re PR testsuite/68913 (gcc.dg/lto/pr61886 FAILs)

PR target/68913
* gcc.dg/lto/pr61886_0.c: Rename the external function called
to fread so that it will be found in all target runtimes.

From-SVN: r232264

re PR target/69053 (ICE in build_vector_from_val)

2016-01-12  Richard Biener  <rguenther@suse.de>

PR tree-optimization/69053
* tree-vect-loop.c (get_initial_def_for_reduction): Properly
convert initial value for cond reductions.

* g++.dg/torture/pr69053.C: New testcase.

From-SVN: r232263

re PR testsuite/69007 (test cases gcc.dg/vect/vect-reduc-dot-s8a.c fails starting with r231815)

2016-01-12  Richard Biener  <rguenther@suse.de>

PR tree-optimization/69007
* tree-vect-patterns.c (vect_vect_recog_func_ptrs): Move
widen_sum after dot_prod and sad.

From-SVN: r232262

re PR tree-optimization/69168 (ICE in vect_analyze_loop_2, at tree-vect-loop.c:2182)

2016-01-12  Richard Biener  <rguenther@suse.de>

PR tree-optimization/69168
* tree-vect-loop.c (vect_analyze_loop_2): Reset both main and
pattern stmt SLP type.
* tree-vect-slp.c (vect_detect_hybrid_slp_stmts): Patterns may
end up unused so cope with that case.

* gcc.dg/torture/pr69168.c: New testcase.

From-SVN: r232261

re PR tree-optimization/69157 (ICE in vect_transform_stmt, at tree-vect-stmts.c:8176)

2016-01-12  Richard Biener  <rguenther@suse.de>

PR tree-optimization/69157
* tree-vect-stmts.c (vectorizable_mask_load_store): Check
stmts def type only during analyze phase.
(vectorizable_call): Likewise.
(vectorizable_simd_clone_call): Likewise.
(vectorizable_conversion): Likewise.
(vectorizable_assignment): Likewise.
(vectorizable_shift): Likewise.
(vectorizable_operation): Likewise.
(vectorizable_store): Likewise.
(vectorizable_load): Likewise.

* gcc.dg/torture/pr69157.c: New testcase.

2016-01-12  Richard Biener  <rguenther@suse.de>

PR tree-optimization/69174
* tree-vect-stmts.c (vect_mark_relevant): Remove excessive vertical
space.
(vectorizable_load): Properly compute the number of loads needed
for permuted strided SLP loads and do not spuriously assign
to SLP_TREE_VEC_STMTS.

* gcc.dg/torture/pr69174.c: New testcase.

From-SVN: r232260

re PR c++/66808 (tree check fail in symbol_table::decl_assembler_name_hash)

PR c++/66808
PR c++/69000
* pt.c (tsubst_decl): If not local_p, clear DECL_TEMPLATE_INFO.

* g++.dg/tls/pr66808.C: New test.
* g++.dg/tls/pr69000.C: New test.

From-SVN: r232259

[PATCH 5/6] [DJGPP] Update DJGPP configuration related files

       * config/i386/djgpp.h (PREFERRED_DEBUGGING_TYPE): Define to DWARF2
       (TARGET_ASM_OUTPUT_IDENT): Define to default_asm_output_ident_directive
       (MD_EXEC_PREFIX): Remove.
       (MD_STARTFILE_PREFIX) Removee.
       (FILE_NAME_ABSOLUTE_P): Remove.
       (CPP_SPEC): Do not read macros from sys/version.h.
       (LINK_COMMAND_SPEC): Remove.
       (LOCAL_INCLUDE_DIR): Remove.
       (TARGET_ASM_NAMED_SECTION): Define to i386_djgpp_asm_named_section
       (TARGET_OS_CPP_BUILTINS): Add DJGPP (non ISO only), __DJGPP, __DJGPP__, unix.
       (POST_LINK_SPEC): Define to invoke stubify after linker
       (LIBSTDCXX): Remove define
       (DBX_REGISTER_NUMBER): Define to svr4_dbx_register_map.
       (DEFAULT_PCC_STRUCT_RETURN): Define to 1.
       (SUBTARGET_OVERRIDE_OPTIONS): Remove warning about -mbnu2210.
       (SUBTARGET_OVERRIDE_OPTIONS): Ignore -fPIC and generate message.
       (SUBTARGET_OVERRIDE_OPTIONS): Default to DWARF2 debugging info.
       (IX86_MAYBE_NO_LIBGCC_TFMODE): Remove.
       (i386_djgpp_asm_named_section): Add propotype of new procedure

       * config/i386/xm-djgpp.h (NATIVE_SYSTEM_HEADER_DIR): Define.
       (MD_EXEC_PREFIX): Define (moved from config/i386/djgpp.h).
       (STANDARD_STARTFILE_PREFIX_1): Define (moved from MD_STARTFILE_PREFIX in config/i386/djgpp.h).
       (STANDARD_STARTFILE_PREFIX_2): Define identical to STANDARD_STARTFILE_PREFIX_1.
       (LOCAL_INCLUDE_DIR): Define (moved from config/i386/djgpp.h).
       (GCC_DRIVER_HOST_INITIALIZATION): Fix reporting fatal installation errors.
       (MAX_OFILE_ALIGNMENT): Define to 128.
       (HAVE_FTW_H): Undefine as DJGPP do not have nftw, but have ftw.h.

       * config/i386/djgpp.c: New file. Add implementation of i386_djgpp_asm_named_section.

       * config/i386/djgpp.opt: Remove obsolete option -mbnu210.

       * config/i386/t-djgpp: New file. Add djgpp.o to EXTRA_OBJS.
       Add rule for building djgpp.o.

From-SVN: r232258

rs6000.c (v2df_reduction_p): New function.

[gcc]

2016-01-11  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>

* config/rs6000/rs6000.c (v2df_reduction_p): New function.
(rtx_is_swappable_p): Reductions are swappable.
(insn_is_swappable_p): V2DF reductions are swappable.

[gcc/testsuite]

2016-01-11  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>

* gcc.target/powerpc/swaps-p8-23.c: New test.
* gcc.target/powerpc/swaps-p8-24.c: Likewise.

From-SVN: r232257

re PR tree-optimization/68356 (FAIL: gcc.dg/torture/pr68264.c   -O*  execution test on x86_64-apple-darwin1(0|4))

PR tree-optimization/68356
* gcc.dg/torture/pr68264.c: Disable exp2 ERANGE test on hppa unix.

From-SVN: r232256

pa.c (pa_emit_move_sequence): Handle floating point reloads for other unsupported memory operands.

* config/pa/pa.c (pa_emit_move_sequence): Handle floating point
reloads for other unsupported memory operands.

From-SVN: r232253

re PR target/69194 (internal compiler error: in extract_insn, at recog.c:2286)

gcc/ChangeLog:

2016-01-12  Kugan Vivekanandarajah  <kuganv@linaro.org>
    Jim Wilson  <jim.wilson@linaro.org>

PR target/69194
* config/arm/arm-builtins.c (arm_expand_neon_args): Call
copy_to_mode_reg instead of force_reg.

gcc/testsuite/ChangeLog:

2016-01-12  Kugan Vivekanandarajah  <kuganv@linaro.org>
    Jim Wilson  <jim.wilson@linaro.org>

PR target/69194
* gcc.target/arm/pr69194.c: New test.

Co-Authored-By: Jim Wilson <jim.wilson@linaro.org>
From-SVN: r232251

Set FLT_EVAL_METHOD to 2 only if 387 FPU is used

When 387 FPU isn't used, there is no excess precision.  We should set
FLT_EVAL_METHOD to 2 only if 387 FPU is used.

gcc/

PR target/69225
* config/i386/i386.h (TARGET_FLT_EVAL_METHOD): Set to 2 only if
TARGET_80387 is true.

gcc/testsuite

PR target/69225
* gcc.target/i386/pr69225-1.c: New test.
* gcc.target/i386/pr69225-2.c: Likewise.
* gcc.target/i386/pr69225-3.c: Likewise.
* gcc.target/i386/pr69225-4.c: Likewise.
* gcc.target/i386/pr69225-5.c: Likewise.
* gcc.target/i386/pr69225-6.c: Likewise.

From-SVN: r232249

Daily bump.

From-SVN: r232248

* zh_CN.po: Update.

From-SVN: r232244

re PR c++/69131 (default constructor of union incorrectly deleted)

PR c++/69131
* method.c (walk_field_subobs): Add dtor_from_ctor parm.
(process_subob_fn): Likewise.  Don't consider triviality if true.
(synthesize_method_walk): Pass it.

From-SVN: r232243

re PR target/67462 (FAIL: gcc.dg/ifcvt-3.c scan-rtl-dump ce1 "3 true changes made")

PR target/67462
* gcc.dg/ifcvt-3.c: Only compile on lp64 targets, include also i?86
if lp64.

From-SVN: r232242

re PR target/69071 (ICE: in decompose, at rtl.h:2107 with -g)

PR target/69071
* lra-eliminations.c (move_plus_up): Only move plus up
if subreg of the constant can be simplified into constant
and use the simplified subreg of the constant instead of
the original constant.

* gcc.dg/pr69071.c: New test.

From-SVN: r232241

PR c++/68795: fix uninitialized close_paren_loc in cp_parser_postfix_expression

gcc/cp/ChangeLog:
PR c++/68795
* parser.c (cp_parser_postfix_expression): Initialize
close_paren_loc to UNKNOWN_LOCATION; only use it if
it has been written to by
cp_parser_parenthesized_expression_list.
(cp_parser_parenthesized_expression_list): Document the behavior
with respect to the CLOSE_PAREN_LOC param.

From-SVN: r232238

re PR c++/69211 (g++ ICE on x86_64-linux-gnu (verify_gimple failed))

PR c++/69211
* cp-gimplify.c (cp_fold): If COMPOUND_EXPR or MODIFY_EXPR
folded operands have side-effects, but folding changed any of them,
build a new tree with the folded operands instead of returning the
unfolded tree.

* g++.dg/opt/pr69211.C: New test.

From-SVN: r232237

fold-const.c (fold_convertible_p): Don't return true for conversion of VECTOR_TYPE to same sized integral type.

* fold-const.c (fold_convertible_p): Don't return true
for conversion of VECTOR_TYPE to same sized integral type.
(fold_convert_loc): Fix up formatting.  Fold conversion of
VECTOR_TYPE to same sized integral type using VIEW_CONVERT_EXPR
instead of NOP_EXPR.

From-SVN: r232236

re PR tree-optimization/69214 (ICE (segfault) at -Os on x86_64-linux-gnu in "fail_abnormal_edge_coalesce")

PR tree-optimization/69214
* tree-vrp.c (simplify_cond_using_ranges): Don't propagate
innerop into a comparison if SSA_NAME_OCCURS_IN_ABNORMAL_PHI.
Formatting fix.

* gcc.c-torture/compile/pr69214.c: New test.

From-SVN: r232235

re PR go/68980 (ps -o cmd in gotest isn't portable)

PR 68980
    libgo/testsuite: portable ps usage in gotest

    gotest is using "ps" to list descendant sleep
    processes in its timeout handling, grepping
    the command name.

    We are currently using the "cmd" ps output
    formatter, which is non-portable. We should
    use "comm" which is part of the POSIX
    standard, and outputs only the command name
    and not the arguments.

    Fixes https://gcc.gnu.org/PR68980

    Reviewed-on: https://go-review.googlesource.com/18426

From-SVN: r232234

re PR tree-optimization/69207 (gcc.target/aarch64/vldN_1.c ICEs at -O3)

PR tree-optimization/69207
* tree-vect-slp.c (vect_get_constant_vectors): For
VECTOR_BOOLEAN_TYPE_P, assert op has integral type instead of
fold_convertible_p to vector_type's element type, and always
use VCE for non-VECTOR_BOOLEAN_TYPE_P.

From-SVN: r232233

allocator_traits<allocator<T>> partial specialization

PR libstdc++/60976
* include/bits/alloc_traits.h (allocator_traits<allocator<_Tp>>):
Define partial specialization.
* testsuite/20_util/shared_ptr/cons/58659.cc: Add construct and
destroy members to std::allocator explicit specialization.

From-SVN: r232232

pr66232-10.c: Do not compile on ia32 target.

* gcc.target/i386/pr66232-10.c: Do not compile on ia32 target.
* gcc.target/i386/pr66232-11.c: Ditto.
* gcc.target/i386/pr66232-12.c: Ditto.
* gcc.target/i386/pr66232-13.c: Ditto.

From-SVN: r232231

re PR tree-optimization/69173 (ICE (segfault) in vinfo_for_stmt)

2016-01-11  Richard Biener  <rguenther@suse.de>

PR tree-optimization/69173
* tree-vect-loop.c (vect_fixup_scalar_cycles_with_patterns): Only
fixup the cycle if all stmts are in a pattern.

* gcc.dg/torture/pr69173.c: New testcase.

From-SVN: r232230

re PR middle-end/68999 (FAIL: gfortran.fortran-torture/execute/save_1.f90 execution)

PR middle-end/68999
* alias.c (base_alias_check): Move check for addresses with
alignment ANDs before the call for compare_base_decls.
(memrefs_conflict_p): Return -1 for different decls
that went through alignment adjustments.

From-SVN: r232229

[AArch64] PR rtl-optimization/68796: Add patterns for QImode and HImode comparison with zero

PR rtl-optimization/68796
* config/aarch64/aarch64.md (*and<mode>_compare0): New pattern.
* config/aarch64/aarch64.c (aarch64_select_cc_mode): Handle HImode
and QImode comparisons against zero with CC_NZmode.
* config/aarch64/iterators.md (short_mask): New mode_attr.

* gcc.target/aarch64/tst_5.c: New test.
* gcc.target/aarch64/tst_6.c: Likewise.

From-SVN: r232228

nvptx plugin: do not force JIT target SM version

When link_ptx runs, a CUDA device is already bound to current thread, so the
driver library knows the target architecture.  There isn't any benefit from
forcing a specific target here; on the contrary, hardcoding sm_30 breaks
offloading on later (Maxwell, sm_5x) devices.

* plugin/plugin-nvptx.c (link_ptx): Do not set CU_JIT_TARGET.

From-SVN: r232227

Remove snprintf from <avx512>_(load|store)<mode>_mask

This patch removes snprintf from <avx512>_(load|store)<mode>_mask
patterns.

* config/i386/sse.md (<avx512>_load<mode>_mask): Remove
snprintf.
(<avx512>_store<mode>_mask): Likewise.

From-SVN: r232225

[RTL-ifcvt] PR rtl-optimization/68841: Make sure one basic block doesn't clobber CC reg usage of the other

        PR rtl-optimization/68841
        * ifcvt.c (struct noce_if_info): Add orig_x field.
        (bbs_ok_for_cmove_arith): Add to_rename parameter.
        Don't record conflicts on to_rename if it's present.
        Allow memory destinations in sets.
        (noce_try_cmove_arith): Call bbs_ok_for_cmove_arith even on simple
        blocks, passing orig_x to the checks.
        (noce_process_if_block): Set if_info->orig_x appropriately.

        * gcc.dg/pr68841.c: New test.
        * gcc.c-torture/execute/pr68841.c: New test.

From-SVN: r232223

Add missing phi args in create_parallel_loop

2016-01-11  Tom de Vries  <tom@codesourcery.com>

PR tree-optimization/69069
* tree-parloops.c (create_parallel_loop): Add missing phi args.

From-SVN: r232221

re PR rtl-optimization/68920 (Undesirable if-conversion for a rarely taken branch)

gcc/

2016-01-11  Yuri Rumyantsev  <ysrumyan@gmail.com>

PR rtl-optimization/68920
* config/i386/i386.c (ix86_option_override_internal): Restrict number
of conditional moves for  RTL if-conversion to 1 for
TARGET_ONE_IF_CONV_INSN.
* config/i386/i386.h (TARGET_ONE_IF_CONV_INSN): New macros.
* config/i386/x86-tune.def (X86_TUNE_ONE_IF_CONV_INSN): New macros.
* params.def (PARAM_MAX_RTL_IF_CONVERSION_INSNS) : Introduce new
parameter to restirct number of conditional moves for
RTL if-conversion.
* doc/invoke.texi (max-rtl-if-conversion-insns): Document it.
* ifcvt.c (bb_ok_for_noce_convert_multiple_sets): Limit number of
conditionl moves.

gcc/testsuite/

2016-01-11  Yuri Rumyantsev  <ysrumyan@gmail.com>

PR rtl-optimization/68920
* gcc.dg/ifcvt-4.c: Add "--param max-rtl-if-conversion-insns=3" option
for ix86 targets.
* gcc.dg/ifcvt-5.c: New test.

From-SVN: r232220

[PR69123] fix handling of MEMs in VTA to avoid dataflow oscillation

The problem arises because we used to drop overwritten MEMs from loc
lists of VALUEs, but not of other onepart variables, and it just so
happens that, by doing so, block 6 in the testcase has no D#5 in its
output in the first pass, because the MEM holding its (previous) value
was correctly dropped from value 88:88, but gains it in the second
pass because D#5 has the MEM location incoming directly in its loc
list, rather than indirectly in a VALUE.

This incorrect binding enables other blocks to believe they have a
tentative binding for D#5 in some cycles, but others, still operating
on the early conclusion, believe there isn't, and they oscillate from
that.

Since we check for escaping MEMs in clobbers, we won't lose anything
relevant by dropping call-clobbered or overwritten MEMs in all onepart
variables, and this ensures the loc intersection operation in onepart
vars won't let a MEM through that wasn't present in earlier
iterations.

for  gcc/ChangeLog

PR bootstrap/69123
* var-tracking.c (drop_overlapping_mem_locs): Operate on all
onepart vars.  Fix typo in comment.  Fix reversed condition in
unshare test.
(dataflow_set_remove_mem_locs): Operate on all onepart vars.

for gcc/testsuite/ChangeLog

PR bootstrap/69123
* g++.dg/pr69123.C: New.

From-SVN: r232218

[PR69123] make dataflow_set_different details more verbose

for  gcc/ChangeLog

PR bootstrap/69123
* var-tracking.c (dump_onepart_variable_differences): New.
(dataflow_set_different): If a detailed dump is requested,
delay early returns and dump differences between onepart
variables present before and after, and added variables.

From-SVN: r232217

re PR target/69010 (Boolean vector constant with a scalar mode is expanded incorrectly)

gcc/

PR target/69010
* expr.c (expand_expr_real_1): For boolean vector constants
with a scalar mode use const_scalar_mask_from_tree.
(const_scalar_mask_from_tree): New.
* optabs.c (expand_vec_cond_mask_expr): Use mask mode
assigned to a mask type to handle constants.

gcc/testsuite/

PR target/69010
* gcc.target/i386/pr69010.c: New test.

From-SVN: r232216

[PR ipa/69044] Do not clone for param removal when not possible

2016-01-11  Martin Jambor  <mjambor@suse.cz>

PR ipa/69044
* ipa-cp.c (estimate_local_effects): Do not clone for removal of
useless parameters if we cannot change function signature.

testsuite/
* gcc.target/i386/chkp-pr69044.c: New test.

From-SVN: r232215

[PR ipa/66616] Copy can_change_signature flag to artificial thunks

2016-01-11  Martin Jambor  <mjambor@suse.cz>

PR ipa/66616
* cgraphclones.c (duplicate_thunk_for_node): Copy can_change_signature
flag.

From-SVN: r232214

Don't allow latch with phi in try_transform_to_exit_first_loop_alt

2016-01-11  Tom de Vries  <tom@codesourcery.com>

PR tree-optimization/69109
* tree-parloops.c (try_transform_to_exit_first_loop_alt): Don't allow
latch with phi.

* gcc.dg/autopar/pr69109-2.c: New test.
* gcc.dg/autopar/pr69109.c: New test.

From-SVN: r232212

Handle case that outer phi res is not used in a phi in gather_scalar_reductions

2016-01-11  Tom de Vries  <tom@codesourcery.com>

PR tree-optimization/69108
* tree-parloops.c (gather_scalar_reductions): Handle case that outer phi
res is not used in a phi.

* gcc.dg/autopar/pr69108.c: New test.

From-SVN: r232211

Fix docs for -frandom-seed.

2016-01-11  Yury Gribov  <y.gribov@samsung.com>

PR 67425
* common.opt (frandom-seed): Fix parameter name.
* doc/invoke.texi (frandom-seed): Ditto and describe parameter.

From-SVN: r232209

Don't parallelize loops if libgomp not supported

2016-01-11  Tom de Vries  <tom@codesourcery.com>

PR tree-optimization/69058
* tree-parloops.c (pass_parallelize_loops::execute): Return 0 if libgomp
not supported.

From-SVN: r232208

arc.opt (mdiv-rem): Add period to the end.

2016-01-11  Andrew Burgess  <andrew.burgess@embecosm.com>

        * config/arc/arc.opt (mdiv-rem): Add period to the end.
        (mcode-density): Likewise.

From-SVN: r232207

Daily bump.

From-SVN: r232206

Fix -Wmisleading indentation false-positive for do-while statement

gcc/c-family/ChangeLog:

PR c++/69029
* c-indentation.c (should_warn_for_misleading_indentation):
Don't warn about do-while statements.

gcc/testsuite/ChangeLog:

PR c++/69029
* c-c++-common/Wisleading-indentation.c: Augment test.

From-SVN: r232202

re PR fortran/69154 (ICE in gfc_trans_where_2, at fortran/trans-stmt.c:5005 on *-linux)

2016-01-10  Thomas Koenig  <tkoenig@gcc.gnu.org>

PR fortran/69154
* frontend-passes.c (in_where):  New variable.
(inline_matmul_assign):  Don't try this if we are within
a WHERE statement.
(gfc_code_walker):  Keep track of in_where.

2016-01-10  Thomas Koenig  <tkoenig@gcc.gnu.org>

PR fortran/69154
* gfortran.dg/inline_matmul_12.f90:  New test.

From-SVN: r232201

re PR fortran/67779 (Strange ordering with strings in extended object)

2016-01-10  Paul Thomas  <pault@gcc.gnu.org>

PR fortran/67779
* trans_array.c (gfc_conv_scalarized_array_ref): Add missing
se->use_offset from condition for calculation of 'base'.

2016-01-10  Paul Thomas  <pault@gcc.gnu.org>

PR fortran/67779
* gfortran.dg/actual_array_offset_1: New test.

From-SVN: r232200

Don't parallelize loops containing phis with addr_exprs

2016-01-10  Tom de Vries  <tom@codesourcery.com>

PR tree-optimization/69062
* tree-parloops.c (loop_has_phi_with_address_arg): New function.
(parallelize_loops): Don't paralelize loop that has phi with address
arg.

* gcc.dg/autopar/pr69062.c: New test.

From-SVN: r232199

Fix scan-tree-dump syntax

gcc/testsuite/
* gcc.dg/vect/slp-perm-1.c: Fix scan-tree-dump syntax.
* gcc.dg/vect/slp-perm-2.c: Likewise.
* gcc.dg/vect/slp-perm-3.c: Likewise.
* gcc.dg/vect/slp-perm-5.c: Likewise.
* gcc.dg/vect/slp-perm-6.c: Likewise.
* gcc.dg/vect/slp-perm-7.c: Likewise.
* gcc.dg/vect/slp-perm-8.c: Likewise.

From-SVN: r232197

Only allow single exit phi for reduction in try_create_reduction_list

2016-01-10  Tom de Vries  <tom@codesourcery.com>

PR tree-optimization/69039
* tree-parloops.c (try_create_reduction_list): Only allow single exit
phi for reduction.

* gcc.dg/autopar/pr69039.c: New test.

From-SVN: r232196

Daily bump.

From-SVN: r232194

re PR middle-end/68743 (FAIL: gfortran.dg/aint_anint_1.f90   -O0  execution test)

PR middle-end/68743
* match.pd: Require target has function_c99_misc before doing
truncl(extend(x)) and trunc(extend(x)) -> extend(truncf(x)), etc.

From-SVN: r232191

configure.ac (isl_options_set_schedule_serialize_sccs): Also use GMPINC.

* configure.ac (isl_options_set_schedule_serialize_sccs): Also
use GMPINC.
* configure: Regenerate.

From-SVN: r232190

re PR c++/69113 (ICE: in vague_linkage_p, at cp/decl2.c:1875 with -fno-weak)

PR c++/69113
* decl2.c (comdat_linkage): Only set DECL_COMDAT if TREE_PUBLIC is set.

* g++.dg/pr69113.C: New test.

From-SVN: r232189

re PR middle-end/50865 (Invalid code generation for INT64_MIN % 1 on x86_64)

PR middle-end/50865
PR tree-optimization/69097
* fold-const.h (expr_not_equal_to): New prototype.
* fold-const.c: Include stringpool.h and tree-ssanames.h.
(expr_not_equal_to): New function.
* match.pd (X % -Y is the same as X % Y): Don't optimize
unless X is known not to be equal to minimum or Y is known
not to be equal to -1.
* tree-vrp.c (simplify_div_or_mod_using_ranges): Add GSI argument.
fold TRUNC_MOD_EXPR if the second argument is not a power of two.
(simplify_stmt_using_ranges): Adjust caller.
(vrp_finalize): Call set_value_range on SSA_NAMEs before calling
substitute_and_fold.

* gcc.c-torture/execute/pr50865.c: New test.
* gcc.c-torture/execute/pr69097-1.c: New test.
* gcc.c-torture/execute/pr69097-2.c: New test.
* gcc.dg/pr69097-1.c: New test.
* gcc.dg/pr69097-2.c: New test.

From-SVN: r232188

re PR tree-optimization/69164 (ICE in create_tmp_var, at gimple-expr.c:468)

PR c++/69164
* class.c (layout_class_type): Use copy_node to copy FIELD_DECLs.

* g++.dg/opt/pr69164.C: New test.

From-SVN: r232187

re PR c++/69158 (ICE in in cxx_eval_indirect_ref, at cp/constexpr.c:2598)

PR c++/69158

* constexpr.c (cxx_fold_indirect_ref): Handle array type differing
in completion.

From-SVN: r232186

Daily bump.

From-SVN: r232184

ipa-icf.c (sem_item_optimizer::merge_classes): Do not ICE on VAR_DECL w/o DECL_NAME.

* ipa-icf.c (sem_item_optimizer::merge_classes): Do not ICE on VAR_DECL
w/o DECL_NAME.

From-SVN: r232179

re PR tree-optimization/69167 (internal compiler error: SSA corruption)

        PR tree-optimization/69167
        * gimple-fold.c (replace_stmt_with_simplification): Also punt if
new SSA_NAME_OCCURS_IN_ABNORMAL_PHI SSA_NAMEs appear in operands of
ops[0] comparison.
* gimple-match-head.c (maybe_push_res_to_seq): Likewise.

* gcc.dg/pr69167.c: New test.

From-SVN: r232178

re PR c++/68449 (ICE in cxx_eval_constant_expression on atomic_load in C++)

PR c++/68449
* constexpr.c (cxx_eval_constant_expression): Handle NULL initializer.

* g++.dg/pr68449.C: New.

From-SVN: r232176

[PATCH] Fix PR68707

gcc/:
* tree-vect-slp.c (vect_analyze_slp_instance): Cancel permuted SLP
instances that can be handled via vect_load_lanes.

gcc/testsuite/:

* lib/target-supports.exp (check_effective_target_vect_load_lanes): New.
* gcc.dg/vect/slp-perm-1.c: Look for vect_load_lanes instead of SLP
on platforms supporting it.
* gcc.dg/vect/slp-perm-2.c: Likewise.
* gcc.dg/vect/slp-perm-3.c: Likewise.
* gcc.dg/vect/slp-perm-5.c: Likewise.
* gcc.dg/vect/slp-perm-7.c: Likewise.
* gcc.dg/vect/slp-perm-8.c: Likewise.
* gcc.dg/vect/slp-perm-6.c: Look for vect_load_lanes in addition to SLP
on platforms supporting it.

Co-Authored-By: Richard Biener <rguenther@suse.de>
From-SVN: r232174

symtab.c (symtab_node::equal_address_to): Return -1 instead of 2 if we can't determine address equivalence.

* symtab.c (symtab_node::equal_address_to): Return -1 instead of 2
if we can't determine address equivalence.
* alias.c (compare_base_decl): Update for changed return value of
symtab_node::equal_address_to.

From-SVN: r232172

constexpr.c (cxx_eval_call_expression): Remove convert_to_void workaround.

* constexpr.c (cxx_eval_call_expression): Remove convert_to_void
workaround.

From-SVN: r232168

PR c++/68983 (BE)

PR c++/67557
gcc/
* function.c (assign_temp): Guard against TREE_ADDRESSABLE types here.
* expr.c (store_field): Not here.
* tree-cfgcleanup.c (fixup_noreturn_call): Don't clear LHS of a
call with TREE_ADDRESSABLE type.
* tree-cfg.c (verify_gimple_call): Adjust.
gcc/cp/
* cvt.c (convert_to_void): Don't strip a TARGET_EXPR of
TREE_ADDRESSABLE type.

From-SVN: r232167

PR c++/68983 (FE)

PR c++/67557
* call.c (unsafe_copy_elision_p): Look through COMPOUND_EXPR.

From-SVN: r232166

Use -std option in test for C99 macros in <cmath>

* testsuite/26_numerics/headers/cmath/
c99_classification_macros_c++.cc: Rename to ...
* testsuite/26_numerics/headers/cmath/
c99_classification_macros_c++98.cc: Here and add -std=gnu++98.
* testsuite/26_numerics/headers/cmath/
c99_classification_macros_c++0x.cc: Rename to ...
* testsuite/26_numerics/headers/cmath/
c99_classification_macros_c++11.cc: Here.

From-SVN: r232165

vxworks.h (VXWORKS_LIBGCC_SPEC): Don't link shared RTPs with libc_internal.

* config/vxworks.h (VXWORKS_LIBGCC_SPEC): Don't link shared RTPs with
libc_internal.

From-SVN: r232164

Fix PR number in previous commit

From-SVN: r232163

Add missing include for std::__addressof

PR libstdc++/69160
* include/bits/uses_allocator.h: Add missing include.

From-SVN: r232162

[PATCH][RS6000] Migrate reduction optabs in paired.md

* gcc.target/rs6000/paired.md (reduc_smax_v2sf): Rename to...
(reduc_smax_scal_v2sf): ...here, make result SFmode, extract element.
(reduc_smin_v2sf): Rename to...
(reduc_smin_scal_v2sf): ...here, make result SFmode, extract element.
(reduc_splus_v2sf): Rename to...
(reduc_plus_scal_v2sf): ...here, make result SFmode, extract element.

From-SVN: r232158

re PR tree-optimization/69162 (ICE in create_tmp_var, at gimple-expr.c:468)

PR tree-optimization/69162
* gimplify.c (gimplify_va_arg_expr): Encode original type of
valist argument in another argument.
(gimplify_modify_expr): Adjust for the above change.  Cleanup.
* tree-stdarg.c (expand_ifn_va_arg_1): Use new 3rd argument
to determine the va_list type, build a MEM_REF instead of
build_fold_indirect_ref.

* gcc.dg/pr69162.c: New test.

From-SVN: r232156

re PR tree-optimization/69172 (ICE in make_ssa_name_fn, at tree-ssanames.c:266)

PR tree-optimization/69172
* gimple-fold.c (gimple_fold_builtin_memory_chk): Pass type to
gimple_build.

* gcc.dg/pr69172.c: New test.

From-SVN: r232155

re PR tree-optimization/67781 (wrong code generated on big-endian with -O1 -fexpensive-optimizations)

2016-01-08  Thomas Preud'homme  <thomas.preudhomme@arm.com>

    gcc/
    PR tree-optimization/67781
    * tree-ssa-math-opts.c (find_bswap_or_nop): Zero out bytes in cmpxchg
    and cmpnop in two steps: first the ones not accessed in original
    gimple expression in a endian independent way and then the ones not
    accessed in the final result in an endian-specific way.

    gcc/testsuite/
    PR tree-optimization/67781
    * gcc.c-torture/execute/pr67781.c: New file.

From-SVN: r232154

re PR tree-optimization/69083 (ICE at -O3 in 64-bit mode on x86_64-linux-gnu (verify_gimple failed))

PR tree-optimization/69083
* tree-vect-slp.c (vect_get_constant_vectors): For
VECTOR_BOOLEAN_TYPE_P assert op is fold_convertible_p to vector_type's
element type.  If op is fold_convertible_p to vector_type's element
type, use NOP_EXPR instead of VCE.

* gcc.dg/vect/pr69083.c: New test.

From-SVN: r232153

Restrict to linux and gnu targets as not all math.h implementations support the issignaling macro.

Restrict to linux and gnu targets as not all math.h implementations support the
issignaling macro. Use -fexcess-precision=standard for compiler options. Use fabs()
only when  __FLT_EVAL_METHOD__ == 0.

gcc/testsuite/
* gcc.dg/pr61441.c: Restrict to linux and gnu targets.
Use -fexcess-precision=standard for compiler options.
Use fabs() only when  __FLT_EVAL_METHOD__ == 0.

From-SVN: r232152

re PR fortran/69128 (OpenMP workshare problem with SUM())

PR fortran/69128
* trans.h (OMPWS_SCALARIZER_BODY): Define.
(OMPWS_NOWAIT): Renumber.
* trans-stmt.c (gfc_trans_where_3): Only set OMPWS_SCALARIZER_WS
if OMPWS_SCALARIZER_BODY is not set already, and set also
OMPWS_SCALARIZER_BODY until the final loop creation.
* trans-expr.c (gfc_trans_assignment_1): Likewise.
* trans-openmp.c (gfc_trans_omp_workshare): Also clear
OMPWS_SCALARIZER_BODY.
* trans-array.c (gfc_trans_scalarized_loop_end): Don't create
OMP_FOR if OMPWS_SCALARIZER_BODY is set.

* gfortran.dg/gomp/pr69128.f90: New test.

From-SVN: r232151

re PR c++/69145 (Bogus 'warning: #pragma implementation for ‘...’ appears after file is included')

PR c++/69145
* files.c (cpp_included_before): If IS_ADHOC_LOC (location), lookup
real location from the line_table.

* g++.dg/ext/pr69145-1.C: New test.
* g++.dg/ext/pr69145-2-very-long-filename.cc: New file.
* g++.dg/ext/pr69145-2.h: New file.

From-SVN: r232150

shrink-wrap: Once more PRs 67778, 68634, and now 68909

If a candidate PRE cannot get the prologue because a block BB is
reachable from it, but PRE does not dominate BB, we try again with the
dominators of PRE.  That "try again" needs to again consider BB though,
we aren't done with it.

PR rtl-optimization/67778
PR rtl-optimization/68634
PR rtl-optimization/68909
* shrink-wrap.c (try_shrink_wrapping): Add comment.  Don't pop
block from the stack until done with it.  Remove a superfluous
bitmap set.  Remove a superfluous bitmap test.

From-SVN: r232148

PR c/68966 - atomic_fetch_* on atomic_bool not diagnosed

gcc/ChangeLog:
* doc/extend.texi (__atomic Builtins, __sync Builtins): Document
constraint on the type of arguments.

gcc/c-family/ChangeLog:
* c-common.c (sync_resolve_size): Reject first argument when it's
a pointer to _Bool.

gcc/testsuite/ChangeLog:
* gcc.dg/atomic-fetch-bool.c: New test.
* gcc.dg/sync-fetch-bool.c: New test.

From-SVN: r232147

Daily bump.

From-SVN: r232146

longlong: fix sh -Wundef builds

This file fails when building for SuperH as it assumes __SHMEDIA__
is always defined.  Update the code to check if it's defined.

From-SVN: r232143

freebsd.h: Rename SUBTARGET_OVERRIDE_OPTIONS to SUBTARGET_OVERRIDE_INTERNAL_OPTIONS.

2016-01-07  Andreas Tobler  <andreast@gcc.gnu.org>

    * config/arm/freebsd.h: Rename SUBTARGET_OVERRIDE_OPTIONS to
    SUBTARGET_OVERRIDE_INTERNAL_OPTIONS. Adjust to check
    unaligned_access on the gcc_options set.
    * config/arm/arm.c (arm_option_override_internal): Use
    SUBTARGET_OVERRIDE_INTERNAL_OPTIONS.

From-SVN: r232141

re PR target/69140 (stack alignment + O1 breaks with Microsoft ABI)

2016-01-07  Uros Bizjak  <ubizjak@gmail.com>

PR target/69140
* config/i386/i386.c (ix86_frame_pointer_required): Enable
frame pointer for TARGET_64BIT_MS_ABI when stack is misaligned.

2016-01-07  Uros Bizjak  <ubizjak@gmail.com>

Revert
2016-01-06  Uros Bizjak  <ubizjak@gmail.com>

PR target/69140
* config/i386/i386.c (ix86_expand_prologue): Declare fs.sp_valid
depending on frame_pointer_needed before remaining integer and SSE
registers are saved.

From-SVN: r232140

Move pr66680.f90 to libgomp

gcc/testsuite/

PR fortran/66680
gfortran.dg/gomp/pr66680.f90: Moved to libgomp.

libgomp/

PR fortran/66680
* testsuite/libgomp.fortran/pr66680.f90: New test.

From-SVN: r232138

re PR target/1078 (Problems with attributes documentation)

2016-01-07  Sandra Loosemore <sandra@codesourcery.com>

PR 1078
gcc/
* doc/extend.texi (Nvidia PDX Function Attributes): New section.

From-SVN: r232136

Add a test for PR66655

PR target/66655
* g++.dg/pr66655.C: New test.
* g++.dg/pr66655_1.cc: Test support file.
* g++.dg/pr66655.h: Test header file.

From-SVN: r232133

Use std::addressof in insert iterators, allocators and promises

PR libstdc++/69105
PR libstdc++/69106
PR libstdc++/69114
* include/bits/stl_iterator.h (back_insert_iterator,
front_insert_iterator, insert_iterator): Use __addressof (LWG 2324).
* include/bits/uses_allocator.h (__use_alloc): Use __addressof.
* include/std/future (__future::base::_State_baseV2::__setter):
Likewise.
* include/std/scoped_allocator (__outermost): Likewise.
* testsuite/20_util/scoped_allocator/69114.cc: New.
* testsuite/20_util/uses_allocator/69114.cc: New.
* testsuite/30_threads/promise/69106.cc: New.

From-SVN: r232129

re PR fortran/66680 (ICE with openmp, a loop and a type bound procedure)

2016-01-07  Dominique d'Humieres <dominiq@lps.ens.fr>

PR fortran/66680
gfortran.dg/gomp/pr66680.f90: New test.

From-SVN: r232127

Add round_nimm_scalar_predicate and use xBm

Add round_nimm_scalar_predicate for scalar SSE integer to floating point
conversions since round_nimm_predicate is for vector operand.

round_constraint can't be used on vector SSE patterns since it is mapped
to "vm".  The "xBm" constraint must be used in this case.

gcc/

PR target/69171
* config/i386/sse.md (<sse>_sqrt<mode>2<mask_name><round_name>):
Use the "xBm" constraint.
(float<sseintvecmodelower><mode>2<mask_name><round_name):
Likewise.
(sse_cvtsi2ss<round_name>): Use round_nimm_scalar_predicate.
(sse_cvtsi2ssq<round_name>): Likewise.
(sse_cvtss2si<round_name>): Likewise.
(sse_cvtss2siq<round_name>): Likewise.
(sse2_cvtsi2sdq<round_name>): Likewise.
(sse2_cvtsd2si<round_name>): Likewise.
(sse2_cvtsd2siq<round_name>): Likewise.
* config/i386/subst.md (round_nimm_scalar_predicate): New
predicate.

gcc/testsuite/

PR target/69171
* gcc.target/i386/pr69171-1.c: New test.
* gcc.target/i386/pr69171-2.c: Likewise.
* gcc.target/i386/pr69171-3.c: Likewise.
* gcc.target/i386/pr69171-4.c: Likewise.
* gcc.target/i386/pr69171-5.c: Likewise.
* gcc.target/i386/pr69171-6.c: Likewise.

From-SVN: r232126

Fix SSA inconsistencies with invalid asm reg names (PR67639)

PR middle-end/67639
* varasm.c (make_decl_rtl): Mark invalid register vars as
DECL_EXTERNAL.

testsuite/
PR middle-end/67639
* c-c++-common/pr67639.c: New test.

From-SVN: r232125

Fix a minor issue found by a static analyzer (PR66206).

PR rtl-optimization/66206
* bt-load.c (find_btr_use): Change first arg to be a pointer to an rtx.
All callers changed.

From-SVN: r232124

re PR tree-optimization/69141 (-O2 -fdump-tree-fre ICEs)

PR tree-optimization/69141
* tree-ssa-pre.c: Include langhooks.h.
(eliminate_dom_walker::before_dom_children): Use
lang_hooks.decl_printable_name instead of
cgraph_node::get ()->name ().

* g++.dg/opt/pr69141.C: New test.

From-SVN: r232123

re PR c/68960 (__attribute__ ((aligned ())) is ignored for OpenMP private variables)

PR middle-end/68960
* gimple-expr.c (copy_var_decl): If var has DECL_USER_ALIGN set, copy
it and DECL_ALIGN too.

* testsuite/libgomp.c/pr68960.c: New test.

From-SVN: r232122

Daily bump.

From-SVN: r232120

* gcc.target/sparc/20151219-1.c: Skip in 64-bit mode.

From-SVN: r232116

correct openacc.h filename

From-SVN: r232114

Reorder function types.

* config/mips/mips-ftypes.def: Sort to lexicographical order.

From-SVN: r232113

Remove noexcept from function template that can throw

PR libstdc++/69092
* include/bits/basic_string.h (basic_string::_S_copy_chars<_Iterator>):
Remove _GLIBCXX_NOEXCEPT.
testsuite/21_strings/basic_string/cons/char/69092.cc: New.

From-SVN: r232112

re PR target/69140 (stack alignment + O1 breaks with Microsoft ABI)

PR target/69140
* config/i386/i386.c (ix86_expand_prologue): Declare fs.sp_valid
depending on frame_pointer_needed before remaining integer and SSE
registers are saved.

testsuite/ChangeLog:

PR target/69140
* gcc.target/i386/pr69140.c: New test

From-SVN: r232111