imm-devirt-1.C: Fix regexp in dg-final.

* g++.dg/ipa/imm-devirt-1.C: Fix regexp in dg-final.
* g++.dg/ipa/imm-devirt-2.C: Likewise.

From-SVN: r212313

re PR c/6940 (taking sizeof array parameter should trigger a warning)

PR c/6940
* doc/invoke.texi: Document -Wsizeof-array-argument.
c-family/
* c.opt (Wsizeof-array-argument): New option.
c/
* c-decl.c (grokdeclarator): Set C_ARRAY_PARAMETER.
* c-tree.h (C_ARRAY_PARAMETER): Define.
* c-typeck.c (c_expr_sizeof_expr): Warn when using sizeof on an array
function parameter.
cp/
* cp-tree.h (DECL_ARRAY_PARAMETER_P): Define.
* decl.c (grokdeclarator): Set DECL_ARRAY_PARAMETER_P.
* typeck.c (cxx_sizeof_expr): Warn when using sizeof on an array
function parameter.
testsuite/
* c-c++-common/Wsizeof-pointer-memaccess1.c: Use
-Wno-sizeof-array-argument.
* c-c++-common/Wsizeof-pointer-memaccess2.c: Likewise.
* g++.dg/warn/Wsizeof-pointer-memaccess-1.C: Likewise.
* gcc.dg/Wsizeof-pointer-memaccess1.c: Likewise.
* g++.dg/torture/Wsizeof-pointer-memaccess1.C: Likewise.
* g++.dg/torture/Wsizeof-pointer-memaccess2.C: Likewise.
* gcc.dg/torture/Wsizeof-pointer-memaccess1.c: Likewise.
* c-c++-common/sizeof-array-argument.c: New test.
* gcc.dg/vla-5.c: Add dg-warnings.
../libgomp/ * testsuite/libgomp.c/appendix-a/a.29.1.c (f): Add dg-warnings.

From-SVN: r212312

Daily bump.

From-SVN: r212311

Restore Solaris/SPARC bootstrap

* configure, config.h.in: Regenerate.
* config/fpu-sysv.h: Include <assert.h>.

From-SVN: r212307

Don't accept isl 0.10.

* configure.ac: Don't accept isl 0.10.
* configure: Regenerate.

From-SVN: r212306

wide-int.h (wide_int_storage): Change declaration from struct to class.

* wide-int.h (wide_int_storage): Change declaration from struct
to class.

From-SVN: r212305

cgraph.c (cgraph_create_indirect_edge): Update call of get_polymorphic_call_info.

* cgraph.c (cgraph_create_indirect_edge): Update call of
get_polymorphic_call_info.
* ipa-utils.h (get_polymorphic_call_info): Add parameter CALL.
(possible_polymorphic_call_targets): Add parameter call.
(decl_maybe_in_construction_p): New predicate.
(get_polymorphic_call_info): Add parameter call;
use decl_maybe_in_construction_p.
* gimple-fold.c (fold_gimple_assign): Update use of
possible_polymorphic_call_targets.
(gimple_fold_call): Likewise.
* ipa-prop.c: Inlcude calls.h
(ipa_binfo_from_known_type_jfunc): Check that known type is record.
(param_type_may_change_p): New predicate.
(detect_type_change_from_memory_writes): Break out from ...
(detect_type_change): ... this one; use
param_type_may_change_p.
(detect_type_change_ssa): Use param_type_may_change_p.
(compute_known_type_jump_func): Use decl_maybe_in_construction_p.

* g++.dg/ipa/devirt-26.C: Update testcase.
* g++.dg/ipa/imm-devirt-1.C: Update testcase.
* g++.dg/ipa/imm-devirt-2.C: Update testcase.

From-SVN: r212304

[ARM] PR target/49423

2014-07-05  Charles Baylis  <charles.baylis@linaro.org>

PR target/49423
* config/arm/arm-protos.h (arm_legitimate_address_p,
arm_is_constant_pool_ref): Add prototypes.
* config/arm/arm.c (arm_legitimate_address_p): Remove static.
(arm_is_constant_pool_ref) New function.
* config/arm/arm.md (unaligned_loadhis, arm_zero_extendhisi2_v6,
arm_zero_extendqisi2_v6): Use Uh constraint for memory operand.
(arm_extendhisi2, arm_extendhisi2_v6): Use Uh constraint for memory
operand. Remove pool_range and neg_pool_range attributes.
(arm_extendqihi_insn, arm_extendqisi, arm_extendqisi_v6): Remove
pool_range and neg_pool_range attributes.
* config/arm/constraints.md (Uh): New constraint.
(Uq): Don't allow constant pool references.

From-SVN: r212303

Daily bump.

From-SVN: r212302

resolve.c (resolve_assoc_var): Fix corank setting.

2014-07-04  Tobias Burnus  <burnus@net-b.de>

        * resolve.c (resolve_assoc_var): Fix corank setting.
        * trans-array.c (gfc_conv_descriptor_token): Change assert.
        for select-type temporaries.
        * trans-decl.c (generate_coarray_sym_init): Skip for
        attr.select_type_temporary.
        * trans-expr.c (gfc_conv_procedure_call): Fix for
        select-type temporaries.
        * trans-intrinsic.c (get_caf_token_offset): Ditto.
        (gfc_conv_intrinsic_caf_get, gfc_conv_intrinsic_caf_send): Set
        the correct dtype.
        * trans-types.h (gfc_get_dtype_rank_type): New.
        * trans-types.c (gfc_get_dtype_rank_type): Ditto.

2014-07-04  Tobias Burnus  <burnus@net-b.de>

        * gfortran.dg/coarray/coindexed_3.f90: New.

From-SVN: r212299

[Patch AArch64_be] Fix some vec_concat big-endian confusions

* config/aarch64/aarch64-simd.md (move_lo_quad_internal_<mode>): New.
(move_lo_quad_internal_be_<mode>): Likewise.
(move_lo_quad_<mode>): Convert to define_expand.
(aarch64_simd_move_hi_quad_<mode>): Gate on BYTES_BIG_ENDIAN.
(aarch64_simd_move_hi_quad_be_<mode>): New.
(move_hi_quad_<mode>): Use appropriate insn for BYTES_BIG_ENDIAN.
(aarch64_combinez<mode>): Gate on BYTES_BIG_ENDIAN.
(aarch64_combinez_be<mode>): New.
(aarch64_combine<mode>): Convert to define_expand.
(aarch64_combine_internal<mode>): New.
  (aarch64_simd_combine<mode>): Remove bogus RTL description.

From-SVN: r212298

Improve documentation of earlyclobber

2014-07-04  Tom de Vries  <tom@codesourcery.com>

* doc/md.texi (@subsection Constraint Modifier Characters): Clarify
combination of earlyclobber and read/write modifiers.

From-SVN: r212297

Fix constraint in vec_unpack_trunc_<mode>

2014-07-04  Tom de Vries  <tom@codesourcery.com>

* config/aarch64/aarch64-simd.md
(define_insn "vec_unpack_trunc_<mode>"): Fix constraint.

From-SVN: r212296

re PR target/61714 (configure --with-arch and --with-cpu are ignored on aarch64)

PR target/61714
* aarch64.h (OPTION_DEFAULT_SPECS): Define.

From-SVN: r212295

re PR middle-end/61654 (ICE in release_function_body, at cgraph.c:1699)

PR middle-end/61654
* cgraphunit.c (expand_thunk): Call free_dominance_info.

* g++.dg/opt/pr61654.C: New test.

From-SVN: r212293

re PR tree-optimization/61684 (ICE at -Os and above on x86_64-linux-gnu in tree check: expected ssa_name, have addr_expr in recognize_single_bit_test, at tree-ssa-ifcombine.c:238)

PR tree-optimization/61684
* tree-ssa-ifcombine.c (recognize_single_bit_test): Make sure
rhs1 of conversion is a SSA_NAME before using SSA_NAME_DEF_STMT on it.

* gcc.c-torture/compile/pr61684.c: New test.

From-SVN: r212290

Move fp-as-gp optimization to nds32-fp-as-gp.c module.

gcc/
* config/nds32/nds32.c (nds32_have_prologue_p): Move to ...
(nds32_symbol_load_store_p): Move to ...
(nds32_fp_as_gp_check_available): Move to ...
* config/nds32/nds32-fp-as-gp.c: ... here.
* config/nds32/nds32-protos.h (nds32_symbol_load_store_p): Remove
extern declaration.

Co-Authored-By: Kito Cheng <kito@0xlab.org>
Co-Authored-By: Monk Chiang <sh.chiang04@gmail.com>
From-SVN: r212288

Move memory related implementation to nds32-memory-manipulation.c module.

gcc/
* config/nds32/nds32.c (nds32_expand_load_multiple): Move to ...
(nds32_expand_store_multiple): Move to ...
(nds32_expand_movmemqi): Move to ...
* config/nds32/nds32-memory-manipulation.c: ... here.

Co-Authored-By: Kito Cheng <kito@0xlab.org>
Co-Authored-By: Monk Chiang <sh.chiang04@gmail.com>
From-SVN: r212287

Move some external functions used by machine description patterns to nds32-md-auxiliary.c module.

gcc/
* config/nds32/nds32.c (nds32_byte_to_size): Move to ...
(nds32_output_casesi_pc_relative): Move to ...
(nds32_output_casesi): Move to ...
(nds32_mem_format): Move to ...
(nds32_output_16bit_store): Move to ...
(nds32_output_16bit_load): Move to ...
(nds32_output_32bit_store): Move to ...
(nds32_output_32bit_load): Move to ...
(nds32_output_32bit_load_s): Move to ...
(nds32_output_stack_push): Move to ...
(nds32_output_stack_pop): Move to ...
* config/nds32/nds32-md-auxiliary.c: ... here.

Co-Authored-By: Kito Cheng <kito@0xlab.org>
Co-Authored-By: Monk Chiang <sh.chiang04@gmail.com>
From-SVN: r212286

Preparation of nds32-pipelines-auxiliary.c module for future implementation.

gcc/
* config/nds32/nds32-pipelines-auxiliary.c: Add comment to describe
the purpose of this file.

Co-Authored-By: Ling-Hua Tseng <uranus@tinlans.org>
From-SVN: r212285

Move cost calculation to nds32-cost.c module.

gcc/
* config/nds32/nds32.c (nds32_rtx_costs): Move implementation to ...
(nds32_address_cost): Move implementation to ...
* config/nds32/nds32-cost.c: ... here.
* config/nds32/nds32-protos.h (nds32_rtx_costs_impl): Declare.
(nds32_address_cost_impl): Declare.

Co-Authored-By: Kito Cheng <kito@0xlab.org>
Co-Authored-By: Monk Chiang <sh.chiang04@gmail.com>
From-SVN: r212284

Move some helper functions of predicates and constraints to nds32-predicates.c module.

gcc/
* config/nds32/nds32.c
(nds32_consecutive_registers_load_store_p): Move to ...
(nds32_valid_multiple_load_store): Move to ...
(nds32_valid_stack_push_pop): Move to ...
(nds32_can_use_bclr_p): Move to ...
(nds32_can_use_bset_p): Move to ...
(nds32_can_use_btgl_p): Move to ...
(nds32_can_use_bitci_p): Move to ...
* config/nds32/nds32-predicates.c: ... here.

Co-Authored-By: Kito Cheng <kito@0xlab.org>
Co-Authored-By: Monk Chiang <sh.chiang04@gmail.com>
From-SVN: r212283

Move builtin functions implementation to nds32-intrinsic.c module.

gcc/
* config/nds32/nds32.c
(nds32_expand_builtin_null_ftype_reg): Move to ...
(nds32_expand_builtin_reg_ftype_imm): Move to ...
(nds32_expand_builtin_null_ftype_reg_imm): Move to ...
(nds32_init_builtins): Move implementation to ...
(nds32_expand_builtin): Move implementation to ...
* config/nds32/nds32-intrinsic.c: ... here.
* config/nds32/nds32-protos.h (nds32_init_builtins_impl): Declare.
(nds32_expand_builtin_impl): Declare.

Co-Authored-By: Kito Cheng <kito@0xlab.org>
Co-Authored-By: Monk Chiang <sh.chiang04@gmail.com>
From-SVN: r212282

Move ISR impelentation to nds32-isr.c module.

gcc/
* config/nds32/nds32.c (nds32_emit_section_head_template): Move to ...
(nds32_emit_section_tail_template): Move to ...
(nds32_emit_isr_jmptbl_section): Move to ...
(nds32_emit_isr_vector_section): Move to ...
(nds32_emit_isr_reset_conten): Move to ...
(nds32_check_isr_attrs_conflict): Move to ...
(nds32_construct_isr_vectors_information): Move to ...
(nds32_asm_file_start): Move implementation to ...
(nds32_asm_file_end): Move implementation to ...
* config/nds32/nds32-isr.c: ... here.
* config/nds32/nds32-protos.h
(nds32_check_isr_attrs_conflict): Declare.
(nds32_construct_isr_vectors_information): Declare.
(nds32_asm_file_start_for_isr): Declare.
(nds32_asm_file_end_for_isr): Declare.

Co-Authored-By: Kito Cheng <kito@0xlab.org>
Co-Authored-By: Monk Chiang <sh.chiang04@gmail.com>
From-SVN: r212281

Add several new files for preparation of providing modules
that are going to be separated from nds32.c source.

gcc/
* config.gcc (nds32*): Add new modules to extra_objs.
(nds32le-*-*): Use t-nds32 makefile fragment for new modules.
(nds32be-*-*): Likewise.
* config/nds32/nds32-cost.c: New file.
* config/nds32/nds32-fp-as-gp.c: New file.
* config/nds32/nds32-intrinsic.c: New file.
* config/nds32/nds32-isr.c: New file.
* config/nds32/nds32-md-auxiliary.c: New file.
* config/nds32/nds32-memory-manipulation.c: New file.
* config/nds32/nds32-pipelines-auxiliary.c: New file.
* config/nds32/nds32-predicates.c: New file.
* config/nds32/t-nds32: New file.

Co-Authored-By: Kito Cheng <kito@0xlab.org>
Co-Authored-By: Monk Chiang <sh.chiang04@gmail.com>
From-SVN: r212280

Normalize interface for all *-dg-runtest.

gcc/testsuite/
* lib/g++-dg.exp (g++-dg-runtest): Change interface to match
dg-runtest's.  Adapt all callers.
* lib/gcc-dg.exp (gcc-dg-runtest): Likewise.
* lib/gfortran-dg.exp (gfortran-dg-runtest): Likewise.
* lib/go-dg.exp (go-dg-runtest): Likewise.
* lib/obj-c++-dg.exp (obj-c++-dg-runtest): Likewise.
* lib/objc-dg.exp (objc-dg-runtest): Likewise.
libffi/
* testsuite/lib/libffi.exp (libffi-dg-runtest): Change interface
match to dg-runtest's.

From-SVN: r212278

Daily bump.

From-SVN: r212277

re PR tree-optimization/61682 (wrong code at -O3 on x86_64-linux-gnu)

PR tree-optimization/61682
* wide-int.cc (wi::mul_internal): Handle high correctly
for umul_ppmm using cases and when one of the operands is
equal to 1.

* gcc.c-torture/execute/pr61682.c: New test.

From-SVN: r212273

libgomp.exp (libgomp_target_compile): If $source matches regex $lang_source_re, add $lang_include_flags to options.

* testsuite/lib/libgomp.exp (libgomp_target_compile): If $source
matches regex $lang_source_re, add $lang_include_flags to options.
* testsuite/libgomp.c/c.exp: Unset lang_include_flags.
* testsuite/libgomp.c++/c++.exp: Likewise.
* testsuite/libgomp.fortran/fortran.exp: Likewise.  Set lang_source_re
and lang_include_flags instead of adding -fintrinsic-modules-path= to
ALWAYS_CFLAGS.
* testsuite/libgomp.graphite/graphite.exp: Unset lang_include_flags.

From-SVN: r212268

Fix the shift patterns, and add test

Firstly, it adds back the split conditions that I accidentally removed.
Without it the dot insns are never generated, or rather, always split
back to a separate compare instruction.

Secondly, the shift amount should be SI always, not GPR, or GCC will
insert a zero-extend at expand time that it cannot get rid of later.

The test tests whether dot-form instructions are generated for both
"dot" and "dot2" cases, that is, with just a CC output or also a GPR
output; for all four basic shifts, with a register amount or an
immediate amount.  It also tests for superfluous zero-extends.  This
also tests if combine "simplifies" the rotates to right-rotates, which
it shouldn't do anymore.

From-SVN: r212267

Explain gfortran-dg-runtest usage in libgomp Fortran testing.

libgomp/
* testsuite/libgomp.fortran/fortran.exp: Explain
gfortran-dg-runtest usage.

From-SVN: r212266

arm.md (arch): Add armv6_or_vfpv3.

* arm.md (arch): Add armv6_or_vfpv3.
(arch_enabled): Add test for the above.
* vfp.md (divsf_vfp, divdf_vfp): Add earlyclobber when code can run
on VFP9.
(sqrtsf_vfp, sqrtdf_vfp): Likewise.

From-SVN: r212265

gcov-io.c (gcov_read_words): Don't call memmove if excess is 0.

* gcov-io.c (gcov_read_words): Don't call memmove if excess is 0.
* data-streamer-in.c (streamer_read_hwi): Shift UHWI 1 instead of
HWI 1 and negate the unsigned value.
* expmed.c (expand_sdiv_pow2): For modes wider than word always
use AND instead of shift.
* wide-int-print.cc (print_decs): Negate UHWI instead of HWI.
c-family/
* c-ada-spec.c (dump_ada_nodes): Don't call qsort if
comments->count <= 1, as comments->entries might be NULL.

From-SVN: r212264

ieee_1.F90: Mark variables as volatile to avoid optimization.

* gfortran.dg/ieee/ieee_1.F90: Mark variables as volatile to
avoid optimization.

From-SVN: r212261

invoke.texi (-fsanitize=bounds): Tweak wording.

* doc/invoke.texi (-fsanitize=bounds): Tweak wording.
(-fsanitize=float-divide-by-zero): Move to the table with
-fsanitize=undefined suboptions.
(-fsanitize=float-cast-overflow): Likewise.

From-SVN: r212260

rs6000.c (rs6000_adjust_atomic_subword): Use BYTES_BIG_ENDIAN rather than WORDS_BIG_ENDIAN to check for byte endianness.

* config/rs6000/rs6000.c (rs6000_adjust_atomic_subword): Use
BYTES_BIG_ENDIAN rather than WORDS_BIG_ENDIAN to check for byte
endianness.

From-SVN: r212258

scanner.c (skip_free_comments): Fix indentation.

2014-07-03  Tobias Burnus  <burnus@net-b.de>

        * scanner.c (skip_free_comments): Fix indentation.

From-SVN: r212257

loop-invariant.c (struct invariant): Add a new member: eqno;

ChangeLog:
2014-07-03  Zhenqiang Chen  <zhenqiang.chen@linaro.org>

* loop-invariant.c (struct invariant): Add a new member: eqno;
(find_identical_invariants): Update eqno;
(create_new_invariant): Init eqno;
(get_inv_cost): Compute comp_cost with eqno;

testsuite/ChangeLog:
2014-07-03  Zhenqiang Chen  <zhenqiang.chen@linaro.org>

* gcc.target/arm/identical-invariants.c: New test.

From-SVN: r212256

Daily bump.

From-SVN: r212253

Only transform rotate to rotatert and v.v. if target has both

Many targets do not have both rotate and rotatert.  Of the 47 targets
in the tree, 17 have both, 9 have only rotate, 2 have only rotatert, and
19 have neither (this is based on "grep -wil" so it can be slightly off).

rs6000 has only rotate, and mips has only rotatert.  For such targets
simplifying rotate to rotatert and vice versa is not simplifying things
at all.  rs6000 has already way too many rotate patterns (some days it
seems like two thousand, but it is somewhat less in reality still); I
would prefer not to double that again.

So, this patch makes genrecog define HAVE_rotate and HAVE_rotatert if
those RTL codes are mentioned anywhere in the machine description, and
then does the transformation in simplify-rtx.c only if both these flags
are set.

From-SVN: r212239

* config/fpu-sysv.h (void set_fpu): Remove stalled comment.

From-SVN: r212235

c-decl.c (duplicate_decls): CLear DECL_STRUCT_FUNCTION before releasing symbol.

* c-decl.c (duplicate_decls): CLear DECL_STRUCT_FUNCTION before
releasing symbol.

* gcc.c-torture/compile/section.c: New testcase.

Co-Authored-By: Chen Gang <gang.chen.5i5j@gmail.com>
From-SVN: r212234

re PR go/61620 (FAIL: go.test/test/fixedbugs/bug242.go execution,  -O2 -g)

PR go/61620

runtime: Don't free tiny blocks in map deletion.

The memory allocator now has a special case for tiny blocks
(smaller than 16 bytes) and they can not be explicitly freed.

From-SVN: r212233

add bugzilla links

From-SVN: r212232

add test

From-SVN: r212231

Support mode toggle.

* mode-switching.c (struct bb_info): Add mode_out, mode_in caches.
(make_preds_opaque): Delete.
(clear_mode_bit, mode_bit_p, set_mode_bit): New macros.
(commit_mode_sets): New function.
(optimize_mode_switching): Handle current_mode to mode_switching_emit.
Process all modes at once.
* basic-block.h (pre_edge_lcm_avs): Declare.
* lcm.c (pre_edge_lcm_avs): Renamed from pre_edge_lcm.
Call clear_aux_for_edges. Fix comments.
(pre_edge_lcm): New wrapper function to call pre_edge_lcm_avs.
(pre_edge_rev_lcm): Idem.
* config/epiphany/epiphany.c (emit_set_fp_mode): Add prev_mode parameter.
* config/epiphany/epiphany-protos.h (emit_set_fp_mode): Idem.
* config/epiphany/resolve-sw-modes.c (pass_resolve_sw_modes::execute): Idem.
* config/i386/i386.c (x96_emit_mode_set): Idem.
* config/sh/sh.c (sh_emit_mode_set): Likewise. Handle PR toggle.
* config/sh/sh.md (toggle_pr):  Defined if TARGET_FPU_SINGLE.
(fpscr_toggle) Disallow from delay slot.
* target.def (emit_mode_set): Add prev_mode parameter.
* doc/tm.texi: Regenerate.

From-SVN: r212230

decl.c (variable_decl): Reject old style initialization for derived type components.

* decl.c (variable_decl): Reject old style initialization
for derived type components.

* gfortran.dg/oldstyle_5.f: New test.

Co-Authored-By: Fritz Reese <Reese-Fritz@zai.com>
From-SVN: r212227

* config/fpu-glibc.h: Fix comment about FE_DENORMAL.

From-SVN: r212226

[AArch64] Removed unused variable i in aarch64_expand_vec_perm

* config/aarch64/aarch64.c (aarch64_expand_vec_perm): Delete unused
variable i.

From-SVN: r212225

configure.host (ieee_flags): Add -mieee for alpha*.

libgfortran/ChangeLog:

* configure.host (ieee_flags): Add -mieee for alpha*.

* config/fpu-glibc.h (support_fpu_rounding_mode): Correctly handle
GFC_FPE_UPWARD, GFC_FPE_DOWNWARD and GFC_FPE_TOWARDZERO.
* config/fpu-aix.h (support_fpu_rounding_mode): Ditto.

testsuite/ChangeLog:

* gfortran.dg/ieee/ieee_1.F90 (dg-additional-options): Remove -O0.
* gfortran.dg/ieee/ieee_rounding_1.f90 (dg-additional-options): Add.

From-SVN: r212224

pt.c (convert_template_argument): Use inform instead of error in three places.

/cp
2014-07-02  Paolo Carlini  <paolo.carlini@oracle.com>

* pt.c (convert_template_argument): Use inform instead of error in
three places.

/testsuite
2014-07-02  Paolo Carlini  <paolo.carlini@oracle.com>

* g++.dg/cpp0x/variadic-ex10.C: Adjust for inform instead of error.
* g++.dg/cpp0x/variadic-ex14.C: Likewise.
* g++.dg/parse/error11.C: Likewise.
* g++.old-deja/g++.brendan/template17.C: Likewise.

From-SVN: r212223

ipa-utils.h (method_class_type, [...]): Constify.

* ipa-utils.h (method_class_type, vtable_pointer_value_to_binfo,
vtable_pointer_value_to_vtable): Constify.
(contains_polymorphic_type_p): Declare.
* ipa-devirt.c (method_class_type, vtable_pointer_value_to_binfo,
vtable_pointer_value_to_vtable): Constify.
(contains_polymorphic_type_p): New predicate.
* ipa-prop.c (ipa_set_jf_known_type): Allow types containing
polymorphic types.
(ipa_set_ancestor_jf): Likewise.
(detect_type_change): Return false in easy cases.
(compute_complex_assign_jump_func): Require type to contain
polymorphic type.
(compute_known_type_jump_func): Likewise.

From-SVN: r212222

get_array.f90: Add missing SYNC ALL.

2014-07-02  Tobias Burnus  <burnus@net-b.de>

        * gfortran.dg/coarray/get_array.f90: Add missing SYNC ALL.
        * gfortran.dg/coarray/send_array.f90: Ditto.
        * gfortran.dg/coarray/sendget_array.f90: Ditto.

From-SVN: r212220

tree.c (decls_same_for_odr, [...]): Remove.

* tree.c (decls_same_for_odr, decls_same_for_odr,
types_same_for_odr): Remove.
(type_in_anonymous_namespace_p): Constify argument.
* tree.h (types_same_for_odr, type_in_anonymous_namespace_p):
Constify.
* ipa-devirt.c (odr_type_d): Add ODR_VIOLATED field.
(main_odr_variant): New function.
(hash_type_name): Make static; update assert; do not ICE on
non-records.
(types_same_for_odr): Bring here from tree.c; simplify
and remove old structural comparing code that doesn't work
for templates.
(odr_hasher::equal): Update assert.
(add_type_duplicate): Return true when bases should be computed;
replace incomplete loader by complete; do not output duplicated
warnings; do not ICE on non-records; set odr_violated flag.
(get_odr_type): Be ready to replace incomplete type by complete
one; work on ODR variants instead of main variants; reorder item
in array so bases have still smaller indexes.
(dump_type_inheritance_graph): Be ready for holdes in odr_types
array.
(possible_polymorphic_call_targets): Do not ICE when BINFO is NULL.

From-SVN: r212218

Daily bump.

From-SVN: r212217

runtime: introduce build targets for running benchmarks

This introduces the "bench" build target, which can be used to run
all benchmarks.

It is also possible to run subsets of benchmarks with the
"package/check" build targets by setting GOBENCH to a matching regex.

From-SVN: r212212

This patch fixes a couple of ICEs when using -gsplit-dwarf.

When compiling a small-enough compilation unit that has no address table
entries, but complex enough that -freorder-blocks-and-partition produces
location lists, dwarf2out_finish does not call index_location_lists, but
optimize_location_lists will later assume that the addr_index_table has
been indexed.
Google ref: b/15417905

When resolve_addr_in_expr replaces a CONST_STRING rtx, it directly
updates the pointer to the old expression with the new one. In the
case of a DW_OP_GNU_addr_index or DW_OP_GNU_const_index, that pointer
may be in an address table entry, which is keyed by the rtx. Instead
of directly replacing the pointer, we need to remove the old address
table entry (i.e., decrement its reference count), and add a new one.
Google ref: b/15957101

gcc/
* dwarf2out.c (remove_addr_table_entry): Remove unnecessary hash table
        lookup.
(resolve_addr_in_expr): When replacing the rtx in a location list
        entry, get a new address table entry.
(dwarf2out_finish): Call index_location_lists even if there are no
        addr_index_table entries yet.

From-SVN: r212211

gcc/testsuite/

* gcc.target/aarch64/aapcs64/aapcs64.exp:
(additional_flags_for_func_ret): New variable based on $additional_flags
with -fno-use-caller-save.
(func-ret-*.c): Use the new variable.

Co-Authored-By: Yufeng Zhang <yufeng.zhang@arm.com>
From-SVN: r212206

re PR c++/59637 ([c++11] ICE with decltype and destructor call in template)

2014-07-01  Paolo Carlini  <paolo.carlini@oracle.com>

PR c++/59637
* g++.dg/cpp0x/decltype60.C: New.

From-SVN: r212205

revert r212197 for apparently not actually being obvious

From-SVN: r212200

Revert "PR preprocessor/60723 - missing system-ness marks for macro tokens"

This reverts commit 747e04f26ac3fb775bfc9af61e9170b9461b6cfc.

From-SVN: r212199

Correct ChangeLog entries from r212041.

From-SVN: r212198

fix -Wunused-parameter warning in winnt.c

gcc/
* config/i386/winnt.c (i386_pe_section_type_flags): Removed name
of unused argument.

From-SVN: r212197

[AArch64] Implement some vca*_f[32,64] intrinsics.

* config/aarch64/arm_neon.h (vcage_f64): New intrinsic.
(vcagt_f64): Likewise.
(vcale_f64): Likewise.
(vcaled_f64): Likewise.
(vcales_f32): Likewise.
(vcalt_f64): Likewise.
(vcaltd_f64): Likewise.
(vcalts_f32): Likewise.

* gcc.target/aarch64/simd/vcage_f64.c: New test.
* gcc.target/aarch64/simd/vcagt_f64.c: Likewise.
* gcc.target/aarch64/simd/vcale_f64.c: Likewise.
* gcc.target/aarch64/simd/vcaled_f64.c: Likewise.
* gcc.target/aarch64/simd/vcales_f32.c: Likewise.
* gcc.target/aarch64/simd/vcalt_f64.c: Likewise.
* gcc.target/aarch64/simd/vcaltd_f64.c: Likewise.
* gcc.target/aarch64/simd/vcalts_f32.c: Likewise.

From-SVN: r212196

pr59867.C: Fix target selector.

2014-07-01  Paolo Carlini  <paolo.carlini@oracle.com>

* g++.dg/cpp1y/pr59867.C: Fix target selector.

From-SVN: r212195

PR preprocessor/60723 - missing system-ness marks for macro tokens

When a system macro is expanded in a non-system file during
out-of-line preprocessing, it can happen that the preprocessor forgets
to emit line markers to express the system-ness status of tokens that
come after the expansion of the macro.

That can lead to situations where the entire non-system file can be
considered as being a system file and thus have its warnings be
discarded during the compilation of the resulting preprocessed file.

My understanding is that this is due to the preprocessor not
systematically detecting (and reporting) the change in system-ness of
tokens.

And this is what this patch does.  Each time the system-ness of a
given token is different from the previous token that was emitted by
the preprocessor, it emits a line marker for the sole purpose of
marking the new system-ness of the subsequent tokens to come.

Bootstrapped and tested on x86_64-unknown-linux-gnu against trunk.

gcc/c-family/ChangeLog:
* c-ppoutput.c (struct print::prev_was_system_token): New data
member.
(init_pp_output): Initialize it.
(maybe_print_line_1, maybe_print_line, print_line_1, print_line)
(do_line_change): Return a flag saying if a line marker was
emitted or not.
(scan_translation_unit): Detect if the system-ness of the token we
are about to emit is different from the one of the previously
emitted token.  If so, emit a line marker.  Avoid emitting
useless adjacent line markers.
(scan_translation_unit_directives_only): Adjust.

gcc/testsuite/ChangeLog:
* gcc.dg/cpp/syshdr{4,5}.{c,h}: New test files.

Signed-off-by: Dodji Seketeli <dodji@redhat.com>
From-SVN: r212194

invoke.texi: Document -Wint-conversion.

* doc/invoke.texi: Document -Wint-conversion.
c-family/
* c.opt (Wint-conversion): New option.
c/
* c-typeck.c (convert_for_assignment): Pass OPT_Wint_conversion
instead of 0 to WARN_FOR_ASSIGNMENT.
testsuite/
* gcc.dg/Wint-conversion.c: New test.

From-SVN: r212193

re PR c/58286 (Need option to make incompatible pointer type warnings into compiler errors)

PR c/58286
* doc/invoke.texi: Document -Wincompatible-pointer-types.
c-family/
* c.opt (Wincompatible-pointer-types): New option.
c/
* c-typeck.c (convert_for_assignment): Pass
OPT_Wincompatible_pointer_types instead of 0 to WARN_FOR_ASSIGNMENT.
testsuite/
* gcc.dg/Wincompatible-pointer-types.c: New test.

From-SVN: r212192

IPA REF alias refactoring

* cgraph.h (iterate_direct_aliases): New function.
(FOR_EACH_ALIAS): New macro iterates all direct aliases for a node.
* cgraph.c (cgraph_for_node_thunks_and_aliases): Usage of
FOR_EACH_ALIAS added.
(cgraph_for_node_and_aliases): Likewise.
* cgraphunit.c (assemble_thunks_and_aliases): Likewise.
* ipa-inline.c (reset_edge_caches): Likewise.
(update_caller_keys): Likewise.
* trans-mem.c (ipa_tm_execute): Likewise.
*varpool.c (varpool_analyze_node): Likewise.
(varpool_for_node_and_aliases): Likewise.
* ipa-ref.h (first_alias): New function.
(last_alias): Likewise.
(has_aliases_p): Likewise.
* ipa-ref.c (ipa_ref::remove_reference): Removal function
is sensitive to IPA_REF_ALIASes.
* symtab.c (symtab_node::add_reference): Node of IPA_REF_ALIAS type
are put at the beginning of the list.
(symtab_node::iterate_direct_aliases): New function.

* lto-partition.c (add_symbol_to_partition_1): Usage of
FOR_EACH_ALIAS added.

From-SVN: r212191

scanasm.exp (dg-function-on-line): Add MicroBlaze specific pattern.

2014-06-30  David Holsgrove <david.holsgrove@xilinx.com>

* gcc/testsuite/lib/scanasm.exp (dg-function-on-line): Add
MicroBlaze specific pattern.

--This line, and those below,

M    ChangeLog
M    lib/scanasm.exp

From-SVN: r212189

* ChangeLog: Correct the PR id from 61061 to 61601.

From-SVN: r212187

re PR c++/58781 (Unicode strings broken in a strange way)

cp/

2014-06-28  Edward Smith-Rowland  <3dw4rd@verizon.net>

PR c++/58781
PR c++/60249
PR c++/59867
* parser.c (cp_parser_userdef_string_literal()): Take a tree
not a cp_token*. (cp_parser_string_literal(): Don't hack
the token stream!

testsuite/

2014-06-28  Edward Smith-Rowland  <3dw4rd@verizon.net>

PR c++/58781
PR c++/60249
PR c++/59867
* testsuite/g++.dg/cpp0x/pr58781.C: New.
* testsuite/g++.dg/cpp0x/pr60249.C: New.
* testsuite/g++.dg/cpp1y/pr59867.C: New.

From-SVN: r212186

re PR testsuite/61061 (FAIL: g++.dg/inherit/covariant7.C)

PR libstdc++/61061
PR libstdc++/61582
* include/bits/regex_automaton.h (_NFA<>::_M_insert_state): Add
a NFA state limit. If it's exceeded, regex_constants::error_space
will be throwed.
* include/bits/regex_automaton.tcc (_StateSeq<>::_M_clone): Use
map (which is sparse) instead of vector. This reduce n times clones'
cost from O(n^2) to O(n).
* include/std/regex: Add map dependency.
* testsuite/28_regex/algorithms/regex_match/ecma/char/61601.cc: New
testcase.

From-SVN: r212185

re PR libstdc++/61424 (std::regex matches right to left, not leftmost longest)

PR libstdc++/61424
* include/bits/regex.tcc (__regex_algo_impl<>): Use DFS for ECMAScript,
not just regex containing back-references.
* include/bits/regex_compiler.tcc (_Compiler<>::_M_disjunction):
exchange _M_next and _M_alt for alternative operator,
making matching from left to right.
* include/bits/regex_executor.h (_State_info<>::_M_get_sol_pos):
Add position tracking fom DFS.
* include/bits/regex_executor.tcc (_Executor<>::_M_main_dispatch,
_Executor<>::_M_dfs): Likewise.
* include/bits/regex_scanner.h: Remove unused enum entry.
* testsuite/28_regex/algorithms/regex_search/61424.cc: New
testcase from PR.

From-SVN: r212184

* MAINTAINERS: Update my email address.

From-SVN: r212183

Daily bump.

From-SVN: r212182

revert: tree-streamer-out.c (pack_ts_type_common_value_fields): Stream if type is complete.

Revert:
* tree-streamer-out.c (pack_ts_type_common_value_fields): Stream if type
is complete.
(write_ts_type_common_tree_pointers): Do not stream fields not set for incomplete
types; do not stream duplicated fields for variants; sanity check that variant
and type match.
(write_ts_type_non_common_tree_pointers): Likewise.
* tree-streamer-in.c (unpack_ts_type_common_value_fields): Mark in TYPE_SIZE whether
type is complete.
(lto_input_ts_type_common_tree_pointers): Do same changes as in
write_ts_type_common_tree_pointers
(lto_input_ts_type_non_common_tree_pointers): Likewise.

* lto.c (lto_copy_fields_not_streamed): New function.
(compare_tree_sccs_1): Do not compare fields shared in between type
and variant.
(lto_read_decls): Fixup types first before inserting into hash.

From-SVN: r212179

var-tracking.c (add_stores): Return instead of asserting if old and new values for conditional store are the...

* var-tracking.c (add_stores): Return instead of asserting if old
and new values for conditional store are the same.

From-SVN: r212178

re PR ipa/61659 (Extra undefined symbol because of devirtualization)

PR c++/61659
PR lto/53808
gcc/cp
* decl2.c (maybe_emit_vtables): Mark all vtable entries if
devirtualizing.
* init.c (build_vtbl_address): Don't mark destructor.
* class.c (finish_struct_1): Add all classes to keyed_classes
if devirtualizing.
libstdc++-v3/
* libsupc++/cxxabi.h (class __pbase_type_info): __pointer_catch
is pure, not inline.

From-SVN: r212174

re PR rtl-optimization/61608 (FAIL: gcc.target/arm/epilog-1.c scan-assembler tests)

PR rtl-opt/61608

        PR target/39284
        * bb-reorder.c (pass_duplicate_computed_gotos::execute): Cleanup
        the cfg if there were any changes.
        * passes.def: Revert move of peephole2 after reorder_blocks;
        move duplicate_computed_gotos before peephole2.

From-SVN: r212172

except.c (emit_note_eh_region_end): New helper function.

* except.c (emit_note_eh_region_end): New helper function.
(convert_to_eh_region_ranges): Use emit_note_eh_region_end to
emit EH_REGION_END note.
* jump.c (cleanup_barriers): Do not split a call and its
corresponding CALL_ARG_LOCATION note.

From-SVN: r212171

ChangeLog: Fix whitespace.

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

From-SVN: r212170

round_4.f90: Skip for powerpc*-*-linux* since the test requires greater precision than the...

2014-06-30  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>

* gfortran.dg/round_4.f90: Skip for powerpc*-*-linux* since the
test requires greater precision than the current PowerPC long
double implementation supports.

From-SVN: r212169

re PR c++/61647 (internal compiler error: in push_access_scope, at cp/pt.c:219 for a c++ header, clang++ 3.4 generate .pch without error)

PR c++/61647
* pt.c (type_dependent_expression_p): Check BASELINK_OPTYPE.

From-SVN: r212168

re PR c++/61566 (ICE in write_unscoped_name)

PR c++/61566
* mangle.c (decl_mangling_context): Look through a TEMPLATE_DECL.

From-SVN: r212165

decl.c (build_ptrmemfunc_type): Don't give a PMF RECORD_TYPE TYPE_BINFO or TYPE_LANG_SPECIFIC.

* decl.c (build_ptrmemfunc_type): Don't give a PMF RECORD_TYPE
TYPE_BINFO or TYPE_LANG_SPECIFIC.
* cp-tree.h (TYPE_PTRMEMFUNC_FLAG): Use TYPE_LANG_FLAG_2.
(TYPE_PTRMEMFUNC_P): Don't expect TYPE_LANG_SPECIFIC.
* typeck.c (build_ptrmemfunc_access_expr): Don't use lookup_member.
* pt.c (unify): Also check whether the argument is a PMF.

From-SVN: r212164

re PR c++/54891 ([C++11] lambda-expression and explicit type conversion (cast notation))

/cp
2014-06-30  Paolo Carlini  <paolo.carlini@oracle.com>

PR c++/54891
* parser.c (cp_parser_tokens_start_cast_expression): In C++11
a '[' can also start a primary-expression.
(cp_parser_cast_expression): Parse a cast-expression only tentatively
when cp_parser_tokens_start_cast_expression returns -1.

/testsuite
2014-06-30  Paolo Carlini  <paolo.carlini@oracle.com>

PR c++/54891
* g++.dg/cpp0x/lambda/lambda-cast1.C: New.

From-SVN: r212162

re PR c++/51400 ([c++0x] ICE with constexpr and attribute noreturn)

/c-family
2014-06-30  Paolo Carlini  <paolo.carlini@oracle.com>

PR c++/51400
* c-common.c (handle_noreturn_attribute, handle_const_attribute):
Do not discard TYPE_QUALS of type.

/testsuite
2014-06-30  Paolo Carlini  <paolo.carlini@oracle.com>

PR c++/51400
* g++.dg/cpp0x/constexpr-attribute3.C: New.

From-SVN: r212155

re PR c++/61539 (ICE: in unify_one_argument, at cp/pt.c:15465)

PR c++/61539
* pt.c (unify_one_argument): Type/expression mismatch just causes
deduction failure.

From-SVN: r212154

semantics.c (simplify_aggr_init_expr): Remove remnants of 2014-04-11 change.

* semantics.c (simplify_aggr_init_expr): Remove remnants of
2014-04-11 change.

From-SVN: r212153

tree-ssa-threadedge.c (simplify_control_stmt_condition): Look deeper into the SSA_NAME_VALUE chain.

tree-optimization/61607
* tree-ssa-threadedge.c (simplify_control_stmt_condition): Look
deeper into the SSA_NAME_VALUE chain.

tree-optimization/61607
* gcc.dg/tree-ssa/pr61607.c: New test.

From-SVN: r212149

convert.c (convert_to_integer): Don't instrument conversions if the function has no_sanitize_undefined attribute.

* convert.c (convert_to_integer): Don't instrument conversions if the
function has no_sanitize_undefined attribute.
* ubsan.c: Don't run the ubsan pass if the function has
no_sanitize_undefined attribute.
c/
* c-decl.c (grokdeclarator): Don't instrument VLAs if the function
has no_sanitize_undefined attribute.
cp/
* cp-gimplify.c (cp_genericize): Don't instrument returns if the
function has no_sanitize_undefined attribute.
* decl.c (compute_array_index_type): Don't instrument VLAs if the
function has no_sanitize_undefined attribute.
testsuite/
* c-c++-common/ubsan/attrib-2.c: New test.
* g++.dg/ubsan/return-3.C: New test.

From-SVN: r212148

invoke.texi (-fsanitize=bounds): Move to the table with -fsanitize=undefined suboptions.

* doc/invoke.texi (-fsanitize=bounds): Move to the table with
-fsanitize=undefined suboptions.

From-SVN: r212146

[AArch64] fix and enable non-const shuffle for bigendian using TBL instruction

* config/aarch64/aarch64-simd.md (vec_perm): Enable for bigendian.
* config/aarch64/aarch64.c (aarch64_expand_vec_perm): Remove assert
against bigendian and adjust indices.

From-SVN: r212142

* doc/install.texi (Specific, aarch64*-*-*): Fix markup.  Reword a bit.

From-SVN: r212140

re PR middle-end/57541 ([Cilkplus]: internal compiler error: in gimplify_expr, at gimplify.c:7809)

PR middle-end/57541
gcc/c/
* c-array-notation.c (fix_builtin_array_notation_fn):
Check for 0 arguments in builtin call. Check that bultin argument is
correct.
* c-parser.c (c_parser_array_notation): Check for incorrect initial
index.

gcc/cpp/
* cp-array-notation.c (expand_sec_reduce_builtin):
Check that bultin argument is correct.
* call.c (build_cxx_call): Check for 0 arguments in builtin call.

gcc/testsuite/
* c-c++-common/cilk-plus/AN/pr57541.c: New case added.
* c-c++-common/cilk-plus/AN/pr57541-2.c: New test.

From-SVN: r212138

[AArch64] Fix register clobber in, aarch64_ashr_sisd_or_int_<mode>3 split.

PR target/61633

The two split patterns associated with
aarch64_ashr_sisd_or_int_<mode>3 split the instruction into a NEG
followed by an SHL.  The split uses one of the input operands as a
scratch register to hold the output of the NEG resulting in register
corruption.

This patch adjusts the splits to use the output operand as the scratch
register.

2014-06-30  Marcus Shawcroft  <marcus.shawcroft@arm.com>

       PR target/61633
       * config/aarch64/aarch64.md (*aarch64_ashr_sisd_or_int_<mode>3):
       Add alternative; make early clobber.  Adjust both split patterns
       to use operand 0 as the working register.

From-SVN: r212137

ira-build.c (sort_conflict_id_map): Don't call qsort if num is 0, as ira_object_id_map might be NULL, or 1.

* ira-build.c (sort_conflict_id_map): Don't call
qsort if num is 0, as ira_object_id_map might be NULL, or 1.

From-SVN: r212136

loop-invariant.c (get_inv_cost): Handle register class.

ChangeLog:
2014-06-30  Zhenqiang Chen  <zhenqiang.chen@linaro.org>

* loop-invariant.c (get_inv_cost): Handle register class.
(gain_for_invariant): Check the register pressure of the inv
and its overlapped register class, other than all.

testsuite/ChangeLog:
2014-06-30  Zhenqiang Chen  <zhenqiang.chen@linaro.org>

* ira-loop-pressure.c: New test.

From-SVN: r212135

Daily bump.

From-SVN: r212132