Enable VNNI support [1/5]

gcc/
* common/config/i386/i386-common.c (OPTION_MASK_ISA_AVX512VNNI_SET,
OPTION_MASK_ISA_AVX512VNNI_UNSET): New.
(ix86_handle_option): Handle -mavx512vnni.
* config/i386/cpuid.h (bit_AVX512VNNI): New bit.
* config/i386/driver-i386.c (host_detect_local_cpu): Handle new bit.
* config/i386/i386-c (__AVX512VNNI__): New.
* config/i386/i386.c (ix86_target_string): Handle new option.
(ix86_valid_target_attribute_inner_p): Handle new option.
* config/i386/i386.h (TARGET_AVX512VNNI, TARGET_AVX512VNNI_P): New.
* config/i386/i386.opt (mavx512vnni): New option.

From-SVN: r255401

compiler: no nil check needed for closure var dereferences

    Add the "no nil check needed" annotation to the dereference operations
    created in Parse::enclosing_var_reference (this is safe since the
    closure object is under control of the compiler, and pointer fields in
    it will always be non-nil).

    Reviewed-on: https://go-review.googlesource.com/81795

From-SVN: r255400

Makefile.am (ARCH_AARCH64_LINUX): Add IFUNC_OPTIONS and libatomic_la_LIBADD.

2017-12-04  Steve Ellcey  <sellcey@cavium.com>

* Makefile.am (ARCH_AARCH64_LINUX): Add IFUNC_OPTIONS and
libatomic_la_LIBADD.
* config/linux/aarch64/host-config.h: New file.
* configure.ac (IFUNC_RESOLVER_ARGS): Define.
(ARCH_AARCH64_LINUX): New conditional for IFUNC builds.
* configure.tgt (aarch64): Set ARCH and try_ifunc.
(aarch64*-*-linux*) Update config_path.
(aarch64*-*-linux*) Set IFUNC_RESOLVER_ARGS.
* libatomic_i.h (GEN_SELECTOR): Add IFUNC_RESOLVER_ARGS argument.
* Makefile.in: Regenerate.
* auto-config.h.in: Regenerate.
* configure: Regenerate.

From-SVN: r255399

Daily bump.

From-SVN: r255398

re PR target/81616 (Update -mtune=generic for the current Intel and AMD processors)

PR target/81616
* athlon.md: Disable for generic.
* haswell.md: Enable for generic.
* i386.c (ix86_sched_init_global): Add core hooks for generic.
* x86-tune-sched.c (ix86_issue_rate): Increase issue rate for generic
to 4.
(ix86_adjust_cost): Move generic to haswell path.

From-SVN: r255395

sparc.c (sparc_do_work_around_errata): Use mem_ref instead of MEM_P in a couple more places.

* config/sparc/sparc.c (sparc_do_work_around_errata): Use mem_ref
instead of MEM_P in a couple more places.  Fix formatting issues.

From-SVN: r255393

Fix warnings in <bits/regex_compiler.tcc>

* include/bits/regex_compiler.tcc: Use C-style comment to work around
PR preprocessor/61638.
(__INSERT_REGEX_MATCHER): Replace GNU extension with __VA_ARGS__.

From-SVN: r255392

Fix -Wunused warnings in libstdc++ headers

* config/io/basic_file_stdio.h (__basic_file): Remove name of unused
parameter.
* include/bits/boost_concept_check.h: Add pragmas to disable
-Wunused-local-typedef warnings.
* include/bits/codecvt.h (codecvt_byname<char16_t, char, mbstate_t>)
(codecvt_byname<char32_t, char, mbstate_t>): Remove name of unused
parameter.
* include/bits/locale_facets_nonio.tcc (time_get::do_get_weekday)
(time_get::do_get_monthname, time_get::do_get_year): Remove unused
variables.
* include/std/bitset (_Base_bitset<0>::_M_getword): Remove name of
unused parameter.
* include/std/streambuf (_IsUnused): Define.
(basic_streambuf::imbue, basic_streambuf::pbackfail)
(basic_streambuf::overflow): Add macro to unused parameters.
* testsuite/24_iterators/operations/prev_neg.cc: Adjust dg-error.

From-SVN: r255391

PR c++/83273 - constexpr if allows non-constant condition

* semantics.c (finish_if_stmt_cond): Use require_constant_expression
rather than is_constant_expression.
* constexpr.c (potential_constant_expression_1) [LAMBDA_EXPR]: Allow
in C++17.

From-SVN: r255390

Fix typos in riscv register save/restore.

gcc/
* config/riscv/riscv.c (riscv_for_each_saved_reg): Use GP_REG_LAST
instead of GP_REG_LAST-1.
(riscv_adjust_libcall_cfi_prologue): Likewise.
(riscv_adjust_libcall_cri_epilogue): Likewise.
* config/riscv/riscv.h (CALL_USED_REGISTERS): Change a6 to t6 in
comment.

From-SVN: r255389

Fix missing newlines from local-pure-const pass dump

I noticed the debugging output from local-pure-const pass is missing a
newline in a couple places, leading to this:

 local analysis of main
   scanning: i ={v} 0;
    Volatile stmt is not const/pure
    Volatile operand is not const/pure  scanning: j ={v} 20;
    Volatile stmt is not const/pure
    Volatile operand is not const/pure  scanning: vol.0_10 ={v} i;
    Volatile stmt is not const/pure

It should've been:

 local analysis of main
   scanning: i ={v} 0;
    Volatile stmt is not const/pure
    Volatile operand is not const/pure
   scanning: j ={v} 20;
    Volatile stmt is not const/pure
    Volatile operand is not const/pure
   scanning: vol.0_10 ={v} i;
    Volatile stmt is not const/pure

gcc/ChangeLog:

2017-12-04  Luis Machado  <luis.machado@linaro.org>

* ipa-pure-const.c (check_decl): Add missing newline.
(state_from_flags): Likewise.

From-SVN: r255388

re PR tree-optimization/78496 (Missed opportunities for jump threading)

PR tree-optimizatin/78496
* gimple-ssa-evrp-analyze.h
(evrp_range_analyzer::get_vr_values): Simplify.
* gimple-ssa-evrp-analyze.c: Corresponding changes.
* tree-ssa-dom.c: Include alloc-pool.h, tree-vrp.h, vr-values.h
and gimple-ssa-evrp-analyze.h.
(dom_opt_dom_walker class): Add evrp_range_analyzer member.
(simplify_stmt_for_jump_threading): Copy a blob of code from
tree-vrp.c to use ranges to simplify statements.
(dom_opt_dom_walker::before_dom_children): Call
evrp_range_analyzer::{enter,record_ranges_from_stmt} methods.
(dom_opt_dom_walker::after_dom_children): Similarly for
evrp_range_analyzer::leave.
(dom_opt_dom_walker::optimize_stmt): Use EVRP ranges to optimize
conditionals.

PR tree-optimization/78496
* gcc.dg/builtin-unreachable-6.c: Disable DOM.
* gcc.dg/builtin-unreachable-6a.c: New test.
* gcc.dg/tree-ssa/20030922-1.c: No longer XFAIL.
* gcc.dg/ssa-dom-branch-1.c: Tweak expected output.

From-SVN: r255387

* gimple-ssa-evrp-analyze.c
(evrp_range_analyzer::extract_range_from_stmt):  Always use
vr_values::update_value_range so preexisting range info is
medged with new range info, even if the new range is VR_VARYING.

From-SVN: r255386

combine: Remove use_crosses_set_p

This removes use_crosses_set_p, and uses modified_between_p instead
everywhere it was used.  This improves optimisation.

* combine.c: Adjust comment.
(use_crosses_set_p): Delete.
(can_combine_p): Use modified_between_p instead of use_crosses_set_p.
(try_combine): Ditto.

From-SVN: r255384

re PR tree-optimization/83255 ([graphite] Wrong code w/ -O1 -floop-nest-optimize)

2017-12-04  Richard Biener  <rguenther@suse.de>

PR tree-optimization/83255
* graphite-isl-ast-to-gimple.c (translate_isl_ast_node_for):
Re-add zero-iteration check.

* gcc.dg/graphite/pr83255.c: New testcase.

From-SVN: r255382

RTEMS/Ada: Account for 64-bit time_t

The Newlib time_t has now 64 bits for RTEMS.

gcc/ada

* gcc-interface/Makefile.in (RTEMS): Use s-osprim__rtems.adb.
* libgnat/s-osprim__rtems.adb: New file.
* libgnarl/s-osinte__rtems.adb (pthread_cond_t): Fix alignment.
(pthread_mutexattr_t): Likewise.
(pthread_rwlockattr_t): Likewise.
(pthread_rwlock_t): Likewise.
(time_t): Use 64-bit integer.

From-SVN: r255380

lra: Clobbers in a parallel are earlyclobbers (PR83245)

The documentation (rtl.texi) says:

  When a @code{clobber} expression for a register appears inside a
  @code{parallel} with other side effects, the register allocator
  guarantees that the register is unoccupied both before and after that
  insn if it is a hard register clobber.

and at least the rs6000 backend relies on that (see PR83245).  This
patch restores that behaviour.

Registers that are also used as operands in the instruction are not
treated as earlyclobber, so such insns also still work (PR80818, an
s390 testcase).

PR rtl-optimization/83245
* lra.c (collect_non_operand_hard_regs): Treat clobbers of non-operand
hard registers as earlyclobber, also if not in an asm.

From-SVN: r255377

re PR bootstrap/83265 (Bootstrap failure on powerpc64)

PR bootstrap/83265
Revert
2017-12-01  Segher Boessenkool  <segher@kernel.crashing.org>

PR target/43871
* config/rs6000/rs6000.c (rs6000_option_override_internal): Set
rs6000_cpu based on cpu_index, not tune_index.

From-SVN: r255376

re PR tree-optimization/83238 ([graphite] ICE in graphite_can_represent_scev, at graphite-scop-detection.c:971)

2017-12-04  Richard Biener  <rguenther@suse.de>

PR tree-optimization/83238
* graphite-scop-detection.c (scop_detection::merge_sese): Make
code match comment, rejecting invalid SESE regions.

* gcc.dg/graphite/pr83238.c: New testcase.

From-SVN: r255375

Require effective target alloca for pr82875.c

2017-12-04  Tom de Vries  <tom@codesourcery.com>

* gcc.dg/pr82875.c: Require effective target alloca.

From-SVN: r255374

Daily bump.

From-SVN: r255372

pa.c (pa_legitimate_address_p): For scaled indexing...

* config/pa/pa.c (pa_legitimate_address_p): For scaled indexing,
require base operand is a REG_POINTER prior to reload on targets
with non-equivalent space registers.

From-SVN: r255369

re PR fortran/36313 ([F03] {MIN,MAX}{LOC,VAL} should accept character arguments)

2017-12-03  Thomas Koenig  <tkoenig@gcc.gnu.org>

PR fortran/36313
* check.c (gfc_check_minval_maxval): Use
int_orLreal_or_char_check_f2003 for array argument.
* iresolve.c (gfc_resolve_maxval): Insert number in
function name for character arguments.
(gfc_resolve_minval): Likewise.
* trans-intrinsic.c (gfc_conv_intrinsic_minmaxloc):
Fix comment.
(gfc_conv_intrinsic_minmaxval): Resort arguments and call library
function if dealing with a character function.

2017-12-03  Thomas Koenig  <tkoenig@gcc.gnu.org>

PR fortran/36313
* Makefile.am: Add new files for character-valued
maxval and minval.
* Makefile.in: Regenerated.
* gfortran.map: Add new functions.
* m4/iforeach-s2.m4: New file.
* m4/ifunction-s2.m4: New file.
* m4/iparm.m4: Add intitval for minval and maxval.
* m4/maxval0s.m4: New file.
* m4/maxval1s.m4: New file.
* m4/minval0s.m4: New file.
* m4/minval1s.m4: New file.
        * generated/maxval0_s1.c: New file.
        * generated/maxval0_s4.c: New file.
        * generated/maxval1_s1.c: New file.
        * generated/maxval1_s4.c: New file.
        * generated/minval0_s1.c: New file.
        * generated/minval0_s4.c: New file.
        * generated/minval1_s1.c: New file.
        * generated/minval1_s4.c: New file.

2017-12-03  Thomas Koenig  <tkoenig@gcc.gnu.org>

PR fortran/36313
* gfortran.dg/maxval_char_1.f90: New test.
* gfortran.dg/maxval_char_2.f90: New test.
* gfortran.dg/maxval_char_3.f90: New test.
* gfortran.dg/maxval_char_4.f90: New test.
* gfortran.dg/minval_char_1.f90: New test.
* gfortran.dg/minval_char_2.f90: New test.
* gfortran.dg/minval_char_3.f90: New test.
* gfortran.dg/minval_char_4.f90: New test.

From-SVN: r255367

Fix typo in ChangeLog.

From-SVN: r255366

re PR fortran/83191 (Writing a namelist with repeated complex numbers)

2017-12-03  Jerry DeLisle  <jvdelisle@gcc.gnu.org>
            Dominique d'Humieres  <dominiq@lps.ens.fr>

        PR libgfortran/83191
        * io/transfer.c (list_formatted_read_scalar): Do not set
        namelist_mode bit here. (namelist_read): Likewise.
        (data_transfer_init): Clear the mode bit here.
        (finalize_transfer): Do set the mode bit just before any calls
        to namelist_read or namelist_write. It can now be referred to
        in complex_write.
        ^ io/write.c (write_complex): Suppress the leading blanks when
        namelist_mode bit is not set to 1.

        * gfortran.dg/namelist_95.f90: New test.

Co-Authored-By: Dominique d'Humieres <dominiq@lps.ens.fr>
From-SVN: r255365

Fix spelling error in log.

From-SVN: r255363

re PR fortran/83225 (runtime error in transfer.c)

2017-12-02  Jerry DeLisle  <jvdelisle@gcc.gnu.org>

PR libgfortran/83225
* io/io.h (is_internal_unit): Use the unit_is_internal bit.
* io/transfer.c (data_transfer_init): Set the bit to true for
internal umits. Use that bit for checks for internal unit
initializations.
* io/unit.c (insert_unit): As a precaution, set the
internal_unit_kind to zero when a unit structure is first created.

From-SVN: r255362

Daily bump.

From-SVN: r255361

ipa-cp.c (ipcp_lattice<valtype>::print): Update dumping.

* ipa-cp.c (ipcp_lattice<valtype>::print): Update dumping.
(update_specialized_profile): Fix updating of counts.
(perhaps_add_new_callers): Likewise.

From-SVN: r255358

re PR target/81616 (Update -mtune=generic for the current Intel and AMD processors)

PR target/81616
* x86-tune.def: Remove obsolette FIXMEs.
(X86_TUNE_PARTIAL_FLAG_REG_STALL): Disable for generic
(X86_TUNE_FUSE_CMP_AND_BRANCH_32, X86_TUNE_FUSE_CMP_AND_BRANCH_64,
X86_TUNE_FUSE_CMP_AND_BRANCH_SOFLAGS, X86_TUNE_FUSE_ALU_AND_BRANCH):
Enable for generic.
(X86_TUNE_PAD_RETURNS): Disable for generic.
* gcc.target/i386/pad-1.c: Compile for amdfam10.
* gcc.target/i386/align-limit.c: Likewise.

From-SVN: r255357

re PR tree-optimization/83170 (ICE: Segmentation fault - during GIMPLE pass: store-merging)

PR tree-optimization/83170
PR tree-optimization/83241
* gimple-ssa-store-merging.c
(imm_store_chain_info::try_coalesce_bswap): Update vuse field from
gimple_vuse (ins_stmt) in case it has changed.
(imm_store_chain_info::output_merged_store): Likewise.

* gcc.dg/store_merging_17.c: New test.

From-SVN: r255356

tree-chkp.c (chkp_compute_bounds_for_assignment): Handle POINTER_DIFF_EXPR.

* tree-chkp.c (chkp_compute_bounds_for_assignment): Handle
POINTER_DIFF_EXPR.

* gcc.target/i386/mpx/pointer-diff-1.c: New test.

From-SVN: r255355

re PR sanitizer/81212 (-Wreturn-type is disabled when used together with -fsanitize=return)

PR c++/81212
* tree-cfg.c (pass_warn_function_return::execute): Handle
__builtin_ubsan_handle_missing_return like __builtin_unreachable
with BUILTINS_LOCATION.

* g++.dg/ubsan/pr81212.C: New test.
* g++.dg/ubsan/return-1.C: Add -Wno-return-type to dg-options.
* g++.dg/ubsan/return-2.C: Likewise.
* g++.dg/ubsan/return-7.C: Likewise.

From-SVN: r255354

re PR target/78643 (ICE in convert_move, at expr.c:230)

PR target/78643
PR target/80583
* expr.c (get_inner_reference): If DECL_MODE of a non-bitfield
is BLKmode for vector field with vector raw mode, use TYPE_MODE
instead of DECL_MODE.

* gcc.target/i386/pr80583.c: New test.

From-SVN: r255353

i386-protos.h (standard_sse_constant_opcode): Change last argument to rtx pointer.

* config/i386/i386-protos.h (standard_sse_constant_opcode): Change
last argument to rtx pointer.
* config/i386/i386.c (standard_sse_constant_opcode): Replace X argument
with OPERANDS.  For AVX+ 128-bit VEX encoded instructions over 256-bit
or 512-bit.  If setting EXT_REX_SSE_REG_P, use EVEX encoded insn
depending on the chosen ISAs.
* config/i386/i386.md (*movxi_internal_avx512f, *movoi_internal_avx,
*movti_internal, *movdi_internal, *movsi_internal, *movtf_internal,
*movdf_internal, *movsf_internal): Adjust standard_sse_constant_opcode
callers.
* config/i386/sse.md (mov<mode>_internal): Likewise.
* config/i386/mmx.md (*mov<mode>_internal): Likewise.

From-SVN: r255352

doc update for -dp

* doc/invoke.texi (-dp): Say that instruction cost is printed as well.

From-SVN: r255351

rs6000: Improve fusion assembler output

This improves the output for load and store fusion a little.  In most
cases it removes the comment output, because that makes the generated
assembler code hard to read, and equivalent info is available with -dp
anyway.  For the vector loads it puts the comment on the second insn,
where it doesn't interfere with other debug comments.

* config/rs6000/rs6000-protos.h (emit_fusion_addis): Remove last two
parameters from prototype.
* config/rs6000/rs6000.c (emit_fusion_addis): Remove last two
parameters.  Don't print a comment.
(emit_fusion_gpr_load): Adjust.
(emit_fusion_load_store): Adjust.
* config/rs6000/rs6000.md (*fusion_p9_<mode>_constant): Adjust.
* config/rs6000/vsx.md (two peepholes): Print the "vector load fusion"
comment on the second line.

gcc/testsuite/
* gcc.target/powerpc/fusion.c: Add -dp to options.  Adjust the expected
output.
* gcc.target/powerpc/fusion3.c: Ditto.
* gcc.target/powerpc/fusion4.c: Ditto.

From-SVN: r255350

rs6000: Set rs6000_cpu correctly (PR43871)

We set rs6000_cpu based on tune_index, but it should be cpu_index.
This patch fixes it.

PR target/43871
* config/rs6000/rs6000.c (rs6000_option_override_internal): Set
rs6000_cpu based on cpu_index, not tune_index.

From-SVN: r255349

final: Improve output for -dp and -fverbose-asm

This improves the assembler output (for -dp and -fverbose-asm) in
several ways.  It always prints the insn_cost.  It does not print
"[length = NN]" but "[c=NN l=NN]", to save space.  It does not add one
to the instruction alternative number (everything else starts counting
those at 0, too).  And finally, it tries to keep things lined up in
columns a bit better.

* final.c (output_asm_name): Print insn_cost.  Shorten output.  Print
which_alternative instead of which_alternative + 1.
(output_asm_insn): Print an extra tab if the template is short.

From-SVN: r255348

runtime: export cgoCheck functions

    The functions cgoCheckPointer and cgoCheckResult are called by code
    generated by cgo. That means that we need to export them using
    go:linkname, as otherwise they are local symbols. The cgo code
    currently uses weak references to only call the symbols if they are
    defined, which is why it has been working--the cgo code has not been
    doing any checks.

    Reviewed-on: https://go-review.googlesource.com/80295

From-SVN: r255347

compiler: avoid GCC middle-end control warnings

    GCC has started emitting "control reaches end of non-void function"
    warnings. Avoid them for Go by 1) marking the builtin function panic
    and the compiler-generated function __go_runtime_error as not
    returning and 2) adding a default case to the switch used for select
    statements that simply calls __builtin_unreachable.

    Fixes golang/go#22767

    Reviewed-on: https://go-review.googlesource.com/80416

* go-gcc.cc (Gcc_backend::Gcc_backend): Define
__builtin_unreachable.
(Gcc_backend::function): Add does_not_return parameter.

From-SVN: r255346

Delete obsolete DWARF1 references.

gcc/
* common.opt (use_gnu_debug_info_extensions): Delete DWARF_DEBUG from
comment.
* config/vx-common.h (DWARF_DEBUGGING_INFO): Delete undef.
* doc/tm.texi.in (PREFERRED_DEBUGGING_TYPE): Delete DWARF_DEBUG
reference.
* doc/tm.texi: Regenerate.

From-SVN: r255345

Daily bump.

From-SVN: r255344

re PR target/81959 (PowerPC __float128 optimization fails with integer PRE_INC addresses)

[gcc]
2017-12-01  Michael Meissner  <meissner@linux.vnet.ibm.com>

PR target/81959
* config/rs6000/rs6000.c (rs6000_address_for_fpconvert): Check for
whether we can allocate pseudos before trying to fix an address.
* config/rs6000/rs6000.md (float_<mode>si2_hw): Make sure the
memory address is indexed or indirect.
(floatuns_<mode>si2_hw2): Likewise.

[gcct/testsuite]
2017-12-01  Michael Meissner  <meissner@linux.vnet.ibm.com>

PR target/81959
* gcc.target/powerpc/pr81959.c: New test.

From-SVN: r255341

compiler: introduce size threshold for nil checks

    Add a new control variable to the Gogo class that stores the size
    threshold for nil checks. This value can be used to control the
    policy for deciding when a given deference operation needs a check and
    when it does not. A size threshold of -1 means that every potentially
    faulting dereference needs an explicit check (and branch to error
    call). A size threshold of K (where K > 0) means that if the size of
    the object being dereferenced is >= K, then we need a check.

    Reviewed-on: https://go-review.googlesource.com/80996

* go-c.h (go_create_gogo_args): Add nil_check_size_threshold
field.
* go-lang.c (go_langhook_init): Set nil_check_size_threshold.

From-SVN: r255340

* doc/extend.texi: Mention C++14 library complex suffix.

From-SVN: r255339

Give #include hints for <complex>.

* name-lookup.c (get_std_name_hint): Add <complex>.
* parser.c (cp_parser_diagnose_invalid_type_name): Call
suggest_alternative_in_explicit_scope.
(cp_parser_namespace_name): Likewise.

From-SVN: r255336

PR c++/79228 - extensions hide C++14 complex literal operators

libcpp/
* expr.c (interpret_float_suffix): Ignore 'i' in C++14 and up.
(interpret_int_suffix): Likewise.
gcc/cp/
* parser.c (cp_parser_userdef_numeric_literal): Be helpful about
'i' in C++14 and up.

From-SVN: r255335

* Makefile.in (TAGS): Add c-family/*.cc.

From-SVN: r255334

[AArch64] Fix address printing on ILP32

Fix address printing for ILP32.  The md file uses 'a' in assembler
templates for symbolic addresses in adrp/add, which end up calling
aarch64_print_operand_address.  However in ILP32 these are not valid
memory addresses (being ptr_mode rather than Pmode), so the assert
triggers.  Since it is incorrect to use symbols in memory addresses
(besides literal pool accesses), change the 'a' to 'c' in the md file.

Skip one failing test in ILP32 which combines the 'p' modifier with the 'a'
assembler template to fake a memory reference.

    gcc/
* config/aarch64/aarch64.md (call_insn): Use %c rather than %a.
(call_value_insn): Likewise.
(sibcall_insn): Likewise.
(sibcall_value_insn): Likewise.
(movsi_aarch64): Likewise.
(movdi_aarch64): Likewise.
(add_losym_): Likewise.
(ldr_got_small_): Likewise.
(ldr_got_small_sidi): Likewise.
(ldr_got_small_28k_): Likewise.
(ldr_got_small_28k_sidi): Likewise.
* config/aarch64/aarch64.c (aarch64_print_address_internal):
Move output_addr_const to symbolic case. Add error check.
    testsuite/
* gcc.dg/asm-4.c: Skip on AArch64 with ILP32 as test is incorrect.

From-SVN: r255333

Fix an undefined behavior in fortran/decl.c

From-SVN: r255332

re PR fortran/83224 (creating character array from elements shorter than declared does not pad with whitespace properly and aborts)

2017-12-01  Thomas Koenig  <tkoenig@gcc.gnu.org>

PR fortran/83224
* frontend-passes.c (create_var): Also handle
character arrays, handling deferred lenghts.

2017-12-01  Thomas Koenig  <tkoenig@gcc.gnu.org>

PR fortran/83224
* gfortran.dg/dependency_51.f90: New test.

From-SVN: r255331

Fix narrowing conversions in string_view types

* include/experimental/string_view (basic_string_view::_S_compare):
Use value-init so narrowing conversions are not ill-formed.
* include/std/string_view (basic_string_view::_S_compare): Likewise.

From-SVN: r255321

Disable -Wliteral-suffix for standard UDLs

* include/bits/basic_string.h (operator""s): Add pragmas to disable
-Wliteral-suffix warnings.
* include/experimental/string_view (operator""sv): Likewise.
* include/std/chrono (operator""h, operator""min, operator""s)
(operator""ms, operator""us, operator""ns): Likewise.
* include/std/complex (operator""if, operator""i, operator""il):
Likewise.
* include/std/string_view (operator""sv): Likewise.
* testsuite/20_util/duration/literals/range.cc: Adjust dg-error.

From-SVN: r255320

S/390: Split MVC instruction for better forwarding

Certain lengths used in an MVC instruction might disable operand
forwarding.  Split MVCs into up to 2 forwardable ones if possible.

gcc/ChangeLog:

2017-12-01  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>

* config/s390/predicates.md (plus16_Q_operand): New predicate.
* config/s390/s390.md: Disable MVC merging peephole if it would
disable operand forwarding.
(new peephole2): Split MVCs if it would turn them into up to 2
forwardable MVCs.

From-SVN: r255319

Add comment to fix -Wfallthrough warning

* include/bits/locale_facets_nonio.tcc (money_get::_M_extract): Add
fallthrough comment.

From-SVN: r255318

Fix -Wempty-body warnings for debug assertions

* include/bits/node_handle.h (_Node_handle_common::operator=)
(_Node_handle_common::_M_swap): Add braces around debug assertions.

From-SVN: r255317

Use const char* to fix -Wwrite-strings warning

* include/ext/ropeimpl.h (rope::_S_dump): Use const char*.

From-SVN: r255316

Add [[noreturn]] attributes to fix warning

* libsupc++/nested_exception.h (__throw_with_nested_impl): Add
noreturn attribute.

From-SVN: r255315

Remove stray semi-colons at namespace scope

* include/bits/regex_executor.tcc (_Executor::_M_rep_once_more):
Remove semi-colon after function body.
* include/bits/uniform_int_dist.h (_Power_of_2): Likewise.

From-SVN: r255314

Use value-init syntax in std::__detail::__airy

* include/tr1/modified_bessel_func.tcc (__airy): Use value-init not
list-init.

From-SVN: r255313

Add noexcept to std::integral_constant members

* include/std/type_traits (integral_constant): Make member functions
noexcept (LWG 2346).
* include/std/utility (integer_sequence): Likewise.

From-SVN: r255312

re PR fortran/82605 ([PDT] ICE in insert_parameter_exprs, at fortran/decl.c:3154)

2017-12-01  Paul Thomas  <pault@gcc.gnu.org>

PR fortran/82605
* resolve.c (get_pdt_constructor): Initialize 'cons' to NULL.
(resolve_pdt): Correct typo in prior comment. Emit an error if
any parameters are deferred and the object is neither pointer
nor allocatable.

PR fortran/82606
* decl.c (gfc_get_pdt_instance): Continue if the parameter sym
is not present or has no name. Select the parameter by name
of component, rather than component order. Remove all the other
manipulations of 'tail' when building the pdt instance.
(gfc_match_formal_arglist): Emit and error if a star is picked
up in a PDT decl parameter list.

PR fortran/82622
* trans-array.c (set_loop_bounds): If a GFC_SS_COMPONENT has an
info->end, use it rather than falling through to
gcc_unreachable.
(structure_alloc_comps): Check that param->name is non-null
before comparing with the component name.
* trans-decl.c (gfc_get_symbol_decl): Do not use the static
initializer for PDT symbols.
(gfc_init_default_dt): Do nothing for PDT symbols.
* trans-io.c (transfer_array_component): Parameterized array
components use the descriptor ubound since the shape is not
available.

PR fortran/82719
PR fortran/82720
* trans-expr.c (gfc_conv_component_ref): Do not use the charlen
backend_decl of pdt strings. Use the hidden component instead.
* trans-io.c (transfer_expr): Do not do IO on "hidden" string
lengths. Use the hidden string length for pdt string transfers
by adding it to the se structure. When finished nullify the
se string length.

PR fortran/82866
* decl.c (gfc_match_formal_arglist): If a name is not found or
star is found, while reading a type parameter list, emit an
immediate error.
(gfc_match_derived_decl): On reading a PDT parameter list, on
failure to match call gfc_error_recovery.

PR fortran/82978
* decl.c (build_struct): Character kind defaults to 1, so use
kind_expr whatever is the set value.
(gfc_get_pdt_instance): Ditto.
* trans-array.c (structure_alloc_comps): Copy the expression
for the PDT string length before parameter substitution. Use
this expression for evaluation and free it after use.

2017-12-01  Paul Thomas  <pault@gcc.gnu.org>

PR fortran/82605
* gfortran.dg/pdt_4.f03 : Incorporate the new error.

PR fortran/82606
* gfortran.dg/pdt_19.f03 : New test.
* gfortran.dg/pdt_21.f03 : New test.

PR fortran/82622
* gfortran.dg/pdt_20.f03 : New test.
* gfortran.dg/pdt_22.f03 : New test.

PR fortran/82719
PR fortran/82720
* gfortran.dg/pdt_23.f03 : New test.

PR fortran/82866
* gfortran.dg/pdt_24.f03 : New test.

PR fortran/82978
* gfortran.dg/pdt_10.f03 : Correct for error in coding the for
kind 4 component and change the kind check appropriately.
* gfortran.dg/pdt_25.f03 : New test.

From-SVN: r255311

Fix bug in an OpenACC async test case

libgomp/
* testsuite/libgomp.oacc-c-c++-common/data-2-lib.c: Add missing
call to acc_wait (1).

From-SVN: r255308

re PR tree-optimization/83232 (fma3d spec2000 regression on zen with -Ofast (generic tuning) after r255268 by missed SLP oppurtunity)

2017-12-01  Richard Biener  <rguenther@suse.de>

PR tree-optimization/83232
* tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Fix
detection of same access. Instead of breaking the group here
do not consider the duplicate.  Add comment explaining real fix.

* gfortran.dg/vect/pr83232.f90: New testcase.

From-SVN: r255307

Fix "central flowgraph" typo in machine desc docs

* doc/md.texi (Insn Splitting): Fix "central flowgraph" typo.

From-SVN: r255305

RTEMS/Ada: Fix some POSIX types

gcc/ada
* libgnarl/s-osinte__rtems.ads (pthread_cond_t): Use correct size and
alignment.
(pthread_mutex_t): Likewise.
(pthread_rwlock_t): Likewise.

From-SVN: r255302

Fix wrong code by arm_final_prescan with fp16 move instructions

ChangeLog entry are as follow:

*** gcc/ChangeLog ***

2017-12-01  Sudakshina Das  <sudi.das@arm.com>

* config/arm/vfp.md (*movhf_vfp_fp16): Add conds attribute.

*** gcc/testsuite/ChangeLog ***

2017-12-01  Sudakshina Das  <sudi.das@arm.com>

* gcc.target/arm/armv8_2-fp16-move-2.c: New test.

From-SVN: r255301

function.h (struct function): Remove cilk_frame_decl, is_cilk_function and calls_cilk_spawn fields.

* function.h (struct function): Remove cilk_frame_decl,
is_cilk_function and calls_cilk_spawn fields.
* tree-inline.h (struct copy_body_data): Remove remap_var_for_cilk
field.
* omp-simd-clone.c (simd_clone_clauses_extract): Don't clear
cilk_elemental field.
* cgraph.h (struct cgraph_simd_clone): Remove cilk_elemental field.
* target.def: Adjust comment.
* config/i386/i386.c (ix86_simd_clone_compute_vecsize_and_simdlen):
Don't test cilk_elemental.
c-family/
* c-attribs.c (c_common_attribute_table): Remove "cilk simd function"
attribute.
(handle_simd_attribute): Don't check for "cilk simd function"
attribute.  Reindent, formatting changes.
cp/
* parser.c (cp_parser_new): Don't clear cilk_simd_fn_info.
(parsing_nsdmi): Adjust comment.
(cp_parser_omp_for_loop_init): Likewise.
* parser.h (struct cp_omp_declare_simd_data): Adjust comment.
(struct cp_parser): Remove cilk_simd_fn_info field.
* cp-tree.h (cilk_valid_spawn): Remove.

From-SVN: r255300

re PR tree-optimization/83233 (gimple-ssa-store-merging.c:183:3: warning: unnamed type with no linkage used to declare variable with linkage)

PR tree-optimization/83233
* gimple-ssa-store-merging.c (nop_stats, bswap_stats): Use
bswap_stat name for the struct.

From-SVN: r255299

re PR c/79153 (-Wimplicit-fallthrough missed warning)

PR c/79153
* tree.h (SWITCH_BREAK_LABEL_P): Define.
* gimplify.c (collect_fallthrough_labels): Handle GIMPLE_BIND
starting with a GIMPLE_SWITCH and ending with GIMPLE_LABEL with
SWITCH_BREAK_LABEL_P set on the label.
(gimplify_switch_expr): Set SWITCH_BREAK_LABEL_P on the label
added for default case if it was missing and not all cases covered.
Wrap GIMPLE_SWITCH and the switch_body_seq into a GIMPLE_BIND if
switch_body_seq ends with a GIMPLE_LABEL with SWITCH_BREAK_LABEL_P
set on the label.
* tree-chrec.c (evolution_function_is_univariate_p): Add return true;
to avoid -Wimplicit-fallthrough warning.
* config/i386/i386.c (ix86_expand_special_args_builtin): Add
FALLTHRU comment to avoid -Wimplicit-fallthrough warning.
c/
* c-parser.c: Include tree-iterator.h.
(c_parser_switch_statement): Emit LABEL_EXPR for the break label
into SWITCH_BODY instead of after it and set SWITCH_BREAK_LABEL_P
on it.
cp/
* cp-gimplify.c (genericize_switch_stmt): Emit LABEL_EXPR for the
break label into SWITCH_BODY instead of after it and set
SWITCH_BREAK_LABEL_P on it.
* parser.c (cp_parser_objc_expression): Add FALLTHRU comment to avoid
-Wimplicit-fallthrough warning.
fortran/
* match.c (gfc_match): Add FALLTHRU comment to avoid
-Wimplicit-fallthrough warning.
testsuite/
* c-c++-common/Wimplicit-fallthrough-7.c: Adjust expected warning
line.
* c-c++-common/Wimplicit-fallthrough-36.c: New test.

From-SVN: r255298

re PR tree-optimization/83221 (qsort comparator not anti-commutative: -2147483648, -2147483648)

PR tree-optimization/83221
* tree-ssa-reassoc.c (sort_by_operand_rank): Shift bb_rank
down by 16.
(init_reassoc): Formatting fix.

From-SVN: r255297

re PR sanitizer/81275 (-fsanitize=thread produce incorrect -Wreturn-type warning)

PR sanitizer/81275
* tree-cfg.c (group_case_labels_stmt): Don't optimize away
C++ FE implicitly added __builtin_unreachable () until -Wreturn-type
is diagnosed.

* c-c++-common/tsan/pr81275.c: Remove dg-skip-if.

From-SVN: r255296

re PR sanitizer/83219 (c-c++-common/ubsan/unreachable-2.c fails starting with r255201)

PR sanitizer/83219
* tree-cfg.c: Include asan.h.
(gimple_seq_unreachable_p): Return false for -fsanitize=unreachable.

From-SVN: r255295

re PR fortran/83224 (creating character array from elements shorter than declared does not pad with whitespace properly and aborts)

2017-12-01  Thomas Koenig  <tkoenig@gcc.gnu.org>

PR fortran/83224
* frontend-passes.c (realloc_string_callback): Handle
case for which the RHS is an array expression.

2017-12-01  Thomas Koenig  <tkoenig@gcc.gnu.org>

PR fortran/83224
* gfortran.dg/dependency_50.f90: New test.

From-SVN: r255294

Fix movdi_internal to return MODE_TI with AVX512

gcc/
* config/i386/i386.md: Fix AVX512 register width in AVX512 instruction.

From-SVN: r255293

Replace __popcntq with two _popcnt32.

gcc/testsuite/
* gcc.target/i386/avx512f-vpcompressb-2.c: Fix popcnt for 64b case.

From-SVN: r255292

Fix wrong instruction vpcmpeqd generation

gcc/
* config/i386/i386.c (standard_sse_constant_opcode): Fix wrong form for
vpcmpeqd instruction.

From-SVN: r255291

Fix registers type for MODE_TI

gcc/
* config/i386/i386.c (standard_sse_constant_opcode): Fix
registers type for 128bit mode.

From-SVN: r255290

_mulkc3.c (__mulkc3): Add forward declaration.

2017-11-30  Michael Meissner  <meissner@linux.vnet.ibm.com>

* config/rs6000/_mulkc3.c (__mulkc3): Add forward declaration.
* config/rs6000/_divkc3.c (__divkc3): Likewise.

From-SVN: r255289

Daily bump.

From-SVN: r255288

re PR c/83222 (Inconsistent "initializer element is not constant" error)

PR c/83222
* c-tree.h (decl_constant_value_1): Declare.
* c-typeck.c (decl_constant_value_1): New function.
(decl_constant_value): Use it.
* c-fold.c (c_fully_fold_internal): If in_init, use
decl_constant_value_1 instead of decl_constant_value.

* gcc.c-torture/compile/pr83222.c: New test.

From-SVN: r255285

spellcheck-tree.c (test_find_closest_identifier): Use ; instead of ;;.

* spellcheck-tree.c (test_find_closest_identifier): Use ; instead
of ;;.
* gengtype-state.c (read_state_pair): Likewise.
* gimple-fold.c (gimple_fold_builtin_string_compare): Likewise.
* sel-sched-dump.c (dump_insn_rtx_1): Likewise.
* ipa-cp.c (intersect_aggregates_with_edge): Likewise.
* ifcvt.c (noce_try_store_flag_constants): Likewise.
* tree-ssa-ccp.c (ccp_finalize): Likewise.
* omp-grid.c (grid_process_kernel_body_copy): Likewise.
* builtins.c (fold_builtin_3): Likewise.
* graphite-scop-detection.c
(scop_detection::stmt_has_simple_data_refs_p): Likewise.
* hsa-gen.c (hsa_function_representation::hsa_function_representation):
Likewise.
c/
* c-parser.c (c_parser_postfix_expression): Use ; instead of ;;.
jit/
* jit-recording.c
(recording::memento_of_new_rvalue_from_const <long>::write_reproducer):
Use ; instead of ;;.
lto/
* lto.c (create_subid_section_table): Use ; instead of ;;.
objc/
* objc-next-runtime-abi-01.c (generate_dispatch_table): Use ; instead
of ;;.

From-SVN: r255284

re PR sanitizer/81697 (Incorrect ASan global variables alignment on arm)

gcc/

2017-11-30  Maxim Ostapenko  <m.ostapenko@samsung.com>

PR sanitizer/81697
* asan.c (asan_protect_global): Add new ignore_decl_rtl_set_p
parameter. Return true if ignore_decl_rtl_set_p is true and other
conditions are satisfied.
* asan.h (asan_protect_global): Add new parameter.
* varasm.c (categorize_decl_for_section): Pass true as second parameter
to asan_protect_global calls.

gcc/testsuite/

2017-11-30  Maxim Ostapenko  <m.ostapenko@samsung.com>

PR sanitizer/81697
* c-c++-common/asan/pr81697.c: New test.

From-SVN: r255283

re PR libgcc/83112 (Silence warnings from PowerPC libgcc float128-ifunc.c compilation)

2017-11-30  Michael Meissner  <meissner@linux.vnet.ibm.com>

PR libgcc/83112
* config/rs6000/float128-ifunc.c (__addkf3_resolve): Use the
correct type for all ifunc resolvers to silence -Wattribute-alias
warnings.  Eliminate the forward declaration of the resolver
functions which is no longer needed.
(__subkf3_resolve): Likewise.
(__mulkf3_resolve): Likewise.
(__divkf3_resolve): Likewise.
(__negkf2_resolve): Likewise.
(__eqkf2_resolve): Likewise.
(__nekf2_resolve): Likewise.
(__gekf2_resolve): Likewise.
(__gtkf2_resolve): Likewise.
(__lekf2_resolve): Likewise.
(__ltkf2_resolve): Likewise.
(__unordkf2_resolve): Likewise.
(__extendsfkf2_resolve): Likewise.
(__extenddfkf2_resolve): Likewise.
(__trunckfsf2_resolve): Likewise.
(__trunckfdf2_resolve): Likewise.
(__fixkfsi_resolve): Likewise.
(__fixkfdi_resolve): Likewise.
(__fixunskfsi_resolve): Likewise.
(__fixunskfdi_resolve): Likewise.
(__floatsikf_resolve): Likewise.
(__floatdikf_resolve): Likewise.
(__floatunsikf_resolve): Likewise.
(__floatundikf_resolve): Likewise.
(__extendkftf2_resolve): Likewise.
(__trunctfkf2_resolve): Likewise.

PR libgcc/83103
* config/rs6000/quad-float128.h (TF): Don't define if long double
is IEEE 128-bit floating point.
(TCtype): Define as either TCmode or KCmode, depending on whether
long double is IEEE 128-bit floating point.
(__mulkc3_sw): Add declarations for software/hardware versions of
complex multiply/divide.
(__divkc3_sw): Likewise.
(__mulkc3_hw): Likewise.
(__divkc3_hw): Likewise.
* config/rs6000/_mulkc3.c (_mulkc3): If we are building ifunc
handlers to switch between using software emulation and hardware
float128 instructions, build the complex multiply/divide functions
for both software and hardware support.
* config/rs6000/_divkc3.c (_divkc3): Likewise.
* config/rs6000/float128-ifunc.c (__mulkc3_resolve): Likewise.
(__divkc3_resolve): Likewise.
(__mulkc3): Likewise.
(__divkc3): Likewise.
* config/rs6000/t-float128-hw (fp128_hardfp_src): Likewise.
(fp128_hw_src): Likewise.
(fp128_hw_static_obj): Likewise.
(fp128_hw_shared_obj): Likewise.
(_mulkc3-hw.c): Create _mulkc3-hw.c and _divkc3-hw.c from
_mulkc3.c and _divkc3.c, changing the function name.
(_divkc3-hw.c): Likewise.
* config/rs6000/t-float128 (clean-float128): Delete _mulkc3-hw.c
and _divkc3-hw.c.

From-SVN: r255282

Cleanup riscv option docs.

gcc/
* doc/invoke.texi (RISC-V Options): Delete nonexistent -mmemcpy and
-mno-memcpy options.  For -mplt, -mfdiv, -mdiv, -msave-restore, and
-mstrict-align, add info on default value.  Delete redundant lines for
-mabi.  Add missing -mexplicit-relocs docs.

From-SVN: r255280

PR c++/82219 - bogus -Wignored-qualifiers with template

* pt.c (tsubst_copy_and_build) [STATIC_CAST_EXPR]: Suppress
-Wignored-qualifiers.

From-SVN: r255279

PR libstdc++/65927 Fix -Wunused-but-set-parameter warning

PR libstdc++/65927
* include/bits/locale_facets_nonio.h (__timepunct::_M_am_pm_format):
Remove statement with no effect.

From-SVN: r255278

Move assertion from to_address to __to_address

2017-11-30  Glen Joseph Fernandes  <glenjofe@gmail.com>

* include/bits/ptr_traits.h (__to_address, to_address): Move static
assertion.
* testsuite/20_util/to_address/1_neg.cc: New test.

From-SVN: r255277

[ARC] Add trap instruction.

2017-11-07  Claudiu Zissulescu  <claziss@synopsys.com>

* config/arc/arc.md (trap): New pattern.

From-SVN: r255276

[ARC] Don't allow the last ZOL insn to be in a delay slot.

The ARC ZOL implementation doesn't allow the last instruction to be a
control instruction or part of a delay slot.  Thus, we add a note to
the last ZOL instruction which will prevent it to finish into a delay
slot.

2017-10-20  Claudiu Zissulescu  <claziss@synopsys.com>

* config/arc/arc.c (hwloop_optimize): Prevent the last
        ZOL instruction to end into a delay slot.
* config/arc/arc.md (cond_delay_insn): Check if the instruction
can be placed into a delay slot against reg_note.
(in_delay_slot): Likewise.

testsuite/
2017-10-20  Claudiu Zissulescu  <claziss@synopsys.com>

* gcc.target/arc/loop-3.c: New test.
* gcc.target/arc/loop-4.c: Likewise.

[FIX][ZOL] fix checking for jumps

From-SVN: r255275

[ARC] Update uses for hw-loop labels.

Make sure we mark the hw-loop labels as beeing used.

gcc/
2017-09-19  Claudiu Zissulescu  <claziss@synopsys.com>

* config/arc/arc.c (hwloop_optimize): Update hw-loop's end/start
labels number of usages.

gcc/testsuite
2017-09-19  Claudiu Zissulescu  <claziss@synopsys.com>

* gcc.target/arc/loop-2.cpp: New test.

From-SVN: r255274

[ARC] Use TARGET_CANNOT_SUBSTITUTE_MEM_EQUIV.

Sometimes the memory equivalent is not valid due to a large offset.
For example replacing the ap register with its fp/sp-equivalent during
LRA step. To solve this we introduced TARGET_CANNOT_SUBSTITUTE_MEM_EQUIV.

gcc/
2017-08-08  Claudiu Zissulescu  <claziss@synopsys.com>

* config/arc/arc.c (arc_cannot_substitute_mem_equiv_p): New function.
(TARGET_CANNOT_SUBSTITUTE_MEM_EQUIV_P): Define.

gcc/testsuite
2017-08-08  Claudiu Zissulescu  <claziss@synopsys.com>

* gcc.target/arc/lra-1.c: New test.

From-SVN: r255273

fold-vec-abs-char-fwrapv.c: Add xxspltib insn to expected output.

[testsuite]

2017-11-29  Will Schmidt  <will_schmidt@vnet.ibm.com>

* gcc.target/powerpc/fold-vec-abs-char-fwrapv.c: Add xxspltib insn
to expected output.
* gcc.target/powerpc/fold-vec-abs-char.c: Add xxspltib insn
to expected output.

From-SVN: r255272

PR libstdc++/83226 avoid forming pointer-to-reference type

PR libstdc++/83226
* include/bits/node_handle.h (_Node_handle::__pointer): Avoid forming
pointer-to-reference types.
* testsuite/23_containers/map/modifiers/insert/83226.cc: New test.

From-SVN: r255271

Remove inv_list.

gcc/c-family/
* c-common.h (inv_list): Remove.

From-SVN: r255270

re PR target/83210 (__builtin_mul_overflow() generates suboptimal code when exactly one argument is the constant 2)

PR target/83210
* internal-fn.c (expand_mul_overflow): Optimize unsigned
multiplication by power of 2 constant into two shifts + comparison.

* gcc.target/i386/pr83210.c: New test.

From-SVN: r255269

re PR target/81616 (Update -mtune=generic for the current Intel and AMD processors)

PR target/81616
* x86-tnue-costs.h (generic_cost): Revise for modern CPUs
* gcc.target/i386/l_fma_double_1.c: Update count of fma instructions.
* gcc.target/i386/l_fma_double_2.c: Update count of fma instructions.
* gcc.target/i386/l_fma_double_3.c: Update count of fma instructions.
* gcc.target/i386/l_fma_double_4.c: Update count of fma instructions.
* gcc.target/i386/l_fma_double_5.c: Update count of fma instructions.
* gcc.target/i386/l_fma_double_6.c: Update count of fma instructions.
* gcc.target/i386/l_fma_float_1.c: Update count of fma instructions.
* gcc.target/i386/l_fma_float_2.c: Update count of fma instructions.
* gcc.target/i386/l_fma_float_3.c: Update count of fma instructions.
* gcc.target/i386/l_fma_float_4.c: Update count of fma instructions.
* gcc.target/i386/l_fma_float_5.c: Update count of fma instructions.
* gcc.target/i386/l_fma_float_6.c: Update count of fma instructions.

From-SVN: r255268

re PR tree-optimization/83202 (Try joining operations on consecutive array elements during tree vectorization)

2017-11-30  Richard Biener  <rguenther@suse.de>

PR tree-optimization/83202
* tree-ssa-loop-ivcanon.c (try_unroll_loop_completely): Add
allow_peel argument and guard peeling.
(canonicalize_loop_induction_variables): Likewise.
(canonicalize_induction_variables): Pass false.
(tree_unroll_loops_completely_1): Pass unroll_outer to disallow
peeling from cunrolli.

* gcc.dg/vect/pr83202-1.c: New testcase.
* gcc.dg/tree-ssa/pr61743-1.c: Adjust.

From-SVN: r255267

compiler: don't make map zero value constant

    The map zero value is a common symbol, and it doesn't really make
    sense to have a constant common symbol. Current GCC has started to
    reject this case, probably as part of the fix for PR 83100.

    Reviewed-on: https://go-review.googlesource.com/80877

From-SVN: r255266