re PR libfortran/69799 (FAIL: gfortran.dg/coarray_allocate_3.f08 at -O1 and above)

PR libfortran/69799
* gfortran.dg/coarray_allocate_5.f08: Add "-latomic" option if
libatomic_available.

From-SVN: r234240

Fix compiling large files

  * line-map.c (new_linemap): Make alloc_size a size_t.

From-SVN: r234239

Fix thinko in indirect_jump_optimize

PR rtl-optimization/69195
PR rtl-optimization/47992
* ira.c (indirect_jump_optimize): Ignore artificial defs.
Add comments.

From-SVN: r234235

re PR c++/70209 (ICE in strip_typedefs, at cp/tree.c:1377)

PR c++/70209
* tree.c (strip_typedefs): Call strip_typedefs again on the
DECL_ORIGINAL_TYPE result.

* g++.dg/ext/attribute-may-alias-4.C: New test.

From-SVN: r234234

re PR c++/70095 ([C++14] Link error on partially specialized variable template)

PR c++/70095
* pt.c (instantiate_decl): Fix call to variable_template_p.

From-SVN: r234231

Fix more P0184R0 references.

From-SVN: r234230

re PR c++/70141 (template parameter not deducible in partial specialization of template inside template)

PR c++/70141
* pt.c (for_each_template_parm_r): Always walk into TYPENAME_TYPE.

From-SVN: r234228

pr63384.C: Add -w to dg-options.

* g++.dg/pr63384.C: Add -w to dg-options.  Remove '-toggle'
in -fvar-tracking-assignments-toggle flag.

From-SVN: r234227

re PR middle-end/69513 (LTO bootstrap fails with bootstrap-profiled during linking gnat1 in stagefeedback)

PR bootstrap/69513
* dwarf2out.c (flush_limbo_die_list): Really flush the limbo list.

From-SVN: r234226

ifcvt-4.c: Add -misel for powerpc*.

* gcc.dg/ifcvt-4.c: Add -misel for powerpc*.

From-SVN: r234224

avr.md (rotl<mode>3): Set mode for operand 2.

* config/avr/avr.md (rotl<mode>3): Set mode for operand 2.

From-SVN: r234223

re PR rtl-optimization/70222 (Test miscompiled with -O1)

PR rtl-optimization/70222
* combine.c (simplify_shift_const_1): For A >> B >> C LSHIFTRT
optimization if mode is different from result_mode, queue up masking
of the result in outer_op.  Formatting fix.

* gcc.c-torture/execute/pr70222-1.c: New test.
* gcc.c-torture/execute/pr70222-2.c: New test.

From-SVN: r234222

re PR middle-end/70239 (ICE in ggc_set_mark)

PR middle-end/70239
* tree-ssa-sccvn.c (VN_INFO_GET): Use safe_grow_cleared instead
of safe_grow.

From-SVN: r234221

Fix pr68714 test for targets not supporting V4SImode.

  * gcc.dg/tree-ssa/pr68714.c: Test during reassoc pass.

From-SVN: r234220

re PR rtl-optimization/69032 (ICE: in cfg_preds_1, at sel-sched-ir.c:4809 with -fsched-pressure -fsel-sched-pipelining -fselective-scheduling)

gcc/

PR rtl-optimization/69032
* sel-sched-ir.c (get_seqno_by_preds): Include both insn and head when
looping backwards over basic block insns.

testsuite/

PR rtl-optimization/69032
* gcc.dg/pr69032.c: New test.

From-SVN: r234219

re PR target/66660 ([ia64] Speculative load not checked before use, leading to a NaT Consumption Vector interruption)

        PR target/66660
        * sel-sched-ir.c (merge_expr): Avoid changing the speculative pattern
        to non-speculative when propagating trap bits.

From-SVN: r234218

re PR rtl-optimization/63384 (scheduler loops on endless fence list with -fselective-scheduling2 on x86)

gcc/

PR rtl-optimization/63384
* sel-sched.c (invoke_aftermath_hooks): Do not decrease issue_more
on DEBUG_INSN_P insns.

testsuite/

PR rtl-optimization/63384
* g++.dg/pr63384.C: New test.

From-SVN: r234217

re PR target/64411 (ICE: in verify_target_availability, at sel-sched.c:1577 with -Os -mcmodel=medium -fPIC -fschedule-insns -fselective-scheduling)

gcc/

        PR target/64411
        * sched-deps.c (get_implicit_reg_pending_clobbers): New function,
        factored out from ...
        (sched_analyze_insn): ... here.
        * sched-int.h (get_implicit_reg_pending_clobbers): Declare it.
        * sel-sched-ir.c (setup_id_implicit_regs): New function, use
        get_implicit_reg_pending_clobbers in it.
        (setup_id_reg_sets): Use setup_id_implicit_regs.
        (deps_init_id): Ditto.

testsuite/

PR target/64411
* gcc.target/i386/pr64411.C: New test.

From-SVN: r234216

Add goacc/kernels-alias{,-2}.f95

2016-03-15  Tom de Vries  <tom@codesourcery.com>

* gfortran.dg/goacc/kernels-alias-2.f95: New test.
* gfortran.dg/goacc/kernels-alias.f95: New test.

From-SVN: r234214

expr.c (cpp_classify_number): Hex floats are new in C++1z.

* libcpp/expr.c (cpp_classify_number): Hex floats are new in C++1z.

* libcpp/init.c (lang_defaults): Likewise.
* gcc/c-family/c-cppbuiltin.c (c_cpp_builtins): Set __cpp_hex_float.

From-SVN: r234213

feat-cxx1z.C: Fix __cpp_fold_expressions value.

* g++.dg/cpp1z/feat-cxx1z.C: Fix __cpp_fold_expressions value.
* g++.dg/cpp1z/fold7.C: Likewise.

From-SVN: r234212

Fix fdump-ipa-all-graph

2016-03-15  Tom de Vries  <tom@codesourcery.com>

PR ipa/70161
* cgraph.c (cgraph_node::get_body): Save, reset and restore
dump_file_name.
* passes.c (execute_one_ipa_transform_pass): Add missing argument to
execute_function_dump.
(execute_one_pass): Don't dump function if it will be dumped after ipa
transform.

From-SVN: r234211

PR c++/53792 - [C++11] improving compiler-time constexpr evaluation

gcc/testsuite/ChangeLog:
2016-03-14  Martin Sebor  <msebor@redhat.com>

PR c++/53792
* g++.dg/cpp0x/constexpr-inline.C: New test.
* g++.dg/cpp0x/constexpr-inline-1.C: Same.

From-SVN: r234208

* c-cppbuiltin.c (c_cpp_builtins): Bump __cpp_range_based_for.

From-SVN: r234207

genrecog: Fix crash on invalid input

If your machine description refers to a non-existent predicate genrecog
crashes.  This fixes it.

* genrecog.c (match_pattern_2): If pred is NULL don't call
safe_predicate_mode on it.

From-SVN: r234206

Daily bump.

From-SVN: r234205

* gcc.dg/torture/pr70083.c: Prune non-standard ABI.

From-SVN: r234202

re PR middle-end/70219 (ICE: in delete_move_and_clobber, at lra-constraints.c:5864 with -O2)

PR middle-end/70219
* lra-constraints.c (delete_move_and_clobber): Change assertion
to also allow dregno == 0.

* gcc.dg/pr70219.c: New test.

From-SVN: r234201

re PR tree-optimization/68714 (less folding of vector comparison)

PR tree-opt/68714

  * tree-ssa-reassoc.c (ovce_extract_ops, optimize_vec_cond_expr): New.
  (can_reassociate_p): Allow ANY_INTEGRAL_TYPE_P.
  (reassociate_bb): Use optimize_vec_cond_expr; avoid
  optimize_range_tests, attempt_builtin_copysign and attempt_builtin_powi
  on vectors.

From-SVN: r234196

P08184R0: Generalizing the Range-Based For Loop

* parser.c (cp_convert_range_for): Set the type of __end separately.
(cp_parser_perform_range_for_lookup): Allow different begin/end
types if they are comparable.

Co-Authored-By: Jason Merrill <jason@redhat.com>
From-SVN: r234191

LRA: identify biggest access mode for hard_regs and use it in split_reg

PR target/70083
* lra-lives.c (process_bb_lives): Also update biggest mode for hard
regs.
(lra_create_live_ranges_1): initialize hard register biggest_mode to
VOIDmode.
* lra-constraints.c (split_reg): For hard regs, try to find the
biggest single-register mode used in the function.

testsuite/
PR target/70083
* gcc.dg/torture/pr70083.c: New test.
* gcc.target/i386/pr70083.c: New test.

From-SVN: r234184

re PR tree-optimization/56365 (Missed opportunities for smin/smax standard name patterns when compiling as C++)

2016-03-14  Richard Biener  <rguenther@suse.de>

PR tree-optimization/56365
* tree-ssa-phiopt.c (minmax_replacement): Handle alternate
constants to compare against.

* gcc.dg/tree-ssa/phi-opt-14.c: New testcase.

From-SVN: r234183

rs6000: Handle "d" output in the bd*z patterns (PR70098)

In the rs6000 port, FLOAT_REGS can contain DImode values when compiling
for 64-bit targets.  Some instructions (like "fcfid" in the testcase,
convert from integer to DP float) only work on floating point registers.
So, we do want to allow DImode in these regs.

Now, in unusual cases IRA will assign FLOAT_REGS to some allocno where
some insns cannot handle FLOAT_REGS there, so they will need a reload.
Maybe IRA can be made smarter, but it isn't doing anything wrong here,
so we should be able to handle it.

The place it goes wrong is in the output of the *ctrdi_internal[1256]
pattern: the "bdz" and "bdnz" instructions.  GCC refuses to do output
reloads on JUMP_INSNs, probably because it is hard to do, needs different
strategies than "normal" reloads do, and it cannot even be done at all
for general patterns.  So JUMP_INSNs need to be able to handle every
possible output for the register class used.

These patterns already handle writing to "c" (the base insn case), and
to "r", "m", and "c" or "l"; all those via splitters.  We just need to
handle "d" as well.  That is what this patch does.  [A predicate in one
of the splitters needs to be touched up so that the correct splitter
is used for the FLOAT_REGS case.]

But, that leaves another problem.  One of the insns that are split to
is a move from a GPR to an FPR.  That work fine on targets with direct
move (which does exactly that), i.e. power8 and up.  But older targets
need memory to do the move, and this splitter runs after reload so
it cannot allocate memory; and allocating memory beforehand for every
bdnz insn is pretty horrible as well.

This patch implements the easy part.  With it, power8 works, where it
didn't before.

PR target/70098
* config/rs6000/rs6000.md (*ctr<mode>_internal1, *ctr<mode>_internal2,
*ctr<mode>_internal5, *ctr<mode>_internal6): Also allow "d" as output.
(define_split for the GPR case): Use int_reg_operand instead of
gpc_reg_operand for the output.

gcc/testsuite/
PR target/70098
* lib/target-supports.exp (check_effective_target_powerpc64_no_dm):
New function.
* g++.dg/pr70098.C: New testcase.

From-SVN: r234181

Unshare create_empty_if_region_on_edge argument

2016-03-14  Tom de Vries  <tom@codesourcery.com>

PR tree-optimization/70045
* graphite-isl-ast-to-gimple.c (graphite_create_new_loop_guard): Unshare
create_empty_if_region_on_edge argument.

* gcc.dg/graphite/pr70045.c: New test.

From-SVN: r234177

Summary: libcc1: rerun configure when gcc/BASE-VER changes

* configure.ac (CONFIG_STATUS_DEPENDENCIES): Substitute.
* configure: Regenerate.
* Makefile.in: Regenerate.

From-SVN: r234176

Daily bump.

From-SVN: r234175

vxworks.h (STACK_CHECK_STATIC_BUILTIN): Define.

* config/arm/vxworks.h (STACK_CHECK_STATIC_BUILTIN): Define.
(STACK_CHECK_PROTECT): Likewise.
* config/i386/vxworks.h (STACK_CHECK_STATIC_BUILTIN): Likewise
(STACK_CHECK_PROTECT): Likewise.
* config/rs6000/vxworks.h (STACK_CHECK_STATIC_BUILTIN): Likewise
(STACK_CHECK_PROTECT): Likewise.
* config/rs6000/vxworksae.h (STACK_CHECK_PROTECT): Likewise.
* config/sparc/vxworks.h (STACK_CHECK_STATIC_BUILTIN): Likewise.
(STACK_CHECK_PROTECT): Likewise.
ada/
* system-vxworks-m68k.ads (Stack_Check_Probes): Set to True.
(Stack_Check_Limits): Set to False.
* system-vxworks-mips.ads (Stack_Check_Probes): Set to True.
(Stack_Check_Limits): Set to False.
* system-vxworks-ppc.ads (Stack_Check_Probes): Set to True.
(Stack_Check_Limits): Set to False.
* system-vxworks-sparcv9.ads (Stack_Check_Probes): Set to True.
(Stack_Check_Limits): Set to False.
* system-vxworks-x86.ads (Stack_Check_Probes): Set to True.
(Stack_Check_Limits): Set to False.

From-SVN: r234171

libgcc_s is required by libstdc++

libgomp/
* testsuite/lib/libgomp.exp (libgomp_init): Potentially append to
always_ld_library_path the path to libgcc_s.

From-SVN: r234170

re PR fortran/69043 (Trying to include a directory causes an infinite loop)

2016-03-13  Jerry DeLisle  <jvdelisle@gcc.gnu.org>
    Jim MacArthur  <jim.macarthur@codethink.co.uk>

PR fortran/69043
* scanner.c (load_file): Check that included file is regular.

PR fortran/69043
* gfortran.dg/include_9.f90: New test.

Co-Authored-By: Jim MacArthur <jim.macarthur@codethink.co.uk>
From-SVN: r234169

re PR fortran/45076 ([OOP] gfortran.dg/dynamic_dispatch_6.f03 ICEs with -fprofile-use)

2016-03-13  Dominique d'Humieres  <dominiq@lps.ens.fr>

PR fortran/45076
gfortran.dg/prof/prof.exp: New script.
gfortran.dg/prof/dynamic_dispatch_6.f03: New test.

From-SVN: r234168

re PR fortran/69520 (Implement reversal of -fcheck options)

2016-03-12  Jerry DeLisle  <jvdelisle@gcc.gnu.org>
    Harold Anlauf  <anlauf@gmx.de>

PR fortran/69520
* invoke.texi: Explain use of the 'no-' construct within the
-fcheck= option.
* options.c (gfc_handle_runtime_check_option): Enable use of
'no-' prefix for the various options with -fcheck= to allow
negating previously enabled check options.

Co-Authored-By: Harald Anlauf <anlauf@gmx.de>
From-SVN: r234167

Daily bump.

From-SVN: r234166

re PR rtl-optimization/69307 (wrong code with -O2 -fselective-scheduling @ armv7a)

PR rtl-optimization/69307
* sel-sched.c (choose_best_pseudo_reg): Properly check for hard
registers in modes that span more than one register.

PR rtl-optimization/69307
* gcc.dg/pr69307.c: New test.

From-SVN: r234163

re PR target/69614 (wrong code with -Os -fno-expensive-optimizations -fschedule-insns -mtpcs-leaf-frame -fira-algorithm=priority @ armv7a)

2016-03-12  Vladimir Makarov  <vmakarov@redhat.com>

PR target/69614
* lra-constraints.c (delete_move_and_clobber): New.
(remove_inheritance_pseudos): Use it.

2016-03-12  Vladimir Makarov  <vmakarov@redhat.com>

PR target/69614
* gcc.target/arm/pr69614.c: New.

From-SVN: r234162

re PR fortran/70031 (Error in recursive module subroutine declaration if declared as "module recursive")

2016-03-12  Paul Thomas  <pault@gcc.gnu.org>

PR fortran/70031
* decl.c (gfc_match_prefix): Treat the 'module' prefix in the
same way as the others, rather than fixing it to come last.
(gfc_match_function_decl, gfc_match_subroutine): After errors
in 'copy_prefix', emit them immediately in the case of module
procedures to prevent a later ICE.

PR fortran/69524
* decl.c (gfc_match_submod_proc): Permit 'module procedure'
declarations within the contains section of modules as well as
submodules.
* resolve.c (resolve_fl_procedure): Likewise.
*trans-decl.c (build_function_decl): Change the gcc_assert to
allow all forms of module procedure declarations within module
contains sections.

2016-03-12  Paul Thomas  <pault@gcc.gnu.org>

PR fortran/70031
* gfortran.dg/submodule_14.f08: New test

PR fortran/69524
* gfortran.dg/submodule_15.f08: New test

From-SVN: r234161

re PR ada/70017 (c52103x and c52104x test failure on s390x)

PR ada/70017
* calls.c (emit_library_call_value_1): Clear the ECF_NOTHROW flag if
the libcall is LCT_THROW.
* explow.c (probe_stack_range): Pass LCT_THROW to emit_library_call
for the checking routine.

From-SVN: r234160

Fix PR c++/70106 (type of parenthesized qualified-id has wrong cv-qualifiers)

gcc/cp/ChangeLog:

PR c++/70106
* semantics.c (force_paren_expr): Just build a PAREN_EXPR when
processing_template_decl and EXPR is a SCOPE_REF.

gcc/testsuite/ChangeLog:

PR c++/70106
* g++.dg/cpp1y/paren3.C: New test.

From-SVN: r234159

Daily bump.

From-SVN: r234158

re PR target/70131 (PowerPC ISA 2.07 is inefficient at doint (float)(int)x.)

[gcc]
2016-03-11  Michael Meissner  <meissner@linux.vnet.ibm.com>

PR target/70131
* config/rs6000/rs6000.md (round32<mode>2_fprs): Do not do the
optimization if we have direct move.
(roundu32<mode>2_fprs): Likewise.

[gcc/testsuite]
2016-03-11  Michael Meissner  <meissner@linux.vnet.ibm.com>

PR target/70131
* gcc.target/powerpc/ppc-round2.c: New test.

From-SVN: r234155

* sv.po: Update.

From-SVN: r234153

re PR target/70123 (Miscompilation of cfitsio testcase on s390x-linux starting with r222144)

PR target/70123
* lra-remat.c (operand_to_remat): Disallow hard regs in the value t
be rematerialized.
(reg_overlap_for_remat_p): Renamed from input_regno_present_p.
Arguments swapped.  All callers changed.  Take reg_renumber into
account, and Calculate and compare register ranges for hard regs.

PR target/70123
* gcc.dg/torture/pr70123.c: New test.

From-SVN: r234152

re PR tree-optimization/70190 (ICE (segfault) at -O2 and -O3 on x86_64-linux-gnu)

PR tree-optimization/70190
* tree-ssa-threadbackward.c (fsm_find_control_statement_thread_paths):
Handle cases where we can not extract the taken edge, even though we
found a constant value.

PR tree-optimization/70190
* gcc.c-torture/compile/pr70190.c: New test.

From-SVN: r234151

re PR tree-optimization/64058 (Performance degradation after r216304)

PR tree-optimization/64058
* tree-ssa-coalesce.c (struct coalesce_pair): Add new field INDEX.
(num_coalesce_pairs): Move up earlier in file.
(find_coalesce_pair): Initialize the INDEX field for each pair
discovered.
(compare_pairs): No longer sort on the elements in each pair.
Instead break ties with the index of the coalesce pair.

From-SVN: r234149

Wmisleading-indentation.c: add more test cases for PR c/68187

gcc/testsuite/ChangeLog:
PR c/68187
* c-c++-common/Wmisleading-indentation.c (test43_a): New test
case.
(test43_b): Likewise.
(test43_c): Likewise.
(test43_d): Likewise.
(test43_e): Likewise.
(test43_f): Likewise.
(test43_g): Likewise.
(test44_a): Likewise.
(test44_b): Likewise.
(test44_c): Likewise.
(test44_d): Likewise.
(test44_e): Likewise.

From-SVN: r234146

Wmisleading-indentation: add reproducer for PR c/70085

PR c/70085 reported a false-positive from -Wmisleading-indentation.

The warning was fixed by the fix for PR c/68187 (r233972), but it seems
worth capturing the reproducer for PR c/70085 as an additional test case,
as it's slightly different to those seen in PR c/68187.

gcc/testsuite/ChangeLog:
PR c/70085
* c-c++-common/Wmisleading-indentation.c (pr70085): New test case.

From-SVN: r234145

[AArch64] Fix gcc.target/aarch64/vect-reduc-or_1.c for -mcpu=cortex-a57

* gcc.target/aarch64/vect-reduc-or_1.c: Add -fno-vect-cost-model to
dg-options.

From-SVN: r234142

[AArch64] PR target/70002: Make aarch64_set_current_function play nice with pragma resetting

PR target/70002
* config/aarch64/aarch64-protos.h
(aarch64_save_restore_target_globals): New prototype.
* config/aarch64/aarch64-c.c (aarch64_pragma_target_parse):
Call the above when popping pragma.
* config/aarch64/aarch64.c (aarch64_save_restore_target_globals):
New function.
(aarch64_set_current_function): Rewrite using the above.

PR target/70002
PR target/69245
* gcc.target/aarch64/pr69245_2.c: New test.

From-SVN: r234141

re PR tree-optimization/70177 (ICE in extract_ops_from_tree starting with r233660)

PR tree-optimization/70177
* gimple-expr.h (extract_ops_from_tree_1): Renamed to ...
(extract_ops_from_tree): ... this.  In the 2 argument
overload remove _1 suffix.
* gimple-expr.c (extract_ops_from_tree_1): Renamed to ...
(extract_ops_from_tree): ... this.
* gimple.c (gimple_build_assign, gimple_assign_set_rhs_from_tree):
Adjust callers.
* tree-ssa-loop-niter.c (derive_constant_upper_bound): Likewise.
* tree-ssa-forwprop.c (defcodefor_name): Call 3 operand
extract_ops_from_tree instead of 2 operand one.

* gcc.dg/pr70177.c: New test.

From-SVN: r234140

Re: Add -funconstrained-commons to work around PR/69368 (and others) in SPEC2006

gcc/testsuite:

* gfortran.dg/unconstrained_commons.f: Widen regexp to match j_<N>.

From-SVN: r234139

Fix PR/70013

gcc:

PR tree-optimization/70013
* tree-sra.c (analyze_access_subtree): Also set grp_unscalarized_data
for constant-pool entries.

gcc/testsuite:

* gcc.dg/tree-ssa/sra-20.c: New.

From-SVN: r234138

re PR rtl-optimization/70174 (ICE at -O1 and above on x86_64-linux-gnu in gen_lowpart_general, at rtlhooks.c:63)

PR rtl-optimization/70174
* expmed.c (store_bit_field_using_insv): Use gen_lowpart_if_possible
followed by gen_lowpart on force_reg instead of just gen_lowpart.

* gcc.dg/pr70174.c: New test.

From-SVN: r234137

re PR tree-optimization/70169 (ICE at -O1 and above on x86_64-linux-gnu in gen_lsm_tmp_name, at tree-ssa-loop.c:791)

PR tree-optimization/70169
* tree-ssa-loop.c (gen_lsm_tmp_name): Handle FUNCTION_DECL and
LABEL_DECL like VAR_DECL.  Emit nothing instead of gcc_unreachable
for unknown codes.

* gcc.dg/pr70169.c: New test.

From-SVN: r234136

re PR rtl-optimization/70160 (gcc ICE at -O2 (seg fault) and above on valid code on x86_64-linux-gnu)

gcc/

PR target/70160
* config/i386/i386.c (scalar_chain::convert_reg): Skip uses
of uninitialized values.

gcc/testsuite/

PR target/70160
* gcc.target/i386/pr70160.c: New test.

Co-Authored-By: Jakub Jelinek <jakub@redhat.com>
From-SVN: r234135

S/390: Fix rounding for _Decimal128 to _Decimal32 conversion

We do not have a direct conversion instruction from 128 bit DFP to 32
bit DFP so this needs to be done in two steps.  The first needs to be
done with the "prepare for shorter precision rounding mode" in order
to produce a correct result.

gcc/ChangeLog:

2016-03-11  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>

* config/s390/s390.md ("trunctddd2"): Turn former define_insn into
define_expand.
("*trunctddd2"): New pattern definition.
("trunctdsd2"): Set prep_for_short_prec rounding mode for the
TD->DD truncation.

gcc/testsuite/ChangeLog:

2016-03-11  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>

* gcc.target/s390/dfp-1.c: New test.

From-SVN: r234134

S/390: Define macros for rounding mode constants

This patch replaces a few magic numbers used for floating point
rounding modes with macros.  This is mostly a NoOp change apart from:

fixuns_truncdddi2, fixuns_trunctddi2, fixuns_trunc<mode>si2: Replace 5
with DFP_RND_TOWARD_0 (which is 9).

5 as well as 9 represent round towards 0 with the difference that for
5 the new DFP quantum exception is enabled as well.  This exception
isn't IEEE754 and we do not have an interface to enable and test it
anyway.  So we do not intend to enable it.  This so far should not
have any noticable effect since the quantum exception was not
observable through the Posix functions.

Some pattern ("fix_trunc<mode>di2") is already using rounding mode 9
correctly.

gcc/ChangeLog:

2016-03-11  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>

* config/s390/s390.md (BFP_RND_*, DFP_RND_*): Add new constant
definitions for BFP and DFP rounding modes.
("fixuns_truncdddi2", "fixuns_trunctddi2")
("fixuns_trunc<BFP:mode><GPR:mode>2", "fixuns_trunc<mode>si2")
("fix_trunc<DSF:mode><GPR:mode>2", "fix_trunc<mode>di2")
("fix_trunctf<mode>2"): Use the new constants instead of magic
numbers.

From-SVN: r234133

S/390: Rename shift_count_or_setmem_operand to setmem_operand

The shift_count_or_setmem_operand predicate is now only used for
setmem patterns anymore.  Rename it together with the related
functions.

gcc/ChangeLog:

2016-03-11  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>

* config/s390/constraints.md: Adjust comment.
("Y"): Adjust comment.  Rename s390_decompose_shift_count to
s390_decompose_addrstyle_without_index.
* config/s390/predicates.md (shift_count_or_setmem_operand):
Rename to setmem_operand.
* config/s390/s390-protos.h
(s390_decompose_shift_count): Rename to
s390_decompose_addrstyle_without_index.
* config/s390/s390.c (s390_decompose_shift_count)
(s390_mem_constraint, print_shift_count_operand)
(print_operand_address, print_operand): Rename
s390_decompose_shift_count to
s390_decompose_addrstyle_without_index and rename
print_shift_count_operand to print_addrstyle_operand troughout the
file.
* config/s390/s390.md ("setmem_long_<P:mode>", "*setmem_long")
("*setmem_long_and", "*setmem_long_31z", "*setmem_long_and_31z"):
Rename shift_count_or_setmem_operand to setmem_operand.
* config/s390/vx-builtins.md ("vec_insert<mode>")
("vec_promote<mode>"): Replace shift_count_or_setmem_operand with
nonmemory_operand.

From-SVN: r234132

Daily bump.

From-SVN: r234130

re PR target/70168 (Wrong code generation in __sync_val_compare_and_swap on PowerPC)

PR target/70168
* config/rs6000/rs6000.c (rs6000_expand_atomic_compare_and_swap):
Handle overlapping retval and newval.

From-SVN: r234126

re PR testsuite/70009 (test case libgomp.oacc-c-c++-common/vprop.c fails starting with its introduction in r233607)

libgomp/
PR testsuite/70009
* testsuite/libgomp.oacc-c-c++-common/vprop.c: Make test data signed.

From-SVN: r234124

* g++.dg/ubsan/pr70135.C: Don't rely on C++17 static_assert.

From-SVN: r234120

re PR target/7044 (vax casesi breakage)

PR target/7044
* config/aarch64/aarch64.c
(aarch64_override_options_after_change_1): When forcing
flag_omit_frame_pointer to be true, use a special value that can
be detected if this function is called again, thus preventing
flag_omit_leaf_frame_pointer from being forced to be false.

* gcc.target/aarch64/pr70044.c: New test.

From-SVN: r234118

re PR c++/70001 (Infinity compilation time)

PR c++/70001
* constexpr.c (cxx_eval_vec_init_1): For pre_init case, reuse
return value from cxx_eval_constant_expression from earlier
elements if it is valid constant initializer requiring no
relocations.

* g++.dg/cpp0x/constexpr-70001-1.C: New test.
* g++.dg/cpp0x/constexpr-70001-2.C: New test.
* g++.dg/cpp0x/constexpr-70001-3.C: New test.

Co-Authored-By: Jakub Jelinek <jakub@redhat.com>
From-SVN: r234117

[AArch64] Fix thinko in handling of -momit-leaf-frame-pointer option

* common/config/aarch64/aarch64-common.c (aarch64_handle_option):
Set x_flag_omit_leaf_frame_pointer when handling
-momit-leaf-frame-pointer.

From-SVN: r234116

re PR ipa/69589 (ICE in initialize_node_lattices, at ipa-cp.c:971)

PR lto/69589
* cgraph.c (cgraph_node::dump): Dump split_part and indirect_call_target.
* cgraph.h (cgraph_node): Add indirect_call_target flag.
* ipa.c (has_addr_references_p): Cleanup.
(is_indirect_call_target_p): New.
(walk_polymorphic_call_targets): Do not mark virtuals that may be
called indirectly as local.
(symbol_table::remove_unreachable_nodes): Compute indirect_call_target.

* g++.dg/lto/pr69589_0.C: New testcase
* g++.dg/lto/pr69589_1.C: New testcase

From-SVN: r234115

re PR lto/69630 (LTO ICE in types_same_for_odr at ipa-devirt.c:402)

PR ipa/69630
* ipa-devirt.c (possible_polymorphic_call_targets): Do not ICE
on cxa_pure_virtual.

From-SVN: r234114

re PR ipa/69589 (ICE in initialize_node_lattices, at ipa-cp.c:971)

PR lto/69589
* tree.c (free_lang_data_in_decl): Clear visibility of TYPE_DECL.

From-SVN: r234113

re PR ipa/69589 (ICE in initialize_node_lattices, at ipa-cp.c:971)

PR lto/69589
* tree.c (need_assembler_name_p): Only record main variant type names.

From-SVN: r234112

re PR c++/70153 (ICE on valid C++ code)

PR c++/70153
* cp-gimplify.c (cp_fold): Handle UNARY_PLUS_EXPR.

* g++.dg/delayedfold/unary-plus1.C: New test.

From-SVN: r234111

pr45701-1.c: Escape brackets.

2016-03-10 Andre Vieira <andre.simoesdiasvieira@arm.com>

    gcc/testsuite/
    * gcc.target/arm/pr45701-1.c: Escape brackets.
    * gcc.target/arm/pr45701-2.c: Likewise.

From-SVN: r234109

re PR target/70113 ([AArch64] -mpc-relative-literal-loads conflicts with fix-cortex-a53-843419)

2016-03-10  Christophe Lyon  <christophe.lyon@linaro.org>

PR target/70113.
* config/aarch64/aarch64.h (TARGET_FIX_ERR_A53_843419_DEFAULT):
Always define to 0 or 1.
(TARGET_FIX_ERR_A53_843419): New macro.
* config/aarch64/aarch64-elf-raw.h
(TARGET_FIX_ERR_A53_843419_DEFAULT): Update for above changes.
* config/aarch64/aarch64-linux.h: Likewise.
* config/aarch64/aarch64.c
(aarch64_override_options_after_change_1): Do not default
aarch64_nopcrelative_literal_loads to true if Cortex-A53 erratum
843419 is on.
(aarch64_attributes): Handle fix-cortex-a53-843419.
(aarch64_can_inline_p): Likewise.
* config/aarch64/aarch64.opt (aarch64_fix_a53_err843419): Save.

From-SVN: r234108

Add -funconstrained-commons to work around PR/69368 (and others) in SPEC2006

gcc/:

        * common.opt (funconstrained-commons, flag_unconstrained_commons): New.
        * tree.c (array_at_struct_end_p): Do not limit to size of decl for
        DECL_COMMONS if flag_unconstrained_commons is set.
        * tree-dfa.c (get_ref_base_and_extent): Likewise.
        * doc/invoke.texi (Optimize Options): Add -funconstrained-commons.
        (funconstrained-commons): Document.

gcc/testsuite:

* gfortran.dg/unconstrained_commons.f: New.

Co-Authored-By: Jakub Jelinek <jakub@redhat.com>
From-SVN: r234106

t-aarch64 (OPTIONS_H_EXTRA): Add aarch64-fusion-pairs.def and aarch64-tuning-flags.def

2016-03-10  Christophe Lyon  <christophe.lyon@linaro.org>

* config/aarch64/t-aarch64 (OPTIONS_H_EXTRA): Add
aarch64-fusion-pairs.def and aarch64-tuning-flags.def

From-SVN: r234105

tree-vect-stmts.c (vectorizable_mask_load_store): Check mask has a proper number of elements.

gcc/

* tree-vect-stmts.c (vectorizable_mask_load_store): Check mask
has a proper number of elements.

From-SVN: r234104

PR69195, Reload confused by invalid reg_equiv

Optimizing indirect jumps to direct jumps, and deleting dead insns can
lead to changes in register lifetimes, which in turn can result in bad
reg_equiv info being passed to reload.  So do these tasks before
calculating reg_equiv info.

gcc/
PR rtl-optimization/69195
PR rtl-optimization/47992
* ira.c (recorded_label_ref): Delete.
(update_equiv_regs): Return void.
(indirect_jump_optimize): New function.
(ira): Call indirect_jump_optimize and delete_trivially_dead_insns
before regstat_compute_ri.  Don't rebuild_jump_labels here.
Delete update_regstat.
gcc/testsuite/
* gcc.dg/pr69195.c: New.
* gcc.dg/pr69238.c: New.

From-SVN: r234101

Xfail pr46032.c scan if alignment requirements not met

2016-03-10  Tom de Vries  <tom@codesourcery.com>

PR testsuite/68915
* gcc.dg/vect/pr46032.c: Xfail scan if alignment requirements not met.

From-SVN: r234100

re PR tree-optimization/70128 (Linux kernel div patching optimized away)

2016-03-10  Richard Biener  <rguenther@suse.de>

PR tree-optimization/70128
* tree-ssa-structalias.c (set_uids_in_ptset): Set
vars_contains_nonlocal for any FUNCTION_DECL or LABEL_DECL.

* gcc.dg/tree-ssa/alias-34.c: New testcase.
* gcc.dg/tree-ssa/alias-35.c: Likewise.

From-SVN: r234099

Daily bump.

From-SVN: r234097

re PR tree-optimization/70152 (gcc ICE at -O3 and above on valid code on x86_64-linux-gnu in "replace_uses_by")

PR tree-optimization/70152
* tree-sra.c (replace_removed_params_ssa_names): Copy over
SSA_NAME_OCCURS_IN_ABNORMAL_PHI from old_name to new_name.

* gcc.dg/pr70152.c: New test.

From-SVN: r234092

re PR target/70086 (ICE: in extract_constrain_insn_cached, at recog.c:2202 (insn does not satisfy its constraints) with -mavx512vl -ffloat-store)

PR target/70086
* config/i386/i386.md (truncdfsf2 splitter): Use gen_vec_concatv2df
instead of gen_sse2_loadlpd.
* config/i386/sse.md (*vec_concatv2df): Rename to...
(vec_concatv2df): ... this.

* gcc.target/i386/pr70086-1.c: New test.
* gcc.target/i386/pr70086-2.c: New test.
* gcc.target/i386/pr70086-3.c: New test.

From-SVN: r234091

re PR middle-end/70127 (wrong code on x86_64-linux-gnu at -O3 in 32-bit and 64-bit modes)

PR tree-optimization/70127
* fold-const.c (operand_equal_p): Revert the 2015-10-28 change.

* gcc.c-torture/execute/pr70127.c: New test.

From-SVN: r234090

c-parser.c (c_parser_oacc_loop): Update cclauses and clauses when calling c_finish_omp_clauses.

gcc/c/
* c-parser.c (c_parser_oacc_loop): Update cclauses and clauses
when calling c_finish_omp_clauses.

gcc/cp/
* parser.c (cp_parser_oacc_loop): Update cclauses and clauses
when calling c_finish_omp_clauses.

gcc/testsuite/
* c-c++-common/goacc/combined-directives-2.c: New test.

From-SVN: r234089

PR c++/70105: prevent nonsensical underline spew for macro expansions

diagnostic_show_locus can sometimes do the wrong thing when handling
expressions built up from macros.

PR c++/70105 (currently marked as a P3 regression) has an example of
a diagnostic where over 500 lines of irrelevant source are printed,
and underlined, giving >1000 lines of useless spew to stderr.

This patch adds extra sanitization to diagnostic-show-locus.c, so that
we only attempt to print underlines and secondary locations if such
locations are "sufficiently sane" relative to the primary location
of a diagnostic.

This "sufficiently sane" condition is implemented by a new helper
function compatible_locations_p, which requires such locations to
have the same macro expansion hierarchy as the primary location,
using linemap_macro_map_loc_unwind_toward_spelling, effectively
mimicing the expansion performed by LRK_SPELLING_LOCATION.

This may be too strong a condition, but it effectively fixes
PR c++/70105, without removing any underlines in my testing.

Successfully bootstrapped&regrtested in combination with the previous
patch on x86_64-pc-linux-gnu; adds 15 new PASS results to g++.sum
and 4 new PASS results to gcc.sum.

gcc/ChangeLog:
PR c/68473
PR c++/70105
* diagnostic-show-locus.c (compatible_locations_p): New function.
(layout::layout): Sanitize ranges using compatible_locations_p.

gcc/testsuite/ChangeLog:
PR c/68473
PR c++/70105
* g++.dg/diagnostic/pr70105.C: New test.
* gcc.dg/plugin/diagnostic-test-expressions-1.c (foo): New decl.
(test_multiple_ordinary_maps): New test function.

libcpp/ChangeLog:
PR c/68473
PR c++/70105
* line-map.c (linemap_macro_map_loc_unwind_toward_spelling): Move
decl...
* include/line-map.h
(linemap_macro_map_loc_unwind_toward_spelling): ...here,
converting from static to extern.

From-SVN: r234088

PR c++/70105: Defer location expansion until diagnostic_show_locus

gcc/ChangeLog:
PR c/68473
PR c++/70105
* diagnostic-show-locus.c (layout_range::layout_range): Replace
location_range param with three const expanded_locations * and a
bool.
(layout::layout): Replace call to
rich_location::lazily_expand_location with get_expanded_location.
Extract the range and perform location expansion here, passing
the results to the layout_range ctor.
* diagnostic.c (source_range::debug): Delete.
* diagnostic.h (diagnostic_expand_location): Reimplement in terms
of rich_location::get_expanded_location.
* gcc-rich-location.c (get_range_for_expr): Delete.
(gcc_rich_location::add_expr): Reimplement to avoid the
rich_location::add_range overload that took a location_range,
passing a location_t instead.

gcc/testsuite/ChangeLog:
PR c/68473
PR c++/70105
* gcc.dg/plugin/diagnostic_plugin_show_trees.c (show_tree):
Drop range information from call to inform_at_rich_loc.
* gcc.dg/plugin/diagnostic_plugin_test_show_locus.c (add_range):
New.
(test_show_locus): Replace calls to rich_location::add_range with
calls to add_range.  Rewrite the tests that used the now-defunct
rich_location ctor taking a source_range.  Simplify other tests
by replacing calls to COMBINE_LOCATION_DATA with calls to
make_location.

libcpp/ChangeLog:
PR c/68473
PR c++/70105
* include/line-map.h (source_range::debug): Delete.
(struct location_range): Update comment.  Replace
expanded_location fields "m_start", "m_finish", and "m_caret" with
a source_location field: "m_loc".
(class rich_location): Reword comment.
(rich_location::get_loc): Reimplement in terms of a new overloaded
variant which takes an unsigned int.
(rich_location::get_loc_addr): Delete.
(rich_location::add_range): Drop params "start" and "finish" in
favor of param "loc".  Drop overloaded variants taking a
source_range or location_range *.
(rich_location::lazily_expand_location): Delete in favor of...
(rich_location::get_expanded_location): New decl.
(rich_location::m_loc): Delete field.
(rich_location::m_column_override): New field.
* line-map.c (rich_location::rich_location):  Drop name of
line_maps * param.  Update initializations for deletion of field
"m_loc" and addition of field "m_column_override".  Reimplement
body as a call to add_range.  Delete overloaded variant taking a
source_range.
(rich_location::get_loc): New function.
(rich_location::lazily_expand_location): Delete in favor of...
(rich_location::get_expanded_location): New function.
(rich_location::override_column): Reimplement.
(rich_location::add_range): Drop params "start" and "finish" in
favor of param "loc".  Eliminate location expansion in favor of
simply storing loc.  Drop overloaded variants taking a
source_range or location_range *.
(rich_location::set_range): Eliminate location expansion.

From-SVN: r234087

re PR tree-optimization/70138 (wrong code at -O3 on x86_64-linux-gnu)

2016-03-09  Richard Biener  <rguenther@suse.de>
Jakub Jelinek  <jakub@redhat.com>

PR tree-optimization/70138
* tree-vect-loop-manip.c (vect_update_ivs_after_vectorizer):
Also skip vect_double_reduction_def.

* gcc.dg/vect/pr70138-1.c: New testcase.
* gcc.dg/vect/pr70138-2.c: Likewise.

Co-Authored-By: Jakub Jelinek <jakub@redhat.com>
From-SVN: r234086

re PR target/70049 (Error: operand size mismatch for `vpextrw' (wrong assembly generated) with -masm=intel)

PR target/70049
* config/i386/sse.md (*vec_extract<mode>): Use %0 instead of %k0
if the operand is "m".

* gcc.target/i386/pr70049.c: New test.

From-SVN: r234085

re PR c/70143 (false strict-aliasing warning)

2016-03-09  Richard Biener  <rguenther@suse.de>

c-family/
PR c/70143
* c-common.c (strict_aliasing_warning): Add back
alias_sets_conflict_p check.

* gcc.dg/Wstrict-aliasing-bogus-upcast.c: New testcase.
* gcc.dg/Wstrict-aliasing-struct-with-char-member.c: Likewise.
* gcc.dg/Wstrict-aliasing-struct-member.c: Remove again.

From-SVN: r234084

* config/nvptx/nvptx.c (nvptx_option_override): Don't kill debug level.

From-SVN: r234083

Really don't xfail gcc.dg/vect/bb-slp-34.c on aarch64, arm

* gcc.dg/vect/bb-slp-34.c: Really don't xfail on aarch64-*-*,
arm-*-*.

From-SVN: r234081

re PR target/67278 (ICE: verify_gimple failed on darwin 14.5 x86_64)

PR middle-end/67278
* gcc.dg/simd-7.c: Add -w -Wno-psabi to dg-options.

From-SVN: r234080

re PR target/67278 (ICE: verify_gimple failed on darwin 14.5 x86_64)

PR target/67278
* gcc.dg/simd-8.c: Add -w -Wno-psabi to dg-options.

PR middle-end/70050
* gcc.dg/pr70050.c: Add -w to dg-options.  Fix up PR number
in testcase comment.

From-SVN: r234079