Daily bump.

From-SVN: r268110

Limit AA walking in IPA summary generation

2019-01-20  Martin Jambor  <mjambor@suse.cz>

PR ipa/87615
* ipa-prop.h (struct ipa_func_body_info): Replaced field aa_walked
with aa_walk_budget.
* cgraph.h (ipa_polymorphic_call_context::get_dynamic_type): Add
aa_walk_budget_p parameter.
* ipa-fnsummary.c (unmodified_parm_1): New parameter fbi.  Limit AA
walk.  Updated all callers.
(unmodified_parm): New parameter fbi, pass it to unmodified_parm_1.
(eliminated_by_inlining_prob): New parameter fbi, pass it on to
unmodified_parm.
(will_be_nonconstant_expr_predicate): New parameter fbi, removed
parameter info.  Extract info from fbi.  Pass fbi to recursive calls
and to unmodified_parm.
(phi_result_unknown_predicate): New parameter fbi, removed parameter
info, updated call to will_be_nonconstant_expr_predicate.
(param_change_prob): New parameter fbi, limit AA walking.
(analyze_function_body): Initialize aa_walk_budget in fbi.  Update
calls to various above functions.
* ipa-polymorphic-call.c (get_dynamic_type): Add aa_walk_budget_p
parameter.  Use it to limit AA walking.
* ipa-prop.c (detect_type_change_from_memory_writes): New parameter
fbi, limit AA walk.
(detect_type_change): New parameter fbi, pass it on to
detect_type_change_from_memory_writes.
(detect_type_change_ssa): Likewise.
(aa_overwalked): Removed.
(parm_preserved_before_stmt_p): Assume fbi is never NULL, stream line
accordingly, adjust to the neew AA limiting scheme.
(parm_ref_data_preserved_p): Likewise.
(ipa_compute_jump_functions_for_edge): Adjust call to
get_dynamic_type.
(ipa_analyze_call_uses): Likewise.
(ipa_analyze_virtual_call_uses): Pass fbi to detect_type_change_ssa.
(ipa_analyze_node): Initialize aa_walk_budget.
(ipcp_transform_function): Likewise.
* tree-ssa-sccvn.c (eliminate_dom_walker::eliminate_stmt): Update call
to get_dynamic_type.

From-SVN: r268107

[D] Don't predefine GNU_StackGrowsDown on every target.

Every back-end defines STACK_GROWS_DOWNWARD, the condition should
instead be  checking whether it was defined as 1 or 0.

gcc/d/ChangeLog:

2019-01-20  Iain Buclaw  <ibuclaw@gdcproject.org>

* d-builtins.cc (d_init_versions): Check value of
STACK_GROWS_DOWNWARD.

From-SVN: r268106

* allocfail.c (main): Increase portability of printf statement.

From-SVN: r268105

[D] Fix identity comparison for structs with complex float fields.

gcc/d/ChangeLog:

2019-01-20  Iain Buclaw  <ibuclaw@gdcproject.org>

* d-codegen.cc (identity_compare_p): Return false if seen built-in
type with padding.
(build_float_identity): Moved here from expr.cc.
(lower_struct_comparison): Handle real and complex types.
* d-tree.h (build_float_identity): New.
* expr.cc (build_float_identity): Move to d-codegen.cc.

gcc/testsuite/ChangeLog:

2019-01-20  Iain Buclaw  <ibuclaw@gdcproject.org>

* gdc.dg/runnable.d: Add more tests for comparing complex types.

From-SVN: r268104

[D] Fix IdentityExp comparison for complex floats.

gcc/d/ChangeLog:

2019-01-20  Johannes Pfau  <johannespfau@gmail.com>

* expr.cc (build_float_identity): New function.
(ExprVisitor::visit(IdentityExp)): Add support for complex types.

gcc/testsuite/ChangeLog:

2019-01-20  Johannes Pfau  <johannespfau@gmail.com>

* gdc.dg/runnable.d: Add tests for comparing complex types.

From-SVN: r268103

vector long long and vector double require vsx support.

gcc/testsuite/ChangeLog

    * gcc.target/powerpc/altivec_vld_vst_addr.c: Remove, split into
    altivec_vld_vst_addr-1.c and altivec_vld_vst_addr-2.c.
    * gcc.target/powerpc/altivec_vld_vst_addr-1.c: New test.
    * gcc.target/powerpc/altivec_vld_vst_addr-2.c: Ditto.

From-SVN: r268102

Daily bump.

From-SVN: r268101

[multiple changes]

2019-01-19  Dominique d'Humieres  <dominiq@gcc.gnu.org>

PR fortran/37835
* resolve.c (resolve_types): Add !flag_automatic.
* symbol.c (gfc_add_save): Silence warnings.

2019-01-18  Dominique d'Humieres  <dominiq@gcc.gnu.org>

PR fortran/37835
* gfortran.dg/no-automatic.f90: New test.

From-SVN: r268098

re PR fortran/77960 (ICE in gfc_conv_ss_startstride, at fortran/trans-array.c:3966)

2019-01-19  Steven G. Kargl  <kargl@gcc.gnu.org>

PR fortran/77960
* io.c (match_io_element): input-item cannot be an external function.

2019-01-19  Steven G. Kargl  <kargl@gcc.gnu.org>

PR fortran/77960
* gfortran.dg/pr77960.f90: New test.

From-SVN: r268097

re PR fortran/56789 (Handling of contiguous dummy arguments)

2018-01-19  Thomas Koenig  <tkoenig@gcc.gnu.org>
Paul Thomas  <pault@gcc.gnu.org>

PR fortran/56789
* trans-expr.c (gfc_conv_procedure_call): Call
gfc_conv_subref_array_arg if the formal arg is contiguous
and the actual arg may not be.

2018-01-19  Thomas Koenig  <tkoenig@gcc.gnu.org>
Paul Thomas  <pault@gcc.gnu.org>

PR fortran/56789
* gfortran.dg/contiguous_3.f90: Make code compilant.  Remove
scan-tree tests that fail with patch.
* gfortran.dg/contiguous_8.f90: New test.

Co-Authored-By: Paul Thomas <pault@gcc.gnu.org>
From-SVN: r268096

aarch64.c (aarch64_stack_protect_guard): Move outside of #if CHECKING_P code.

* config/aarch64/aarch64.c (aarch64_stack_protect_guard): Move
outside of #if CHECKING_P code.

From-SVN: r268095

Claw back some of the code size regression in 548.exchange2_r

This patch tries harder to detect cases in which the inner dimension
of an array access is invariant, such as:

     x(i, :) = 100

It fixes some of the code size regression in 548.exchange2_r, with
size improving by 5% compared to before the patch.  Of the two other
SPEC 2017 tests affected by loop versioning, 554.roms_r improved by a
trivial amount (0.3%) and 549.fotonik3d_r didn't change.  All three
results are with -Ofast -flto.

2019-01-19  Richard Sandiford  <richard.sandiford@arm.com>

gcc/
* gimple-loop-versioning.cc (loop_versioning::dump_inner_likelihood):
New function, split out from...
(loop_versioning::analyze_stride): ...here.
(loop_versioning::find_per_loop_multiplication): Use gassign.
(loop_versioning::analyze_term_using_scevs): Return a success code.
(loop_versioning::analyze_arbitrary_term): New function.
(loop_versioning::analyze_address_fragment): Use
analyze_arbitrary_term if all else fails.

gcc/testsuite/
* gfortran.dg/loop_versioning_1.f90: Bump the number of identified
inner strides.
* gfortran.dg/loop_versioning_9.f90: New test.
* gfortran.dg/loop_versioning_10.f90: Likewise.

From-SVN: r268093

re PR fortran/88871 (ICE segmentation fault in f951)

2019-01-17  Thomas Koenig  <tkoenig@gcc.gnu.org>

PR fortran/88871
* resolve.c (resolve_ref): Fix logic for removal of
reference.

From-SVN: r268092

re PR fortran/88902 (ICE: Segmentation fault (in DFS::DFS_write_tree_body))

PR fortran/88902
* trans-decl.c (gfc_get_symbol_decl): Don't add length to function
or parent function if it has been added there already.

* gfortran.dg/pr88902.f90: New test.

From-SVN: r268091

attr-nonstring-3.c: Remove an xfail.

gcc/testsuite/ChangeLog:

* c-c++-common/attr-nonstring-3.c: Remove an xfail.

From-SVN: r268090

Daily bump.

From-SVN: r268089

PR libstdc++/88782 avoid ODR problems in std::make_shared

The old version of _Sp_counted_ptr_inplace::_M_get_deleter (up to GCC
8.2.0) expects to be passed a real std::typeinfo object, so mixing that
with the new definition of the __shared_ptr constructor (which always
passes the fake tag) leads to accessing the fake object as a real
std::typeinfo. Instead of trying to make it safe to mix the old and new
definitions, just stop using that function. By passing a reference to
__shared_ptr::_M_ptr to the __shared_count constructor it can be set
directly, without needing to obtain the pointer via the _M_get_deleter
back-channel. This avoids a virtual dispatch (which fixes PR 87514).

This means that code built against new libstdc++ headers doesn't use
_M_get_deleter at all, and so make_shared works the same whether RTTI is
enabled or not.

Also change _M_get_deleter so that it checks for a real type_info object
even when RTTI is disabled, by calling a library function. Unless
libstdc++ itself is built without RTTI that library function will be
able to test if it's the right type_info. This means the new definition
of _M_get_deleter can handle both the fake type_info tag and a real
type_info object, even if built without RTTI.

If linking to objects built against older versions of libstdc++ then if
all objects use -frtti or all use -fno-rtti, then the caller of
_M_get_deleter and the definition of _M_get_deleter will be consistent
and it will work. If mixing -frtti with -fno-rtti it can still fail if
the linker picks an old definition of _M_get_deleter and an old
__shared_ptr constructor that are incompatible. In that some or all
objects might need to be recompiled.

PR libstdc++/87514
PR libstdc++/87520
PR libstdc++/88782
* config/abi/pre/gnu.ver (GLIBCXX_3.4.26): Export new symbol.
* include/bits/shared_ptr.h
(shared_ptr(_Sp_make_shared_tag, const Alloc&, Args&&...))
(allocate_shared): Change to use new tag type.
* include/bits/shared_ptr_base.h (_Sp_make_shared_tag::_S_eq):
Declare new member function.
(_Sp_alloc_shared_tag): Define new type.
(_Sp_counted_ptr_inplace): Declare __shared_count<_Lp> as a friend.
(_Sp_counted_ptr_inplace::_M_get_deleter) [!__cpp_rtti]: Use
_Sp_make_shared_tag::_S_eq to check type_info.
(__shared_count(Ptr, Deleter),__shared_count(Ptr, Deleter, Alloc)):
Constrain to prevent being called with _Sp_alloc_shared_tag.
(__shared_count(_Sp_make_shared_tag, const _Alloc&, Args&&...)):
Replace constructor with ...
(__shared_count(Tp*&, _Sp_alloc_shared_tag<_Alloc>, Args&&...)): Use
reference parameter so address of the new object can be returned to
the caller. Obtain the allocator from the tag type.
(__shared_ptr(_Sp_make_shared_tag, const Alloc&, Args&&...)): Replace
constructor with ...
(__shared_ptr(_Sp_alloc_shared_tag<Alloc>, Args&&...)): Pass _M_ptr
to the __shared_count constructor.
(__allocate_shared): Change to use new tag type.
* src/c++11/shared_ptr.cc (_Sp_make_shared_tag::_S_eq): Define.

From-SVN: r268086

PR c++/88875 - error with explicit list constructor.

In my patch for CWG issue 2267, I changed reference_binding to clear
CONSTRUCTOR_IS_DIRECT_INIT on the argument init-list.  But that breaks if
there's another candidate for which CONSTRUCTOR_IS_DIRECT_INIT is correct.
So instead, let's encode in the conversion that we want to override the
flag.

* call.c (reference_binding): Don't modify EXPR.  Set
need_temporary_p on the ck_user conversion for a temporary.
(convert_like_real): Check it.

From-SVN: r268085

libgo: update to Go1.12beta2

    Reviewed-on: https://go-review.googlesource.com/c/158019

gotools/:
* Makefile.am (go_cmd_vet_files): Update for Go1.12beta2 release.
(GOTOOLS_TEST_TIMEOUT): Increase to 600.
(check-runtime): Export LD_LIBRARY_PATH before computing GOARCH
and GOOS.
(check-vet): Copy golang.org/x/tools into check-vet-dir.
* Makefile.in: Regenerate.

gcc/testsuite/:
* go.go-torture/execute/names-1.go: Stop using debug/xcoff, which
is no longer externally visible.

From-SVN: r268084

rs6000: Fix *movsi_from_df (PR88892)

The memory store instructions (stfs[u][x], stxssp[x]) can result in
garbage if the value to be stored isn't already a valid single
precision floating point number.  So we cannot use this here.

PR target/88892
* config/rs6000/rs6000.md (*movsi_from_df): Allow only register
operands.

From-SVN: r268083

re PR libbacktrace/88890 (libbacktrace on 32-bit system with _FILE_OFFSET_BITS == 64)

PR libbacktrace/88890
* mmapio.c (backtrace_get_view): Change size parameter to
uint64_t.  Check that value fits in size_t.
* read.c (backtrace_get_view): Likewise.
* internal.h (backtrace_get_view): Update declaration.
* elf.c (elf_add): Pass shstrhdr->sh_size to backtrace_get_view.

From-SVN: r268082

re PR c++/86926 (ICE for a recursive generic lambda)

PR c++/86926
* g++.dg/cpp1z/constexpr-lambda23.C: New test.

From-SVN: r268080

Update PR middle-end/88587 tests

It is wrong to use -m32 in dg-options.  { target ia32 } should be used
instead.  Also add -fno-pic to g++.target/i386/pr88587.C since it is
invalid with PIC.

PR middle-end/88587
* g++.target/i386/pr88587.C (dg-do): Add { target ia32 }.
(dg-options): Replace -m32 with -fno-pic.
* gcc.target/i386/mvc13.c (dg-do): Add { target ia32 }.
(dg-options): Remove -m32.

From-SVN: r268079

re PR tree-optimization/88903 (wrong-code with SLP vectorized shift)

2019-01-18  Richard Biener  <rguenther@suse.de>

PR tree-optimization/88903
* tree-vect-stmts.c (vectorizable_shift): Verify we see all
scalar stmts a SLP shift amount is composed of when detecting
shifts by scalars.

* gcc.dg/vect/pr88903-1.c: New testcase.
* gcc.dg/vect/pr88903-2.c: Likewise.

From-SVN: r268076

c-family: Update unaligned adress of packed member check

Check unaligned pointer conversion and strip NOPS.

gcc/c-family/

PR c/51628
PR c/88664
* c-common.h (warn_for_address_or_pointer_of_packed_member):
Remove the boolean argument.
* c-warn.c (check_address_of_packed_member): Renamed to ...
(check_address_or_pointer_of_packed_member): This.  Also
warn pointer conversion.
(check_and_warn_address_of_packed_member): Renamed to ...
(check_and_warn_address_or_pointer_of_packed_member): This.
Also warn pointer conversion.
(warn_for_address_or_pointer_of_packed_member): Remove the
boolean argument.  Don't check pointer conversion here.

gcc/c

PR c/51628
PR c/88664
* c-typeck.c (convert_for_assignment): Upate the
warn_for_address_or_pointer_of_packed_member call.

gcc/cp

PR c/51628
PR c/88664
* call.c (convert_for_arg_passing): Upate the
warn_for_address_or_pointer_of_packed_member call.
* typeck.c (convert_for_assignment): Likewise.

gcc/testsuite/

PR c/51628
PR c/88664
* c-c++-common/pr51628-33.c: New test.
* c-c++-common/pr51628-35.c: New test.
* c-c++-common/pr88664-1.c: Likewise.
* c-c++-common/pr88664-2.c: Likewise.
* gcc.dg/pr51628-34.c: Likewise.

From-SVN: r268075

PR target/88799 Add +mp and +sec extensions to ARMv7-a

Most armv7-a implementations support a number of basic extensions to
the architecture which are not particularly important to the compiler,
but can matter if code contains inline assembly.  This patch adds
support for these extensions, based on the capabilities that GAS
already provides for the appropriate CPUs.  For the purposes of
multilib selection we ignore these extensions entirely and map the
extended architecture versions down to the base versions we have
already support for.

gcc:
PR target/88799
* config/arm/arm-cpus.in (mp): New feature.
(sec): New feature.
(fgroup ARMv7ve): Add mp and sec features.
(arch armv7-a): Add options to allow mp and sec extensions.
(cpu generic-armv7-a): Add options to allow mp and sec extensions.
(cpu cortex-a5, cpu cortex-7, cpu cortex-a9): Add mp and sec
extenstions to the base architecture.
(cpu cortex-a8): Add sec extension to the base architecture.
(cpu marvell-pj4): Add mp and sec extensions to the base architecture.
* config/arm/t-aprofile (MULTILIB_MATCHES): Map all armv7-a arch
variants down to the base v7-a varaint.
* config/arm/t-multilib (v7_a_arch_variants): New variable.
* doc/invoke.texi (ARM Options): Add +mp and +sec to the list
of permitted extensions for -march=armv7-a and for
-mcpu=generic-armv7-a.

testsuite:
* gcc.target/arm/multilib.exp (config "aprofile"): Add tests for
mp and sec extensions to armv7-a.

From-SVN: r268072

Bump version of __gcov_indirect_call_profiler function as there was ABI change.

2019-01-18  Martin Liska  <mliska@suse.cz>

* params.def: Fix comment.
* tree-profile.c (gimple_init_gcov_profiler): Bump function
name.
(gimple_gen_ic_func_profiler): Likewise.
2019-01-18  Martin Liska  <mliska@suse.cz>

* gcc.dg/no_profile_instrument_function-attr-1.c: Update
expected function name.
2019-01-18  Martin Liska  <mliska@suse.cz>

* libgcov-profiler.c (__gcov_indirect_call_profiler_v2): Rename
to ...
(__gcov_indirect_call_profiler_v3): ... this.
* libgcov.h (__gcov_indirect_call_profiler_v2): Likewise.
(__gcov_indirect_call_profiler_v3): Likewise.
* Makefile.in: Bump function name.

From-SVN: r268071

Update error message prefix in libgcov profiling.

2019-01-18  Martin Liska  <mliska@suse.cz>

* libgcov-driver.c (GCOV_PROF_PREFIX): Define.
(gcov_version): Use in gcov_error.
(merge_one_data): Likewise.
(dump_one_gcov): Likewise.

From-SVN: r268070

Describe better version mismatch in libgcov driver.

2019-01-18  Martin Liska  <mliska@suse.cz>

* libgcov-driver.c (gcov_version_string): New function.
(gcov_version): Convert version integer into string.

From-SVN: r268069

[AArch64] Add support for system register based stack protector canary access

2019-01-18  Ramana Radhakrishnan  <ramana.radhakrishnan@arm.com>

        * config/aarch64/aarch64-opts.h (enum stack_protector_guard): New
        * config/aarch64/aarch64.c (aarch64_override_options_internal): Handle
        and put in error checks for stack protector guard options.
        (aarch64_stack_protect_guard): New.
        (TARGET_STACK_PROTECT_GUARD): Define.
        * config/aarch64/aarch64.md (UNSPEC_SSP_SYSREG): New.
        (reg_stack_protect_address<mode>): New.
        (stack_protect_set): Adjust for SSP_GLOBAL.
        (stack_protect_test): Likewise.
        * config/aarch64/aarch64.opt (-mstack-protector-guard-reg): New.
        (-mstack-protector-guard): Likewise.
        (-mstack-protector-guard-offset): Likewise.

From-SVN: r268068

re PR tree-optimization/86214 (Strongly increased stack usage)

PR tree-optimization/86214
* tree-inline.h (struct copy_body_data): Add
add_clobbers_to_eh_landing_pads member.
* tree-inline.c (add_clobbers_to_eh_landing_pad): New function.
(copy_edges_for_bb): Call it if EH edge destination is <
id->add_clobbers_to_eh_landing_pads.  Fix a comment typo.
(expand_call_inline): Set id->add_clobbers_to_eh_landing_pads
if flag_stack_reuse != SR_NONE and clear it afterwards.

* g++.dg/opt/pr86214-1.C: New test.
* g++.dg/opt/pr86214-2.C: New test.

From-SVN: r268067

[ARM][testsuite] follow-up to PR target/52813 and target/11807 fix.

2019-01-18  Christophe Lyon  <christophe.lyon@linaro.org>

* gcc.target/arm/pr77904.c: Add dg-warning for sp clobber.

From-SVN: r268066

PR target/85596 Add --with-multilib-list doc for aarch64

2019-01-18  Christophe Lyon  <christophe.lyon@linaro.org>

PR target/85596
* doc/install.texi (with-multilib-list): Document for aarch64.

From-SVN: r268065

re PR target/88734 (AArch64's ACLE intrinsics give an ICE instead of compile error when option mismatch.)

PR target/88734
* config/arm/arm_neon.h: Fix #pragma GCC target syntax - replace
(("..."))) with ("...").

From-SVN: r268064

Reapply:

2018-12-15  Jakub Jelinek  <jakub@redhat.com>

PR target/88489
* gcc.target/i386/avx512vl-vfixupimmsd-2.c: New test.
* gcc.target/i386/avx512vl-vfixupimmss-2.c: New test.

From-SVN: r268063

Document atomic fetch and nand

Copy code example for fetch and nand from "Legacy __sync Built-in
Functions for Atomic Memory Access" to "Built-in Functions for Memory
Model Aware Atomic Operations".

gcc/

* doc/extend.texi (Built-in Functions for Memory Model Aware
Atomic Operations): Document atomic fetch and nand.

From-SVN: r268062

RTEMS/Ada: Remove RTEMS quirk

The new network stack (libbsd) supports IPv6.

gcc/ada

* s-oscons-tmplt.c (AF_INET6): Do not undefine for RTEMS.

From-SVN: r268061

Reset proper type on vector types (PR middle-end/88587).

2019-01-18  Martin Liska  <mliska@suse.cz>
    Richard Biener  <rguenther@suse.de>

PR middle-end/88587
* cgraph.h (create_version_clone_with_body): Add new argument
with attributes.
* cgraphclones.c (cgraph_node::create_version_clone): Add
DECL_ATTRIBUTES to a newly created decl.  And call
valid_attribute_p so that proper cl_target_optimization_node
is set for the newly created declaration.
* multiple_target.c (create_target_clone): Set DECL_ATTRIBUTES
for declaration.
(expand_target_clones): Do not call valid_attribute_p, it must
be already done.
* tree-inline.c (copy_decl_for_dup_finish): Reset mode for
vector types.
2019-01-18  Martin Liska  <mliska@suse.cz>

PR middle-end/88587
* g++.target/i386/pr88587.C: New test.
* gcc.target/i386/mvc13.c: New test.

Co-Authored-By: Richard Biener <rguenther@suse.de>
From-SVN: r268060

Add self to MAINTAINERS

2019-01-18  Li Jia He  <helijia@gcc.gnu.org>

    * MAINTAINERS (Write After Approval): Add myself.

From-SVN: r268059

PR c++/86205 - ICE with ?: of throw and template-id.

My patch for 64372 removed a bogus lvalue-rvalue conversion for one arm of a
?: expression where the other arm is a throw.  But we still need to require
any overload to be resolved, even though we aren't getting that from
decay_conversion anymore.

* pt.c (resolve_nondeduced_context_or_error): Split out from...
* typeck.c (decay_conversion): ...here.
* call.c (build_conditional_expr_1): Use it.

From-SVN: r268058

re PR go/88202 (FAIL: runtime/pprof)

PR go/88202
    runtime: in sigprof, skip to sigtrampgo if we don't find sigtramp

    Fixes https://gcc.gnu.org/PR88202

    Reviewed-on: https://go-review.googlesource.com/c/158218

From-SVN: r268057

libphobos: Add Fiber/Thread support for StackGrowsUp.

The StackGrowsDown version being turned off for hppa targets.

After other fixes in the compiler, this allows core.thread unittests to
all pass, as well as the garbage collector to work correctly.

Backported from upstream druntime 2.084.

Reviewed-on: https://github.com/dlang/druntime/pull/2410

From-SVN: r268056

libphobos: Add platform bindings for hppa-linux-gnu.

Backported from upstream druntime 2.084

Reviewed-on: https://github.com/dlang/druntime/pull/2402

From-SVN: r268055

re PR fortran/88898 ([Regression 9] gomp is broken by r268045)

2018-01-17  Steve Ellcey  <sellcey@cavium.com>

PR fortran/88898
* gfortran.dg/gomp/declare-simd-2.f90: Add aarch64 target specifier to
warning checks.
* gfortran.dg/gomp/pr79154-1.f90: Ditto.
* gfortran.dg/gomp/pr83977.f90: Ditto.

From-SVN: r268054

Daily bump.

From-SVN: r268053

libphobos: Add OSX bindings for core.sys.posix.aio

Backported from upstream druntime 2.084

Reviewed-on: https://github.com/dlang/druntime/pull/2342

From-SVN: r268050

re PR target/88734 (AArch64's ACLE intrinsics give an ICE instead of compile error when option mismatch.)

PR target/88734
* config/aarch64/arm_neon.h: Fix #pragma GCC target syntax - replace
(("..."))) with ("...").  Use arch=armv8.2-a+sha3 instead of
arch=armv8.2-a+crypto for vsha512hq_u64 etc. intrinsics.

From-SVN: r268049

PR middle-end/88273 - [8/9 Regression] warning: 'memcpy' offset [-527, -529]

PR middle-end/88273 - [8/9 Regression] warning: 'memcpy' offset [-527, -529]
is out of the bounds [0, 16]

gcc/ChangeLog:

PR middle-end/88273
* gimple-ssa-warn-restrict.c (builtin_memref::extend_offset_range):
Handle anti-ranges the same as no range at all.

gcc/testsuite/ChangeLog:

PR middle-end/88273
* gcc.dg/Warray-bounds-38.c: New test.

From-SVN: r268048

Improve the C error for mismatched array string literal initialization.

* c-typeck.c (digest_init): Revised the error message produced for
ill-formed cases of array initialization with a string literal.
(error_init): Make variadic.

Co-Authored-By: Jason Merrill <jason@redhat.com>
From-SVN: r268047

PR c++/86740, ICE with constexpr if and nested generic lambdas.

When we partially instantiate the constexpr if, we walk through its body to
see what it uses from the enclosing local_specializations.  That walk was
overlooking the use of 'count' in the captures of the innermost lambda,
because we weren't walking into the capture list.

* tree.c (cp_walk_subtrees): Handle LAMBDA_EXPR.

From-SVN: r268046

pr60823-1.c: Add aarch64 specific warning checks and assembler scans.

2018-01-17  Steve Ellcey  <sellcey@cavium.com>

* c-c++-common/gomp/pr60823-1.c: Add aarch64 specific
warning checks and assembler scans.
* c-c++-common/gomp/pr60823-3.c: Ditto.
* c-c++-common/gomp/pr63328.c: Ditto.
* g++.dg/gomp/declare-simd-1.C: Ditto.
* g++.dg/gomp/declare-simd-3.C: Ditto.
* g++.dg/gomp/declare-simd-4.C: Ditto.
* g++.dg/gomp/declare-simd-7.C: Ditto.
* g++.dg/gomp/pr88182.C: Ditto.
* g++.dg/vect/simd-clone-7.cc: Ditto.
* gcc.dg/gomp/declare-simd-1.c: Ditto.
* gcc.dg/gomp/declare-simd-3.c: Ditto.
* gcc.dg/gomp/pr59669-2.c: Ditto.
* gcc.dg/gomp/pr87895-1.c: Ditto.
* gcc.dg/gomp/pr87895-2.c: Ditto.
* gcc.dg/gomp/simd-clones-2.c: Ditto.
* gfortran.dg/gomp/declare-simd-2.f90: Ditto.
* gfortran.dg/gomp/pr79154-1.f90: Ditto.
* gfortran.dg/gomp/pr83977.f90: Ditto.

From-SVN: r268045

aarch64.c (cgraph.h): New include.

2018-01-17  Steve Ellcey  <sellcey@cavium.com>

* config/aarch64/aarch64.c (cgraph.h): New include.
(intl.h): New include.
(supported_simd_type): New function.
(currently_supported_simd_type): Ditto.
(aarch64_simd_clone_compute_vecsize_and_simdlen): Ditto.
(aarch64_simd_clone_adjust): Ditto.
(aarch64_simd_clone_usable): Ditto.
(TARGET_SIMD_CLONE_COMPUTE_VECSIZE_AND_SIMDLEN): New macro.
(TARGET_SIMD_CLONE_ADJUST): Ditto.
(TARGET_SIMD_CLONE_USABLE): Ditto.
* config/i386/i386.c (ix86_simd_clone_adjust): Add definition check.
* omp-simd-clone.c (expand_simd_clones): Add targetm.simd_clone.adjust
call.

From-SVN: r268044

decl.c (grokdeclarator): Use typespec_loc in error messages about 'auto' and trailing return type.

/cp
2019-01-17  Paolo Carlini  <paolo.carlini@oracle.com>

* decl.c (grokdeclarator): Use typespec_loc in error messages
about 'auto' and trailing return type.

/testsuite
2019-01-17  Paolo Carlini  <paolo.carlini@oracle.com>

* g++.dg/cpp0x/auto52.C: Test locations too.
* g++.dg/cpp0x/trailing2.C: Likewise.
* g++.dg/cpp1y/auto-fn18.C: Likewise.
* g++.dg/cpp1y/auto-fn25.C: Likewise.
* g++.dg/cpp1y/auto-fn52.C: Likewise.
* g++.dg/cpp1y/auto-fn53.C: Likewise.
* g++.dg/cpp1y/auto-fn54.C: Likewise.

From-SVN: r268043

C++: Fix ICE when adding overloaded operator via using_decl (PR c++/88699)

PR c++/88699 reports an ICE within this assertion in add_method:

  gcc_assert (!current_fns || !DECL_DESTRUCTOR_P (method));

when adding an overloaded operator to a class via a using_decl, due to
DECL_DESTRUCTOR_P requiring a FUNCTION_DECL, but "method" being a
USING_DECL.

This patch weakens the assertion to avoid testing DECL_DESTRUCTOR_P
for the case where "via_using" is true, fixing the ICE.

gcc/cp/ChangeLog:
PR c++/88699
* class.c (add_method): Don't use DECL_DESTRUCTOR_P on
USING_DECLs.

gcc/testsuite/ChangeLog:
PR c++/88699
* g++.dg/template/pr88699.C: New test.

From-SVN: r268041

PR tree-optimization/88800 - Spurious -Werror=array-bounds for non-taken branch

gcc/ChangeLog:

PR tree-optimization/88800
* gimple-fold.c (gimple_fold_builtin_memory_op): Avoid checking
NO_WARNING bit here.  Avoid folding out-of-bounds calls.
* gimple-ssa-warn-restrict.c (maybe_diag_offset_bounds): Remove
redundant argument.  Add new argument and issue diagnostics under
its control.  Detect out-of-bounds access even with warnings
disabled.
(check_bounds_or_overlap): Change return type.  Add argument.
(wrestrict_dom_walker::check_call): Adjust.
* gimple-ssa-warn-restrict.h (check_bounds_or_overlap): Add argument.
* tree-ssa-strlen.c (handle_builtin_strcpy): Adjust to change in
check_bounds_or_overlap's return value.
(handle_builtin_stxncpy): Same.
(handle_builtin_strcat): Same.

gcc/testsuite/ChangeLog:

PR tree-optimization/88800
* c-c++-common/Wrestrict.c: Adjust.
* gcc.dg/Warray-bounds-37.c: New test.
* gcc.dg/builtin-memcpy-2.c: New test.
* gcc.dg/builtin-memcpy.c: New test.

From-SVN: r268037

Fix filesystem::equivalent for mingw

* src/c++17/fs_ops.cc
(equivalent(const path&, const path&, error_code&))
[_GLIBCXX_FILESYSTEM_IS_WINDOWS]: Use GetFileInformationByHandle to
compare files instead of relying on incomplete info returned by stat.

From-SVN: r268036

PR libstdc++/88884 fix filesystem::absolute("//") for mingw

PR libstdc++/88884
* src/c++17/fs_ops.cc (absolute(const path&, error_code&)): Do nothing
if the path is already absolute.
[_GLIBCXX_FILESYSTEM_IS_WINDOWS]: Normalize root-directory.
[!_GLIBCXX_FILESYSTEM_IS_WINDOWS]: Pass error_code to current_path.

From-SVN: r268035

PR libstdc++/88881 adjust filesystem::status and tests for mingw semantics

On Windows stat("foo/bar/../.") will resolve to "foo" even if that is a
non-directory and "foo/bar" does not exist. This is the expected
behaviour and consistent with boost::filesystem, so don't try to correct
it. The only unwanted behaviour is that stat("baz/") fails due to a
mingw bug (fixed in mingw-w64 v6.0.0) so add a workaround.

PR libstdc++/88881
* src/c++17/fs_ops.cc (canonical(const path&, error_code&))
[_GLIBCXX_FILESYSTEM_IS_WINDOWS]: Normalize path, to match behaviour
of filesystem::exists.
(create_directories(const path&, error_code&)): Add assertions.
(status(const path&, error_code&)) [_GLIBCXX_FILESYSTEM_IS_WINDOWS]:
Add workaround for bug in _wstat for paths with trailing slash.
* testsuite/27_io/filesystem/operations/create_directories.cc: Adjust
for expected behaviour on mingw.
* testsuite/experimental/filesystem/operations/create_directories.cc:
Likewise.
* testsuite/27_io/filesystem/operations/temp_directory_path.cc: Use
"TMP" instead of "TMPDIR" and clean environment before each test. Do
not test permissions on mingw targets.

From-SVN: r268034

Fix Arm testcase by using NEON.

gcc/testsuite/ChangeLog:

PR target/88850
* gcc.target/arm/pr51968.c: Use neon intrinsics.

From-SVN: r268033

[libbacktrace] Add btest_dwz test-case

Add test-case to verify that libbacktrace can read debug info that was
compressed with dwz.

2019-01-17  Tom de Vries  <tdevries@suse.de>

PR libbacktrace/82857
* configure.ac (DWZ): Set with AC_CHECK_PROG.
(HAVE_DWZ): Set with AM_CONDITIONAL.
* configure: Regenerate.
* Makefile.am (TESTS): Add btest_dwz.
* Makefile.in: Regenerate.

From-SVN: r268032

[libbacktrace] Handle DW_FORM_GNU_ref_alt

Handle DW_FORM_GNU_ref_alt which references the .debug_info section in the
.gnu_debugaltlink file.

2019-01-17  Tom de Vries  <tdevries@suse.de>

PR libbacktrace/82857
* dwarf.c (enum attr_val_encoding): Add ATTR_VAL_REF_ALT_INFO.
(read_attribute): Handle DW_FORM_GNU_ref_alt using
ATTR_VAL_REF_ALT_INFO.
(read_referenced_name_from_attr): Handle DW_FORM_GNU_ref_alt.

From-SVN: r268031

[libbacktrace] Add find_unit

Add a function that finds the unit given an offset into .debug_info.

2019-01-17  Tom de Vries  <tdevries@suse.de>

* dwarf.c (struct unit): Add low_offset and high_offset fields.
(struct unit_vector): New type.
(struct dwarf_data): Add units and units_counts fields.
(find_unit): New function.
(find_address_ranges): Add and handle unit_tag parameter.
(build_address_map): Add and handle units_vec parameter.
(build_dwarf_data): Pass units_vec to build_address_map.  Store resulting
units vector.

From-SVN: r268030

Port testsuite to GCN

This collection of miscellaneous patches configures the testsuite to run on AMD
GCN in a standalone (i.e. not offloading) configuration.  It assumes you have
your Dejagnu set up to run binaries via the gcn-run tool.

2019-01-17  Andrew Stubbs  <ams@codesourcery.com>
    Kwok Cheung Yeung  <kcy@codesourcery.com>
    Julian Brown  <julian@codesourcery.com>
    Tom de Vries  <tom@codesourcery.com>

gcc/
* doc/sourcebuild.texi: Document dg-add-options sqrt_insn.

gcc/testsuite/
* gcc.dg/20020312-2.c: Add amdgcn support.
* gcc.dg/Wno-frame-address.c: Disable on amdgcn.
* gcc.dg/builtin-apply2.c: Likewise.
* gcc.dg/torture/stackalign/builtin-apply-2.c: Likewise.
* gcc.dg/gimplefe-28.c: Add dg-add-options for sqrt_insn.
* gcc.dg/intermod-1.c: Add -mlocal-symbol-id on amdgcn.
* gcc.dg/memcmp-1.c: Increase timeout factor.
* gcc.dg/pr59605-2.c: Addd -DMAX_COPY=1025 on amdgcn.
* gcc.dg/sibcall-10.c: xfail on amdgcn.
* gcc.dg/sibcall-9.c: Likewise.
* gcc.dg/tree-ssa/gen-vect-11c.c: Likewise.
* gcc.dg/tree-ssa/pr84512.c: Likewise.
* gcc.dg/tree-ssa/loop-1.c: Adjust expectations for amdgcn.
* gfortran.dg/bind_c_array_params_2.f90: Likewise.
* lib/target-supports.exp (check_effective_target_trampolines):
Configure amdgcn.
(check_profiling_available): Likewise.
(check_effective_target_global_constructor): Likewise.
(check_effective_target_return_address): Likewise.
(check_effective_target_fopenacc): Likewise.
(check_effective_target_fopenmp): Likewise.
(check_effective_target_vect_int): Likewise.
(check_effective_target_vect_intfloat_cvt): Likewise.
(check_effective_target_vect_uintfloat_cvt): Likewise.
(check_effective_target_vect_floatint_cvt): Likewise.
(check_effective_target_vect_floatuint_cvt): Likewise.
(check_effective_target_vect_simd_clones): Likewise.
(check_effective_target_vect_shift): Likewise.
(check_effective_target_whole_vector_shift): Likewise.
(check_effective_target_vect_bswap): Likewise.
(check_effective_target_vect_shift_char): Likewise.
(check_effective_target_vect_long): Likewise.
(check_effective_target_vect_float): Likewise.
(check_effective_target_vect_double): Likewise.
(check_effective_target_vect_perm): Likewise.
(check_effective_target_vect_perm_byte): Likewise.
(check_effective_target_vect_perm_short): Likewise.
(check_effective_target_vect_widen_mult_qi_to_hi): Likewise.
(check_effective_target_vect_widen_mult_hi_to_si): Likewise.
(check_effective_target_vect_widen_mult_qi_to_hi_pattern): Likewise.
(check_effective_target_vect_widen_mult_hi_to_si_pattern): Likewise.
(check_effective_target_vect_natural_alignment): Likewise.
(check_effective_target_vect_fully_masked): Likewise.
(check_effective_target_vect_element_align): Likewise.
(check_effective_target_vect_masked_store): Likewise.
(check_effective_target_vect_scatter_store): Likewise.
(check_effective_target_vect_condition): Likewise.
(check_effective_target_vect_cond_mixed): Likewise.
(check_effective_target_vect_char_mult): Likewise.
(check_effective_target_vect_short_mult): Likewise.
(check_effective_target_vect_int_mult): Likewise.
(check_effective_target_sqrt_insn): Likewise.
(check_effective_target_vect_call_sqrtf): Likewise.
(check_effective_target_vect_call_btrunc): Likewise.
(check_effective_target_vect_call_btruncf): Likewise.
(check_effective_target_vect_call_ceil): Likewise.
(check_effective_target_vect_call_floorf): Likewise.
(check_effective_target_lto): Likewise.
(check_vect_support_and_set_flags): Likewise.
(check_effective_target_vect_stridedN): Enable when fully masked is
available.
(add_options_for_sqrt_insn): New procedure.

Co-Authored-By: Julian Brown <julian@codesourcery.com>
Co-Authored-By: Kwok Cheung Yeung <kcy@codesourcery.com>
Co-Authored-By: Tom de Vries <tom@codesourcery.com>
From-SVN: r268028

Ignore LLVM's blank lines.

The GCN toolchain must use the LLVM assembler and linker because there's no
binutils port.  The LLVM tools do not have the same diagnostic style as
binutils, so the "blank line(s) in output" tests are inappropriate (and very
noisy).

The LLVM tools also have different command line options, so it's not possible
to autodetect object formats in the same way.

This patch addresses both issues.

2019-01-17  Andrew Stubbs  <ams@codesourcery.com>

gcc/
* doc/sourcebuild.texi: Document dg-require-effective-target
llvm_binutils and offload_gcn.

gcc/testsuite/
* lib/file-format.exp (gcc_target_object_format): Handle AMD GCN.
* lib/gcc-dg.exp (gcc-dg-prune): Ignore blank lines from the LLVM
linker.
* lib/target-supports.exp (check_effective_target_llvm_binutils): New.

From-SVN: r268027

Testsuite: GCN is always PIE.

The GCN/HSA loader ignores the load address and uses a random location, so we
build all GCN binaries as PIE, by default.

This patch makes the necessary testsuite adjustments to make this work
correctly.

2019-01-17  Andrew Stubbs  <ams@codesourcery.com>

gcc/testsuite/
* gcc.dg/graphite/scop-19.c: Check pie_enabled.
* gcc.dg/pic-1.c: Disable on amdgcn.
* gcc.dg/pic-2.c: Disable on amdgcn.
* gcc.dg/pic-3.c: Disable on amdgcn.
* gcc.dg/pic-4.c: Disable on amdgcn.
* gcc.dg/pie-3.c: Disable on amdgcn.
* gcc.dg/pie-4.c: Disable on amdgcn.
* gcc.dg/uninit-19.c: Check pie_enabled.
* lib/target-supports.exp (check_effective_target_pie): Add amdgcn.

From-SVN: r268026

Add dg-require-effective-target exceptions

There are a number of tests that fail because they assume that exceptions are
available, but GCN does not support them, yet.

This patch adds "dg-require-effective-target exceptions" in all the affected
tests.  There's probably an automatic way to test for exceptions, but the
current implementation simply says that AMD GCN does not support them.  This
should ensure that no other targets are affected by the change.

2019-01-17  Andrew Stubbs  <ams@codesourcery.com>
    Kwok Cheung Yeung  <kcy@codesourcery.com>
    Julian Brown  <julian@codesourcery.com>
    Tom de Vries  <tom@codesourcery.com>

gcc/
* doc/sourcebuild.texi: Document dg-required-effective-target
exceptions.

gcc/testsuite/
* c-c++-common/ubsan/pr71512-1.c: Require exceptions.
* c-c++-common/ubsan/pr71512-2.c: Require exceptions.
* gcc.c-torture/compile/pr34648.c: Require exceptions.
* gcc.c-torture/compile/pr41469.c: Require exceptions.
* gcc.dg/20111216-1.c: Require exceptions.
* gcc.dg/cleanup-10.c: Require exceptions.
* gcc.dg/cleanup-11.c: Require exceptions.
* gcc.dg/cleanup-12.c: Require exceptions.
* gcc.dg/cleanup-13.c: Require exceptions.
* gcc.dg/cleanup-5.c: Require exceptions.
* gcc.dg/cleanup-8.c: Require exceptions.
* gcc.dg/cleanup-9.c: Require exceptions.
* gcc.dg/gomp/pr29955.c: Require exceptions.
* gcc.dg/lto/pr52097_0.c: Require exceptions.
* gcc.dg/nested-func-5.c: Require exceptions.
* gcc.dg/pch/except-1.c: Require exceptions.
* gcc.dg/pch/valid-2.c: Require exceptions.
* gcc.dg/pr41470.c: Require exceptions.
* gcc.dg/pr42427.c: Require exceptions.
* gcc.dg/pr44545.c: Require exceptions.
* gcc.dg/pr47086.c: Require exceptions.
* gcc.dg/pr51481.c: Require exceptions.
* gcc.dg/pr51644.c: Require exceptions.
* gcc.dg/pr52046.c: Require exceptions.
* gcc.dg/pr54669.c: Require exceptions.
* gcc.dg/pr56424.c: Require exceptions.
* gcc.dg/pr64465.c: Require exceptions.
* gcc.dg/pr65802.c: Require exceptions.
* gcc.dg/pr67563.c: Require exceptions.
* gcc.dg/tree-ssa/pr41469-1.c: Require exceptions.
* gcc.dg/tree-ssa/ssa-dse-28.c: Require exceptions.
* gcc.dg/vect/pr46663.c: Require exceptions.
* lib/target-supports.exp (check_effective_target_exceptions): New.

Co-Authored-By: Julian Brown <julian@codesourcery.com>
Co-Authored-By: Kwok Cheung Yeung <kcy@codesourcery.com>
Co-Authored-By: Tom de Vries <tom@codesourcery.com>
From-SVN: r268025

GCN back-end config

This patch contains the configuration adjustments needed to enable the GCN
back-end.

The new configure check for dlopen is required to allow building the new
gcn-run tool.  This tool uses libdl to load the HSA runtime libraries, which
are required to run programs on the GPU.  The tool is disabled if libdl is not
available.

2019-01-17  Andrew Stubbs  <ams@codesourcery.com>
    Kwok Cheung Yeung  <kcy@codesourcery.com>
    Julian Brown  <julian@codesourcery.com>
    Tom de Vries  <tom@codesourcery.com>
    Jan Hubicka  <hubicka@ucw.cz>
    Martin Jambor  <mjambor@suse.cz>

* configure.ac: Likewise.
* configure: Regenerate.
* contrib/config-list.mk: Add amdgcn-amdhsa.

gcc/
* config.gcc: Add amdgcn*-*-amdhsa configuration.
* configure.ac: Check for dlopen.
* configure: Regenerate.

Co-Authored-By: Jan Hubicka <hubicka@ucw.cz>
Co-Authored-By: Julian Brown <julian@codesourcery.com>
Co-Authored-By: Kwok Cheung Yeung <kcy@codesourcery.com>
Co-Authored-By: Martin Jambor <mjambor@suse.cz>
Co-Authored-By: Tom de Vries <tom@codesourcery.com>
From-SVN: r268024

GCN back-end code

This patch contains the major part of the GCN back-end.  The machine
description has been broken out to avoid the mailing list size limit.

The back-end contains various bits that support OpenACC and OpenMP, but the
middle-end and libgomp patches are missing, as is mkoffload.  I include them
here because they're harmless and carving up the files seems like unnecessary
effort.  The remaining offload support will be posted at a later date.

The gcn-run.c is a separate tool that can run a GCN program on a GPU using
the ROCm drivers and HSA runtime libraries.

2019-01-17  Andrew Stubbs  <ams@codesourcery.com>
    Kwok Cheung Yeung  <kcy@codesourcery.com>
    Julian Brown  <julian@codesourcery.com>
    Tom de Vries  <tom@codesourcery.com>
    Jan Hubicka  <hubicka@ucw.cz>
    Martin Jambor  <mjambor@suse.cz>

gcc/
* common/config/gcn/gcn-common.c: New file.
* config/gcn/driver-gcn.c: New file.
* config/gcn/gcn-builtins.def: New file.
* config/gcn/gcn-hsa.h: New file.
* config/gcn/gcn-modes.def: New file.
* config/gcn/gcn-opts.h: New file.
* config/gcn/gcn-passes.def: New file.
* config/gcn/gcn-protos.h: New file.
* config/gcn/gcn-run.c: New file.
* config/gcn/gcn-tree.c: New file.
* config/gcn/gcn.c: New file.
* config/gcn/gcn.h: New file.
* config/gcn/gcn.opt: New file.
* config/gcn/t-gcn-hsa: New file.

Co-Authored-By: Jan Hubicka <hubicka@ucw.cz>
Co-Authored-By: Julian Brown <julian@codesourcery.com>
Co-Authored-By: Kwok Cheung Yeung <kcy@codesourcery.com>
Co-Authored-By: Martin Jambor <mjambor@suse.cz>
Co-Authored-By: Tom de Vries <tom@codesourcery.com>
From-SVN: r268023

GCN machine description

This patch contains the machine description portion of the GCN back-end.  I've
broken it out mainly to avoid the mailing list size limit.

2019-01-17  Andrew Stubbs  <ams@codesourcery.com>
    Kwok Cheung Yeung  <kcy@codesourcery.com>
    Julian Brown  <julian@codesourcery.com>
    Tom de Vries  <tom@codesourcery.com>
    Jan Hubicka  <hubicka@ucw.cz>
    Martin Jambor  <mjambor@suse.cz>

gcc/
* config/gcn/constraints.md: New file.
* config/gcn/gcn-valu.md: New file.
* config/gcn/gcn.md: New file.
* config/gcn/predicates.md: New file.

Co-Authored-By: Jan Hubicka <hubicka@ucw.cz>
Co-Authored-By: Julian Brown <julian@codesourcery.com>
Co-Authored-By: Kwok Cheung Yeung <kcy@codesourcery.com>
Co-Authored-By: Martin Jambor <mjambor@suse.cz>
Co-Authored-By: Tom de Vries <tom@codesourcery.com>
From-SVN: r268022

GCN libgcc.

This patch contains the GCN port of libgcc.

2019-01-17  Andrew Stubbs  <ams@codesourcery.com>
    Kwok Cheung Yeung  <kcy@codesourcery.com>
    Julian Brown  <julian@codesourcery.com>
    Tom de Vries  <tom@codesourcery.com>

libgcc/
* config.host: Recognize amdgcn*-*-amdhsa.
* config/gcn/crt0.c: New file.
* config/gcn/lib2-divmod-hi.c: New file.
* config/gcn/lib2-divmod.c: New file.
* config/gcn/lib2-gcn.h: New file.
* config/gcn/sfp-machine.h: New file.
* config/gcn/t-amdgcn: New file.

Co-Authored-By: Julian Brown <julian@codesourcery.com>
Co-Authored-By: Kwok Cheung Yeung <kcy@codesourcery.com>
Co-Authored-By: Tom de Vries <tom@codesourcery.com>
From-SVN: r268021

GCN libgfortran.

This patch contains the GCN port of libgfortran.  We use the minimal
configuration created for NVPTX.  That's all that's required, besides the
target-independent bug fixes posted already.

2019-01-17  Andrew Stubbs  <ams@codesourcery.com>
    Kwok Cheung Yeung  <kcy@codesourcery.com>
    Julian Brown  <julian@codesourcery.com>
    Tom de Vries  <tom@codesourcery.com>

libgfortran/
* configure.ac: Use minimal mode for amdgcn.
* configure: Regenerate.

Co-Authored-By: Julian Brown <julian@codesourcery.com>
Co-Authored-By: Kwok Cheung Yeung <kcy@codesourcery.com>
Co-Authored-By: Tom de Vries <tom@codesourcery.com>
From-SVN: r268020

gimple-ssa-isolate-paths.c (stmt_uses_name_in_undefined_way): Replace flag_non_call_exceptions with cfun->can_throw_non_call_exceptions.

* gimple-ssa-isolate-paths.c (stmt_uses_name_in_undefined_way): Replace
flag_non_call_exceptions with cfun->can_throw_non_call_exceptions.
(stmt_uses_0_or_null_in_undefined_way): Likewise.
* tree-ssa-alias.c (same_addr_size_stores_p): Likewise.

From-SVN: r268018

Rename stack-clash protection CFA register to avoid clash

gcc/ChangeLog:

PR target/88851
* config/aarch64/aarch64.md (STACK_CLASH_SVE_CFA_REGNUM): New.
* config/aarch64/aarch64.c (aarch64_allocate_and_probe_stack_space): Use
it and document registers.

gcc/testsuite/ChangeLog:

PR target/88851
* gcc.target/aarch64/stack-check-cfa-3.c: Update test.

From-SVN: r268017

[PR c++/86610] lambda captures in templates

https://gcc.gnu.org/ml/gcc-patches/2019-01/msg00948.html
PR c++/86610
* semantics.c (process_outer_var_ref): Only skip dependent types
in templates.

PR c++/86610
* g++.dg/cpp0x/pr86610.C: New.

From-SVN: r268016

[AArch64] Initial -mcpu=ares tuning

This patch adds a tuning struct for the Arm Ares CPU and uses it for -m{cpu,tune}=ares.
The tunings are an initial attempt and may be improved upon in the future, but they serve
as a decent starting point for GCC 9.

With this I see a 1.3% improvement on SPEC2006 int and 0.3% on SPEC2006 fp with -mcpu=ares.
On SPEC2017 I see a 0.6% improvement in intrate and changes in the noise for fprate.

        * config/aarch64/aarch64.c (ares_tunings): Define.
        * config/aarch64/aarch64-cores.def (ares): Use the above.

From-SVN: r268015

avx512f-vfixupimmpd-2.c: Fix the test cases for VFIXUPIMM* intrinsics.

gcc/testsuite/ChangeLog
2019-01-17  Wei Xiao  <wei3.xiao@intel.com>

* gcc.target/i386/avx512f-vfixupimmpd-2.c: Fix the test cases for
VFIXUPIMM* intrinsics.
* gcc.target/i386/avx512f-vfixupimmps-2.c: Ditto.
* gcc.target/i386/avx512f-vfixupimmsd-2.c: Ditto.
* gcc.target/i386/avx512f-vfixupimmss-2.c: Ditto.

From-SVN: r268014

re PR target/88794 (fixupimm intrinsics are unusable)

gcc/ChangeLog
2019-01-17  Wei Xiao  <wei3.xiao@intel.com>

PR target/88794
Revert:

2018-11-06  Wei Xiao  <wei3.xiao@intel.com>

* config/i386/avx512fintrin.h: Update VFIXUPIMM* intrinsics.
(_mm512_fixupimm_round_pd): Update parameters and builtin.
(_mm512_maskz_fixupimm_round_pd): Ditto.
(_mm512_fixupimm_round_ps): Ditto.
(_mm512_maskz_fixupimm_round_ps): Ditto.
(_mm_fixupimm_round_sd): Ditto.
(_mm_maskz_fixupimm_round_sd): Ditto.
(_mm_fixupimm_round_ss): Ditto.
(_mm_maskz_fixupimm_round_ss): Ditto.
(_mm512_fixupimm_pd): Ditto.
(_mm512_maskz_fixupimm_pd): Ditto.
(_mm512_fixupimm_ps): Ditto.
(_mm512_maskz_fixupimm_ps): Ditto.
(_mm_fixupimm_sd): Ditto.
(_mm_maskz_fixupimm_sd): Ditto.
(_mm_fixupimm_ss): Ditto.
(_mm_maskz_fixupimm_ss): Ditto.
(_mm512_mask_fixupimm_round_pd): Update builtin.
(_mm512_mask_fixupimm_round_ps): Ditto.
(_mm_mask_fixupimm_round_sd): Ditto.
(_mm_mask_fixupimm_round_ss): Ditto.
(_mm512_mask_fixupimm_pd): Ditto.
(_mm512_mask_fixupimm_ps): Ditto.
(_mm_mask_fixupimm_sd): Ditto.
(_mm_mask_fixupimm_ss): Ditto.
* config/i386/avx512vlintrin.h:
(_mm256_fixupimm_pd): Update parameters and builtin.
(_mm256_maskz_fixupimm_pd): Ditto.
(_mm256_fixupimm_ps): Ditto.
(_mm256_maskz_fixupimm_ps): Ditto.
(_mm_fixupimm_pd): Ditto.
(_mm_maskz_fixupimm_pd): Ditto.
(_mm_fixupimm_ps): Ditto.
(_mm_maskz_fixupimm_ps): Ditto.
(_mm256_mask_fixupimm_pd): Update builtin.
(_mm256_mask_fixupimm_ps): Ditto.
(_mm_mask_fixupimm_pd): Ditto.
(_mm_mask_fixupimm_ps): Ditto.
* config/i386/i386-builtin-types.def: Add new types and remove useless ones.
* config/i386/i386-builtin.def: Update builtin definitions.
* config/i386/i386.c: Handle new builtin types and remove useless ones.
* config/i386/sse.md: Update VFIXUPIMM* patterns.
(<avx512>_fixupimm<mode>_maskz<round_saeonly_expand_name>): Update.
(<avx512>_fixupimm<mode><sd_maskz_name><round_saeonly_name>): Update.
(<avx512>_fixupimm<mode>_mask<round_saeonly_name>): Update.
(avx512f_sfixupimm<mode>_maskz<round_saeonly_expand_name>): Update.
(avx512f_sfixupimm<mode><sd_maskz_name><round_saeonly_name>): Update.
(avx512f_sfixupimm<mode>_mask<round_saeonly_name>): Update.
* config/i386/subst.md:
(round_saeonly_sd_mask_operand4): Add new subst_attr.
(round_saeonly_sd_mask_op4): Ditto.
(round_saeonly_expand_operand5): Ditto.
(round_saeonly_expand): Update.

gcc/testsuite/ChangeLog
2019-01-17  Wei Xiao  <wei3.xiao@intel.com>

PR target/88794
Revert:
2018-11-06  Wei Xiao  <wei3.xiao@intel.com>

* gcc.target/i386/avx-1.c: Update tests for VFIXUPIMM* intrinsics.
* gcc.target/i386/avx512f-vfixupimmpd-1.c: Ditto.
* gcc.target/i386/avx512f-vfixupimmpd-2.c: Ditto.
* gcc.target/i386/avx512f-vfixupimmps-1.c: Ditto.
* gcc.target/i386/avx512f-vfixupimmsd-1.c: Ditto.
* gcc.target/i386/avx512f-vfixupimmsd-2.c: Ditto.
* gcc.target/i386/avx512f-vfixupimmss-1.c: Ditto.
* gcc.target/i386/avx512f-vfixupimmss-2.c: Ditto.
* gcc.target/i386/avx512vl-vfixupimmpd-1.c: Ditto.
* gcc.target/i386/avx512vl-vfixupimmps-1.c: Ditto.
* gcc.target/i386/sse-13.c: Ditto.
* gcc.target/i386/sse-14.c: Ditto.
* gcc.target/i386/sse-22.c: Ditto.
* gcc.target/i386/sse-23.c: Ditto.
* gcc.target/i386/testimm-10.c: Ditto.
* gcc.target/i386/testround-1.c: Ditto.

From-SVN: r268013

re PR target/88794 (fixupimm intrinsics are unusable)

2019-01-17  Wei Xiao  <wei3.xiao@intel.com>

        PR target/88794
        Revert:
        2018-11-12  Wei Xiao  <wei3.xiao@intel.com>

        * config/i386/sse.md: Combine VFIXUPIMM* patterns
        (<avx512>_fixupimm<mode>_maskz<round_saeonly_expand_name>): Update.
        (<avx512>_fixupimm<mode><sd_maskz_name><round_saeonly_name>): Update.
        (<avx512>_fixupimm<mode>_mask<round_saeonly_name>): Remove.
        (avx512f_sfixupimm<mode>_maskz<round_saeonly_expand_name>): Update.
        (avx512f_sfixupimm<mode><sd_maskz_name><round_saeonly_name>): Update.
        (avx512f_sfixupimm<mode>_mask<round_saeonly_name>): Remove.

From-SVN: r268012

re PR target/88794 (fixupimm intrinsics are unusable)

gcc/ChangeLog:
2019-01-17  Wei Xiao  <wei3.xiao@intel.com>

        PR target/88794
        Revert:
        2018-12-15  Jakub Jelinek  <jakub@redhat.com>

        PR target/88489
        * config/i386/sse.md (UNSPEC_SFIXUPIMM): New unspec enumerator.
        (avx512f_sfixupimm<mode><mask_name><round_saeonly_name>): Use it
        instead of UNSPEC_FIXUPIMM.

gcc/testsuite/ChangeLog:
2019-01-17  Wei Xiao  <wei3.xiao@intel.com>

        PR target/88794
        Revert:
        2018-12-15  Jakub Jelinek  <jakub@redhat.com>

        PR target/88489
        * gcc.target/i386/avx512vl-vfixupimmsd-2.c: New test.
        * gcc.target/i386/avx512vl-vfixupimmss-2.c: New test.

From-SVN: r268011

re PR lto/86736 (g++.dg/asan/pr81021.C   -O2 -flto -flto-partition=none ICE at dwarf2out.c:31111)

2019-01-17  Richard Biener  <rguenther@suse.de>

PR lto/86736
* dwarf2out.c (want_pubnames): Never generate pubnames sections
and friends for the LTO part of debug info.

From-SVN: r268010

re PR tree-optimization/86214 (Strongly increased stack usage)

PR tree-optimization/86214
* cfgexpand.c (add_stack_var_conflict): Don't add any conflicts
if x == y.

From-SVN: r268009

re PR rtl-optimization/88870 (ICE: Segmentation fault (in df_worklist_propagate_backward))

PR rtl-optimization/88870
* dce.c (deletable_insn_p): Never delete const/pure calls that can
throw if we can't alter the cfg or delete dead exceptions.
(mark_insn): Don't call find_call_stack_args for such calls.

* gcc.dg/pr88870.c: New test.

From-SVN: r268008

[rs6000] Support more prototypes for vec_ld and vec_st.

2019-01-17  Kewen Lin  <linkw@gcc.gnu.org>

gcc/ChangeLog
    * doc/extend.texi: Add four new prototypes for vec_ld and seven new
    prototypes for vec_st.
    * config/rs6000/rs6000-c.c (altivec_overloaded_builtins): Add entries
    for scalar address type variants of altivec_vec_ld/altivec_vec_st,
    mainly on signed/unsigned long long and double.

gcc/testsuite/ChangeLog
    * gcc.target/powerpc/altivec_vld_vst_addr.c: New test.

From-SVN: r268007

[PR87768] reset location wrapper suppression when reentering top level

Concepts-checking and other kinds of early tsubsting may often take
place while location wrappers are suppressed, e.g. because we've
triggered template instantiation within template parameter lists.

With that, exprs that are usually wrapped by VIEW_CONVERT_EXPRs
location wrappers may end up wrapped by NON_LVALUE_EXPRs that are not
marked as location wrappers.  If such NON_LVALUE_EXPRs tsubsted exprs
undergo another round of tsubsting, say for constraint checking, or
even for another round of specialization, they will be rejected by
tsubst_copy_and_build.

This patch arranges for suppress_location_wrappers to be saved and
reset when pushing to the top level, and restored when popping from
it.

for  gcc/cp/ChangeLog

PR c++/87768
* cp-tree.h (saved_scope): Add suppress_location_wrappers.
* name-lookup.c (do_push_to_top_level): Save and reset it.
(do_pop_from_top_level): Restore it.

for  gcc/testsuite/ChangeLog

PR c++/87768
* g++.dg/concepts/pr87768.C: New.

From-SVN: r268006

[PR86648] use auto identifier for class placeholder templates

dwarf2out recognizes unspecified auto types by the identifier.  C++
template class placeholders are unspecified auto types that take the
identifier of the class rather than those used by preexisting auto
types, so dwarf2out ICEs when it finds one of those.  Alas, they may
be visible to dwarf2out, since the types of e.g. static data members
of templates are only deduced at member instantiation, i.e., if the
data member is actually referenced, but the data member is added as a
field, still with unspecified auto placeholder type, when the
enclosing class is instantiated.

I've changed placeholder creator to use an auto identifier instead,
which allowed dropping the placeholder test in C++'s is_auto (alas, it
can't be used in dwarf2out, think LTO).  To avoid losing information
in error messages and dumps and whatnot, I've added code to recognize
placeholders for template classes say A and print them out as
A<...auto...>.

for  gcc/cp/ChangeLog

PR c++/86648
        * pt.c (make_template_placeholder): Use auto_identifier.
        (is_auto): Drop CLASS_PLACEHOLDER_TEMPLATE test.
        * error.c (dump_type): Handle template placeholders.
        * cxx-pretty-print.c (pp_cx_unqualified_id): Likewise.

for  gcc/testsuite/ChangeLog

PR c++/86648
        * gcc.dg/cpp1z/pr86648.C: New.

From-SVN: r268005

[PR88146] avoid diagnostics diffs if cdtor_returns_this

Diagnostics for testsuite/g++.dg/cpp0x/inh-ctor32.C varied across
platforms.  Specifically, on ARM, the diagnostics within the subtest
derived_ctor::inherited_derived_ctor::constexpr_noninherited_ctor did
not match those displayed on other platforms, and the test failed.

The difference seemed to have to do with locations assigned to ctors,
but it was more subtle: on ARM, the instantiation of bor's template
ctor was nested within the instantiation of bar's template ctor
inherited from bor.  The reason turned out to be related with the
internal return type of ctors: arm_cxx_cdtor_returns_this is enabled
for because of AAPCS, while cxx.cdtor_returns_this is disabled on most
other platforms.  While convert_to_void returns early with a VOID
expr, the non-VOID return type of the base ctor CALL_EXPR causes
convert_to_void to inspect the called decl for nodiscard attributes:
maybe_warn_nodiscard -> cp_get_fndecl_from_callee ->
maybe_constant_init -> cxx_eval_outermost_constant_expr ->
instantiate_constexpr_fns -> nested instantiation.

The internal return type assigned to a cdtor should not affect
instantiation (constexpr or template) decisions, IMHO.  We know it
affects diagnostics, but I have a hunch this might bring deeper issues
with it, so I've arranged for the CALL_EXPR handler in convert_to_void
to disregard cdtors, regardless of the ABI.

for  gcc/cp/ChangeLog

PR c++/88146
* cvt.c (convert_to_void): Handle all cdtor calls as if
returning void.

From-SVN: r268004

[PATCH, rs6000, testsuite] Fix PR87306

    PR target/87306
    * gcc.dg/vect/bb-slp-pow-1.c: Modify to reflect that
    the loop is not vectorized on POWER unless hardware
    misaligned loads are available.

From-SVN: r268003

runtime: dropg before CAS g status to _Grunnable/_Gwaiting

    Currently, we dropg (which clears gp.m) after we CAS the g status
    to _Grunnable or _Gwaiting. Immediately after CASing the g status,
    another thread may CAS it to _Gscan status and scan its stack.
    With precise stack scan, it accesses gp.m in order to switch to g
    and back (in doscanstackswitch). This races with dropg. If
    doscanstackswitch reads gp.m, then dropg runs, when we restore
    the m at the end of the scan it will set to a stale value. Worse,
    if dropg runs after doscanstackswitch sets the new m, gp will be
    running with a nil m.

    To fix this, we do dropg before CAS g status to _Grunnable or
    _Gwaiting. We can do this safely if we are CASing from _Grunning,
    as we own the g when it is in _Grunning. There is one case where
    we CAS from _Gsyscall to _Grunnable. It is not safe to dropg when
    it is in _Gsyscall, as precise stack scan needs to read gp.m in
    order to signal the m. So we need to introduce a transient state,
    _Gexitingsyscall, between _Gsyscall and _Grunnable, where the GC
    should not scan its stack.

    In is a little unfortunate that we have to add another g status.
    We could reuse an existing one (e.g. _Gcopystack), but it is
    clearer and safer to just use a new one, as Austin suggested.

    Reviewed-on: https://go-review.googlesource.com/c/158157

From-SVN: r268001

Daily bump.

From-SVN: r267999

[libbacktrace] Handle DW_FORM_GNU_strp_alt

Handle DW_FORM_GNU_strp_alt which references the .debug_str section in the
.gnu_debugaltlink file.

2019-01-17  Tom de Vries  <tdevries@suse.de>

PR libbacktrace/82857
* dwarf.c (read_attribute): Handle DW_FORM_GNU_strp_alt
using altlink.

From-SVN: r267996

[libbacktrace] Handle alt FORMS without .gnu_debugaltlink

Handle DW_FORM_GNU_strp_alt and DW_FORM_GNU_ref_alt references robustly in
presence of missing .gnu_debugaltlink file.

2019-01-17  Tom de Vries  <tdevries@suse.de>

* dwarf.c (enum attr_val_encoding): Add ATTR_VAL_NONE.
(read_attribute): Add altlink parameter.  Handle missing altlink for
DW_FORM_GNU_strp_alt and DW_FORM_GNU_ref_alt.
(find_address_ranges, build_address_map, build_dwarf_data): Add and
handle altlink parameter.
(read_referenced_name, read_function_entry): Add argument to
read_attribute call.

From-SVN: r267995

[libbacktrace] Add altlink field to struct dwarf_data

Add an altlink field to struct dwarf_data, and initialize it with the pointer
to the struct dwarf_data for the .gnu_debugaltlink.

2019-01-17  Tom de Vries  <tdevries@suse.de>

* dwarf.c (struct dwarf_data): Add altlink field.
(backtrace_dwarf_add): Add and handle fileline_altlink parameter.
* elf.c (elf_add): Add argument to backtrace_dwarf_add call.
(phdr_callback, backtrace_initialize): Add argument to elf_add calls.
* internal.h (backtrace_dwarf_add): Add fileline_altlink parameter.
* pecoff.c (coff_add): Add argument to backtrace_dwarf_add call.
* xcoff.c (xcoff_add): Same.

From-SVN: r267994

[libbacktrace] Return struct dwarf_data pointer from elf_add

Allow the caller of elf_add access to the struct dwarf_data pointer
corresponding to the added elf.

2019-01-17  Tom de Vries  <tdevries@suse.de>

* internal.h (backtrace_dwarf_add): Add fileline_entry parameter.
* dwarf.c (backtrace_dwarf_add): Add and handle fileline_entry parameter.
* elf.c (elf_add): Add and handle fileline_entry parameter.  Add
argument to backtrace_dwarf_add call.
(phdr_callback, backtrace_initialize): Add argument to elf_add calls.
* pecoff.c (coff_add): Add argument to backtrace_dwarf_add call.
* xcoff.c (xcoff_add): Same.

From-SVN: r267993

[libbacktrace] Read .gnu_debugaltlink

Read the elf file pointed at by the .gnu_debugaltlink section, and verify that
the build id matches.

2019-01-17  Tom de Vries  <tdevries@suse.de>

* elf.c (elf_add): Add and handle with_buildid_data and
with_buildid_size parameters.  Handle .gnu_debugaltlink section.
(phdr_callback, backtrace_initialize): Add arguments to elf_add calls.

From-SVN: r267992

Fix failing filesystem tests on mingw targets

* config/abi/pre/gnu.ver (GLIBCXX_3.4.26): Add exports for fstream
constructors and open members taking wide strings. Fix patterns for
filesystem::path members to match wstring_view parameters. Add
exports for shared_ptr members used by directory iterators.
* src/c++17/fs_ops.cc (remove(const path&, error_code&)): Clear the
error code parameter if the file doesn't exist.
* src/filesystem/ops.cc (remove(const path&, error_code&)):
Likewise.
* testsuite/27_io/filesystem/operations/canonical.cc: Fix expected
values for mingw targets, where "/" is not an absolute path. Do not
test symlinks on mingw targets.
* testsuite/experimental/filesystem/operations/canonical.cc: Likewise.
* testsuite/27_io/filesystem/operations/copy.cc: Do not test symlinks
on mingw targets.
* testsuite/experimental/filesystem/operations/copy.cc: Likewise.
* testsuite/27_io/filesystem/operations/create_directories.cc: Check
that each component of the path is created.
* testsuite/experimental/filesystem/operations/create_directories.cc:
Likewise.
* testsuite/27_io/filesystem/operations/exists.cc: Do not test
permissions on mingw targets.
* testsuite/experimental/filesystem/operations/exists.cc: Likewise.
* testsuite/27_io/filesystem/operations/is_empty.cc: Likewise.
* testsuite/experimental/filesystem/operations/is_empty.cc: Likewise.
* testsuite/27_io/filesystem/operations/permissions.cc: XFAIL for
mingw targets.
* testsuite/experimental/filesystem/operations/permissions.cc:
Likewise.
* testsuite/27_io/filesystem/operations/remove.cc: Do not test
symlinks or permissions on mingw targets.
* testsuite/experimental/filesystem/operations/remove.cc: Likewise.
* testsuite/27_io/filesystem/operations/remove_all.cc: Do not test
symlinks on mingw targets.
* testsuite/experimental/filesystem/operations/remove_all.cc:
Likewise.
* testsuite/27_io/filesystem/operations/status.cc: Do not test
permissions on mingw targets.
* testsuite/27_io/filesystem/operations/weakly_canonical.cc: Do not
test symlinks on mingw targets.
* testsuite/experimental/filesystem/operations/space.cc: Fix test
for mingw targets.

From-SVN: r267991

syscall: mark C syscall functions noescape

    Many C syscall functions take pointer arguments. The pointers
    don't escape in the C functions. Mark the C functions noescape so
    calling them doesn't need allocation.

    Reviewed-on: https://go-review.googlesource.com/c/158158

From-SVN: r267989

[libbacktrace] Factor out read_referenced_name_from_attr

Factor out the common handling of DW_AT_abstract_origin and
DW_AT_specification from read_function_entry and read_referenced_name.

2019-01-16  Tom de Vries  <tdevries@suse.de>

* dwarf.c (read_referenced_name_from_attr): New function.  Factor out
of ...
  (read_referenced_name): ... here, and ...
(read_function_entry): ... here.

From-SVN: r267986

[D] Fix failing EH execution test on i386.

Turn off partitioning unless it was explicitly requested, as it doesn't
work with D exception chaining, where personality routines use LSDA to
determine whether two thrown exceptions are in the same context.

The following distills what was failing in the D testsuite.
```
try {
  try {
    fn();  // throws "1"
  }
  finally {
    throw new Exception("2");
  }
}
catch (Exception e) {
  assert(e.msg == "1");
  assert(e.next.msg == "2");
}
```

gcc/d/ChangeLog:

PR d/87824
* d-lang.cc (d_post_options): Disable implicit
-forder-blocks-and-partition.

From-SVN: r267985

Fix ICE due to "combine" creating unreachable EH blocks (PR target/88861)

PR target/88861 reports an ICE in "ce2" due to an unreachable
basic block.

The block becomes unreachable in "combine" when delete_noop_moves
deletes an insn with a REG_EH_REGION, deleting the EH edge, the
only edge leading to the basic block.

Normally, rest_of_handle_combine would call cleanup_cfg, deleting
unreachable blocks, if combine_instructions returns true, and
combine_instructions does return true for some cases of edge-removal,
but it doesn't for this case, leading to the ICE.

This patch updates delete_noop_moves so that it returns true if
it deletes any edges, and passes that through to combine_instructions,
so that it too will return true if any edges were deleted, ensuring that
cleanup_cfg will be called by rest_of_handle_combine for this case,
deleting the now-unreachable block, and fixing the ICE.

gcc/ChangeLog:
PR target/88861
* combine.c (delete_noop_moves): Convert to "bool" return,
returning true if any edges are eliminated.
(combine_instructions): Also return true if delete_noop_moves
returns true.

gcc/testsuite/ChangeLog:
PR target/88861
* g++.dg/torture/pr88861.C: New test.

From-SVN: r267984