re PR target/77834 (ICE: in make_decl_rtl, at varasm.c:1311 with -O -ftree-pre -mstringop-strategy=libcall)

PR target/77834
* alias.c (nonoverlapping_memrefs_p): Return 0 if exprx or expry
doesn't have rtl set.

* gcc.dg/pr77834.c: New test.

From-SVN: r241859

re PR c++/67980 (left shift count is negative [-Wshift-count-negative] generated for unreachable code)

/cp
2016-11-04  Paolo Carlini  <paolo.carlini@oracle.com>

PR c++/67980
* pt.c (tsubst_expr, case IF_STMT): Use fold_non_dependent_expr
to suppress unwanted warnings.

/testsuite
2016-11-04  Paolo Carlini  <paolo.carlini@oracle.com>

PR c++/67980
* g++.dg/cpp1y/pr67980.C: New.

From-SVN: r241858

rs6000.c (gimple-ssa.h): New #include.

[gcc]

2016-11-04  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>

* config/rs6000/rs6000.c (gimple-ssa.h): New #include.
(TARGET_GIMPLE_FOLD_BUILTIN): Define as
rs6000_gimple_fold_builtin.
(rs6000_gimple_fold_builtin): New function.  Add handling for
early expansion of vector addition builtins.

[gcc/testsuite]

2016-11-04  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>

* gcc.target/powerpc/fold-vec-add-1.c: New.
* gcc.target/powerpc/fold-vec-add-2.c: New.
* gcc.target/powerpc/fold-vec-add-3.c: New.
* gcc.target/powerpc/fold-vec-add-4.c: New.
* gcc.target/powerpc/fold-vec-add-5.c: New.
* gcc.target/powerpc/fold-vec-add-6.c: New.
* gcc.target/powerpc/fold-vec-add-7.c: New.

From-SVN: r241857

expr.h (copy_blkmode_from_reg): Delete.

* expr.h (copy_blkmode_from_reg): Delete.
* expr.c (copy_blkmode_from_reg): Make static.

From-SVN: r241856

defaults.h (LOAD_EXTEND_OP): Define if not already defined.

* defaults.h (LOAD_EXTEND_OP): Define if not already defined.
* combine.c (LOAD_EXTEND_OP): Delete.
(simplify_comparison): Fix comment about LOAD_EXTEND_OP.
* cse.c (LOAD_EXTEND_OP): Delete.
* fold-const.c (LOAD_EXTEND_OP): Likewise.
* fwprop.c (free_load_extend): Remove #ifdef LOAD_EXTEND_OP/#endif.
* postreload.c (LOAD_EXTEND_OP): Delete.
* reload.c (push_reload): Remove #ifdef LOAD_EXTEND_OP/#endif.
Convert conditional compilation based on WORD_REGISTER_OPERATIONS.
(find_reloads): Likewise.
* reload1.c (eliminate_regs_1): Likewise.
* rtlanal.c (nonzero_bits1): Remove #ifdef LOAD_EXTEND_OP/#endif.
(num_sign_bit_copies1): Likewise.

From-SVN: r241855

Correct typo in changelog author

gcc/testsuite/ChangeLog
* Update toma.tobacu to toma.tabacu

From-SVN: r241852

Start adding target-specific selftests

gcc/ChangeLog:
* config/i386/i386.c: Include "selftest.h" and "selftest-rtl.h".
(selftest::ix86_test_dumping_hard_regs): New function.
(selftest::ix86_run_selftests): New function.
(TARGET_RUN_TARGET_SELFTESTS): When CHECKING_P, wire this up to
selftest::ix86_run_selftests.
* doc/tm.texi.in (TARGET_RUN_TARGET_SELFTESTS): New.
* doc/tm.texi: Regenerate
* selftest-rtl.h: New file.
* rtl-tests.c: Include "selftest-rtl.h".
(selftest::assert_rtl_dump_eq): Make non-static.
(ASSERT_RTL_DUMP_EQ): Move to selftest-rtl.h.
(selftest::test_dumping_regs): Update comment.
* selftest-run-tests.c: Include "target.h".
(selftest::run_tests): If non-NULL, call
targetm.run_target_selftests.
* target.def (run_target_selftests): New hook.

From-SVN: r241851

MIPS: Handle -mbranch-likely in testsuite options

gcc/testsuite/
* gcc.target/mips/mips.exp (mips-dg-options): Downgrade to R5
for -mbranch-likely, infer -mno-branch-likely for R6.

From-SVN: r241850

Add support for ARM Cortex-M33 processor

2016-11-04  Thomas Preud'homme  <thomas.preudhomme@arm.com>

    gcc/
    * config/arm/arm-arches.def (armv8-m.main+dsp): Set Cortex-M33 as
    representative core for this architecture.
    * config/arm/arm-cores.def (cortex-m33): Define new processor.
    * config/arm/arm-tables.opt: Regenerate.
    * config/arm/arm-tune.md: Likewise.
    * config/arm/bpabi.h (BE8_LINK_SPEC): Add Cortex-M33 to the list of
    valid -mcpu options.
    * doc/invoke.texi (ARM Options): Document new Cortex-M33 processor.

From-SVN: r241849

Add support for ARM Cortex-M23 processor

2016-11-04  Thomas Preud'homme  <thomas.preudhomme@arm.com>

    gcc/
    * config/arm/arm-arches.def (armv8-m.base): Set Cortex-M23 as
    representative core for this architecture.
    * config/arm/arm-cores.def (cortex-m23): Define new processor.
    * config/arm/arm-tables.opt: Regenerate.
    * config/arm/arm-tune.md: Likewise.
    * config/arm/arm.c (arm_v6m_tune): Add Cortex-M23 to the list of cores
    this tuning parameters apply to in the comment.
    * config/arm/bpabi.h (BE8_LINK_SPEC): Add Cortex-M23 to the list of
    valid -mcpu options.
    * doc/invoke.texi (ARM Options): Document new Cortex-M23 processor.

From-SVN: r241848

fold-const.c (fold_cond_expr_with_comparison): Remove call to pedantic_non_lvalue_loc.

* fold-const.c (fold_cond_expr_with_comparison): Remove call
to pedantic_non_lvalue_loc.  Remove useless code for lvalue
where cond_expr can't be a lvalue.

From-SVN: r241847

re PR c++/71973 (c++ handles built-in functions inconsistently)

2016-11-04  Bernd Edlinger  <bernd.edlinger@hotmail.de>

        PR c++/71973
        * g++.dg/cpp1y/lambda-generic-udt.C: Fix builtin function declaration.
        * g++.dg/init/new15.C: Likewise.
        * g++.dg/ipa/inline-1.C: Likewise.
        * g++.dg/ipa/inline-2.C: Likewise.
        * g++.dg/lto/20080908-1_0.C: Likewise.
        * g++.dg/tc1/dr20.C: Likewise.
        * g++.dg/tree-ssa/inline-1.C: Likewise.
        * g++.dg/tree-ssa/inline-2.C: Likewise.
        * g++.old-deja/g++.law/except1.C: Likewise.
        * g++.old-deja/g++.other/vbase5.C: Likewise.
        * obj-c++.dg/lto/trivial-1_0.mm: Likewise.

From-SVN: r241846

re PR tree-optimization/78210 (slsr-8.c scan-tree-dump-times optimized fails)

2016-11-04  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>

PR tree-optimization/78210
* gcc.dg/tree-ssa/slsr-8.c: Fix slsr scan to include the
possibility of widening multiplies.

From-SVN: r241845

PR c++/78198 - inherited template ctor with default arg

* call.c (convert_default_arg): Look through inheriting ctors.

From-SVN: r241843

[ARC] Various small miscellaneous fixes.

gcc/
2016-11-04  Claudiu Zissulescu  <claziss@synopsys.com>

* config/arc/arc.c (arc_process_double_reg_moves): Use
gen_dexcl_2op call.
* config/arc/arc.md (movsi_insn): Disable unsupported move
instructions for ARCv2 cores.
(movdi): Use prepare_move_operands.
(movsf, movdf): Use move_dest_operand predicate.
* config/arc/constraints.md (Chs): Enable when barrel shifter is
present.
* config/arc/fpu.md (divsf3): Change to divsf3_fpu.
* config/arc/fpx.md (dexcl_3op_peep2_insn): Dx data register is
also a destination.
(dexcl_3op_peep2_insn_nores): Likewise.
* config/arc/arc.h (SHIFT_COUNT_TRUNCATED): Define to one.
(LINK_COMMAND_SPEC): Remove.

From-SVN: r241842

re PR tree-optimization/78185 (Wrong branch optimization with -O1 on x86/x86_64)

2016-11-04  Richard Biener  <rguenther@suse.de>

PR middle-end/78185
* loop-invariant.c (find_exits): Record entering inner
loops as possibly exiting to handle infinite sub-loops.
* tree-ssa-loop-im.c: Include tree-ssa-loop-niter.h.
(fill_always_executed_in_1): Honor infinite child loops.

* gcc.dg/pr78185.c: New testcase.

From-SVN: r241841

Fix wrong patch committed

From-SVN: r241840

ref-3.C: XFAIL AIX.

        * g++.dg/debug/dwarf2/ref-3.C: XFAIL AIX.
        * g++.dg/debug/dwarf2/ref-4.C: XFAIL AIX.

From-SVN: r241838

Daily bump.

From-SVN: r241837

re PR target/78192 (extract from vector registers to int results in wrong data order)

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

PR target/78192
* config/rs6000/vsx.md (vsx_extract_<mode>_di): The element number
has already been adjusted for endianness, so don't adjust it any
further.

From-SVN: r241834

re PR bootstrap/77993 (bootstrap failure on PowerPC/Linux)

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

PR target/77993
* config/rs6000/rs6000.h (FLOAT128_IBM_P): Do not allow IFmode or
ICmode unless we have standard PowerPC floating point.
* config/rs6000/rs6000.md (FP iterator): Likewise.
(FMOVE128 iterator): Likewise.

From-SVN: r241833

re PR debug/28767 (GCC should output DW_TAG_ptr_to_member for member functions)

PR debug/28767
PR debug/56974
* langhooks.h (struct lang_hooks_for_types): Add type_dwarf_attribute
langhook.
* langhooks.c (lhd_type_dwarf_attribute): New function.
* langhooks-def.h (lhd_type_dwarf_attribute): Declare.
(LANG_HOOKS_TYPE_DWARF_ATTRIBUTE): Define.
(LANG_HOOKS_FOR_TYPES_INITIALIZER): Add
LANG_HOOKS_TYPE_DWARF_ATTRIBUTE.
(check_qualified_type, check_aligned_type): Call it.
* dwarf2out.c (modified_type_die): Don't use type_main_variant
for FUNCTION_TYPE or METHOD_TYPE, instead walk over variants with
check_base_type and check_lang_type.
(gen_ptr_to_mbr_type_die): If lookup_type_die is already non-NULL,
return early.  For pointer-to-data-member add DW_AT_use_location
attribute.
(gen_subroutine_type_die): Add DW_AT_{,rvalue_}reference attribute
if needed.
(gen_type_die_with_usage): Don't use type_main_variant
for FUNCTION_TYPE or METHOD_TYPE, instead walk over variants with
check_base_type and check_lang_type.  Formatting fixes. Call
get_debug_type langhook.
cp/
* tree.c (cp_check_qualified_type): Use check_base_type and
TYPE_QUALS comparison instead of check_qualified_type.
(cxx_type_hash_eq): Return false if type_memfn_rqual don't match.
* cp-objcp-common.c (cp_get_debug_type): New function.
(cp_decl_dwarf_attribute): Don't handle types here.
(cp_type_dwarf_attribute): New function.
* cp-objcp-common.h (cp_get_debug_type, cp_type_dwarf_attribute):
Declare.
(LANG_HOOKS_GET_DEBUG_TYPE, LANG_HOOKS_TYPE_DWARF_ATTRIBUTE):
Define.
testsuite/
* g++.dg/debug/dwarf2/ptrdmem-1.C: New test.
* g++.dg/debug/dwarf2/ref-3.C: New test.
* g++.dg/debug/dwarf2/ref-4.C: New test.
* g++.dg/debug/dwarf2/refqual-1.C: New test.
* g++.dg/debug/dwarf2/refqual-2.C: New test.

From-SVN: r241832

Use type_hash_eq langhook in check_qualified_type.

gcc/
* tree.c (check_lang_type): New.
(check_qualified_type): Use it.
(check_aligned_type): Use it.
* tree.h: Declare it.
gcc/cp/
* tree.c (cp_check_qualified_type): Call check_base_type instead
of check_qualified_type.
(cxx_type_hash_eq): Check ref-qualifiers.
* typeck.c (apply_memfn_quals): No need to mess with TYPE_CANONICAL.

From-SVN: r241831

* c-cppbuiltin.c (c_cpp_builtins): Correct __cpp_inheriting_constructors.

From-SVN: r241830

Add missing hunk from previous commit.

From-SVN: r241829

Default missing exponents to 0 with -fdec.

gcc/fortran/
* gfortran.texi: Document.
* gfortran.h (gfc_dt): New field default_exp.
* primary.c (match_real_constant): Default exponent with -fdec.
* io.c (match_io): Set dt.default_exp with -fdec.
* ioparm.def (IOPARM_dt_default_exp): New.
* trans-io.c (build_dt): Set IOPARM_dt_default_exp with -fdec.

libgfortran/io/
* io.h (IOPARM_DT_DEFAULT_EXP): New flag bit.
* list_read.c (parse_real, read_real): Allow omission of exponent with
IOPARM_DT_DEFAULT_EXP.
* read.c (read_f): Ditto.

gcc/testsuite/gfortran.dg/
* dec_exp_1.f90, dec_exp_2.f90, dec_exp_3.f90: New testcases.

From-SVN: r241828

[arm] Eliminate SUBTARGET_CPU_DEFAULT

* config.gcc (arm-wrs-vxworks): Set target_cpu_cname.
(arm*-freebsd*): Likewise.
(arm*-*-netbsdelf*): Likewise.
(arm*-*-linux*): Likewise.
(arm*-*-uclinux*eabi*): Likewise.
(arm*-*-phoenix*): Likewise.
(arm*-*-eabi*, arm*-*-symbianelf*, arm*-*-rtems*): Likewise.
(arm*-*-*): Don't clobber target_cpu_cname when --with-cpu is not
specified.  Default to arm6 if target_cpu_cname is not set.
* arm/arm.c (arm_option_override): Simplify logic.  Assert that the
default cpu has been correctly configured.
* arm/arm.h (TARGET_CPU_DEFAULT): Delete.
(target_cpus): Delete TARGET_CPU_generic, add TARGET_CPU_num_cores.
* arm/freebsd.h (SUBTARGET_CPU_DEFAULT): Delete.
* arm/linux-eabi.h (SUBTARGET_CPU_DEFAULT): Delete.
* arm/linux-elf.h (SUBTARGET_CPU_DEFAULT): Delete.
* arm/symbian.h (SUBTARGET_CPU_DEFAULT): Delete.
* arm/unknown-elf.h (SUBTARGET_CPU_DEFAULT): Delete.

From-SVN: r241827

[DWARF] Enable DW_CFA_VAL_EXPRESSION support

gcc/
* reg-notes.def (CFA_VAL_EXPRESSION): New entry.
* dwarf2cfi.c (dwarf2out_frame_debug_cfa_val_expression): New function.
(dwarf2out_frame_debug): Support REG_CFA_VAL_EXPRESSION.
(output_cfa_loc): Support DW_CFA_val_expression.
(output_cfa_loc_raw): Likewise.
(output_cfi): Likewise.
(output_cfi_directive): Likewise.
* dwarf2out.c (dw_cfi_oprnd1_desc): Support DW_CFA_val_expression.
(dw_cfi_oprnd2_desc): Likewise.
(mem_loc_descriptor): Recognize new pattern generated for value
expression.

From-SVN: r241826

combine lhs zero_extract fix (PR78186)

PR rtl-optimization/78186
* combine.c (change_zero_ext): Mask the RHS of a zero_extract as
well, when converting to IOR.

From-SVN: r241824

Support legacy PARAMETER statements with -std=legacy.

gcc/fortran/
* decl.c (gfc_match_parameter): Allow omitted '()' with -std=legacy.
* parse.c (decode_statement): Match "parameter" before assignments.
* gfortran.texi: Document.

gcc/testsuite/gfortran.dg/
* dec_parameter_1.f: New test.
* dec_parameter_2.f90: Likewise.
* dec_parameter_3.f90: Likewise.
* dec_parameter_4.f90: Likewise.

From-SVN: r241823

sparc.md (vec_interleave_lowv8qi): Delete.

* config/sparc/sparc.md (vec_interleave_lowv8qi): Delete.
(vec_interleave_highv8qi): Likewise.

From-SVN: r241822

Make direct emission of time profiler counter

* libgcov-profiler.c (__gcov_time_profiler): Remove.
(__gcov_time_profiler_atomic): Likewise.
* profile.c (instrument_values): Fix coding style.
(branch_prob): Use renamed function.
* tree-profile.c (init_ic_make_global_vars): Likewise.
(gimple_init_edge_profiler): Rename to
gimple_init_gcov_profiler.
tree_time_profiler_counter variable declaration.
(gimple_gen_time_profiler): Rewrite to do a direct gimple code
emission.
* value-prof.h: Remove an argument.
* gcc.dg/no_profile_instrument_function-attr-1.c: Update scanned
output.
* gcc.dg/tree-prof/time-profiler-3.c: New test.

From-SVN: r241821

re PR libgcc/78067 (libgcc2 calls count_leading_zero with 0)

2016-11-03  Bernd Edlinger  <bernd.edlinger@hotmail.de>

        PR libgcc/78067
        * libgcc2.c (__floatdisf, __floatdidf): Avoid undefined results from
        count_leading_zeros.

testsuite:
2016-11-03  Bernd Edlinger  <bernd.edlinger@hotmail.de>

        PR libgcc/78067
        * gcc.dg/torture/fp-int-convert.h: Add more conversion tests.

From-SVN: r241817

rs6000.c (rs6000_xcoff_declare_object_name): Use symtab_node::get_create.

2016-11-03  Richard Biener  <rguenther@suse.de>

        * config/rs6000/rs6000.c (rs6000_xcoff_declare_object_name): Use
        symtab_node::get_create.

From-SVN: r241816

[rtlanal] Fix WORD_REGISTER_OPERATIONS condition in nonzero_bits

* rtlanal.c (nonzero_bits1): Fix WORD_REGISTER_OPERATIONS condition.
Move comments into more natural position.

From-SVN: r241815

[ARC] Define SIZE_TYPE and PTRDIFF_TYPE correctly.

This silences tons of -Wformat= warnings when building ARC Linux kernel
with gcc 6.x (and restores the ARC gcc 4.8.x behaviour) which had
similar fix.

gcc/
2016-11-03  Vineet Gupta  <vgupta@synopsys.com>

* config/arc/arc.h (SIZE_TYPE): Define as unsigned int.
(PTRDIFF_TYPE): Define as int.

From-SVN: r241812

make targetm.gen_ccmp{first,next} take rtx_insn **

gcc/ChangeLog:

2016-11-03  Trevor Saunders  <tbsaunde+gcc@tbsaunde.org>

* ccmp.c (expand_ccmp_expr_1): Adjust.
(expand_ccmp_expr): Likewise.
(expand_ccmp_next): Likewise.
* config/aarch64/aarch64.c (aarch64_gen_ccmp_next): Likewise.
(aarch64_gen_ccmp_first): Likewise.
* doc/tm.texi: Regenerate.
* target.def (gen_ccmp_first): Change argument types to rtx_insn *.
(gen_ccmp_next): Likewise.

From-SVN: r241811

tree-vect-loop.c (destroy_loop_vec_info): Handle cond_expr.

* tree-vect-loop.c (destroy_loop_vec_info): Handle cond_expr.
(vect_is_simple_reduction): Swap cond_reduction by inversion.

From-SVN: r241810

Fix bougs PR 78170 failure for avr

The test defines int bitfields that are more than 16 bits wide, and this
fails for the avr with its 16 bit integer size. Therefore, require
effective target int32plus.

gcc/testsuite/

2016-11-03  Senthil Kumar Selvaraj  <senthil_kumar.selvaraj@atmel.com>

* gcc.c-torture/execute/pr78170.c: Require int32plus.

From-SVN: r241809

Daily bump.

From-SVN: r241808

i386.c (ix86_init_libfuncs): New.

* config/i386/i386.c (ix86_init_libfuncs): New.  Call
darwin_rename_builtins here.
(ix86_expand_divmod_libfunc): New.
(TARGET_INIT_LIBFUNCS): Unconditionally define to ix86_init_libfuncs.
(TARGET_EXPAND_DIVMOD_LIBFUNC): Define.

From-SVN: r241805

Makefile.in (LIB2_DIVMOD_FUNCS): Add _divmoddi4.

* Makefile.in (LIB2_DIVMOD_FUNCS): Add _divmoddi4.
* libgcc2.c (__divmoddi4): New function.
* libgcc2.h (__divmoddi4): Declare.
* libgcc-std.ver.in (GCC_7.0.0): New. Add __PFX_divmoddi4
and __PFX_divmodti4.

From-SVN: r241804

nvptx.c (PTX_GANG_DEFAULT): Set to zero.

gcc/
* config/nvptx/nvptx.c (PTX_GANG_DEFAULT): Set to zero.

libgomp/
* plugin/plugin-nvptx.c (nvptx_exec): Interrogate board attributes
to determine default geometry.
* testsuite/libgomp.oacc-c-c++-common/loop-auto-1.c: Set gang
dimension.

Co-Authored-By: Nathan Sidwell <nathan@acm.org>
From-SVN: r241803

xtensa: don't xfail gcc.c-torture/compile/20001226-1.c

With jump trampolines implemented in binutils since 2.25 and enabled by
default this test no longer fails on xtensa.

2016-11-02  Max Filippov  <jcmvbkbc@gmail.com>
gcc/testsuite/
* gcc.c-torture/compile/20001226-1.c: Don't xfail on xtensa.

From-SVN: r241801

xtensa: fix ICE on pr59037.c test

xtensa gcc gets ICE on pr59037.c test because its xtensa_output_literal
function cannot handle integer literals of sizes other than 4 and 8,
whereas the test uses 16-byte int vector.
Split integer literal formatting into the recursive function
xtensa_output_integer_literal_parts capable of handling literals of any
power of 2 size not less than 4.

2016-11-02  Max Filippov  <jcmvbkbc@gmail.com>
gcc/
* config/xtensa/xtensa.c (xtensa_output_integer_literal_parts):
New function.
(xtensa_output_literal): Use xtensa_output_integer_literal_parts
to format MODE_INT and MODE_PARTIAL_INT literals.

From-SVN: r241800

rs6000: Disable shrink-wrap-separate for abi=spe (PR78168)

With the SPE ABI, if we wrap GPRs we need to handle the upper half of the
extended 64-bit registers as well, which we cannot easily do.  So, this
patch disables separate shrink-wrapping for the SPE ABI.

PR target/78168
* config/r6000/rs6000.c (rs6000_get_separate_components): Return
NULL if TARGET_SPE_ABI.

From-SVN: r241799

Fix for big-endian gcc.c-torture/execute/pr55750.c

* gimple-ssa-store-merging.c (encode_tree_to_bitpos): Don't forget to
clear padding bits even when they're less than a byte.

From-SVN: r241798

Fix failing vminnm/vmaxnm test on ARM.

* gcc.target/arm/simd/vmaxnm_f32_1.c (dg-require-effective-target):
Check for arm_v8_neon_hw.
* gcc.target/arm/simd/vmaxnmq_f32_1.c (dg-require-effective-target):
Likewise.
* gcc.target/arm/simd/vminnm_f32_1.c (dg-require-effective-target):
Likewise.
* gcc.target/arm/simd/vminnmq_f32_1.c(dg-require-effective-target):
Likewise.

From-SVN: r241797

gimple-ssa-store-merging.c: Include gimplify-me.h.

2016-11-02  Richard Biener  <rguenther@suse.de>

* gimple-ssa-store-merging.c: Include gimplify-me.h.
(imm_store_chain_info::output_merged_stores): Force base_addr
to be proper GIMPLE for a MEM_REF address.
(pass_store_merging::execute): Restrict negative bitpos
handling to non-MEM_REF bases.  Remove TREE_THIS_VOLATILE
check.  Take into account non-NULL_TREE offset if the base
is already addressable.

* gcc.dg/store_merging_8.c: New testcase.

From-SVN: r241796

New warning -Wargument-mismatch for function argument mismatches.

gcc/fortran/
* lang.opt, invoke.texi: New argument -Wargument-mismatch.
* interface.c (compare_parameter, compare_actual_formal,
gfc_check_typebound_override, argument_rank_mismatch): Control argument
mismatch warnings with -Wargument-mismatch.
* resolve.c (resolve_structure_cons, resolve_global_procedure): Ditto.

gcc/testsuite/gfortran.dg/
* warn_argument_mismatch_1.f90: New test.

From-SVN: r241795

Allow warning flags to associate through gfc_error.

gcc/fortran/
* gfortran.h (gfc_error): New declaration for gfc_error with 'opt'.
* error.c (gfc_error): Add optional 'opt' argument.
* error.c (gfc_notify_std): Call fully-qualified gfc_error.

From-SVN: r241794

Add LE/BE SHA1H patterns with a V2SI input.

Add LE/BE SHA1H patterns with a V2SI input.  This avoids unnecessary
DUPs when using intrinsics like vsha1h_u32 (vgetq_lane_u32 (x, 0)).

    gcc/
        * config/aarch64/aarch64-simd.md (aarch64_crypto_sha1hv4si):
        New pattern.
        (aarch64_be_crypto_sha1hv4si): New pattern.

From-SVN: r241791

The add expander still contains some expansion code that was required for the previous prolog/epilog code...

The add expander still contains some expansion code that was required for the
previous prolog/epilog code, but which is no longer needed.  I also noticed
that the current version splits off immediates from frame addressing
instructions, which doesn't seem a good idea.  Avoiding this resulted in small
codesize improvements.

    gcc/
        * config/aarch64/aarch64.md (add<mode>3): Remove
        redundant code.  Don't split frame based additions.

From-SVN: r241790

gimple-ssa-store-merging.c (struct store_immediate_info): Remove redundant val and dest members.

2016-11-02  Richard Biener  <rguenther@suse.de>

* gimple-ssa-store-merging.c (struct store_immediate_info): Remove
redundant val and dest members.
(store_immediate_info::store_immediate_info): Adjust.
(merged_store_group::merged_store_group): Adjust.
(merged_store_group::apply_stores): Likewise.
(struct imm_store_chain_info): Add base_addr field.
(imm_store_chain_info::imm_store_chain_info): New constructor.
(imm_store_chain_info::terminate_and_process_chain): Do not pass base.
(imm_store_chain_info::output_merged_store): Likewise.  Use
addr_base which is already the address.
(imm_store_chain_info::output_merged_stores): Likewise.
(pass_tree_store_merging::terminate_all_aliasing_chains): Take
imm_store_chain_info instead of base.  Fix alias check.
(pass_tree_store_merging::terminate_and_release_chain): Likewise.
(imm_store_chain_info::coalesce_immediate_stores): Adjust.
(pass_store_merging::execute): Refuse to operate on TARGET_MEM_REF.
use the address of the base and adjust for other changes.

From-SVN: r241789

prod-options.c: XFAIL AIX.

        * gcc.dg/debug/dwarf2/prod-options.c: XFAIL AIX.
        * gcc.dg/debug/dwarf2/pr71855.c: XFAIL AIX.
        * gcc.dg/addr_equal_1.c: XFAIL AIX.
        * gcc.dg/pr65779.c: XFAIL AIX.
        * gcc.dg/pr70405.c: XFAIL AIX.

From-SVN: r241787

inline-var-1.C: XFAIL AIX.

        * g++.dg/debug/dwarf2/inline-var-1.C: XFAIL AIX.
        * g++.dg/debug/dwarf2/ref-2.C: XFAIL AIX.
        * g++.dg/debug/dwarf2/constexpr-var-1.C: XFAIL AIX.

From-SVN: r241786

Fix linker detection in check_gc_sections_available

* lib/target-supports.exp (check_gc_sections_available): Use
-print-prog-name=ld to determine linker used.

From-SVN: r241783

Fix host_size_t_cst_p predicat

* fold-const-call.c (host_size_t_cst_p): Test whether
t is convertible to size_t.
* gcc.dg/tree-ssa/builtins-folding-gimple-ub.c (main): Add
test case.

From-SVN: r241780

PR tree-optimization/78170: Truncate sign-extended padding when encoding bitfields

PR tree-optimization/78170
* gimple-ssa-store-merging.c (encode_tree_to_bitpos): Truncate padding
introduced by native_encode_expr on little-endian as well.

* gcc.c-torture/execute/pr78170.c: New test.

From-SVN: r241779

PR tree-optimization/78162: Reject negative offsets in store merging early

PR tree-optimization/78162
* gimple-ssa-store-merging.c (execute): Mark stores with bitpos < 0
as invalid.

* gcc.c-torture/compile/pr78162.c: New test.

From-SVN: r241778

[AArch64] Add function comments to some prologue/epilogue helpers

* config/aarch64/aarch64.c (aarch64_register_saved_on_entry): Add
function comment.
(aarch64_next_callee_save): Likewise.
(aarch64_pushwb_single_reg): Likewise.
(aarch64_gen_storewb_pair): Likewise.
(aarch64_push_regs): Likewise.
(aarch64_gen_loadwb_pair): Likewise.
(aarch64_pop_regs): Likewise.
(aarch64_gen_store_pair): Likewise.
(aarch64_gen_load_pair): Likewise.
(aarch64_save_callee_saves): Likewise.
(aarch64_restore_callee_saves): Likewise.

From-SVN: r241777

re PR tree-optimization/78035 (Inconsistency between address comparison and alias analysis)

2016-11-02  Richard Biener  <rguenther@suse.de>

PR tree-optimization/78035
PR tree-optimization/77964
* gimple-pretty-print.c (pp_points_to_solution): Print
vars_contains_interposable.
* tree-ssa-alias.c: Include varasm.h.
(ptrs_compare_unequal): Check vars_contains_interposable and
decl_binds_to_current_def_p.
(dump_points_to_solution): Dump vars_contains_interposable.
* tree-ssa-alias.h (struct pt_solution): Add vars_contains_interposable
flag.
* tree-ssa-structalias.c: Include varasm.h.
(set_uids_in_ptset): Record whether vars contains a
not decl_binds_to_current_def_p variable in vars_contains_interposable.
(ipa_escaped_pt): Update initializer.

* gcc.target/i386/pr78035.c: New testcase.

From-SVN: r241776

re PR middle-end/78047 (Chromium apparently gets miscompiled)

2016-11-02  Richard Biener  <rguenther@suse.de>

PR tree-optimization/78047
* tree-ssa-structalias.c (push_fields_onto_fieldstack): Initialize
fake field at offset zero conservatively regarding to may_have_pointers.

From-SVN: r241775

tree-vrp.c (evrp_dom_walker::before_dom_children): Call infer_value_range on stmt ops and update value-ranges.

2016-11-02  Richard Biener  <rguenther@suse.de>

* tree-vrp.c (evrp_dom_walker::before_dom_children): Call
infer_value_range on stmt ops and update value-ranges.
Dump visited stmts and blocks.
(evrp_dom_walker::push_value_range): Dump changes.
(evrp_dom_walker::pop_value_range): Likewise.
(evrp_dom_walker::try_find_new_range): Avoid noop changes.

* gcc.dg/tree-ssa/vrp111.c: New testcase.
* gcc.dg/tree-ssa/pr20702.c: Disable EVRP.
* gcc.dg/tree-ssa/pr21086.c: Likewise.
* gcc.dg/tree-ssa/pr58480.c: Likewise.
* gcc.dg/tree-ssa/vrp08.c: Likewise.

From-SVN: r241774

remove cast from prev_nonnote_insn_bb

gcc/ChangeLog:

2016-11-01  Trevor Saunders  <tbsaunde+gcc@tbsaunde.org>

* emit-rtl.c (prev_nonnote_insn_bb): Change argument type to
rtx_insn *.
* rtl.h (prev_nonnote_insn_bb): Adjust prototype.

From-SVN: r241773

remove cast in delete_insn_chain

gcc/ChangeLog:

2016-11-01  Trevor Saunders  <tbsaunde+gcc@tbsaunde.org>

* cfgrtl.c (delete_insn_chain): Change argument type to rtx_insn *
and adjust for that.
* cfgrtl.h (delete_insn_chain): Adjust prototype.

From-SVN: r241772

remove cast to rtx_insn * in remove_note

gcc/ChangeLog:

2016-11-01  Trevor Saunders  <tbsaunde+gcc@tbsaunde.org>

* config/rl78/rl78.c (gen-and_emit_move): Change argument type
to rtx_insn *.
(transcode_memory_rtx): Likewise.
(move_to_acc): Likewise.
(move_from_acc): Likewise.
(move_acc_to_reg): Likewise.
(move_to_x): Likewise.
(move_to_hl): Likewise.
(move_to_de): Likewise.
* config/rs6000/rs6000.c (emit_frame_save): Likewise.
(rs6000_emit_savres_rtx): Likewise.
(rs6000_emit_prologue): Likewise.
* reorg.c (update_reg_unused_notes): Likewise.
* rtl.h (remove_note): Adjust prototype.
* rtlanal.c (remove_note): Make argument type rtx_insn *.

From-SVN: r241771

split up some variables to use rtx_insn * more

gcc/ChangeLog:

2016-11-01  Trevor Saunders  <tbsaunde+gcc@tbsaunde.org>

* config/alpha/alpha.c (alpha_legitimize_address_1): Split up
variables so some can be rtx_insn *.
(alpha_emit_xfloating_libcall): Likewise.
* config/mips/mips.c (mips_call_tls_get_addr): Likewise.
(mips_legitimize_tls_address): Likewise.
* optabs.c (expand_binop): Likewise.
* reload1.c (gen_reload): Likewise.

From-SVN: r241770

split up the trial variable in reorg.c:relax_delay_slots to use rtx_insn * more

gcc/ChangeLog:

2016-11-01  Trevor Saunders  <tbsaunde+gcc@tbsaunde.org>

* reorg.c (relax_delay_slots): Split up the trial variable.

From-SVN: r241769

use rtx_insn * in various places where it is obvious

gcc/ChangeLog:

2016-11-01  Trevor Saunders  <tbsaunde+gcc@tbsaunde.org>

* config/arc/arc.c (arc_emit_call_tls_get_addr): Make the type
of variables rtx_insn *.
* config/arm/arm.c (arm_call_tls_get_addr): Likewise.
(legitimize_tls_address): Likewise.
* config/bfin/bfin.c (hwloop_optimize): Likewise.
(bfin_gen_bundles): Likewise.
* config/c6x/c6x.c (reorg_split_calls): Likewise.
(c6x_reorg): Likewise.
* config/frv/frv.c (frv_reorder_packet): Likewise.
* config/i386/i386.c (ix86_split_idivmod): Likewise.
* config/ia64/ia64.c (ia64_expand_compare): Likewise.
* config/m32c/m32c.c (m32c_prepare_shift): Likewise.
* config/mn10300/mn10300.c: Likewise.
* config/rl78/rl78.c: Likewise.
* config/s390/s390.c (s390_fix_long_loop_prediction): Likewise.
* config/sh/sh-mem.cc (sh_expand_cmpstr): Likewise.
(sh_expand_cmpnstr): Likewise.
(sh_expand_strlen): Likewise.
(sh_expand_setmem): Likewise.
* config/sh/sh.md: Likewise.
* emit-rtl.c (emit_pattern_before): Likewise.
* except.c: Likewise.
* final.c: Likewise.
* jump.c: Likewise.

From-SVN: r241768

MAINTAINERS (Write After Approval): Update email address,

2016-11-01  Josh Conner  <joshconner@google.com>

* MAINTAINERS (Write After Approval): Update email address,

From-SVN: r241767

c-cppbuiltin.c (c_cpp_builtins): Update __cpp_inheriting_constructors.

* c-cppbuiltin.c (c_cpp_builtins): Update
__cpp_inheriting_constructors.

From-SVN: r241766

Implement P0136R1, Rewording inheriting constructors.

gcc/c-family/
* c.opt (-fnew-inheriting-ctors): New.
* c-opts.c: Default to on for ABI 11+.
gcc/cp/
* call.c (enum rejection_reason_code): Add rr_inherited_ctor.
(inherited_ctor_rejection): New.
(add_function_candidate): Reject inherited ctors for copying.
(enforce_access): Use strip_inheriting_ctors.
(print_z_candidate): Likewise.  Handle rr_inherited_ctor.
(convert_like_real): Avoid copying inheriting ctor parameters.
(build_over_call): Likewise.  A base ctor inheriting from vbase
has no parms.  Sorry about varargs.
(joust): A local constructor beats inherited with the same convs.
* class.c (add_method): Handle hiding inheriting ctors.
(one_inherited_ctor): Handle new semantics.
(add_implicitly_declared_members): Pass using_decl down.
(build_clone): A base ctor inheriting from vbase has no parms.
* cp-tree.h (DECL_INHERITED_CTOR): Store this instead of the base.
(SET_DECL_INHERITED_CTOR): Likewise.
(DECL_INHERITED_CTOR_BASE): Adjust.
* constexpr.c: Adjust.
* error.c (dump_function_decl): Decorate inheriting ctors.
* init.c (emit_mem_initializers): Suppress access control in
inheriting ctor.
* mangle.c (write_special_name_constructor): Handle new inheriting
ctor mangling.
* method.c (strip_inheriting_ctors, inherited_ctor_binfo)
(ctor_omit_inherited_parms, binfo_inherited_from): New.
(synthesized_method_walk): Use binfo_inherited_from.  Suppress
access control in inheriting ctor.
(deduce_inheriting_ctor): Deleted if ambiguous ctor inheritance.
(maybe_explain_implicit_delete): Explain ambigous ctor inheritance.
(add_one_base_init, do_build_copy_constructor): Adjust.
(locate_fn_flags, explain_implicit_non_constexpr): Adjust.
(implicitly_declare_fn): Adjust.
(get_inherited_ctor): Remove.
* name-lookup.c (do_class_using_decl): Check for indirect ctor
inheritance.
* optimize.c (cdtor_comdat_group): Adjust for new mangling.
(maybe_clone_body): Handle omitted parms in base clone.
(maybe_thunk_body): Don't thunk if base clone omits parms.
* pt.c (tsubst_decl): Adjust.
(instantiate_template_1): Suppress access control in inheriting
ctor.
(fn_type_unification): Do deduction with inherited ctor.
* tree.c (special_function_p): Adjust.
gcc/
* tree-inline.c (copy_tree_body_r): Only copy the taken branch of
a COND_EXPR with constant condition.
libiberty/
* cp-demangle.c (d_ctor_dtor_name): Handle inheriting constructor.

From-SVN: r241765

Daily bump.

From-SVN: r241764

* gfortran.dg/pr70937.f90: require-effective-target lto.

From-SVN: r241761

libiberty: Fix memory leak in ada_demangle when symbol cannot be demangled.

When a symbol cannot be demangled in ada_demangle a new demangled VEC
will be allocated without deleting the demangled VEC already in use.

Running testsuite/test-demangle under valgrind will show the leak for
this entry in testsuite/demangle-expected:

    # Elaborated flag (not demangled)
    --format=gnat
    x_E
    <x_E>

 11 bytes in 1 blocks are definitely lost in loss record 1 of 1
    at 0x4C27BE3: malloc (vg_replace_malloc.c:299)
    by 0x413FE7: xmalloc (xmalloc.c:148)
    by 0x4025EC: ada_demangle (cplus-dem.c:930)
    by 0x402C59: cplus_demangle (cplus-dem.c:892)
    by 0x400FEC: main (test-demangle.c:317)

libiberty/ChangeLog:

* cplus-dem.c (ada_demangle): Initialize demangled to NULL and
XDELETEVEC demangled when unknown.

From-SVN: r241760

dwarf2out.c (gen_variable_die): Remove again origin_die variable and its initialization.

* dwarf2out.c (gen_variable_die): Remove again origin_die variable
and its initialization.

From-SVN: r241759

Fix dwarf2out.c non-HAVE_AS_LEB128 bootstrap failure

* dwarf2out.c (output_rnglists): Wrap basebuf, len in
HAVE_AS_LEB128.

From-SVN: r241758

libgo: fix GOARCH_PHYSPAGESIZE for ia64

    From Andreas Schwab.

    Reviewed-on: https://go-review.googlesource.com/32537

From-SVN: r241757

re PR fortran/78178 (ICE in WHERE statement with diagnostic)

2016-11-01  Thomas Koenig  <tkoenig@gcc.gnu.org>

PR fortran/78178
* match.c (match_simple_where):  Fill in locus for assigment
in simple WHERE statement.

2016-11-01  Thomas Koenig  <tkoenig@gcc.gnu.org>

PR fortran/69544
* gfortran.dg/where_6.f90:  New test.

From-SVN: r241756

dwarf2out.c (add_name_and_src_coords_attributes): Add NO_LINKAGE_NAME argument, don't call add_linkage_name if it is true.

* dwarf2out.c (add_name_and_src_coords_attributes): Add NO_LINKAGE_NAME
argument, don't call add_linkage_name if it is true.
(gen_variable_die): For C++ inline static data members, consider the
initial call when old_die is NULL to be declaration and call
add_name_and_src_coords_attributes in that case with true as
NO_LINKAGE_NAME.  Add DW_AT_inline attribute if needed.
(gen_member_die): For C++ inline static data members, emit a
definition DIE right away in DW_TAG_compile_unit context.
cp/
* cp-objcp-common.c (cp_decl_dwarf_attribute): Handle DW_AT_inline.
testsuite/
* g++.dg/debug/dwarf2/inline-var-1.C: New test.

From-SVN: r241753

re PR target/78166 (hash.c:1887:1: error: unrecognizable insn)

PR target/78166
* config/pa/pa.md: Add new shift/add patterns to handle
(plus (mult (reg) (mem_shadd_operand)) (reg)) source operand.

From-SVN: r241749

xtensa: Fix PR target/78118

It started failing after the following commit: 32e90dc6a0cda45 ("PR
rtl-optimization/61047").

The change that made xtensa backend go ICE looks completely unrelated,
and indeed, the issue is caused by the side effect of
compute_frame_size() function call hidden in the
INITIAL_ELIMINATION_OFFSET macro. This call updates the value of the
xtensa_current_frame_size static variable, used in "return" instruction
predicate. Prior to the change the value of xtensa_current_frame_size was
set to 0 after the end of epilogue generation, which enabled the "return"
instruction for the CALL0 ABI, but after the change the additional
INITIAL_ELIMINATION_OFFSET calls make xtensa_current_frame_size non-zero
and "return" pattern unavailable.

Get rid of the global xtensa_current_frame_size and
xtensa_callee_save_size variables by moving them into the
machine_function structure. Implement predicate for the "return" pattern
as a function. Don't communicate completion of epilogue generation
through zeroing of xtensa_current_frame_size, add explicit epilogue_done
variable to the machine_function structure. Don't update stack frame
layout after the completion of reload.

2016-11-01  Max Filippov  <jcmvbkbc@gmail.com>
gcc/
* config/xtensa/xtensa-protos.h
(xtensa_use_return_instruction_p): New prototype.
* config/xtensa/xtensa.c (xtensa_current_frame_size,
xtensa_callee_save_size): Remove.
(struct machine_function): Add new fields: current_frame_size,
callee_save_size, frame_laid_out and epilogue_done.
(compute_frame_size, xtensa_expand_prologue,
xtensa_expand_epilogue): Replace xtensa_callee_save_size with
cfun->machine->callee_save_size and xtensa_current_frame_size
with cfun->machine->current_frame_size.
(compute_frame_size): Update cfun->machine->frame_laid_out and
don't update frame layout after reload completion.
(xtensa_expand_epilogue): Set cfun->machine->epilogue_done
instead of zeroing xtensa_current_frame_size.
(xtensa_use_return_instruction_p): New function.
* config/xtensa/xtensa.h (xtensa_current_frame_size): Remove
declaration.
(INITIAL_ELIMINATION_OFFSET): Use return value of
compute_frame_size instead of xtensa_current_frame_size value.
* config/xtensa/xtensa.md ("return" pattern): Use new predicate
function xtensa_use_return_instruction_p instead of inline code.

From-SVN: r241748

tree.h (BLOCK_IN_COLD_SECTION_P): Define.

* tree.h (BLOCK_IN_COLD_SECTION_P): Define.
* final.c (final_scan_insn): Set BLOCK_IN_COLD_SECTION_P.
* dwarf2out.c (rnglist_idx): New variable.
(struct dw_ranges): Add label, idx and maybe_new_sec fields.
(DEBUG_RNGLISTS_SECTION): Define.
(ranges_base_label): New variable.
(size_of_die) <case dw_val_class_range_list>: If using
DW_FORM_rnglistx, count size of uleb128 of range list index.
(value_format) <case dw_val_class_range_list>: For
-gdwarf-5 -gsplit-dwarf return DW_FORM_rnglistx.
(output_range_list_offset): Handle -gdwarf-5 .debug_rnglists
offsets.  Multiply dwarf < 5 offsets by 2 * DWARF_ADDR_SIZE.
(add_ranges_num): Remove useless prototype.  Don't multiply
by 2 * DWARF2_ADDR_SIZE.  Add maybe_new_sec argument, adjust
for new fields added to dw_ranges struct.
(add_ranges): Add maybe_new_sec argument and pass it
through to add_ranges_num.
(note_rnglist_head): New function.
(add_ranges_by_labels): Pass true as maybe_new_sec to
add_ranges_num, call note_rnglist_head on the head of the list.
(output_ranges): Add function comment.  Switch to
.debug_ranges section here and emit .Ldebug_ranges0 label.
(index_rnglists, output_rnglists): New functions.
(gen_subprogram_die): Formatting fixes.
(add_high_low_attributes): Don't divide offsets
by 2 * DWARF2_ADDR_SIZE.  Call note_rnglist_head on the
first list element or when pointing into the middle of
a list.  Pass true as second argument to add_ranges on the
first block fragment after cold/hot section switch.
(init_sections_and_labels): For -gdwarf-5 use .debug_rnglists
section instead of .debug_ranges.  Initialize
ranges_base_label if -gdwarf-5 -gsplit-dwarf.
(dwarf2out_finish): For -gdwarf-5 -gsplit-dwarf call
index_rnglists and add DW_AT_rnglists_base attr.  Don't switch
to dwarf_ranges_section here or emit .Ldebug_ranges0 label.
Call output_rnglists for -gdwarf-5.
(dwarf2out_c_finalize): Clear rnglist_idx.

From-SVN: r241747

hashtab.h: Use standard GPLv3 with runtime exception boilerplate.

* hashtab.h: Use standard GPLv3 with runtime exception
boilerplate.

From-SVN: r241746

re PR fortran/69544 (Internal compiler error with -Wall and where)

2016-11-01  Thomas Koenig  <tkoenig@gcc.gnu.org>

PR fortran/69544
* match.c (gfc_match_where):  Fill in locus for assigment
in simple WHERE statement.

2016-11-01  Thomas Koenig  <tkoenig@gcc.gnu.org>

PR fortran/69544
* gfortran.dg/where_5.f90:  New test.

From-SVN: r241745

Mark fallthrough cases with gcc_fallthrough.

* gcc/combine.c (simplify_compare_const): Add gcc_fallthrough.

From-SVN: r241744

runtime: recreate function called by cgo -gccgo

    When using cgo -gccgo calls to C.GoString, C.GoStringN, and C.GoBytes
    are turned into calls to __go_byte_array_to_string and
    __go_string_to_byte_array.  Those functions were removed when the string
    code was copied from Go 1.7, but we still need them for cgo.  While cgo
    should be updated, old versions will exist for some time.

    Reviewed-on: https://go-review.googlesource.com/32474

From-SVN: r241743

libgo: only rebuild package if dependent .gox has changed

    Use mvifdiff and stamp files to track whether a .gox file has changed.
    When package A depends on package B, and we rebuild package B, only
    rebuild package A if package B's .gox file changes.  This is safe
    because when package A imports package B it only reads package B's .gox
    file.  This means that changes that do not affect export information
    will not cause dependent packages to be recompiled.

    Reviewed-on: https://go-review.googlesource.com/32476

From-SVN: r241742

libgo: minor Makefile improvements

    Suggested by Ralph Corderoy.

    Reviewed-on: https://go-review.googlesource.com/32414

From-SVN: r241741

re PR go/78145 (Several go.test tests fail with error: integer constant overflow on 32bit targets)

PR go/78145
    compiler: don't put print/println constants into temporaries

    It's not necessary, and it breaks setting their type to int64/uint64
    when appropriate.

    This fixes GCC PR 78145.

    Reviewed-on: https://go-review.googlesource.com/32475

From-SVN: r241740

* class.c (declared_access): Split out from handle_using_decl.

From-SVN: r241739

* es.po: Update.

From-SVN: r241737

arm-c.c (arm_cpu_builtins): New macro definition.

gcc/

2016-09-30  Bilyan Borisov  <bilyan.borisov@arm.com>
    Tamar Christina <tamar.christina@arm.com>

* config/arm/arm-c.c (arm_cpu_builtins): New macro definition.
* config/arm/arm_neon.h (vmaxnm_f32): New intrinsinc.
(vmaxnmq_f32): Likewise.
(vminnm_f32): Likewise.
(vminnmq_f32): Likewise.
* config/arm/arm_neon_builtins.def (vmaxnm): New builtin.
(vminnm): Likewise.
* config/arm/neon.md (neon_<fmaxmin_op><mode>, VCVTF): New
expander.

gcc/testsuite/

2016-09-30  Bilyan Borisov  <bilyan.borisov@arm.com>

* gcc.target/arm/simd/vmaxnm_f32_1.c: New.
* gcc.target/arm/simd/vmaxnmq_f32_1.c: Likewise.
* gcc.target/arm/simd/vminnm_f32_1.c: Likewise.
* gcc.target/arm/simd/vminnmq_f32_1.c: Likewise.

Co-Authored-By: Tamar Christina <tamar.christina@arm.com>
From-SVN: r241736

Minor tweaks

From-SVN: r241734

target-supports.exp: Normalize order of i?86 and x86_64 targets.

* lib/target-supports.exp: Normalize order of i?86 and x86_64 targets.
Whitespace fixes.
(check_effective_target_vect_cmdline_needed): Check
is-effective-target ia32 for x86 targets.
(check_effective_target_vect_simd_clones): Simplify condition.
(check_effective_target_vect_double): Ditto.
(check_effective_target_vect_aligned_arrays): Check
is-effective-target ia32 for x86 targets.  Simplify condition.
(check_effective_target_vect_multiple_sizes): Simplify condition.
(check_effective_target_sqrt_insn): Add i?86-*-* target.
(check_effective_target_sync_int_128): Simplify condition.
(check_effective_target_sync_int_128_runtime): Ditto.
(check_effective_target_sync_long_long_runtime): Ditto.
(check_effective_target_divmod): Add i?86-*-* target.

From-SVN: r241733

vsx.md (VSX_EXTRACT_FL): New iterator for all binary floating point types supported by the hardware...

[gcc]
2016-10-31  Michael Meissner  <meissner@linux.vnet.ibm.com>

* config/rs6000/vsx.md (VSX_EXTRACT_FL): New iterator for all
binary floating point types supported by the hardware except for
double.
(vsx_xvcvsxwdp_df): Provide scalar result alternative to the
vector instruction for optimizing extracting a SImode from a
V4SImode vector and converting it to floating point.
(vsx_xvcvuxwdp_df): Likewise.
(vsx_extract_si): On ISA 3.0, allow extract target and temporary
registers to be any VSX register.  Move stores to the end of the
constraints.
(vsx_extract_si_<uns>float_df): New combiner pattern and splitter
to optimize extracting a SImode from a V4SImode vector and
converting it to a binary floating point type supported by the
hardware.  Use the vector converts instead of extracting the
element, sign extending it, and then converting it to double.
Other floating point types  than double first convert to double,
then the double is converted to that type.
(vsx_extract_si_<uns>float_<mode>): Likewise.

[gcc/testsuite]
2016-10-31  Michael Meissner  <meissner@linux.vnet.ibm.com>

* gcc.target/powerpc/vsx-extract-4.c: New test.
* gcc.target/powerpc/vsx-extract-5.c: Likewise.

From-SVN: r241731

Daily bump.

From-SVN: r241730

driver-aarch64.c (host_detect_local_cpu): Rewrite handling of part num to handle the case where multiple implementers...

2016-10-31  Andrew Pinski  <apinski@cavium.com>

        * config/aarch64/driver-aarch64.c (host_detect_local_cpu):
        Rewrite handling of part num to handle the case where
        multiple implementers share the same part num.

From-SVN: r241726

dwarf2out.c (DWARF_COMPILE_UNIT_HEADER_SIZE): Adjust for -gdwarf-5.

* dwarf2out.c (DWARF_COMPILE_UNIT_HEADER_SIZE): Adjust for -gdwarf-5.
(DWARF_COMDAT_TYPE_UNIT_HEADER_SIZE): Likewise.

Co-Authored-By: Jakub Jelinek <jakub@redhat.com>
From-SVN: r241725