IPA-CP Alignment lattice cleanup

2015-10-05  Martin Jambor  <mjambor@suse.cz>
    Jan Hubicka  <hubicka@ucw.cz>

* ipa-cp.c (ipcp_alignment_lattice): New type.
(ipcp_param_lattices): Use the above to represent alignment.
(ipcp_alignment_lattice::print): New function.
(print_all_lattices): Use it to print alignment information.
(ipcp_alignment_lattice::top_p): New function.
(ipcp_alignment_lattice::bottom_p): Likewise.
(ipcp_alignment_lattice::set_to_bottom): Likewise.
(ipcp_alignment_lattice::meet_with_1): Likewise.
(ipcp_alignment_lattice::meet_with): Two new overloaded functions.
(set_all_contains_variable): Use set_to_bottom of alignment lattice.
(initialize_node_lattices): Likewise.
(propagate_alignment_accross_jump_function): Work with the new class
for alignment lattices.
(propagate_constants_accross_call): Pass only the alignment lattice to
propagate_alignment_accross_jump_function.
(ipcp_store_alignment_results): Work with the new class for alignment
lattices.

testsuite/
* gcc.dg/ipa/propalign-4.c: New test.
* gcc.dg/ipa/propalign-5.c: Likewise.

Co-Authored-By: Jan Hubicka <hubicka@ucw.cz>
From-SVN: r228491

re PR middle-end/67821 (ICE in duplicate_ssa_name_range_info)

PR tree-optimization/67821
* tree-ssanames.c (duplicate_ssa_name_range_info): Remove an assert.

* gcc.dg/torture/pr67821-2.c: New test.
* gcc.dg/torture/pr67821.c: New test.

From-SVN: r228490

Move runtime transactional memory tests to libitm testsute.

From-SVN: r228489

[PR other/65021] mkoffloads -save-temps handling, and cleanup cleanup

gcc/
PR other/65021
* config/i386/intelmic-mkoffload.c (mkoffload_atexit): Rename
function to...
(mkoffload_cleanup): ... this.  Adjust all users.
(maybe_unlink): Look at save_temps and verbose flags instead of
debug flag.
(main): Parse "-save-temps" flag.
(generate_target_descr_file, generate_target_offloadend_file)
(generate_host_descr_file, prepare_target_image): Pass it on.
* config/nvptx/mkoffload.c (tool_cleanup): Implement.
(mkoffload_cleanup): New function.
(maybe_unlink): Look at save_temps and verbose flags instead of
debug flag.
(main): Instead of calling utils_cleanup, register atexit handler
for mkoffload_cleanup.
(main): Parse "-save-temps" flag.
(compile_native, main): Pass it on.
* lto-wrapper.c (compile_offload_image): Likewise.

From-SVN: r228488

make more gimple_x_ptr accessors const correct

gcc/ChangeLog:

2015-10-05  Trevor Saunders  <tbsaunde+gcc@tbsaunde.org>

* gimple.h (gimple_op_ptr): Require a non const gimple *.
(gimple_assign_lhs_ptr): Likewise.
(gimple_assign_rhs1_ptr): Likewise.
(gimple_assign_rhs2_ptr): Likewise.
(gimple_assign_rhs3_ptr): Likewise.
(gimple_call_lhs_ptr): Likewise.
(gimple_call_fn_ptr): Likewise.
(gimple_call_chain_ptr): Likewise.
(gimple_call_arg_ptr): Likewise.
(gimple_cond_lhs_ptr): Likewise.
(gimple_cond_rhs_ptr): Likewise.
(gimple_switch_index_ptr): Likewise.
(gimple_return_retval_ptr): Likewise.

From-SVN: r228487

remove unused gasm accessors

gcc/ChangeLog:

2015-10-05  Trevor Saunders  <tbsaunde+gcc@tbsaunde.org>

* gimple.h (gimple_asm_input_op_ptr): Remove.
(gimple_asm_output_op_ptr): Likewise.

From-SVN: r228486

remove gimple_location_ptr ()

gcc/ChangeLog:

2015-10-05  Trevor Saunders  <tbsaunde+gcc@tbsaunde.org>

* gimple.h (gimple_location_ptr): Remove.
* tree-vrp.c (check_all_array_refs): Adjust.

From-SVN: r228485

make build_uses store tree * instead of tree

gcc/ChangeLog:

2015-10-05  Trevor Saunders  <tbsaunde+gcc@tbsaunde.org>

* tree-ssa-operands.c (build_uses): store tree * instead of
tree.
(finalize_ssa_uses): Adjust.
(append_use): Likewise.
(verify_ssa_operands): Likewise.

From-SVN: r228484

Add a build_real_truncate helper function

...which simplifies the match.pd patterns I'm about to add.

Bootstrapped & regression-tested on x86_64-linux-gnu.

gcc/
* real.h (build_real_truncate): Declare.
* tree.c (build_real_truncate): New function.
(strip_float_extensions): Use it.
* builtins.c (fold_builtin_cabs, fold_builtin_sqrt, fold_builtin_cbrt)
(fold_builtin_hypot, fold_builtin_pow): Likewise.
* match.pd: Likewise.

From-SVN: r228483

[AArch64] Fix output assembly bug under TLSIE ILP32

2015-10-05 James Greenhalgh <james.greenhalgh@arm.com>
   Jiong Wang  <jiong.wang@arm.com>

* config/aarch64/aarch64.md (tlsie_tiny_sidi): Replace "<w>" with "w".

Co-Authored-By: Jiong Wang <jiong.wang@arm.com>
From-SVN: r228482

cpuinfo.c (get_intel_cpu): Detect "skylake-avx512".

libgcc/
* config/i386/cpuinfo.c (get_intel_cpu): Detect "skylake-avx512".
gcc/testsuite/
* gcc.target/i386/builtin_target.c: Add check for "skylake-avx512".

From-SVN: r228481

Cleanup directives.

From-SVN: r228479

Remove REAL_VALUE_FROM_CONST_DOUBLE

To maintain symmetry after the previous removal of
CONST_DOUBLE_FROM_REAL_VALUE, this patch also gets rid of
REAL_VALUE_FROM_CONST_DOUBLE.  All the macro did was copy the
contents of CONST_DOUBLE_REAL_VALUE into a temporary real_value
structure.  In many cases there was no need for this temporary
and we could simply use the CONST_DOUBLE_REAL_VALUE directly.
For that reason this patch is less automatic than the others.

Bootstrapped & regression-tested on x86_64-linux-gnu.  Also tested by
building one target per CPU directory and checking that there were
no new warnings and no changes in testsuite output at -O2.

gcc/
* real.h (REAL_VALUE_FROM_CONST_DOUBLE): Delete.
* config/aarch64/aarch64.c (aarch64_float_const_zero_rtx_p)
(aarch64_print_operand, aarch64_float_const_representable_p)
(aarch64_output_simd_mov_immediate): Use CONST_DOUBLE_REAL_VALUE
instead of REAL_VALUE_FROM_CONST_DOUBLE.
* config/arc/arc.c (arc_print_operand): Likewise.
* config/arm/arm.c (arm_const_double_rtx, vfp3_const_double_index)
(neon_valid_immediate, arm_print_operand, arm_emit_fp16_const)
(vfp3_const_double_for_fract_bits, vfp3_const_double_for_bits):
Likewise.
* config/arm/arm.md (*arm32_movhf, consttable_4, consttable_8)
(consttable_16): Likewise.
* config/arm/vfp.md (*movhf_vfp_neon, *movhf_vfp): Likewise.
* config/avr/avr.c (avr_print_operand): Likewise.
* config/bfin/bfin.md: Likewise (in a define_split).
* config/c6x/c6x.md: Likewise (in a define_split).
* config/cr16/cr16.c (cr16_const_double_ok): Likewise.
(cr16_print_operand): Likewise.
* config/cris/cris.c (cris_print_operand): Likewise.
* config/epiphany/epiphany.c (epiphany_print_operand): Likewise.
* config/fr30/fr30.c (fr30_print_operand): Likewise.
(fr30_const_double_is_zero): Likewise.
* config/frv/frv.c (frv_print_operand, output_move_single): Likewise.
* config/frv/frv.md: Likewise (in a define_split).
* config/frv/predicates.md (int_2word_operand): Likewise.
* config/h8300/h8300.c (h8300_print_operand): Likewise.
* config/i386/i386.c (standard_80387_constant_p): Likewise.
(ix86_print_operand, ix86_split_to_parts): Likewise.
* config/i386/i386.md: Likewise (in a define_split).
* config/ia64/ia64.c (ia64_split_tmode, ia64_print_operand): Likewise.
* config/iq2000/iq2000.md (movsf_lo_sum, movsf_high): Likewise.
* config/m32r/m32r.c (easy_df_const, m32r_print_operand): Likewise.
* config/m68k/m68k.c (handle_move_double, standard_68881_constant_p)
(print_operand): Likewise.
* config/m68k/m68k.md (movsf_cf_hard, movdf_cf_hard): Likewise.
* config/mep/mep.md: Likewise (in define_split).
* config/microblaze/microblaze.c (microblaze_const_double_ok)
(print_operand): Likewise.
* config/mips/mips.md (consttable_float): Likewise.
* config/mmix/mmix.c (mmix_intval): Likewise.
* config/mn10300/mn10300.c (mn10300_print_operand): Likewise.
* config/nvptx/nvptx.c (nvptx_print_operand): Likewise.
* config/pa/pa.c (pa_singlemove_string): Likewise.
* config/pdp11/pdp11.c (pdp11_expand_operands): Likewise.
(pdp11_asm_print_operand, legitimate_const_double_p): Likewise.
* config/rs6000/rs6000.c (num_insns_constant, rs6000_emit_cmove)
(output_toc): Likewise.
* config/rs6000/rs6000.md: Likewise (in define_splits).
* config/rx/rx.c (rx_print_operand): Likewise.
* config/s390/s390.c (s390_output_pool_entry): Likewise.
* config/sh/sh.c (fp_zero_operand, fp_one_operand): Likewise.
* config/sh/sh.md (consttable_sf, consttable_df): Likewise
(and also in define_splits).
* config/sparc/sparc.c (fp_sethi_p, fp_mov_p): Likewise.
(fp_high_losum_p): Likewise.
* config/sparc/sparc.md (*movsf_insn, *movsf_lo_sum): Likewise.
(*movsf_high): Likewise.
* config/spu/spu.c (const_double_to_hwint): Likewise.
* config/v850/v850.c (const_double_split): Likewise.
* config/vax/vax.c (vax_float_literal): Likewise.
* config/visium/visium.c (visium_expand_copysign): Likewise.
* config/visium/visium.md: Likewise (in define_split).
* config/xtensa/predicates.md (const_float_1_operand): Likewise.
* config/xtensa/xtensa.c (print_operand): Likewise.
(xtensa_output_literal): Likewise.
* cprop.c (implicit_set_cond_p): Likewise.
* dwarf2out.c (insert_float): Likewise.
* expmed.c (expand_mult, make_tree): Likewise.
* expr.c (compress_float_constant): Likewise.
* rtlanal.c (split_double): Likewise.
* simplify-rtx.c (avoid_constant_pool_reference): Likewise.
(simplify_const_unary_operation, simplify_binary_operation_1)
(simplify_const_binary_operation): Likewise.
(simplify_const_relational_operation): Likewise.
* varasm.c (output_constant_pool_2): Likewise.

From-SVN: r228478

Remove remaining uses of CONST_DOUBLE_FROM_REAL_VALUE

This patch replaces all uses of CONST_DOUBLE_FROM_REAL_VALUE
with the already-existing const_double_from_real_value.

Bootstrapped & regression-tested on x86_64-linux-gnu.  Also tested by
building one target per CPU directory and checking that there were
no new warnings and no changes in testsuite output at -O2.

gcc/
* real.h (CONST_DOUBLE_ATOF): Use const_double_from_real_value
instead of CONST_DOUBLE_FROM_REAL_VALUE.
(CONST_DOUBLE_FROM_REAL_VALUE): Delete.
* config/c6x/c6x.md (divsf3, divdf3): Use const_double_from_real_value
instead of CONST_DOUBLE_FROM_REAL_VALUE.
* config/epiphany/epiphany.md (fixuns_truncsfsi2): Likewise.
* config/i386/i386.c (standard_80387_constant_rtx): Likewise.
(ix86_expand_builtin, ix86_emit_i387_log1p, ix86_emit_i387_round)
(ix86_emit_swsqrtsf): Likewise.
* config/ia64/ia64.c (ia64_expand_builtin): Likewise.
* config/mips/mips.md (fixuns_truncdfsi2, fixuns_truncdfdi2)
(fixuns_truncsfsi2, fixuns_truncsfdi2): Likewise.
* config/pa/pa.c (pa_expand_builtin): Likewise.
* config/rs6000/rs6000.c (rs6000_load_constant_and_splat): Likewise.
(rs6000_scale_v2df): Likewise.
* config/rs6000/rs6000.md (*cmptf_internal2): Likewise.
* config/s390/s390.md (fixuns_truncdddi2, fixuns_trunctddi2)
(fixuns_trunc<BFP:mode><GPR:mode>2): Likewise.
* config/s390/vx-builtins.md (vec_ctd_s64, vec_ctd_u64, vec_ctsl)
(vec_ctul): Likewise.
* config/sparc/sparc.c (sparc_emit_fixunsdi): Likewise.
* config/spu/spu.c (hwint_to_const_double, spu_float_const): Likewise.
* config/spu/spu.md (floatunsdisf2, floatunstisf2): Likewise.
* cse.c (fold_rtx): Likewise.
* emit-rtl.c (immed_double_const): Likewise (in comments).
(init_emit_once): Likewise.
* expr.c (compress_float_constant, expand_expr_real_1)
(const_vector_from_tree): Likewise.
* optabs.c (expand_float, expand_fix): Likewise.
* reg-stack.c (reg_to_stack): Likewise.
* simplify-rtx.c (avoid_constant_pool_reference): Likewise.
(simplify_const_unary_operation, simplify_binary_operation_1)
(simplify_const_binary_operation, simplify_relational_operation)
(simplify_immed_subreg): Likewise.

From-SVN: r228477

Remove remaining uses of REAL_ARITHMETIC

This patch replaces all remaining uses of the old target macro
REAL_ARITHMETIC with calls to the (now generic) real_arithmetic
function.

Bootstrapped & regression-tested on x86_64-linux-gnu.  Also tested by
building one target per CPU directory and checking that there were
no new warnings and no changes in testsuite output at -O2.

gcc/ada/
* gcc-interface/trans.c (convert_with_check): Use real_arithmetic
instead of REAL_ARITHMETIC.

gcc/
* doc/tm.texi.in (REAL_ARITHMETIC): Delete.
* doc/tm.texi: Regenerate.
* real.h (REAL_ARITHMETIC): Delete.
* config/i386/i386.c (ix86_expand_lround, ix86_expand_round)
(ix86_expand_round_sse4): Use real_arithmetic instead of
REAL_ARITHMETIC.
* config/i386/sse.md (round<mode>2): Likewise.
* rtl.h (rtx_to_tree_code): Likewise (in comment).
* explow.c (rtx_to_tree_code): Likewise (in comment).
* match.pd: Likewise.
* simplify-rtx.c (simplify_binary_operation_1): Likewise.
* tree-ssa-math-opts.c (representable_as_half_series_p): Likewise.
(expand_pow_as_sqrts): Likewise.
* tree-pretty-print.c (dump_generic_node): Remove code that
was conditional on REAL_ARITHMETIC being undefined.

From-SVN: r228476

Replace REAL_VALUES_LESS with real_less

This patch continues the removal of real-related macros by
replacing REAL_VALUES_LESS with real_less.

Bootstrapped & regression-tested on x86_64-linux-gnu.  Also tested by
building one target per CPU directory and checking that there were
no new warnings and no changes in testsuite output at -O2.

gcc/ada/
* gcc-interface/trans.c (convert_with_check): Use real_less instead
of REAL_VALUES_LESS.

gcc/
* doc/tm.texi.in (REAL_VALUES_LESS): Delete.
* doc/tm.texi: Regenerate.
* real.h (real_less): Declare.
(REAL_VALUES_LESS): Delete.
* real.c (real_less): New function.
(real_compare): Use it.
* config/m68k/m68k.c (floating_exact_log2): Use real_less instead
of REAL_VALUES_LESS.
* config/microblaze/microblaze.c (microblaze_const_double_ok):
Likewise.
* fold-const.c (fold_convert_const_int_from_real): Likewise.
* simplify-rtx.c (simplify_const_unary_operation): Likewise.
(simplify_const_relational_operation): Likewise.
* tree-call-cdce.c (check_pow): Likewise.
(gen_conditions_for_pow_cst_base): Likewise.

From-SVN: r228475

Remove remaining uses of REAL_VALUES_IDENTICAL

This patch continues the removal of real-related macros.
We already had both the old-style REAL_VALUES_IDENTICAL and the
new-style real_identical, so this patch replaces all remaining
uses of the former with the latter.

Bootstrapped & regression-tested on x86_64-linux-gnu.  Also tested by
building one target per CPU directory and checking that there were
no new warnings and no changes in testsuite output at -O2.

gcc/
* real.h (REAL_VALUES_IDENTICAL): Delete.
* config/m68k/m68k.c (standard_68881_constant_p): Use real_identical
instead of REAL_VALUES_IDENTICAL.
* fold-const.c (operand_equal_p): Likewise.
* ipa-icf.c (sem_variable::equals): Likewise.
* tree-complex.c (some_nonzerop): Likewise.
(expand_complex_multiplication): Likewise.
* tree.c (simple_cst_equal): Likewise.
* varasm.c (compare_constant): Likewise.
---
 gcc/config/m68k/m68k.c | 5 ++---
 gcc/fold-const.c       | 3 +--
 gcc/ipa-icf.c          | 4 ++--
 gcc/real.h             | 1 -
 gcc/tree-complex.c     | 4 ++--
 gcc/tree.c             | 2 +-
 gcc/varasm.c           | 2 +-
 7 files changed, 9 insertions(+), 12 deletions(-)

From-SVN: r228474

Replace REAL_VALUES_EQUAL with real_equal

Richard B suggested we should replace dconsthalf etc. with
dconst<1, 2> ().  When I tried that, the extra comma caused problems
with some lingering uses of the old target macros for handling reals
(e.g. REAL_ARITHMETIC instead of real_arithmetic), since the constant
was then treated as two macro parameters.  It would have been possible
to add an extra level of brackets to avoid this, but I thought I might
as well take the opportunity to remove the macros instead.  (Note that
I'm only removing macros that caused a problem directly, or are closely
related to ones that did.)

This first patch replaces REAL_VALUES_EQUAL with a real_equal function.
The prototype is the same as for real_identical, which has already
undergone a half-transition in this direction.

Bootstrapped & regression-tested on x86_64-linux-gnu.  Also tested by
building one target per CPU directory and checking that there were
no new warnings and no changes in testsuite output at -O2.

gcc/c-family/
* c-lex.c (interpret_float): Use real_equal instead of
REAL_VALUES_EQUAL.

gcc/c/
* c-typeck.c (c_tree_equal): Use real_equal instead of
REAL_VALUES_EQUAL.

gcc/cp/
* tree.c (cp_tree_equal): Use real_equal instead of
REAL_VALUES_EQUAL.

gcc/
* real.h (real_equal): Declare.
(REAL_VALUES_EQUAL): Delete.
* real.c (real_equal): New function.
(real_compare): Use it.
* doc/tm.texi.in (REAL_VALUES_EQUAL): Delete.
* doc/tm.texi: Regenerate.
* builtins.c (fold_builtin_pow, fold_builtin_load_exponent): Use
real_equal instead of REAL_VALUES_EQUAL.
* config/aarch64/aarch64.c (aarch64_float_const_zero_rtx_p): Likewise.
* config/arm/arm.c (arm_const_double_rtx, neon_valid_immediate)
(fp_const_from_val): Likewise.
* config/fr30/fr30.c (fr30_const_double_is_zero): Likewise.
* config/m68k/m68k.c (standard_68881_constant_p): Likewise.
(floating_exact_log2): Likewise.
* config/sh/sh.c (fp_zero_operand, fp_one_operand): Likewise.
* config/vax/vax.c (vax_float_literal): Likewise.
* config/xtensa/predicates.md (const_float_1_operand): Likewise.
* cprop.c (implicit_set_cond_p): Likewise.
* expmed.c (expand_mult): Likewise.
* fold-const.c (const_binop): Likewise.
* simplify-rtx.c (simplify_binary_operation_1): Likewise.
(simplify_const_binary_operation): Likewise.
(simplify_const_relational_operation): Likewise.
* tree-call-cdce.c (check_pow): Likewise.
(gen_conditions_for_pow_cst_base): Likewise.
* tree-inline.c (estimate_num_insns): Likewise.
* tree-ssa-dom.c (record_equality): Likewise.
* tree-ssa-math-opts.c (representable_as_half_series_p): Likewise.
(gimple_expand_builtin_pow): Likewise.
(pass_optimize_widening_mul::execute): Likewise.
* tree-ssa-uncprop.c (associate_equivalences_with_edges): Likewise.
* tree-vect-patterns.c (vect_recog_pow_pattern): Likewise.
* tree.c (real_zerop, real_onep, real_minus_onep): Likewise.
---
 gcc/builtins.c                  | 14 +++++++-------
 gcc/c-family/c-lex.c            |  6 +++---
 gcc/c/c-typeck.c                |  2 +-
 gcc/config/aarch64/aarch64.c    |  2 +-
 gcc/config/arm/arm.c            |  6 +++---
 gcc/config/fr30/fr30.c          |  2 +-
 gcc/config/m68k/m68k.c          |  6 +++---
 gcc/config/sh/sh.c              |  4 ++--
 gcc/config/vax/vax.c            |  4 ++--
 gcc/config/xtensa/predicates.md |  2 +-
 gcc/cp/tree.c                   |  2 +-
 gcc/cprop.c                     |  2 +-
 gcc/doc/tm.texi                 |  7 -------
 gcc/doc/tm.texi.in              |  7 -------
 gcc/expmed.c                    |  2 +-
 gcc/fold-const.c                |  2 +-
 gcc/real.c                      | 10 +++++++++-
 gcc/real.h                      |  4 ++--
 gcc/simplify-rtx.c              | 16 ++++++++--------
 gcc/tree-call-cdce.c            |  4 ++--
 gcc/tree-inline.c               |  5 +++--
 gcc/tree-ssa-dom.c              |  2 +-
 gcc/tree-ssa-math-opts.c        | 16 ++++++++--------
 gcc/tree-ssa-uncprop.c          |  2 +-
 gcc/tree-vect-patterns.c        |  4 ++--
 gcc/tree.c                      |  6 +++---
 26 files changed, 67 insertions(+), 72 deletions(-)

From-SVN: r228473

re PR ipa/67783 (quadratic time consumption in IPA inlining with -O1 and higher)

2015-10-05  Richard Biener  <rguenther@suse.de>

PR ipa/67783
* ipa-inline-analysis.c (estimate_function_body_sizes): Only
consider loop header PHI defs as IVs.

From-SVN: r228472

tree-ssa-pre.c (create_component_ref_by_pieces_1): Move call handling ...

2015-10-05  Richard Biener  <rguenther@suse.de>

* tree-ssa-pre.c (create_component_ref_by_pieces_1): Move
call handling ...
(create_expression_by_pieces): ... here and build GIMPLE
calls directly.  Use gimple_build API and avoid force_gimple_operand.
(insert_into_preds_of_block): Simplify.
(do_regular_insertion): Add comment.

From-SVN: r228471

[Patch ARM/ AArch64] Fix typo in vcvt_f16.c testcase .

This test worked by accident. While looking at why this was failing randomly in my builds, I discovered a bug in the way in which the testcases were written up in this case.

2015-10-05  Ramana Radhakrishnan  <ramana.radhakrishnan@arm.com>

* gcc.target/aarch64/advsimd-intrinsics/vcvt_f16.c (TEST_MSG): Fix typo.
(exec_vcvt): Add comments.

From-SVN: r228470

re PR c++/67844 (Cannot make tuple of class with template constructor)

2015-10-05  Ville Voutilainen  <ville.voutilainen@gmail.com>

PR 67844.
* include/std/tuple (_TC::_NonNestedTuple): Eagerly reject
conversions from tuple types same as the target tuple.
* testsuite/20_util/tuple/67844.cc: New.
* testsuite/20_util/tuple/cons/nested_tuple_construct.cc: Add
a missing copyright header.

From-SVN: r228468

Daily bump.

From-SVN: r228465

Implement N4514, C++ Extensions for Transactional Memory.

gcc/
* builtins.def (BUILT_IN_ABORT): Add transaction_pure attribute.
gcc/c-family/
* c-common.c (c_common_reswords): Add C++ TM TS keywords.
(c_common_attribute_table): Add transaction_safe_dynamic.
transaction_safe now affects type identity.
(handle_tm_attribute): Handle transaction_safe_dynamic.
* c-common.h (enum rid): Add RID_ATOMIC_NOEXCEPT,
RID_ATOMIC_CANCEL, RID_SYNCHRONIZED.
(OBJC_IS_CXX_KEYWORD): Add RID_SYNCHRONIZED.
(D_TRANSMEM): New.
* c-cppbuiltin.c (c_cpp_builtins): Define __cpp_transactional_memory.
* c-pretty-print.c (pp_c_attributes_display): Don't print
transaction_safe in C++.
gcc/c/
* c-parser.c (c_lex_one_token): Handle @synchronized.
* c-decl.c (match_builtin_function_types): A declaration of a built-in
can change whether the function is transaction_safe.
gcc/cp/
* cp-tree.h (struct cp_declarator): Add tx_qualifier field.
(BCS_NORMAL, BCS_TRANSACTION): New enumerators.
* lex.c (init_reswords): Limit TM kewords to -fgnu-tm.
* parser.c (cp_lexer_get_preprocessor_token): Fix @synchronized.
(make_call_declarator): Take tx_qualifier.
(cp_parser_tx_qualifier_opt): New.
(cp_parser_lambda_declarator_opt): Use it.
(cp_parser_direct_declarator): Likewise.
(cp_parser_statement): Handle atomic_noexcept, atomic_cancel.
(cp_parser_compound_statement): Change in_try parameter to bcs_flags.
(cp_parser_std_attribute): Map optimize_for_synchronized to
transaction_callable.
(cp_parser_transaction): Take the token.  Handle atomic_noexcept.
* lambda.c (maybe_add_lambda_conv_op): Handle transaction-safety.
* call.c (enum conversion_kind): Add ck_tsafe.
(standard_conversion): Handle transaction-safety conversion.
(convert_like_real, resolve_address_of_overloaded_function): Likewise.
(check_methods): Diagnose transaction_safe_dynamic on non-virtual
function.
(look_for_tm_attr_overrides): Don't inherit transaction_safe_dynamic.
* cvt.c (tx_safe_fn_type_p, tx_unsafe_fn_variant)
(can_convert_tx_safety): New.
* typeck.c (composite_pointer_type): Handle transaction-safety.
* name-lookup.h (enum scope_kind): Add sk_transaction.
* name-lookup.c (begin_scope): Handle it.
* semantics.c (begin_compound_stmt): Pass it.
* decl.c (check_previous_goto_1): Check it.
(struct named_label_entry): Add in_transaction_scope.
(poplevel_named_label_1): Set it.
(check_goto): Check it.
(duplicate_decls): A specialization can be transaction_safe
independently of its template.
(grokdeclarator): Handle tx-qualifier.
* rtti.c (ptr_initializer): Handle transaction-safe.
* search.c (check_final_overrider): Check transaction_safe_dynamic.
Don't check transaction_safe.
* mangle.c (write_function_type): Mangle transaction_safe here.
(write_CV_qualifiers_for_type): Not here.
(write_type): Preserve transaction_safe when stripping attributes.
* error.c (dump_type_suffix): Print transaction_safe.
libiberty/
* cp-demangle.c (d_cv_qualifiers): Dx means transaction_safe.
(cplus_demangle_type): Let d_cv_qualifiers handle it.
(d_dump, d_make_comp, has_return_type, d_encoding)
(d_count_templates_scopes, d_print_comp_inner)
(d_print_mod_list, d_print_mod, d_print_function_type)
(is_ctor_or_dtor): Handle DEMANGLE_COMPONENT_TRANSACTION_SAFE.

From-SVN: r228462

re PR rtl-optimization/67447 (ICE in extract_constrain_insn (reload_cse_simplify_operands): insn does not satisfy its constraints)

PR rtl-optimization/67447
* gcc.target/i386/pr67447.c: New test.

From-SVN: r228461

i386.c (ix86_nsaved_regs): Use GENERAL_REGNO_P to check for general register.

* config/i386/i386.c (ix86_nsaved_regs): Use GENERAL_REGNO_P to
check for general register.
(ix86_emit_save_regs): Ditto.
(ix86_emit_save_regs_using_mov): Ditto.
(ix86_emit_restore_regs_using_pop): Ditto.
(ix86_emit_restore_regs_using_mov): Ditto.

From-SVN: r228460

vect-pack-trunc-1.c: Require avx512bw effective target.

* gcc.target/i386/vect-pack-trunc-1.c: Require avx512bw
effective target.
* gcc.target/i386/vect-pack-trunc-2.c: Ditto.
* gcc.target/i386/vect-perm-even-1.c: Ditto.
* gcc.target/i386/vect-perm-odd-1.c: Ditto.
* gcc.target/i386/vect-unpack-1.c: Ditto.
* gcc.target/i386/vect-unpack-2.c: Ditto.

From-SVN: r228459

Remove dead code

gcc/fortran/
* match.c (gfc_match_common): Remove dead variable old_blank_common.

From-SVN: r228458

Fix fortran common-related error recovery ICE.

Upon reverting a symbol in a common block (after throwing an error),
the compiler was ICEing because the symbol's common_block field was set,
but the symbol was not in the common block's list of symbols.

Fixed by both adding the symbol to the common block list and setting
the symbol's common_block field at the same time.
Furthermore, the gfc_add_in_common call is delayed and its result is
ignored, so that its error messages are ignored and the compiler has
the opportunity to give a better error message.
Another gfc_add_in_common call is added later during resolution
to emit again the missing errors.

PR fortran/67758
gcc/fortran/
* match.c (gfc_match_common): Delay the common_block pointer
assignment after error checking.
Delay the call to gfc_add_in_common attribute after the handling
of array specs.
* resolve.c (resolve_common_vars): Call gfc_add_in_common again.
gcc/testsuite/
* gfortran.dg/common_24.f: New.

From-SVN: r228457

Factor access to the common block head symbol

* resolve.c (resolve_common_vars): Move access to the common
block's head symbol inside the function.
(resolve_common_blocks, resolve_types): Update callers.

From-SVN: r228456

Daily bump.

From-SVN: r228454

Makefile.in (insn-latencytab.o): Remove -Wno-duplicated-cond.

* Makefile.in (insn-latencytab.o): Remove -Wno-duplicated-cond.
(insn-dfatab.o): Likewise.

From-SVN: r228451

xtensa: add uclinux support

2015-10-03  Max Filippov  <jcmvbkbc@gmail.com>
gcc/
* config.gcc (xtensa*-*-uclinux*): New configuration.
* config/xtensa/uclinux.h: New file.
* config/xtensa/uclinux.opt: New file.

libgcc/
* config.host (xtensa*-*-uclinux*): New configuration.

From-SVN: r228450

Document value of __cplusplus for C++14

* doc/cpp.texi (Standard Predefined Macros): Document value of
__cplusplus for C++14.

From-SVN: r228448

Remove pretty printing for 'any' with allocators

* python/libstdcxx/v6/printers.py (StdExpAnyPrinter): Remove support
for _Manager_alloc.

From-SVN: r228447

Fix comments saying why headers are included

* src/c++98/locale.cc: Fix comment.
* src/c++98/locale_init.cc: Likewise.

From-SVN: r228446

pr67756.c: Fixed warnings.

2015-10-03  Bernd Edlinger  <bernd.edlinger@hotmail.de>

        * gcc.target/arm/pr67756.c: Fixed warnings.

From-SVN: r228445

Daily bump.

From-SVN: r228442

Fix testsuite failures with --disable-wchar_t

* testsuite/21_strings/basic_string/literals/types.cc: Guard use of
wchar_t with _GLIBCXX_USE_WCHAR_T.
* testsuite/21_strings/basic_string/literals/values.cc: Likewise.
* testsuite/21_strings/basic_string/requirements/citerators.cc:
Likewise.
* testsuite/22_locale/messages/13631.cc: Likewise.
* testsuite/experimental/string_view/literals/types.cc: Likewise.
* testsuite/experimental/string_view/literals/values.cc: Likewise.

From-SVN: r228439

Enable dual ABI for Filesystem library

* src/filesystem/Makefile.am: Add cxx11_abi_sources.
* src/filesystem/Makefile.in: Regenerate.
* src/filesystem/cow-dir.cc: New.
* src/filesystem/cow-ops.cc: New.
* src/filesystem/cow-path.cc: New.
* src/filesystem/dir.cc: Define macro for new ABI.
* src/filesystem/ops.cc: Likewise.
* src/filesystem/path.cc: Likewise.

From-SVN: r228437

minimal.cc: Minor tweaks to match narrow char version.

* testsuite/21_strings/basic_string/allocator/wchar_t/minimal.cc:
Minor tweaks to match narrow char version.

From-SVN: r228436

Fix tests that fail with old std::string

* testsuite/21_strings/basic_string/allocator/char/minimal.cc: Guard
explicit instantiation with check for new ABI.
* testsuite/21_strings/basic_string/allocator/wchar_t/minimal.cc:
Likewise. Use wchar_t as char_type.

From-SVN: r228435

Use noexcept instead of _GLIBCXX_NOEXCEPT

* include/bits/basic_string.h [!_GLIBCXX_USE_CXX11_ABI]
(basic_string::front() const, basic_string::back() const): Use
noexcept instead of _GLIBCXX_NOEXCEPT macro.
(__versa_string::front, __versa_string::back): Likewise.

From-SVN: r228434

* acinclude.m4 (GLIBCXX_ENABLE_DEBUG_FLAGS): Fix comment.

From-SVN: r228433

Help the offload gcc driver find the right assembler

gcc/
* gcc.c (process_command): Use spec_machine rather than
spec_host_machine to build tooldir_prefix2.

From-SVN: r228429

nvptx offloading linking

gcc/
* config/nvptx/mkoffload.c (Kind, Vis): Remove enums.
(Token, Stmt): Remove structs.
(decls, vars, fns): Remove variables.
(alloc_comment, append_stmt, is_keyword): Remove macros.
(tokenize, write_token, write_tokens, alloc_stmt, rev_stmts)
(write_stmt, write_stmts, parse_insn, parse_list_nosemi)
(parse_init, parse_file): Remove functions.
(read_file): Accept a pointer to a length and store into it.
(process): Don't try to parse the input file, just write it out as
a string, but looking for maps.  Also write out the length.
(main): Don't use "-S" to compile PTX code.

libgomp/
* oacc-ptx.h: Remove file, moving its content into...
* config/nvptx/fortran.c: ... here...
* config/nvptx/oacc-init.c: ..., here...
* config/nvptx/oacc-parallel.c: ..., and here.
* config/nvptx/openacc.f90: New file.
* plugin/plugin-nvptx.c: Don't include "oacc-ptx.h".
(link_ptx): Don't link in predefined bits of PTX code.

Co-Authored-By: Bernd Schmidt <bernds@codesourcery.com>
From-SVN: r228418

Allow Filesystem TS to compile without wchar_t

* include/bits/locale_conv.h [!_GLIBCXX_USE_WCHAR_T]
(__do_str_codecvt, __str_codecvt_in, __str_codecvt_out): Enable.
* include/experimental/fs_path.h [!_GLIBCXX_USE_WCHAR_T]
(path::wstring, path::generic_wstring): Disable.
* src/filesystem/path.cc (path::_S_convert_loc)
[!_GLIBCXX_USE_WCHAR_T]: Skip conversion.
* testsuite/experimental/filesystem/path/assign/assign.cc: Check for
wchar_t support.
* testsuite/experimental/filesystem/path/concat/strings.cc: Likewise.
* testsuite/experimental/filesystem/path/construct/range.cc: Likewise.

From-SVN: r228417

testsuite_fs.h (nonexistent_path): Use less generic name for paths.

* testsuite/util/testsuite_fs.h (nonexistent_path): Use less generic
name for paths. Prefer snprintf to sprintf.

From-SVN: r228416

Re: [PATCH] Improve DOM's optimization of control statements

* tree-ssa-dom.c (optimize_stmt): Note when loop structures need
fixups.

From-SVN: r228415

[PR target/67822] OpenMP offloading to nvptx fails

gcc/
PR target/67822
* config/nvptx/mkoffload.c (main): Scan the argument vector for
-fopenmp, and skip generating an offloading image if specified.

From-SVN: r228414

system.h (ROUND_UP): New macro definition.

* system.h (ROUND_UP): New macro definition.
(ROUND_DOWN): Ditto.
* ggc-page.c (ROUND_UP): Remove local macro definition.
(PAGE_ALIGN): Implement using ROUND_UP macro.

* config/i386/i386.h (PUSH_ROUNDING): Implement using ROUND_UP macro.
* config/i386/i386.c (function_arg_advance_64): Use ROUND_UP macro
to align values.
(ix86_compute_frame_layout): Ditto.
(ix86_expand_prologue): Ditto.
(ix86_adjust_stack_and_probe): Use ROUND_DOWN macro
to round down values.
(expand_set_or_movmem_via_rep): Ditto.

From-SVN: r228410

re PR c/67730 (No warning when returning NULL in void function)

PR c/67730
* c-typeck.c (convert_for_assignment): Use the expansion point
location throughout.

* gcc.dg/pr67730-1.c: New test.
* gcc.dg/pr67730-2.c: New test.
* gcc.dg/pr67730.h: New test.

From-SVN: r228408

re PR c/67819 (-Wduplicated-cond should take macros into account)

* genemit.c (gen_exp): Remove -Wduplicated-cond hack.

* c.opt (Wduplicated-cond): Don't enable by -Wall anymore.

* c-c++-common/Wduplicated-cond-2.c: Skip until PR67819 is resolved.

From-SVN: r228405

PR libstdc++/67747 use readdir instead of readdir_r

PR libstdc++/67747
* src/filesystem/dir.cc (native_readdir): Remove.
(_Dir::advance): Use readdir instead of native_readdir.
(recursive_directory_iterator(const path&, directory_options,
error_code*)): Use swap instead of reset.

From-SVN: r228404

reject loops early where ivs cannot be represented

From-SVN: r228403

fix PR67754

From-SVN: r228402

add testcase

From-SVN: r228401

outline functions from stmt_simple_for_scop_p

Outlined functions from stmt_simple_for_scop_p. No functional changes intended.
Passes regtest and bootstrap.

gcc/ChangeLog:

2015-10-01  Aditya Kumar  <hiraditya@msn.com>

        * graphite-scop-detection.c (stmt_has_side_effects): New function
          outlined from stmt_simple_for_scop_p.
        (graphite_can_represent_stmt): Same.
        (stmt_simple_for_scop_p): Moved code out of this function for better
        readability.

From-SVN: r228400

AVX-512.

AVX-512. Add missing features to cpuinfo.c
gcc/
* config/i386/i386.c (processor_features): Add F_AVX512VBMI,
F_AVX512IFMA.
(isa_names_table): Handle F_AVX512VBMI and F_AVX512IFMA.
libgcc/
* config/i386/cpuinfo.c (processor_features): Add
FEATURE_AVX512VBMI and FEATURE_AVX512VBMI.
testsuite/
* gcc.target/i386/builtin_target.c: Handle "avx512ifma"
and "avx512vbmi".

From-SVN: r228399

[AArch64] Use default_elf_asm_named_section instead of special cased hook

* config/aarch64/aarch64-elf.h (TARGET_ASM_NAMED_SECTION): Delete.

From-SVN: r228397

re PR rtl-optimization/67756 (ICE compiling Linux Kernel fs/namei.c on ARM)

2015-10-02  Vladimir Makarov  <vmakarov@redhat.com>

PR rtl-optimization/67756
* lra-constraints.c (match_reload): Add a new parameter.  Use it
for creating a pseudo with the same value.
(curr_insn_transform): Pass a new argument to match_reload.

2015-10-02  Vladimir Makarov  <vmakarov@redhat.com>

PR rtl-optimization/67756
* gcc.target/arm/pr67756.c: New.

From-SVN: r228396

[AArch64] Add more TLS local executable testcases

2015-10-02  Jiong Wang  <jiong.wang@arm.com>

* gcc.target/aarch64/tlsle12_tiny_1.c: New testcase for tiny model.
* gcc.target/aarch64/tlsle24_tiny_1.c: Likewise.
* gcc.target/aarch64/tlsle_sizeadj_tiny_1.c: TLS size truncation test
for tiny model.
* gcc.target/aarch64/tlsle_sizeadj_small_1.c: TLS size truncation test
for small model.

From-SVN: r228395

Update extract_even_odd w/ AVX-512BW insns.

gcc/
* config/i386/i386.c (expand_vec_perm_even_odd_trunc): New.
(expand_vec_perm_even_odd_1): Handle V64QImode.
(ix86_expand_vec_perm_const_1): Try expansion with
expand_vec_perm_even_odd_trunc as well.
* config/i386/sse.md (VI124_AVX512F): Rename to ...
(define_mode_iterator VI124_AVX2_24_AVX512F_1_AVX512BW): This. Extend
to V54QI.
(define_mode_iterator VI248_AVX2_8_AVX512F): Rename to ...
(define_mode_iterator VI248_AVX2_8_AVX512F_24_AVX512BW): This. Extend
to V32HI and V16SI.
(define_insn "avx512bw_<code>v32hiv32qi2"): Unhide pattern name.
(define_expand "vec_pack_trunc_<mode>"): Update iterator name.
(define_expand "vec_unpacks_lo_<mode>"): Ditto.
(define_expand "vec_unpacks_hi_<mode>"): Ditto.
(define_expand "vec_unpacku_lo_<mode>"): Ditto.
(define_expand "vec_unpacku_hi_<mode>"): Ditto.

gcc/testsuite/
* gcc.target/i386/vect-pack-trunc-1.c: New test.
* gcc.target/i386/vect-pack-trunc-2.c: Ditto.
* gcc.target/i386/vect-perm-even-1.c: Ditto.
* gcc.target/i386/vect-perm-odd-1.c: Ditto.
* gcc.target/i386/vect-unpack-1.c: Ditto.
* gcc.target/i386/vect-unpack-2.c: Ditto.

From-SVN: r228394

invoke.texi: Mention -mavx512vl, -mavx512bw, -mavx512dq, -mavx521vbmi, -mavx512ifma.

gcc/
* doc/invoke.texi: Mention -mavx512vl, -mavx512bw, -mavx512dq,
-mavx521vbmi, -mavx512ifma. Add missing opindex-es.

From-SVN: r228393

re PR c/59218 (atomic transactions: accesses to volatiles not disallowed in transaction_safe code)

PR c/59218

* trans-mem.c (volatile_lvalue_p): Rename from volatile_var_p.
(diagnose_tm_1_op): Also diagnose volatile accesses in
transaction_safe function.

From-SVN: r228392

* system.h (malloc.h): Don't include obsolete header.

From-SVN: r228391

[AArch64] Use default_elf_asm_named_section instead of special cased hook

Hi,

The hook for TARGET_ASM_NAMED_SECTION was defined separately in
the backend around the time frame for GCC 4.7 under the assumption
that '@' would be used as a comment character in the binutils port.

This was indeed true in versions of the AArch64 port that never made it
into the FSF tree. However this practice was killed
before the binutils port made it upstream. Doing archaeology into
various revisions including the first commit of upstream binutils,
talking to Marcus about the history of this and actually testing the
first commit of binutils, I can confidently say that the upstream binutils
port never had any use of '@' as a comment character for AArch64
However we never got rid of the special cased handling in
GCC and the duplication of code in the AArch64 backend.

This was found when I was playing with Virtual Table verification
on ARM and AArch64 and discovered ICEs which were similar but manifested
in different places begging the obvious question.

Tested on aarch64-none-elf with no regressions.

Ok to apply ?

Ramana

2015-10-02  Ramana Radhakrishnan  <ramana.radhakrishnan@arm.com>

* config/aarch64/aarch64.c (aarch64_elf_asm_named_section): Delete.
(TLS_SECTION_ASM_FLAG): Delete.

From-SVN: r228390

re PR c/64249 (Missing warning for if (A) else if (A))

PR c/64249
* c-common.c (warn_duplicated_cond_add_or_warn): New function.
* c-common.h (warn_duplicated_cond_add_or_warn): Declare.
* c.opt (Wduplicated-cond): New option.

* c-parser.c (c_parser_statement_after_labels): Add CHAIN parameter
and pass it down to c_parser_if_statement.
(c_parser_else_body): Add CHAIN parameter and pass it down to
c_parser_statement_after_labels.
(c_parser_if_statement): Add CHAIN parameter.  Add code to warn about
duplicated if-else-if conditions.

* parser.c (cp_parser_statement): Add CHAIN parameter and pass it
down to cp_parser_selection_statement.
(cp_parser_selection_statement): Add CHAIN parameter.  Add code to
warn about duplicated if-else-if conditions.
(cp_parser_implicitly_scoped_statement): Add CHAIN parameter and pass
it down to cp_parser_statement.

* doc/invoke.texi: Document -Wduplicated-cond.
* Makefile.in (insn-latencytab.o): Use -Wno-duplicated-cond.
(insn-dfatab.o): Likewise.
* genemit.c (gen_exp): Rewrite condition to avoid -Wduplicated-cond
warning.

* c-c++-common/Wduplicated-cond-1.c: New test.
* c-c++-common/Wduplicated-cond-2.c: New test.
* c-c++-common/Wduplicated-cond-3.c: New test.
* c-c++-common/Wduplicated-cond-4.c: New test.
* c-c++-common/Wmisleading-indentation.c (fn_37): Avoid
-Wduplicated-cond warning.

From-SVN: r228388

[PATCH][AARCH64]Add csneg3_uxtw_insn pattern

gcc/

2015-10-02  Renlin Li <renlin.li@arm.com>

* config/aarch64/aarch64.md (csneg3_uxtw_insn): New pattern.

gcc/testsuite/

2015-10-02  Renlin Li <renlin.li@arm.com>

* gcc.target/aarch64/csneg-1.c: Update test.

From-SVN: r228387

sh.md: Add new unnamed split pattern to handle movt-movt sequences.

gcc/
* config/sh/sh.md: Add new unnamed split pattern to handle movt-movt
sequences.

From-SVN: r228386

[PATCH][AARCH64][PR66776]Add cmovdi_insn_uxtw pattern.

gcc/

2015-10-02  Renlin Li  <renlin.li@arm.com>

PR target/66776
* config/aarch64/aarch64.md (cmovdi_insn_uxtw): New pattern.

gcc/testsuite/

2015-10-02  Renlin Li  <renlin.li@arm.com>

PR target/66776
* gcc.target/aarch64/pr66776.c: New.

From-SVN: r228384

MAINTAINERS: Update email address.

From-SVN: r228383

lang.opt (fbuiltin-printf): Accept.

* gcc-interface/lang.opt (fbuiltin-printf): Accept.
* gcc-interface/misc.c (gnat_handle_option): Ignore it.
* gcc-interface/lang-specs.h (@ada, @adawhy, @adascil): Propagate -f*
flags as -g* and -m* flags past -gnatez.

Co-Authored-By: Olivier Hainque <hainque@adacore.com>
From-SVN: r228381

MAINTAINERS (Write After Approval): Add myself.

2015-10-02  Louis Krupp <louis.krupp@zoho.com>

* MAINTAINERS  (Write After Approval): Add myself.

From-SVN: r228380

trans.c (Pragma_to_gnu): Replace linear search with call to find_opt and issue warnings if...

* gcc-interface/trans.c (Pragma_to_gnu) <Pragma_Warnings>: Replace
linear search with call to find_opt and issue warnings if the -W
switch is not appropriate.

From-SVN: r228379

Back out improper commit

From-SVN: r228378

ada-tree.h (DECL_RESTRICTED_ALIASING_P): New flag.

* gcc-interface/ada-tree.h (DECL_RESTRICTED_ALIASING_P): New flag.
* gcc-interface/decl.c (gnat_to_gnu_param): For parameters passed by
reference but whose type isn't by-ref and whose mechanism hasn't been
forced to by-ref, set the DECL_RESTRICTED_ALIASING_P flag directly on
them instead of changing their type.
* gcc-interface/trans.c (scan_rhs_r): New helper function.
(independent_iterations_p): New predicate.
(Loop_Statement_to_gnu): For a loop with an iteration scheme, set an
ivdep pragma if the iterations are independent.

From-SVN: r228377

Delete undesirable svn:executable property

From-SVN: r228376

[RTL ifcvt] PR 67786, 67787: Check that intermediate instructions in the basic block don't clobber a reg used in condition

        PR rtl-optimization/67786
        PR rtl-optimization/67787
        * ifcvt.c (bb_valid_for_noce_process_p): Reject basic block if
        it modifies a reg used in the condition calculation.

        * gcc.dg/pr67786.c: New test.
        * gcc.dg/pr67787.c: Likewise.

From-SVN: r228375

[Patch AArch64] Improve SIMD concatenation with zeroes

gcc/

* config/aarch64/aarch64-simd.md (*aarch64_combinez<mode>): Add
alternatives for reads from memory and moves from general-purpose
registers.
(*aarch64_combinez_be<mode>): Likewise.

gcc/testsuite/

* gcc.target/aarch64/vect_combine_zeroes_1.c: New.

From-SVN: r228374

trans.c (find_loop_for): Use FOR_EACH_VEC_ELT_REVERSE.

* gcc-interface/trans.c (find_loop_for): Use FOR_EACH_VEC_ELT_REVERSE.
(Loop_Statement_to_gnu): Use FOR_EACH_VEC_ELT.
(finalize_nrv): Use FOR_EACH_VEC_SAFE_ELT and initialize field.
(build_stmt_group): Add blank lines.

From-SVN: r228373

re PR target/51726 (LTO and attribute 'selectany')

        PR target/51726
        * g++.dg/ext/selectany2.C: Allow uninitialized variable case.

From-SVN: r228371

re PR target/51726 (LTO and attribute 'selectany')

PR target/51726
* config/i386/winnt.c (ix86_handle_selectany_attribute): Handle
selectany within this function without need to keep attribute.
(i386_pe_encode_section_info): Remove selectany-code.

From-SVN: r228370

tree-ssa-sccvn.c (has_VN_INFO): New function.

2015-10-02  Richard Biener  <rguenther@suse.de>

* tree-ssa-sccvn.c (has_VN_INFO): New function.
(free_scc_vn): Use it.
(visit_use): Remove dead code and refactor to use gassign
and use less indentation.

From-SVN: r228369

[multiple changes]

2015-10-01  Louis Krupp <louis.krupp@zoho.com>

PR fortran/62242
PR fortran/52332
* trans-array.c
(store_backend_decl): Create new gfc_charlen instance if requested
(get_array_ctor_all_strlen): Call store_backend_decl requesting
new gfc_charlen
(trans_array_constructor): Call store_backend_decl requesting
new gfc_charlen if get_array_ctor_strlen was called
(gfc_add_loop_ss_code): Don't try to convert non-constant length

2015-10-01  Louis Krupp     <louis.krupp@zoho.com>

PR fortran/62242
PR fortran/52332
* gfortran.dg/string_array_constructor_1.f90: New.
* gfortran.dg/string_array_constructor_2.f90: New.
* gfortran.dg/string_array_constructor_3.f90: New.

From-SVN: r228368

Approved by FX

From-SVN: r228367

rs6000: Add "cannot_copy" attribute, use it (PR67788, PR67789)

After the shrink-wrapping patches the prologue will often be pushed
"deeper" into the function, which in turn means the software trace cache
pass will more often want to duplicate the basic block containing the
prologue.  This caused failures for 32-bit SVR4 with -msecure-plt PIC.

This configuration uses the load_toc_v4_PIC_1 instruction, which creates
assembler labels without using the normal machinery for that.  If now
the compiler decides to duplicate the insn, it will emit the same label
twice.  Boom.

It isn't so easy to fix this to use labels the compiler knows about (let
alone test that properly).  Instead, this patch wires up a "cannot_copy"
attribute to be used by TARGET_CANNOT_COPY_P, and sets that attribute on
these insns we do not want copied.

2015-10-01  Segher Boessenkool  <segher@kernel.crashing.org>

PR target/67788
PR target/67789
* config/rs6000/rs6000.c (TARGET_CANNOT_COPY_INSN_P): New.
(rs6000_cannot_copy_insn_p): New function.
* config/rs6000/rs6000.md (cannot_copy): New attribute.
(load_toc_v4_PIC_1_normal): Set cannot_copy.
(load_toc_v4_PIC_1_476): Ditto.

gcc/testsuite/
PR target/67788
PR target/67789
* gcc.target/powerpc/pr67789.c: New testcase.

From-SVN: r228366

PR fortran.67802

2015-10-01  Steven G. Kargl  <kargl@gcc.gnu.org>

PR fortran.67802
* decl.c (add_init_expr_to_sym): Numeric constant for character
length must be an INTEGER.

2015-10-01  Steven G. Kargl  <kargl@gcc.gnu.org>

PR fortran.67802
* gfortran.dg/pr67802.f90: New test.

From-SVN: r228365

re PR fortran/66979 (gfortran internal compiler error with malformed FLUSH statement)

2015-10-01  Steven G. Kargl  <kargl@gcc.gnu.org>

PR fortran/66979
* io.c (gfc_resolve_filepos): Check for a UNIT number.  Add a nearby
missing 'return false'.

2015-10-01  Steven G. Kargl  <kargl@gcc.gnu.org>

PR fortran/66979
gfortran.dg/pr66979.f90: new test.

From-SVN: r228364

re PR fortran/67616 (ICE on data initialization of type variable in block)

2015-10-01  Steven G. Kargl  <kargl@gcc.gnu.org>

PR fortran/67616
* primary.c (gfc_match_structure_constructor): Use a possibly
host-associated symtree to prevent ICE.

2015-10-01  Steven G. Kargl  <kargl@gcc.gnu.org>

PR fortran/67616
* gfortran.dg/pr67616.f90: New test.

From-SVN: r228363

Daily bump.

From-SVN: r228362

use sese_l throughout scop-detection

Use sese_l throughout SCoP detection and create vec<scop_p> at the very end when
all SCoPs have been identified. 'struct sese_l' is very lightweight (two
pointers) compared to 'struct scop'.

No functional change intended. Passes regtest and bootstrap.

gcc/ChangeLog:

2015-10-01  Aditya Kumar  <hiraditya@msn.com>

        * graphite-scop-detection.c (struct sese_l): New conversion constructor
          so that this type can be pushed into a vec.
        (class scop_builder): use sese_l to collect scops.
        (get_scops): New getter function.
        (remove_intersecting_scops): Use sese_l instead of scops_p.
        (intersects): Same.
        (add_scop): Same.
        (subsumes): Same.
        (remove_subscops): Same.
        (build_scops): Add scops to vec<scops_p> once all the scops have been
        detected.

From-SVN: r228359

Rename gimple_bb to gimple_poly_bb

Renaming gimple_bb to gimple_poly_bb because there is a function gimple_bb
by the same name in gimple.h. No functional change intended.
Passes regtest and bootstrap.

gcc/ChangeLog:

2015-10-01  Aditya Kumar  <hiraditya@msn.com>

        * graphite-isl-ast-to-gimple.c (class translate_isl_ast_to_gimple):
Renamed type from gimple_bb_p to gimple_poly_bb_p.
        (translate_isl_ast_node_user): Same.
        * graphite-poly.c (new_poly_bb): Same.
        * graphite-poly.h (gbb_from_bb): Same.
        * sese.h: Same.
        * graphite-sese-to-poly.c (new_gimple_bb): gimple_bb_p -> gimple_poly_bb_p
        (build_scop_scattering): Same.
        (find_params_in_bb): Same.
        (add_conditions_to_domain): Same.
        (sese_dom_walker::before_dom_children): Same.
        (analyze_drs_in_stmts): Same.
        (new_pbb_from_pbb): Same.
        (free_data_refs_aux): New pointer to type base_alias_pair.
        * graphite-sese-to-poly.h: Same.
        * sese.c (if_region_set_false_region): Fixed Indentation.
        (move_sese_in_condition): Same.

From-SVN: r228358

correctly handle non affine data references

2015-10-01  Sebastian Pop  <s.pop@samsung.com>
            Aditya Kumar  <aditya.k7@samsung.com>
        PR tree-optimization/66980
        * graphite-scop-detection.c (stmt_has_simple_data_refs_p): Return false
        when data reference analysis has failed.

Co-Authored-By: Aditya Kumar <aditya.k7@samsung.com>
From-SVN: r228357

re PR c++/54430 ([C++11] For-Loop: Scope of iterating variable begins too early)

PR c++/54430

/cp
2015-10-01  Ville Voutilainen  <ville.voutilainen@gmail.com>

PR c++/54430
* name-lookup.c (push_binding): Make non-static.
* name-lookup.h (push_binding): Declare it.
* parser.c (cp_parser_range_for): Use it, get the range
declaration away from the scope until the range expression
has been parsed, then restore the declaration.

/testsuite
2015-10-01  Ville Voutilainen  <ville.voutilainen@gmail.com>

PR c++/54430
* g++.dg/cpp0x/range-for30.C: New.

From-SVN: r228354

Do not describe -std=c11 etc. as experimental in c.opt help text.

I noticed that c.opt still described -std=c11 and related options as
experimental in the --help text.  This patch fixes this.

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

* c.opt (std=c11): Do not describe as experimental.
(std=gnu11): Likewise.
(std=iso9899:2011): Likewise.

From-SVN: r228349

* doc/html/manual/errno.html: Add new file.

From-SVN: r228348

call scev analysis in scop-detection as in sese-to-poly

Before our rewrite of the scop detection, we used to not have a valid SESE
region under hand, and so we used to do more ad-hoc analysis of data references
by trying to prove that at all levels of a loop nest the data references would
be still valid.

Now that we have a valid SESE region, we can call the scev analysis in the same
way on the same computed loop nest in the scop-detection as in the sese-to-poly.

Next step will be to cache the data references analyzed in the scop detection
and not compute the same info in sese-to-poly.

The patch fixes block-1.f90 that used to ICE on x86_64-linux when compiled with
-m32.  Patch passed bootstrap with BOOT_CFLAGS="-g -O2 -fgraphite-identity
-floop-nest-optimize" and check on x86_64-linux using ISL-0.15.

2015-09-28  Sebastian Pop  <s.pop@samsung.com>
Aditya Kumar  <aditya.k7@samsung.com>

PR tree-optimization/67754
* graphite-scop-detection.c (stmt_has_simple_data_refs_p): Call
scev analysis on the same loop nest as analyze_drs_in_stmts.
* graphite-sese-to-poly.c (outermost_loop_in_sese_1): Moved and renamed...
(try_generate_gimple_bb): Call outermost_loop_in_sese.
(analyze_drs_in_stmts): Same.
* sese.c (outermost_loop_in_sese): ...here.

Co-Authored-By: Aditya Kumar <aditya.k7@samsung.com>
From-SVN: r228347

add recursion on the inner loops

We now check that all data references in the current loop and inner loops
contained within loop are valid in an outer region before declaring that the
outer loop is a valid scop.

2015-09-30  Sebastian Pop  <s.pop@samsung.com>
    Aditya Kumar  <aditya.k7@samsung.com>

PR tree-optimization/67754
* graphite-scop-detection.c (loop_body_is_valid_scop): Add missing
recursion on the inner loops.

Co-Authored-By: Aditya Kumar <aditya.k7@samsung.com>
From-SVN: r228346

remove many typedefs

gcc/ChangeLog:

2015-10-01  Trevor Saunders  <tbsaunde+gcc@tbsaunde.org>

* cfganal.c, compare-elim.c, coverage.c, cprop.c, df-scan.c,
function.c, read-rtl.c, statistics.c, trans-mem.c, tree-if-conv.c,
tree-into-ssa.c, tree-loop-distribution.c, tree-ssa-coalesce.c,
tree-ssa-loop-ivopts.c, tree-ssa-reassoc.c, tree-ssa-strlen.c,
tree-ssa-tail-merge.c, tree-vrp.c, var-tracking.c: Remove
unneeded typedefs.

From-SVN: r228344