re PR sanitizer/81604 (Ubsan type reporting can be bogus in some cases)

PR sanitizer/81604
* ubsan.c (ubsan_type_descriptor): For UBSAN_PRINT_ARRAY don't
change type to the element type, instead add eltype variable and
use it where we are interested in the element type.

* c-c++-common/ubsan/pr81604.c: New test.

From-SVN: r250728

re PR tree-optimization/81603 (Various compiler UB on very large constant offsets)

PR tree-optimization/81603
* ipa-polymorphic-call.c
(ipa_polymorphic_call_context::ipa_polymorphic_call_context): Perform
offset arithmetic in offset_int, bail out if the resulting bit offset
doesn't fit into shwi.

From-SVN: r250727

Remove a Java-specific hunk.

2017-07-31  Martin Liska  <mliska@suse.cz>

* gimplify.c (mostly_copy_tree_r): Remove Java specific hunk.
(gimplify_save_expr): Fix comment.

From-SVN: r250726

Daily bump.

From-SVN: r250725

i386: Update INCOMING_FRAME_SP_OFFSET for exception handler

Since there is an extra error code passed to the exception handler,
INCOMING_FRAME_SP_OFFSET is return address plus error code for the
exception handler.  This patch updates INCOMING_FRAME_SP_OFFSET to
the correct value for the exception handler.

This patch exposed a bug in DWARF stack frame CFI generation, which
assumes that INCOMING_FRAME_SP_OFFSET is the same for all functions:

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=81570

It sets and caches the incoming stack frame offset with the same
INCOMING_FRAME_SP_OFFSET for all functions.  When there are both
exception handler and normal function in the same input, the wrong
incoming stack frame offset is used for exception handler or normal
function, which leads to

FAIL: gcc.dg/guality/pr68037-1.c   -O2 -flto -fuse-linker-plugin -fno-fat-lto-objects  line 33 error == 0x12345670
FAIL: gcc.dg/guality/pr68037-1.c   -O2 -flto -fuse-linker-plugin -fno-fat-lto-objects  line 33 frame->ip == 0x12345671
FAIL: gcc.dg/guality/pr68037-1.c   -O2 -flto -fuse-linker-plugin -fno-fat-lto-objects  line 33 frame->cs == 0x12345672
FAIL: gcc.dg/guality/pr68037-1.c   -O2 -flto -fuse-linker-plugin -fno-fat-lto-objects  line 33 frame->flags == 0x12345673
FAIL: gcc.dg/guality/pr68037-1.c   -O2 -flto -fuse-linker-plugin -fno-fat-lto-objects  line 33 frame->sp == 0x12345674
FAIL: gcc.dg/guality/pr68037-1.c   -O2 -flto -fuse-linker-plugin -fno-fat-lto-objects  line 33 frame->ss == 0x12345675

With the patch for PR 81570:

https://gcc.gnu.org/ml/gcc-patches/2017-07/msg01851.html

applied, there are no regressions on i686 and x86-64.

gcc/

PR target/79793
* config/i386/i386.c (ix86_function_arg): Update arguments for
exception handler.
(ix86_compute_frame_layout): Set the initial stack offset to
INCOMING_FRAME_SP_OFFSET.  Update red-zone offset with
INCOMING_FRAME_SP_OFFSET.
(ix86_expand_epilogue): Don't pop the 'ERROR_CODE' off the
stack before exception handler returns.
* config/i386/i386.h (INCOMING_FRAME_SP_OFFSET): Add the
the 'ERROR_CODE' for exception handler.

gcc/testsuite/

PR target/79793
* gcc.dg/guality/pr68037-1.c: Update gdb breakpoints.
* gcc.target/i386/interrupt-5.c (interrupt_frame): New struct.
(foo): Check the builtin return address against the return address
in interrupt frame.
* gcc.target/i386/pr79793-1.c: New test.
* gcc.target/i386/pr79793-2.c: Likewise.

From-SVN: r250721

i386.h (ASM_PRINTF_EXTENSIONS): New macro.

* config/i386/i386.h (ASM_PRINTF_EXTENSIONS): New macro.
(ASM_OUTPUT_REG_PUSH): Rewrite with new operand modifiers.
(ASM_OUTPUT_REG_POP): Ditto.
* config/i386/i386.c (ix86_asm_output_function_label): Use fputs
instead of asm_fprintf to output pure string.

From-SVN: r250720

c-format.c (asm_fprintf_char_table): Add 'z' to format_chars.

* c-format.c (asm_fprintf_char_table): Add 'z' to format_chars.

From-SVN: r250719

Daily bump.

From-SVN: r250718

c-ada-spec.c (dump_generic_ada_node): Take into account signedness for enumeral types.

* c-ada-spec.c (dump_generic_ada_node): Take into account signedness
for enumeral types.
(print_ada_declaration): Add missing guard for record types.

From-SVN: r250714

debug.h (struct gcc_debug_hooks): Add IMPLICIT argument to imported_module_or_decl hook.

* debug.h (struct gcc_debug_hooks): Add IMPLICIT argument
to imported_module_or_decl hook.
(debug_nothing_tree_tree_tree_bool): Remove.
(debug_nothing_tree_tree_tree_bool_bool): New declaration.
* debug.c (do_nothing_debug_hooks): Use
debug_nothing_tree_tree_tree_bool_bool instead of
debug_nothing_tree_tree_tree_bool.
* vmsdbgout.c (vmsdbg_debug_hooks): Likewise.
* dbxout.c (dbx_debug_hooks, xcoff_debug_hooks): Likewise.
* sdbout.c (sdb_debug_hooks): Likewise.
* dwarf2out.c (dwarf2_lineno_debug_hooks): Likewise.
(gen_namespace_die): Add DW_AT_export_symbols attribute if
langhook wants it.
(dwarf2out_imported_module_or_decl): Add IMPLICIT argument,
if true, -gdwarf-5 and decl will have DW_AT_export_symbols
attribute, don't add anything.
cp/
* cp-objcp-common.c (cp_decl_dwarf_attribute): Handle
DW_AT_export_symbols.
* name-lookup.c (emit_debug_info_using_namespace): Add IMPLICIT
argument, pass it through to the debug hook.
(finish_namespace_using_directive): Adjust
emit_debug_info_using_namespace caller.
(push_namespace): Likewise.  Call it after setting
DECL_NAMESPACE_INLINE_P.
(cp_emit_debug_info_for_using): Pass false as new argument to
the imported_module_or_decl debug hook.
fortran/
* trans-decl.c (gfc_trans_use_stmts): Pass false as new argument to
the imported_module_or_decl debug hook.
ada/
* gcc-interface/utils.c (gnat_write_global_declarations): Pass false
as new argument to the imported_module_or_decl debug hook.
testsuite/
* g++.dg/debug/dwarf2/inline-ns-1.C: New test.
* g++.dg/debug/dwarf2/inline-ns-2.C: New test.

From-SVN: r250713

use c++ for fold_buildN_loc

gcc/ChangeLog:

2017-07-28  Trevor Saunders  <tbsaunde+gcc@tbsaunde.org>

* fold-const.c (fold_build1_stat_loc): Remove _stat from name.
(fold_build2_stat_loc): Likewise.
(fold_build3_stat_loc): Likewise.
* fold-const.h (fold_build1, fold_build2, fold_build3): Adjust.
(fold_build1_loc): Remove macro.
(fold_build2_loc): Likewise.
(fold_build3_loc): Likewise.

From-SVN: r250712

use c++ for gimple_build_debug_bind{,_source}

gcc/ChangeLog:

2017-07-28  Trevor Saunders  <tbsaunde+gcc@tbsaunde.org>

* gimple.c (gimple_build_debug_bind_stat): Remove _stat from name.
(gimple_build_debug_bind_source_stat): Likewise.
* gimple.h (gimple_build_debug_bind): Remove macro.
(gimple_build_debug_bind_source): Likewise.

From-SVN: r250711

use c++ for bitmap_initialize

gcc/ChangeLog:

2017-07-28  Trevor Saunders  <tbsaunde+gcc@tbsaunde.org>

* bitmap.c (bitmap_alloc): Adjust.
(bitmap_gc_alloc): Likewise.
* bitmap.h (bitmap_initialize_stat): Remove _stat from name.

From-SVN: r250710

simplify the bitmap alloc_stat functions with c++

gcc/fortran/ChangeLog:

2017-07-28  Trevor Saunders  <tbsaunde+gcc@tbsaunde.org>

* resolve.c (find_reachable_labels): Adjust.

gcc/ChangeLog:

2017-07-28  Trevor Saunders  <tbsaunde+gcc@tbsaunde.org>

* bitmap.c (bitmap_obstack_alloc_stat): Rename to bitmap_alloc.
(bitmap_gc_alloc_stat): Rename to bitmap_gc_alloc.
* bitmap.h (bitmap_obstack_alloc_stat): Adjust prototype.
(bitmap_gc_alloc_stat): Likewise.
(BITMAP_ALLOC, BITMAP_GGC_ALLOC): Adjust.

From-SVN: r250709

replace shallow_copy_rtx_stat with c++

gcc/ChangeLog:

2017-07-28  Trevor Saunders  <tbsaunde+gcc@tbsaunde.org>

* rtl.c (shallow_copy_rtx_stat): Remove _stat from name.
* rtl.h (shallow_copy_rtx): Remove macro.

From-SVN: r250708

replace rtx_alloc_stat with c++

gcc/ChangeLog:

2017-07-28  Trevor Saunders  <tbsaunde+gcc@tbsaunde.org>

* emit-rtl.c (gen_raw_REG): Adjust.
* gengenrtl.c (gendef): Likewise.
* rtl.c (rtx_alloc_stat): Remove _stat from name.
* rtl.h (rtx_alloc): Remove macro.

From-SVN: r250707

use c++ for build_tree_list{,_vec}_stat

gcc/ChangeLog:

2017-07-28  Trevor Saunders  <tbsaunde+gcc@tbsaunde.org>

* tree.c (build_tree_list_vec_stat): Remove _stat from name.
(build_tree_list_stat): Likewise.
* tree.h (build_tree_list): Remove macro.
(build_tree_list_vec): Likewise.

From-SVN: r250706

use C++ for {make,build}_vector_stat

gcc/ChangeLog:

2017-07-28  Trevor Saunders  <tbsaunde+gcc@tbsaunde.org>

* tree.c (make_vector_stat): Remove _stat from name.
(build_vector_stat): Likewise.
* tree.h (make_vector_stat): Remove macro.
(build_vector_stat): Likewise.

From-SVN: r250705

remove unused build_var_debug_value prototype

gcc/ChangeLog:

2017-07-28  Trevor Saunders  <tbsaunde+gcc@tbsaunde.org>

* tree.h (build_var_debug_value): Remove prototype.

From-SVN: r250704

use c++ for tree_cons_stat

gcc/ChangeLog:

2017-07-28  Trevor Saunders  <tbsaunde+gcc@tbsaunde.org>

* tree.c (tree_cons_stat): Remove _stat from name.
* tree.h (tree_cons): Remove macro.

From-SVN: r250703

use c++ instead of build_vl_exp_stat

gcc/ChangeLog:

2017-07-28  Trevor Saunders  <tbsaunde+gcc@tbsaunde.org>

* tree.c (build_vl_exp_stat): Remove _stat from name.
* tree.h (build_vl_exp): Remove macro.

From-SVN: r250702

use c++ instead of build_decl_stat

gcc/ChangeLog:

2017-07-28  Trevor Saunders  <tbsaunde+gcc@tbsaunde.org>

* tree.c (build_decl_stat): Remove _stat from name.
* tree.h (build_decl): Remove macro.

From-SVN: r250701

replace gimple_alloc_stat with c++

gcc/ChangeLog:

2017-07-28  Trevor Saunders  <tbsaunde+gcc@tbsaunde.org>

* gimple.c (gimple_build_with_ops_stat): Adjust.
(gimple_alloc_stat): Remove _stat from name.
* gimple.h (gimple_alloc): Remove macro.

From-SVN: r250700

use c++ instead of {make,grow}_tree_vec_stat

gcc/ChangeLog:

2017-07-28  Trevor Saunders  <tbsaunde+gcc@tbsaunde.org>

* tree.c (make_tree_vec_stat): Remove _stat from name.
(grow_tree_vec_stat): Likewise.
* tree.h (make_tree_vec_stat): Adjust prototype.
(grow_tree_vec_stat): Likewise.
(make_tree_vec): Remove macro.
(grow_tree_vec): Likewise.

From-SVN: r250699

use c++ instead of buildN_stat{,_loc}

gcc/ChangeLog:

2017-07-28  Trevor Saunders  <tbsaunde+gcc@tbsaunde.org>

* fold-const.c (fold_build1_stat_loc): Adjust.
(fold_build2_stat_loc): Likewise.
(fold_build3_stat_loc): Likewise.
* tree.c (build0_stat): Remove _stat from name.
(build1_stat): Likewise.
(build2_stat): Likewise.
(build3_stat): Likewise.
(build4_stat): Likewise.
(build5_stat): Likewise.
* tree.h (build1_loc): Remove macro, and rename _stat function
to this.
(build2_loc): Likewise.
(build3_loc): Likewise.
(build4_loc): Likewise.
(build5_loc): Likewise.

From-SVN: r250698

use c++ for make_int_cst_stat

gcc/ChangeLog:

2017-07-28  Trevor Saunders  <tbsaunde+gcc@tbsaunde.org>

* tree.c (make_int_cst_stat): Remove _stat from name.
* tree.h (make_int_cst_stat): Adjust prototype.
(make_int_cst): Remove macro.

From-SVN: r250697

use cxx instead of make_tree_binfo_stat

gcc/ChangeLog:

2017-07-28  Trevor Saunders  <tbsaunde+gcc@tbsaunde.org>

* tree.c (make_tre_binfo_stat): Remove _stat from name.
* tree.h (make_tree_binfo_stat): Adjust prototype.
(make_tree_binfo): Remove.

From-SVN: r250696

use c++ instead of _stat for copy_node_stat

gcc/cp/ChangeLog:

2017-07-28  Trevor Saunders  <tbsaunde+gcc@tbsaunde.org>

* lex.c (copy_decl): Adjust.
(copy_type): Likewise.

gcc/ChangeLog:

2017-07-28  Trevor Saunders  <tbsaunde+gcc@tbsaunde.org>

* tree.c (copy_node_stat): Rename to copy_node.
(build_distinct_type_copy): Adjust.
* tree.h (copy_node_stat): Adjust prototype.
(copy_node): Remove macro.

From-SVN: r250695

use c++ instead of make_node_stat

gcc/ChangeLog:

2017-07-28  Trevor Saunders  <tbsaunde+gcc@tbsaunde.org>

* tree.c (make_node_stat): rename to make_node.
(build_tree_list_stat): Adjust.
(build0_stat): Likewise.
(build2_stat): Likewise.
(build3_stat): Likewise.
(build4_stat): Likewise.
(build5_stat): Likewise.
(build_decl_stat): Likewise.
* tree.h (make_node_stat): Adjust prototype.
(make_node): remove macro.

From-SVN: r250694

Daily bump.

From-SVN: r250693

ppc-auxv.h (PPC_FEATURE2_DARN): New define.

gcc/
* config/rs6000/ppc-auxv.h (PPC_FEATURE2_DARN): New define.
(PPC_FEATURE2_SCV): Likewise.
* config/rs6000/rs6000.c (cpu_supports_info): Use them.

gcc/testsuite/
* gcc.target/powerpc/cpu-builtin-1.c (darn, scv): Add tests.

From-SVN: r250687

compiler: add backend type conversion

    Tweak Interface_field_reference_expression::do_get_backend to apply an
    additional backend type conversion to the returned result.  This is
    needed due to the fact that the top level type of the expression is a
    function descriptor, however the value being manufactured is a pointer
    to <thunk-descriptor, value> struct.

    Reviewed-on: https://go-review.googlesource.com/51712

From-SVN: r250686

configure.ac (target-libffi): Don't disable for AIX.

2017-07-28  Tony Reix  <tony.reix@atos.net>

        * configure.ac (target-libffi): Don't disable for AIX.
        (go): Don't add to unsupported_languages on AIX.
        (target-libgo): Don't disable for AIX.
        * configure: Regenerate.

From-SVN: r250685

* xcoff.c: Don't leak a file descriptor if an archive is malformed.

From-SVN: r250684

compiler: track placeholder pointer types for conversion

    We recently started walking through the hash table of pointer types to
    finalize them.  Unfortunately it is possible to create a new pointer
    type while finalizing an existing one (test case: test/fixedbugs/issue5291)
    and that breaks the iteration.  So, instead, keep a list of
    placeholder pointer types, and iterate through them while permitting
    the list to be extended as we go.

    Reviewed-on: https://go-review.googlesource.com/51771

From-SVN: r250683

compiler: use a single temporary for calls with multiple results

    For calls that return multiple results we used to create a temporary
    of struct type to hold the results, and also create a separate
    temporary for each result.  Then the call expression would copy each
    result out of the struct to the temporary, and Call_result_expression
    would refer to the desired temporary.

    Simplify this to just use a single temporary of struct type, and
    change Call_result_expression to fetch a field of the struct.

    This may reduce some incorrect tree sharing in the backend code.

    Reviewed-on: https://go-review.googlesource.com/51770

From-SVN: r250682

2017-07-28  Tamar Christina  <tamar.christina@arm.com>

* config/aarch64/aarch64.c
(aarch64_internal_mov_immediate): Add new special pattern.
* config/aarch64/aarch64.md (*movdi_aarch64):
Add reg/32bit const mov case.

gcc/testsuite/
2017-07-28  Tamar Christina  <tamar.christina@arm.com>

* gcc.target/aarch64/int_mov_immediate_1.c: New.

From-SVN: r250680

pr80815-3.c: Require vect_perm.

gcc/testsuite
* gcc.dg/vect/pr80815-3.c: Require vect_perm.

From-SVN: r250676

Fix pid_t printing

* fileline.c (fileline_initialize): Print pid_t as long.

From-SVN: r250675

dbl_mov_immediate_1.c: New.

2017-07-28  Tamar Christina  <tamar.christina@arm.com>
    Bilyan Borisov  <bilyan.borisov@arm.com>

* gcc.target/aarch64/dbl_mov_immediate_1.c: New.
* gcc.target/aarch64/flt_mov_immediate_1.c: New.
* gcc.target/aarch64/f16_mov_immediate_1.c: New.
* gcc.target/aarch64/f16_mov_immediate_2.c: New.
* gcc.target/aarch64/pr63304_1.c: Changed to double.

Co-Authored-By: Bilyan Borisov <bilyan.borisov@arm.com>
From-SVN: r250674

aarch64.md (mov<mode>): Generalize.

2017-07-28  Tamar Christina  <tamar.christina@arm.com>
            Richard Sandiford <richard.sandiford@linaro.org>

        * config/aarch64/aarch64.md (mov<mode>): Generalize.
        (*movhf_aarch64, *movsf_aarch64, *movdf_aarch64):
        Add integer and movi cases.
        (movi-split-hf-df-sf split, fp16): New.
        (enabled): Added TARGET_FP_F16INST.
        * config/aarch64/iterators.md (GPF_HF): New.
        * config/aarch64/predicates.md (aarch64_reg_or_fp_float): New.

Co-Authored-By: Richard Sandiford <richard.sandiford@linaro.org>
From-SVN: r250673

2017-07-28  Tamar Christina  <tamar.christina@arm.com>

* config/aarch64/aarch64.c
(aarch64_simd_container_mode): Add prototype.
(aarch64_expand_mov_immediate): Add HI support.
(aarch64_reinterpret_float_as_int, aarch64_float_const_rtx_p: New.
(aarch64_can_const_movi_rtx_p): New.
(aarch64_preferred_reload_class):
Remove restrictions of using FP registers for certain SIMD operations.
(aarch64_rtx_costs): Added new cost for CONST_DOUBLE moves.
(aarch64_valid_floating_const): Add integer move validation.
(aarch64_simd_imm_scalar_p): Remove.
(aarch64_output_scalar_simd_mov_immediate): Generalize function.
(aarch64_legitimate_constant_p): Expand list of supported cases.
* config/aarch64/aarch64-protos.h
(aarch64_float_const_rtx_p, aarch64_can_const_movi_rtx_p): New.
(aarch64_reinterpret_float_as_int): New.
(aarch64_simd_imm_scalar_p): Remove.
* config/aarch64/constraints.md (Uvi): New.
(Dd): Split into Ds and new Dd.
* config/aarch64/aarch64.md (*movsi_aarch64):
Add SIMD mov case.
(*movdi_aarch64): Add SIMD mov case.

From-SVN: r250672

tree-predcom.c: (struct chain): Handle store-store chain in which stores for elimination only...

* tree-predcom.c: (struct chain): Handle store-store chain in which
stores for elimination only store loop invariant values.
(execute_pred_commoning_chain): Ditto.
(prepare_initializers_chain_store_elim): Ditto.
(prepare_finalizers): Ditto.
(is_inv_store_elimination_chain): New function.
(initialize_root_vars_store_elim_1): New function.

From-SVN: r250670

tree-predcom.c: Revise general description of the pass.

* tree-predcom.c: Revise general description of the pass.
(enum chain_type): New enum type for store elimination.
(struct chain): New field supporting store elimination.
(struct component): Ditto.
(dump_chain): Dump store-stores chain.
(release_chain): Release resources.
(split_data_refs_to_components): Compute and create component
contains only stores for elimination.
(get_chain_last_ref_at): New function.
(make_invariant_chain): Initialization.
(make_rooted_chain): Specify chain type in parameter and record it.
(add_looparound_copies): Skip for store-stores chain.
(determine_roots_comp): Compute type of chain and pass it to
make_rooted_chain.
(initialize_root_vars_store_elim_2): New function.
(finalize_eliminated_stores): New function.
(remove_stmt): Handle store for elimination.
(execute_pred_commoning_chain): Execute predictive commoning on
store-store chains.
(determine_unroll_factor): Skip unroll for store-stores chain.
(prepare_initializers_chain_store_elim): New function.
(prepare_initializers_chain): Hanlde store-store chain.
(prepare_finalizers_chain, prepare_finalizers): New function.
(tree_predictive_commoning_loop): Return integer value indicating
if loop is unrolled or lcssa form is corrupted.
(tree_predictive_commoning): Rewrite for lcssa form if necessary.

gcc/testsuite
* gcc.dg/tree-ssa/predcom-dse-1.c: New test.
* gcc.dg/tree-ssa/predcom-dse-2.c: New test.
* gcc.dg/tree-ssa/predcom-dse-3.c: New test.
* gcc.dg/tree-ssa/predcom-dse-4.c: New test.
* gcc.dg/tree-ssa/predcom-dse-5.c: New test.
* gcc.dg/tree-ssa/predcom-dse-6.c: New test.
* gcc.dg/tree-ssa/predcom-dse-7.c: New test.
* gcc.dg/tree-ssa/predcom-dse-8.c: New test.
* gcc.dg/tree-ssa/predcom-dse-9.c: New test.
* gcc.dg/tree-ssa/predcom-dse-10.c: New test.
* gcc.dg/tree-ssa/predcom-dse-11.c: New test.

From-SVN: r250669

tree-predcom.c (initialize_root): Delete.

* tree-predcom.c (initialize_root): Delete.
(execute_pred_commoning_chain): Initialize root vars and replace
reference of non-combined chain directly, rather than call above
function.

From-SVN: r250668

tree-predcom.c (ref_at_iteration): Add parameter NITERS.

* tree-predcom.c (ref_at_iteration): Add parameter NITERS.  Compute
memory reference to DR at (NITERS + ITERS)-th iteration of loop.

From-SVN: r250667

tree-predcom.c (struct chain): New field init_seq.

* tree-predcom.c (struct chain): New field init_seq.
(release_chain): Release init_seq.
(prepare_initializers_chain): Record intialization stmts in above
field.
(insert_init_seqs): New function.
(tree_predictive_commoning_loop): Call insert_init_seqs.

From-SVN: r250666

* tree-predcom.c (determine_roots_comp): Skip trivial components.

From-SVN: r250665

match.pd: Remove superfluous :c.

2017-07-28  Richard Biener  <rguenther@suse.de>

* match.pd: Remove superfluous :c.
* genmatch.c (simplify::id): Add member.
(lower_commutative, lower_opt_convert, lower_cond, lower_for):
Copy id.
(current_id): New global.
(dt_node::parent): Move from ...
(dt_operand::parent): ... here.  Add for_id member.
(is_a_helper <dt_operand *>::test): DT_TRUE is also a dt_operand.
(decision_tree::find_node): Relax order requirement when
merging DT_TRUE nodes to ones inbetween the current simplify
and the one we try to merge with.  Add diagnostic whenever
we need to enforce pattern order by not merging.
(decision_tree::insert): Set current_id.
(decision_tree::print_node): Dump parent node and for_id.
(parser::last_id): Add member.
(parser::push_simplify): Assign unique id.
(parser::parser): Initialize last_id.

From-SVN: r250664

Set DECL_VALUE_EXPR after a debug stmt is generated (PR sanitizer/81340).

2017-07-28  Martin Liska  <mliska@suse.cz>

PR sanitizer/81340
* sanopt.c (sanitize_rewrite_addressable_params): Set VALUE_EXPR after
gimple_build_debug_bind.
2017-07-28  Martin Liska  <mliska@suse.cz>

PR sanitizer/81340
* g++.dg/asan/pr81340.C: New test.

From-SVN: r250660

re PR middle-end/81502 (In some cases the data is moved to memory unnecessarily [partial regression])

2017-07-28  Richard Biener  <rguenther@suse.de>

PR tree-optimization/81502
* match.pd: Add pattern combining BIT_INSERT_EXPR with
BIT_FIELD_REF.
* tree-cfg.c (verify_expr): Verify types of BIT_FIELD_REF
size/pos operands.
(verify_gimple_assign_ternary): Likewise for BIT_INSERT_EXPR pos.
* gimple-fold.c (maybe_canonicalize_mem_ref_addr): Use bitsizetype
for BIT_FIELD_REF args.
* fold-const.c (make_bit_field_ref): Likewise.
* tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.

* gcc.target/i386/pr81502.c: New testcase.

From-SVN: r250659

re PR sanitizer/80998 (Implement -fsanitize=pointer-overflow)

PR sanitizer/80998
* sanopt.c (pass_sanopt::execute): Handle IFN_UBSAN_PTR.
* tree-ssa-alias.c (call_may_clobber_ref_p_1): Likewise.
* flag-types.h (enum sanitize_code): Add SANITIZER_POINTER_OVERFLOW.
Or it into SANITIZER_UNDEFINED.
* ubsan.c: Include gimple-fold.h and varasm.h.
(ubsan_expand_ptr_ifn): New function.
(instrument_pointer_overflow): New function.
(maybe_instrument_pointer_overflow): New function.
(instrument_object_size): Formatting fix.
(pass_ubsan::execute): Call instrument_pointer_overflow
and maybe_instrument_pointer_overflow.
* internal-fn.c (expand_UBSAN_PTR): New function.
* ubsan.h (ubsan_expand_ptr_ifn): Declare.
* sanitizer.def (__ubsan_handle_pointer_overflow,
__ubsan_handle_pointer_overflow_abort): New builtins.
* tree-ssa-tail-merge.c (merge_stmts_p): Handle IFN_UBSAN_PTR.
* internal-fn.def (UBSAN_PTR): New internal function.
* opts.c (sanitizer_opts): Add pointer-overflow.
* lto-streamer-in.c (input_function): Handle IFN_UBSAN_PTR.
* fold-const.c (build_range_check): Compute pointer range check in
integral type if pointer arithmetics would be needed.  Formatting
fixes.
gcc/testsuite/
* c-c++-common/ubsan/ptr-overflow-1.c: New test.
* c-c++-common/ubsan/ptr-overflow-2.c: New test.
libsanitizer/
* ubsan/ubsan_handlers.cc: Cherry-pick upstream r304461.
* ubsan/ubsan_checks.inc: Likewise.
* ubsan/ubsan_handlers.h: Likewise.

From-SVN: r250656

Do not handle VLA in sanitization (PR sanitizer/81460).

2017-07-28  Martin Liska  <mliska@suse.cz>

PR sanitizer/81460
* sanopt.c (sanitize_rewrite_addressable_params): Do not rewrite
parameters that are of a variable-length.
2017-07-28  Martin Liska  <mliska@suse.cz>

PR sanitizer/81460
* gcc.dg/asan/pr81460.c: New test.

From-SVN: r250655

[PowerPC/RTEMS] Add 64-bit support using ELFv2 ABI

Add 64-bit support for RTEMS using the ELFv2 ABI with 64-bit long
double.

gcc/
* config.gcc (powerpc-*-rtems*): Remove rs6000/eabi.h.  Add
rs6000/biarch64.h.
* config/rs6000/rtems.h (ASM_DECLARE_FUNCTION_SIZE): New macro.
(ASM_OUTPUT_SPECIAL_POOL_ENTRY_P): Likewise.
(CRT_CALL_STATIC_FUNCTION): Likewise.
(ASM_DEFAULT_SPEC): New define.
(ASM_SPEC32): Likewise.
(ASM_SPEC64): Likewise.
(ASM_SPEC_COMMON): Likewise.
(ASM_SPEC): Likewise.
(INVALID_64BIT): Likewise.
(LINK_OS_DEFAULT_SPEC): Likewise.
(LINK_OS_SPEC32): Likewise.
(LINK_OS_SPEC64): Likewise.
(POWERPC_LINUX): Likewise.
(PTRDIFF_TYPE): Likewise.
(RESTORE_FP_PREFIX): Likewise.
(RESTORE_FP_SUFFIX): Likewise.
(SAVE_FP_PREFIX): Likewise.
(SAVE_FP_SUFFIX): Likewise.
(SIZE_TYPE): Likewise.
(SUBSUBTARGET_OVERRIDE_OPTIONS): Likewise.
(TARGET_64BIT): Likewise.
(TARGET_64BIT): Likewise.
(TARGET_AIX): Likewise.
(WCHAR_TYPE_SIZE): Likewise.
(WCHAR_TYPE): Undefine.
(TARGET_OS_CPP_BUILTINS): Add 64-bit PowerPC defines.
(CPP_OS_DEFAULT_SPEC): Use previous CPP_OS_RTEMS_SPEC.
(CPP_OS_RTEMS_SPEC): Delete.
(SUBSUBTARGET_EXTRA_SPECS): Remove cpp_os_rtems.  Add
asm_spec_common, asm_spec32, asm_spec64, link_os_spec32, and
link_os_spec64.
* config/rs6000/t-rtems: Add mcpu=e6500/m64 multilibs.

libgcc/
* config/rs6000/ibm-ldouble.c: Disable if defined __rtems__.

From-SVN: r250652

re PR tree-optimization/81578 (ICE in omp_reduction_init_op)

PR tree-optimization/81578
* tree-parloops.c (build_new_reduction): Bail out if
reduction_code isn't one of the standard OpenMP reductions.
Move the details printing after that decision.

* gcc.dg/pr81578.c: New test.

From-SVN: r250651

re PR tree-optimization/81573 (wrong code at -O3 on x86_64-linux-gnu)

2017-07-28  Richard Biener  <rguenther@suse.de>

PR tree-optimization/81573
* gcc.dg/torture/pr81573.c: Explicitely use signed char.

From-SVN: r250650

go-backend.c (go_write_export_data): Use EXCLUDE section for AIX.

* go-backend.c (go_write_export_data): Use EXCLUDE section for
AIX.

From-SVN: r250648

Daily bump.

From-SVN: r250647

predicates.md (volatile_mem_operand): Remove code related to reload_in_progress.

* config/rs6000/predicates.md (volatile_mem_operand): Remove code
related to reload_in_progress.
(splat_input_operand): Likewise.
* config/rs6000/rs6000-protos.h (rs6000_secondary_memory_needed_rtx):
Delete prototype.
* config/rs6000/rs6000.c (machine_function): Remove sdmode_stack_slot
field.
(TARGET_EXPAND_TO_RTL_HOOK): Delete.
(TARGET_INSTANTIATE_DECLS): Likewise.
(legitimate_indexed_address_p): Delete reload_in_progress code.
(rs6000_debug_legitimate_address_p): Likewise.
(rs6000_eliminate_indexed_memrefs): Likewise.
(rs6000_emit_le_vsx_store): Likewise.
(rs6000_emit_move_si_sf_subreg): Likewise.
(rs6000_emit_move): Likewise.
(register_to_reg_type): Likewise.
(rs6000_pre_atomic_barrier): Likewise.
(rs6000_machopic_legitimize_pic_address): Likewise.
(rs6000_allocate_stack_temp): Likewise.
(rs6000_address_for_fpconvert): Likewise.
(rs6000_address_for_altivec): Likewise.
(rs6000_secondary_memory_needed_rtx): Delete function.
(rs6000_check_sdmode): Likewise.
(rs6000_alloc_sdmode_stack_slot): Likewise.
(rs6000_instantiate_decls): Likewise.
* config/rs6000/rs6000.h (SECONDARY_MEMORY_NEEDED_RTX): Delete.
* config/rs6000/rs6000.md (splitter for *movsi_got_internal):
Delete reload_in_progress.
(*vec_reload_and_plus_<mptrsize>): Likewise.
* config/rs6000/vsx.md (vsx_mul_v2di): Likewise.
(vsx_div_v2di): Likewise.
(vsx_udiv_v2di): Likewise.

From-SVN: r250638

rs6000.opt (mlra): Replace with stub.

gcc/

* config/rs6000/rs6000.opt (mlra): Replace with stub.
* config/rs6000/rs6000-cpus.def (POWERPC_MASKS): Delete OPTION_MASK_LRA.
* config/rs6000/rs6000.c (TARGET_LRA_P): Delete.
(rs6000_debug_reg_global): Delete print of LRA status.
(rs6000_option_override_internal): Delete dead LRA related code.
(rs6000_lra_p): Delete function.
* doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mlra.

gcc/testsuite/

* g++.dg/pr69667.C: Remove option -mlra.
* gcc.target/powerpc/dform-1.c: Likewise.
* gcc.target/powerpc/dform-2.c: Likewise.
* gcc.target/powerpc/dform-3.c: Likewise.
* gcc.target/powerpc/p8vector-int128-1.c: Likewise.
* gcc.target/powerpc/p9-vparity.c: Likewise.
* gcc.target/powerpc/pr63491.c: Likewise.
* gcc.target/powerpc/pr67808.c: Likewise.
* gcc.target/powerpc/pr68805.c: Likewise.
* gcc.target/powerpc/pr69461.c: Likewise.
* gcc.target/powerpc/pr71680.c: Likewise.
* gcc.target/powerpc/pr77289.c: Likewise.
* gcc.target/powerpc/pr78458.c: Likewise.
* gcc.target/powerpc/pr78543.c: Likewise.
* g++.dg/pr71294.C: Remove option -mno-lra.
* gcc.target/powerpc/pr71656-1.c: Likewise.
* gcc.target/powerpc/pr71656-2.c: Likewise.
* gcc.target/powerpc/pr71698.c: Likewise.

From-SVN: r250637

re PR c/45784 (gcc OpenMP - error: invalid controlling predicate)

PR c/45784
* c-omp.c (c_finish_omp_for): If the condition is wrapped in
rhs of COMPOUND_EXPR(s), skip them and readd their lhs into
new COMPOUND_EXPRs around the rhs of the comparison.

* testsuite/libgomp.c/pr45784.c: New test.
* testsuite/libgomp.c++/pr45784.C: New test.

From-SVN: r250635

Add RTEMS support

gcc/ChangeLog

2017-07-27  Sebastian Huber  <sebastian.huber@embedded-brains.de>

* config.gcc (riscv*-*-elf*): Add (riscv*-*-rtems*).
* config/riscv/rtems.h: New file.

From-SVN: r250632

[PATCH][AArch64] Fix missing optimization for CMP+AND

During combine GCC tries to merge CMP (with zero) and AND into a TST. However,
in cases where an ANDS operand is not compatible, this was being missed. Adding
a define_split where this operand was moved to a register seems to help out.

Committed on behalf of Sudi Das

---
gcc/

2017-07-27  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
    Sudakshina Das  <sudi.das@arm.com>

* config/aarch64/aarch64.md
(define_split for and<mode>3nr_compare): Move
non aarch64_logical_operand to a register.
(define_split for and_<SHIFT:optab><mode>3nr_compare0): Move non
register immediate operand to a register.
* config/aarch64/predicates.md (aarch64_mov_imm_operand): New.

gcc/testsuite

2017-07-27  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
    Sudakshina Das  <sudi.das@arm.com>

* gcc.target/aarch64/tst_imm_split_1.c: New Test.

Co-Authored-By: Sudakshina Das <sudi.das@arm.com>
From-SVN: r250631

re PR c/81417 (-Wsign-compare should print types being compared)

PR c/81417
* c-warn.c (warn_for_sign_compare): Tweak the warning message.  Print
the types.

* c-c++-common/Wsign-compare-1.c: New test.
* g++.dg/warn/Wsign-compare-2.C: Update dg-warning.
* g++.dg/warn/Wsign-compare-4.C: Likewise.
* g++.dg/warn/Wsign-compare-6.C: Likewise.
* g++.dg/warn/compare1.C: Likewise.
* gcc.dg/compare1.c: Likewise.
* gcc.dg/compare2.c: Likewise.
* gcc.dg/compare4.c: Likewise.
* gcc.dg/compare5.c: Likewise.
* gcc.dg/pr35430.c: Likewise.
* gcc.dg/pr60087.c: Likewise.

From-SVN: r250630

re PR middle-end/81564 (ICE in group_case_labels_stmt())

gcc/
PR middle-end/81564
* tree-cfg.c (group_case_labels_stmt): Handle already deleted blocks.

gcc/testsuite/
PR middle-end/81564
* gcc.dg/pr81564.c: New test.

From-SVN: r250628

re PR tree-optimization/81573 (wrong code at -O3 on x86_64-linux-gnu)

2017-07-27  Richard Biener  <rguenther@suse.de>

PR tree-optimization/81573
PR tree-optimization/81494
* tree-vect-loop.c (vect_create_epilog_for_reduction): Handle
multi defuse cycle case.

* gcc.dg/torture/pr81573.c: New testcase.

From-SVN: r250627

re PR tree-optimization/81571 (ICE at -O3 in both 32-bit and 64-bit modes (internal compiler error: in as_a, at is-a.h:192))

2017-07-27  Richard Biener  <rguenther@suse.de>

PR tree-optimization/81571
* tree-vect-slp.c (vect_build_slp_tree): Properly verify reduction
PHIs.

* gcc.dg/torture/pr81571.c: New testcase.

From-SVN: r250626

Validate that Init value is within range defined by IntegerRange.

2017-07-27  Martin Liska  <mliska@suse.cz>

* opt-functions.awk: Add validation of value of Init.
* optc-gen.awk: Pass new argument.

From-SVN: r250625

sparc.c (sparc_option_override): Set MASK_FSMULD flag earlier and only if MASK_FPU is set.

* config/sparc/sparc.c (sparc_option_override): Set MASK_FSMULD flag
earlier and only if MASK_FPU is set.  Adjust formatting.

From-SVN: r250623

Fix indirect call optimization done by autoFDO.

2017-07-27  Martin Liska  <mliska@suse.cz>

* auto-profile.c (autofdo_source_profile::update_inlined_ind_target):
Fix wrong condition.

From-SVN: r250622

Initialize counters in autoFDO to zero, not to uninitialized.

2017-07-27  Martin Liska  <mliska@suse.cz>

* auto-profile.c (afdo_annotate_cfg): Assign zero counts to
BBs and edges seen by autoFDO.

From-SVN: r250621

re PR middle-end/81502 (In some cases the data is moved to memory unnecessarily [partial regression])

2017-07-27  Richard Biener  <rguenther@suse.de>

PR tree-optimization/81502
* tree-ssa.c (non_rewritable_lvalue_p): Handle BIT_INSERT_EXPR
with incompatible but same sized type.
(execute_update_addresses_taken): Likewise.

* gcc.target/i386/vect-insert-1.c: New testcase.

From-SVN: r250620

[Patch (preapproved)] Guard Copy Header pass on

While answering a user question on the equivalence of
-ftree-loop-vectorize + -ftree-slp-vectorize and -ftree-vectorize I
spotted one case which broke the equivalence. pass_ch::process_loop_p
was guarded on flag_tree_vectorize, meaning you would get it for
-ftree-vectorize, but not for -ftree-loop-vectorize/-ftree-slp-vectorize.

This patch fixes that, getting rid of the only use of flag_tree_vectorize
in the code base.

gcc/

* tree-ssa-loop-ch.c (pass_ch::process_loop_p): Guard on
flag_tree_loop_vectorize rather than flag_tree_vectorize.

From-SVN: r250619

S/390: Fix PR81534

The HI/QI atomic_fetch_<atomic><mode>" expander accepted symbolic
references and emitted CAS patterns whose insn predicates rejected them.

Fixed by allowing symbolic references there as well.  Reload will get
rid of them due to the constraint letter.

Regression tested on s390x.

gcc/ChangeLog:

2017-07-27  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>

PR target/81534
* config/s390/s390.md ("*atomic_compare_and_swap<mode>_1")
("*atomic_compare_and_swapdi_2", "*atomic_compare_and_swapsi_3"):
Change s_operand to memory_operand.

gcc/testsuite/ChangeLog:

2017-07-27  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>

PR target/81534
* gcc.target/s390/pr81534.c: New test.

From-SVN: r250617

[rs6000] Avoid rotates of floating-point modes

The little-endian VSX code uses rotates to swap the two 64-bit halves of
128-bit scalar modes.  This is fine for TImode and V1TImode, but it
isn't really valid to use RTL rotates on floating-point modes like
KFmode and TFmode, and doing that triggered an assert added by the
SVE series.  This patch uses bit-casts to V1TImode instead.

2017-07-27  Richard Sandiford  <richard.sandiford@linaro.org>

gcc/
* config/rs6000/rs6000-protos.h (rs6000_emit_le_vsx_permute): Declare.
* config/rs6000/rs6000.c (rs6000_gen_le_vsx_permute): Replace with...
(rs6000_emit_le_vsx_permute): ...this.  Take the destination as input.
Emit instructions rather than returning an expression.  Handle TFmode
and KFmode by casting to TImode.
(rs6000_emit_le_vsx_load): Update to use rs6000_emit_le_vsx_permute.
(rs6000_emit_le_vsx_store): Likewise.
* config/rs6000/vsx.md (VSX_TI): New iterator.
(*vsx_le_permute_<mode>): Use it instead of VSX_LE_128.
(*vsx_le_undo_permute_<mode>): Likewise.
(*vsx_le_perm_load_<mode>): Use rs6000_emit_le_vsx_permute to
emit the split sequence.
(*vsx_le_perm_store_<mode>): Likewise.

From-SVN: r250615

re PR tree-optimization/81555 (Wrong code at -O1)

PR tree-optimization/81555
PR tree-optimization/81556
* tree-ssa-reassoc.c (rewrite_expr_tree): Add NEXT_CHANGED argument,
if true, force CHANGED for the recursive invocation.
(reassociate_bb): Remember original length of ops array, pass
len != orig_len as NEXT_CHANGED in rewrite_expr_tree call.

* gcc.c-torture/execute/pr81555.c: New test.
* gcc.c-torture/execute/pr81556.c: New test.

From-SVN: r250609

attribs.c (decl_attributes): Imply noinline, noclone and no_icf attributes for noipa attribute.

* attribs.c (decl_attributes): Imply noinline, noclone and no_icf
attributes for noipa attribute.  For naked attribute use
lookup_attribute first before lookup_attribute_spec.
* final.c (rest_of_handle_final): Disable IPA RA for functions with
noipa attribute.
* ipa-visibility.c (non_local_p): Fix comment typos.  Return true
for functions with noipa attribute.
(cgraph_externally_visible_p): Return true for functions with noipa
attribute.
* cgraph.c (cgraph_node::get_availability): Return AVAIL_INTERPOSABLE
for functions with noipa attribute.
* doc/extend.texi: Document noipa function attribute.
* tree-ssa-structalias.c (refered_from_nonlocal_fn): Set *nonlocal_p
also for functions with noipa attribute.
(ipa_pta_execute): Set nonlocal_p also for nodes with noipa attribute.
c-family/
* c-attribs.c (c_common_attribute_table): Add noipa attribute.
(handle_noipa_attribute): New function.
testsuite/
* gcc.dg/attr-noipa.c: New test.
* gcc.dg/ipa/ipa-pta-18.c: New test.
* gcc.dg/ipa/ipa-sra-11.c: New test.

From-SVN: r250607

Daily bump.

From-SVN: r250601

aarch64.c (thunderx_vector_cost): Decrease cost of vec_unalign_load_cost and vec_unalign_store_cost.

2017-07-26  Andrew Pinski  <apinski@cavium.com>

        * config/aarch64/aarch64.c (thunderx_vector_cost): Decrease cost of
        vec_unalign_load_cost and vec_unalign_store_cost.

From-SVN: r250597

rs6000-cpus.def (ISA_2_7_MASKS_SERVER): Delete -mvsx-small-integer option.

[gcc]
2017-07-26  Michael Meissner  <meissner@linux.vnet.ibm.com>

* config/rs6000/rs6000-cpus.def (ISA_2_7_MASKS_SERVER): Delete
-mvsx-small-integer option.
(ISA_3_0_MASKS_IEEE): Likewise.
(OTHER_VSX_VECTOR_MASKS): Likewise.
(POWERPC_MASKS): Likewise.
* config/rs6000/rs6000.opt (-mvsx-small-integer): Likewise.
* config/rs6000/rs6000.c (rs6000_hard_regno_mode_ok): Simplify
code, only testing for DImode being allowed in non-VSX floating
point registers.
(rs6000_init_hard_regno_mode_ok): Change TARGET_VSX_SMALL_INTEGER
to TARGET_P8_VECTOR test.  Remove redundant VSX test inside of
another VSX test.
(rs6000_option_override_internal): Delete -mvsx-small-integer.
(rs6000_expand_vector_set): Change TARGET_VSX_SMALL_INTEGER to
TARGET_P8_VECTOR test.
(rs6000_secondary_reload_simple_move): Likewise.
(rs6000_preferred_reload_class): Delete TARGET_VSX_SMALL_INTEGER,
since TARGET_P9_VECTOR was already tested.
(rs6000_opt_masks): Remove -mvsx-small-integer.
* config/rs6000/vsx.md (vsx_extract_<mode>): Delete
TARGET_VSX_SMALL_INTEGER, since a test for TARGET_P9_VECTOR was
used.
(vsx_extract_<mode>_p9): Delete TARGET_VSX_SMALL_INTEGER, since a
test for TARGET_VEXTRACTUB was used, and that uses
TARGET_P9_VECTOR.
(p9 extract splitter): Likewise.
(vsx_extract_<mode>_di_p9): Likewise.
(vsx_extract_<mode>_store_p9): Likewise.
(vsx_extract_si): Delete TARGET_VSX_SMALL_INTEGER, since a test
for TARGET_P9_VECTOR was used.  Delete code that is now dead with
the elimination of TARGET_VSX_SMALL_INTEGER.
(vsx_extract_<mode>_p8): Likewise.
(vsx_ext_<VSX_EXTRACT_I:VS_scalar>_fl_<FL_CONV:mode>): Likewise.
(vsx_ext_<VSX_EXTRACT_I:VS_scalar>_ufl_<FL_CONV:mode>): Likewise.
(vsx_set_<mode>_p9): Likewise.
(vsx_set_v4sf_p9): Likewise.
(vsx_set_v4sf_p9_zero): Likewise.
(vsx_insert_extract_v4sf_p9): Likewise.
(vsx_insert_extract_v4sf_p9_2): Likewise.
* config/rs6000/rs6000.md (sign extend splitter): Change
TARGET_VSX_SMALL_INTEGER to TARGET_P8_VECTOR test.
(floatsi<mode>2_lfiwax_mem): Likewise.
(floatunssi<mode>2_lfiwzx_mem): Likewise.
(float<QHI:mode><FP_ISA3:mode>2): Delete TARGET_VSX_SMALL_INTEGER,
since a test for TARGET_P9_VECTOR was used.
(float<QHI:mode><FP_ISA3:mode>2_internal): Likewise.
(floatuns<QHI:mode><FP_ISA3:mode>2): Likewise.
(floatuns<QHI:mode><FP_ISA3:mode>2_internal): Likewise.
(fix_trunc<mode>si2): Change TARGET_VSX_SMALL_INTEGER to
TARGET_P8_VECTOR test.
(fix_trunc<mode>si2_stfiwx): Likewise.
(fix_trunc<mode>si2_internal): Likewise.
(fix_trunc<SFDF:mode><QHI:mode>2): Delete
TARGET_VSX_SMALL_INTEGER, since a test for TARGET_P9_VECTOR was
used.
(fix_trunc<SFDF:mode><QHI:mode>2_internal): Likewise.
(fixuns_trunc<mode>si2): Change TARGET_VSX_SMALL_INTEGER to
TARGET_P8_VECTOR test.
(fixuns_trunc<mode>si2_stfiwx): Likewise.
(fixuns_trunc<SFDF:mode><QHI:mode>2): Delete
TARGET_VSX_SMALL_INTEGER, since a test for TARGET_P9_VECTOR was
used.
(fixuns_trunc<SFDF:mode><QHI:mode>2_internal): Likewise.
(fctiw<u>z_<mode>_smallint): Delete TARGET_VSX_SMALL_INTEGER,
since a test for TARGET_P9_VECTOR was used.
(splitter for loading small constants): Likewise.

[gcc/testsuite]
2017-07-25  Michael Meissner  <meissner@linux.vnet.ibm.com>

* gcc.target/powerpc/vsx-himode.c: Delete -mvsx-small-integer
option.
* gcc.target/powerpc/vsx-himode2.c: Likewise.
* gcc.target/powerpc/vsx-himode3.c: Likewise.
* gcc.target/powerpc/vsx-qimode.c: Likewise.
* gcc.target/powerpc/vsx-qimode2.c: Likewise.
* gcc.target/powerpc/vsx-qimode3.c: Likewise.
* gcc.target/powerpc/vsx-simode.c: Likewise.
* gcc.target/powerpc/vsx-simode2.c: Likewise.
* gcc.target/powerpc/vsx-simode3.c: Likewise.

From-SVN: r250595

PR libstdc++/53984 fix failing test

PR libstdc++/53984
* testsuite/27_io/basic_fstream/53984.cc: Fix test.

From-SVN: r250594

re PR go/81548 ("make distclean" does not clean all of gotools/)

PR go/81548
* Makefile.am (MOSTLYCLEANFILES): Add *.sent.
* Makefile.in: Rebuild.

From-SVN: r250593

aarch64.c (thunderx_vector_cost): Fix vec_fp_stmt_cost.

2017-07-26  Andrew Pinski  <apinski@cavium.com>

        * config/aarch64/aarch64.c (thunderx_vector_cost): Fix
        vec_fp_stmt_cost.

From-SVN: r250592

re PR c++/71570 (ICE on invalid variable capture in cxx_incomplete_type_diagnostic, at cp/typeck2.c:55)

/cp
2017-07-26  Paolo Carlini  <paolo.carlini@oracle.com>

PR c++/71570
* lambda.c (add_capture): Early return if we cannot capture by
reference.

/testsuite
2017-07-26  Paolo Carlini  <paolo.carlini@oracle.com>

PR c++/71570
* g++.dg/cpp0x/lambda/lambda-ice17.C: New.

From-SVN: r250591

configure.ac: Check for XCOFF32/XCOFF64.

* configure.ac: Check for XCOFF32/XCOFF64.  Check for loadquery.
* filetype.awk: Separate AIX XCOFF32 and XCOFF64.
* xcoff.c: Add support for AIX XCOFF32 and XCOFF64 formats.
* configure, config.h.in: Regenerate.

From-SVN: r250590

runtime: handle Alpha GNU/Linux in getSiginfo

    Patch by Uros Bizjak.

    Reviewed-on: https://go-review.googlesource.com/51370

From-SVN: r250588

x86: Properly check saved register CFA offset

X86 prologue saves register at CFA offset.  Since its location on stack
is computed as CFA - its CFA_OFFSET, CFA_OFFSET points the end of the
saved register area on stack.  This patch updates sp_valid_at and
fp_valid_at to properly check saved register CFA offset.

gcc/

PR target/81563
* config/i386/i386.c (sp_valid_at): Properly check CFA offset.
(fp_valid_at): Likewise.

gcc/testsuite/

PR target/81563
* gcc.target/i386/pr81563.c: New test

From-SVN: r250587

[Patch AArch64 obvious] Unify address costs to generic_addrcost_table

The special case address cost tables for Cortex-A57 and qdf24xx are no
different from the generic address cost table. We should just use the
address cost table directly. If this changes in future, a core is welcome
to add new address cost tables.

gcc/

* config/aarch64/aarch64.c (cortexa57_addrcost_table): Remove.
(qdf24xx_addrcost_table): Likewise.
(cortexa57_tunings): Update to use generic_branch_cost.
(cortexa72_tunings): Likewise.
(cortexa73_tunings): Likewise.
(qdf24xx_tunings): Likewise.

From-SVN: r250585

P0702R1 - List deduction of vector.

* pt.c (do_class_deduction): Special-case deduction from a single
element of related type.

From-SVN: r250584

PR c++/67054 - Inherited ctor with non-default-constructible members

        PR c++/67054 - Inherited ctor with non-default-constructible members
        * method.c (walk_field_subobs) Consider member initializers (NSDMIs)
when deducing an inheriting constructor.

From-SVN: r250583

[Patch AArch64 Obvious] Unify branch costs to generic_branch_cost

All the cores in AArch64 use the pair {1, 3} for their branch costs. As
that is covered by generic_branch_cost, we can just use that directly and
save the tiny amount of redundant code. If in future any core wants to
modify this, they can always add a special-case branch-cost back.

gcc/

* config/aarch64/aarch64.c (cortexa57_branch_cost): Remove.
(thunderx2t99_branch_cost): Likewise.
(cortexa35_tunings): Update to use generic_branch_cost.
(cortexa53_tunings): Likewise.
(cortexa57_tunings): Likewise.
(cortexa72_tunings): Likewise.
(cortexa73_tunings): Likewise.
(thunderx2t99_tunings): Likewise.

From-SVN: r250582

53984.cc: Fix typo in dg-require directive.

2017-07-26  Paolo Carlini  <paolo.carlini@oracle.com>

* testsuite/27_io/basic_fstream/53984.cc: Fix typo in dg-require
directive.

From-SVN: r250575

[SPARC] Add -mfsmuld option

Add the -mfsmuld option to control the generation of the FsMULd
instruction.  In general, this instruction is available in architecture
version V8 and V9 CPUs with FPU.  Some CPUs of this category do not
support this instruction properly, e.g. AT697E, AT697F and UT699.  Some
CPUs of this category do not implement it in hardware, e.g. LEON3/4 with
GRFPU-lite.

gcc/
* config/sparc/sparc.c (dump_target_flag_bits): Dump MASK_FSMULD.
(sparc_option_override): Honour MASK_FSMULD.
* config/sparc/sparc.h (MASK_FEATURES): Add MASK_FSMULD.
* config/sparc/sparc.md (muldf3_extend): Use TARGET_FSMULD.
* config/sparc/sparc.opt (mfsmuld): New option.
* doc/invoke.texi (mfsmuld): Document option.

From-SVN: r250570

Disable pr79041-2.c with -mabi=ilp32.

    gcc/testsuite/
PR target/79041
* gcc.target/aarch64/pr79041-2.c: Don't run in ILP32.

From-SVN: r250567

re PR middle-end/70992 (Infinite recursion between fold_build2_stat_loc and fold_binary_loc w/ -fwrapv)

PR middle-end/70992
* tree.c (build2_stat): Don't set TREE_CONSTANT on divisions by zero.

* gcc.dg/overflow-warn-1.c: Adjust dg-error.
* gcc.dg/overflow-warn-2.c: Likewise.
* gcc.dg/overflow-warn-3.c: Likewise.
* gcc.dg/overflow-warn-4.c: Likewise.
* gcc.dg/torture/pr70992-2.c: New test.
* gcc.dg/torture/pr70992.c: New test.

From-SVN: r250566

gimple-match-head.c (do_valueize): Return OP if valueize returns NULL_TREE.

2017-07-26  Richard Biener  <rguenther@suse.de>

* gimple-match-head.c (do_valueize): Return OP if valueize
returns NULL_TREE.
(get_def): New helper to get at the def stmt of a SSA name
if valueize allows.
* genmatch.c (dt_node::gen_kids_1): Use get_def instead of
do_valueize to get at the def stmt.
(dt_operand::gen_gimple_expr): Simplify do_valueize calls.

* gcc/testsuite/gcc.dg/pr70920-2.c: Adjust for transform already
happening in ccp1.
* gcc/testsuite/gcc.dg/pr70920-4.c: Likewise.

From-SVN: r250565

Fix PR46932: Block auto increment on frame pointer

Block auto increment on frame pointer references.  This is never
beneficial since the SFP expands into SP+C or FP+C during register
allocation.  The generated code for the testcase is now as expected:

str x30, [sp, -32]!
strb w0, [sp, 31]
add x0, sp, 31
bl foo3
ldr x30, [sp], 32
ret

    gcc/
PR middle-end/46932
* auto-inc-dec.c (parse_add_or_inc): Block autoinc on sfp.

    gcc/testsuite/
PR middle-end/46932
* gcc.dg/pr46932.c: New testcase.

From-SVN: r250564

Move non-local goto expansion after parm_birth_insn (PR sanitize/81186).

2017-07-26  Martin Liska  <mliska@suse.cz>

PR sanitize/81186
* function.c (expand_function_start): Make expansion of
nonlocal_goto_save_area after parm_birth_insn.
2017-07-26  Martin Liska  <mliska@suse.cz>

PR sanitize/81186
* gcc.dg/asan/pr81186.c: New test.

From-SVN: r250561

[SPARC] Drop superfluous MASK_FPU enable

All TARGET_DEFAULT defines set MASK_FPU.  There is no need to set it in
some CPU target flags enable.

gcc/
* config/sparc/sparc.c (sparc_option_override): Remove MASK_FPU
from all CPU target flags enable members.

From-SVN: r250557

genmatch.c (dt_simplify::gen): Make iterator vars const.

2017-07-25  Richard Biener  <rguenther@suse.de>

* genmatch.c (dt_simplify::gen): Make iterator vars const.
(decision_tree::gen): Make 'type' const.
(write_predicate): Likewise.

From-SVN: r250556