re PR tree-optimization/79389 (30% performance regression in SciMark2 MonteCarlo)

2017-02-24  Richard Biener  <rguenther@suse.de>

PR tree-optimization/79389
* gimple-ssa-split-paths.c (is_feasible_trace): Verify more
properly that a threading opportunity exists.  Detect conditional
copy/constant propagation opportunities.

* gcc.dg/tree-ssa/split-path-10.c: New testcase.

From-SVN: r245696

Daily bump.

From-SVN: r245695

re PR c++/79361 (ICE redefining a template function as defaulted or deleted)

/cp
2017-02-23  Paolo Carlini  <paolo.carlini@oracle.com>

PR c++/79361
* pt.c (register_specialization): Check duplicate_decls return value
for error_mark_node and pass it back.

/testsuite
2017-02-23  Paolo Carlini  <paolo.carlini@oracle.com>

PR c++/79361
* g++.dg/cpp0x/pr79361-1.C: New.
* g++.dg/cpp0x/pr79361-2.C: Likewise.

From-SVN: r245692

visium.md (type): Add trap.

* config/visium/visium.md (type): Add trap.
(b): New mode attribute.
(*btst): Rename into...
(*btst<mode>): ...this and adjust.
(*cbranchsi4_btst_insn): Rename into...
(*cbranch<mode>4_btst_insn): ...this and adjust.
(trap): New define_insn.

From-SVN: r245691

re PR tree-optimization/79389 (30% performance regression in SciMark2 MonteCarlo)

PR tree-optimization/79389
* ifcvt.c (struct noce_if_info): Add rev_cond field.
(noce_reversed_cond_code): New function.
(noce_emit_store_flag): Use rev_cond if non-NULL instead of
reversed_comparison_code.  Formatting fix.
(noce_try_store_flag): Test rev_cond != NULL in addition to
reversed_comparison_code.
(noce_try_store_flag_constants): Likewise.
(noce_try_store_flag_mask): Likewise.
(noce_try_addcc): Use rev_cond if non-NULL instead of
reversed_comparison_code.
(noce_try_cmove_arith): Likewise.  Formatting fixes.
(noce_try_minmax, noce_try_abs): Clear rev_cond.
(noce_find_if_block): Initialize rev_cond.
(find_cond_trap): Call noce_get_condition with then_bb == trap_bb
instead of false as last argument never attempt to reverse it
afterwards.

From-SVN: r245690

re PR tree-optimization/79663 (r244815 causes 10% regression for spec1k/172.mgrid on AArch64)

2017-01-21  Bin Cheng  <bin.cheng@arm.com>

PR tree-optimization/79663
* tree-predcom.c (combine_chains): Process refs in reverse order
only for ZERO length chains, and add explaining comment.

From-SVN: r245689

re PR tree-optimization/79578 (Unnecessary instructions in generated code)

PR tree-optimization/79578
* tree-ssa-dse.c (clear_bytes_written_by): Use OEP_ADDRESS_OF
in call to operand_equal_p.

From-SVN: r245688

Ensure <experimental/iterator> includes <iterator>

* include/experimental/iterator: Include <iterator>.
* testsuite/experimental/iterator/requirements.cc: Check for contents
of <iterator>.

From-SVN: r245686

PR 68749: S/390: Disable ifcvt-4.c for -m31.

gcc/testsuite/ChangeLog:

2017-02-23  Dominik Vogt  <vogt@linux.vnet.ibm.com>

PR 68749
* gcc.dg/ifcvt-4.c: Disable for -m31, use -march=z196.

From-SVN: r245684

re PR c/79684 (Conditional jump or move depends on uninitialised value in GIMPLE FE)

2017-02-23  Richard Biener  <rguenther@suse.de>

PR c/79684
* gimple-parser.c (c_parser_gimple_statement): Use set_error
to initialize c_exprs to return.
(c_parser_gimple_binary_expression): Likewise.
(c_parser_gimple_unary_expression): Likewise.
(c_parser_gimple_postfix_expression): Likewise.

From-SVN: r245681

re PR target/71017 (libgcc/config/i386/cpuinfo.c:346:17: runtime error: left shift of 1 by 31 places cannot be represented in type 'int')

2017-01-23  Dominique d'Humieres  <dominiq@lps.ens.fr>

PR target/71017
* config/i386/cpuid.h: Fix another undefined behavior.

From-SVN: r245680

re PR tree-optimization/79683 (SLP vectorizer drops gs: prefix)

2017-02-23  Richard Biener  <rguenther@suse.de>

PR tree-optimization/79683
* tree-vect-stmts.c (vect_analyze_stmt): Do not overwrite
vector types for data-refs.

* gcc.target/i386/pr79683.c: New testcase.

From-SVN: r245679

Fix typo in ChangeLog

From-SVN: r245678

Change default of param not being smaller that min.

2017-02-23  Martin Liska  <mliska@suse.cz>

* params.def (PARAM_MIN_NONDEBUG_INSN_UID): Change default to 0.

From-SVN: r245677

re PR middle-end/79665 (gcc's signed (x*x)/200 is slower than clang's)

PR middle-end/79665
* internal-fn.c (get_range_pos_neg): Moved to ...
* tree.c (get_range_pos_neg): ... here.  No longer static.
* tree.h (get_range_pos_neg): New prototype.
* expr.c (expand_expr_real_2) <case TRUNC_DIV_EXPR>: If both arguments
are known to be in between 0 and signed maximum inclusive, try to
expand both unsigned and signed divmod and use the cheaper one from
those.

From-SVN: r245676

re PR tree-optimization/79578 (Unnecessary instructions in generated code)

PR tree-optimization/79578
* tree-ssa-dse.c (clear_bytes_written_by): Use operand_equal_p
to compare base operands.

PR tree-optimization/79578
* g++.dg/tree-ssa/ssa-dse-3.C: New test.

From-SVN: r245675

PR c++/79679 - missing destructor for argument

PR c++/79679 - missing destructor for argument
* call.c (build_over_call): Don't pass tf_no_cleanup to argument
conversions.

From-SVN: r245672

Daily bump.

From-SVN: r245671

rs6000: Fix fsel pattern (PR79211)

The fsel define_insn uses fpr_reg_operand for its predicates.  This
won't work because passes can put a hard register in the operands: in
the testcase, combine likes to forward the parameter registers to what
then is still an smin, and then split1 uses "*s<minmax><mode>3_fpr"
(which has gpc_reg_operand).  And then we have a GPR in the operand,
which does not match fpr_reg_operand.

It seems to me the predicates should be gpc_reg_operand here as well.
This patch changes that.

PR target/79211
* config/rs6000/rs6000.md (*fsel<SFDF:mode><SFDF2:mode>4): Use
gpc_reg_operand instead of fpr_reg_operand.

From-SVN: r245667

Fix MIPS o32 calling convention for MSA and FP vector types

gcc/
* config/mips/mips.c (mips_return_in_memory): Force FP
vector types to be returned in memory for o32 ABI.

gcc/testsuite/

* gcc.target/mips/msa-fp-cc.c: New test.

From-SVN: r245666

* pt.c (do_class_deduction): Handle 0 argument case.

From-SVN: r245665

re PR c++/79664 (ICE with #pragma omp parallel in constexpr function)

PR c++/79664
* parser.c (cp_parser_omp_teams, cp_parser_omp_target): Use
SET_EXPR_LOCATION on OMP_TARGET/OMP_TEAMS tree.
* constexpr.c (potential_constant_expression_1): Handle
OMP_*, OACC_* and CILK_* trees.  Use error_at with
EXPR_LOC_OR_LOC (t, input_location) computed early
instead of error, or error_at with location_of (t).

* g++.dg/gomp/teams-1.C: Adjust expected diagnostic location.
* g++.dg/cpp1y/constexpr-throw.C: Likewise.
* g++.dg/gomp/pr79664.C: New test.

From-SVN: r245662

dwarf2out.c (gen_variable_die): For -gdwarf-5...

* dwarf2out.c (gen_variable_die): For -gdwarf-5, use DW_TAG_variable
instead of DW_TAG_member for static data member declarations and don't
set no_linkage_name for static inline data members.
(gen_member_die): For -gdwarf-5 don't change DW_TAG_variable
to DW_TAG_member.

* g++.dg/debug/dwarf2/inline-var-2.C: New test.

From-SVN: r245661

re PR c/79662 (ICE on invalid code in convert_arguments in c/c-typeck.c:3452)

PR c/79662
* c-typeck.c (convert_arguments): Handle error_mark_node.

* gcc.dg/enum-incomplete-4.c: New test.

From-SVN: r245660

instead of -Wnormalized=<options>

2017-02-22  Martin Liska  <mliska@suse.cz>

* doc/invoke.texi: Replace inequality signs with square brackets
for -Wnornalized.
2017-02-22  Martin Liska  <mliska@suse.cz>

* c.opt: Replace inequality signs with square brackets
for -Wnornalized.

From-SVN: r245659

re PR c++/79657 (ICE on invalid c++ code in finish_underlying_type cp/semantics.c:3849)

PR c++/79657
* semantics.c (finish_underlying_type): Bail out for incomplete enums.

* g++.dg/ext/underlying_type12.C: New test.

From-SVN: r245658

re PR c++/79653 (ICE on invalid c++ code in cp_check_const_attributes in cp/decl2.c:1423)

PR c++/79653
* parser.c (cp_parser_std_attribute_spec): Don't build the attribute
if the alignas expression is erroneous.
* pt.c (tsubst_attribute): If tsubst_pack_expansion fails, return
error_mark_node.

* g++.dg/cpp0x/alignas10.C: New test.
* g++.dg/cpp0x/alignas9.C: New test.

From-SVN: r245657

re PR tree-optimization/68644 (FAIL: gcc.dg/tree-ssa/ivopts-lt-2.c scan-tree-dump-times ivopts "PHI <p_" 1)

2017-02-22  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>

PR tree-optimization/68644
* gcc.dg/tree-ssa/ivopts-lt-2.c: Skip for powerpc*-*-*.

From-SVN: r245656

Support WORD_REGISTER_OPERATIONS requirements in simplify_operand_subreg

gcc/
PR target/78660
* lra-constraints.c (simplify_operand_subreg): Handle
WORD_REGISTER_OPERATIONS targets.

From-SVN: r245655

re PR target/70465 (Poor code for x87 asm)

PR target/70465
* reg-stack.c (emit_swap_insn): Treat (float_extend:?F (mem:?F))
and (const_double:?F) like (mem:?F) for the purpose of fxch %st(1)
elimination by swapping fld*.

* gcc.target/i386/pr70465-2.c: New test.

From-SVN: r245654

pr61441.c: Use dg-add-options ieee.

* gcc.dg/pr61441.c: Use dg-add-options ieee.

From-SVN: r245652

ChangeLog: Add missing entry.

* ChangeLog: Add missing entry.
* testsuite/ChangeLog: Ditto.

From-SVN: r245651

re PR middle-end/79673 (GIMPLE verification fails when compiling code with __seg_gs)

2017-02-22  Richard Biener  <rguenther@suse.de>

PR tree-optimization/79673
* tree-ssa-pre.c (compute_avail): Use wide_int_to_tree to
convert the [TARGET_]MEM_REF offset INTEGER_CST, scrapping off
irrelevant address-space qualifiers and avoiding a
ADDR_SPACE_CONVERT_EXPR from fold_convert.

* gcc.target/i386/pr79673.c: New testcase.

From-SVN: r245649

re PR tree-optimization/79666 (wrong code (SIGFPE) at -O2 on x86_64-linux-gnu (in both 32-bit and 64-bit modes))

2017-02-22  Richard Biener  <rguenther@suse.de>

PR tree-optimization/79666
* tree-vrp.c (extract_range_from_binary_expr_1): Make sure
to not symbolically negate if that may introduce undefined
overflow.

* gcc.dg/torture/pr79666.c: New testcase.

From-SVN: r245648

Remove wrong assert about gcov_type (PR lto/79587).

2017-02-22  Martin Liska  <mliska@suse.cz>

PR lto/79587
* data-streamer-in.c (streamer_read_gcov_count): Remove assert.
* data-streamer-out.c (streamer_write_gcov_count_stream):
Likewise.
* value-prof.c (stream_out_histogram_value): Make assert more
precise based on type of counter.
2017-02-22  Martin Liska  <mliska@suse.cz>

PR lto/79587
* gcc.dg/tree-prof/pr79587.c: New test.

From-SVN: r245647

Daily bump.

From-SVN: r245646

PR c++/50308 - wrong deprecated warning with ADL

PR c++/17729 - duplicate deprecated warning
* semantics.c (finish_id_expression): Only call mark_used on a
function if we aren't building a call.

From-SVN: r245643

PR c++/41727 - ICE with partial spec of partial instantiation

* pt.c (process_partial_specialization): For now, don't check more
specialized if there is more than one level of args.

From-SVN: r245642

re PR c++/79535 (ICE in verify_ctor_sanity, at cp/constexpr.c:2636)

PR c++/79535
* cp-tree.h (maybe_reject_flexarray_init): Declare.
* init.c (maybe_reject_flexarray_init): No longer static.
Add check for current_function_decl.
* parser.c (cp_parser_late_parse_one_default_arg): Reject
a default mem-initializer for a flexible array.

* g++.dg/ext/flexary23.C: New test.

From-SVN: r245641

re PR rtl-optimization/79593 (Poor/Worse code generation for FPU on versions after 6)

PR target/79593
* config/i386/i386.md (standard_x87sse_constant_load splitter):
Use nonimmediate_operand instead of memory_operand for operand 1.
(float-extend standard_x87sse_constant_load splitter): Ditto.

testsuite/ChangeLog:

PR target/79593
* gcc.target/i386/pr79593.c: New test.

From-SVN: r245640

re PR c++/79654 (ICE on invalid c++ code in register_dtor_fn in cp/decl.c:7877)

PR c++/79654
* decl.c (cp_finish_decomp): Don't set decl's type to error_mark_node
on error.
* pt.c (tsubst_decomp_names): Return error_mark_node if the first
decl after the decomposition artificial decl has error_mark_node.
* decl2.c (prune_vars_needing_no_initialization): Use error_operand_p
instead of just == error_mark_node comparison.

* g++.dg/cpp1z/decomp26.C: New test.

Co-Authored-By: Paolo Carlini <paolo.carlini@oracle.com>
From-SVN: r245639

re PR sanitizer/79589 (ICE in gimplify_compound_expr (gimplify.c:5712) with -fsanitize=undefined)

PR sanitizer/79589
* decl.c: Include gimplify.h.
(cp_finish_decomp): Make sure there is no sharing of trees
in between DECL_VALUE_EXPR of decomposition decls.

* g++.dg/ubsan/pr79589.C: New test.

From-SVN: r245638

re PR tree-optimization/79621 (Missed path isolation opportunity)

PR tree-optimization/79621
* gimple-ssa-isolate-paths.c (find_implicit_erroneous_behavior): Ignore
blocks with edges to themselves.

PR tree-optimization/79621
* gcc.c-torture/compile/pr79621.c: New test.

From-SVN: r245637

re PR c++/79655 (ICE on invalid c++ code in cxx_eval_store_expression in cp/constexpr.c:3464)

PR c++/79655
* constexpr.c (cxx_eval_array_reference): Diagnose negative subscript.

* g++.dg/cpp1y/constexpr-79655.C: New test.

From-SVN: r245636

re PR c++/79639 (ICE with -O and constexpr)

PR c++/79639
* constexpr.c (cxx_eval_store_expression): If *valp is a PTRMEM_CST,
call cplus_expand_constant on it first.

* g++.dg/cpp1y/constexpr-79639.C: New test.

From-SVN: r245635

re PR target/79633 (ICE in gimple_call_arg, at gimple.h:3163)

PR target/79633
* tree-chkp-opt.c (chkp_optimize_string_function_calls): Use
is_gimple_call instead of comparing gimple_code with GIMPLE_CALL.
Use gimple_call_builtin_p.

* gcc.target/i386/mpx/pr79633.c: New test.

From-SVN: r245634

re PR target/79570 (ICE in sel-sched-ir.c:4534 in pr69956.c)

PR target/79570
* sel-sched.c (moveup_expr_cached): Don't call sel_bb_head
on temporarily removed DEBUG_INSNs.

* gcc.dg/pr79570.c: New test.

From-SVN: r245633

re PR c++/79641 (ICE with const variable and attribute)

PR c++/79641
* c-attribs.c (handle_mode_attribute): Use build_qualified_type to
preserve quals.

* c-c++-common/pr79641.c: New test.

From-SVN: r245632

re PR middle-end/79649 (Memset pattern in named address space crashes compiler or generates wrong code)

PR tree-optimization/79649
* tree-loop-distribution.c (classify_partition): Give up on
non-generic address space loads/stores.

* gcc.target/i386/pr79649.c: New test.

From-SVN: r245631

loop.texi (Loop manipulation): Remove nonexistent tree_ssa_loop_version from the documentation.

* doc/loop.texi (Loop manipulation): Remove nonexistent
tree_ssa_loop_version from the documentation.
* cfgloopmanip.c (loop_version): Document CONDITION_BB argument.

From-SVN: r245630

re PR target/79494 (ICE in maybe_record_trace_start, at dwarf2cfi.c:2330)

PR target/79494
* config/i386/i386.c (ix86_expand_split_stack_prologue): Call
make_reg_eh_region_note_nothrow_nononlocal on call_insn.
* config/rs6000/rs6000.c: Include except.h.
(rs6000_expand_split_stack_prologue): Call
make_reg_eh_region_note_nothrow_nononlocal on the call insn.

* gcc.dg/pr79494.c: New test.

From-SVN: r245629

[PR 79579] Avoid segfault on NULL ipa_edge_args_vector

2017-02-21  Martin Jambor  <mjambor@suse.cz>

PR lto/79579
* ipa-prop.c (ipa_prop_write_jump_functions): Bail out if no edges
have been analyzed.

From-SVN: r245628

[PATCH] Properly deprecate -fipa-cp-alignment

2017-02-21  Martin Jambor  <mjambor@suse.cz>

* common.opt (-fipa-cp-alignment): Mark as ignored and preserved
for backward compatibility only.
* doc/invoke.texi (Option Summary): Remove all references to
-fipa-cp-alignment.

From-SVN: r245627

Revert r245598

gcc/
PR target/78660
Revert:
2017-02-20  Matthew Fortune  <matthew.fortune@imgtec.com>

* lra-constraints.c (curr_insn_transform): Handle
WORD_REGISTER_OPERATIONS requirements when reloading SUBREGs.

From-SVN: r245626

Add -Wno-psabi to diagnostic-test-expressions-1.c options

Without this the test fails on 32-bit PowerPC.

gcc/testsuite/
* gcc.dg/plugin/diagnostic-test-expressions-1.c: Add -Wno-psabi
to dg-options.

From-SVN: r245625

Fix typo in -masm-dialect option values.

2017-02-21  Martin Liska  <mliska@suse.cz>

* config/i386/i386.opt: Replace -masm-dialect with -masm.

From-SVN: r245624

[PR translation/79638] "%ntid.y" confuses gcc.pot generation

gcc/
* config/nvptx/nvptx.c (ENTRY_TEMPLATE): Single out "%ntid.y".

From-SVN: r245623

Skip gcc.dg/gimplefe-26.c on non-c99 runtime targets

* gcc.dg/gimplefe-26.c: Require c99_runtime.

From-SVN: r245622

re PR ada/67205 (eliminate No_Implicit_Dynamic_Code restriction violations)

PR ada/67205
* config/arm/arm.c (TARGET_CUSTOM_FUNCTION_DESCRIPTORS): Define.
(arm_function_ok_for_sibcall): Return false for an indirect call by
descriptor if all the argument registers are used.
(arm_relayout_function): Use FUNCTION_ALIGNMENT macro to adjust the
alignment of the function.

From-SVN: r245621

re PR tree-optimization/61441 (ARM aarch64 fails to quiet signaling NaN)

PR tree-optimization/61441
* simplify-rtx.c (simplify_const_unary_operation): For
-fsignaling-nans and sNaN operand, return NULL_RTX rather than
the sNaN unmodified.

From-SVN: r245620

re PR go/79642 (space instead of tab in lang.opt)

PR go/79642
* lang.opt (-fgo-relative-import-path): Change space to tab.

From-SVN: r245619

Daily bump.

From-SVN: r245618

* de.po: Update.

From-SVN: r245615

* de.po, fr.po, sv.po: Update.

From-SVN: r245614

Makefile.in (BUILD_SYSTEM_HEADER_DIR): New make variabe.

2017-02-20  Bernd Edlinger  <bernd.edlinger@hotmail.de>

        * Makefile.in (BUILD_SYSTEM_HEADER_DIR): New make variabe.
        (LIMITS_H_TEST, if_multiarch, stmp-fixinc): Use BUILD_SYSTEM_HEADER_DIR
        instead of SYSTEM_HEADER_DIR.

From-SVN: r245613

PR c++/78139 - destructor needed by new-expression

* call.c (build_special_member_call): Use tf_no_cleanup.

From-SVN: r245612

strncmp-2.c: Portability fixes.

2017-02-14  Aaron Sawdey  <acsawdey@linux.vnet.ibm.com>

        * gcc.dg/strncmp-2.c: Portability fixes.

From-SVN: r245608

re PR target/78056 (build failure on Power7)

gcc/testsuite/ChangeLog:

2017-02-20  Kelvin Nilsen  <kelvin@gcc.gnu.org>

PR target/78056
* gcc.target/powerpc/pr78056-8.c: Remove.

From-SVN: r245607

invoke.texi (use-after-scope-direct-emission-threshold): Fix typos and grammar, use active voice, and clarify.

* doc/invoke.texi (use-after-scope-direct-emission-threshold):
Fix typos and grammar, use active voice, and clarify.

Co-Authored-By: Martin Liska <mliska@suse.cz>
From-SVN: r245606

re PR sanitizer/79558 (ICE: Segfault in ubsan_type_descriptor, at ubsan.c:412)

PR sanitizer/79558
* ubsan.c (ubsan_type_descriptor): Check if TYPE_MAX_VALUE is null.

* c-c++-common/ubsan/bounds-14.c: New test.

From-SVN: r245605

re PR middle-end/79537 (ICE in gimplify_expr, at gimplify.c:12009)

PR middle-end/79537
* gimplify.c (gimplify_expr): Handle unused *&&L;.

* gcc.dg/comp-goto-4.c: New.

From-SVN: r245604

re PR fortran/79599 (typo in diagnostic gfc_error ("DTIO dummy argument at %L be a scalar")

2017-02-20  Paul Thomas  <pault@gcc.gnu.org>

PR fortran/79599
* interface.c (check_dtio_arg_TKR_intent): Supply 'must'
missing from error message.

2017-02-20  Paul Thomas  <pault@gcc.gnu.org>

PR fortran/79523
* interface.c (gfc_find_typebound_dtio_proc): Guard test for
flavor attribute by checking that symbol is resolved.

From-SVN: r245603

re PR target/79568 (ICE in extract_insn, at recog.c:2311 for pr70325.c (with -mavx512bw))

PR target/79568
* config/i386/i386.c (ix86_expand_builtin): Handle
OPTION_MASK_ISA_AVX512VL and OPTION_MASK_ISA_64BIT in
ix86_builtins_isa[fcode].isa as a requirement of those
flags and any other flag in the bitmask.
(ix86_init_mmx_sse_builtins): Use 0 instead of
~OPTION_MASK_ISA_64BIT as mask.
* config/i386/i386-builtin.def (__builtin_ia32_rdtsc,
__builtin_ia32_rdtscp, __builtin_ia32_pause, __builtin_ia32_bsrsi,
__builtin_ia32_rdpmc, __builtin_ia32_rolqi, __builtin_ia32_rolhi,
__builtin_ia32_rorqi, __builtin_ia32_rorhi): Likewise.

* gcc.target/i386/pr79568-1.c: New test.
* gcc.target/i386/pr79568-2.c: New test.
* gcc.target/i386/pr79568-3.c: New test.

From-SVN: r245602

Ensure the mode used to create split registers is suppported

gcc/
PR target/78012
* lra-constraints.c (split_reg): Check requested split mode
is supported by the register.

From-SVN: r245601

Partial revert of r243782 to restore previous behavior

gcc/
* lra-constraints.c (simplify_operand_subreg): Remove early
return false.

From-SVN: r245600

Tighten condition for converting SUBREG reloads from OP_OUT to OP_INOUT

gcc/
PR target/78660
* lra-constraints.c (curr_insn_transform): Tighten condition
for converting SUBREG reloads from OP_OUT to OP_INOUT.

From-SVN: r245599

Handle WORD_REGISTER_OPERATIONS when reloading (subreg (reg))

gcc/
PR target/78660
* lra-constraints.c (curr_insn_transform): Handle
WORD_REGISTER_OPERATIONS requirements when reloading SUBREGs.

From-SVN: r245598

re PR fortran/79382 (DTIO ICE)

2017-02-16  Paul Thomas  <pault@gcc.gnu.org>

PR fortran/79382
* decl.c (access_attr_decl): Test for presence of generic DTIO
interface and emit error if not present.

2017-02-16  Paul Thomas  <pault@gcc.gnu.org>

PR fortran/79382
* io/transfer.c (check_dtio_proc): New function.
(formatted_transfer_scalar_read): Use it.
(formatted_transfer_scalar_write): ditto.

2017-02-16  Paul Thomas  <pault@gcc.gnu.org>

PR fortran/79382
* gfortran.dg/dtio_10.f90 : Change test of error message.
* gfortran.dg/dtio_23.f90 : New test.
* gfortran.dg/dtio_24.f90 : New test.

From-SVN: r245596

re PR fortran/79434 ([submodules] separate module procedure breaks encapsulation)

2017-02-20  Paul Thomas  <pault@gcc.gnu.org>

PR fortran/79434
* parse.c (check_component, parse_union): Whitespace.
(set_syms_host_assoc): For a derived type, check if the module
in which it was declared is one of the submodule ancestors. If
it is, make the components public. Otherwise, reset attribute
'host_assoc' and set 'use-assoc' so that encapsulation is
preserved.

2017-02-20  Paul Thomas  <pault@gcc.gnu.org>

PR fortran/79434
* gfortran.dg/submodule_25.f08 : New test.

From-SVN: r245595

PR c++/78282 - auto template and pack expansion

* pt.c (find_parameter_packs_r): Don't walk into the type of
templates other than template template-parameters.

From-SVN: r245594

PR c++/79606 - ICE with this->base_member in NSDMI

* class.c (build_base_path): Check processing_template_decl.

From-SVN: r245593

PR c++/79607 - ICE with T{} initializer

* decl.c (type_dependent_init_p): Check the type of a CONSTRUCTOR.

From-SVN: r245592

PR c++/79566 - elaborated-type-specifier in range for

* parser.c (cp_parser_simple_declaration): Fix check for type
definition.

From-SVN: r245591

PR c++/79400 - confusing suggestion of 'noexcept'

* parser.c (cp_parser_exception_specification_opt): Remove
suggestion for deprecated dynamic exception-specification.

From-SVN: r245590

PR c++/79470 - partial ordering with reference parameters

* pt.c (unify) [INDIRECT_REF]: Handle pack expansions.

From-SVN: r245589

PR c++/79500 - ICE with non-template deduction guide

* pt.c (do_class_deduction): Use STRIP_TEMPLATE rather than
DECL_TEMPLATE_RESULT.

From-SVN: r245588

PR c++/79580 - ICE with compound literal

* parser.c (cp_parser_class_head): If we're in the middle of an
expression, use ts_within_enclosing_non_class.

From-SVN: r245587

PR c++/79503 - inherited ctor taking base class

* call.c (add_function_candidate): Also check that
DECL_INHERITED_CTOR_BASE is reference-related to the parameter type.

From-SVN: r245586

Daily bump.

From-SVN: r245585

re PR fortran/79447 ([F08] gfortran rejects valid & accepts invalid internal subprogram in a submodule)

2017-02-19  Paul Thomas  <pault@gcc.gnu.org>

PR fortran/79447
* decl.c (gfc_set_constant_character_len): Whitespace.
(gfc_match_end): Catch case where a procedure is contained in
a module procedure and ensure that 'end procedure' is the
correct termination.

2017-02-19  Paul Thomas  <pault@gcc.gnu.org>

PR fortran/79447
* gfortran.dg/submodule_24.f08 : New test.

From-SVN: r245582

re PR fortran/79229 (ICE in gfc_trans_assignment_1 with -fcheck=mem)

gcc/testsuite/ChangeLog:

2017-02-19  Andre Vehreschild  <vehre@gcc.gnu.org>

PR fortran/79229
* gfortran.dg/class_allocate_24.f90: New test.

gcc/fortran/ChangeLog:

2017-02-19  Andre Vehreschild  <vehre@gcc.gnu.org>

PR fortran/79229
* trans-expr.c (gfc_trans_assignment_1): Deref indirect refs when
compiling with -fcheck=mem to check the pointer and not the data.

From-SVN: r245581

re PR fortran/79402 (ICE with submodules: module procedure interface defined in parent module)

2017-02-19  Paul Thomas  <pault@gcc.gnu.org>

PR fortran/79402
* resolve.c (fixup_unique_dummy): New function.
(gfc_resolve_expr): Call it for dummy variables with a unique
symtree name.

2017-02-19  Paul Thomas  <pault@gcc.gnu.org>

PR fortran/79402
* gfortran.dg/submodule_23.f90: New test.

From-SVN: r245580

re PR fortran/79335 (Conditional jump or move depends on uninitialised in value  get_scalar_to_descriptor_type(tree_node*, symbol_attribute) (trans-expr.c:53))

gcc/fortran/ChangeLog:

2017-02-19  Andre Vehreschild  <vehre@gcc.gnu.org>

PR fortran/79335
* trans-array.c (duplicate_allocatable_coarray): Ensure attributes
passed are properly initialized.
(structure_alloc_comps): Same.
* trans-expr.c (gfc_trans_structure_assign): Same.

From-SVN: r245579

re PR c++/79380 (ICE on C++11 code with illegal alignas specifier on x86_64-linux-gnu: unexpected expression ‘f’ of kind overload)

/cp
2017-02-19  Paolo Carlini  <paolo.carlini@oracle.com>

PR c++/79380
* typeck.c (cxx_alignas_expr): Reject a non-integral alignas
argument.

/testsuite
2017-02-19  Paolo Carlini  <paolo.carlini@oracle.com>

PR c++/79380
* g++.dg/cpp0x/alignas8.C: New.

From-SVN: r245578

Revert:

2016-05-30  Uros Bizjak  <ubizjak@gmail.com>

* config/i386/sync.md (mfence_nosse): Use "lock orl $0, -4(%esp)".

From-SVN: r245577

C++17 GB50 resolution (P0505R0)

2017-02-19  Dinka Ranns  <dinka.ranns@googlemail.com>

C++17 GB50 resolution
* include/std/chrono (duration::operator++()): Add
_GLIBCXX17_CONSTEXPR.
(duration::operator++(int)): Likewise.
(duration::operator--()): Likewise.
(duration::operator--(int)): Likewise.
(duration::operator+=(const duration&)): Likewise.
(duration::operator-=(const duration&)): Likewise.
(duration::operator*=(const rep&)): Likewise.
(duration::operator/=(const rep&)): Likewise.
(duration::operator%=(const rep&)): Likewise.
(duration::operator%=(const duration&)): Likewise.
(time_point::operator+=(const duration&)): Likewise.
(time_point::operator-=(const duration&)): Likewise.
* testsuite/20_util/duration/arithmetic/constexpr_c++17.cc: New test.
* testsuite/20_util/duration/literals/range.cc: Adjust dg-error.
* testsuite/20_util/time_point/arithmetic/constexpr.cc: New test.

From-SVN: r245575

PR c++/69523 make -Wliteral-suffix control warning

gcc:

PR c++/69523
* doc/invoke.texi (C++ Dialect Options) [-Wliteral-suffix]: Update
description.

gcc/cp:

2017-02-19  Eric Fiselier  <eric@efcs.ca>
    Jonathan Wakely  <jwakely@redhat.com>

PR c++/69523
* parser.c (cp_parser_unqualified_id): Use OPT_Wliteral_suffix to
control warning about literal suffix identifiers without a leading
underscore.

gcc/testsuite:

2017-02-19  Eric Fiselier  <eric@efcs.ca>
    Jonathan Wakely  <jwakely@redhat.com>

PR c++/69523
* g++.dg/cpp0x/Wliteral-suffix2.C: New test.

From-SVN: r245574

* doc/xml/manual/debug.xml: Adjust link to ThreadSanitizer.

From-SVN: r245572

gimple-parser.c (c_parser_gimple_postfix_expression): Check return value of c_parser_parse_ssa_name against error_mark_node and emit...

2017-02-19  Prathamesh Kulkarni  <prathamesh.kulkarni@linaro.org>

* gimple-parser.c (c_parser_gimple_postfix_expression): Check return
value of c_parser_parse_ssa_name against error_mark_node and emit
error if ssa name is anonymous and written as default definition.

From-SVN: r245571

gimple-pretty-print.c (dump_ternary_rhs): Adjust gimple dump format for FMA_EXPR.

2017-02-19  Prathamesh Kulkarni  <prathamesh.kulkarni@linaro.org>

* gimple-pretty-print.c (dump_ternary_rhs): Adjust gimple dump format
for FMA_EXPR.

c/
* gimple-parser.c (c_parser_gimple_postfix_expression): Handle
FMA_EXPR.

testsuite/
* gcc.dg/gimplefe-26.c: New test.

From-SVN: r245570