rs6000.c (rs6000_gimple_fold_builtin): Add handling for early expansion of vec_min and vec_max builtins.

[gcc]

2017-06-02  Will Schmidt  <will_schmidt@vnet.ibm.com>
* config/rs6000/rs6000.c (rs6000_gimple_fold_builtin): Add handling
for early expansion of vec_min and vec_max builtins.
(builtin_function_type): Add min/max unsigned variants to those
identified as having unsigned arguments.

[gcc/testsuite]

2017-06-02  Will Schmidt  <will_schmidt@vnet.ibm.com>

*  testsuite/gcc.target/powerpc/fold-vec-minmax-char.c: New.
*  testsuite/gcc.target/powerpc/fold-vec-minmax-floatdouble.c: New.
*  testsuite/gcc.target/powerpc/fold-vec-minmax-int.c: New.
*  testsuite/gcc.target/powerpc/fold-vec-minmax-longlong.c: New.
*  testsuite/gcc.target/powerpc/fold-vec-minmax-short.c: New.

From-SVN: r248834

vx-common.h (DWARF_UNWIND_INFO): Switch #define to 1.

2017-06-02  Olivier Hainque  <hainque@adacore.com>

* config/vx-common.h (DWARF_UNWIND_INFO): Switch #define to 1.

From-SVN: r248833

[AArch64] Emit tighter strong atomic compare-exchange loop when comparing against zero

* config/aarch64/aarch64.c (aarch64_split_compare_and_swap):
Emit CBNZ inside loop when doing a strong exchange and comparing
against zero.  Generate the CC flags after the loop.

* gcc.target/aarch64/atomic_cmp_exchange_zero_strong_1.c: New test.

From-SVN: r248832

dwarf2out.c (DWARF_INITIAL_LENGTH_SIZE_STR): New.

        * dwarf2out.c (DWARF_INITIAL_LENGTH_SIZE_STR): New.
        (dl_section_ref): New.
        (dwarf2out_finish): Copy debug_line_section_label to dl_section_ref.
        On AIX, append an expression to subtract the size of the
        section length to dl_section_ref.

From-SVN: r248831

rs6000.c (rs6000_gimple_fold_builtin): Add handling for early expansion of vector absolute builtins.

[gcc]

2017-05-31  Will Schmidt  <will_schmidt@vnet.ibm.com>

* config/rs6000/rs6000.c (rs6000_gimple_fold_builtin): Add handling
for early expansion of vector absolute builtins.

[gcc/testsuite]

2017-05-31  Will Schmidt  <will_schmidt@vnet.ibm.com>

* gcc.target/powerpc/fold-vec-abs-char.c: New.
* gcc.target/powerpc/fold-vec-abs-floatdouble.c: New.
* gcc.target/powerpc/fold-vec-abs-int.c: New.
* gcc.target/powerpc/fold-vec-abs-longlong.c: New.
* gcc.target/powerpc/fold-vec-abs-short.c: New.
* gcc.target/powerpc/fold-vec-abs-char-fwrapv.c: New.
* gcc.target/powerpc/fold-vec-abs-int-fwrapv.c: New.
* gcc.target/powerpc/fold-vec-abs-longlong-fwrapv.c: New.
* gcc.target/powerpc/fold-vec-abs-short-fwrapv.c: New.

From-SVN: r248830

Remove lang_type_ptrmem.

Changlogs:
cp/
Remove lang_type_ptrmem.
* cp-tree.h (lang_type_header): Remove is_lang_type_class. Move
into ...
(lang_type_class): ... this.  Reorder bitfields.  Rename to ...
(lang_type): ... this.  Delete old definition.
(lang_type_ptrmem): Delete.
(LANG_TYPE_CLASS_CHECK): Simply get TYPE_LANG_SPECIFIC.  Adjust
uses.
(LANG_TYPE_PTRMEM_CHECK): Delete.
(TYPE_GET_PTRMEMFUNC_TYPE, TYPE_SET_PTRMEMFUNC_TYPE): Delete.
(TYPE_PTRMEMFUNC_TYPE): New.  Use TYPE_LANG_SLOT_1.
* decl.c (build_ptrmemfunc_type): Adjust.
* lex.c (copy_lang_type): Remove lang_type_ptrmem handling.
(maybe_add_lang_type_raw): Don't set u.c.h.is_lang_type_class.

objcp/
* objcp-decl.h (SIZEOF_OBJC_TYPE_LANG_SPECIFIC): Use lang_type.
(ALLOC_OBJC_TYPE_LANG_SPECIFIC): Use it.  Don't set
u.c.h.is_lang_type_class.

From-SVN: r248827

Remove lang_type_ptrmem.

cp/
Remove lang_type_ptrmem.
* cp-tree.h (lang_type_header): Remove is_lang_type_class. Move
into ...
(lang_type_class): ... this.  Reorder bitfields.  Rename to ...
(lang_type): ... this.  Delete old definition.
(lang_type_ptrmem): Delete.
(LANG_TYPE_CLASS_CHECK): Simply get TYPE_LANG_SPECIFIC.  Adjust
uses.
(LANG_TYPE_PTRMEM_CHECK): Delete.
(TYPE_GET_PTRMEMFUNC_TYPE, TYPE_SET_PTRMEMFUNC_TYPE): Delete.
(TYPE_PTRMEMFUNC_TYPE): New.  Use TYPE_LANG_SLOT_1.
* decl.c (build_ptrmemfunc_type): Adjust.
* lex.c (copy_lang_type): Remove lang_type_ptrmem handling.
(maybe_add_lang_type_raw): Don't set u.c.h.is_lang_type_class.

objcp/
* objcp-decl.h (SIZEOF_OBJC_TYPE_LANG_SPECIFIC): Use lang_type.
(ALLOC_OBJC_TYPE_LANG_SPECIFIC): Use it.  Don't set
u.c.h.is_lang_type_class.

From-SVN: r248826

tree-vect-slp.c (vect_detect_hybrid_slp_2): Match up what we consider a relevant use stmt with vect_detect_hybrid_slp_stmts.

2017-06-02  Richard Biener  <rguenther@suse.de>

* tree-vect-slp.c (vect_detect_hybrid_slp_2): Match up
what we consider a relevant use stmt with vect_detect_hybrid_slp_stmts.

From-SVN: r248825

re PR tree-optimization/80948 (test case gcc.dg/torture/pr68017.c fails with ICE starting with r248771)

2017-06-02  Richard Biener  <rguenther@suse.de>

PR tree-optimization/80948
* tree-tailcall.c (find_tail_calls): Track stmts to move in
stmt order as well.

From-SVN: r248824

Remove libsupc++ makefile targets setting -std=gnu++11 or -std=gnu++14

* libsupc++/Makefile.am: Remove custom targets for files that need to
be compiled as C++11 or C++14.
* libsupc++/Makefile.in: Regenerate.
* libsupc++/del_ops.cc: Use pragma to disable -Wsized-deallocation
warnings.
* libsupc++/del_opvs.cc: Likewise.

From-SVN: r248823

name-lookup.h (cp_binding_level): Lose namespaces field.

* name-lookup.h (cp_binding_level): Lose namespaces field.
* name-lookup.c (add_decl_to_level): Chain namespaces on the names
list.
(suggest_alternatives_for): Adjust for namespace list.  Do
breadth-first search.
* decl2.c (collect_source_refs): Namespaces are on the regulr
list.
(collect_ada_namespace): Likewise.

* g++.dg/pr45330.C: Adjust.  Check breadth-firstness.

From-SVN: r248821

tree-vect-loop.c (vect_analyze_loop_operations): Not relevant PHIs are ok.

2017-06-02  Richard Biener  <rguenther@suse.de>

* tree-vect-loop.c (vect_analyze_loop_operations): Not relevant
PHIs are ok.
* tree-vect-stmts.c (process_use): Do not mark backedge defs
for inductions as relevant.

From-SVN: r248820

tree-vect-loop.c (get_initial_def_for_induction): Inline into ...

2017-06-02  Richard Biener  <rguenther@suse.de>

* tree-vect-loop.c (get_initial_def_for_induction): Inline into ...
(vectorizable_induction): ... this.  Remove dead code.

From-SVN: r248819

* builtins. (expand_builtin_alloca): Remove second parameter and
infer its value from the first parameter instead.
(expand_builtin) <BUILT_IN_ALLOCA>: Adjust call to above.

From-SVN: r248818

re PR libstdc++/80721 (Sorting/Merging of free EH-emergency buffer may wrong or uncomplete)

2017-06-02  Richard Biener  <rguenther@suse.de>
Markus Eisenmann  <meisenmann.lba@fh-salzburg.ac.at>

PR libstdc++/80721
* libsupc++/eh_alloc.cc (pool::free): Keep list properly
sorted and add missing freelist item merging cases.

Co-Authored-By: Markus Eisenmann <meisenmann.lba@fh-salzburg.ac.at>
From-SVN: r248816

re PR rtl-optimization/80903 (ICE: internal consistency failure (error: invalid rtl sharing found in the insn))

PR rtl-optimization/80903
* loop-doloop.c (add_test): Unshare sequence.

* gcc.dg/pr80903.c: New test.

From-SVN: r248815

re PR fortran/80918 (Assumed size whole array rejected in depend clause)

PR fortran/80918
* openmp.c (resolve_omp_clauses): Fix a typo.

* gfortran.dg/gomp/pr80918.f90: New test.

From-SVN: r248812

invoke.texi: Document the -Wsizeof-pointer-div warning.

gcc:
2017-06-02  Bernd Edlinger  <bernd.edlinger@hotmail.de>

        * doc/invoke.texi: Document the -Wsizeof-pointer-div warning.

gcc/c-family:
2017-06-02  Bernd Edlinger  <bernd.edlinger@hotmail.de>

        * c.opt (Wsizeof-pointer-div): New warning option.

gcc/c:
2017-06-02  Bernd Edlinger  <bernd.edlinger@hotmail.de>

        * c-parser.c (c_parser_binary_expression): Implement the
        -Wsizeof_pointer_div warning.
        (c_parser_postfix_expression): Allow SIZEOF_EXPR as expr.original_code
        from a parenthesized expression.
        (c_parser_expr_list): Use c_last_sizeof_loc.
        * c-tree.h (c_last_sizeof_loc): New external.
        * c-typeck.c (c_last_sizeof_loc): New variable.
        (c_expr_sizeof_expr, c_expr_sizeof_type): Assign c_last_sizeof_loc.

gcc/cp:
2017-06-02  Bernd Edlinger  <bernd.edlinger@hotmail.de>

        * typeck.c (cp_build_binary_op): Implement the -Wsizeof_pointer_div
        warning.

gcc/testsuite:
2017-06-02  Bernd Edlinger  <bernd.edlinger@hotmail.de>

        * c-c++-common/Wsizeof-pointer-div.c: New test.
        * gcc.dg/Wsizeof-pointer-memaccess1.c: Add test cases with parens.
        * gcc.dg/torture/Wsizeof-pointer-memaccess1.c: Likewise.
        * gcc.target/i386/sse-init-v4hi-1.c: Fix test case.
        * gcc.target/i386/sse-init-v4sf-1.c: Likewise.
        * gcc.target/i386/sse-set-ps-1.c: Likewise.
        * gcc.target/i386/sse2-init-v16qi-1.c: Likewise.
        * gcc.target/i386/sse2-init-v2di-1.c: Likewise.
        * gcc.target/i386/sse2-init-v4si-1.c: Likewise.
        * gcc.target/i386/sse2-init-v8hi-1.c: Likewise.
        * gcc.target/i386/sse2-set-epi32-1.c: Likewise.
        * gcc.target/i386/sse2-set-epi64x-1.c: Likewise.
        * gcc.target/i386/sse4_1-init-v16qi-1.c: Likewise.
        * gcc.target/i386/sse4_1-init-v2di-1.c: Likewise.
        * gcc.target/i386/sse4_1-init-v4sf-1.c: Likewise.
        * gcc.target/i386/sse4_1-init-v4si-1.c: Likewise.
        * gcc.target/i386/sse4_1-set-epi32-1.c: Likewise.
        * gcc.target/i386/sse4_1-set-epi64x-1.c: Likewise.
        * gcc.target/i386/sse4_1-set-ps-1.c: Likewise.
        * libgomp.c/pr39591-2.c: Likewise.
        * libgomp.c/pr39591-3.c: Likewise.

From-SVN: r248811

Daily bump.

From-SVN: r248810

fold-vec-logical-ors-longlong.c: Update the target to power8-vector.

[gcc/testsuite]

2017-05-26  Will Schmidt  <will_schmidt@vnet.ibm.com>

* gcc.target/powerpc/fold-vec-logical-ors-longlong.c:
Update the target to power8-vector.

From-SVN: r248805

* es.po: Update.

From-SVN: r248802

testsuite: ensure GCC_COLORS is unset

gcc/testsuite/ChangeLog:
* lib/gcc-dg.exp: Ensure GCC_COLORS is unset.

From-SVN: r248800

i386.c (x86_64_ms_sysv_extra_clobbered_registers): Make static.

2017-06-01  Bernd Edlinger  <bernd.edlinger@hotmail.de>

        * config/i386/i386.c (x86_64_ms_sysv_extra_clobbered_registers): Make
        static.
        (xlogue_layout::get_stack_space_used, xlogue_layout::s_instances,
        xlogue_layout::get_instance, logue_layout::xlogue_layout,
        sp_valid_at, fp_valid_at, choose_basereg): Formatting.
        (xlogue_layout::get_stub_rtx): Make static.
        (xlogue_layout::get_stub_name): Avoid const-cast, make static.
        (xlogue_layout::compute_stub_managed_regs): Rename to...
        (xlogue_layout::count_stub_managed_regs): ...this.
        (xlogue_layout::is_stub_managed_reg): New function.
        (xlogue_layout::m_stub_names): Rename to...
        (xlogue_layout::s_stub_names): ...this, make static.
        (xlogue_layout::STUB_INDEX_OFFSET, xlogue_layout::MIN_REGS,
        xlogue_layout::MAX_REGS, xlogue_layout::MAX_EXTRA_REGS,
        xlogue_layout::VARIANT_COUNT, xlogue_layout::STUB_NAME_MAX_LEN,
        xlogue_layout::s_stub_names): Instantiate statics.
        (stub_managed_regs): Remove.
        (ix86_save_reg): Use xlogue_layout::compute_stub_managed_regs.
        (disable_call_ms2sysv_xlogues): Rename to...
        (warn_once_call_ms2sysv_xlogues): ...this, and warn only once.
        (ix86_initial_elimination_offset, ix86_expand_call): Fix call_ms2sysv
        warning logic.
        (ix86_static_chain): Make sure that ix86_static_chain_on_stack can't
        change after reload_completed.
        (ix86_can_use_return_insn_p): Use the ix86_frame data structure
        directly.
        (ix86_expand_prologue): Likewise.
        (ix86_expand_epilogue): Likewise.
        (ix86_expand_split_stack_prologue): Likewise.
        (ix86_compute_frame_layout): Remove frame parameter ...
        (TARGET_COMPUTE_FRAME_LAYOUT): ... and export it as a target hook.
        (ix86_finalize_stack_realign_flags): Call ix86_compute_frame_layout
        only if necessary.
        (ix86_init_machine_status): Don't set use_fast_prologue_epilogue_nregs.
        (ix86_frame): Move from here ...
        * config/i386/i386.h (ix86_frame): ... to here.
        (machine_function): Remove use_fast_prologue_epilogue_nregs, cache the
        complete ix86_frame data structure instead.  Remove some_ld_name.

From-SVN: r248798

DWARF: add DW_AT_location for global decls with DECL_VALUE_EXPR

In GNAT, we materialize renamings that cannot be described in standard
DWARF as synthetic variables that describe how to fetch the renamed
object.  Look for "___XR" in gcc/ada/exp_dbug.ads for more details about
this convention.

In order to have a location for these variables in the debug info (GDB
requires it not to discard the variable) but also to avoid allocating
runtime space for them, we make these variable hold a DECL_VALUE_EXPR
tree.  However, since GCC 7, the DWARF back-end no longer generates a
DW_AT_location attribute for those.  This patch is an attempt to restore
this attribute.

gcc/

* dwarf2out.c (dwarf2out_late_global_decl): Add locations for
symbols that hold a DECL_VALUE_EXPR.

gcc/testsuite/

* debug12.adb, debug12.ads: New testcase.

From-SVN: r248792

Fix cross compilation to Solaris

(GLIBCXX_CROSSCONFIG): Handle *-solaris* like *-linux* etc.
Remove *-solaris* section.
* configure: Regenerate.

From-SVN: r248791

[PR 80898] Propagate grp_write from disqualified SRA candidates

2017-06-01  Martin Jambor  <mjambor@suse.cz>

PR tree-optimization/80898
* tree-sra.c (process_subtree_disqualification): Removed.
(disqualify_candidate): Do not acll
process_subtree_disqualification.
(subtree_mark_written_and_enqueue): New function.
(propagate_all_subaccesses): Set grp_write of LHS subtree if the
RHS has been disqualified and re-queue LHS if necessary.  Apart
from that, ignore disqualified RHS.

testsuite/
* gcc.dg/tree-ssa/pr80898.c: New test.
* gcc.dg/tree-ssa/pr80898-2.c: Likewise.

From-SVN: r248790

S/390: Don't fetch the return address early with ooo

We used to load the return address slot some time in advance.  This
helped on older machines to resolve the data dependencies in time.
However, it is pointless on out of order CPUs.  Disabled with that
patch.

gcc/ChangeLog:

2017-06-01  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>

* config/s390/s390.c (s390_emit_epilogue): Disable early return
address fetch for z10 or later.

From-SVN: r248789

re PR c++/80812 (ICE: in build_value_init_noctor, at cp/init.c:483)

PR c++/80812

cp/

PR c++/80812
* method.c (constructible_expr): Strip array types before calling
build_value_init.

libstdc++/

PR c++/80812
* testsuite/20_util/is_constructible/80812.cc: New.

From-SVN: r248788

re PR ada/80921 (cross compiling fails to build Ada shared libraries)

PR ada/80921
* configure.ac (default_gnatlib_target): Remove bogus condition.
(have_getipinfo): Tweak.
* configure: Regenerate.

From-SVN: r248785

re PR c++/80896 ([[nodiscard]] is ignored for functions returning references)

/cp
2017-06-01  Paolo Carlini  <paolo.carlini@oracle.com>

PR c++/80896
* cvt.c (convert_to_void): Possibly call maybe_warn_nodiscard
for case INDIRECT_REF too in the main switch.

/testsuite
2017-06-01  Paolo Carlini  <paolo.carlini@oracle.com>

PR c++/80896
* g++.dg/cpp1z/nodiscard5.C: New.

From-SVN: r248784

[ARC] Fix tst_movb pattern.

The tst_movb pattern is missing guarding when spitting.

gcc/
2017-06-01  Claudiu Zissulescu  <claziss@synopsys.com>

* config/arc/arc.md (tst_movb): Add guard when splitting.

testsuite/
2017-06-01  Claudiu Zissulescu  <claziss@synopsys.com>

* gcc.target/arc/pr9001195952.c: New test.

From-SVN: r248783

[ARC] Test against frame_pointer_needed in arc_can_eliminate.

arc_can_eliminate is using arc_frmae_pointer_required() which is wrong
as the frame_pointer_needed can be set on different conditions. Fix it
by calling arc_frame_pointer_needed().

gcc/
2017-06-01  Claudiu Zissulescu  <claziss@synopsys.com>

* config/arc/arc.c (arc_can_eliminate): Test against
arc_frame_pointer_needed.

From-SVN: r248782

[ARC] Prevent moving stores to the frame before the stack adjustment.

If the stack pointer is needed, emit a special barrier that will prevent
the scheduler from moving stores to the frame before the stack adjustment.

For example:

[snip]
mov_s fp,sp  ; frame pointer is set here
[snip]
st r1,[fp,-24] ; frame pointer is used here
[snip]
sub_s sp,sp,0x20 ; stack pointer adjusted

So we can easily see that any interrupt between the `st` and `sub`
instruction will lead to faulty code as the interrupt routine will use
a faulty sp register, and, potentially, overwriting the value stored
by 'st' instruction. Thus, adding a scheduler barrier will force the
compiler to emit the `sub` instruction before the store one.

2017-06-01  Claudiu Zissulescu  <claziss@synopsys.com>

* config/arc/arc.c (arc_expand_prologue): Emit a special barrier
to prevent store reordering.
* config/arc/arc.md (UNSPEC_ARC_STKTIE): Define.
(type): Add block type.
(stack_tie): Define special instruction to be used in
expand_prologue.

From-SVN: r248781

[ARC] Update (non)commutative_binary_comparison patterns.

gcc/
2017-06-01  Claudiu Zissulescu  <claziss@synopsys.com>

* config/arc/arc.md (commutative_binary_comparison): Remove 'I'
constraint. It is not valid for the pattern.
(noncommutative_binary_comparison): Likewise.

From-SVN: r248780

[ARC] Change predicate movv2hi to avoid scaled addresses.

2017-06-01  Claudiu Zissulescu  <claziss@synopsys.com>

* config/arc/simdext.md (movv2hi_insn): Change predicate to avoid
scaled addresses.

From-SVN: r248779

[ARC] Allow r30 to be used by the reg-alloc.

gcc/
2018-06-01  Claudiu Zissulescu  <claziss@synopsys.com>

* config/arc/arc.c (arc_conditional_register_usage): Allow r30 to
be used by the reg-alloc.

From-SVN: r248778

[ARC] Avoid use of hard registers before reg-alloc.

gcc/
2017-06-01  Claudiu Zissulescu  <claziss@synopsys.com>

* config/arc/arc.md (mulsi3): Avoid use of hard registers before
reg-alloc when having mul64 or mul32x16 instructions.
(mulsidi3): Likewise.
(umulsidi3): Likewise.
(mulsi32x16): New pattern.
(mulsi64): Likewise.
(mulsidi64): Likewise.
(umulsidi64): Likewise.
(MUL32x16_REG): Define.
(mul64_600): Use MUL32x16_REG.
(mac64_600): Likewise.
(umul64_600): Likewise.
(umac64_600): Likewise.

From-SVN: r248777

[ARC] Make mulsi for A700 pattern commutative.

gcc/
2017-06-01  Claudiu Zissulescu  <claziss@synopsys.com>

* config/arc/arc.md (mulsi3_700): Make it commutative.

From-SVN: r248776

Require label_values for some test-cases

2017-06-01  Tom de Vries  <tom@codesourcery.com>

* c-c++-common/pr43395.c: Add dg-require-effective-target label_values.
* gcc.c-torture/compile/asmgoto-1.c: Same.
* gcc.dg/20000707-1.c: Same.
* gcc.dg/pr38700.c: Same.
* gcc.dg/pr70169.c: Same.
* gcc.dg/pr80112.c: Same.
* gcc.dg/torture/pr51071-2.c: Same.
* gcc.dg/torture/pr51071.c: Same.
* gcc.dg/tree-ssa/alias-34.c: Same.

From-SVN: r248775

sparc.md (*zero_extendsidi2_insn_sp64): Set insn type for movstouw.

gcc/ChangeLog:
* config/sparc/sparc.md (*zero_extendsidi2_insn_sp64): Set insn
type for movstouw.
(*sign_extendsidi2_insn): Likewise for movstosw.

From-SVN: r248774

DWARF: for variants, produce unsigned discr. when debug type is unsigned

In Ada, the Character type is supposed to be unsigned.  However,
depending on the sign of C char types, GNAT can materialize it as a
signed type for code generation purposes.  When this is the case, GNAT
also attach a debug type to it so it is represented as an unsigned base
type in the debug information.

This change adapts record variant parts processing in the DWARF back-end
so that when the debug type of discriminant is unsigned while
discriminant values are signed themselves, we output unsigned
discriminant values in DWARF.

gcc/

* dwarf2out.c (get_discr_value): Call the get_debug_type hook on
the type of the input discriminant value.  Convert the
discriminant value of signedness vary.

gcc/testsuite/

* gnat.dg/debug11.adb: New testcase.

From-SVN: r248773

c.opt (Wcatch-value): New shortcut for Wcatch-value=1.

        * c.opt (Wcatch-value): New shortcut for Wcatch-value=1.
        (Wcatch-value=1): Enable by -Wall.

        * doc/invoke.texi (-Wcatch-value): Document new shortcut.
        Add to -Wall section.

From-SVN: r248772

re PR middle-end/66313 (Unsafe factorization of a*b+a*c)

2017-06-01  Richard Biener  <rguenther@suse.de>

PR middle-end/66313
* fold-const.c (fold_plusminus_mult_expr): If the factored
factor may be zero use a wrapping type for the inner operation.
* tree-tailcall.c (independent_of_stmt_p): Pass in to_move bitmap
and handle moved defs.
(process_assignment): Properly guard the unary op case.  Return a
tri-state indicating that moving the stmt before the call may allow
to continue.  Pass through to_move.
(find_tail_calls): Handle moving unrelated defs before
the call.

* c-c++-common/ubsan/pr66313.c: New testcase.
* gcc.dg/tree-ssa/loop-15.c: Adjust.

From-SVN: r248771

Daily bump.

From-SVN: r248770

Fix up dg options to resolve make check failures Bill Seurer found
when compiling --with-cpu=power6.

2017-05-31  Steven Munroe  <munroesj@gcc.gnu.org>

* gcc.target/powerpc/bmi2-pdep32-1.c: Add -mcpu=power7 to
        dg-options.  Change dg-require-effective-target powerpc_vsx_ok
        to vsx_hw.  Add dg-skip-if directive to disable this test if
-mcpu overridden.
* gcc.target/powerpc/bmi2-pdep64-1.c: Likewise.
* gcc.target/powerpc/bmi2-pext32-1.c: Likewise.
* gcc.target/powerpc/bmi2-pext64-1.c: Likewise.
* gcc.target/powerpc/bmi2-pext64-1a.c: Add -mcpu=power7
to dg-option.  Add dg-skip-if directive to disable this test
for darwin.  Add dg-skip-if directive to disable this test if
-mcpu overridden.

From-SVN: r248766

libgo: support for sparc64 GNU/Linux

    Fix lfstack code to work with sparc64 GNU/Linux address map.

    Force alignment of epollevent.  To make this work reliably, pass
    GOARCH explicitly to mkrsysinfo.sh.

    Patch by Vladimir Mezentsev.

    Reviewed-on: https://go-review.googlesource.com/44494

From-SVN: r248765

Fix changelog of previous commit, the correct version is:

PR target/80618
* config/rs6000/vector.md (*vector_uneq<mode>): Write the nor in the
splitter result in the canonical way.

From-SVN: r248764

rs6000: Don't write "nor" as (not (ior () ())) (PR80618)

The canonical RTL for "nor" is (and (not ()) (not ())), and that is
indeed what we use in boolccv2df3_internal1.  So, the splitter for
*vector_uneq<mode> should use that form, not (not (ior () ())), which
does not match any pattern.

PR target/80618
* config/rs6000/rs6000.md (*vector_uneq<mode>): Write the nor in the
splitter result in the canonical way.

From-SVN: r248763

i386.md (*zero_extendsidi2): Enable alternative (?r, *Yj) also for 32bit target.

* config/i386/i386.md (*zero_extendsidi2): Enable alternative (?r, *Yj)
also for 32bit target.  Update insn attributes.
(zero-extendsidi2 splitter): Allow all registers for operand 1.

From-SVN: r248757

* config/i386/avx512fintrin.h (_mm_mask_max_sd)
        (_mm_maskz_max_sd, _mm_mask_max_ss, _mm_maskz_max_ss)
        (_mm_mask_min_sd, _mm_maskz_min_sd, _mm_mask_min_ss)
        (_mm_maskz_min_ss): New intrinsics.

testsuite/ChangeLog:

* gcc.target/i386/avx512f-vmaxsd-1.c (_mm_mask_max_sd)
(_mm_maskz_max_sd): Test new intrinsics.
* gcc.target/i386/avx512f-vmaxsd-2.c (_mm_mask_max_sd)
(_mm_maskz_max_sd): Test new intrinsics.
* gcc.target/i386/avx512f-vmaxss-1.c (_mm_mask_max_ss)
(_mm_maskz_max_ss): Test new intrinsics.
* gcc.target/i386/avx512f-vmaxss-2.c (_mm_mask_max_ss)
(_mm_maskz_max_ss): Test new intrinsics.
* gcc.target/i386/avx512f-vminsd-1.c (_mm_mask_min_sd)
(_mm_maskz_min_sd): Test new intrinsics.
* gcc.target/i386/avx512f-vminsd-2.c (_mm_mask_min_sd)
(_mm_maskz_min_sd): Test new intrinsics.
* gcc.target/i386/avx512f-vminss-1.c (_mm_mask_min_ss)
(_mm_maskz_min_ss): Test new intrinsics.
* gcc.target/i386/avx512f-vminss-2.c (_mm_mask_min_ss)
(_mm_maskz_min_ss): Test new intrinsics.

From-SVN: r248756

PR c++/80840 - ICE with constexpr and reference

* pt.c (convert_nontype_argument): Don't test whether a decl is
value-dependent when binding to a reference.

From-SVN: r248749

cp-tree.h (lang_decl_slector): New enum.

* cp-tree.h (lang_decl_slector): New enum.
(lang_decl_base): Make selector an enum.  Drop decomposition_p
field.
(lang_decl): Use enum for discrimination.
(LANG_DECL_FN_CHECK, LANG_DECL_NS_CHECK, LANG_DECL_PARM_CHECK,
LANG_DECL_DEOMP_CHECK): Use enum.
(DECL_DECOMPOSITION_P): Use selector value.
(SET_DECL_DECOMPOSITION_P): Delete.
(retrofit_lang_decl): Lose SEL parm.
(fit_decomposition_lang_decl): Declare.
* decl.c (cp_finish_decomp, grokdeclarator): Use
fit_decomposition_lang_decl.
* lex.c (maybe_add_lang_decl_raw): New. Broken out of
retrofit_lang_decl.
(set_decl_linkage): New.  Broken out of retrofit_lang_decl.  Use
enum.
(fit_decomposition_lang_decl): Likewise.
(retrofit_lang_decl): Use worker functions.
(cxx_dup_lang_specific_decl): Use selector enum.
(maybe_add_lang_type_raw): New.  Broken out of ...
(cxx_make_type_name): ... here.  Call it.

From-SVN: r248748

* g++.dg/lookup/lambda1.C New.

From-SVN: r248747

ivopt_mult_4.c: Explicitly check comparison on pointer should not be replaced.

gcc/testsuite
* gcc.dg/tree-ssa/ivopt_mult_4.c: Explicitly check comparison
on pointer should not be replaced.

From-SVN: r248746

Doxygen: transform ENUM_BITFIELD and comments starting with '/**'.

2017-05-31  Martin Liska  <mliska@suse.cz>

* filter_params.py:
Transform ENUM_BITFIELD and comments starting with '/**'

From-SVN: r248741

Change comment style to one we normally use.

2017-05-31  Martin Liska  <mliska@suse.cz>

* tree-vect-loop.c (vect_create_epilog_for_reduction):
Change comment style to one we normally use.
(vectorizable_reduction): Likewise.
(vectorizable_induction): Likewise.
* tree-vect-stmts.c (vectorizable_mask_load_store): Likewise.
(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.
* tree-vectorizer.h: Likewise.

From-SVN: r248740

Port Doxygen support script from Perl to Python; add unittests

2017-05-31  David Malcolm  <dmalcolm@redhat.com>
    Martin Liska  <mliska@suse.cz>

* filter_params.py: New, porting the perl script to python,
adding a test suite.
* filter_gcc_for_doxygen_new: New file.

Co-Authored-By: Martin Liska <mliska@suse.cz>
From-SVN: r248739

GIMPLEFE: Handle missing labels in goto statements

gcc/c/

PR testsuite/80580
* gimple-parser.c (c_parser_gimple_if_stmt): Check for empty labels.

gcc/testsuite/

PR testsuite/80580
* gcc.dg/gimplefe-error-7.c: New test.

From-SVN: r248738

Fix configure.ac to respect --{enable,disable}-werror option.

2017-05-31  Martin Liska  <mliska@suse.cz>

* configure.ac: Add handling of stage2_werror_flags to
action-if-given and to action-if-not-given.
* configure: Regenerate.

From-SVN: r248737

passes.c: implement pre-ICE emergency dumping

* passes.c (emergency_dump_function): New.
* tree-pass.h (emergency_dump_function): Declare.
* plugin.c (plugins_internal_error_function): Remove.
* plugin.h (plugins_internal_error_function): Remove declaration.
* toplev.c (internal_error_function): New static function.  Use it...
(general_init): ...here.

From-SVN: r248736

[ARC] Recognise add_n and sub_n in combine again

Since the combine pass canonicalises shift-add insns using plus and
ashift (as opposed to plus and mult which it previously used to do), it
no longer creates *add_n or *sub_n insns, as the patterns match plus and
mult only. The outcome of this is that some opportunities to generate
add{1,2,3} and sub{1,2,3} instructions are missed.

This change adds additional *add_n and *sub_n insns that match the
plus-ashift pattern. The original *add_n and *sub_n insns are still left
in, as they are sometimes generated later on by constant propagation.
The idea of adding these insns is modelled on the changes in:

  https://gcc.gnu.org/ml/gcc-patches/2015-05/msg01882.html

which addresses a similar issue for the PA target.

For the small test cases that are added, even if the combine pass misses
the opportunity to generate addN or subN, constant propagation manages
to do so, so the rtl of the combine pass is checked.

gcc/ChangeLog:

        * config/arc/arc.c (arc_print_operand): Handle constant operands.
        (arc_rtx_costs): Add costs for new patterns.
        * config/arc/arc.md: Additional *add_n and *sub_n patterns.
        * config/arc/predicates.md: Add _1_2_3_operand predicate.

gcc/testsuite/ChangeLog:

        * gcc.target/arc/add_n-combine.c: New test.
        * gcc.target/arc/sub_n-combine.c: New test.

From-SVN: r248735

PR libstdc++/80893 Fix null dereference in vector<bool>

PR libstdc++/80893
* include/bits/stl_bvector.h (vector<bool>::_M_initialize): Avoid
null pointer dereference when size is zero.
* testsuite/23_containers/vector/bool/80893.cc: New.
* testsuite/util/testsuite_allocator.h (PointerBase::PointerBase):
Add non-explicit constructor from nullptr.
(PointerBase::derived() const): Add const-qualified overload.

From-SVN: r248734

[1/2] Add get_next_strinfo helper function

This patch just adds a helper function for getting the next strinfo
in a chain, since part 2 adds another place where we do that.

2017-05-16  Richard Sandiford  <richard.sandiford@linaro.org>

gcc/
* tree-ssa-strlen.c (get_next_strinfo): New function.
(get_stridx_plus_constant): Use it.
(zero_length_string): Likewise.
(adjust_related_strinfos): Likewise.
(adjust_last_stmt): Likewise.

From-SVN: r248732

re PR target/80880 (internal compiler error: in ix86_expand_builtin)

2017-05-31  Richard Biener  <rguenther@suse.de>

PR target/80880
* config/i386/i386.c (ix86_expand_builtin): Remove assert
for arg being an SSA name when expanding IX86_BUILTIN_BNDRET.

* gcc.target/i386/pr80880.c: New testcase.

From-SVN: r248731

Alternative check for vector refs with same alignment

vect_find_same_alignment_drs uses the ddr dependence distance
to tell whether two references have the same alignment.  Although
that's safe with the current code, there's no particular reason
why a dependence distance of 0 should mean that the accesses start
on the same byte.  E.g. a reference to a full complex value could
in principle depend on a reference to the imaginary component.
A later patch adds support for this kind of dependence.

On the other side, checking modulo vf is pessimistic when the step
divided by the element size is a factor of 2.

This patch instead looks for cases in which the drs have the same
base, offset and step, and for which the difference in their constant
initial values is a multiple of the alignment.

2017-05-03  Richard Sandiford  <richard.sandiford@linaro.org>

gcc/
* tree-vect-data-refs.c (vect_find_same_alignment_drs): Remove
loop_vinfo argument and use of dependence distance vectors.
Check instead whether the two references differ only in their
initial value and assume that they have the same alignment if the
difference is a multiple of the vector alignment.
(vect_analyze_data_refs_alignment): Update call accordingly.

gcc/testsuite/
* gcc.dg/vect/vect-103.c: Update wording of dump message.

From-SVN: r248730

Fix typo in a comment in cpuid.h (PR target/79155).

2017-05-31  Martin Liska  <mliska@suse.cz>

PR target/79155
* config/i386/cpuid.h: Fix typo in a comment in cpuid.h.

From-SVN: r248729

* tree-vect-loop-manip.c (create_intersect_range_checks_index)
(create_intersect_range_checks): Move from ...
* tree-data-ref.c (create_intersect_range_checks_index)
(create_intersect_range_checks): ... to here.
(create_runtime_alias_checks): New function factored from ...
* tree-vect-loop-manip.c (vect_create_cond_for_alias_checks): ...
here.  Call above function.
* tree-data-ref.h (create_runtime_alias_checks): New function.

From-SVN: r248726

tree-data-ref.c (prune_runtime_alias_test_list): Relax minimal segment length for dr_b and compute it in wide_int.

* tree-data-ref.c (prune_runtime_alias_test_list): Relax minimal
segment length for dr_b and compute it in wide_int.

gcc/testsuite
* gcc.dg/vect/pr80815-3.c: New test.

From-SVN: r248725

MAINTAINERS: Add self to Write After Approval

2017-05-31  Prakhar Bahuguna  <prakhar.bahuguna@arm.com>

* MAINTAINERS: Add self to Write After Approval

From-SVN: r248723

re PR tree-optimization/80906 (ICE in copy_loop_close_phi_args, at graphite-isl-ast-to-gimple.c:2094)

2017-05-31  Richard Biener  <rguenther@suse.de>

PR tree-optimization/80906
* graphite-isl-ast-to-gimple.c (copy_loop_close_phi_nodes): Get
and pass through iv_map.
(copy_bb_and_scalar_dependences): Adjust.
(translate_pending_phi_nodes): Likewise.
(copy_loop_close_phi_args): Handle code-generating IVs instead
of ICEing.

* gcc.dg/graphite/pr80906.c: New testcase.

From-SVN: r248722

configure.ac (strnlen): Add to AC_CHECK_DECLS.

* configure.ac (strnlen): Add to AC_CHECK_DECLS.
* configure: Likewise.
* config.in: Add HACE_DECL_STRNLEN.

From-SVN: r248721

Daily bump.

From-SVN: r248719

Test if host compiler supports -std=c++11 in ms-sysv.exp

2017-05-30  Tom de Vries  <tom@codesourcery.com>

PR testsuite/80910
* gcc.target/x86_64/abi/ms-sysv/ms-sysv.exp: Exit with status
unsupported if host compiler does not support c++11.
(host_supports_c++11): New proc.

From-SVN: r248701

re PR c/80731 (poor -Woverflow warnings, missing detail)

gcc/testsuite/ChangeLog:
PR c/80731
* g++.dg/ext/utf16-4.C: Relax test.
* gcc.dg/fixed-point/int-warning.c: Adjust.

From-SVN: r248700

PR c++/80856 - ICE with local extern in template

* semantics.c (finish_call_expr): Replace a local extern overload
set in a template with the IDENTIFIER_NODE.

From-SVN: r248699

C++ template type diff printing

gcc/ChangeLog:
* diagnostic-color.c (color_dict): Add "type-diff".
(parse_gcc_colors): Update comment.
* doc/invoke.texi (Diagnostic Message Formatting Options): Add
-fdiagnostics-show-template-tree and -fno-elide-type.
(GCC_COLORS): Add type-diff to example.
(type-diff=): New.
(-fdiagnostics-show-template-tree): New.
(-fno-elide-type): New.
* pretty-print.c (pp_format): Pass quote and formatters[argno] to
the pp_format_decoder callback.  Call any m_format_postprocessor's
"handle" method.
(pretty_printer::pretty_printer): Initialize
m_format_postprocessor.
(pretty_printer::~pretty_printer): Delete any
m_format_postprocessor.
* pretty-print.h (printer_fn): Add bool and const char **
parameters.
(class format_postprocessor): New class.
(struct pretty_printer::format_decoder): Document the new
parameters.
(struct pretty_printer::m_format_postprocessor): New field.
* tree-diagnostic.c (default_tree_printer): Update for new
bool and const char ** params.
* tree-diagnostic.h (default_tree_printer): Likewise.

gcc/c/ChangeLog:
* c-objc-common.c (c_tree_printer): Gain bool and const char **
parameters.

gcc/c-family/ChangeLog:
* c-format.c (gcc_cxxdiag_char_table): Add 'H' and 'I' to
format_chars.
* c.opt (fdiagnostics-show-template-tree): New option.
(felide-type): New option.

gcc/cp/ChangeLog:
* call.c (perform_implicit_conversion_flags): Convert
"from %qT to %qT" to "from %qH to %qI" in diagnostic.
(print_conversion_rejection): Replace pairs of %qT with
%qH and %qI in various places.
(build_user_type_conversion_1): Likewise.
(build_integral_nontype_arg_conv): Likewise.
(build_conditional_expr_1): Likewise.
(convert_like_real): Likewise.
(convert_arg_to_ellipsis): Likewise.
(joust): Likewise.
(initialize_reference): Likewise.
* cvt.c (cp_convert_to_pointer): Likewise.
(cp_convert_to_pointer): Likewise.
(convert_to_reference): Likewise.
(ocp_convert): Likewise.
* error.c (cp_printer): Gain bool and const char ** parameters.
(struct deferred_printed_type): New struct.
(class cxx_format_postprocessor): New class.
(cxx_initialize_diagnostics): Wire up a cxx_format_postprocessor
to pp->m_format_postprocessor.
(comparable_template_types_p): New function.
(newline_and_indent): New function.
(arg_to_string): New function.
(print_nonequal_arg): New function.
(print_template_differences): New function.
(type_to_string_with_compare): New function.
(print_template_tree_comparison): New function.
(append_formatted_chunk): New function.
(add_quotes): New function.
(cxx_format_postprocessor::handle): New function.
(defer_phase_2_of_type_diff): New function.
(cp_printer): Add "quoted" and "buffer_ptr" params.  Implement
%H and %I.
* typeck.c (cp_build_binary_op): Replace pairs of %qT with
%qH and %qI in various places.
(convert_member_func_to_ptr): Likewise.
(build_reinterpret_cast_1): Likewise.
(convert_for_assignment): Likewise.
* typeck2.c (check_narrowing): Likewise.

gcc/fortran/ChangeLog:
* error.c (gfc_format_decoder): Update for new bool and
const char ** params.

gcc/testsuite/ChangeLog:
* g++.dg/plugin/plugin.exp (plugin_test_list): Add...
* g++.dg/plugin/show-template-tree-color-no-elide-type.C: New
test case.
* g++.dg/plugin/show-template-tree-color.C: New test case.
* g++.dg/plugin/show_template_tree_color_plugin.c: New plugin.
* g++.dg/template/show-template-tree-2.C: New test case.
* g++.dg/template/show-template-tree-3.C: New test case.
* g++.dg/template/show-template-tree-4.C: New test case.
* g++.dg/template/show-template-tree-no-elide-type.C: New test case.
* g++.dg/template/show-template-tree.C: New test case.

From-SVN: r248698

Remove gen_cell_microcode, warn_cell_microcode

-mno-gen-cell-microcode is a pain to handle correctly: it causes
different code generation for some very basic patterns, even patterns
specific to that option.  It also requires marking up many patterns,
which is a pain when adding new patterns or modifying existing ones
(first non-trivial step is finding the Cell BE manual!)

-mwarn-cell-microcode is very expensive, even more so after my recent
fix for PR43763; and it used to ICE for seven years before that fix.

This patch removes both these command line options (it leaves the
positive form of -mgen-cell-microcode, doing nothing, for
compatibility).

Where cc_reg_not_micro_cr0_operand was used, we now get the regular
cc_reg_not_cr0_operand.

One testcase used -mgen-cell-microcode and one its negation; both
are adjusted.

* config/rs6000/predicates.md (cc_reg_not_micro_cr0_operand): Delete.
(lwa_operand): Delete rs6000_gen_cell_microcode test.
* config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
rs6000_gen_cell_microcode code.
(rs6000_final_prescan_insn): Delete.
(rs6000_opt_vars): Delete the "gen-cell-microcode" and
"warn-cell-microcode" entries.
* config/rs6000/rs6000.h (FINAL_PRESCAN_INSN): Delete.
* config/rs6000/rs6000.md: Delete rs6000_gen_cell_microcode tests
throughout.  Change cc_reg_not_micro_cr0_operand to
cc_reg_not_cr0_operand throughout.
(*extendhi<mode>2_noload): Delete.
* config/rs6000/rs6000.opt (mgen-cell-microcode): Replace by stub.
(mwarn-cell-microcode): Delete.
* doc/invoke.texi (RS/6000 and PowerPC Options): Delete
-mgen-cell-microcode and -mwarn-cell-microcode.

gcc/testsuite/
* gcc.target/powerpc/shift-dot.c: Delete -mgen-cell-microcode from
dg-options.
* gfortran.dg/pr80107.f: Delete testcase.

From-SVN: r248695

Kill IDENTIFIER_NAMESPACE_BINDINGS

Kill IDENTIFIER_NAMESPACE_BINDINGS
* cp-tree.h (lang_identifier): Delete namespace_bindings.
(IDENTIFIER_NAMESPACE_BINDINGS): Delete.
(lang_decl_ns): Add bindings.
(DECL_NAMESPACE_BINDINGS): New.
* lex.c (retrofit_lang_decl): Create namespace hash table.
* name-lookup.c (find_namespace_slot): Change to use hash-map.
* ptree.c (cxx_print_binding): Delete.
(cxx_print_identifier): Remove NAMESPACE_BINDING printing.

From-SVN: r248694

cp-tree.def (OVERLOAD): Fix comment.

* cp-tree.def (OVERLOAD): Fix comment.
* cp-tree.h: Fix comments and whitespace.
* error.c (dump_decl): Use pp_cxx_colon_colon, ovl_scope.
* name-lookup.c (add_decl_to_level): Assert not class.
(check_local_shadow): Use OVL_P.
(pushdecl_with_scope_1): Rename to ...
(do_pushdecl_with_Scope): ... here.
(do_nonmember_using_decl): Use qualified_namespace_lookup return
value.
(push_class_level_binding_1): Use OVL_P.
(pushdecl_namespace_level): Use do_pushdecl_with_scope.
(pushtag_1): Rename to ...
(do_pushtag): ... here.  Adjust do_pushdecl_with_scope call.
(pushtag): Adjust.
(store_class_bindings): Do not time here.
* name-lookup.h (pushdecl_outermost_localscope): Reorder.
* pt.c (listify): Declare argvec at point of initialization.

From-SVN: r248693

ChangeLog: Add missing entry.

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

From-SVN: r248692

re PR target/80833 (32-bit x86 causes store-forwarding stalls for int64_t -> xmm)

PR target/80833
* config/i386/constraints.md (Yd): New constraint.
(Ye): Ditto.
* config/i386/i386.md (*movti_internal): Add (?r, Ye)
and (?Yd, r) alternatives.  Update insn attributes.
* config/i386/i386.md (*movti_internal): Add (?r, *Ye)
and (?*Yd, r) alternatives.  Update insn attributes.
(double-mode inter-unit splitters): Add new GR<->XMM splitters.

testsuite/ChangeLog:

PR target/80833
* gcc.target/i386/pr80833-1.c: New test.
* gcc.target/i386/pr80833-2.c: Ditto.

From-SVN: r248691

gimplify_modify_expr: avoid DECL_DEBUG_EXPR links across functions

An upcoming patch exposes a bug in gimplify_modify_expr.  There, we try
not to create DECL_DEBUG_EXPR links across functions, however we don't
check that *FROM_P actually belongs to the current function before
modifying it.  This patch fixes this oversight.

gcc/

* gimplify.c (gimplify_modify_expr): Don't create a
DECL_DEBUG_EXPR link if *FROM_P does not belong to the current
function.

From-SVN: r248688

re PR c++/80913 (Infinite loop in cc1plus with stat hack patch)

PR c++/80913
* name-lookup.c (add_decl_to_level): Assert not making a circular
chain.
(update_binding): Don't prematurely slide artificial decl.

* g++.dg/lookup/pr80913.C: New.

From-SVN: r248687

Fix ARM bootstrap failure due to an odd warning:

/src/gcc/gcc/config/arm/arm-builtins.c: In function 'rtx_def* arm_expand_builtin(tree, rtx, rtx, machine_mode, int)':
/src/gcc/gcc/config/arm/arm-builtins.c:3056:46: error: type qualifiers ignored on cast result type [-Werror=ignored-qualifiers]
     if (d->code == (const enum arm_builtins) fcode)
                                              ^~~~~

Avoid the warning by removing const, and bootstrap is OK again.

Committed as trivial patch.

    gcc/
* config/arm/arm-builtins.c (arm_expand_builtin): Remove const.

From-SVN: r248686

tree-vectorizer.h (struct _stmt_vec_info): Add reduc_type and reduc_def fields.

2017-05-30  Richard Biener  <rguenther@suse.de>

* tree-vectorizer.h (struct _stmt_vec_info): Add reduc_type
and reduc_def fields.
(STMT_VINFO_REDUC_TYPE): New define.
(STMT_VINFO_REDUC_DEF): Likewise.
(vect_force_simple_reduction): Adjust prototype.
* tree-vect-loop.c (vect_analyze_scalar_cycles_1): Adjust.
(vect_is_simple_reduction): Remove check_reduction argument.
(vect_force_simple_reduction): Adjust and set
STMT_VINFO_REDUC_TYPE and STMT_VINFO_REDUC_DEF.
(vectorizable_reduction): Do not re-do reduction analysis
but use STMT_VINFO_REDUC_TYPE and STMT_VINFO_REDUC_DEF.
* tree-parloops.c (gather_scalar_reductions): Adjust.

From-SVN: r248685

t-vxworks (LIBGCC2_INCLUDES): Remove extraneous dollar sign before $(MULTIDIR).

2017-05-30  Olivier Hainque  <hainque@adacore.com>

libgcc/
* config/t-vxworks (LIBGCC2_INCLUDES): Remove extraneous
dollar sign before $(MULTIDIR).

From-SVN: r248684

re PR libgomp/80822 (libgomp incorrect affinity when OMP_PLACES=threads)

PR libgomp/80822
* config/linux/affinity.c (gomp_affinity_init_level_1): New function.
(gomp_affinity_init_level): Use it.  Always analyze the core and thread
sibling lists, depending on level just pick up what CPUs to put
together into a place vs. whether add multiple ordered places.

From-SVN: r248683

re PR tree-optimization/80901 (ICE on valid code at -Os and above on x86_64-linux-gnu: in verify_loop_structure, at cfgloop.c:1644)

2017-05-30  Richard Biener  <rguenther@suse.de>

PR middle-end/80901
* cfgexpand.c (expand_gimple_cond): Match up loop fixup with
split_edge code.

* gcc.dg/torture/pr80901.c: New testcase.

From-SVN: r248681

Vector peeling cost model 6/6

gcc/ChangeLog:

2017-05-24  Robin Dapp  <rdapp@linux.vnet.ibm.com>

* tree-vect-data-refs.c (vect_get_peeling_costs_all_drs):
Introduce unknown_misalignment parameter and remove vf.
(vect_peeling_hash_get_lowest_cost):
Pass unknown_misalignment parameter.
(vect_enhance_data_refs_alignment):
Fix unsupportable data ref treatment.

From-SVN: r248680

Vector peeling cost model 5/6

gcc/testsuite/ChangeLog:

2017-05-30  Robin Dapp  <rdapp@linux.vnet.ibm.com>

* gcc.target/s390/vector/vec-nopeel-2.c: New test.

From-SVN: r248679

Vector peeling cost model 4/6

gcc/ChangeLog:

2017-05-30  Robin Dapp  <rdapp@linux.vnet.ibm.com>

* tree-vect-data-refs.c (vect_get_data_access_cost):
Workaround for SLP handling.
(vect_enhance_data_refs_alignment):
Compute costs for doing no peeling at all, compare to the best
peeling costs so far and avoid peeling if cheaper.

From-SVN: r248678

Vector peeling cost model 3/6

gcc/ChangeLog:

2017-05-30  Robin Dapp  <rdapp@linux.vnet.ibm.com>

* tree-vect-data-refs.c (vect_peeling_hash_choose_best_peeling):
Return peeling info and set costs to zero for unlimited cost
model.
(vect_enhance_data_refs_alignment): Also inspect all datarefs
with unknown misalignment. Compute and costs for unknown
misalignment, compare them to the costs for known misalignment
and choose the cheapest for peeling.

From-SVN: r248677

Vector peeling cost model 2/6

gcc/ChangeLog:

2017-05-30  Robin Dapp  <rdapp@linux.vnet.ibm.com>

* tree-vect-data-refs.c (vect_update_misalignment_for_peel): Rename.
(vect_get_peeling_costs_all_drs): Create function.
(vect_peeling_hash_get_lowest_cost):
Use vect_get_peeling_costs_all_drs.
(vect_peeling_supportable): Create function.
(vect_enhance_data_refs_alignment): Use vect_peeling_supportable.

From-SVN: r248676

Vector peeling cost model 1/6

gcc/ChangeLog:

2017-05-30  Robin Dapp  <rdapp@linux.vnet.ibm.com>

* tree-vect-data-refs.c (vect_compute_data_ref_alignment): Create
DR_HAS_NEGATIVE_STEP.
(vect_update_misalignment_for_peel): Define DR_MISALIGNMENT.
(vect_enhance_data_refs_alignment): Use.
(vect_duplicate_ssa_name_ptr_info): Use.
* tree-vectorizer.h (dr_misalignment): Use.
(known_alignment_for_access_p): Use.

From-SVN: r248675

re PR target/78838 (msp430 option -mcode-region=either, -ffunction-sections, and interrupt function attributes cause incorrect section to be created)

PR target/78838
gcc * config/msp430/msp430.c (gen_prefix): Return NULL when section name is
.lowtext.
(has_section_name): New function.

testsuite * gcc.target/msp430/interrupt_fn_placement.c: New test.

From-SVN: r248674

Fix parenthesis in auto-profile.c (PR other/80909).

2017-05-30  Martin Liska  <mliska@suse.cz>

PR other/80909
* auto-profile.c (get_function_decl_from_block): Fix
parenthesis.

From-SVN: r248647

re PR tree-optimization/80876 (ICE in verify_loop_structure, at cfgloop.c:1644 (error: loop 1's latch does not have an edge to its header))

2017-05-30  Richard Biener  <rguenther@suse.de>

PR middle-end/80876
* cfgexpand.c (expand_gimple_cond): Fixup preserving loops again.

* gcc.dg/torture/pr80876.c: New testcase.

From-SVN: r248633

Remove dump_file_info::dump_file_info and replace it with macro.

2017-05-30  Martin Liska  <mliska@suse.cz>

* dumpfile.c: Use newly added macro DUMP_FILE_INFO.
* dumpfile.h (struct dump_file_info): Remove ctors.

From-SVN: r248617

Compare predictor values with these defined in predict.def.

2017-05-30  Martin Liska  <mliska@suse.cz>

* analyze_brprob.py: Add new argument to parse and modify
predict.def file.
* analyze_brprob_spec.py: Likewise.

From-SVN: r248600

Fix coding style.

2017-05-30  Martin Liska  <mliska@suse.cz>

* predict.def: Fix GNU coding style.

From-SVN: r248599