[PR 69920] Prevent SRA from leaving a removed SSA_NAME in IL

2016-02-26  Martin Jambor  <mjambor@suse.cz>

PR middle-end/69920
* tree-sra.c (sra_modify_assign): Do not remove loads of
uninitialized aggregates to SSA_NAMEs.

testsuite/
* gcc.dg/torture/pr69932.c: New test.
* gcc.dg/torture/pr69936.c: Likewise.

From-SVN: r233753

S/390: PR69709 Fix risbg splitter

This fixes a wrong code generation problem with the splitters introduced
with that patch: https://gcc.gnu.org/ml/gcc-patches/2015-07/msg01840.html

The target operand is used as temporary.  This fails if it matches the
source of the left shift which is read after writing the temporary.

Thanks to Dominik for debugging it and thanks to Richard for the fix!

Bootstrapped and regtested on s390x with-arch=z13.

Bye,

-Andreas-

gcc/ChangeLog:

2016-02-26  Richard Henderson  <rth@redhat.com>

PR target/69709
* config/s390/s390.md (risbg and risbgn splitters): Allocate new
pseudo in case the target rtx matches the source of the left
shift.

gcc/testsuite/ChangeLog:

2016-02-26  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>

PR target/69709
* gcc.target/s390/pr69709.c: New test.

From-SVN: r233752

[hsa/69568] Fix ld instruction type for packed data

2016-02-26  Martin Jambor  <mjambor@suse.cz>

PR hsa/69568
* hsa.h (hsa_type_packed_p): Declare.
* hsa.c (hsa_type_packed_p): New function.
* hsa-gen.c (mem_type_for_type): Use unsigned type for packed
loads.
(gen_hsa_insns_for_store): Use hsa_type_packed_p.
* hsa-brig.c (emit_basic_insn): Likewise.

From-SVN: r233751

[hsa/69674] Make testsuite libgomp.c/for-3.c compile with -m32

2016-02-26  Martin Jambor  <mjambor@suse.cz>

pr hsa/69674
* hsa-gen.c (gen_hsa_phi_from_gimple_phi): Use proper hsa type for
pointers.
(gen_hsa_addr): Allow integer constants in TMR_INDEX2.

From-SVN: r233750

[hsa] Satisfy conditional move operand type constrains

2016-02-26  Martin Jambor  <mjambor@suse.cz>

* hsa.h (is_a_helper): New overload for hsa_op_immed for
hsa_op_with_type operands.
(hsa_unsigned_type_for_type): Declare.
* hsa.c (hsa_unsigned_type_for_type): New function.
* hsa-gen.c (gen_hsa_binary_operation): Use hsa_unsigned_type_for_type.
(gen_hsa_insns_for_operation_assignment): Satisfy constrains of
the finalizer.  Do not emit extra move.

From-SVN: r233749

[hsa] Fail in presence of atomic operations in private segment

2016-02-26  Martin Jambor  <mjambor@suse.cz>

* hsa-gen.c (gen_hsa_ternary_atomic_for_builtin): Fail in presence of
atomic operations in private segment.

From-SVN: r233748

re PR go/69966 (libgo: Port syscall.SetsockoptUcred from golang)

PR go/69966
    syscall: Add new Getsockopt functions.

    Add GetsockoptICMPv6Filter, GetsockoptIPv6MTUInfo, GetsockoptUcred as
    appropriate.  These functions exist in the master library.

    For GCC PR 69966.

    Reviewed-on: https://go-review.googlesource.com/19960

From-SVN: r233747

[omp, hsa] Do not gridify simd constructs

2016-02-26  Martin Jambor  <mjambor@suse.cz>

* omp-low.c (grid_find_ungridifiable_statement): Store problematic
statements to wi->info.  Also disallow omp simd constructs.
(grid_target_follows_gridifiable_pattern): Use wi.info to dump reason
for not gridifying.  Dump special string for omp_for.

From-SVN: r233746

[AArch64] Set TREE_TARGET_GLOBALS in aarch64_set_current_function when new tree is the default node to recalculate optab availability

PR target/69245
* config/aarch64/aarch64.c (aarch64_set_current_function):
Save/restore target globals when switching to
target_option_default_node.

* gcc.target/aarch64/pr69245_1.c: New test.

From-SVN: r233745

[AArch64] PR target/69613: Return zero TARGET_SHIFT_TRUNCATION_MASK when SHIFT_COUNT_TRUNCATED is false

PR target/69613
* config/aarch64/aarch64.c (aarch64_shift_truncation_mask):
Return 0 if !SHIFT_COUNT_TRUNCATED.

* gcc.dg/torture/pr69613.c: New test.

From-SVN: r233744

re PR rtl-optimization/69891 (wrong code with -mstringop-strategy=libcall @ i686)

PR rtl-optimization/69891
* dse.c (scan_insn): If we can't figure out memset arguments
or they are non-constant, call clear_rhs_from_active_local_stores.

* gcc.target/i386/pr69891.c: New test.

Co-Authored-By: Eric Botcazou <ebotcazou@adacore.com>
From-SVN: r233743

Fix ChangeLog formatting

From-SVN: r233742

Update ChangeLog to mention pr61397

From-SVN: r233740

p8vector-ldst.c: Adjust to test desired functionality for both 32-bit and 64-bit.

2016-02-26  Michael Meissner <meissner@linux.vnet.ibm.com>
    Bill Schmidt  <wschmidt@linux.vnet.ibm.com>

* gcc.target/powerpc/p8vector-ldst.c: Adjust to test desired
functionality for both 32-bit and 64-bit.

Co-Authored-By: Bill Schmidt <wschmidt@linux.vnet.ibm.com>
From-SVN: r233739

Mention clog10{,f,l} in documentation (Builtins section)

* doc/extend.texi: Mention clog10, clog10f an clog10l
in Builtins section.

From-SVN: r233738

Replace ENABLE_CHECKING with CHECKING_P in dwarf2out

* dwarf2out.c (new_loc_descr): Replace ENABLE_CHECKING with
CHECKING_P.
(resolve_args_picking_1): Likewise.
* dwarf2out.h (struct GTY): Likewise.

From-SVN: r233737

Replace ENABLE_CHECKING macro with flag_checking in HSA

* hsa-gen.c (generate_hsa): Replace ENABLE_CHECKING macro
with flag_checking.
* hsa-regalloc.c (linear_scan_regalloc): Likewise.

From-SVN: r233736

Mention --enable-valgrind-annotations in install.texi

    Martin Liska  <mliska@suse.cz>

* doc/install.texi: Mention --enable-valgrind-annotations.

Co-Authored-By: Martin Liska <mliska@suse.cz>
From-SVN: r233735

re PR target/69551 (Wrong code with single element vector insert)

2016-02-26  Richard Biener  <rguenther@suse.de>

PR tree-optimization/69551
* tree-ssa-structalias.c (get_constraint_for_ssa_var): When
looking through aliases adjust DECL_PT_UID to refer to the
ultimate alias target.

* gcc.dg/torture/pr69951.c: New testcase.

From-SVN: r233734

re PR c++/69889 (ICE: in assign_temp, at function.c:961)

PR c++/69889

* cp-tree.h (AGGR_INIT_FROM_THUNK_P): New.
* tree.c (build_aggr_init_expr): Set it.
* semantics.c (simplify_aggr_init_expr): Check it.
* cp-gimplify.c (cp_genericize_r): Don't walk into
a call/aggr_init from a thunk.

From-SVN: r233733

Daily bump.

From-SVN: r233732

Do not gather mem stats in run_exit_handles (PR

PR middle-end/69919
* alloc-pool.c (after_memory_report): New variable.
* alloc-pool.h (base_pool_allocator ::release): Do not use
the infrastructure if after_memory_report.
* toplev.c (toplev::main): Mark after memory report.

From-SVN: r233722

re PR c++/69842 (Parameter deduction in polymorphic lambdas)

PR c++/69842
* method.c (forward_parm): Handle parameter packs.
* lambda.c (maybe_add_lambda_conv_op): Use it for them.

From-SVN: r233719

re PR c++/67364 ("accessing uninitialized member" error in constexpr context)

PR c++/67364
* constexpr.c (cxx_eval_component_reference): Don't complain about
unevaluated empty classes.

From-SVN: r233716

re PR c++/68049 (template instantiation involving may_alias defines symbol twice)

PR c++/68049
* tree.c (strip_typedefs): Use DECL_ORIGINAL_TYPE.

From-SVN: r233715

re PR tree-optimization/48795 (-Warray-bounds false positive)

2016-02-25  Richard Biener  <rguenther@suse.de>

PR tree-optimization/48795
* tree-vrp.c (check_array_ref): Use array_at_struct_end_p.

* gcc.dg/Warray-bounds-18.c: New testcase.

From-SVN: r233714

Require init_priority support for g++.dg/ext/attr-constructor1.C

* g++.dg/ext/attr-constructor1.C: Require init_priority support.

From-SVN: r233713

re PR driver/68463 (Offloading fails when some objects are compiled with LTO and some without)

gcc/
PR driver/68463
* config/gnu-user.h (CRTOFFLOADBEGIN): Define.  Add crtoffloadbegin.o if
offloading is enabled and -fopenacc or -fopenmp is specified.
(CRTOFFLOADEND): Likewise.
(GNU_USER_TARGET_STARTFILE_SPEC): Add CRTOFFLOADBEGIN.
(GNU_USER_TARGET_ENDFILE_SPEC): Add CRTOFFLOADEND.
* lto-wrapper.c (offloadbegin, offloadend): Remove static vars.
(offload_objects_file_name): New static var.
(tool_cleanup): Remove offload_objects_file_name file.
(find_offloadbeginend): Replace with ...
(find_crtoffloadtable): ... this.
(run_gcc): Remove offload_argc and offload_argv.
Get offload_objects_file_name from -foffload-objects=... option.
Read names of object files with offload from this file, pass them to
compile_images_for_offload_targets.  Don't call find_offloadbeginend and
don't pass offloadbegin and offloadend to the linker.  Don't pass
offload non-LTO files to the linker, because now they're not claimed.
libgcc/
PR driver/68463
* Makefile.in (crtoffloadtable$(objext)): New rule.
* configure.ac (extra_parts): Add crtoffloadtable$(objext) if
enable_offload_targets is not empty.
* configure: Regenerate.
* offloadstuff.c: Move __OFFLOAD_TABLE__ from crtoffloadend to
crtoffloadtable.
libgomp/
PR driver/68463
* testsuite/libgomp.oacc-c-c++-common/parallel-dims-2.c: Remove.
lto-plugin/
PR driver/68463
* lto-plugin.c (struct plugin_offload_file): New.
(offload_files): Change type.
(offload_files_last, offload_files_last_obj): New.
(offload_files_last_lto): New.
(free_2): Adjust accordingly.
(all_symbols_read_handler): Don't add offload files to lto_arg_ptr.
Don't call free_1 for offload_files.  Write names of object files with
offloading to the temporary file.  Add new option to lto_arg_ptr.
(claim_file_handler): Don't claim file if it contains offload sections
without LTO sections.  If it contains offload sections, add to the list.

From-SVN: r233712

re PR lto/69630 (LTO ICE in types_same_for_odr at ipa-devirt.c:402)

PR ipa/69630
* ipa-devirt.c (possible_polymorphic_call_targets): Do not ICE
on builtin_unreachable.

From-SVN: r233711

re PR rtl-optimization/69896 (wrong code with -frename-registers @ x64_64)

PR rtl-optimization/69896
* regcprop.c: Include cfgrtl.h.
(copyprop_hardreg_forward_1): If noop_p insn uses narrower
than remembered mode, either delete it (if noop_move_p), or
treat like copy_p but not noop_p instruction.

* gcc.dg/pr69896.c: New test.

From-SVN: r233692

re PR c++/69736 ("error: too few arguments to function" in c++14 but not c++11)

Fix PR c++/69736

gcc/cp/ChangeLog:

PR c++/69736
* cp-tree.h (REF_PARENTHESIZED_P): Adjust documentation.
(maybe_undo_parenthesized_ref): Declare.
* semantics.c (maybe_undo_parenthesized_ref): Split out from
check_return_expr.
(finish_call_expr): Use it.
* typeck.c (check_return_expr): Use it.
* pt.c (tsubst_copy_and_build) [INDIRECT_REF]: Retain the
REF_PARENTHESIZED_P flag.

gcc/testsuite/ChangeLog:

PR c++/69736
* g++.dg/cpp1y/paren2.C: New test.

From-SVN: r233691

Daily bump.

From-SVN: r233690

PR c/51147 - attribute((mode(byte))) on an enum generates wrong code

gcc/testsuite/ChangeLog:
2016-02-24  Martin Sebor  <msebor@redhat.com>

        PR c/51147
        * gcc.dg/enum-mode-2.c: New test.

From-SVN: r233686

re PR debug/69705 (segfault in libgomp.fortran/task1.f90 with -flto -fno-use-linker-plugin -fno-toplevel-reorder -O1 -g)

PR debug/69705
* dwarf2out.c (gen_variable_die): Work around buggy LTO
- allow NULL decl for Fortran DW_TAG_common_block variables.

From-SVN: r233685

re PR c++/69922 (Bogus -Wnonnull-compare for: ... ? static_cast<T*>(this) : nullptr)

PR c++/69922
* class.c (build_base_path): Set TREE_NO_WARNING on the null_test.
Avoid folding it.
* init.c (build_vec_delete_1, build_delete): Don't fold the non-NULL
tests.
* cp-gimplify.c (cp_fold): For TREE_NO_WARNING comparisons with NULL,
unless they are folded into INTEGER_CST, error_mark_node or some
comparison with NULL, avoid folding them and use either the original
comparison or non-folded comparison of folded arguments.
* cp-ubsan.c (cp_ubsan_instrument_vptr): Set TREE_NO_WARNING on the
comparison, don't fold the comparison right away.

* g++.dg/warn/Wnonnull-compare-6.C: New test.
* g++.dg/warn/Wnonnull-compare-7.C: New test.
* g++.dg/ubsan/pr69922.C: New test.

From-SVN: r233684

re PR c/69819 (ICE on invalid code on x86_64-linux-gnu in tree check: expected function_type or method_type, have array_type in function_args_iter_init, at tree.h:4536)

PR c/69819
* c-decl.c (finish_decl): Don't update the copy of the type of a
different decl type.

* gcc.dg/pr69819.c: New test.

From-SVN: r233683

PR c++/69323 - errors

* friend.c (make_friend_class): Likewise.
* decl.c (lookup_and_check_tag): Diagnose invalid dependent friend.

From-SVN: r233682

PR c++/69323 - valid

* pt.c (instantiate_class_template_1): Set
processing_template_decl before substituting friend_type.

From-SVN: r233681

* doc/invoke.texi: Adjust -flifetime-dse documentation.

From-SVN: r233680

PR c++/69912 - [6 regression] ICE in build_ctor_subob_ref initializing

PR c++/69912 - [6 regression] ICE in build_ctor_subob_ref initializing
a flexible array member

gcc/testsuite/ChangeLog:
2016-02-24  Martin Sebor  <msebor@redhat.com>

PR c++/69912
* g++.dg/ext/flexary15.C: New test.

gcc/cp/ChangeLog:
2016-02-24  Martin Sebor  <msebor@redhat.com>

PR c++/69912
* tree.c (build_ctor_subob_ref): Compare types' main variants
        instead of the types as they are.

From-SVN: r233678

Avoid making unportable assumptions about the relationship between SIZE_MAX
and UINT_MAX.

gcc/testsuite/ChangeLog:
        * gcc/testsuite/gcc.dg/builtins-68.c: Avoid making unportable
        assumptions about the relationship between SIZE_MAX and UINT_MAX.
        * gcc/testsuite/g++.dg/ext/builtin_alloca.C: Same.

From-SVN: r233677

re PR debug/61033 (Infinite loop in variable tracking)

PR debug/61033
* g++.dg/pr61033.C: Add a regression testcase for PR debug/61033.

Co-Authored-By: Charles Baylis <charles.baylis@linaro.org>
From-SVN: r233675

Add -flifetime-dse=1.

gcc/
* common.opt (flifetime-dse): Add -flifetime-dse=1.
gcc/cp/
* decl.c (start_preparsed_function): Condition ctor clobber on
flag_lifetime_dse > 1.

From-SVN: r233672

* cp-gimplify.c (cp_fold): Don't fold constexpr calls if -fno-inline.

From-SVN: r233671

runtime: lock M during cgo call

    cgo should lock the M.

    See also https://golang.org/cl/18882 .

    Reviewed-on: https://go-review.googlesource.com/18883

From-SVN: r233670

re PR tree-optimization/69760 (Wrong 64-bit memory address caused by an unneeded overflowing 32-bit integer multiplication on x86_64 under -O2 and -O3 code optimization)

2016-02-24  Richard Biener  <rguenther@suse.de>
Jakub Jelinek  <jakub@redhat.com>

PR middle-end/69760
* tree-scalar-evolution.c (interpret_rhs_expr): Re-write
conditionally executed ops to well-defined overflow behavior.

* gcc.dg/torture/pr69760.c: New testcase.

Co-Authored-By: Jakub Jelinek <jakub@redhat.com>
From-SVN: r233669

Adapt gcc.dg/debug/dwarf2/prod-options.c for Solaris assembler

* gcc.dg/debug/dwarf2/prod-options.c: Use different DW_AT_producer
pattern on Solaris with as.

From-SVN: r233668

[ARM] Tests for arm_restrict_it patterns in thumb2.md

* gcc.target/arm/cond_sub_restrict_it.c: New test.
* gcc.target/arm/condarith_restrict_it.c: Likewise.
* gcc.target/arm/movcond_restrict_it.c: Likewise.
* gcc.target/arm/negscc_restrict_it.c: Likewise.

From-SVN: r233667

libstdc++/69939 Qualify get and forward

PR libstdc++/69939
* include/experimental/tuple (__apply_impl): Qualify get and forward.

From-SVN: r233666

re PR middle-end/69915 (ICE: SIGSEGV with -O -ftracer with broken backtrace)

PR middle-end/69915
* tree.c (build_vector_from_ctor): Fix handling of VECTOR_CST
elements.

* gcc.dg/pr69915.c: New test.

From-SVN: r233664

[gcse] PR rtl-optimization/69886: Check target mode in can_assign_to_reg_without_clobbers_p

PR rtl-optimization/69886
* gcse.c (can_assign_to_reg_without_clobbers_p): Accept mode
argument.  Use it when checking validity of set instructions.
(want_to_gcse_p): Pass mode to can_assign_to_reg_without_clobbers_p.
(compute_ld_motion_mems): Update can_assign_to_reg_without_clobbers_p
callsite.
* rtl.h (can_assign_to_reg_without_clobbers_p): Update prototype.
* store-motion.c (find_moveable_store): Update
can_assign_to_reg_without_clobbers_p callsite.

* gcc.dg/torture/pr69886.c: New test.

From-SVN: r233662

re PR tree-optimization/68963 (O3 vs. O2 discards part of loop and terminates early)

2016-02-24  Richard Biener  <rguenther@suse.de>

PR middle-end/68963
* tree-ssa-loop-niter.c (derive_constant_upper_bound_ops): Fix
bogus check.
(record_nonwrapping_iv): Do not fall back to the low/high bound
for non-constant IV bases if the stmt is not always executed.

* gcc.dg/torture/pr68963.c: New testcase.

From-SVN: r233660

[ARM] Add initial support for the Cortex-A32

* config/arm/arm-cores.def (cortex-a32): New entry.
* config/arm/arm-tables.opt: Regenerate.
* config/arm/arm-tune.md: Regenerate.
* config/arm/bpabi.h (BE8_LINK_SPEC): Add mcpu=cortex-a32.
* config/arm/t-aprofile: Handle mcpu=cortex-a32.
* doc/invoke.texi (ARM Options): Document cortex-a32 as value
for -mcpu and -mtune.

From-SVN: r233659

[ARM] PR target/69875 Fix atomic_loaddi expansion

PR target/69875
* config/arm/arm.h (TARGET_HAVE_LPAE): Define.
* config/arm/unspecs.md (VUNSPEC_LDRD_ATOMIC): New value.
* config/arm/sync.md (arm_atomic_loaddi2_ldrd): New pattern.
(atomic_loaddi_1): Delete.
(atomic_loaddi): Rewrite expander using the above changes.

* gcc.target/arm/atomic_loaddi_acquire.x: New file.
* gcc.target/arm/atomic_loaddi_relaxed.x: Likewise.
* gcc.target/arm/atomic_loaddi_seq_cst.x: Likewise.
* gcc.target/arm/atomic_loaddi_1.c: New test.
* gcc.target/arm/atomic_loaddi_2.c: Likewise.
* gcc.target/arm/atomic_loaddi_3.c: Likewise.
* gcc.target/arm/atomic_loaddi_4.c: Likewise.
* gcc.target/arm/atomic_loaddi_5.c: Likewise.
* gcc.target/arm/atomic_loaddi_6.c: Likewise.
* gcc.target/arm/atomic_loaddi_7.c: Likewise.
* gcc.target/arm/atomic_loaddi_8.c: Likewise.
* gcc.target/arm/atomic_loaddi_9.c: Likewise.

From-SVN: r233658

re PR c/69918 (gcc.dg/torture/builtin-integral-1.c FAILs)

PR c/69918
* params.def (PARAM_MAX_SSA_NAME_QUERY_DEPTH): Bump default from
2 to 3.

From-SVN: r233657

re PR middle-end/69909 (wrong code with -Os and vectors @ x86_64)

PR middle-end/69909
* expr.c (expand_expr_real_1) <normal_inner_ref>: Avoid
set_mem_attributes if tem is SSA_NAME which got expanded
as a MEM.

* gcc.dg/torture/pr69909.c: New test.

Co-Authored-By: Richard Biener <rguenther@suse.de>
From-SVN: r233656

re PR tree-optimization/69907 (wrong code at -O3 on x86_64-linux-gnu)

2016-02-24  Richard Biener  <rguenther@suse.de>

PR tree-optimization/69907
* tree-vect-stmts.c (vectorizable_load): Check for gaps at the
end of permutations for BB vectorization.

* gcc.dg/vect/bb-slp-pr69907.c: New testcase.
* gcc.dg/vect/bb-slp-34.c: XFAIL.
* gcc.dg/vect/bb-slp-pr68892.c: Likewise.

From-SVN: r233655

arm-c.c (arm_option_override): Initialize target_option_current_node.

2016-02-24  Christian Bruel  <christian.bruel@st.com>

* config/arm/arm-c.c (arm_option_override): Initialize
target_option_current_node.
* config/arm/arm.c (arm_pragma_target_parse): Replace
build_target_option_node call by target_option_current_node.
Set target_option_current_node.
Fix comments.

From-SVN: r233654

Daily bump.

From-SVN: r233652

re PR fortran/61156 (Internal compiler error for Fortran files when specifying a file instead of an include directory with -I)

2016-02-23  Jerry DeLisle  <jvdelisle@gcc.gnu.org>

PR fortran/61156
* scanner.c (add_path_to_list): If include path is not a directory,
issue a fatal error.

PR fortran/61156
* gfortran.dg/include_6.f90: Update test.

From-SVN: r233649

re PR target/69810 (PowerPC64: unrecognizable insn)

PR target/69810
* config/rs6000/rs6000.md (zero_extendqi<mode>2_dot): Convert from
define_insn_and_split to define_insn.
(zero_extendqi<mode>2_dot2): Same.
(extendqi<mode>2_dot): Same.
(extendqi<mode>2_dot2): Same.

From-SVN: r233648

Tweak the pipeline model for Exynos M1

gcc/
* config/aarch64/aarch64.c (exynosm1_tunings): Enable fusion of AES{D,E}
and AESMC pairs.
* config/arm/exynos-m1.md: Change cost of STP, fix bypass for stores
and add bypass for AES{D,E} and AESMC pairs.

From-SVN: r233647

aarch64.c (exynosm1_tunings): Enable the Newton series for reciprocal square root in Exynos M1.

        * gcc/config/aarch64/aarch64.c (exynosm1_tunings): Enable the Newton
        series for reciprocal square root in Exynos M1.

From-SVN: r233646

Document __STDCPP_WANT_MATH_SPEC_FUNCS__ macro

* doc/xml/manual/using.xml: Document __STDCPP_WANT_MATH_SPEC_FUNCS__.
* doc/html/*: Regenerate.

From-SVN: r233645

libstdc++/69893 make <tr1/cmath> work with C++11

PR libstdc++/69893
* include/tr1/cmath (acosh, asinh, atanh, cbrt, copysign, erf, erfc,
exp2, expm1, fdim, fma, fmax, fmin, hypot, ilogb, lgamma, llrint,
llround, log1p, log2, logb, lrint, lround, nan, nearbyint, nextafter,
nexttoward, remainder, remquo, rint, round, scalbln, scalbn, tgamma,
trunc) [__cplusplus >= 201103L]: Import from namespace std.
(fabs) [__cplusplus < 201103L]: Import from namespace std.
* include/tr1/complex (acosh, asinh, atanh) [__cplusplus >= 201103L]:
Likewise.
* testsuite/tr1/headers/c++200x/complex.cc: Add std::fabs to global
namespace before including TR1 headers.
* testsuite/tr1/headers/c++200x/math.cc: New test.

From-SVN: r233644

re PR objc/69844 (Possibly bogus error: unknown type name in ObjC code)

PR objc/69844
* c-parser.c (c_parser_for_statement): Properly handle ObjC classes
in id_kind reclassification.

* objc.dg/pr69844.m: New test.

From-SVN: r233643

re PR fortran/69456 (Namelist value with trailing sign is ignored without error)

2016-02-23  Jerry DeLisle  <jvdelisle@gcc.gnu.org>

PR libgfortran/69456
* io/list_read.c (read_real): If digit is missing from exponent issue
an error. (parse_real): Likewise and adjusted error message to clarify
it is part of a complex number.
(nml_read_obj): Bump item count and add comment that this is used to
identify which item in a namelist read has a problem.

PR libgfortran/69456
* gfortran.dg/namelist_89.f90: New test.
* gfortran.dg/pr59700.f90: Update test..

From-SVN: r233641

PR middle-end/69780 - [4.9/5/6 Regression] ICE on __builtin_alloca_with_align

PR middle-end/69780 - [4.9/5/6 Regression] ICE on __builtin_alloca_with_align
with small alignment
PR c/69759 - __builtin_alloca and __builtin_alloca_with_align undocumented

gcc/c-family/ChangeLog:
* c-common.c (check_builtin_function_arguments): Validate and reject
invalid arguments to __builtin_alloca_with_align.

gcc/ChangeLog:
* doc/extend.texi (Other Builtins): Document __builtin_alloca and
__builtin_alloca_with_align.

gcc/testsuite/ChangeLog:
* g++.dg/ext/builtin_alloca.C: New test.
* gcc.dg/builtins-68.c: New test.

From-SVN: r233640

Remove __seg_tls before first release.

  * config/i386/i386-c.c (ix86_target_macros): Remove __SEG_TLS.
  (ix86_register_pragmas): Remove __seg_tls.
  * config/i386/i386-protos.h (ADDR_SPACE_SEG_TLS): Remove.
  * config/i386/i386.c (ix86_print_operand_address_as): Don't handle it.
  (ix86_addr_space_subset_p, TARGET_ADDR_SPACE_SUBSET_P): Remove.
  (ix86_addr_space_convert, TARGET_ADDR_SPACE_CONVERT): Remove.
  (ix86_addr_space_debug, TARGET_ADDR_SPACE_DEBUG): Remove.
  * doc/extend.texi (__seg_tls): Remove item.
testsuite/
  * gcc.target/i386/addr-space-3.c: Remove test.

From-SVN: r233639

PR preprocessor/69126: avoid comparing ad-hoc and non-ad-hoc locations

gcc/testsuite/ChangeLog:
PR preprocessor/69126
PR preprocessor/69543
* c-c++-common/pr69126-2-long.c: New test.
* c-c++-common/pr69126-2-short.c: New test.
* c-c++-common/pr69543-1.c: Remove xfail.

libcpp/ChangeLog:
PR preprocessor/69126
PR preprocessor/69543
* line-map.c (linemap_compare_locations): At the function top,
replace inlined bodies of get_location_from_adhoc_loc with calls
to get_location_from_adhoc_loc.  Add a pair of calls to
get_location_from_adhoc_loc at the bottom of the function, to
avoid meaningless comparisons of ad-hoc and non-ad-hoc locations.

From-SVN: r233638

Add test coverage for _Pragma (PR preprocessor 69126, 69543, 69558)

We had some regressions in the ability for _Pragma to disable a warning
(PR preprocessor/69126, PR preprocessor/69543, PR preprocessor/69558).

This patch attempts to add more test coverage for this, for the
various combinations of:
  - various warnings:
    -Wunused-variable
    -Wuninitialized
    -Wdeprecated-declarations
  - various combinations of location of _Pragma relative to location of
    the warning:
     - _Pragma is in a macro, warning isn't a macro
     - neither is in a macro
     - _Pragma isnt't in a macro, warning is in a macro
     - in different macros
     - both in the same macro
  - C vs C++ frontend.

It adds some XFAILs:
  - pr69543-1.c for C++ (fixed in the followup patch)
  - pr69543-3.c for both C and C++
  - pr69543-4.c for both C and C++
  - pr69558.c for C++ (moving it from gcc.dg to c-c++-common,
    marking it as xfail for C++ for now)

gcc/testsuite/ChangeLog:
PR preprocessor/69126
PR preprocessor/69543
PR preprocessor/69558
* c-c++-common/pr69126.c (MACRO_1, test_1): New.
(f): Rename to...
(test_2): ...this, and add leading comment.
(MACRO_3, test_3): New.
(MACRO_4A, MACRO_4B, test_4): New.
(MACRO): Rename to...
(MACRO_5): ...this.
(g): Rename to...
(test_5): ...this, updating for renaming of MACRO, and
add leading comment.
* c-c++-common/pr69543-1.c: New.
* c-c++-common/pr69543-2.c: New.
* c-c++-common/pr69543-3.c: New.
* c-c++-common/pr69543-4.c: New.
* c-c++-common/pr69558-1.c: New.
* c-c++-common/pr69558-2.c: New.
* c-c++-common/pr69558-3.c: New.
* c-c++-common/pr69558-4.c: New.
* gcc.dg/pr69558.c: Move to...
* c-c++-common/pr69558.c: ...here.  Add dg-bogus directives, with
xfail for c++.

From-SVN: r233637

re PR libstdc++/69881 (with gcc-6 of today building gcc-4.9 fails)

2016-02-23  Bernd Edlinger  <bernd.edlinger@hotmail.de>

        PR libstdc++/69881
        * include/c_global/cstdarg: Undefine __need___va_list.
        * include/c_global/cstddef: Undefine all kinds of __need_*.

From-SVN: r233636

alloc-pool.h (struct allocation_object): Make id member conditional on CHECKING_P again.

2016-02-23  Richard Biener  <rguenther@suse.de>

* alloc-pool.h (struct allocation_object): Make id member
conditional on CHECKING_P again.
(get_instance): Adjust.
(base_pool_allocator): Likewise.

From-SVN: r233635

Use plain -fopenacc to enable OpenACC kernels processing

gcc/
* tree-parloops.c (create_parallel_loop, gen_parallel_loop)
(parallelize_loops): In OpenACC kernels mode, set n_threads to
zero.
(pass_parallelize_loops::gate): In OpenACC kernels mode, gate on
flag_openacc.
* tree-ssa-loop.c (gate_oacc_kernels): Likewise.
gcc/testsuite/
* c-c++-common/goacc/kernels-counter-vars-function-scope.c: Adjust
to -ftree-parallelize-loops/-fopenacc changes.
* c-c++-common/goacc/kernels-double-reduction-n.c: Likewise.
* c-c++-common/goacc/kernels-double-reduction.c: Likewise.
* c-c++-common/goacc/kernels-loop-2.c: Likewise.
* c-c++-common/goacc/kernels-loop-3.c: Likewise.
* c-c++-common/goacc/kernels-loop-g.c: Likewise.
* c-c++-common/goacc/kernels-loop-mod-not-zero.c: Likewise.
* c-c++-common/goacc/kernels-loop-n.c: Likewise.
* c-c++-common/goacc/kernels-loop-nest.c: Likewise.
* c-c++-common/goacc/kernels-loop.c: Likewise.
* c-c++-common/goacc/kernels-one-counter-var.c: Likewise.
* c-c++-common/goacc/kernels-reduction.c: Likewise.
* gfortran.dg/goacc/kernels-loop-inner.f95: Likewise.
* gfortran.dg/goacc/kernels-loops-adjacent.f95: Likewise.
libgomp/
* oacc-parallel.c (GOACC_parallel_keyed): Initialize dims.
* plugin/plugin-nvptx.c (nvptx_exec): Provide default values for
dims.
* testsuite/libgomp.oacc-c-c++-common/kernels-loop-2.c: Adjust to
-ftree-parallelize-loops/-fopenacc changes.
* testsuite/libgomp.oacc-c-c++-common/kernels-loop-3.c: Likewise.
* testsuite/libgomp.oacc-c-c++-common/kernels-loop-and-seq-2.c:
Likewise.
* testsuite/libgomp.oacc-c-c++-common/kernels-loop-and-seq-3.c:
Likewise.
* testsuite/libgomp.oacc-c-c++-common/kernels-loop-and-seq-4.c:
Likewise.
* testsuite/libgomp.oacc-c-c++-common/kernels-loop-and-seq-5.c:
Likewise.
* testsuite/libgomp.oacc-c-c++-common/kernels-loop-and-seq-6.c:
Likewise.
* testsuite/libgomp.oacc-c-c++-common/kernels-loop-and-seq.c:
Likewise.
* testsuite/libgomp.oacc-c-c++-common/kernels-loop-collapse.c:
Likewise.
* testsuite/libgomp.oacc-c-c++-common/kernels-loop-g.c: Likewise.
* testsuite/libgomp.oacc-c-c++-common/kernels-loop-mod-not-zero.c:
Likewise.
* testsuite/libgomp.oacc-c-c++-common/kernels-loop-n.c: Likewise.
* testsuite/libgomp.oacc-c-c++-common/kernels-loop-nest.c:
Likewise.
* testsuite/libgomp.oacc-c-c++-common/kernels-loop.c: Likewise.
* testsuite/libgomp.oacc-c-c++-common/kernels-reduction.c:
Likewise.

From-SVN: r233634

mem-stats.h (struct mem_usage): Use PRIu64 for printing size_t.

2016-02-23  Richard Biener  <rguenther@suse.de>

* mem-stats.h (struct mem_usage): Use PRIu64 for printing size_t.
* bitmap.h (struct bitmap_usage): Likewise.
(bitmap_move): Declare.
* bitmap.c (register_overhead): Take size_t argument.
(bitmap_move): New function.
* df-problems.c (df_rd_transfer_function): Use bitmap_move
to properly account overhead.
* tree.c (free_node): Use tree_size.

From-SVN: r233633

Require alias support in gcc.target/i386/chkp-hidden-def.c

* gcc.target/i386/chkp-hidden-def.c: Require alias support.

From-SVN: r233632

re PR c++/69902 (Bogus -Wnonnull-compare for: dynamic_cast<T*>(&ref) == nullptr)

PR c++/69902
* fold-const.c (fold_truth_not_expr): Propagate TREE_NO_WARNING
when inverting comparison.

* g++.dg/warn/Wnonnull-compare-5.C: New test.

From-SVN: r233631

re PR c/69900 (Unhelpful diagnostic about Ignored options)

PR c/69900
* common.opt (Wunreachable-code): Add Warning flag.

* gcc.dg/pr69900.c: New test.

From-SVN: r233630

PR c/69911 Check main_input_filename and DECL_SOURCE_FILE are not NULL.

DECL_SOURCE_FILE can be NULL (for example when DECL_ARTIFICIAL).
So make sure to check before comparing against main_input_filename.

gcc/ChangeLog

PR c/69911
* cgraphunit.c (check_global_declaration): Check main_input_filename
and DECL_SOURCE_FILE are not NULL.

Co-Authored-By: Jakub Jelinek <jakub@redhat.com>
From-SVN: r233627

[pr 69666] No SRA default_def replacements for unscalarizable regions

2016-02-23  Martin Jambor  <mjambor@suse.cz>

PR tree-optimization/69666
* tree-sra.c (sra_modify_assign): Do not attempt to create
default_def replacements for unscalarizable regions.

testsuite/
* gcc.dg/tree-ssa/pr69666.c: New test.

From-SVN: r233626

re PR fortran/67451 ([F08] ICE with sourced allocation from coarray.)

gcc/testsuite/ChangeLog:

2016-02-23  Andre Vehreschild  <vehre@gcc.gnu.org>

PR fortran/67451
* gfortran.dg/coarray_allocate_5.f08: New test.

gcc/fortran/ChangeLog:

2016-02-23  Andre Vehreschild  <vehre@gcc.gnu.org>

PR fortran/67451
* trans-array.c (gfc_array_allocate): Take the attributes from the
expression to allocate and not from the source=-expression.

From-SVN: r233625

S/390: Move vcond-shift.c to vector subdir.

gcc/testsuite/ChangeLog:

* gcc.target/s390/vcond-shift.c: Move to ...
* gcc.target/s390/vector/vcond-shift.c: ... here.

From-SVN: r233624

S/390: Move movstr-2.c into vector subdir.

gcc/testsuite/ChangeLog:

* gcc.target/s390/md/movstr-2.c: Move and rename to ...
* gcc.target/s390/vector/stpcpy-1.c: ... this one.

From-SVN: r233623

S/390: Turn movstr-1.c into compile only test.

gcc/testsuite/ChangeLog:

* gcc.target/s390/md/movstr-1.c: Turn into compile test.

From-SVN: r233622

sso tests need int32 machines.

From-SVN: r233621

Daily bump.

From-SVN: r233620

* sv.po, vi.po: Update.

From-SVN: r233617

PR28901 Add two levels for -Wunused-const-variable.

There is some controversy about enabling -Wunused-const-variable for all
unused static const variables because some feel there are too many errors
exposed in header files. Create two levels for -Wunused-const-variable.
One level to only check for unused static const variables in the main
compilation file. Which is enabled by -Wunused-variable. And a second
level that also checks for unused static const variables in included
files. Which must be explicitly enabled.

gcc/ChangeLog

PR c/28901
* cgraphunit.c (check_global_declaration): Check level of
warn_unused_const_variable and main_input_filename.
* doc/invoke.texi (Warning Options): Add -Wunused-const-variable=.
(-Wunused-variable): For C implies -Wunused-const-variable=1.
(-Wunused-const-variable): Explain levels 1 and 2.

gcc/c-family/ChangeLog

PR c/28901
* c.opt (Wunused-const-variable): Turn into Alias for...
(Wunused-const-variable=): New option.

gcc/testsuite/ChangeLog

PR c/28901
* gcc.dg/unused-variable-3.c: New test.

From-SVN: r233616

re PR target/69888 (ICE: SIGSEGV in decide_alg (i386.c:26169) due to infinite (?) recursion with -minline-all-stringops -mmemset-strategy=no_stringop:-1:noalign)

PR target/69888
* config/i386/i386.c (decide_alg): Ensure we don't recurse with
identical arguments.  Formatting and spelling fixes.

* gcc.target/i386/pr69888.c: New test.

From-SVN: r233615

re PR target/69888 (ICE: SIGSEGV in decide_alg (i386.c:26169) due to infinite (?) recursion with -minline-all-stringops -mmemset-strategy=no_stringop:-1:noalign)

PR target/69888
* config/i386/i386.c (decide_alg): Ensure we don't recurse with
identical arguments.  Formatting and spelling fixes.

* gcc.target/i386/pr69888.c: New test.

From-SVN: r233614

re PR bootstrap/69885 (ICE in maybe_legitimize_operand, at optabs.c:6903 on m68k-linux-gnu)

PR target/69885
* doc/md.texi (ashl@var{m}3): Document that mode of operand 2 must
be specified.

From-SVN: r233613

Add terminating line to output as prune can remove last \r from output.

From-SVN: r233612

re PR target/69894 (dependency of gcc-plugin.h not installed on aarch64-linux-gnu)

PR target/69894
PR target/69895
* config/m68k/t-opts (OPTIONS_H_EXTRA): Add m68k-microarchs.def
and m68k-devices.def.
* config/c6x/t-c6x (OPTIONS_H_EXTRA): Add c6x-isas.def.
* config/aarch64/t-aarch64 (OPTIONS_H_EXTRA): Add aarch64-arches.def.

From-SVN: r233611

MAINTAINERS: add myself to libcpp and diagnostic messages

2016-02-22  David Malcolm  <dmalcolm@redhat.com>

* MAINTAINERS (libcpp): Add myself.
(diagnostic messages): Likewise.

From-SVN: r233608

nvptx.c (nvptx_gen_shuffle): Add support for QImode and HImode registers.

gcc/
* config/nvptx/nvptx.c (nvptx_gen_shuffle): Add support for QImode
and HImode registers.

libgomp/
* testsuite/libgomp.oacc-c-c++-common/vprop.c: New test.

From-SVN: r233607

re PR tree-optimization/69882 (Excessive reduction statements generated by SLP)

2016-02-22  Richard Biener  <rguenther@suse.de>

PR tree-optimization/69882
* tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Properly
preserve permutations present because of gaps.
(vect_supported_load_permutation_p): Always continue checking
permutations after vect_attempt_slp_rearrange_stmts.

* gfortran.dg/vect/pr69882.f90: New testcase.

From-SVN: r233605

tree-vect-loop.c (vect_estimate_min_profitable_iters): Dump min_profitable_estimate, rather than min_profitable_iters.

* tree-vect-loop.c (vect_estimate_min_profitable_iters): Dump
min_profitable_estimate, rather than min_profitable_iters.

From-SVN: r233604

re PR bootstrap/69885 (ICE in maybe_legitimize_operand, at optabs.c:6903 on m68k-linux-gnu)

PR target/69885
* config/m68k/m68k.md (ashldi3, ashrdi3, lshrdi3): Use
SImode for last match_operand.

* gcc.dg/pr69885.c: New test.

From-SVN: r233603

HSA: fix emission of clrsb{l,ll} builtins

* hsa-gen.c (gen_hsa_clrsb): In case of zero value,
return bitsize - 1 as the return value.

From-SVN: r233602

re PR rtl-optimization/69806 ([SH] Combine doesn't see constant)

gcc/
PR target/69806
PR target/54089
* config/sh/sh.c (sh_lshrsi_clobbers_t_reg_p, sh_dynamicalize_shift_p):
Handle negative shift counts.
* config/sh/sh.md (ashlsi3, lshrsi3_n, lshrsi3_n_clobbers_t): Don't use
force_reg on the shift constant.
(lshrsi3): Likewise.  Expand into lshrsi3_n* instead of lshrsi3_d.
(lshrsi3_d): Handle negative shift counts.

gcc/testsuite/
PR target/69806
PR target/54089
* gcc.target/sh/pr54089-10.c: New.

From-SVN: r233601