Add gcc.dg/vect/pr62171.c

2015-09-28  Tom de Vries  <tom@codesourcery.com>

* gcc.dg/vect/pr62171.c: New test.

From-SVN: r228193

update a few places for the change from gimple_statement_base to gimple

gcc/ChangeLog:

2015-09-28  Trevor Saunders  <tbsaunde+gcc@tbsaunde.org>

* doc/gimple.texi: Update references to gimple_statement_base.
* gdbhooks.py: Likewise.
* gimple.h: Likewise.

From-SVN: r228192

Use leon3 target for native LEON on Linux.

2015-09-28  Daniel Cederman  <cederman@gaisler.com>

Use leon3 target for native LEON on Linux. Linux requires LEON version 3 or
above with CASA support.

gcc/
* config/sparc/driver-sparc.c: map LEON to leon3

From-SVN: r228185

Make muser-mode the default for LEON3

2015-09-28  Daniel Cederman  <cederman@gaisler.com>

Make muser-mode the default for LEON3

The muser-mode flag causes the CASA instruction for LEON3 to use the
user mode ASI. This is the correct behavior for almost all LEON3 targets.
For this reason it makes sense to make user mode the default.

gcc/
* config/sparc/sparc.opt: Rename mask from USER_MODE to SV_MODE
  and make it inverse to change default
* config/sparc/sync.md: Only use supervisor ASI for CASA when in
  supervisor mode
* doc/invoke.texi: Document change of default

From-SVN: r228184

Do not use floating point registers when compiling with -msoft-float for SPARC

2015-09-28  Daniel Cederman  <cederman@gaisler.com>

Do not use floating point registers when compiling with -msoft-float for SPARC

__builtin_apply* and __builtin_return accesses the floating point registers on
SPARC even when compiling with -msoft-float.

gcc/
* config/sparc/sparc.c (sparc_function_value_regno_p): Do not return
true on %f0 for a target without FPU.
* config/sparc/sparc.md (untyped_call): Do not save %f0 for a target
without FPU.
(untyped_return): Do not load %f0 for a target without FPU.

From-SVN: r228183

aarch64.md (prefetch): Change the predicate of operand 0 to register_operand.

2015-09-28  Andrew Pinski  <apinski@cavium.com>

* config/aarch64/aarch64.md (prefetch):
Change the predicate of operand 0 to register_operand.

From-SVN: r228182

Daily bump.

From-SVN: r228181

predicates.md (register_sse4nonimm_operand): New predicate.

* config/i386/predicates.md (register_sse4nonimm_operand): New
predicate.
* config/i386/sse.md (PEXTR_MODE12): New mode iterator.
(*vec_extract<mode>): Use PEXTR_MODE12 instead of VI12_128 mode.
Use register_sse4nonimm_operand as operand 0 predicate.
(*vec_extractv8hi_sse2): Remove insn pattern.
(*vec_extract<PEXTR_MODE12:mode>_zext): Merge insn pattern from
*vec_extractv8hi_zext and *vec_extractv16qi_zext patterns.

From-SVN: r228178

pr44641.C: Revert line number change.

        * g++.dg/debug/dwarf2/pr44641.C: Revert line number change. Remove
        skip on AIX.  XFAIL individual line tests.

From-SVN: r228177

re PR target/67391 ([SH] Convert clrt addc to normal add insn)

gcc/
PR target/67391
* config/sh/sh-protos.h (sh_lra_p): Declare.
* config/sh/sh.c (sh_lra_p): Make non-static.
* config/sh/sh.md (addsi3): Use arith_reg_dest for operands[0] and
arith_reg_operand for operands[1].  Remove TARGET_SHMEDIA case.
Expand into addsi3_scr if operands[2] if needed.
(*addsi3_compact): Rename to *addsi3_compact_lra.  Use
arith_reg_operand for operands[1].  Allow it only when LRA is enabled.
(addsi3_scr, *addsi3): New insn_and_split patterns.

Co-Authored-By: Kaz Kojima <kkojima@gcc.gnu.org>
From-SVN: r228176

revert to assign_parms assignments using default defs

Revert the fragile and complicated changes to assign_parms designed to
enable it to use RTL assigments chosen by cfgexpand, and instead have
cfgexpand use the RTL assignments by assign_parms, keying them off of
the default defs that are now necessarily introduced for each parm and
result.  The possible lack of a default def was already a problem, and
the fallbacks in place were not enough, as shown by PR67312.  We now
have checking asserts in set_rtl that verify that we're assigning to
each var a piece of RTL that matches the expectations set forth by
use_register_for_decl.

for  gcc/ChangeLog

PR rtl-optimization/64164
PR tree-optimization/67312
PR middle-end/67340
PR middle-end/67490
PR bootstrap/67597
* cfgexpand.c (parm_in_stack_slot_p): Remove.
(ssa_default_def_partition): Remove.
(get_rtl_for_parm_ssa_default_def): Remove.
(set_rtl): Check that RTL assignments match expectations.
Loop on SUBREGs, CONCATs and PARALLELs subexprs.  Set only the
default def location for params and results.  Record SSA names
or types in REG and MEM attrs, respectively.
(set_parm_rtl): New.
(expand_one_ssa_partition): Drop logic that assigned MEMs with
unassigned addresses.
(adjust_one_expanded_partition_var): Don't accept NULL RTL on
deferred stack alloc vars.
(expand_used_vars): Skip partitions holding parm default defs.
Move adjust_one_expanded_partition_var loop...
(pass_expand::execute): ... here.  Drop redundant assert.
Adjust comments before the final loop over all ssa names.
Require assigned rtl of parms and results to match exactly.
Reset its attributes to match them, not any other variables in
the same partition.
(expand_debug_expr): Use entry value for PARM's default defs
only iff they have zero nondebug uses.
* cfgexpand.h (parm_in_stack_slot_p): Remove.
(get_rtl_for_parm_ssa_default_def): Remove.
(set_parm_rtl): Declare.
* doc/invoke.texi: Improve wording.
* explow.c (promote_decl_mode): Fix promote_function_mode for
result decls not by reference.
(promote_ssa_mode): Disregard BLKmode from promote_decl, and
bypass TYPE_MODE to get the actual vector mode.
* function.c: Include tree-dfa.h.  Revert 2015-08-14's and
2015-08-19's changes as follows.  Drop include of
basic-block.h and df.h.
(rtl_for_parm): Remove.
(maybe_reset_rtl_for_parm): Remove.
(parm_in_unassigned_mem_p): Remove.
(use_register_for_decl): Add logic for RESULT_DECLs matching
assign_parms' behavior.
(split_complex_args): Revert.
(assign_parms_augmented_arg_list): Revert.  Add comment
referencing the logic above.
(assign_parm_adjust_stack_rtl): Revert.
(assign_parm_setup_block): Revert.  Use set_parm_rtl instead
of SET_DECL_RTL.  Set up a REG if the parm demands so.
(assign_parm_setup_reg): Revert.  Consolidated SET_DECL_RTL
calls into a single set_parm_rtl.  Set up a temporary RTL
temporarily for expand_assignment.
(assign_parm_setup_stack): Revert.  Use set_parm_rtl.
(assign_parms_unsplit_complex): Revert.  Use set_parm_rtl.
(assign_bounds): Revert.
(assign_parms): Revert.  Use set_parm_rtl.
(allocate_struct_function): Relayout result and parms of
non-abstruct functions.
(expand_function_start): Revert.  Use set_parm_rtl.  If the
result is not a hard reg, create a pseudo from the promoted
mode of the default def.  Promote static chain mode.
* tree-outof-ssa.c (remove_ssa_form): Drop unused
partition_has_default_def.  Set up
partitions_for_parm_default_defs.
(finish_out_of_ssa): Remove partition_has_default_def.
Release partitions_for_parm_default_defs.
* tree-outof-ssa.h (struct ssaexpand): Remove
partition_has_default_def.  Add
partitions_for_parm_default_defs.
* tree-ssa-coalesce.c: Include tree-dfa.h, tm_p.h and
stor-layout.h.
(build_ssa_conflict_graph): Fix conflict-detection of default
defs of even unused default defs of params and results.
(for_all_parms): New.
(create_default_def): New.
(register_default_def): New.
(coalesce_with_default): New.
(create_outofssa_var_map): Create default defs for all parms
and results, and register their partitions.  Add GIMPLE_RETURN
operands as coalesce candidates with results.  Add default
defs of each parm or result as coalesce candidates with its
other defs.  Mark each result def, and each default def of
parms, as used_in_copy.
(gimple_can_coalesce_p): Call it.  Call use_register_for_decl
with the ssa names, even anonymous ones.  Drop
parm_in_stack_slot_p calls.  Require same signedness and
alignment.
(coalesce_ssa_name): Add coalesce candidates for all defs of
each parm and result, even unused ones.
(parm_default_def_partition_arg): New type.
(set_parm_default_def_partition): New.
(get_parm_default_def_partitions): New.
* tree-ssa-coalesce.h (get_parm_default_def_partitions): New.
* tree-ssa-live.c (partition_view_init): Regard unused defs of
parms and results as used.
(verify_live_on_entry): Don't error out just because they're
not live.

for  gcc/testsuite/ChangeLog

PR rtl-optimization/64164
PR tree-optimization/67312
* gcc.dg/pr67312.c: New.  From Zdenek Sojka.
* gcc.target/i386/stackalign/return-4.c: Add -O.

From-SVN: r228175

* g++.dg/debug/dwarf2/pr44641.C: Add one to expected line numbers.

From-SVN: r228174

Daily bump.

From-SVN: r228173

Fix missing deep copy when assigning a DT constructor to an array

This adds the missing deep copy when assiging a constructor of a derived
type with allocatable components to an array.

The check for constantness is removed so that the deep_copy argument passed
to gfc_trans_scalar_assign is set to true.

PR fortran/67721
gcc/fortran/
* trans-expr.c (gfc_trans_assignment_1): Remove the non-constantness
condition guarding deep copy.
gcc/testsuite/
* gfortran.dg/alloc_comp_deep_copy_3.f03: New.

From-SVN: r228170

re PR fortran/67567 (resolve.c: gfc_error called with iface->module == NULL)

2013-09-26  Paul Thomas  <pault@gcc.gnu.org>

PR fortran/67567
* resolve.c (resolve_fl_procedure): For module procedures, take
the parent module name and the submodule name from the name of
the namespace.

From-SVN: r228169

pr64935-1.c: XFAIL on AIX.

        * gcc.dg/pr64935-1.c: XFAIL on AIX.
        * gcc.dg/pr64935-2.c: XFAIL on AIX.
        * gcc.dg/debug/dwarf2/dwarf2-macro2.c: XFAIL Start new file on AIX.
        * gcc.dg/debug/dwarf2/dwarf2-macro2.c: XFAIL At line number on AIX.
        * gcc.dg/debug/dwarf2/global-used-types.c: XFAIL ascii on AIX.
        * gcc.dg/debug/dwarf2/noreturn-function-keyword.c: XFAIL
        DW_AT_noreturn on AIX.
        * gcc.dg/debug/dwarf2/noreturn-function-attribute.c: XFAIL
        DW_AT_noreturn on AIX.
        * gcc.dg/debug/dwarf2/lang-c89.c: XFAIL DW_AT_language on AIX.
        * gcc.dg/debug/dwarf2/stacked-qualified-types-3.c: XFAIL DW_TAG on AIX.
        * g++.dg/debug/dwarf2/deleted-member-function.C: XFAIL
        DW_AT_GNU_deleted on AIX.
        * g++.dg/debug/dwarf2/noreturn-function.C: XFAIL DW_AT_noreturn on AIX.
        * g++.dg/debug/dwarf2/template-params-6.C: XFAIL ascii on AIX.
        * g++.dg/debug/dwarf2/dwarf4-nested.C: XFAIL ascii on AIX.
        * g++.dg/debug/dwarf2/pubnames-2.C: Skip on AIX.
        * g++.dg/debug/dwarf2/pr44641.C: Skip on AIX.
        * g++.dg/debug/dwarf2/pr46527.C: XFAIL on AIX.
        * g++.dg/debug/dwarf2/imported-decl-2.C: XFAIL ascii on AIX.
        * g++.dg/debug/dwarf2/namespace-1.C: XFAIL ascii on AIX.
        * g++.dg/debug/dwarf2/global-used-types-1.C: XFAIL ascii on AIX.
        * g++.dg/debug/dwarf2/lambda1.C: XFAIL ascii on AIX.
        * g++.dg/debug/dwarf2/pubnames-3.C: Skip on AIX.
        * g++.dg/debug/dwarf2/pr61433.C: XFAIL on AIX.
        * g++.dg/debug/dwarf2/nested-4.C: XFAIL on AIX.
        * g++.dg/debug/dwarf2/template-func-params-6.C: XFAIL ascii on AIX.
        * g++.dg/debug/dwarf2/lineno-simple1.C: XFAIL on AIX.

From-SVN: r228168

dwarf2out.c (XCOFF_DEBUGGING_INFO): Default 0 definition.

        * dwarf2out.c (XCOFF_DEBUGGING_INFO): Default 0 definition.
        (HAVE_XCOFF_DWARF_EXTRAS): Default to 0 definition.
        (output_fde): Don't output length for debug_frame on AIX.
        (output_call_frame_info): Don't output length for debug_frame on AIX.
        (have_macinfo): Force to False for XCOFF_DEBUGGING_INFO and not
        HAVE_XCOFF_DWARF_EXTRAS.
        (add_AT_loc_list): Return early if XCOFF_DEBUGGING_INFO and not
        HAVE_XCOFF_DWARF_EXTRAS.
        (output_compilation_unit_header): Don't output length on AIX.
        (output_pubnames): Don't output length on AIX.
        (output_aranges): Delete argument. Compute length locally. Don't
        output length on AIX.
        (output_line_info): Don't output length on AIX.
        (dwarf2out_finish): Don't compute aranges_length.
        * dwarf2asm.c (XCOFF_DEBUGGING_INFO): Default 0 definition.
        (dw2_asm_output_nstring): Emit .byte not .ascii on AIX.
        * config/rs6000/rs6000.c (rs6000_output_dwarf_dtprel): Emit correct
        symbol decoration for AIX.
        (rs6000_xcoff_debug_unwind_info): New.
        (rs6000_xcoff_asm_named_section): Emit .dwsect pseudo-op
        for SECTION_DEBUG.
        (rs6000_xcoff_declare_function_name): Emit different
        .function pseudo-op when DWARF2_DEBUG. Don't call
        xcoffout_declare_function for DWARF2_DEBUG.
        * config/rs6000/xcoff.h (TARGET_DEBUG_UNWIND_INFO):
        Redefine.
        * config/rs6000/aix71.h: New.
        * configure.ac (gcc_cv_as_aix_dwloc): Check AIX as for DWARF
        locations support.
        * configure: Regenerate.
        * config.gcc (powerpc-ibm-aix[789]+): New stanza for AIX 7.1+ with
        DWARF support.

From-SVN: r228167

[PATCH] Fix undefined behaviour in arc port

* config/arc/arc.c (arc_output_addsi): Fix left shift undefined
behaviour.
* config/arc/constraints.md (Cca, C2a): Fix left shift undefined
behaviour.

From-SVN: r228166

[PATCH] Fix undefined behaviour in SH port

    [PATCH] Fix undefined behaviour in SH port
* config/sh/sh.h (CONST_OK_FOR_J16): Fix left shift undefined
behaviour

From-SVN: r228165

[PATCH] Fix undefined behaviour in mips port

[PATCH] Fix undefined behaviour in mips port
* config/mips/mips.c (mips_compute_frame_info): Fix left shift
undefined behaviour.

From-SVN: r228164

[PATCH] Fix undefined behaviour in cris port

[PATCH] Fix undefined behaviour in cris port
* config/cris/cris.md (asrandb): Fix left shift undefined
behaviour.
(asrandw): Likewise.

From-SVN: r228163

Daily bump.

From-SVN: r228161

re PR fortran/67614 (ICE on using arithmetic if with null)

2015-09-25  Steven G. Kargl  <kargl@gcc.gnu.org>

PR fortran/67614
* resolve.c (gfc_resolve_code): Prevent ICE for invalid EXPR_NULL.

2015-09-25  Steven G. Kargl  <kargl@gcc.gnu.org>

PR fortran/67614
* gfortran.dg/pr67614.f90: New test.

From-SVN: r228156

re PR fortran/67525 (ICE on select type with improper selector)

2015-09-25  Steven G. Kargl  <kargl@gcc.gnu.org>

PR fortran/67525
* parse.c (match_deferred_characteristics): Remove an assert, which
allows an invalid SELECT TYPE selector to be detected.

2015-09-25  Steven G. Kargl  <kargl@gcc.gnu.org>

PR fortran/67525
* gfortran.dg/pr67525.f90: New test.

From-SVN: r228155

Fix type in the changelog entry,

From-SVN: r228154

re PR target/61578 (Code size increase for ARM thumb compared to 4.8.x when compiling with -Os)

2015-09-25  Vladimir Makarov  <vmakarov@redhat.com>

PR target/61578
* lra-constarints.c (match_reload): Check presence of the input pseudo
in the output pseudo.

From-SVN: r228153

Add PR fortran/55603 working test

gcc/testsuite/
PR fortran/55603
* gfortran.dg/allocatable_function_9.f90: New.

From-SVN: r228151

invoke.texi (-fsanitize): Minor wording tweak.

2015-09-25  Tobias Burnus  <burnus@net-b.de>

* doc/invoke.texi (-fsanitize): Minor wording tweak.

From-SVN: r228148

invoke.texi (-fsanitize): Update URLs.

2015-09-25  Tobias Burnus  <burnus@net-b.de>

        * doc/invoke.texi (-fsanitize): Update URLs.

From-SVN: r228144

* tr.po: Update.

From-SVN: r228138

Fix old ChangeLog entry to include global variable name that was changed.

From-SVN: r228137

This patch unsets -freorder-blocks-and-partition when -fprofile-use is not specified.

This patch unsets -freorder-blocks-and-partition when -fprofile-use
is not specified. Function splitting was not actually being performed
in that case, as probably_never_executed_bb_p does not distinguish
any basic blocks as being cold vs hot when there is no profile data.
Leaving it enabled, however, causes the assembly code generator to create
(empty) cold sections and labels, leading to unnecessary size overhead.

2015-09-25  Teresa Johnson  <tejohnson@google.com>

* opts.c (finish_options): Unset -freorder-blocks-and-partition
if not using profile.

From-SVN: r228136

Avoid creating dangling references in case of nested tuples
for tuple constructors that construct from other tuples.

2015-09-25  Ville Voutilainen  <ville.voutilainen@gmail.com>

Avoid creating dangling references in case of nested tuples
for tuple constructors that construct from other tuples.
* include/std/tuple (_TC::_NonNestedTuple): New.
* include/std/tuple (tuple::_TNTC): New.
* include/std/tuple (tuple(const tuple<_UElements...>&),
tuple(tuple<_UElements...>&&): Use _TNTC.
* testsuite/20_util/tuple/cons/nested_tuple_construct.cc: New.

From-SVN: r228134

PR pretty-print/67567 do not pass NULL as a string

Fortran passes NULL where a non-null string is expected by the pretty-printer,
which causes a sanitizer warning. This could have been found earlier by using
gcc_checking_assert. Even if the assertion is false, the result is just an
incomplete diagnostic, thus it seems more user-friendly to assert only when
checking. I do not have any idea how to properly fix the Fortran bug, thus this
patch simply works-around it.

gcc/fortran/ChangeLog:

2015-09-25  Manuel López-Ibáñez  <manu@gcc.gnu.org>

PR pretty-print/67567
* resolve.c (resolve_fl_procedure): Work-around when iface->module
== NULL.

gcc/ChangeLog:

2015-09-25  Manuel López-Ibáñez  <manu@gcc.gnu.org>

PR pretty-print/67567
* pretty-print.c (pp_string): Add gcc_checking_assert.
* pretty-print.h (output_buffer_append_r): Likewise.

From-SVN: r228131

re PR target/67675 ([SH] Improve __builtin_strcmp alignment test)

gcc/
PR target/67675
* config/sh/sh-mem.cc (sh_expand_cmpstr): Check alignment of addr1 and
addr2 individually.  Don't emit logical or insn if one is known to
be aligned approriately.
(sh_expand_cmpnstr): Likewise.

gcc/testsuite/
PR target/67675
* gcc.target/sh/pr67675.c: New.

From-SVN: r228118

[AArch64] Force __builtin_aarch64_fp[sc]r argument into a REG

The testcase triggered an ICE because the builtin expansion
code passed the output of expand_normal directly to the SET_FP[SC]R
generator, without forcing it into a register first.

gcc/
* config/aarch64/aarch64-builtins.c (aarch64_expand_builtin): Force
__builtin_aarch64_fp[sc]r arguments into a register.

gcc/testsuite/
* gcc.target/aarch64/fpcr_fpsr_1.c: New file.

From-SVN: r228116

Rename IA MCU processor lakemount to lakemont

IA MCU processor name is lakemont, not lakemount.

gcc/

* config.gcc (x86_archs): Replace lakemount with lakemont.
(with_cpu): Likewise.
(with_arch): Likewise.
* config/i386/i386-c.c (ix86_target_macros_internal): Replace
PROCESSOR_LAKEMOUNT with PROCESSOR_LAKEMONT.  Replace
__tune_lakemount__ with __tune_lakemont__.
* config/i386/i386.c (lakemount_cost): Renamed to ...
(lakemont_cost): This.
(m_LAKEMOUNT): Renamed to ...
(m_LAKEMONT): This.
(initial_ix86_arch_features): Replace m_LAKEMOUNT with m_LAKEMONT.
(processor_target_table): Replace "lakemount" with "lakemont".
(processor_alias_table): Likewise.
(ix86_issue_rate): Replace PROCESSOR_LAKEMOUNT with
PROCESSOR_LAKEMONT.
(ix86_adjust_cost): Likewise.
(ia32_multipass_dfa_lookahead): Likewise.
* config/i386/i386.h (processor_type): Likewise.
* config/i386/x86-tune.def: Replace m_LAKEMOUNT with m_LAKEMONT.
* doc/invoke.texi: Replace lakemount with lakemont.  Replace
Lakemount with Lakemont.

gcc/testsuite/

* gcc.target/i386/pr66749.c (dg-options): Replace
-mtune=lakemount with -mtune=lakemont.
* gcc.target/i386/pr66821.c (dg-options): Likewise.
* gcc.target/i386/pr67329.c (dg-options): Likewise.

From-SVN: r228115

c-ubsan.c (ubsan_instrument_division): Remove unnecessary code.

* c-ubsan.c (ubsan_instrument_division): Remove unnecessary code.
(ubsan_instrument_shift): Likewise.

* c-c++-common/ubsan/bounds-11.c: New test.
* c-c++-common/ubsan/bounds-12.c: New test.

From-SVN: r228114

re PR sanitizer/64906 (-fsanitize=integer-divide-by-zero creates false -Wmaybe-uninitialized warning)

PR sanitizer/64906
* c-ubsan.c (ubsan_instrument_division): Also pre-evaluate OP1.

* c-c++-common/ubsan/pr64906.c: New test.

From-SVN: r228112

Change IA MCU processor from iamcu to lakemount

The first IA MCU processor will be Lakemount.  This patch changes IA MCU
processor name from iamcu to lakemount.

gcc/

* config.gcc (x86_archs): Replace iamcu with lakemount.
(with_cpu): Likewise.
(with_arch): Likewise.
* doc/invoke.texi: Likewise.
* config/i386/i386-c.c (ix86_target_macros_internal): Replace
PROCESSOR_IAMCU with PROCESSOR_LAKEMOUNT.  Replace
__tune_iamcu__ with __tune_lakemount__.
* config/i386/i386.c (iamcu_cost): Renamed to ...
(lakemount_cost): This.
(m_IAMCU): Renamed to ...
(m_LAKEMOUNT): This.
(initial_ix86_arch_features): Replace m_IAMCU with m_LAKEMOUNT.
(processor_target_table): Replace "iamcu" with "lakemount".
(processor_alias_table): Likewise.
(ix86_issue_rate): Replace PROCESSOR_IAMCU with
PROCESSOR_LAKEMOUNT.
(ix86_adjust_cost): Likewise.
(ia32_multipass_dfa_lookahead): Likewise.
* config/i386/i386.h (processor_type): Likewise.
* config/i386/x86-tune.def: Replace m_IAMCU with m_LAKEMOUNT.

gcc/testsuite/

* gcc.target/i386/pr66749.c (dg-options): Replace -mtune=iamcu
with -mtune=lakemount.
* gcc.target/i386/pr66821.c (dg-options): Likewise.
* gcc.target/i386/pr67329.c (dg-options): Likewise.

From-SVN: r228109

Daily bump.

From-SVN: r228108

pa-linux.h (HAVE_sync_compare_and_swapdi): Define.

* config/pa/pa-linux.h (HAVE_sync_compare_and_swapdi): Define.
* config/pa/pa-protos.h (pa_maybe_emit_compare_and_swap_exchange_loop):
Declare.
* config/pa/pa.c (pa_init_libfuncs): Init sync libfuncs up to 8 bytes.
(pa_expand_compare_and_swap_loop): New.
(pa_maybe_emit_compare_and_swap_exchange_loop): New.
* config/pa/pa.md (atomic_storeqi, atomic_storehi, atomic_storesi,
atomic_storesf, atomic_loaddf, atomic_storedf): New expanders.
(atomic_loaddf_1, atomic_storedf_1): New insn patterns.
(atomic_loaddi, atomic_loaddi_1, atomic_storedi, atomic_storedi_1):
Revise.

From-SVN: r228104

re PR other/57195 (Mode attributes with specific mode iterator can not be used as mode iterators in *.md files)

2015-09-24  Michael Collison  <michael.collison@linaro.org>

PR other/57195
* read-md.c (read_name): Allow mode iterators inside angle
brackets in rtl expressions.

From-SVN: r228102

re PR target/61578 (Code size increase for ARM thumb compared to 4.8.x when compiling with -Os)

2015-09-24  Vladimir Makarov  <vmakarov@redhat.com>

PR target/61578
* ira-color.c (update_allocno_cost): Add parameter.
(update_costs_from_allocno): Decrease conflict cost.  Pass the new
parameter.

From-SVN: r228097

fdiagnostics-color=never does not disable color for some diagnostics

Actually, I was trying to reject non-warning options as argument to
-Werror=. However, the new test fails because -fdiagnostics-color=never is
always placed by the driver after the warning options when calling the compiler
proper. This patch prunes all -fdiagnostics-color from the command-line but the
last one, which is moved to the first position.

gcc/ChangeLog:

2015-09-24  Manuel López-Ibáñez  <manu@gcc.gnu.org>

PR driver/67640
* opts-common.c (prune_options): Discard all -fdiagnostics-color
but the last one, which is moved to the front to be processed
first.
* opts.c (enable_warning_as_error): Reject options that do not
control warnings.

gcc/testsuite/ChangeLog:

2015-09-24  Manuel López-Ibáñez  <manu@gcc.gnu.org>

PR driver/67640
* gcc.dg/Werror-13.c: New test.

From-SVN: r228094

[AArch64] Handle const address in aarch64_print_operand

2015-09-24  Jiong Wang  <jiong.wang@arm.com>

* config/aarch64/aarch64.c (aarch64_print_operand): Add "CONST" support.

From-SVN: r228093

[AArch64] Delete aarch64_symbol_context which is not used

2015-09-24  Jiong Wang  <jiong.wang@arm.com>

* config/aarch64/aarch64-protos.h (aarch64_symbol_context): Delete.
* config/aarch64/aarch64.c (aarch64_expand_mov_immediate): Likewise.
(aarch64_cannot_force_const_mem): Likewise.
(aarch64_classify_address): Likewise.
(aarch64_classify_symbolic_expression): Likewise.
(aarch64_print_operand): Likewise.
(aarch64_classify_symbol): Likewise.
(aarch64_mov_operand_p): Likewise.
* config/aarch64/predicates.md (aarch64_valid_symref): Likewise.
(aarch64_mov_operand): Likewise.

From-SVN: r228092

Leave moved-from std::deque in a valid state

PR libstdc++/67707
* include/bits/stl_deque.h (_Deque_base::_M_move_impl): Initialize
empty object.
* testsuite/23_containers/deque/allocator/move.cc: Check moved-from
deque.

From-SVN: r228090

Fix aarch64/target_attr_10 test following r227997.

gcc/testsuite/ChangeLog:

* gcc.target/aarch64/target_attr_10.c (foo): Use dg-message for note.

From-SVN: r228088

runtime: Fix recent lfstack change on Solaris.

    Reviewed-on: https://go-review.googlesource.com/14922

From-SVN: r228087

rs6000: Fix -mdebug=stack code for spe_gp_offset

2015-09-23  Segher Boessenkool  <segher@kernel.crashing.org>

* config/rs6000/rs6000.c (debug_stack_info): Invert the test
for info->spe_gp_size.

From-SVN: r228086

re PR lto/67699 (ICE (segfault) compiling a const array with -flto)

2015-09-24  Richard Biener  <rguenther@suse.de>

PR lto/67699
* lto-cgraph.c (compute_ltrans_boundary): Do not stream
abstract origins.

* g++.dg/pr67699.C: New testcase.

From-SVN: r228084

ARM: fp16 Fix PR 67624 - Incorrect conversion of float Infinity to __fp16

PR libgcc/67624
libgcc:
* config/arm/fp16.c (__gnu_f2h_internal): Handle infinity correctly.
gcc/testsuite:
* gcc.target/arm/fp16-inf.c: New test.

From-SVN: r228082

Additional changes to switch from gimple to gimple *

gcc/
* tree-object-size.c (plus_stmt_object_size)
(cond_expr_object_size): Change the formal parameters from gimple
to gimple *.
* tree-ssa-sccvn.h (vn_nary_op_insert_stmt): Likewise.
* tree-ssa-sccvn.c (vn_nary_op_insert_stmt): Make it static.
* tree-ssa-sccvn.h (vn_nary_op_insert_stmt): Don't declare.

From-SVN: r228080

avx512vbmi-check.h (main): Fix register name while checking for AVX-512VBMI presence.

gcc/testsuite/
* gcc.target/i386/avx512vbmi-check.h (main): Fix register
name while checking for AVX-512VBMI presence.

From-SVN: r228079

Support PIE on Solaris

gcc/testsuite:
* lib/target-supports.exp (check_effective_target_pie): Check for
PIE support on Solaris 11.x and 12.

libgcc:
* config.host (*-*-solaris2*): Add t-crtstuff-pic to tmake_file.
Add crtbeginS.o, crtendS.o to extra_parts if libgcc_cv_solaris_crts.
* config/sol2/gmon.c: (monstartup): Don't write trailing NUL of
messages.
(internal_mcount): Likewise.
* config/sol2/t-sol2 (crtp.o, crtpg.o, gmon.o): Compile with
crt_compile, add CRTSTUFF_T_CFLAGS_S.

gcc:
* configure.ac (gcc_cv_ld_pie): Check for gld >= 2.26 on Solaris.
Check for ld -type pie on Solaris 11.x and 12.
* configure: Regenerate.
* config.in: Regenerate.

* gcc.c (LD_PIE_SPEC): Allow redefinition.

* config/sol2.h (STARTFILE_CRTBEGIN_SPEC): Define.
(STARTFILE_SPEC): Use it.
(ENDFILE_CRTEND_SPEC): Define.
(ENDFILE_SPEC): Use it and ENDFILE_ARCH_SPEC.
(SUBTARGET_EXTRA_SPECS): Add STARTFILE_CRTBEGIN_SPEC,
ENDFILE_ARCH_SPEC, ENDFILE_CRTEND_SPEC.
[HAVE_LD_PIE && HAVE_SOLARIS_CRTS] (LD_PIE_SPEC): Define.
(!(HAVE_LD_PIE && HAVE_SOLARIS_CRTS)] (LINK_PIE_SPEC): Define.
* config/i386/sol2.h (ENDFILE_SPEC): Remove.
(ENDFILE_ARCH_SPEC): Define.
* config/sparc/sol2.h (ENDFILE_ARCH_SPEC): Define.

From-SVN: r228078

Use CRTs provided by Solaris

gcc:
* configure.ac (gcc_cv_solaris_crts): New test.
* configure. Regenerate.
* config.in: Regenerate.
* config/sol2.h (STARTFILE_SPEC): Simplify, provide
HAVE_SOLARIS_CRTS variant.

libgcc:
* configure.ac (libgcc_cv_solaris_crts): New test.
* configure: Regenerate.
* config.in: Regenerate.
* config/sol2/crtp.c, config/sol2/crtpg.c: New files.
* config/gmon-sol2.c: Rename to ...
* config/sol2/gmon.c: ... this.
Include auto-target.h.
(internal_mcount): Wrap setup handling in !HAVE_SOLARIS_CRTS.
* config/t-sol2: Rename to ...
* config/sol2/t-sol2: ... this.
(gmon.o): Reflect renaming.
(crtp.o, crtpg.o): New rules.
* config.host (*-*-solaris2*): Reflect renaming.
Use system CRTs if present.
Remove default CRT case.

From-SVN: r228077

[tree-inline][obvious] Delete redundant count_insns_seq

* tree-inline.h (count_insns_seq): Delete prototype.
(estimate_num_insns_seq): Define prototype.
* tree-inline.c (count_insns_seq): Delete.
(estimate_num_insns_seq): Remove static qualifier.
* tree-eh.c (decide_copy_try_finally): Replace use of count_insns_seq
with estimate_num_insns_seq.

From-SVN: r228075

tree-ssa-sccvn.h (vn_reference_op_struct): Add clique and base members.

2015-09-24  Richard Biener  <rguenther@suse.de>

* tree-ssa-sccvn.h (vn_reference_op_struct): Add clique and base
members.
* tree-ssa-sccvn.c (copy_reference_ops_from_ref): Record clique
and base for MEM_REF and TARGET_MEM_REF.  Handle BIT_FIELD_REF
offset.
(ao_ref_init_from_vn_reference): Record clique and base in the
built base.
* tree-ssa-pre.c (create_component_ref_by_pieces_1): Likewise

* g++.dg/tree-ssa/restrict3.C: New testcase.

From-SVN: r228074

re PR c/48885 (missed optimization with restrict qualifier?)

2015-09-24  Richard Biener  <rguenther@suse.de>

PR tree-optimization/48885
* tree-ssa-structalias.c (visit_loadstore): Handle default defs
as not including any restrict tags from other pointers.

* gcc.dg/tree-ssa/restrict-6.c: New testcase.

From-SVN: r228073

ChangeLog: Fix whitespace.

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

From-SVN: r228072

Fix a -Wmisleading-indentation false-negative

gcc/c-family/ChangeLog:

* c-indentation.c (should_warn_for_misleading_indentation):
Compare next_stmt_vis_column with guard_line_first_nws instead
of with guard_line_vis_column.

gcc/testsuite/ChangeLog:

* c-c++-common/Wmisleading-indentation.c: Augment test.

From-SVN: r228071

Daily bump.

From-SVN: r228070

linux-atomic.c (__kernel_cmpxchg2): Reorder error checks.

* config/pa/linux-atomic.c (__kernel_cmpxchg2): Reorder error checks.
(__sync_fetch_and_##OP##_##WIDTH): Change result to match type of
__kernel_cmpxchg2.
(__sync_##OP##_and_fetch_##WIDTH): Likewise.
(__sync_val_compare_and_swap_##WIDTH): Likewise.
(__sync_bool_compare_and_swap_##WIDTH): Likewise.
(__sync_lock_test_and_set_##WIDTH): Likewise.
(__sync_lock_release_##WIDTH): Likewise.
(__sync_fetch_and_##OP##_4): Change result to match type of
__kernel_cmpxchg.
(__sync_##OP##_and_fetch_4): Likewise.
(__sync_val_compare_and_swap_4): Likewise.
(__sync_bool_compare_and_swap_4): likewise.
(__sync_lock_test_and_set_4): Likewise.
(__sync_lock_release_4): Likewise.
(FETCH_AND_OP_2): Add long long variants.
(OP_AND_FETCH_2): Likewise.
(COMPARE_AND_SWAP_2 ): Likewise.
(SYNC_LOCK_TEST_AND_SET_2): Likewise.
(SYNC_LOCK_RELEASE_2): Likewise.
(__sync_bool_compare_and_swap_##WIDTH): Correct return.

From-SVN: r228065

re PR go/67695 (Please improve POSIX shell compatibility of libgo/mksysinfo.sh)

PR go/67695
    mksysinfo.sh: Use = with test rather than ==.

    Fixes GCC PR 67695.

    Reviewed-on: https://go-review.googlesource.com/14858

From-SVN: r228064

Fix small typos in the coding rule enforcement warnings.

/cp
2015-09-23  Ville Voutilainen  <ville.voutilainen@gmail.com>

Fix small typos in the coding rule enforcement warnings.
* parser.c (cp_parser_namespace_definition): Replace 'namepace'
with 'namespace'.

/testsuite
2015-09-23  Ville Voutilainen  <ville.voutilainen@gmail.com>

Fix small typos in the coding rule enforcement warnings.
* g++.dg/diagnostic/disable.C: Replace 'namepace'
with 'namespace'.

From-SVN: r228061

runtime: rewrite lfstack packing/unpacking to look more like that in Go

    Reviewed-on: https://go-review.googlesource.com/13037

From-SVN: r228057

Fix --enable-offload-targets/-foffload handling, pt. 2

gcc/
* gcc.c (handle_foffload_option): Don't lose the trailing NUL
character when appending to offload_targets.

From-SVN: r228054

Fix --enable-offload-targets/-foffload handling, pt. 1

gcc/
* configure.ac (offload_targets, OFFLOAD_TARGETS): Separate
offload targets by commas, not colons.
* config.in: Regenerate.
* configure: Likewise.
* gcc.c (driver::maybe_putenv_COLLECT_LTO_WRAPPER): Due to that,
instead of setting up the default offload targets here...
(process_command): ..., do it here.
libgomp/
* plugin/configfrag.ac (OFFLOAD_TARGETS): Clarify that offload
targets are separated by commas.
* config.h.in: Regenerate.

From-SVN: r228053

Refactor omp_reduction_init: omp_reduction_init_op

gcc/
* omp-low.h (omp_reduction_init_op): Declare.
* omp-low.c (omp_reduction_init_op): New, broken out of ...
(omp_reduction_init): ... here.  Call it.
* tree-parloops.c (initialize_reductions): Use
omp_reduction_init_op.

Co-Authored-By: Nathan Sidwell <nathan@codesourcery.com>
From-SVN: r228052

re PR middle-end/67662 (-fsanitize=undefined cries wolf for X - 1 + X when X is 2**30)

2015-09-23   Richard Biener  <rguenther@suse.de>

PR middle-end/67662
* fold-const.c (fold_binary_loc): Do not reassociate two vars with
undefined overflow unless they will cancel out.

* gcc.dg/ubsan/pr67662.c: New testcase.

From-SVN: r228051

AVX-512. Fix emit in '*<mshift><mode>3' pattern.

gcc/
 * config/i386/i386.md (define_insn "*<mshift><mode>3"): Fix
 insn emit.

From-SVN: r228050

[c-family/49654/49655] reject invalid options in pragma diagnostic

Use find_opt instead of linear search through options in
handle_pragma_diagnostic (PR 49654) and reject non-warning options and
options not valid for the current language (PR 49655).

gcc/testsuite/ChangeLog:

2015-09-23  Manuel López-Ibáñez  <manu@gcc.gnu.org>

PR c/49655
* gcc.dg/pragma-diag-6.c: New test.

gcc/ChangeLog:

2015-09-23  Manuel López-Ibáñez  <manu@gcc.gnu.org>

PR c/49655
* opts.h (write_langs): Declare.
* opts-global.c (write_langs): Make it extern.

gcc/c-family/ChangeLog:

2015-09-23  Manuel López-Ibáñez  <manu@gcc.gnu.org>

PR c/49654
PR c/49655
* c-pragma.c (handle_pragma_diagnostic): Detect non-warning
options and options not valid for the current language.

From-SVN: r228049

re PR target/67391 ([SH] Convert clrt addc to normal add insn)

gcc/
PR target/67391
* config/sh/sh.md (addsi3, *addsi3_compact): Don't check for overlapping
regs when matching the pattern.

From-SVN: r228046

[AArch64] Fix vcvt_high_f64_f32 and vcvt_figh_f32_f64 intrinsics.

gcc/

* config/aarch64/aarch64-simd.md
(aarch64_float_truncate_hi_v4sf): Rewrite as an expand.
(aarch64_float_truncate_hi_v4sf_le): New.
(aarch64_float_truncate_hi_v4sf_be): Likewise.

gcc/testsuite/

* gcc.target/aarch64/advsimd-intrinsics/vcvt_high_1.c: New.

From-SVN: r228044

Limit number of symlinks that canonical() will resolve

* src/filesystem/ops.cc (canonical): Simplify error handling and
limit number of symlinks that can be resolved.

From-SVN: r228043

Fix semantics of Filesystem TS directory iterators

[class.directory_iterator] p4 and [directory_iterator.members] p4
require that only the default constructor and ignored permission denied
errors can create the end iterator.

* acinclude.m4 (GLIBCXX_CHECK_FILESYSTEM_DEPS): Remove _GLIBCXX_
prefix from HAVE_STRUCT_DIRENT_D_TYPE.
* config.h.in: Regenerate.
* configure: Regenerate.
* include/experimental/fs_dir.h (operator==, operator==):
Use owner_before instead of pointer equality.
(directory_iterator(std::shared_ptr<_Dir>, error_code*)): Remove.
* src/filesystem/dir.cc (ErrorCode): Remove.
(_Dir::advance): Change ErrorCode parameter to error_code*, add
directory_options parameter and check it on error.
(opendir): Rename to open_dir to avoid clashing with macro. Change
ErrorCode parameter to error_code*.
(make_shared_dir): Remove.
(native_readdir) [_GLIBCXX_FILESYSTEM_IS_WINDOWS]: Don't set errno.
(directory_iterator(std::shared_ptr<_Dir>, error_code*)): Remove.
(directory_iterator(const path&, directory_options, error_code*)):
Pass options to _Dir::advance and create non-end iterator on error.
(recursive_directory_iterator(const path&, directory_options,
error_code*)): Clear error_code on ignored error, create non-end
iterator otherwise.
(recursive_directory_iterator::increment): Pass _M_options to
_Dir::advance.
(recursive_directory_iterator::pop): Likewise.
* testsuite/experimental/filesystem/iterators/directory_iterator.cc:
New.
* testsuite/experimental/filesystem/iterators/
recursive_directory_iterator.cc: New.

From-SVN: r228042

Fix filesystem::create_directories() function

* src/filesystem/ops.cc (is_dot, is_dotdot): Define new helpers.
(create_directories): Fix error handling.
* testsuite/experimental/filesystem/operations/create_directories.cc:
New.

From-SVN: r228041

tree-ssa-structalias.c (intra_create_variable_infos): Build representatives for all restrict qualified pointer destinations.

2015-09-23  Richard Biener  <rguenther@suse.de>

* tree-ssa-structalias.c (intra_create_variable_infos): Build
representatives for all restrict qualified pointer destinations.

* g++.dg/tree-ssa/restrict2.C: Un-XFAIL testcase.

From-SVN: r228040

AVX-512. Add kshift[lr][bwdq]. Fix iterator.

gcc/
* config/i386/i386.md (define_code_attr mshift): New.
(define_mode_iterator SWI1248_AVX512BW): Rename ...
(SWI1248_AVX512BW): ... to this. Make QI enabled for TARGET_AVX512DQ
only.
(define_insn "*k<logic><mode>"): Use new iterator name.
(define_insn "*<mshift><mode>3"): New.

From-SVN: r228034

Fix use of valgrind API

gcc/
PR middle-end/67649
* memory-block.h (memory_block_pool::allocate): Use valgrind API to
mark the block as accessible.

From-SVN: r228033

Daily bump.

From-SVN: r228031

Fix -Wmisleading-indentation false-positive

gcc/c-family/ChangeLog:

* c-indentation.c (should_warn_for_misleading_indentation):
Float out and consolidate the calls to get_visual_column that
are passed guard_exploc as an argument.  Compare
next_stmt_vis_column with guard_line_first_nws instead of with
body_line_first_nws.

gcc/testsuite/ChangeLog:

* c-c++-common/Wmisleading-indentation.c: Augment test.

From-SVN: r228027

shrink-wrap: Handle multiple predecessors of prologue

The caller of try_shrink_wrapping wants to be returned a single edge to
put the prologue on.  To make that work even if there are multiple edges
(all pointing to the PRO block) that need the prologue, add a new block
that becomes the destination of all such edges, and then jumps to PRO.

In the general case, some edges to PRO will need to be redirected, and
not all edges *can* be redirected.  This adds a can_get_prologue function
that detects such cases.  This then happily can also handle the "prologue
clobbers some reg that is live on the edge we want to insert it on" case.

Not all EDGE_CROSSING edges can be redirected, so handle those the same
as EDGE_COMPLEX edges.

2015-09-22  Segher Boessenkool  <segher@kernel.crashing.org>

* function.c (thread_prologue_and_epilogue_insns): Delete
orig_entry_edge argument to try_shrink_wrapping.
* shrink-wrap.c (can_get_prologue): New function.
(can_dup_for_shrink_wrapping): Also handle EDGE_CROSSING.
(try_shrink_wrapping): Delete orig_entry_edge argument.  Use
can_get_prologue where needed.  Remove code that finds a single
edge for the prologue.  Remove code that tests if any reg clobbered
by the prologue is live on the prologue edge.  Remove code that finds
the new prologue edge after duplicating blocks.  Make a new prologue
block and edge.
* shrink-wrap.h (try_shrink_wrapping): Delete orig_entry_edge argument.

From-SVN: r228022

[PATCH] Remove undefined behaviour from hppa backend

* config/pa/pa.h (MIN_LEGIT_64BIT_CONST_INT: Avoid undefined
behavior.

From-SVN: r228021

c.opt (Wmultiple-inheritance, [...]): New C++ warnings.

c-family/
* c.opt (Wmultiple-inheritance, Wvirtual-inheritance, Wtemplates,
Wnamespaces): New C++ warnings.

cp/
* decl.c (xref_basetypes): Check virtual and/or multiple
inheritance warning.
* parser.c (cp_parser_namespace_definition): Check namespaces
warning.
* pt.c (push_template_decl_real): Check templates warning.

* doc/invoke.texi  (-Wmultiple-inheritance, -Wvirtual-inheritance,
-Wtemplates, -Wnamespaces): Document.

testsuite/
* g++.dg/diagostic/disable.C: New.

From-SVN: r228019

Disconnect -Wabi=<n> from -fabi-compat-version=<n>.

gcc/c-family/
* c-common.h (abi_compat_version_crosses): New.
(warn_abi_version): Declare.
* c-common.c: Define it.
* c-opts.c (c_common_post_options): Handle it.
flag_abi_compat_version defaults to 8.
gcc/cp/
* mangle.c (abi_warn_or_compat_version_crosses): New.
Use it instead of abi_version_crosses.
(mangle_decl): Deal with -fabi-compat-version and -Wabi separately.

From-SVN: r228017

Handle restrict pointer references as restrict in AA

2015-09-22  Tom de Vries  <tom@codesourcery.com>

PR tree-optimization/67671
* tree-ssa-structalias.c (create_variable_info_for_1): Handle restrict
pointer references as restrict.

* g++.dg/pr67671.C: New test.

From-SVN: r228015

Update Andrew Cagney's address.

From-SVN: r228014

nios2.c (nios2_legitimize_address): When handling 'reg + reloc' cases...

2015-09-22  Chung-Lin Tang  <cltang@codesourcery.com>

* config/nios2/nios2.c (nios2_legitimize_address): When handling
'reg + reloc' cases, allow first operand to be non-REG, and use
force_reg() to enforce address pattern.

From-SVN: r228012

re PR target/67480 (AVX512 bitwise logic operations pattern is incorrect)

PR target/67480

gcc/
* config/i386/sse.md (define_mode_iterator VI48_AVX_AVX512F): New.
(define_mode_iterator VI12_AVX_AVX512F): New.
(define_insn "<mask_codefor><code><mode>3<mask_name>"): Change
all iterators to VI48_AVX_AVX512F. Extract remaining modes ...
(define_insn "*<code><mode>3"): ... Into new pattern using
VI12_AVX_AVX512F iterators without masking.

gcc/testsuite/
* gcc.target/i386/pr67480.c: New test.

From-SVN: r228010

AVX-512. Introduce SKylake server CPU.

gcc/
* config.gcc: Support "skylake-avx512".
        * config/i386/i386-c.c (ix86_target_macros_internal): Handle
        PROCESSOR_SKYLAKE_AVX512.
        * config/i386/i386.c (m_SKYLAKE_AVX512): Define.
        (processor_target_table): Add "skylake-avx512".
        (PTA_SKYLAKE_AVX512): Define.
(ix86_option_override_internal): Add "skylake_avx512".
        (fold_builtin_cpu): Handle "skylake_avx512", add F_AVX512VL
F_AVX512BW, F_AVX512DQ, F_AVX512ER, F_AVX512PF, F_AVX512CD.
        * config/i386/i386.h (TARGET_SKYLAKE_AVX512): Define.
        (processor_type): Add PROCESSOR_SKYLAKE_AVX512.
* doc/invoke.texi (skylake-avx512): New.

libgcc/
* libgcc/config/i386/cpuinfo.c (enum processor_features): Add
FEATURE_AVX512VL, FEATURE_AVX512BW, FEATURE_AVX512DQ,
FEATURE_AVX512CD, FEATURE_AVX512ER, FEATURE_AVX512PF.
(get_available_features): Habdle new features.

gcc/testsuite/
        * gcc.target/i386/funcspec-5.c: Test avx512vl, avx512bw,
avx512dq, avx512cd, avx512er, avx512pf and skylake-avx512.
* gcc.target/i386/builtin_target.c: Test  avx512vl, avx512bw,
avx512dq, avx512cd, avx512er and avx512pf.

From-SVN: r228009

AVX-512. Fix patterns for kunpck insns.

gcc/
* gcc/config/i386/i386.md (define_insn "kunpckhi"): Fix
operand in pattern.
(define_insn "kunpcksi"): Ditto.
(define_insn "kunpckdi"): Ditto.

From-SVN: r228005

AVX-512. Fix kxnor insn split.

gcc/
* gcc/config/i386/i386.md (define_split not/xor SWI1248x): Use
iterator instead of fixed modes.

From-SVN: r228003

[AArch64] Use atomic load-operate instructions for update-fetch patterns.

2015-09-22  Matthew Wahab  <matthew.wahab@arm.com>

* config/aarch64/aarch64-protos.h (aarch64_gen_atomic_ldop):
Adjust declaration.
* config/aarch64/aarch64.c (aarch64_emit_bic): New.
(aarch64_gen_atomic_ldop): Adjust comment.  Add parameter
out_result.  Update to support update-fetch operations.
* config/aarch64/atomics.md (aarch64_atomic_exchange<mode>_lse):
Adjust for change to aarch64_gen_atomic_ldop.
(aarch64_atomic_<atomic_optab><mode>_lse): Likewise.
(aarch64_atomic_fetch_<atomic_optab><mode>_lse): Likewise.
(atomic_<atomic_optab>_fetch<mode>): Change to an expander.
(aarch64_atomic_<atomic_optab>_fetch<mode>): New.
(aarch64_atomic_<atomic_optab>_fetch<mode>_lse): New.

gcc/testsuite
2015-09-22  Matthew Wahab  <matthew.wahab@arm.com>

* gcc.target/aarch64/atomic-inst-ldadd.c: Add tests for
update-fetch operations.
* gcc.target/aarch64/atomic-inst-ldlogic.c: Likewise.

From-SVN: r228002

[AArch64] Use atomic load-operate instructions for fetch-update patterns.

gcc/
2015-09-22  Matthew Wahab  <matthew.wahab@arm.com>

* config/aarch64/aarch64-protos.h
(aarch64_atomic_ldop_supported_p): Declare.
* config/aarch64/aarch64.c (aarch64_atomic_ldop_supported_p): New.
(enum aarch64_atomic_load_op_code): New.
(aarch64_emit_atomic_load_op): New.
(aarch64_gen_atomic_ldop): Update to support load-operate
patterns.
* config/aarch64/atomics.md (atomic_<atomic_optab><mode>): Change
to an expander.
(aarch64_atomic_<atomic_optab><mode>): New.
(aarch64_atomic_<atomic_optab><mode>_lse): New.
(atomic_fetch_<atomic_optab><mode>): Change to an expander.
(aarch64_atomic_fetch_<atomic_optab><mode>): New.
(aarch64_atomic_fetch_<atomic_optab><mode>_lse): New.

gcc/testsuite/
2015-09-22  Matthew Wahab  <matthew.wahab@arm.com>

* gcc.target/aarch64/atomic-inst-ldadd.c: New.
* gcc.target/aarch64/atomic-inst-ldlogic.c: New.

From-SVN: r228001

[AArch64] Add atomic load-operate instructions.

2015-09-22  Matthew Wahab  <matthew.wahab@arm.com>

* config/aarch64/aarch64/atomics.md (UNSPECV_ATOMIC_LDOP): New.
(UNSPECV_ATOMIC_LDOP_OR): New.
(UNSPECV_ATOMIC_LDOP_BIC): New.
(UNSPECV_ATOMIC_LDOP_XOR): New.
(UNSPECV_ATOMIC_LDOP_PLUS): New.
(ATOMIC_LDOP): New.
(atomic_ldop): New.
(aarch64_atomic_load<atomic_ldop><mode>): New.

From-SVN: r228000

[AArch64] Make BIC, other logical instructions, available.

2015-09-22  Matthew Wahab  <matthew.wahab@arm.com>

* config/aarch64/aarch64.md
(<LOGICAL:optab>_one_cmpl_<SHIFT:optab><mode>3): Make a named
pattern.

From-SVN: r227999

[AArch64] Use atomic instructions for swap and fetch-update operations.

gcc/
2015-09-22  Matthew Wahab  <matthew.wahab@arm.com>

* config/aarch64/aarch64-protos.h (aarch64_gen_atomic_ldop):
Declare.
* config/aarch64/aarch64.c (aarch64_emit_atomic_swap): New.
(aarch64_gen_atomic_ldop): New.
(aarch64_split_atomic_op): Fix whitespace and add a comment.
* config/aarch64/atomics.md (UNSPECV_ATOMIC_SWP): New.
(aarch64_compare_and_swap<mode>_lse): Fix some whitespace.
(atomic_exchange<mode>): Replace with an expander.
(aarch64_atomic_exchange<mode>): New.
(aarch64_atomic_exchange<mode>_lse): New.
(aarch64_atomic_<atomic_optab><mode>): Fix some whitespace.
(aarch64_atomic_swp<mode>): New.

gcc/testsuite/
2015-09-22  Matthew Wahab  <matthew.wahab@arm.com>

* gcc.target/aarch64/atomic-inst-ops.inc: (TEST_MODEL): New.
(TEST_ONE): New.
        * gcc.target/aarch64/atomic-inst-swap.c: New.

From-SVN: r227998

[PATCH tree-inline] Do not say "called from here" with UNKNOWN_LOCATION

In https://sourceware.org/ml/libc-alpha/2014-12/msg00300.html, we give a
"called from here" note without actually having a location, which looks
strange. I haven't been able to generate such a testcase. If this happens,
try to at least point to the current function being called. If that location is
also unknown, skip the extra note.

gcc/ChangeLog:

2015-09-22  Manuel López-Ibáñez  <manu@gcc.gnu.org>

* tree-inline.c (expand_call_inline): Use inform for extra note.
Do not give a note with UNKNOWN_LOCATION.
Replace input_location with gimple_location (stmt).
Use true/false instead of TRUE/FALSE.

gcc/testsuite/ChangeLog:

2015-09-22  Manuel López-Ibáñez  <manu@gcc.gnu.org>

* gcc.target/i386/inline_error.c (int bar): Use dg-message for note.
* gcc.target/i386/pr57756.c (static __inline int caller): Likewise.
* gcc.target/i386/pr59789.c (f1): Likewise.
* gcc.target/i386/intrinsics_5.c (__m128i foo): Likewise.
* gcc.target/i386/intrinsics_6.c: Likewise.
* gcc.dg/winline-5.c (int t): Likewise.
* gcc.dg/winline-9.c (t): Likewise.
* gcc.dg/always_inline2.c (q): Likewise.
* gcc.dg/winline-2.c (inline int t): Likewise.
* gcc.dg/winline-6.c: Likewise.
* gcc.dg/winline-10.c (void g): Likewise.
* gcc.dg/pr49243.c (void parse): Likewise.
* gcc.dg/always_inline3.c (q2): Likewise.
* gcc.dg/winline-3.c: Likewise.
* gcc.dg/winline-7.c (inline void *t): Likewise.

From-SVN: r227997