_ref): Tag IVAR refs with specific metadata.

From-SVN: r276847

Add check for prefixed addresses.

2019-10-10  Michael Meissner  <meissner@linux.ibm.com>

* config/rs6000/rs6000.c (quad_address_p): Add check for prefixed
addresses.
(mem_operand_gpr): Add check for prefixed addresses.
(mem_operand_ds_form): Add check for prefixed addresses.
(rs6000_legitimate_offset_address_p): If we support prefixed
addresses, check for a 34-bit offset instead of 16-bit.
(rs6000_legitimate_address_p): Add check for prefixed addresses.
Do not allow load/store with update if the address is prefixed.
(rs6000_mode_dependent_address):  If we support prefixed
addresses, check for a 34-bit offset instead of 16-bit.

From-SVN: r276846

Documentation hyperlinks for [-Wname-of-option] (PR 87488)

This patch uses the support for pretty-printing escaped URLs added in
the previous patch (PR 87488) so that in a sufficiently capable terminal
the [-Wname-of-option] emitted at the end of each diagnostic becomes a
hyperlink to the documentation for that option on the GCC website.

I've tested it with Fedora 30's GNOME Terminal (3.32.2 with VTE 0.56.3);
the option text is printed with a dotted underline, hovering shows the
URL, and on right-clicking it offers menu items to visit/copy the URL.

gcc/ChangeLog:
PR 87488
* Makefile.in (CFLAGS-opts.o): Pass in DOCUMENTATION_ROOT_URL via
-D.
* configure.ac (--with-documentation-root-url): New option.
* configure: Regenerate.
* diagnostic-format-json.cc (json_end_diagnostic): If there is an
option URL, add it as a new string field of the diagnostic option.
* diagnostic.c (diagnostic_initialize): Initialize get_option_url.
(print_option_information): If get_option_url is non-NULL, call
it, and if the result is non-NULL, potentially emit an escape
sequence to markup the option text with the resulting URL.
* diagnostic.h (diagnostic_context::get_option_url): New callback.
* doc/invoke.texi (-fdiagnostics-format=): Add "option_url" to
example of JSON output.
* opts-diagnostic.h (get_option_url): New decl.
* opts.c (get_option_url): New function.
* toplev.c (general_init): Initialize the get_option_url callback.

gcc/testsuite/ChangeLog:
PR 87488
* c-c++-common/diagnostic-format-json-2.c: Expect an "option_url"
field.
* c-c++-common/diagnostic-format-json-3.c: Likewise.
* gfortran.dg/diagnostic-format-json-2.F90: Likewise.
* gfortran.dg/diagnostic-format-json-3.F90: Likewise.
* jit.dg/test-error-array-bounds.c (create_code): Ensure that
error messages don't contain escaped URLs.

From-SVN: r276843

[PATCH 1/3] S/390: Do not use signaling vector comparisons on z13

z13 supports only non-signaling vector comparisons.  This means we
cannot vectorize LT, LE, GT, GE and LTGT when compiling for z13.  Notify
middle-end about this by using more restrictive operator predicate in
vcond<V_HW:mode><V_HW2:mode>.

gcc/ChangeLog:

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

PR target/77918
* config/s390/vector.md (vcond_comparison_operator): New
predicate.
(vcond<V_HW:mode><V_HW2:mode>): Use vcond_comparison_operator.

From-SVN: r276842

pretty-print: support URL escape sequences (PR 87488)

https://gist.github.com/egmontkob/eb114294efbcd5adb1944c9f3cb5feda
describes an emerging standard for embedding URLs in escape sequences
for marking up text output.  This is supported e.g. by recent releases
of GNOME Terminal.

This patch adds support to our pretty-printing framework for emitting
URLs.

A followup patch uses this to add URLs to the pertinent documentation
for the output of -fdiagnostics-show-option.

gcc/ChangeLog:
PR 87488
* common.opt (fdiagnostics-urls=): New option.
(diagnostic-url.h): Add SourceInclude.
(diagnostic_url_rule): New enum.
* diagnostic-color.c: Include "diagnostic-url.h".
(diagnostic_urls_enabled_p): New function.
* diagnostic-url.h: New file.
* diagnostic.c: Include "diagnostic-url.h".
(diagnostic_urls_init): New function.
* diagnostic.h (diagnostic_urls_init): New decl.
* doc/invoke.texi (Diagnostic Message Formatting Options): Add
-fdiagnostics-urls to the list.
(-fdiagnostics-urls): New option.
* gcc.c (driver_handle_option): Handle OPT_fdiagnostics_urls_.
(driver::global_initializations): Call diagnostic_urls_init.
* opts-global.c (init_options_once): Likewise.
* opts.c (common_handle_option): Handle OPT_fdiagnostics_urls_.
* pretty-print.c (pretty_printer::pretty_printer): Initialize
show_urls.
(pp_begin_url): New function.
(pp_end_url): New function.
(selftest::test_urls): New selftest.
(selftest::pretty_print_c_tests): Call it.
* pretty-print.h (pretty_printer::show_urls): New field.
(pp_begin_url): New decl.
(pp_end_url): New decl.

gcc/testsuite/ChangeLog:
PR 87488
* lib/prune.exp (TEST_ALWAYS_FLAGS): Add -fdiagnostics-urls=never.

From-SVN: r276841

PR libstdc++/91057 fix bootstrap failure on powerpc

PR libstdc++/91057
* src/c++98/locale.cc [_GLIBCXX_LONG_DOUBLE_COMPAT]
(find_ldbl_sync_facet): Fix parameter type and missing return.

From-SVN: r276840

re PR target/92022 (ICE in alpha_handle_trap_shadows, at config/alpha/alpha.c:8847)

PR target/92022
* config/alpha/alpha.c (alpha_handle_trap_shadows): Skip DEBUG_INSN.

testsuite/ChangeLog:

PR target/92022
* g++.dg/pr92022.C: New test.

From-SVN: r276839

[Ada] Handling up-level references in loops within library-level declarations

2019-10-10  Gary Dismukes  <dismukes@adacore.com>

gcc/ada/

* exp_ch7.adb (Check_Unnesting_In_Decls_Or_Stmts): When
encountering a loop at the top level of a package declaration
list (that is, within the declarations of a package spec or
body) that has nested subprograms, call Unnest_Loop to create a
new library-level procedure that will contain the loop, to allow
for proper handling of up-level references from within nested
subprograms, such as to loop parameters.
(Unnest_Loop): New procedure that takes a loop statement and
creates a new procedure body to enclose the loop statement,
along with generating a call to the procedure.

From-SVN: r276836

[Ada] Ensure constructor is a C++ constructor

2019-10-10  Arnaud Charlet  <charlet@adacore.com>

gcc/ada/

* freeze.adb (Freeze_Subprogram): Ensure constructor is a C++
constructor.

From-SVN: r276835

[Ada] Add pragma Preelaborable_Initialization to Stream_IO.File_Type

2019-10-10  Gary Dismukes  <dismukes@adacore.com>

gcc/ada/

* libgnat/a-ststio.ads (File_Type): Apply pragma
Preelaborable_Initialization to the type.

From-SVN: r276834

[Ada] Fix inlining of subprograms with deep param/result in GNATprove

2019-10-10  Yannick Moy  <moy@adacore.com>

gcc/ada/

* inline.adb (Can_Be_Inlined_In_GNATprove_Mode): Do not peek
under private types whose completion is SPARK_Mode Off.

From-SVN: r276833