re PR c++/83059 (ICE on invalid C++ code: in tree_to_uhwi, at tree.c:6633)

PR c++/83059
* c-common.c (get_atomic_generic_size): Use TREE_INT_CST_LOW
instead of tree_to_uhwi, formatting fix.
* config/i386/i386.c (ix86_memmodel_check): Start
-Winvalid-memory-model diagnostics with lowercase letter.

* c-c++-common/pr83059.c: New test.

From-SVN: r254990

re PR debug/82718 (Bad DWARF5 .debug_loclists generation)

PR debug/82718
* dwarf2out.c (dw_loc_list): If crtl->has_bb_partition, temporarily
set in_cold_section_p to the partition containing loc_list->first.
When seeing loc_list->last_before_switch node, update secname and
perform range_across_switch second partition handling only after that.

* gcc.dg/debug/dwarf2/pr82718-1.c: New test.
* gcc.dg/debug/dwarf2/pr82718-2.c: New test.

From-SVN: r254989

re PR debug/82933 (valgrind error in set_cur_line_info_table with -g)

PR debug/82933
* run-rtl-passes.c: Include debug.h.
(run_rtl_passes): Call debug_hooks->assembly_start.
* dwarf2out.c (dwarf2out_assembly_start): Return early if invoked
multiple times.

* gcc.dg/rtl/x86_64/pr82933.c: New test.

From-SVN: r254987

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

PR target/82981
* internal-fn.c (expand_mul_overflow): Use OPTAB_WIDEN instead of
OPTAB_DIRECT in calls to expand_simple_binop.

From-SVN: r254986

RTEMS: Enable some libstdc++ features

libstdc++/

* acinclude.m4 (GLIBCXX_ENABLE_LIBSTDCXX_TIME): Add rtems*.
(GLIBCXX_ENABLE_FILESYSTEM_TS): Likewise.
* configure: Regenerate.

From-SVN: r254985

libgo: fix makefile buglet

    Fix a small bug in the libgo Makefile recipe that constructs the
    directory from which to pick up libgcc_s.so ; the gccgo invocation
    with -print-libgcc-file-name was missing the flags, which meant that
    for -m32 builds we'd see the 64-bit libgcc dir.

    Reviewed-on: https://go-review.googlesource.com/78836

From-SVN: r254984

compiler: report error for ++/-- applied to a non-numeric type

    This avoids a compiler crash.

    Fixes GCC PR 83071.

    Reviewed-on: https://go-review.googlesource.com/78875

From-SVN: r254983

re PR libfortran/78549 (Very slow formatted internal file output)

2017-11-20  Jerry DeLisle  <jvdelisle@gcc.gnu.org>

PR libgfortran/78549
* io/io.h (newunit_free): Add declaration. Clean some whitespace.
* io/transfer.c (st_read_done, st_write_done): Call newunit_free.
* io/unit.c (newunit_free): Change type from static void to void.

From-SVN: r254982

Use -Wtraditional for "would be stringified in traditional C" (PR preprocessor/81794)

libcpp/ChangeLog:

2017-03-24  Eric Gallager  <egall@gwmail.gwu.edu>

PR preprocessor/81794
* macro.c (check_trad_stringification): Have warning be controlled
by -Wtraditional.

gcc/testsuite/ChangeLog:

2017-09-17  Eric Gallager  <egall@gwmail.gwu.edu>

PR preprocessor/81794
* gcc.dg/pragma-diag-7.c: Update to include check for
stringification.

From-SVN: r254981

C/C++: more stdlib header hints (PR c/81404)

This patch extends the C frontend's "knowledge" of the C stdlib within
get_c_name_hint to cover some more macros and functions, covering
a case reported in PR c/81404 ("INT_MAX"), so that rather than printing:

  t.c:5:12: error: 'INT_MAX' undeclared here (not in a function); did you mean '__INT_MAX__'?
   int test = INT_MAX;
              ^~~~~~~
              __INT_MAX__

we instead print:

  t.c:5:12: error: 'INT_MAX' undeclared here (not in a function)
   int test = INT_MAX;
              ^~~~~~~
  t.c:5:12: note: 'INT_MAX' is defined in header '<limits.h>'; did you forget to '#include <limits.h>'?
  t.c:1:1:
  +#include <limits.h>

  t.c:5:12:
    int test = INT_MAX;
               ^~~~~~~

It also adds generalizes some of the code for this (and for the "std::"
namespace hints in the C++ frontend), moving it to a new
c-family/known-headers.cc and .h, and introducing a class known_headers.
This currently just works by scanning a hardcoded array of known
name/header associations, but perhaps in the future could be turned
into some kind of symbol database so that the compiler could record API
uses and use that to offer suggestions e.g.

foo.cc: error: 'myapi::foo' was not declared in this scope
foo.cc: note: 'myapi::foo" was declared in header 'myapi/private.h'
(included via 'myapi/public.h') when compiling 'bar.cc'; did you forget to
'#include "myapi/public.h"'?

or somesuch.

In any case, moving this to a class gives an easier way to locate the
hardcoded knowledge about the stdlib.

The patch also adds similar code to the C++ frontend covering
unqualified names in the standard library, so that rather than just
e.g.:

  t.cc:19:13: error: 'NULL' was not declared in this scope
   void *ptr = NULL;
               ^~~~

we can emit:

  t.cc:19:13: error: 'NULL' was not declared in this scope
   void *ptr = NULL;
               ^~~~
  t.cc:19:13: note: 'NULL' is defined in header '<cstddef>'; did you forget
  to '#include <cstddef>'?
  t.cc:1:1:
  +#include <cstddef>

  t.cc:19:13:
   void *ptr = NULL;
               ^~~~

(Also XFAIL for PR c++/80567 added for the C++ testcase; this is a
separate pre-existing bug exposed by the testcase for PR 81404).

gcc/ChangeLog:
PR c/81404
* Makefile.in (C_COMMON_OBJS): Add c-family/known-headers.o.

gcc/c-family/ChangeLog:
PR c/81404
* known-headers.cc: New file, based on material from c/c-decl.c.
(suggest_missing_header): Copied as-is.
(get_stdlib_header_for_name): New, based on get_c_name_hint but
heavily edited to add C++ support.  Add some knowledge about
<limits.h>, <stdint.h>, and <wchar.h>.
* known-headers.h: Likewise.

gcc/c/ChangeLog:
PR c/81404
* c-decl.c: Include "c-family/known-headers.h".
(get_c_name_hint): Rename to get_stdlib_header_for_name and move
to known-headers.cc.
(class suggest_missing_header): Move to known-header.h.
(lookup_name_fuzzy): Call get_c_stdlib_header_for_name rather
than get_c_name_hint.

gcc/cp/ChangeLog:
PR c/81404
* name-lookup.c: Include "c-family/known-headers.h"
(lookup_name_fuzzy): Call get_cp_stdlib_header_for_name and
potentially return a new suggest_missing_header hint.

gcc/testsuite/ChangeLog:
PR c/81404
* g++.dg/spellcheck-stdlib.C: New.
* gcc.dg/spellcheck-stdlib.c (test_INT_MAX): New.

From-SVN: r254980

C: hints for missing stdlib includes for macros and types

The C frontend already "knows" about many common functions in
the C standard library:

  test.c: In function 'test':
  test.c:3:3: warning: implicit declaration of function 'printf' [-Wimplicit-function-declaration]
     printf ("hello world\n");
     ^~~~~~
  test.c:3:3: warning: incompatible implicit declaration of built-in function 'printf'
  test.c:3:3: note: include '<stdio.h>' or provide a declaration of 'printf'

and which header file they are in.

However it doesn't know about various types and macros:

test.c:1:13: error: 'NULL' undeclared here (not in a function)
 void *ptr = NULL;
             ^~~~

This patch uses the name_hint/deferred_diagnostic machinery to
add hints for missing C standard library headers for some of the
most common type and macro names.

For example, the above becomes:
test.c:1:13: error: 'NULL' undeclared here (not in a function)
 void *ptr = NULL;
             ^~~~
test.c:1:13: note: 'NULL' is defined in header '<stddef.h>'; did you forget to '#include <stddef.h>'?

gcc/c/ChangeLog:
* c-decl.c (get_c_name_hint): New function.
(class suggest_missing_header): New class.
(lookup_name_fuzzy): Call get_c_name_hint and use it to
suggest missing headers to the user.

gcc/testsuite/ChangeLog:
* gcc.dg/spellcheck-stdlib.c: New test case.

From-SVN: r254979

C++: provide macro used-before-defined hint (PR c++/72786)

This patch uses the name_hint/deferred_diagnostic to provide
a message in the C++ frontend if a macro is used before it is defined
e.g.:

test.c:6:24: error: expected ';' at end of member declaration
   virtual void clone() const OVERRIDE { }
                        ^~~~~
                             ;
test.c:6:30: error: 'OVERRIDE' does not name a type
   virtual void clone() const OVERRIDE { }
                              ^~~~~~~~
test.c:6:30: note: the macro 'OVERRIDE' had not yet been defined
test.c:15:0: note: it was later defined here
 #define OVERRIDE override

It's possible to do it from the C++ frontend as tokenization happens
up-front (and hence the macro already exists when the above is parsed);
I attempted to do it from the C frontend, but because the C frontend only
tokenizes on-demand during parsing, the macro isn't known about until
later.

gcc/cp/ChangeLog:
PR c++/72786
* name-lookup.c (class macro_use_before_def): New class.
(lookup_name_fuzzy): Detect macro that were used before being
defined, and report them as such.

gcc/ChangeLog:
PR c++/72786
* spellcheck.h (best_match::blithely_get_best_candidate): New
accessor.

gcc/testsuite/ChangeLog:
PR c++/72786
* g++.dg/spellcheck-macro-ordering-2.C: New test case.
* g++.dg/spellcheck-macro-ordering.C: Add dg-message directives
for macro used-before-defined.

libcpp/ChangeLog:
PR c++/72786
* include/cpplib.h (cpp_macro_definition_location): New decl.
* macro.c (cpp_macro_definition): New function.

From-SVN: r254978

re PR target/81356 (__builtin_strcpy is not good for copying an empty string on aarch64)

2017-11-20  Steve Ellcey  <sellcey@cavium.com>

PR target/81356
* gfortran.dg/pr45636.f90 (aarch64*-*-*): Remove from xfail list.

From-SVN: r254977

Daily bump.

From-SVN: r254976

Avoid duplicate visibility warning.

* decl2.c (constrain_class_visibility): Don't warn about artificial
fields.

From-SVN: r254973

streambuf_iterator.h (istreambuf_iterator<>): Declare std::advance for istreambuf_iterator of char types to be friend.

2017-11-20  François Dumont  <fdumont@gcc.gnu.org>

* include/bits/streambuf_iterator.h (istreambuf_iterator<>): Declare
std::advance for istreambuf_iterator of char types to be friend.
(std::advance(istreambuf_iterator&, _Distance)): New overload.
* include/std/streambuf (basic_streambuf<>): Declare std::advance for
istreambuf_iterator of char types to be friend.
* testsuite/22_locale/money_get/get/char/9.cc: Have
istreambuf_iterator created on the fly when calling money_get<>::get.
* testsuite/22_locale/money_get/get/wchar_t/9.cc: Likewise.
* testsuite/24_iterators/istreambuf_iterator/debug/1_neg.cc: New.
* testsuite/24_iterators/istreambuf_iterator/debug/2_neg.cc: New.
* testsuite/25_algorithms/advance/istreambuf_iterators/char/1.cc: New.
* testsuite/25_algorithms/advance/istreambuf_iterators/char/1_neg.cc:
New.
* testsuite/25_algorithms/advance/istreambuf_iterators/char/2.cc: New.
* testsuite/25_algorithms/advance/istreambuf_iterators/char/2_neg.cc:
New.
* testsuite/25_algorithms/advance/istreambuf_iterators/char/3_neg.cc:
New.
* testsuite/25_algorithms/advance/istreambuf_iterators/wchar_t/1.cc:
New.
* testsuite/25_algorithms/advance/istreambuf_iterators/wchar_t/1_neg.cc:
New.
* testsuite/25_algorithms/advance/istreambuf_iterators/wchar_t/2.cc:
New.
* testsuite/25_algorithms/advance/istreambuf_iterators/wchar_t/2_neg.cc:
New.
* testsuite/25_algorithms/advance/istreambuf_iterators/wchar_t/3_neg.cc:
New.
* testsuite/25_algorithms/find/istreambuf_iterators/char/2.cc: Leverage
on std::advance overload.
* testsuite/25_algorithms/find/istreambuf_iterators/wchar_t/2.cc:
Likewise.

From-SVN: r254972

Fix failing tests caused by duplicate dg-options

* testsuite/special_functions/18_riemann_zeta/check_value.cc: Fix
duplicate dg-options directive.
* testsuite/tr1/5_numerical_facilities/special_functions/
20_riemann_zeta/check_value_neg.cc: Likewise.

From-SVN: r254971

i386.c (parse_mtune_ctrl_str): Start diagnostics with lower case letter.

* config/i386/i386.c (parse_mtune_ctrl_str): Start diagnostics
with lower case letter.

From-SVN: r254970

i386.md (bswaphi2): New expander.

* config/i386/i386.md (bswaphi2): New expander.
(*bswaphi2_movbe): New insn pattern.
(bswaphi -> rorhi pepehole2): New peephole pattern.

testsuite/ChangeLog:

* gcc.target/i386/movbe-5.c: New test.

From-SVN: r254967

re PR fortran/79072 (ICE with class(*) pointer function result and character value)

2017-11-20  Paul Thomas  <pault@gcc.gnu.org>

PR fortran/79072
* trans-expr.c (trans_class_vptr_len_assignment): Set from_len
if the temporary is unlimited polymorphic.
* trans-stmt.c (trans_associate_var): Use the fake result decl
to obtain the 'len' field from an explicit function result when
in that function scope.

2017-11-20  Paul Thomas  <pault@gcc.gnu.org>

PR fortran/79072
* gfortran.dg/class_result_5.f90: New test.

From-SVN: r254966

RISC-V: Implement __umulsidi3, umul_ppmm and __muluw3

2017-11-20  Kito Cheng  <kito.cheng@gmail.com>

        * longlong.h [__riscv] (__umulsidi3): Define.
        [__riscv] (umul_ppmm): Likewise.
        [__riscv] (__muluw3): Likewise.

From-SVN: r254965

P0329R4: Designated Initialization

P0329R4: Designated Initialization
* parser.c (cp_parser_initializer_clause): List in comment grammar
designated-initializer-list.
(cp_parser_initializer_list): Allow .identifier = without pedwarn for
C++2A, parse .identifier { ... }.  Improve location_t argument to
pedwarn.  Add pedwarn for [cst] = designators.  Diagnose ... in
designated initializer list.  Diagnose mixing designated and
non-designated initializer clauses for C++2A.  Diagnose duplicated
identifiers in designators.
* name-lookup.h (search_anon_aggr): New declaration.
* name-lookup.c (fields_linear_search): Use search_anon_aggr.
(search_anon_aggr): New function.
* typeck2.c (process_init_constructor_record): Allow designator
to skip over some non-static data members.  Handle anonymous
aggregates.  Add diagnostics for designator order not matching
member declaration order.

* g++.dg/ext/desig2.C: Adjust comment, no sorry about designator
refering to second member.
(b): New variable and associated expected diagnostic.
* g++.dg/ext/desig4.C: For C++2A expect diagnostics.
* g++.dg/ext/desig5.C: Add dg-do dg-compile and empty dg-options.
* g++.dg/ext/desig8.C: Likewise.
* g++.dg/ext/desig9.C: New test.
* g++.dg/ext/pr27019.C: Don't expect any diagnostics.
* g++.dg/init/error2.C: Adjust expected diagnostics.
* g++.dg/cpp0x/desig1.C: Add dg-options with -pedantic, expect
warning on C99 designators.
* g++.dg/cpp2a/desig1.C: New test.
* g++.dg/cpp2a/desig2.C: New test.
* g++.dg/cpp2a/desig3.C: New test.
* g++.dg/cpp2a/desig4.C: New test.
* g++.dg/cpp2a/desig5.C: New test.
* g++.dg/cpp2a/desig6.C: New test.

From-SVN: r254964

c-family: add name_hint/deferred_diagnostic

In various places we use lookup_name_fuzzy to provide a hint,
and can report messages of the form:
  error: unknown foo named 'bar'
or:
  error: unknown foo named 'bar'; did you mean 'SUGGESTION?

This patch provides a way for lookup_name_fuzzy to provide
both the suggestion above, and (optionally) additional hints
that can be printed e.g.

  note: did you forget to include <SOME_HEADER.h>?

This patch provides the mechanism and ports existing users
of lookup_name_fuzzy to the new return type.
There are no uses of such hints in this patch, but followup
patches provide various front-end specific uses of this.

gcc/c-family/ChangeLog:
* c-common.h (enum lookup_name_fuzzy_kind): Move to name-hint.h.
(lookup_name_fuzzy): Likewise.  Convert return type from
const char * to name_hint.  Add location_t param.
* name-hint.h: New header.

gcc/c/ChangeLog:
* c-decl.c: Define INCLUDE_UNIQUE_PTR before including system.h.
Include "c-family/name-hint.h"
(implicit_decl_warning): Convert "hint" from
const char * to name_hint.  Pass location to
lookup_name_fuzzy.  Suppress any deferred diagnostic if the
warning was not printed.
(undeclared_variable): Likewise for "guessed_id".
(lookup_name_fuzzy): Convert return type from const char *
to name_hint.  Add location_t param.
* c-parser.c: Define INCLUDE_UNIQUE_PTR before including system.h.
Include "c-family/name-hint.h"
(c_parser_declaration_or_fndef): Convert "hint" from
const char * to name_hint.  Pass location to lookup_name_fuzzy.
(c_parser_parameter_declaration): Likewise.

gcc/cp/ChangeLog:
* name-lookup.c: Define INCLUDE_UNIQUE_PTR before including system.h.
Include "c-family/name-hint.h"
(suggest_alternatives_for): Convert "fuzzy_name" from const char *
to name_hint, and rename to "hint".  Pass location to
lookup_name_fuzzy.
(lookup_name_fuzzy): Convert return type from const char *
to name_hint.  Add location_t param.
* parser.c: Define INCLUDE_UNIQUE_PTR before including system.h.
Include "c-family/name-hint.h"
(cp_parser_diagnose_invalid_type_name): Convert
"suggestion" from const char * to name_hint, and rename to "hint".
Pass location to lookup_name_fuzzy.

From-SVN: r254963

Makefile.in (OBJS): Add gimple-ssa-evrp-analyze.o.

* Makefile.in (OBJS): Add gimple-ssa-evrp-analyze.o.
* gimple-ssa-evrp-analyze.c: New file pulled from gimple-ssa-evrp.c.
* gimple-ssa-evrp-analyze.h: New file pulled from gimple-ssa-evrp.c.
* gimple-ssa-evrp.c: Remove bits moved into new files.  Include
gimple-ssa-evrp-analyze.h.

From-SVN: r254961

gimple-ssa-evrp.c (evrp_dom_walker::before_dom_children): Do not set BB_VISITED here.

* gimple-ssa-evrp.c (evrp_dom_walker::before_dom_children): Do not
set BB_VISITED here.
(evrp_range_analyzer::enter): Set BB_VISITED here instead.

From-SVN: r254960

[PR c++/82878] pass-by-invisiref in lambda

https://gcc.gnu.org/ml/gcc-patches/2017-11/msg01115.html
PR c++/82878
PR c++/78495
* call.c (build_call_a): Don't set CALL_FROM_THUNK_P for inherited
ctor.
* cp-gimplify.c (cp_genericize_r): Restore THUNK dereference
inhibibition check removed in previous c++/78495 change.

PR c++/82878
* g++.dg/cpp0x/pr82878.C: New.
* g++.dg/cpp1z/inh-ctor38.C: Check moves too.

From-SVN: r254958

Implement LWG 2353

* include/bits/stl_iterator_base_funcs.h (next):
Use InputIterator instead of ForwardIterator.
* testsuite/24_iterators/operations/lwg2353.cc: New.
* testsuite/24_iterators/operations/next_neg.cc: Remove.

From-SVN: r254957

tree-predcom.c: Add general comment on Store-Store chains.

* tree-predcom.c: Add general comment on Store-Store chains.
(split_data_refs_to_components): Postpone clearing eliminate_store_p
flag in component.
(get_chain_last_ref_at): Rename into...
(get_chain_last_write_at): ...this.
(get_chain_last_write_before_load): New function.
(add_ref_to_chain): Promote type of chain from CT_STORE_LOAD to
CT_STORE_STORE when write reference is added.
(determine_roots_comp): Support load ref in CT_STORE_STORE chains.
(is_inv_store_elimination_chain): Update get_chain_last_write_at call.
(initialize_root_vars_store_elim_1): Ditto.
(initialize_root_vars_store_elim_2): Ditto.  Replace rhs once default
definition is created.
(execute_pred_commoning_chain): Support load ref in CT_STORE_STORE
chain by replacing it with dominant stored value.

gcc/testsuite
* gcc.dg/tree-ssa/predcom-dse-12.c: New test.

From-SVN: r254956

* tree-predcom.c (add_ref_to_chain): Remove check on distance.

From-SVN: r254955

VRP: x+1 and -x cannot be INT_MIN

2017-11-20  Marc Glisse  <marc.glisse@inria.fr>

gcc/
* vr-values.c (extract_range_from_binary_expr): Use a full range
for VR_VARYING.

gcc/testsuite/
PR testsuite/82951
* gcc.c-torture/execute/20040409-1.c: Move invalid tests...
* gcc.c-torture/execute/20040409-1w.c: ... here with -fwrapv.
* gcc.c-torture/execute/20040409-2.c: Move invalid tests...
* gcc.c-torture/execute/20040409-2w.c: ... here with -fwrapv.
* gcc.c-torture/execute/20040409-3.c: Move invalid tests...
* gcc.c-torture/execute/20040409-3w.c: ... here with -fwrapv.
* gcc.dg/tree-ssa/cmpmul-1.c: Tweak condition.
* gcc.dg/tree-ssa/vrp118.c: New file.

From-SVN: r254954

Add g++.dg/pr82836.C requirements

* g++.dg/pr82836.C: Require int128, __float128 support.
Add __float128 options.
(size_t): Define using __SIZE_TYPE__.

From-SVN: r254953

re PR c++/82781 (Vector extension operators return wrong result in constexpr)

PR c++/82781
* constexpr.c (cxx_eval_vector_conditional_expression): New function.
(cxx_eval_constant_expression) <case VEC_COND_EXPR>: Use it instead
of cxx_eval_conditional_expression.

* g++.dg/ext/constexpr-pr82781.C: New test.

From-SVN: r254952

re PR bootstrap/83015 (bootstrap comparison failure on ia64)

PR bootstrap/83015
* config/cr16/unwind-cr16.c (uw_install_context): Add FRAMES
parameter.
* config/xtensa/unwind-dw2-xtensa.c: Likewise
* config/ia64/unwind-ia64.c: Add frames parameter.
* unwind-sjlj.c: Likewise.

From-SVN: r254951

[ARM] Do no clobber r4 in Armv8-M nonsecure call

Expanders for Armv8-M nonsecure call unnecessarily clobber r4 despite
the libcall they perform not writing to r4.  Furthermore, the
requirement for the branch target address to be in r4 as expected by
the libcall is modeled in a convoluted way in the define_insn patterns:
the address is a register match_operand constrained by the match_dup
for the clobber which is guaranteed to be r4 due to the expander.

This patch simplifies all this by simply requiring the address to be in
r4 and removing the clobbers. Expanders are left alone because
cmse_nonsecure_call_clear_caller_saved relies on branch target memory
attributes which would be lost if expanding to reg:SI R4_REGNUM.

2017-11-20  Thomas Preud'homme  <thomas.preudhomme@arm.com>

    gcc/
    * config/arm/arm.md (R4_REGNUM): Define constant.
    (nonsecure_call_internal): Remove r4 clobber.
    (nonsecure_call_value_internal): Likewise.
    * config/arm/thumb1.md (nonsecure_call_reg_thumb1_v5): Remove second
    clobber and resequence match_operands.
    (nonsecure_call_value_reg_thumb1_v5): Likewise.
    * config/arm/thumb2.md (nonsecure_call_reg_thumb2): Likewise.
    (nonsecure_call_value_reg_thumb2): Likewise.

From-SVN: r254950

[testsuite,arm] [Diagnostic Patch] don't print column zero

2017-11-20  Christophe Lyon  <christophe.lyon@linaro.org>

gcc/testsuite/
* gcc.target/arm/pr69180.c: Use -: for no column in expected
warnings.

From-SVN: r254949

re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)

PR tree-optimization/78821
* gimple-ssa-store-merging.c (find_bswap_or_nop_load): Give up
if base is TARGET_MEM_REF.  If base is not MEM_REF, set base_addr
to the address of the base rather than the base itself.
(find_bswap_or_nop_1): Just use pointer comparison for vuse check.
(find_bswap_or_nop_finalize): New function.
(find_bswap_or_nop): Use it.
(bswap_replace): Return a tree rather than bool, change first
argument from gimple * to gimple_stmt_iterator, allow inserting
into an empty sequence, allow ins_stmt to be NULL - then emit
all stmts into gsi.  Fix up MEM_REF address gimplification.
(pass_optimize_bswap::execute): Adjust bswap_replace caller.
(struct store_immediate_info): Add N and INS_STMT non-static
data members.
(store_immediate_info::store_immediate_info): Initialize them
from newly added ctor args.
(merged_store_group::apply_stores): Formatting fixes.  Sort by
bitpos at the end.
(stmts_may_clobber_ref_p): For stores call also
refs_anti_dependent_p.
(gather_bswap_load_refs): New function.
(imm_store_chain_info::try_coalesce_bswap): New method.
(imm_store_chain_info::coalesce_immediate_stores): Use it.
(split_group): Handle LROTATE_EXPR and NOP_EXPR rhs_code specially.
(imm_store_chain_info::output_merged_store): Fail if number of
new estimated stmts is bigger or equal than old.  Handle LROTATE_EXPR
and NOP_EXPR rhs_code.
(pass_store_merging::process_store): Compute n and ins_stmt, if
ins_stmt is non-NULL and the store rhs is otherwise invalid, use
LROTATE_EXPR rhs_code.  Pass n and ins_stmt to store_immediate_info
ctor.
(pass_store_merging::execute): Calculate dominators.

* gcc.dg/store_merging_16.c: New test.

From-SVN: r254948

tree-ssa-math-opts.c (nop_stats, [...]): Moved to ...

* tree-ssa-math-opts.c (nop_stats, bswap_stats, struct symbolic_number,
BITS_PER_MARKER, MARKER_MASK, MARKER_BYTE_UNKNOWN, HEAD_MARKER, CMPNOP,
CMPXCHG, do_shift_rotate, verify_symbolic_number_p,
init_symbolic_number, find_bswap_or_nop_load, perform_symbolic_merge,
find_bswap_or_nop_1, find_bswap_or_nop, pass_data_optimize_bswap,
class pass_optimize_bswap, bswap_replace,
pass_optimize_bswap::execute): Moved to ...
* gimple-ssa-store-merging.c: ... this file.
Include optabs-tree.h.
(nop_stats, bswap_stats, do_shift_rotate, verify_symbolic_number_p,
init_symbolic_number, find_bswap_or_nop_load, perform_symbolic_merge,
find_bswap_or_nop_1, find_bswap_or_nop, bswap_replace): Put into
anonymous namespace, remove static keywords.
(pass_optimize_bswap::gate): Test BITS_PER_UNIT == 8 here...
(pass_optimize_bswap::execute): ... rather than here.  Formatting fix.

From-SVN: r254947

re PR bootstrap/83062 (Bootstrap failure: libsanitizer/tsan/tsan_rtl.h:713:44: error: inlining failed in call to always_inline ‘void __tsan::MemoryRead(__tsan::ThreadState*, __sanitizer::uptr, __sanitizer: :uptr, int)’: caller is not optimized)

PR bootstrap/83062
* ipa-inline.c (can_inline_edge_p): Fix typo in previous patch.

From-SVN: r254946

vec.h (debug_helper): New function.

* vec.h (debug_helper): New function.
(DEFINE_DEBUG_VEC): New macro.
* hash-set.h (debug_helper): New function.
(DEFINE_DEBUG_HASH_SET): New macro.
* cfg.c (debug_slim (edge)): New function.
Call DEFINE_DEBUG_VEC for edges.
Call DEFINE_DEBUG_HASH_SET for edges.
* cfghooks.c (debug_slim (basic_block)): New function.
Call DEFINE_DEBUG_VEC for basic blocks.
Call DEFINE_DEBUG_HASH_SET for basic blocks.
* print-tree.c (debug_slim): New function to handle trees.
Call DEFINE_DEBUG_VEC for trees.
Call DEFINE_DEBUG_HASH_SET for trees.
(debug (vec<tree, va_gc>) &): Remove.
(debug (<vec<tree, va_gc>) *): Remove.
* print-rtl.c (debug_slim): New function to handle const_rtx.
Call DEFINE_DEBUG_VEC for rtx_def.
Call DEFINE_DEBUG_VEC for rtx_insn.
Call DEFINE_DEBUG_HASH_SET for rtx_def.
Call DEFINE_DEBUG_HASH_SET for rtx_insn.
* sel-sched-dump.c (debug (vec<rtx_insn *> &): Remove.
(debug (vec<rtx_insn *> *ptr): Remove.
(debug_insn_vector): Remove.
* stor-layout.c (debug_rli): Call debug() instead of debug_vec_tree.

From-SVN: r254945

Fix comparison mode in simplify_ternary_operation

2017-11-20  Tom de Vries  <tom@codesourcery.com>

PR rtl-optimization/82020
* simplify-rtx.c (simplify_ternary_operation): Fix comparison mode of
IF_THEN_ELSE condition.

From-SVN: r254944

Daily bump.

From-SVN: r254943

re PR ada/83016 (gnat1: warning: command line option ‘-nostdinc++’ is valid for C++/ObjC++ but not for Ada)

PR ada/83016
* gnatlink.adb (Process_Args): Accept multiple switches for --LINK.
(Usage): Adjust.
* gcc-interface/Makefile.in (GCC_LINK): Remove $(ADA_INCLUDES).
(common-tools): Pass $(CC) as --GCC= and $(GCC_LINK) as --LINK= in
the invocations of $(GNATLINK).
(../../gnatdll$(exeext)): Likewise.
(../../vxaddr2line$(exeext)): Likewise.
(gnatmake-re): Likewise.
(gnatlink-re): Likewise.

From-SVN: r254940

re PR target/82281 (Bulldozer/Zen tuning: uses XMM for single 64-bit integer AND, even with a simple mask)

PR target/82281
* gcc.target/i386/pr82281.c: New testcase.

From-SVN: r254939

tree-ssa-dom.c (record_equivalences_from_phis): Fix handling of degenerates resulting from ignoring an edge.

        * tree-ssa-dom.c (record_equivalences_from_phis): Fix handling
        of degenerates resulting from ignoring an edge.

From-SVN: r254938

re PR ipa/81360 (ice in estimate_edge_growth, at ipa-inline.h:86)

PR ipa/81360
* ipa-inline.c (can_inline_edge_p): Also check that caller is optimized
* gcc.c-torture/compile/pr81360.c: New testcase.

From-SVN: r254937

re PR fortran/78990 (ICE when assigning polymorphic array function result)

2017-11-19  Paul Thomas  <pault@gcc.gnu.org>

PR fortran/78990
* expr.c (gfc_is_class_array_function): Renamed from
'gfc_is_alloc_class_array_function' and modified to return true
for pointers as well as allocatable results.
* gfortran.h : Change of name for prototype of above function.
* trans-array.c (gfc_add_loop_ss_code): Force finalization of
class array results.
(build_class_array_ref): Change assertion into a condition.
(build_class_array_ref): Set the se class_vptr for class array
function results.
(gfc_walk_function_expr): Reference gfc_is_class_array_function
as above.
* trans-decl.c (get_proc_result): Move it up before
gfc_trans_deferred_vars.
(gfc_trans_deferred_vars): Nullify explicit return class arrays
on entry.
* trans-expr.c (gfc_conv_class_to_class): Allow conversion of
class array functions that have an se class_vptr and use it
for the result vptr.
(gfc_conv_subref_array_arg): Rename reference to the above
function.
(gfc_conv_procedure_call): Ditto. Add the se pre block to the
loop pre block before the function is evaluated. Do not
finalize class pointer results.
(arrayfunc_assign_needs_temporary, gfc_trans_assignment_1) More
renamed references.
* trans-intrinsic.c (gfc_conv_intrinsic_size): Ditto.

2017-11-19  Paul Thomas  <pault@gcc.gnu.org>

PR fortran/78990
* gfortran.dg/class_67.f90: New test.

From-SVN: r254936

re PR ipa/83001 (ICE in edge_badness, at ipa-inline.c:1025)

PR ipa/83001
* profile-count.c (profile_count::to_sreal_scale): Fix return value
for uninitialied counts.

From-SVN: r254935

re PR ipa/60243 (IPA is slow on large cgraph tree)

PR ipa/60243
* tree-inline.c (estimate_num_insns): Set to 1 at least.

From-SVN: r254934

re PR target/82713 (ICE in ix86_builtin_vectorization_cost, at config/i386/i386.c:44475)

PR target/82713
* i386.c (ix86_builtin_vectorization_cost): Be ready for insane
types.

From-SVN: r254933

re PR c/66618 (Failure to diagnose non-constant initializer for static object with -O1)

PR c/66618
PR c/69960
c-family/
* c-common.h (c_fully_fold): Add LVAL argument defaulted to false.
c/
* c-parser.c (c_parser_omp_atomic): Pass true as LVAL to c_fully_fold
where needed.
* c-typeck.c (build_unary_op, build_modify_expr, build_asm_expr,
handle_omp_array_sections): Likewise.
(digest_init): Don't call decl_constant_value_for_optimization.
* c-tree.h (decl_constant_value_for_optimization): Removed.
* c-fold.c (c_fold_array_ref): New function.
(c_fully_fold_internal): Add LVAL argument, propagate it through
recursive calls.  For VAR_P call decl_constant_value and
unshare if not LVAL and either optimizing or IN_INIT.  Remove
decl_constant_value_for_optimization calls.  If IN_INIT and not LVAL,
fold ARRAY_REF with STRING_CST and INTEGER_CST operands.
(c_fully_fold): Add LVAL argument, pass it through to
c_fully_fold_internal.
(decl_constant_value_for_optimization): Removed.
cp/
* cp-gimplify.c (c_fully_fold): Add LVAL argument, call
cp_fold_maybe_rvalue instead of cp_fold_rvalue and pass it !LVAL.
testsuite/
* gcc.dg/pr69960.c: New test.
* gcc.dg/pr66618.c: New test.
* gcc.dg/pr66618-2.c: New test.

From-SVN: r254930

[arc] Remove semicolon after do while (0) in FUNCTION_PROFILER

2017-11-19  Tom de Vries  <tom@codesourcery.com>

* config/arc/arc.h (FUNCTION_PROFILER): Remove semicolon after
"do while (0)".

From-SVN: r254929

[phoenix] Remove semicolon after do {} while (0) in TARGET_OS_CPP_BUILTINS

2017-11-19  Tom de Vries  <tom@codesourcery.com>

* config/phoenix.h (TARGET_OS_CPP_BUILTINS): Remove semicolon after
"do {} while (0)".

From-SVN: r254928

[visium] Remove semicolon after ASM_OUTPUT_CASE_END

2017-11-19  Tom de Vries  <tom@codesourcery.com>

* config/visium/visium.h (ASM_OUTPUT_CASE_END): Remove semicolon after
macro body.

From-SVN: r254927

[ft32, spu] Remove semicolon after do {} while (0) in REGISTER_TARGET_PRAGMAS

2017-11-19  Tom de Vries  <tom@codesourcery.com>

* config/ft32/ft32.h (REGISTER_TARGET_PRAGMAS): Remove semicolon after
"do {} while (0)".
* config/spu/spu.h (REGISTER_TARGET_PRAGMAS): Same.

From-SVN: r254926

[mcore] Remove semicolon after do {} while (0) in MCORE_EXPORT_NAME

2017-11-19  Tom de Vries  <tom@codesourcery.com>

* config/mcore/mcore-elf.h (MCORE_EXPORT_NAME): Remove semicolon after
"do {} while (0)".
* config/mcore/mcore.h (ASM_OUTPUT_ALIGNED_COMMON): After missing
semicolon after MCORE_EXPORT_NAME call.

From-SVN: r254925

PR libstdc++/pr66689 - comp_ellint_3 and ellint_3 return garbage values

2017-11-18  Edward Smith-Rowland  <3dw4rd@verizon.net>

PR libstdc++/pr66689 - comp_ellint_3 and ellint_3 return garbage values
* include/tr1/ell_integral.tcc: Correct the nu sign convention
in ellint_3 and comp_ellint_3.
* testsuite/tr1/5_numerical_facilities/special_functions/
06_comp_ellint_3/check_value.cc: Regen with correct values.
* testsuite/tr1/5_numerical_facilities/special_functions/
14_ellint_3/check_value.cc: Ditto.
* testsuite/special_functions/06_comp_ellint_3/check_value.cc: Ditto.
* testsuite/special_functions/13_ellint_3/check_value.cc: Ditto.
* testsuite/tr1/5_numerical_facilities/special_functions/
01_assoc_laguerre/check_value.cc: Regen.
* testsuite/tr1/5_numerical_facilities/special_functions/
02_assoc_legendre/check_value.cc: Regen.
* testsuite/tr1/5_numerical_facilities/special_functions/
03_beta/check_value.cc: Regen.
* testsuite/tr1/5_numerical_facilities/special_functions/
04_comp_ellint_1/check_value.cc: Regen.
* testsuite/tr1/5_numerical_facilities/special_functions/
05_comp_ellint_2/check_value.cc: Regen.
* testsuite/tr1/5_numerical_facilities/special_functions/
07_conf_hyperg/check_value.cc: Regen.
* testsuite/tr1/5_numerical_facilities/special_functions/
08_cyl_bessel_i/check_value.cc: Regen.
* testsuite/tr1/5_numerical_facilities/special_functions/
09_cyl_bessel_j/check_value.cc: Regen.
* testsuite/tr1/5_numerical_facilities/special_functions/
10_cyl_bessel_k/check_value.cc: Regen.
* testsuite/tr1/5_numerical_facilities/special_functions/
11_cyl_neumann/check_value.cc: Regen.
* testsuite/tr1/5_numerical_facilities/special_functions/
12_ellint_1/check_value.cc: Regen.
* testsuite/tr1/5_numerical_facilities/special_functions/
13_ellint_2/check_value.cc: Regen.
* testsuite/tr1/5_numerical_facilities/special_functions/
15_expint/check_value_neg.cc: Regen.
* testsuite/tr1/5_numerical_facilities/special_functions/
16_hermite/check_value.cc: Regen.
* testsuite/tr1/5_numerical_facilities/special_functions/
17_hyperg/check_value.cc: Regen.
* testsuite/tr1/5_numerical_facilities/special_functions/
18_laguerre/check_value.cc: Regen.
* testsuite/tr1/5_numerical_facilities/special_functions/
19_legendre/check_value.cc: Regen.
* testsuite/tr1/5_numerical_facilities/special_functions/
20_riemann_zeta/check_value_neg.cc: Regen.
* testsuite/tr1/5_numerical_facilities/special_functions/
21_sph_bessel/check_value.cc: Regen.
* testsuite/tr1/5_numerical_facilities/special_functions/
22_sph_legendre/check_value.cc: Regen.
* testsuite/tr1/5_numerical_facilities/special_functions/
23_sph_neumann/check_value.cc: Regen.
* testsuite/ext/special_functions/conf_hyperg/check_value.cc: Regen.
* testsuite/ext/special_functions/hyperg/check_value.cc: Regen.
* testsuite/special_functions/01_assoc_laguerre/check_value.cc: Regen.
* testsuite/special_functions/02_assoc_legendre/check_value.cc: Regen.
* testsuite/special_functions/03_beta/check_value.cc: Regen.
* testsuite/special_functions/04_comp_ellint_1/check_value.cc: Regen.
* testsuite/special_functions/05_comp_ellint_2/check_value.cc: Regen.
* testsuite/special_functions/07_cyl_bessel_i/check_value.cc: Regen.
* testsuite/special_functions/08_cyl_bessel_j/check_value.cc: Regen.
* testsuite/special_functions/09_cyl_bessel_k/check_value.cc: Regen.
  * testsuite/special_functions/10_cyl_neumann/check_value.cc: Regen.
* testsuite/special_functions/11_ellint_1/check_value.cc: Regen.
* testsuite/special_functions/12_ellint_2/check_value.cc: Regen.
* testsuite/special_functions/14_expint/check_value.cc: Regen.
* testsuite/special_functions/15_hermite/check_value.cc: Regen.
* testsuite/special_functions/16_laguerre/check_value.cc: Regen.
* testsuite/special_functions/17_legendre/check_value.cc: Regen.
* testsuite/special_functions/18_riemann_zeta/check_value.cc: Regen.
* testsuite/special_functions/19_sph_bessel/check_value.cc: Regen.
* testsuite/special_functions/20_sph_legendre/check_value.cc: Regen.
* testsuite/special_functions/21_sph_neumann/check_value.cc: Regen.
* testsuite/tr1/5_numerical_facilities/special_functions/
06_comp_ellint_3/pr66689.cc: New.
* testsuite/tr1/5_numerical_facilities/special_functions/
14_ellint_3/pr66689.cc: New.
* testsuite/special_functions/06_comp_ellint_3/pr66689.cc: New.
* testsuite/special_functions/13_ellint_3/pr66689.cc: New.

From-SVN: r254924

Daily bump.

From-SVN: r254923

[vms] Add missing vmsdbgout_early_finish

2017-11-19  Tom de Vries  <tom@codesourcery.com>

PR target/82961
* vmsdbgout.c (vmsdbgout_early_finish): New function.
(vmsdbg_debug_hooks): Set early_finish field to vmsdbgout_early_finish.

From-SVN: r254920

cgraphclones.c (cgraph_edge::clone): Rename gcov_count to prof_count.

* cgraphclones.c (cgraph_edge::clone): Rename gcov_count to prof_count.
(cgraph_edge::clone): Cleanup updating of profile.
* ipa-cp.c (update_profiling_info): Likewise.
* ipa-inline-transform.c (inline_transform): Likewise.
* ipa-inline.c (inline_small_functions): Add missing space to dump.
* ipa-split.c (execute_split_functions): Do not split when function
is cold.
* predict.c (estimate_bb_frequencies): Cleanup updating of profile.
* profile-count.c (profile_count::dump): Add global0.
(profile_count::to_cgraph_frequency): Do not ICE when entry is
undefined.
(profile_count::to_sreal_scale): Likewise.
(profile_count::adjust_for_ipa_scaling): Fix typo in comment.
(profile_count::combine_with_ipa_count): New function.
* profile-count.h (profile_guessed_global0adjusted): New.
(profile_count::adjusted_zero): New.
(profile_count::global0adjusted): New.
(profile_count::combine_with_ipa_count): New.
* tree-inline.c (copy_edges_for_bb): Add NUM/DEN arugment;
correct profile of return block of split functions.
(copy_cfg_body): Remove unused profile_count.
(copy_body): Likewise.
(expand_call_inline): Update.
(tree_function_versioning): Update.

From-SVN: r254919

PR 44292 Handle large record lengths

Now that the ABI supports large record lengths, there's a few places
in libgfortran where we need to use larger types. For internal units
which by definition are in-memory, it's enought to use ptrdiff_t, for
external units gfc_offset.

Regtested on x86_64-pc-linux-gnu?

libgfortran/ChangeLog:

2017-11-19  Janne Blomqvist  <jb@gcc.gnu.org>

        PR fortran/44292
* io/transfer.c (skip_record): Use gfc_offset to handle large
records.
(next_record_r): Likewise.
(sset): Likewise.
(next_record_w): Use gfc_offset/ptrdiff_t appropriately.

From-SVN: r254918

specfun.h: Expose airy_ai and airy_bi.

2017-11-18  Edward Smith-Rowland  <3dw4rd@verizon.net>

* include/bits/specfun.h: Expose airy_ai and airy_bi.
* include/tr1/modified_bessel_func.tcc: Treat NaN and inf arg, return.
* testsuite/ext/special_functions/airy_ai/check_nan.cc: New.
* testsuite/ext/special_functions/airy_ai/check_value.cc: New.
* testsuite/ext/special_functions/airy_ai/compile.cc: New.
* testsuite/ext/special_functions/airy_bi/check_nan.cc: New.
* testsuite/ext/special_functions/airy_bi/check_value.cc: New.
* testsuite/ext/special_functions/airy_bi/compile.cc: New.

From-SVN: r254917

PR 83036 Make NEXTREC specifier for INQUIRE work for large record numbers

This is accomplished by making the NEXTREC specifier be a 8 byte
integer where supported.

I wasn't able to come up with a testcase that does not create a large
file that could be added to the testsuite, but here's one which
creates a 2 GB file:

program nextrec
  implicit none
  integer(8) :: ii, n
  open(10, file="foo.dat", recl=1, access="direct", form="unformatted", &
       status="replace")
  do ii = 1, huge(1) + 2_8
     write(10, rec=ii) 'a'
  end do
  inquire(10, nextrec=n)
  if (n /= huge(1) + 3_8) then
     call abort()
  end if
  close(10, status="delete")
end program nextrec

Regtested on x86_64-pc-linux-gnu.

gcc/fortran/ChangeLog:

2017-11-18  Janne Blomqvist  <jb@gcc.gnu.org>

PR fortran/83036
* ioparm.def (IOPARM): Make nextrec a pintio.

libgfortran/ChangeLog:

2017-11-18  Janne Blomqvist  <jb@gcc.gnu.org>

PR fortran/83036
* io/io.h: Make nextrec a GFC_IO_INT*.
---
 gcc/fortran/ioparm.def | 2 +-
 libgfortran/io/io.h    | 3 +--
 2 files changed, 2 insertions(+), 3 deletions(-)

diff --git a/gcc/fortran/ioparm.def b/gcc/fortran/ioparm.def
index 5fc04bc..59cc7cd 100644
--- a/gcc/fortran/ioparm.def
+++ b/gcc/fortran/ioparm.def
@@ -62,7 +62,7 @@ IOPARM (inquire, exist, 1 << 7,  pint4)
 IOPARM (inquire, opened, 1 << 8,  pint4)
 IOPARM (inquire, number, 1 << 9,  pint4)
 IOPARM (inquire, named, 1 << 10, pint4)
-IOPARM (inquire, nextrec, 1 << 11, pint4)
+IOPARM (inquire, nextrec, 1 << 11, pintio)
 IOPARM (inquire, recl_out, 1 << 12, pintio)
 IOPARM (inquire, strm_pos_out, 1 << 13, pintio)
 IOPARM (inquire, file, 1 << 14, char1)
diff --git a/libgfortran/io/io.h b/libgfortran/io/io.h
index 3330bce..d29b112 100644
--- a/libgfortran/io/io.h
+++ b/libgfortran/io/io.h
@@ -388,8 +388,7 @@ typedef struct
 {
   st_parameter_common common;
   GFC_INTEGER_4 *exist, *opened, *number, *named;
-  GFC_INTEGER_4 *nextrec;
-  GFC_IO_INT  *recl_out, *strm_pos_out;
+  GFC_IO_INT *nextrec, *recl_out, *strm_pos_out;
   CHARACTER1 (file);
   CHARACTER2 (access);
   CHARACTER1 (form);
--
2.7.4

From-SVN: r254916

PR 44292 Enable large record lengths in OPEN and INQUIRE statements

This is a straightforward change that we can do now that the ABI has
been bumped (again!).

Regtested on x86_64-pc-linux-gnu.

gcc/fortran/ChangeLog:

2017-11-18  Janne Blomqvist  <jb@gcc.gnu.org>

PR fortran/44292
* ioparm.def (IOPARM): Make recl_in intio, and recl_out pintio.

libgfortran/ChangeLog:

2017-11-18  Janne Blomqvist  <jb@gcc.gnu.org>

PR fortran/44292
* io/io.h: Make recl_in a GC_IO_INT and recl_out a type
GFC_IO_INT*.

gcc/testsuite/ChangeLog:

2017-11-18  Janne Blomqvist  <jb@gcc.gnu.org>

PR fortran/44292
* gfortran.dg/large_recl.f90: New test.

From-SVN: r254915

re PR fortran/83012 (Simply contiguous pointer function not recognized as contiguous)

2017-11-18  Thomas Koenig  <tkoenig@gcc.gnu.org>

PR fortran/83012
* expr.c (gfc_is_simply_contiguous): If a function call through a
class variable is done through a reference, check the function's
interface.

2017-11-18  Thomas Koenig  <tkoenig@gcc.gnu.org>

PR fortran/83012
* gfortran.dg/contiguous_5.f90: New test.

From-SVN: r254914

hash-set.h (hash_set::empty): New.

* hash-set.h (hash_set::empty): New.
* tree-ssa-threadbackward.h: Delete.
* tree-ssa-threadbackward.c (class thread_jumps): New.
Move max_threaded_paths into class.
(fsm_find_thread_path): Remove arguments that are now in class.
(profitable_jump_thread_path): Rename to...
(thread_jumps::profitable_jump_thread_path): ...this.
(convert_and_register_jump_thread_path): Rename to...
(thread_jumps::convert_and_register_current_path): ...this.
(check_subpath_and_update_thread_path): Rename to...
(thread_jumps::check_subpath_and_update_thread_path): ...this.
(register_jump_thread_path_if_profitable): Rename to...
(thread_jumps::register_jump_thread_path_if_profitable): ...this.
(handle_phi): Rename to...
(thread_jumps::handle_phi): ...this.
(handle_assignment): Rename to...
(thread_jumps::handle_assignment): ...this.
(fsm_find_control_statement_thread_paths): Rename to...
(thread_jumps::fsm_find_control_statement_thread_paths): ...this.
(find_jump_threads_backwards): Rename to...
(thread_jumps::find_jump_threads_backwards): ...this.
Initialize path local data.
(pass_thread_jumps::execute): Call find_jump_threads_backwards
from within thread_jumps class.
(pass_early_thread_jumps::execute): Same.

From-SVN: r254913

Daily bump.

From-SVN: r254912

* cgraph.c (cgraph_node::dump): Do IPA sanity checking on IPA counts.

From-SVN: r254909

Enable building libitm with Intel CET

libitm/
* Makefile.in: Regenerate.
* acinclude.m4: Add enable.m4 and cet.m4.
* config/x86/sjlj.S: Include cet.h.
(_ITM_beginTransaction): Add _CET_ENDBR.
Save Shadow Stack pointer.
(GTM_longjmp): Add _CET_ENDBR. Restore Shadow Stack pointer.
* config/x86/target.h (struct gtm_jmpbuf):
Add new field for Shadow Stack pointer.
* configure: Regenerate.
* configure.ac: Set CET_FLAGS. Update XCFLAGS.
* configure.ac: Update libtool_VERSION for x86.
* testsuite/Makefile.in: Regenerate.

From-SVN: r254908

Add extra field to gtm_jmpbuf on x86 only

Expand the gtm_jmpbuf structure by one word field to add
Intel CET support further. The code in sjlj.S already
allocates more space on the stack then gtm_jmpbuf needs.
Use this extra space to absorb the new field.

The structure is allocated on the stack in such a way
that eip/rsp field is overlapped with return address on
the stack. Locate the new field right before eip/rsp so
code that accesses buffer fields relative to address of
gtm_jmpbuf has its offsets unchanged.

* libitm/config/x86/target.h: Add new field (ssp).
* libitm/config/x86/sjlj.S: Change offsets.

From-SVN: r254907

fmls.c: New test.

2017-11-17  Steve Ellcey  <sellcey@cavium.com>

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

From-SVN: r254906

aarch64-simd.md (fnma<mode>4): Move neg operator to canonical location.

2017-11-17  Steve Ellcey  <sellcey@cavium.com>

* config/aarch64/aarch64-simd.md (fnma<mode>4): Move neg operator
to canonical location.

From-SVN: r254905

Enable building libobjc with Intel CET

libobjc/
* Makefile.in: Regenerate.
* aclocal.m4: Likeiwse.
* configure: Likewise.
* configure.ac: Set CET_FLAGS. Update XCFLAGS.

From-SVN: r254904

pr81356.c: New test.

2017-11-17  Steve Ellcey  <sellcey@cavium.com>

* gcc.target/aarch64/pr81356.c: New test.

From-SVN: r254903

Enable building libgfortran with Intel CET

libgfortran/
* acinclude.m4: Add enable.m4, cet.m4.
* Makefile.in: Regenerate.
* configure: Likewise.
* configure.ac: Set CET_FLAGS. Update AM_FCFLAGS,
AM_CFLAGS, CFLAGS.

From-SVN: r254902

re PR target/81356 (__builtin_strcpy is not good for copying an empty string on aarch64)

2017-11-17  Steve Ellcey  <sellcey@cavium.com>

PR target/81356
* config/aarch64/aarch64.c (aarch64_use_by_pieces_infrastructure_p):
Remove.
(TARGET_USE_BY_PIECES_INFRASTRUCTURE_P): Remove define.

From-SVN: r254901

Enable building libmpx with Intel CET

libmpx/
* Makefile.in: Regenerate.
* acinclude.m4: Add enable.m4 and cet.m4.
* configure: Regenerate.
* configure.ac: Set CET_FLAGS. Update XCFLAGS.
* mpxrt/Makefile.am: Update libmpx_la_CFLAGS.
* mpxrt/Makefile.in: Regenerate.
* mpxwrap/Makefile.am: Add AM_CFLAGS. Update
libmpxwrappers_la_CFLAGS.
* mpxwrap/Makefile.in: Regenerate.

From-SVN: r254900

Enable building libquadmath with Intel CET

libquadmath/
* Makefile.am: Update AM_CFLAGS.
* Makefile.in: Regenerate:
* acinclude.m4: Add enable.m4 and cet.m4.
* configure: Regenerate.
* configure.ac: Set CET_FLAGS. Update XCFLAGS.

From-SVN: r254899

Enable building libssp with Intel CET

libssp/
* Makefile.am: Update AM_CFLAGS, update
libssp_nonshared_la_CFLAGS.
* Makefile.in: Regenerate.
* configure: Likewise.
* aclocal.m4: Likewise.
* configure.ac: Set CET_FLAGS. Update XCFLAGS.

From-SVN: r254898

Enable building libvtv with Intel CET

libvtv/
* acinclude.m4: Add enable.m4 and cet.m4.
* Makefile.in: Regenerate.
* testsuite/Makefile.in: Likewise.
* configure: Likewise.
* configure.ac: Set CET_FLAGS. Update XCFLAGS.
* testsuite/libvtv.cc/vtv.exp: Add scanlang.exp.

From-SVN: r254897

Enable building libsanitizer with Intel CET

libsanitizer/
* acinclude.m4: Add enable.m4 and cet.m4.
* Makefile.in: Regenerate.
* asan/Makefile.am: Update AM_CXXFLAGS.
* asan/Makefile.in: Regenerate.
* configure: Likewise.
* configure.ac: Set CET_FLAGS. Update EXTRA_CFLAGS,
EXTRA_CXXFLAGS, EXTRA_ASFLAGS.
* interception/Makefile.am: Update AM_CXXFLAGS.
* interception/Makefile.in: Regenerate.
* libbacktrace/Makefile.am: Update AM_CFLAGS, AM_CXXFLAGS.
* libbacktrace/Makefile.in: Regenerate.
* lsan/Makefile.am: Update AM_CXXFLAGS.
* lsan/Makefile.in: Regenerate.
* sanitizer_common/Makefile.am: Update AM_CXXFLAGS,
AM_CCASFLAGS.
* sanitizer_common/sanitizer_linux_x86_64.S: Include cet.h.
Add _CET_ENDBR macro.
* sanitizer_common/Makefile.in: Regenerate.
* tsan/Makefile.am: Update AM_CXXFLAGS.
* tsan/Makefile.in: Regenerate.
* tsan/tsan_rtl_amd64.S Include cet.h. Add _CET_ENDBR macro.
* ubsan/Makefile.am: Update AM_CXXFLAGS.
* ubsan/Makefile.in: Regenerate.

From-SVN: r254896

Enable building libstdc++-v3 with Intel CET

libstdc++-v3/
* acinclude.m4: Add cet.m4.
* configure.ac: Set CET_FLAGS. Update EXTRA_CFLAGS,
EXTRA_CXX_FLAGS.
* libsupc++/Makefile.am: Use Add EXTRA_CFLAGS.
* Makefile.in: Regenerate.
* configure: Likewise.
* doc/Makefile.in: Likewise.
* include/Makefile.in: Likewise.
* libsupc++/Makefile.in: Likewise.
* po/Makefile.in: Likewise.
* python/Makefile.in: Likewise.
* src/Makefile.in: Likewise.
* src/c++11/Makefile.in: Likewise.
* src/c++98/Makefile.in: Likewise.
* src/filesystem/Makefile.in: Likewise.
* testsuite/Makefile.in: Likewise.

From-SVN: r254895

Enable building libgomp with Intel CET

libgomp/
* configure.ac: Set CET_FLAGS, update XCFLAGS and FCFLAGS.
* acinclude.m4: Add cet.m4.
* configure: Regenerate.
* Makefile.in: Likewise.
* testsuite/Makefile.in: Likewise.

From-SVN: r254894

Enable building libatomic with Intel CET

libatomic/
* configure.ac: Set CET_FLAGS, update XCFLAGS.
* acinclude.m4: Add cet.m4 and enable.m4.
* configure: Regenerate.
* Makefile.in: Likewise.
* testsuite/Makefile.in: Likewise.

From-SVN: r254893

Enable building libbacktrace with Intel CET

libbacktrace/
* configure.ac: Add CET_FLAGS to EXTRA_FLAGS.
* aclocal.m4: Regenerate.
* Makefile.in: Likewise.
* configure: Likewise.

From-SVN: r254892

Use rcrt1.o%s/grcrt1.o%s to relocate static PIE

crt1.o is used to create dynamic and non-PIE static executables.  Static
PIE needs to link with rcrt1.o, instead of crt1.o, which is also used by
musl libc and OpenBSD:

https://gcc.gnu.org/ml/gcc/2015-06/msg00008.html

to relocate static PIE at run-time.  When -pg is used with -static-pie,
grcrt1.o should be used.

* config/gnu-user.h (GNU_USER_TARGET_STARTFILE_SPEC): Use
rcrt1.o%s/grcrt1.o%s for -static-pie.

From-SVN: r254890

i386.c (ix86_multiplication_cost, [...]): Break out from ...

* i386.c (ix86_multiplication_cost, ix86_division_cost,
ix86_shift_rotate_cost): Break out from ...
(ix86_rtx_costs): ... here.
(ix86_add_stmt_cost): Use rtx cost machinery to compute cost of
vector operations.

* gcc.target/i386/xop-imul64-vector.c: Disable vectorizer costmodel.

From-SVN: r254889

predict.c (determine_unlikely_bbs): Set cgraph node count to 0 when entry block was promoted unlikely.

* predict.c (determine_unlikely_bbs): Set cgraph node count to 0
when entry block was promoted unlikely.
(estimate_bb_frequencies): Increase frequency scale.
* profile-count.h (profile_count): Export precision info.
* gcc.dg/tree-ssa/dump-2.c: Fixup template for profile precision
changes.
* gcc.dg/tree-ssa/pr77445-2.c: Fixup template for profile precision
changes.

From-SVN: r254888

tree-tailcall.c (eliminate_tail_call): Be more careful about not disturbin profile of entry block.

* tree-tailcall.c (eliminate_tail_call): Be more careful about not
disturbin profile of entry block.

From-SVN: r254887

ipa-fnsummary.c (estimate_node_size_and_time): Be more tolerant for roundoff errors.

* ipa-fnsummary.c (estimate_node_size_and_time): Be more tolerant for
roundoff errors.

From-SVN: r254886

ipa-cp.c (update_profiling_info): Handle conversion to local profile.

* ipa-cp.c (update_profiling_info): Handle conversion to local
profile.
* tree-cfg.c (execute_fixup_cfg): Do fixup same way as inliner does.

From-SVN: r254885

gimple-ssa-evrp.c (class evrp_range_analyzer): New class extracted from evrp_dom_walker class.

* gimple-ssa-evrp.c (class evrp_range_analyzer): New class extracted
from evrp_dom_walker class.  Various methods moved into new class.
(evrp_range_analyzer::evrp_range_analyzer): Constructor for new class.
(evrp_range_analyzer::enter): New method.
(evrp_range_analyzer::leave): New method.
(evrp_dom_walker): Remove delegators no longer needed by this class.
Replace vr_values data member with evrp_range_analyzer

From-SVN: r254884

gimple-ssa-evrp.c (evrp_dom_walker::record_ranges_from_phis): New method extracted from evrp_dom_walker::before_dom_children.

* gimple-ssa-evrp.c (evrp_dom_walker::record_ranges_from_phis): New
method extracted from evrp_dom_walker::before_dom_children.
(evrp_dom_walker::record_ranges_from_stmt): Likewise.
(evrp_dom_walker::record_ranges_from_incoming_edge): Likewise.

From-SVN: r254883

gimple-ssa-evrp.c (evrp_dom_walker): Add cleanup method.

* gimple-ssa-evrp.c (evrp_dom_walker): Add cleanup method.
Add private copy constructor and move assignment operators.
Privatize methods and class data where trivially possible.
(evrp_dom_walker::cleanup): New function, extracted from
execute_early_vrp.  Simplify access to class data.

From-SVN: r254882

[PR c++/82836] Fixe testcase

https://gcc.gnu.org/ml/gcc-patches/2017-11/msg01515.html
* g++.dg/pr82836.C: Fix for c++17.

From-SVN: r254881

vr-values.h (get_output_for_vrp): Prototype.

* vr-values.h (get_output_for_vrp): Prototype.
* vr-values.c (get_output_for_vrp): New function extracted from
vrp_visit_assignment_or_call and extract_range_from_stmt.
(vrp_visit_assignment_or_call): Use get_output_for_vrp.  Simplify.

From-SVN: r254880

[AArch64] Adjust tuning parameters for Falkor

Disabling software prefetching and switching the autoprefetcher to weak improves
CPU2017 rate and speed benchmarks for both int and fp sets on Falkor.

SPECrate 2017 fp is up 0.38%
SPECspeed 2017 fp is up 0.54%
SPECrate 2017 int is up 3.02%
SPECspeed 2017 int is up 3.16%

There are only a couple individual regressions. The biggest one being about 4%
in parest.

For SPEC2006, we've noticed the following:

SPECint is up 0.91%
SPECfp is stable

In the case of SPEC2006 we noticed both a big regression in mcf (about 20%)
and a big improvement for hmmer (about 40%).

Since the overall result is positive, we would like to make these new tuning
settings the default for Falkor.

We may revisit the software prefetcher setting in the future, in case we
can adjust it enough so it provides us a good balance between improvements and
regressions (mcf). But for now it is best if it stays off.

2017-11-17  Luis Machado  <luis.machado@linaro.org>

gcc/
* config/aarch64/aarch64.c
(qdf24xx_prefetch_tune) <default_opt_level>: Set to -1.
(qdf24xx_tunings) <autoprefetcher_model>: Set to
tune_params::AUTOPREFETCHER_WEAK.

From-SVN: r254879

re PR target/82641 (Unable to enable crc32 for a certain function with target attribute on ARM (aarch32))

2017-11-17  Tamar Christina  <tamar.christina@arm.com>

PR target/82641
* config/arm/arm.c (arm_valid_target_attribute_rec):
Parse "arch=" and "+<ext>".
(arm_valid_target_attribute_tree): Re-init global options.
(arm_option_override): Make non-static.
(arm_options_perform_arch_sanity_checks): Make errors fatal.
* gcc/config/arm/arm-c.c (__ARM_FEATURE_CMSE): Support undef.
(__ARM_FEATURE_CRC32): Support undef.
* config/arm/arm_acle.h (__ARM_FEATURE_CRC32): Replace with pragma.
* doc/extend.texi (ARM Function Attributes): Add pragma and target.

gcc/testsuite/
2017-11-17  Tamar Christina  <tamar.christina@arm.com>

PR target/82641
* gcc.target/arm/pragma_arch_attribute.c: New.

From-SVN: r254878

gdbinit.in: add "break-on-diagnostic" command

gcc/ChangeLog:
* gdbinit.in (break-on-diagnostic): New command.

From-SVN: r254877

Add Intel CET support for EH in libgcc.

Control-flow Enforcement Technology (CET), published by Intel,
introduces the Shadow Stack feature, which ensures a return from a
function is done to exactly the same location from where the function
was called. When EH is present the control-flow transfer may skip some
stack frames and the shadow stack has to be adjusted not to signal a
violation of a control-flow transfer. It's done by counting a number
of skiping frames and adjasting shadow stack pointer by this number.

Having new semantic of the 'ret' instruction if CET is supported in HW
the 'ret' instruction cannot be generated in ix86_expand_epilogue when
we are returning after EH is processed. Added a code in
ix86_expand_epilogue to adjust Shadow Stack pointer and to generate an
indirect jump instead of 'ret'. As sp register is used during this
adjustment thus the argument in pro_epilogue_adjust_stack is changed
to update cfa_reg based on whether control-flow instrumentation is set.
Without updating the cfa_reg field there is an assert later in dwarf2
pass related to mismatch the stack register and cfa_reg value.

gcc/
* config/i386/i386.c (ix86_expand_epilogue): Change simple
return to indirect jump for EH return if control-flow protection
is enabled. Change explicit 'false' argument in
pro_epilogue_adjust_stack with a value of flag_cf_protection.
* config/i386/i386.md (simple_return_indirect_internal): Remove
SImode restriction to support 64-bit.

libgcc/
* config/i386/linux-unwind.h: Include
config/i386/shadow-stack-unwind.h.
* config/i386/shadow-stack-unwind.h: New file.
* unwind-dw2.c: (uw_install_context): Add a frame parameter and
pass it to _Unwind_Frames_Extra.
* unwind-generic.h (_Unwind_Frames_Extra): New.
* unwind.inc (_Unwind_RaiseException_Phase2): Add frames_p
parameter. Add local variable frames to count number of frames.
(_Unwind_ForcedUnwind_Phase2): Likewise.
(_Unwind_RaiseException): Add local variable frames to count
number of frames, pass it to _Unwind_RaiseException_Phase2 and
uw_install_context.
(_Unwind_ForcedUnwind): Likewise.
(_Unwind_Resume): Likewise.
(_Unwind_Resume_or_Rethrow): Likewise.

From-SVN: r254876

combine: Add added_notes_insn

This patch makes combine reconsider insns it added notes to.  This
matters for example if the note is a REG_DEAD; without the note the
setter of the register has to be kept around in the result of
combinations, so it cannot be a 2->1 combination, and the cost of
the result is higher than without that extra set, so try_combine may
refuse the combination with the set, but allow it without the set.

This fixes a regression for powerpc: pr69946.c has started to fail
after the bitfield expansion changes.  GCC used to generate

        lwz 3,0(9)
        rlwinm 3,3,12,20,23
        ori 3,3,0x11
        rotldi 3,3,52
        bl bar

but now it does

        lwz 3,0(9)
        rldicr 3,3,32,3
        srdi 3,3,48
        ori 3,3,0x110
        sldi 3,3,48
        bl bar

(an instruction too many).  After this patch it is

        lwz 3,0(9)
        rlwinm 3,3,16,16,19
        ori 3,3,0x110
        sldi 3,3,48
        bl bar

(the testcase still does not pass, it looks for very specific insns).

* combine.c (added_notes_insn): New.
(try_combine): Handle added_notes_insn like added_links_insn.
Rewrite return value code.
(distribute_notes): Set added_notes_insn to the earliest insn we added
a note to.

From-SVN: r254875