re PR tree-optimization/64193 (Decreased performance after r173250)

2014-12-09  Richard Biener  <rguenther@suse.de>

PR tree-optimization/64193
* tree-ssa-alias.c (walk_non_aliased_vuses): Add valueize parameter
and valueize the VUSE before looking up the def stmt.
* tree-ssa-alias.h (walk_non_aliased_vuses): Adjust prototype.
* tree-ssa-sccvn.c (vn_reference_lookup_pieces): Pass vn_valueize
to walk_non_aliased_vuses.
(vn_reference_lookup): Likewise.
* tree-ssa-dom.c (lookup_avail_expr): Pass NULL as valueize
callback to walk_non_aliased_vuses.

* gcc.dg/tree-ssa/ssa-fre-43.c: New testcase.

From-SVN: r218515

Local Vim config with GNU formatting.

2014-12-09  Laurynas Biveinis  <laurynas.biveinis@gmail.com>
    Yury Gribov  <y.gribov@samsung.com>

/
* .gitignore: Added .local.vimrc and .lvimrc.
* Makefile.tpl (vimrc, .lvimrc, .local.vimrc): New targets.
* Makefile.in: Regenerate.

contrib/
* vimrc: New file.

From-SVN: r218514

re PR tree-optimization/64199 (ICE: tree check: expected class 'constant', have 'binary' (plus_expr) in fold_binary_loc, at fold-const.c:10404 with -ffast-math -frounding-math)

2014-12-09  Richard Biener  <rguenther@suse.de>

PR middle-end/64199
* fold-const.c (fold_binary_loc): Use TREE_OVERFLOW_P.

* gcc.dg/torture/pr64199.c: New testcase.

From-SVN: r218513

re PR libstdc++/64203 (shared_mutex compile errors on bare-metal targets)

PR libstdc++/64203
* include/std/shared_mutex: Fix preprocessor conditions.
* testsuite/experimental/feat-cxx14.cc: Check conditions.

From-SVN: r218511

re PR tree-optimization/64191 (indirect clobbers messes up dead code elimination in loop calling dtor)

2014-12-09  Richard Biener  <rguenther@suse.de>

PR tree-optimization/64191
* tree-vect-stmts.c (vect_stmt_relevant_p): Clobbers are
not relevant (nor are their uses).

From-SVN: r218509

lto-partition.c (privatize_symbol_name): Correctly privatize instrumentation clones.

gcc/

* lto/lto-partition.c (privatize_symbol_name): Correctly
privatize instrumentation clones.

gcc/testsuite/

* gcc.dg/lto/lto.exp: Load mpx-dg.exp.
* gcc.dg/lto/chkp-privatize_0.c: New.
* gcc.dg/lto/chkp-privatize_1.c: New.

From-SVN: r218508

lto-cgraph.c (input_cgraph_1): Don't break existing instrumentation clone references.

* lto-cgraph.c (input_cgraph_1): Don't break existing
instrumentation clone references.
* lto/lto-symtab.c (lto_cgraph_replace_node): Redirect
instrumented_version references appropriately.

From-SVN: r218507

re PR bootstrap/63995 (Bootstrap error with -mmpx -fcheck-pointer-bounds)

gcc/

PR bootstrap/63995
* tree-chkp.c (chkp_make_static_bounds): Share bounds var
between nodes sharing assembler name.

gcc/testsuite/

PR bootstrap/63995
* g++.dg/dg.exp: Add mpx-dg.exp.
* g++.dg/pr63995-1.C: New.

From-SVN: r218506

re PR target/64204 (gcc.dg/c11-atomic-2.c fails on powerpc 64-bit little endian after -mupper-regs patches went in)

2014-12-08  Michael Meissner  <meissner@linux.vnet.ibm.com>

PR target/64204
* config/rs6000/rs6000.c (rs6000_emit_move): Do not split TFmode
constant moves if -mupper-regs-df.

* config/rs6000/rs6000.md (mov<mode>_64bit_dm): Optimize moving
0.0L to TFmode.
(movtd_64bit_nodm): Likewise.
(mov<mode>_32bit, FMOVE128 case): Likewise.

From-SVN: r218505

simplify-rtx.c (simplify_relational_operation_1): Handle simplification identities for BICS patterns.

2014-12-08  Sandra Loosemore  <sandra@codesourcery.com>

gcc/
* simplify-rtx.c (simplify_relational_operation_1): Handle
simplification identities for BICS patterns.

gcc/testsuite/
* gcc.target/aarch64/bics_4.c: New.

From-SVN: r218503

move gimple_canonical_types htab out of gc memory

lto/

* lto.c (read_cgraph_and_symbols): allocate gimple_canonical_types
htab with malloc instead of ggc.

From-SVN: r218502

remove param_is from nvptx

gcc/

* config/nvptx/nvptx.c: Convert htabs to hash_table.

From-SVN: r218501

Daily bump.

From-SVN: r218500

re PR target/64226 (Secondary reload incorrect TOC address)

        PR target/64226
        * config/rs6000/rs6000.c (rs6000_secondary_reload_inner)
        [SYMBOL_REF]: Do not explicitly call create_TOC_reference for
        TARGET_TOC. Always use rs6000_emit_move.

From-SVN: r218497

DWARFv5 Emit DW_TAG_atomic_type for C11 _Atomic.

This implements the DW_TAG_atomic_type for C11 _Atomic proposal as adopted
in the latest DWARF5 draft. http://dwarfstd.org/ShowIssue.php?issue=131112.1

gcc/ChangeLog

PR debug/60782
* dwarf2out.c (modified_type_die): Handle TYPE_QUAL_ATOMIC.

gcc/testsuite/ChangeLog

PR debug/60782
* gcc.dg/debug/dwarf2/atomic.c: New test.
* gcc.dg/debug/dwarf2/stacked-qualified-types-3.c: Likewise.

include/ChangeLog

PR debug/60782
* dwarf2.def: Add DWARFv5 DW_TAG_atomic_type.

From-SVN: r218496

Fix typo in ChangeLog

From-SVN: r218495

re PR inline-asm/61692 (ICE in extract_insn in recog.c for asm with many parameters)

        PR target/61692
* cfgexpand.c (expand_asm_operands): Count all inline asm params.

PR target/61692
        * gcc.dg/pr61692.c: New test.

From-SVN: r218494

Added myself to Write After Approval list

From-SVN: r218491

PR jit/63854: Introduce xstrdup_for_dump

gcc/ChangeLog:
PR jit/63854
* cgraph.h (xstrdup_for_dump): New function.
* cgraph.c (cgraph_node::get_create): Replace use of xstrdup
within fprintf with xstrdup_for_dump.
(cgraph_edge::make_speculative): Likewise.
(cgraph_edge::resolve_speculation): Likewise.
(cgraph_edge::redirect_call_stmt_to_callee): Likewise.
(cgraph_node::dump): Likewise.
* cgraphclones.c (symbol_table::materialize_all_clones): Likewise.
* ipa-cp.c (perhaps_add_new_callers): Likewise.
* ipa-inline.c (report_inline_failed_reason): Likewise.
(want_early_inline_function_p): Likewise.
(edge_badness): Likewise.
(update_edge_key): Likewise.
(flatten_function): Likewise.
(inline_always_inline_functions): Likewise.
* ipa-profile.c (ipa_profile): Likewise.
* ipa-prop.c (ipa_print_node_jump_functions): Likewise.
(ipa_make_edge_direct_to_target): Likewise.
(remove_described_reference): Likewise.
(propagate_controlled_uses): Likewise.
* ipa-utils.c (ipa_merge_profiles): Likewise.

From-SVN: r218490

libgccjit++.h: use indentation to show inheritance

gcc/jit/ChangeLog:
* libgccjit++.h: Indent the forward declarations of the classes to
show the inheritance hierarchy.

From-SVN: r218489

Make jit/notes.txt better reflect current status quo

gcc/jit/ChangeLog:
* notes.txt: Show the beginning and ending of
recording::context::compile vs playback::context::compile.  Show
the creation and unlinking of the tempdir.  Show toplev::finalize.
Move "RELEASE MUTEX" to the correct location.  Show
gcc_jit_result_release, and indicate where the
dlopen/dlsym/dlclose occur.

From-SVN: r218488

re PR ipa/64049 (r215898 caused wrong code at -O3)

2014-12-08  Bernd Edlinger  <bernd.edlinger@hotmail.de>

        PR ipa/64049
        * ipa-polymorphic-call.c
        (pa_polymorphic_call_context::ipa_polymorphic_call): Allow RESULT_DECL.

testsuite/ChangeLog:
2014-12-08  Bernd Edlinger  <bernd.edlinger@hotmail.de>

        PR ipa/64049
        * g++.dg/ipa/pr64049.h: New.
        * g++.dg/ipa/pr64049-1.C: New.
        * g++.dg/ipa/pr64049-2.C: New.

From-SVN: r218487

Bics instruction generation for aarch64

gcc/

* config/aarch64/aarch64.md (and_one_cmpl<mode>3_compare0_no_reuse):
New define_insn.
* (and_one_cmpl_<SHIFT:optab><mode>3_compare0_no_reuse):
Likewise.

gcc/testsuite/

* gcc.target/aarch64/bics_3.c : New testcase.

From-SVN: r218486

re PR go/64198 (ICE in gofrontend)

PR go/64198
compiler: Don't crash on invalid ++.

From-SVN: r218485

arm_neon.h (vrecpe_u32, [...]): Rewrite using builtin functions.

        * config/aarch64/arm_neon.h (vrecpe_u32, vrecpeq_u32): Rewrite using
        builtin functions.
        (vfma_f32, vfmaq_f32, vfmaq_f64, vfma_n_f32, vfmaq_n_f32, vfmaq_n_f64,
        vfms_f32, vfmsq_f32, vfmsq_f64): Likewise.
        (vhsub_s8, vhsub_u8, vhsub_s16, vhsub_u16, vhsub_s32, vhsub_u32,
        vhsubq_s8, vhsubq_u8, vhsubq_s16, vhsubq_u16, vhsubq_s32, vhsubq_u32,
        vsubhn_s16, vsubhn_u16, vsubhn_s32, vsubhn_u32, vsubhn_s64, vsubhn_u66,
        vrsubhn_s16, vrsubhn_u16, vrsubhn_s32, vrsubhn_u32, vrsubhn_s64,
        vrsubhn_u64, vsubhn_high_s16, vsubhn_high_u16, vsubhn_high_s32,
        vsubhn_high_u32, vsubhn_high_s64, vsubhn_high_u64, vrsubhn_high_s16,
        vrsubhn_high_u16, vrsubhn_high_s32, vrsubhn_high_u32, vrsubhn_high_s64,
        vrsubhn_high_u64): Likewise.
        * config/aarch64/iterators.md (VDQ_SI): New mode iterator.
        * config/aarch64/aarch64.md (define_c_enum "unspec"): Add UNSPEC_URECPE.
        * config/aarch64/aarch64-simd.md (aarch64_urecpe<mode>): New pattern.
        * config/aarch64/aarch64-simd-builtins.def (shsub, uhsub, subhn, rsubhn,
        subhn2, rsubhn2, urecpe): New builtins.

Co-Authored-By: Haijian Zhang <z.zhanghaijian@huawei.com>
Co-Authored-By: Jiji Jiang <jiangjiji@huawei.com>
Co-Authored-By: Pengfei Sui <suipengfei@huawei.com>
From-SVN: r218484

Enable non-const v64qi permutations.

gcc/
* config/i386/i386.c (ix86_expand_vec_perm_vpermi2): Handle v64qi.
* config/i386/sse.md (VEC_PERM_AVX2): Add v64qi.

From-SVN: r218483

Enable const permutations for V64QImode.

gcc/
* config/i386/i386.c (expand_vec_perm_broadcast_1): Handle v64qi.
(expand_vec_perm_vpermi2_vpshub2): New.
(ix86_expand_vec_perm_const_1): Use it.
(ix86_vectorize_vec_perm_const_ok): Handle v64qi.
* config/i386/sse.md (VEC_PERM_CONST): Add v64qi.

From-SVN: r218482

tree-chkp.c (chkp_build_returned_bound): Don't predict return bounds for strchr calls.

* tree-chkp.c (chkp_build_returned_bound): Don't predict
return bounds for strchr calls.

From-SVN: r218479

tree-chkp.c (chkp_call_returns_bounds_p): New.

gcc/

* tree-chkp.c (chkp_call_returns_bounds_p): New.
(chkp_build_returned_bound): Use zero bounds as
returned by calls not returning bounds.

gcc/testsuite/

* gcc.target/i386/chkp-bndret.c: New.
* gcc.target/i386/chkp-strchr.c: New.

From-SVN: r218478

builtins.c (fold_builtin_0): Remove unused ignore parameter.

2014-12-08  Richard Biener  <rguenther@suse.de>

* builtins.c (fold_builtin_0): Remove unused ignore parameter.
(fold_builtin_1): Likewise.
(fold_builtin_3): Likewise.
(fold_builtin_varargs): Likewise.
(fold_builtin_2): Likewise.  Do not fold stpcpy here.
(fold_builtin_n): Adjust.
(fold_builtin_stpcpy): Move to gimple-fold.c.
(gimple_fold_builtin_stpcpy): Moved and gimplified from builtins.c.
(gimple_fold_builtin): Fold stpcpy here.

From-SVN: r218477

don't ICE when section attribute is used on things in comdats

gcc/

PR ipa/63621
* symtab.c (symtab_node::verify): Check for section attribute before
asserting something isn't in a section and a comdat group.

From-SVN: r218476

Daily bump.

From-SVN: r218475

re PR target/53513 ([SH] Add support for fpchg insn and improve fenv support)

gcc/testsuite/
PR target/53513
* gcc.target/sh/pr54602-4.c: Fix matching of rte-nop sequence.

From-SVN: r218472

re PR target/50751 (SH Target: Displacement addressing does not work for QImode and HImode)

gcc/
PR target/50751
* config/sh/sh.md (extendqihi2): Allow only for TARGET_SH1.

From-SVN: r218469

compare-elim.c: Fix head comment.

* compare-elim.c: Fix head comment.
(conforming_compare): Remove redundant test.
(can_eliminate_compare): New function extracted from...
(before_dom_children): ...here.  Use it, replace direct uses of
flag_non_call_exceptions and tidy up.
(maybe_select_cc_mode): Tidy up.

From-SVN: r218468

re PR testsuite/64038 (FAIL: gcc.dg/ipa/ipa-icf-5.c (test for excess errors))

PR testsuite/64038
* gcc.dg/ipa/ipa-icf-5.c: Require c99_runtime.  Add c99_runtime options.
* gcc.dg/ipa/ipa-icf-7.c: Likewise.

From-SVN: r218467

h8300.exp: Fix duplicated text.

gcc/testsuite/
* gcc.target/h8300/h8300.exp: Fix duplicated text.
* gcc.target/h8300/pragma-isr.c: Likewise.
* gcc.target/h8300/pragma-isr2.c: Likewise.

From-SVN: r218465

aarch64-simd.md (clrsb<mode>2, [...]): New patterns.

        * config/aarch64/aarch64-simd.md (clrsb<mode>2, popcount<mode>2): New
        patterns.
        * config/aarch64/aarch64-simd-builtins.def (clrsb, popcount): New
        builtins.
        * config/aarch64/arm_neon.h (vcls_s8, vcls_s16, vcls_s32, vclsq_s8,
        vclsq_s16, vclsq_s32, vcnt_p8, vcnt_s8, vcnt_u8, vcntq_p8, vcntq_s8,
        vcntq_u8): Rewrite using builtin functions.

Co-Authored-By: Shanyao Chen <chenshanyao@huawei.com>
From-SVN: r218464

[ARM,AArch64][testsuite] Fix vaddl and vaddw tests

2014-12-07  Christophe Lyon  <christophe.lyon@linaro.org>

* gcc.target/aarch64/advsimd-intrinsics/vaddhn.c: Actually execute
the test.
* gcc.target/aarch64/advsimd-intrinsics/vaddl.c: Actually execute
the test. Fix expected output.
* gcc.target/aarch64/advsimd-intrinsics/vaddw.c: Likewise.

From-SVN: r218463

symtab.c (symtab_node::equal_address_to): New function.

* symtab.c (symtab_node::equal_address_to): New function.
* cgraph.h (symtab_node::equal_address_to): Declare.
* fold-const.c (fold_comparison, fold_binary_loc): Use it.

* c-family/c-common.c: Refuse weaks for symbols that can not change
visibility.

* gcc.dg/addr_equal-1.c: New testcase.

From-SVN: r218462

* doc/invoke.texi (Warning Options): Fix spelling and grammar.

From-SVN: r218461

Daily bump.

From-SVN: r218460

extend jump thread for finite state automata

PR tree-optimization/54742
* params.def (max-fsm-thread-path-insns, max-fsm-thread-length,
max-fsm-thread-paths): New.

* doc/invoke.texi (max-fsm-thread-path-insns, max-fsm-thread-length,
max-fsm-thread-paths): Documented.

* tree-cfg.c (split_edge_bb_loc): Export.
* tree-cfg.h (split_edge_bb_loc): Declared extern.

* tree-ssa-threadedge.c (simplify_control_stmt_condition): Restore the
original value of cond when simplification fails.
(fsm_find_thread_path): New.
(fsm_find_control_statement_thread_paths): New.
(thread_through_normal_block): Call find_control_statement_thread_paths.

* tree-ssa-threadupdate.c (dump_jump_thread_path): Pretty print
EDGE_FSM_THREAD.
(verify_seme): New.
(duplicate_seme_region): New.
(thread_through_all_blocks): Generate code for EDGE_FSM_THREAD edges
calling duplicate_seme_region.

* tree-ssa-threadupdate.h (jump_thread_edge_type): Add EDGE_FSM_THREAD.

* testsuite/gcc.dg/tree-ssa/ssa-dom-thread-6.c: New test.
* testsuite/gcc.dg/tree-ssa/ssa-dom-thread-7.c: New test.

Co-Authored-By: Brian Rzycki <b.rzycki@samsung.com>
Co-Authored-By: Sebastian Pop <s.pop@samsung.com>
From-SVN: r218451

error.c (gfc_error_check): Use bool not int.

2014-12-06  Tobias Burnus  <burnus@net-b.de>

        * error.c (gfc_error_check): Use bool not int.
        * gfortran.h (gfc_error_check): Update prototype.
        * match.c (gfc_match_if): Update call.
        * parse.c (decode_statement, decode_omp_directive,
        decode_gcc_attribute): Ditto.

From-SVN: r218450

error.c (gfc_buffer_error, [...]): Use bool not

2014-12-06  Tobias Burnus  <burnus@net-b.de>
            Manuel López-Ibáñez  <manu@gcc.gnu.org>

        * error.c (gfc_buffer_error, gfc_error_flag_test): Use bool not
        * int.
        (buffer_flag): Remove static variable.
        (buffered_p): Add static variable.
        (gfc_error_init_1): Call gfc_buffer_error.
        (gfc_warning_1, gfc_warning, gfc_warning_now_1, gfc_error,
        gfc_error_now_1): Update for static variable change.
        * gfortran.h (gfc_buffer_error, gfc_error_flag_test): Update
        prototype.
        * parse.c (use_modules, decode_specification_statement,
        next_fixed, next_statement, match_deferred_characteristics):
        Update calls.
        * decl.c (gfc_match_data_decl): Ditto.
        * match.c (gfc_match_name): Ditto.

Co-Authored-By: Manuel López-Ibáñez <manu@gcc.gnu.org>
From-SVN: r218449

clarification

From-SVN: r218448

re PR tree-optimization/64183 (Complete unroll doesn't happen for a while-loop)

PR tree-optimization/64183
* c-gimplify.c (c_gimplify_expr): Don't convert the RHS of a
shift-expression if it is integer_type_node.  Use types_compatible_p.

* gcc.dg/tree-ssa/pr64183.c: New test.

From-SVN: r218447

Don't assert "alg != libcall" for -minline-stringops-dynamically

gcc/

PR target/64200
* config/i386/i386.c (decide_alg): Don't assert "alg != libcall"
for TARGET_INLINE_STRINGOPS_DYNAMICALLY.

gcc/testsuite/

PR target/64200
* gcc.target/i386/memcpy-strategy-4.c: New test.

From-SVN: r218446

Fix ChangeLog format.

From-SVN: r218444

Daily bump.

From-SVN: r218443

re PR sanitizer/64170 (ICE compiling Linux Kernel drivers/media/rc/imon.c in imon_incoming_packet)

PR sanitizer/64170
* sanopt.c (maybe_optimize_asan_check_ifn): If base_checks is
non-NULL, call maybe_get_dominating_check on it even if g is
non-NULL.

* gcc.dg/asan/pr64170.c: New test.

From-SVN: r218440

md.texi: Note problems using function calls to determine insn lengths and point readers...

        * doc/md.texi: Note problems using function calls to determine
        insn lengths and point readers to a potential workaround.

From-SVN: r218439

* combine.c (is_parallel_of_n_reg_sets)
(can_split_parallel_of_n_reg_sets): Only define if !HAVE_cc0.

From-SVN: r218436

aarch64-simd-builtins.def (bswap): Use CF2 rather than CF10 so 2 is appended on the code.

2014-12-05  Andrew Pinski  <apinski@cavium.com>

        * config/aarch64/aarch64-simd-builtins.def (bswap): Use CF2 rather
        than CF10 so 2 is appended on the code.
        * config/aarch64/aarch64-simd.md (bswap<mode>): Rename to ...
        (bswap<mode>2): This so it matches for the optabs.

From-SVN: r218435

regrename.c (find_best_rename_reg): Rename to ...

2014-12-05  Thomas Preud'homme  <thomas.preudhomme@arm.com>

    gcc/
    * regrename.c (find_best_rename_reg): Rename to ...
    (find_rename_reg): This. Also add a parameter to skip tick check.
    * regrename.h: Likewise.
    * config/c6x/c6x.c (try_rename_operands): Adapt to above renaming.

From-SVN: r218434

re PR ipa/64192 (r218369 causes some regressions with -m32.)

2014-12-05  Martin Jambor  <mjambor@suse.cz>

PR ipa/64192
* ipa-prop.c (ipa_compute_jump_functions_for_edge): Convert alignment
from bits to bytes after checking they are byte-aligned.

From-SVN: r218433

[AArch64]Clarify the usage of SCHED in AARCH64_CORE macro

gcc/

2014-12-05  Renlin Li  <renlin.li@arm.com>

* config/aarch64/aarch64-opts.h (AARCH64_CORE): Rename IDENT to SCHED.
* config/aarch64/aarch64.h (AARCH64_CORE): Likewise.
* config/aarch64/aarch64.c (AARCH64_CORE): Rename X to IDENT,
IDENT to SCHED.

From-SVN: r218432

re PR target/53199 (__builtin_bswap64 and __builtin_bswap32 generate errors if -mcpu=power6)

2014-12-05  Michael Meissner  <meissner@linux.vnet.ibm.com>

PR target/53199
* gcc.target/powerpc/pr53199.c: Limit tests to 64-bit.

From-SVN: r218431

[AArch64]load store pair optimization using sched_fusion pass.

From-SVN: r218430

defaults.h: (DWARF_REG_TO_UNWIND_COLUMN): Define default.

2014-12-05  Olivier Hainque  <hainque@adacore.com>

        gcc/
        * defaults.h: (DWARF_REG_TO_UNWIND_COLUMN): Define default.
        * dwarf2cfi.c (init_one_dwarf_reg_size): Honor
        DWARF_REG_TO_UNWIND_COLUMN.

        libgcc/
        * unwind-dw2.c (DWARF_REG_TO_UNWIND_COLUMN): Remove default def,
        now provided by defaults.h.

From-SVN: r218429

dwarf2cfi.c (init_one_dwarf_reg_size): New helper, processing one particular reg for expand_builtin_init_dwarf_reg_sizes.

2014-12-05  Olivier Hainque  <hainque@adacore.com>

        * dwarf2cfi.c (init_one_dwarf_reg_size): New helper, processing
        one particular reg for expand_builtin_init_dwarf_reg_sizes.
        (expand_builtin_init_dwarf_reg_sizes): Rework to use helper and
        account for dwarf register spans.

From-SVN: r218428

pr63341-1.c: Remove explicit "dg-do run".

2014-12-05  Sandra Loosemore  <sandra@codesourcery.com>

gcc/testsuite/
* gcc.dg/vect/pr63341-1.c: Remove explicit "dg-do run".
* gcc.dg/vect/pr63341-2.c: Likewise.
* gcc.dg/vect/pr63379.c: Likewise.
* gcc.dg/vect/pr63605.c: Likewise.

From-SVN: r218427

re PR target/64003 (valgrind complains about get_attr_length_nobnd in insn-attrtab.c from i386.md)

PR target/64003
* config/i386/i386.md (*jcc_1_bnd): New.
(*jcc_2_bnd): New.
(jump_bnd): New.
(*jcc_1): Remove bnd prefix.
(*jcc_2): Likewise.
(jump): Likewise.

From-SVN: r218426

[AArch64] Use selected cpu's tuning when no tuning parameter is specified

gcc/

2014-12-05  Renlin Li  <renlin.li@arm.com>

* config/aarch64/aarch64.c (aarch64_parse_cpu): Don't define
selected_tune.
(aarch64_override_options): Use selected_cpu's tuning.

From-SVN: r218425

xcoff.h (ASM_OUTPUT_ALIGNED_LOCAL): Append alignment to section name.

* config/rs6000/xcoff.h (ASM_OUTPUT_ALIGNED_LOCAL): Append
alignment to section name. Increase default alignment to word.

From-SVN: r218424

re PR fortran/60414 (internal compiler error: tree check)

2014-12-05  Andre Vehreschild  <vehre@gmx.de>

PR fortran/60414
* interface.c (compare_parameter): Remove class argument rank
check short circuit.

2014-12-05  Andre Vehreschild  <vehre@gmx.de>

PR fortran/60414
* gfortran.dg/unlimited_polymorphism_18.f90: New test.

From-SVN: r218422

cgraph.h (cgraph_node): New method expand_all_artificial_thunks.

2014-12-05  Martin Jambor  <mjambor@suse.cz>

* cgraph.h (cgraph_node): New method expand_all_artificial_thunks.
(cgraph_edge): New method redirect_callee_duplicating_thunks.
* cgraphclones.c (duplicate_thunk_for_node): Donot expand newly
created thunks.
(redirect_edge_duplicating_thunks): Turned into edge method
redirect_callee_duplicating_thunks.
(cgraph_node::expand_all_artificial_thunks): New method.
(create_clone): Call expand_all_artificial_thunks.
* ipa-cp.c (perhaps_add_new_callers): Call
redirect_callee_duplicating_thunks instead of redirect_callee.
Also call expand_all_artificial_thunks.

From-SVN: r218417

hashtable_policy.h (_Map_base::operator[], [...]): Simplify definitions with trailing return types.

* include/bits/hashtable_policy.h (_Map_base::operator[],
_Map_base::at): Simplify definitions with trailing return types.

From-SVN: r218416

re PR target/64056 (gcc.target/i386/chkp-strlen-4.c etc. FAIL)

gcc/

PR target/64056
* doc/sourcebuild.texi: Add mempcpy and stpcpy for Effective-Target Keywords.

gcc/testsuite/

PR target/64056
* lib/target-supports.exp (check_effective_target_stpcpy): New.
* gcc.target/i386/chkp-strlen-2.c: Add stpcpy target check.

From-SVN: r218415

Daily bump.

From-SVN: r218413

This test should have been added in revision 218407.

gcc/testsuite/

2014-12-05  Manuel López-Ibáñez  <manu@gcc.gnu.org>

* gfortran.dg/line_length_5.f90: New test.

From-SVN: r218410

error.c (gfc_diagnostic_build_locus_prefix): Use diagnostic_expand_location.

gcc/fortran/ChangeLog:

2014-12-05  Manuel López-Ibáñez  <manu@gcc.gnu.org>

* error.c (gfc_diagnostic_build_locus_prefix): Use
diagnostic_expand_location.

gcc/ChangeLog:

2014-12-05  Manuel López-Ibáñez  <manu@gcc.gnu.org>

* diagnostic.h (diagnostic_expand_location): New inline function.
* diagnostic.c (diagnostic_build_prefix): Use it.
(diagnostic_show_locus): Likewise.

From-SVN: r218409

Always define HAVE_LD_PIE_COPYRELOC

PR bootstrap/64189
* configure.ac (HAVE_LD_PIE_COPYRELOC): Always define.
* configure: Regenerated.

From-SVN: r218408

line-map.c (linemap_position_for_loc_and_offset): Add new linemap_assert_fails.

libcpp/ChangeLog:

2014-12-05  Manuel López-Ibáñez  <manu@gcc.gnu.org>

* line-map.c (linemap_position_for_loc_and_offset): Add new
linemap_assert_fails.

gcc/fortran/ChangeLog:

2014-12-05  Manuel López-Ibáñez  <manu@gcc.gnu.org>

* scanner.c (gfc_next_char_literal): Use gfc_warning_now.
(load_file): Use the line length as the column hint for
linemap_line_start. Reserve a location for the highest column of
the line.

From-SVN: r218407

diagnostic.c (diagnostic_color_init): New.

gcc/ChangeLog:

2014-12-04  Manuel López-Ibáñez  <manu@gcc.gnu.org>

* diagnostic.c (diagnostic_color_init): New.
* diagnostic.h: Declare.
* gcc.c (driver::global_initializations): Use it.
(driver_handle_option): Handle -fdiagnostics-color_.
* toplev.c: Do not include diagnostic-color.h.
(process_options): Do not initialize color diagnostics here.
* common.opt (fdiagnostics-color=): Add Driver.
* opts-global.c (init_options_once): Initialize color here.
* opts.c (common_handle_option): Use diagnostics_color_init.
* diagnostic-color.h: Fix comment.

From-SVN: r218406

Avoid overuse of name "buffer" in tree-pretty-print.c

gcc/ChangeLog:
* tree-pretty-print.c (INDENT): Rename "buffer" to "pp".
(NIY): Likewise.
(buffer): Rename this variable to...
(tree_pp): ...this.

(do_niy): Rename param from "buffer" to "pp".
(dump_decl_name): Likewise.
(dump_function_name): Likewise.
(dump_function_declaration): Likewise.
(dump_array_domain): Likewise.
(dump_omp_clause): Likewise.
(dump_omp_clauses): Likewise.
(dump_location): Likewise.
(dump_block_node): Likewise.
(dump_generic_node): Likewise.
(print_declaration): Likewise.
(print_struct_decl): Likewise.
(print_call_name): Likewise.
(pretty_print_string): Likewise.
(newline_and_indent): Likewise.

(print_generic_decl): Update for renaming of "buffer" to
"tree_pp".
(print_generic_stmt): Likewise.
(print_generic_stmt_indented): Likewise.
(print_generic_expr): Likewise.
(maybe_init_pretty_print): Likewise.

From-SVN: r218405

PR jit/63854: Fix double-initialization within tree-pretty-print.c

gcc/ChangeLog:
PR jit/63854
* tree-pretty-print.c: Eliminate include of <new>.
(buffer): Convert this variable from a pretty_printer to a
pretty_printer *.
(initialized): Eliminate this variable in favor of the NULL-ness
of "buffer".
(print_generic_decl): Update for "buffer" becoming a pointer.
(print_generic_stmt): Likewise.
(print_generic_stmt_indented): Likewise.
(print_generic_expr): Likewise.
(maybe_init_pretty_print): Likewise, allocating "buffer" on the
heap and using its non-NULL-ness to ensure idempotency.

From-SVN: r218404

PR jit/63854: Fix leak of ipa hooks

gcc/ChangeLog:
PR jit/63854
* ipa-prop.c (ipa_register_cgraph_hooks): Guard insertion of
ipa_add_new_function on function_insertion_hook_holder being
non-NULL.
* ipa-reference.c (ipa_reference_c_finalize): Remove
node_removal_hook_holder and node_duplication_hook_holder if
they've been added to symtab.
* toplev.c (toplev::finalize): Call ipa_reference_c_finalize
before cgraph_c_finalize so that the former can access "symtab".

From-SVN: r218403

re PR c++/64029 (const int (&in)[]{1,2,3,4,5}; results in internal compiler error: Segmentation fault)

PR c++/64029
* decl.c (grok_reference_init): Complete array type.

From-SVN: r218402

re PR c++/64080 ([C++14]ICE in cxx_eval_store_expression)

PR c++/64080
* constexpr.c (cxx_eval_store_expression): Handle non-decl store
targets.

From-SVN: r218401

fix comment

From-SVN: r218400

Update docs to reflect use of gimple subclasses

gcc/ChangeLog:
* doc/cfg.texi (GIMPLE statement iterators): Add note about
gphi_iterator, and use one in the example.
* doc/gimple.texi (Tuple specific accessors): Add missing
GIMPLE_GOTO section and menu item.
(gimple_build_asm, gimple gimple_build_assign_with_ops)
gimple_call_mark_uninlinable, gimple_call_cannot_inline_p): Remove
description of removed functions.
(gimple_build_assign, gimple_build_bind, gimple_build_call,
gimple_build_call_from_tree, gimple_build_call_vec,
gimple_build_catch, gimple_build_cond,
gimple_build_cond_from_tree, gimple_build_debug_bind,
gimple_build_eh_filter, gimple_build_label, gimple_build_goto,
gimple_build_omp_atomic_load, gimple_build_omp_atomic_store,
gimple_build_omp_continue, gimple_build_omp_critical,
gimple_build_omp_for, gimple_build_omp_parallel,
gimple_build_omp_sections, gimple_build_omp_single,
gimple_build_return, gimple_build_resx, gimple_build_switch,
gimple_build_try): Update return type within description to
reflect changes in gimple.h to using gimple subclasses.
(gimple_build_asm_vec): Update return type, params and
description.
(gimple_asm_ninputs): Update param.
(gimple_asm_noutputs, gimple_asm_nclobbers, gimple_asm_input_op
gimple_asm_set_input_op, gimple_asm_output_op
gimple_asm_set_output_op, gimple_asm_clobber_op,
gimple_asm_set_clobber_op, gimple_asm_string,
gimple_asm_volatile_p, gimple_asm_set_volatile, gimple_bind_vars,
gimple_bind_set_vars, gimple_bind_append_vars, gimple_bind_body,
gimple_bind_set_body, gimple_bind_add_stmt, gimple_bind_add_seq,
gimple_bind_block, gimple_bind_set_block, gimple_call_set_fn,
gimple_call_return_type, gimple_call_set_chain,
gimple_call_set_tail, gimple_call_tail_p,
gimple_call_copy_skip_args, gimple_catch_types,
gimple_catch_types_ptr, gimple_catch_handler,
gimple_catch_set_types, gimple_catch_set_handler,
gimple_cond_set_code, gimple_cond_set_lhs, gimple_cond_set_rhs,
gimple_cond_true_label, gimple_cond_set_true_label,
gimple_cond_set_false_label, gimple_cond_false_label,
gimple_cond_make_false, gimple_cond_make_true,
gimple_eh_filter_set_types, gimple_eh_filter_set_failure,
gimple_eh_must_not_throw_fndecl,
gimple_eh_must_not_throw_set_fndecl, gimple_label_label,
gimple_label_set_label, gimple_goto_set_dest,
gimple_omp_atomic_load_set_lhs, gimple_omp_atomic_load_lhs,
gimple_omp_atomic_load_set_rhs, gimple_omp_atomic_load_rhs,
gimple_omp_atomic_store_set_val, gimple_omp_atomic_store_val,
gimple_omp_continue_control_def,
gimple_omp_continue_control_def_ptr,
gimple_omp_continue_set_control_def,
gimple_omp_continue_control_use,
gimple_omp_continue_control_use_ptr,
gimple_omp_continue_set_control_use, gimple_omp_critical_name,
gimple_omp_critical_name_ptr, gimple_omp_critical_set_name,
gimple_omp_parallel_clauses_ptr, gimple_omp_parallel_set_clauses,
gimple_omp_parallel_child_fn, gimple_omp_parallel_child_fn_ptr,
gimple_omp_parallel_set_child_fn, gimple_omp_parallel_data_arg,
gimple_omp_parallel_data_arg_ptr,
gimple_omp_parallel_set_data_arg, gimple_omp_single_set_clauses,
gimple_phi_set_result, gimple_phi_set_arg, gimple_resx_region,
gimple_resx_set_region, gimple_return_retval,
gimple_return_set_retval, gimple_switch_num_labels,
gimple_switch_set_num_labels, gimple_switch_index,
gimple_switch_set_index, gimple_switch_label,
gimple_switch_set_label, gimple_switch_default_label,
gimple_switch_set_default_label, gimple_try_set_eval,
gimple_try_set_cleanup): Update initial param within description
to reflect changes in gimple.h to using gimple subclasses.
(Adding a new GIMPLE statement code): Update to reflect gimple
statement subclassing.

From-SVN: r218399

x86-64: Optimize access to globals in PIE with copy reloc

Normally, with -fPIE/-fpie, GCC accesses globals that are extern to the
module using the GOT.  This is two instructions, one to get the address
of the global from the GOT and the other to get the value.  If it turns
out that the global gets defined in the executable at link-time, it still
needs to go through the GOT as it is too late then to generate a direct
 access.

Examples:

foo.cc
------
int a_glob;
int main () {
  return a_glob; // defined in this file
}

With -O2 -fpie -pie, the generated code directly accesses the global via
PC-relative insn:

5e0   <main>:
   mov    0x165a(%rip),%eax        # 1c40 <a_glob>

foo.cc
------

extern int a_glob;
int main () {
  return a_glob; // defined in this file
}

With -O2 -fpie -pie, the generated code accesses global via GOT using
two memory loads:

6f0  <main>:
   mov    0x1609(%rip),%rax   # 1d00 <_DYNAMIC+0x230>
   mov    (%rax),%eax

This is true even if in the latter case the global was defined in the
executable through a different file.

Some experiments on google benchmarks shows that the extra memory loads
affects performance by 1% to 5%.

Solution - Copy Relocations:

When the linker supports copy relocations, GCC can always assume that
the global will be defined in the executable.  For globals that are truly
extern (come from shared objects), the linker will create copy relocations
and have them defined in the executable. Result is that no global access
needs to go through the GOT and hence improves performance.

This optimization only applies to undefined, non-weak global data.
Undefined, weak global data access still must go through the GOT.

This patch checks if linker supports PIE with copy reloc, which is
enabled in gold and bfd linker in bininutils 2.25, at configure time
and enables this optimization if the linker support is available.

gcc/

* configure.ac (HAVE_LD_PIE_COPYRELOC): Defined to 1 if
Linux/x86-64 linker supports PIE with copy reloc.
* config.in: Regenerated.
* configure: Likewise.

* config/i386/i386.c (legitimate_pic_address_disp_p): Allow
pc-relative address for undefined, non-weak, non-function
symbol reference in 64-bit PIE if linker supports PIE with
copy reloc.

* doc/sourcebuild.texi: Document pie_copyreloc target.

gcc/testsuite/

* gcc.target/i386/pie-copyrelocs-1.c: New test.
* gcc.target/i386/pie-copyrelocs-2.c: Likewise.
* gcc.target/i386/pie-copyrelocs-3.c: Likewise.
* gcc.target/i386/pie-copyrelocs-4.c: Likewise.

* lib/target-supports.exp (check_effective_target_pie_copyreloc):
New procedure.

Co-Authored-By: H.J. Lu <hongjiu.lu@intel.com>
From-SVN: r218397

re PR middle-end/56917 (-ftrapv detects a overflow wrongly.)

PR middle-end/56917
* fold-const.c (fold_unary_loc): Perform the negation in A's type
when transforming ~ (A - 1) or ~ (A + -1) to -A.

* c-c++-common/ubsan/pr56917.c: New test.

From-SVN: r218395

Makefile.in: Remove CLOOGLIB and CLOOGINC.

2014-12-04  Tobias Burnus  <burnus@net-b.de>

        * Makefile.in: Remove CLOOGLIB and CLOOGINC.

From-SVN: r218389

check_GNU_style.sh: Warn for incorrect number of spaces in function call only if 0 or 2+ spaces found.

2014-12-04  Thomas Preud'homme  <thomas.preudhomme@arm.com>

    contrib/
    * check_GNU_style.sh: Warn for incorrect number of spaces in function
    call only if 0 or 2+ spaces found.

From-SVN: r218382

match-and-simplify.texi: Update for recent changes.

2014-12-04  Richard Biener  <rguenther@suse.de>

* doc/match-and-simplify.texi: Update for recent changes.

From-SVN: r218372

ipa-prop.h (ipa_alignment): New type.

2014-12-04  Martin Jambor  <mjambor@suse.cz>

* ipa-prop.h (ipa_alignment): New type.
(ipa_jump_func): New field alignment.
(ipcp_transformation_summary) New type.
(ipcp_grow_transformations_if_necessary): Declare.
(ipa_node_agg_replacements): Removed.
(ipcp_transformations): Declare.
(ipcp_get_transformation_summary): New function.
(ipa_get_agg_replacements_for_node): Use it.
* ipa-cp.c (ipcp_param_lattices): New field alignment.
(print_all_lattices): Also print alignment.
(alignment_bottom_p): New function.
(set_alignment_to_bottom): Likewise.
(set_all_contains_variable): Also set alignment to bottom.
(initialize_node_lattices): Likewise.
(propagate_alignment_accross_jump_function): New function.
(propagate_constants_accross_call): Call it.
(ipcp_store_alignment_results): New function.
(ipcp_driver): Call it.
* ipa-prop.c (ipa_node_agg_replacements): Removed.
(ipcp_transformations): New.
(ipa_print_node_jump_functions_for_edge): Also print alignment.
(ipa_set_jf_unknown): New function.
(detect_type_change_from_memory_writes): Use ipa_set_jf_unknown.
(ipa_compute_jump_functions_for_edge): Also calculate alignment.
(update_jump_functions_after_inlining): Use ipa_set_jf_unknown.
(ipcp_grow_transformations_if_necessary): New function.
(ipa_set_node_agg_value_chain): Use ipcp_transformations.
(ipa_node_removal_hook): Likewise.
(ipa_node_duplication_hook): Also duplicate alignment results.
(ipa_write_jump_function): Also stream alignments.
(ipa_read_jump_function): Use ipa_set_jf_unknown, also stream
alignments.
(write_agg_replacement_chain): Renamed to
write_ipcp_transformation_info, also stream alignments.
(read_agg_replacement_chain): Renamed to
read_ipcp_transformation_info, also stream alignments.
(ipa_prop_write_all_agg_replacement): Renamed to
ipcp_write_transformation_summaries. Stream always.
(ipa_prop_read_all_agg_replacement): Renamed to
ipcp_read_transformation_summaries.
(ipcp_update_alignments): New function.
(ipcp_transform_function): Call it, free also alignments.

testsuite/
* gcc.dg/ipa/propalign-1.c: New test.
* gcc.dg/ipa/propalign-2.c: Likewise.

From-SVN: r218369

gimple-fold.c (replace_stmt_with_simplification): Properly fail when maybe_push_res_to_seq fails.

2014-12-04  Richard Biener  <rguenther@suse.de>

* gimple-fold.c (replace_stmt_with_simplification): Properly
fail when maybe_push_res_to_seq fails.

From-SVN: r218353

Add prefetch support for aarch64

From-SVN: r218349

install.texi: Remove mentions of cloog and ppl.

* doc/install.texi: Remove mentions of cloog and ppl.
* doc/invoke.texi: Likewise

From-SVN: r218348

re PR c++/56493 (Performance regression in google dense hashmap)

PR c++/56493
* convert.c (convert_to_real, convert_to_expr, convert_to_complex):
Handle COMPOUND_EXPR.

* c-c++-common/pr56493.c: New test.

From-SVN: r218345

[testsuite] Cache return value of check_effective_target_tiny

* lib/target-supports.exp (check_effective_target_tiny): Cache
predicate value.

From-SVN: r218344

builtins.c (target_newline): Export.

2014-12-04  Richard Biener  <rguenther@suse.de>

* builtins.c (target_newline): Export.
(target_percent_s_newline): Likewise.
(fold_builtin_1): Do not fold printf functions here.
(fold_builtin_2): Likewise.
(fold_builtin_3): Likewise, do not fold strncat.
(fold_builtin_strncat): Move to gimple-fold.c.
(fold_builtin_printf): Likewise.
* builtins.h (target_newline): Declare.
(target_percent_s_newline): Likewise.
* gimple-fold.c (gimple_fold_builtin_strncat): Move from
builtins.c and gimplify.
(gimple_fold_builtin_printf): Likewise.
(gimple_fold_builtin): Fold strncat, printf, printf_unlocked,
vprintf, printf_chk and vprintf_chk here.

From-SVN: r218343

re PR testsuite/64145 (gcc.dg/graphite/isl-codegen-loop-dumping.c UNRESOLVED after r217315.)

PR testsuite/64145
* gcc.dg/graphite/isl-codegen-loop-dumping.c: Restore
-fgraphite-identity.

From-SVN: r218342

isl.m4: Don't pass -DCLOOG_INT_GMP on islinc.

* config/isl.m4: Don't pass -DCLOOG_INT_GMP on islinc.
* configure: Regenerated.

From-SVN: r218341

re PR libstdc++/64140 (match_results.prefix() returns an incorrect result if regex_iterator holds a zero-length match)

PR libstdc++/64140
* include/bits/regex.tcc (regex_iterator<>::operator++): Update
prefix.matched after modifying prefix.first.
* testsuite/28_regex/iterators/regex_iterator/char/64140.cc: New
testcase.

From-SVN: r218340

rs6000.md (floatsidf2_internal): Use std::swap.

        * config/rs6000/rs6000.md (floatsidf2_internal): Use std::swap.
        (floatunssidf2_internal): Same.
        * config/rs6000/rs6000.c (rs6000_emit_vector_compare): Same.
        (rs6000_emit_int_cmove): Same.
        (rs6000_sched_reorder): Same.
        (altivec_expand_vec_perm_const): Same.
        (rs6000_expand_vec_perm_const_1): Same.

From-SVN: r218339

Daily bump.

From-SVN: r218338

re PR rtl-optimization/64010 ([msp430-elf] struct function dereference clobbers parameter passed to function)

PR rtl-optimization/64010
* reload.c (push_reload): Before reusing a register contained
in an operand as input reload register, ensure that it is not
used in CALL_INSN_FUNCTION_USAGE.

From-SVN: r218335