re PR testsuite/90615 (several vector test cases fail starting with r271553)

2019-05-27  Richard Biener  <rguenther@suse.de>

PR testsuite/90615
* gcc.dg/vect/costmodel/ppc/costmodel-vect-31b.c: Add
-fno-tree-loop-distribute-patterns.
* gcc.dg/vect/costmodel/ppc/costmodel-vect-33.c: Likewise.
* gcc.dg/vect/costmodel/ppc/costmodel-vect-76a.c: Likewise.
* gcc.dg/vect/costmodel/ppc/costmodel-vect-76b.c: Likewise.
* gcc.dg/vect/costmodel/ppc/costmodel-vect-76c.c: Likewise.
* gcc.target/powerpc/vsx-vectorize-1.c: Likewise.

From-SVN: r271645

Daily bump.

From-SVN: r271644

re PR go/90635 (typo in libgo/configure.ac)

PR go/90635
    libgo: correct typo in USE_LIBFFI AM_CONDITIONAL

    Only affects the case of passing --without-libffi to configure.

    Fixes https://gcc.gnu.org/PR90635

    Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/178998

From-SVN: r271640

re PR go/90614 (gcc-9.1.0/libgo/go/syscall/wait.c:54:22: error: unused parameter ‘w’ [-Werror=unused-parameter] Continued (uint32_t *w))

PR go/90614
    syscall: avoid unused parameter error if WIFCONTINUED not defined

    Fixes https://gcc.gnu.org/PR90614

    Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/178997

From-SVN: r271638

i386.md (@leave_<mode>): New expander.

* config/i386/i386.md (@leave_<mode>): New expander.
(*leave): Rename from leave.
(*leave_rex64): Rename from leave_rex64.
(@monitorx_<mode>): Rename from monitorx_<mode>.
(@clzero_<mode>): Rename from clzero_<mode>.
* config/i386/sse.md (@sse3_monitor_<mode>): Rename from
sse3_monitor_<mode>.
* config/i386/i386.c (*ix86_gen_leave): Remove indirect function.
(*ix86_gen_monitor): Ditto.
(*ix86_gen_monitorx): Ditto.
(*ix86_gen_clzero): Ditto.
(*ix86_gen_one_cmpl2): Ditto.
(ix86_emit_leave): Use gen_leave instead of ix86_gen_leave.
* config/i386/i386-expand.c (ix86_expand_builtin)
<case IX86_BUILTIN_MONITOR>: Use gen_sse3_monitor
instead of ix86_gen_monitor.
<case IX86_BUILTIN_MONITORX>: Use gen_monitorx
instead of ix86_gen_monitorx.
<case IX86_BUILTIN_CLZERO>: Use gen_clzero
instead of ix86_gen_clzero.
* config/i386/i386-options.c (ix86_option_override_internal):
Do not initialize ix86_gen_leave, ix86_gen_monitor,
ix86_gen_monitorx, ix86_gen_clzero and ix86_gen_one_cmpl2.

* config/i386/i386.md (@tls_global_dynamic_64_<mode>):
Rename from tls_global_dynamic_64_<mode>.
(@tls_local_dynamic_base_64_<mode>): Rename from
tls_local_dynamic_base_64_<mode>.
* config/i386/i386.c (*ix86_gen_tls_global_dynamic_64):
Remove indirect function.
(*ix86_gen_tls_local_dynamic_base_64): Ditto.
(legitimize_tls_address): Use gen_tls_global_dynamic_64 function
instead of ix86_gen_tls_global_dynamic_64.
Use gen_tls_local_dynamic_base_64 instead of
ix86_gen_tls_local_dynamic_base_64.
* config/i386/i386-options.c (ix86_option_override_internal):
Do not initialize ix86_gen_tls_global_dynamic_64 and
ix86_gen_tls_local_dynamic_base_64.

* config/i386/i386.md (@pro_epilogue_adjust_stack_add_<mode>)
Rename from pro_epilogue_adjust_stack_<mode>_add.
(@pro_epilogue_adjust_stack_sub_<mode>)
Rename from pro_epilogue_adjust_stack_<mode>_sub.
(@allocate_stack_worker_probe_<mode>):
Rename from allocate_stack_worker_probe_<mode>.
(allocate_stack): Use gen_allocate_stack_worker_probe.
(probe_stack): Use gen_probe_stack_1.
(@probe_stack_1_<mode>): Rename from probe_stack_<mode>.
(@adjust_stack_and_probe_<mode>): Rename from
adjust_stack_and_probe<mode>.
(@probe_stack_range_<mode>): Rename from probe_stack_range<mode>.
(stack_protect_set): Use gen_stack_protect_set_1.
(@stack_protect_set_1_<mode>): Rename from stack_protect_set_<mode>.
(stack_protect_test): Use gen_stack_protect_test_1.
(@stack_protect_test_1_<mode>): Rename from stack_protect_test_<mode>.
* config/i386/i386.c (*ix86_gen_allocate_stack_worker):
Remove indirect function.
(*ix86_gen_adjust_stack_and_probe): Ditto.
(*ix86_gen_probe_stack_range): Ditto.
(pro_epilogue_adjust_stack): Use gen_pro_epilogue_adjust_stack_add
instead of gen_pro_epilogue_adjust_stack_{si,di}_add.
(ix86_adjust_stack_and_probe_stack_clash): Use
gen_adjust_stack_and_probe instead of ix86_gen_adjust_stack_and_probe.
(ix86_adjust_stack_and_probe): Ditto.
(ix86_emit_probe_stack_range): Use gen_probe_stack_range instead
of ix86_gen_probe_stack_range.
(ix86_expand_prologue):  Use gen_pro_epilogue_adjust_stack_sub
instead of gen_pro_epilogue_adjust_stack_{si,di}_sub.
* config/i386/x86-tune-sched.c (ix86_macro_fusion_pair_p):
Include insn-opinit.h.  Use code_for_stack_protect_test_1 instead of
CODE_FOR_stack_protect_test_{si,di}.
* config/i386/i386-options.c (ix86_option_override_internal):
Do not initialize ix86_gen_allocate_stack_worker,
ix86_gen_adjust_stack_and_probe and ix86_gen_probe_stack_range.

From-SVN: r271636

appendix_contributing.xml: Update pointer to C++ standard at ansi.org.

* doc/xml/manual/appendix_contributing.xml: Update pointer to
C++ standard at ansi.org.

From-SVN: r271635

darwin, x86, testsuite - Amend three PIE tests,

Darwin requires PIC code in order to support PIE, amend the test scan-asms
to match this.

2019-05-26  Iain Sandoe  <iain@sandoe.co.uk>

* gcc.target/i386/pr39013-1.c: Adjust scan-asms for PIE to
account for PIC code on Darwin.
* gcc.target/i386/pr39013-2.c: Likewise.
* gcc.target/i386/pr64317.c: Likewise.

From-SVN: r271634

invoke.texi (Link Options): Many editorial changes around -flinker-output.

* doc/invoke.texi (Link Options): Many editorial changes around
-flinker-output.

From-SVN: r271633

linux-unwind.h (pa32_fallback_frame_state): Add cast.

* config/pa/linux-unwind.h (pa32_fallback_frame_state): Add cast.

From-SVN: r271631

re PR fortran/90539 (481.wrf slowdown by 25% on Intel Kaby with -Ofast -march=native starting with r271377)

2019-05-26  Thomas Koenig  <tkoenig@gcc.gnu.org>

PR fortran/90539
* trans-types.c (get_formal_from_actual_arglist): Set rank
and lower bound for assumed size arguments.

From-SVN: r271630

Added ChangeLog entry for last commit.

From-SVN: r271629

Remove Solaris 10-only note about -mvect8-ret-in-mem

From-SVN: r271628

Daily bump.

From-SVN: r271627

darwin, x86, testsuite - Amend popcnt match string for Darwin.

Darwin uses a different spelling for popcnt (popcnt instead of popcntw).
Check for this in the test.

2019-05-25  Iain Sandoe  <iain@sandoe.co.uk>

* gcc.target/i386/pr59874-3.c: Use the spelling of popcnt
expected for Darwin.

From-SVN: r271623

darwin, x86, testsuite - Check for an expected error on Darwin.

The test requests an alignment which exceeds the maximum object
file aligment for Darwin, rather than skipping it - test that we
see the expected error.

2019-05-25  Iain Sandoe  <iain@sandoe.co.uk>

* gcc.target/i386/pr89261.c: Test that the alignment required
by the test correctly produces the expected error on Darwin.

From-SVN: r271622

darwin, x86, testsuite - require alias on a test using it.

2019-05-25  Iain Sandoe  <iain@sandoe.co.uk>

* gcc.target/i386/pr82659-3.c: Require alias support.

From-SVN: r271621

PR c++/90572 - wrong disambiguation in friend declaration.

* parser.c (cp_parser_constructor_declarator_p): Don't allow missing
typename for friend declarations.

* g++.dg/cpp2a/typename16.C: New test.
* g++.dg/parse/friend13.C: New test.

From-SVN: r271619

Daily bump.

From-SVN: r271618

re PR target/90530 (Invalid SUBREG insn generated by reload)

PR target/90530
* config/pa/pa.c (pa_can_change_mode_class): Accept mode changes from
DImode to SImode in floating-point registers on 64-bit target.
* config/pa/pa.md (umulsidi3): Change nonimmediate_operand to
register_operand in xmpyu patterns.

From-SVN: r271611

tree-core.h (enum omp_clause_code): Add OMP_CLAUSE__CONDTEMP_.

* tree-core.h (enum omp_clause_code): Add OMP_CLAUSE__CONDTEMP_.
* tree.h (OMP_CLAUSE_DECL): Use OMP_CLAUSE__CONDTEMP_ instead of
OMP_CLAUSE__REDUCTEMP_.
* tree.c (omp_clause_num_ops, omp_clause_code_name): Add
OMP_CLAUSE__CONDTEMP_.
(walk_tree_1): Handle OMP_CLAUSE__CONDTEMP_.
* tree-pretty-print.c (dump_omp_clause): Likewise.
* tree-nested.c (convert_nonlocal_omp_clauses,
convert_local_omp_clauses): Likewise.
* gimplify.c (enum gimplify_omp_var_data): Use hexadecimal constants
instead of decimal.  Add GOVD_LASTPRIVATE_CONDITIONAL.
(gimplify_scan_omp_clauses): Don't reject lastprivate conditional
on OMP_FOR.
(gimplify_omp_for): Warn and disable conditional modifier from
lastprivate on loop iterators.
* omp-general.h (struct omp_for_data): Add lastprivate_conditional
member.
* omp-general.c (omp_extract_for_data): Initialize it.
* omp-low.c (struct omp_context): Add lastprivate_conditional_map
member.
(delete_omp_context): Delete it.
(lower_lastprivate_conditional_clauses): New function.
(lower_lastprivate_clauses): Add BODY_P and CSTMT_LIST arguments,
handle lastprivate conditional clauses.
(lower_reduction_clauses): Add CLIST argument, emit it into
the critical section if any.
(lower_omp_sections): Adjust lower_lastprivate_clauses and
lower_reduction_clauses callers.
(lower_omp_for_lastprivate): Add CLIST argument, pass it through
to lower_lastprivate_clauses.
(lower_omp_for): Call lower_lastprivate_conditional_clauses, adjust
lower_omp_for_lastprivate and lower_reduction_clauses callers, emit
clist into a critical section if not emitted there already by
lower_reduction_clauses.
(lower_omp_taskreg, lower_omp_teams): Adjust lower_reduction_clauses
callers.
(lower_omp_1): Handle GIMPLE_ASSIGNs storing into lastprivate
conditional variables.
* omp-expand.c (determine_parallel_type): Punt if OMP_CLAUSE__CONDTEMP_
clause is present.
(expand_omp_for_generic, expand_omp_for_static_nochunk,
expand_omp_for_static_chunk): Handle lastprivate conditional.
(expand_omp_for): Handle fd.lastprivate_conditional like
fd.have_reductemp.
gcc/testsuite/
* c-c++-common/gomp/lastprivate-conditional-2.c (foo): Don't expect
sorry for omp for.
* c-c++-common/gomp/lastprivate-conditional-3.c: New test.
libgomp/
* testsuite/libgomp.c-c++-common/lastprivate-conditional-1.c: New test.
* testsuite/libgomp.c-c++-common/lastprivate-conditional-2.c: New test.

From-SVN: r271610

Fix std::midpoint(T*, T*) for reversed arguments

* include/std/numeric (midpoint(T*, T*)): Fix incorrect result.
* testsuite/26_numerics/midpoint/pointer.cc: Change "compile" test
to "run".

From-SVN: r271606

GCN: Wait for exit value to write before exiting.

2019-05-24  Andrew Stubbs  <ams@codesourcery.com>

gcc/
* config/gcn/gcn-run.c (main): Set a non-zero return value if the
kernel does not exit cleanly.
* config/gcn/gcn.md (gcn_return): Insert s_waitcnt before s_dcache_wb.

From-SVN: r271605

Revert "* gimplify.c (gimplify_cond_expr): Don't check TREE_ADDRESSABLE."

This reverts commit ac0f04360cc04e5b80a7d74f7edc47e395d2e744.

From-SVN: r271604

Fix broken shared_ptr test

* testsuite/20_util/shared_ptr/cons/alias-rval.cc: Fix test.
* testsuite/20_util/shared_ptr/cons/alias.cc: Remove unused function.

From-SVN: r271603

Makefile.am (BUILDTESTS): Remove test_elf, add test_elf_32 and test_elf_64.

* Makefile.am (BUILDTESTS): Remove test_elf, add test_elf_32 and
test_elf_64.
* Makefile.in: Regenerate.

From-SVN: r271602

re PR middle-end/90607 (gcc.dg/pr53265.c FAILs)

2019-05-24  Richard Biener  <rguenther@suse.de>

PR testsuite/90607
* tree-loop-distribution.c (struct partition): Add location
member.
(partition_alloc): Initialize all fields.
(generate_memset_builtin): Use the location recorded in the
partition for the generated call.
(generate_memcpy_builtin): Likewise.
(classify_partition): Record the location of a single store
as location for the partition.

* gcc.dg/pr53265.c: Amend for new expected diagnostic.

From-SVN: r271601

Fix 64-bit addition in prologue.

2019-05-24  Andrew Stubbs  <ams@codesourcery.com>

gcc/
* config/gcn/gcn.c (gcn_expand_prologue): Use gen_addsi3_scalar_carry
for lo-part.

From-SVN: r271600

[aarch64] Change two function declaration types

Commit r271514 missed changing the type of two functions in
aarch64-protos.h.  The function definitions had been updated to use
uint64_t while the function declarations had been missed.
They were missed since I only tested the patch on aarch64 where
`unsigned long` is the same as `uint64_t`.

This patch updates these declarations in aarch64-protos.h.

Tested by building an aarch64 cross-compiler on arm-none-linux-gnu (so
that `unsigned long` and `uint64_t` are different and would give error
messages), and bootstrapping on aarch64-none-linux-gnu.
Also manually tested command line options to see that
-march=armv8-a+typo prints out the expected flags while using the new
feature flags does not complain about missing flags.

gcc/ChangeLog:

2019-05-24  Matthew Malcomson  <matthew.malcomson@arm.com>

PR target/90588
* common/config/aarch64/aarch64-common.c
(aarch64_rewrite_selected_cpu): Change local temporary variable
type from unsigned long to uint64_t.
* config/aarch64/aarch64-protos.h (aarch64_parse_extension,
aarch64_get_extension_string_for_isa_flags): Change declaration to
match new definition by replacing unsigned long with uint64_t.

From-SVN: r271599

re PR tree-optimization/90106 (builtin sqrt() ignoring libm's sqrt call result)

PR tree-optimization/90106
PR testsuite/90517
* gcc.dg/cdce1.c: Don't scan-assembler, instead -fdump-tree-optimized
and scan-tree-dump for tail call.
* gcc.dg/cdce2.c: Likewise.

From-SVN: r271598

re PR libgomp/90585 (libgomp hsa plugin ftbfs in the x32 multilib variant)

PR libgomp/90585
* plugin/plugin-hsa.c: Include gstdint.h.  Include inttypes.h only if
HAVE_INTTYPES_H is defined.
(print_uint64_t): New typedef.
(PRIu64): Define if HAVE_INTTYPES_H is not defined.
(print_kernel_dispatch, run_kernel): Use PRIu64 macro instead of
"lu", cast uint64_t HSA_DEBUG and fprintf arguments to print_uint64_t.
(release_kernel_dispatch): Likewise.  Cast shadow->debug to uintptr_t
before casting to void *.
* plugin/plugin-nvptx.c: Include gstdint.h instead of stdint.h.
* oacc-mem.c: Don't include config.h nor stdint.h.
* target.c: Don't include config.h.
* oacc-cuda.c: Likewise.
* oacc-host.c: Don't include stdint.h.

From-SVN: r271597

re PR target/90568 (stack protector should use cmp or sub, not xor, to allow macro-fusion on x86)

PR target/90568
* config/i386/x86-tune-sched.c (ix86_macro_funsion_pair_p): Call
gen_attr_type just once instead of 4-7 times.  Formatting fixes.
Handle stack_protect_test_<mode> codegen similarly to corresponding
sub instruction.

From-SVN: r271596

re PR tree-optimization/88440 (size optimization of memcpy-like code)

2019-05-23  Richard Biener  <rguenther@suse.de>

PR tree-optimization/88440
* opts.c (default_options_table): Enable -ftree-loop-distribute-patterns
at -O[2s]+.
* tree-loop-distribution.c (generate_memset_builtin): Fold the
generated call.
(generate_memcpy_builtin): Likewise.
(distribute_loop): Pass in whether to only distribute patterns.
(prepare_perfect_loop_nest): Also allow size optimization.
(pass_loop_distribution::execute): When optimizing a loop
nest for size allow pattern replacement.

* gcc.dg/tree-ssa/ldist-37.c: New testcase.
* gcc.dg/tree-ssa/ldist-38.c: Likewise.
* gcc.dg/vect/vect.exp: Add -fno-tree-loop-distribute-patterns.
* gcc.dg/tree-ssa/ldist-37.c: Adjust.
* gcc.dg/tree-ssa/ldist-38.c: Likewise.
* g++.dg/tree-ssa/pr78847.C: Likewise.
* gcc.dg/autopar/pr39500-1.c: Likewise.
* gcc.dg/autopar/reduc-1char.c: Likewise.
* gcc.dg/autopar/reduc-7.c: Likewise.
* gcc.dg/tree-ssa/ivopts-lt-2.c: Likewise.
* gcc.dg/tree-ssa/ivopts-lt.c: Likewise.
* gcc.dg/tree-ssa/predcom-dse-1.c: Likewise.
* gcc.dg/tree-ssa/predcom-dse-2.c: Likewise.
* gcc.dg/tree-ssa/predcom-dse-3.c: Likewise.
* gcc.dg/tree-ssa/predcom-dse-4.c: Likewise.
* gcc.dg/tree-ssa/prefetch-7.c: Likewise.
* gcc.dg/tree-ssa/prefetch-8.c: Likewise.
* gcc.dg/tree-ssa/prefetch-9.c: Likewise.
* gcc.dg/tree-ssa/scev-11.c: Likewise.
* gcc.dg/vect/costmodel/i386/costmodel-vect-31.c: Likewise.
* gcc.dg/vect/costmodel/i386/costmodel-vect-33.c: Likewise.
* gcc.dg/vect/costmodel/x86_64/costmodel-vect-31.c: Likewise.
* gcc.dg/vect/costmodel/x86_64/costmodel-vect-33.c: Likewise.
* gcc.target/i386/pr30970.c: Likewise.
* gcc.target/i386/vect-double-1.c: Likewise.
* gcc.target/i386/vect-double-2.c: Likewise.
* gcc.dg/tree-ssa/gen-vect-2.c: Likewise.
* gcc.dg/tree-ssa/gen-vect-26.c: Likewise.
* gcc.dg/tree-ssa/gen-vect-28.c: Likewise.
* gcc.dg/tree-ssa/gen-vect-32.c: Likewise.
* gfortran.dg/vect/vect-5.f90: Likewise.
* gfortran.dg/vect/vect-8.f90: Likewise.

From-SVN: r271595

add missed Changelog.

From-SVN: r271592

Darwin, x86, testsuite - Make match strings more specific.

Some of the i386.exp tests fail on Darwin (and at least one passes
incorrectly on Linux) because their scan-asm match strings are too
general.  In some cases the strings also match instances in the .file
and size directives or in comment output.  This patch makes the match
strings more specific.

2019-05-24  Iain Sandoe  <iain@sandoe.co.uk>

* gcc.target/i386/pconfig-1.c: Scan for the string in the generated
code, not in comments or miscellaneous directives. Adjust expected
count.
* gcc.target/i386/pr18041-1.c: Likewise.
* gcc.target/i386/pr18041-2.c: Likewise.
* gcc.target/i386/wbinvd-1.c: Likewise.
* gcc.target/i386/wbnoinvd-1.c: Likewise.
* gcc.target/i386/pr66819-3.c: Specifically, check that there is no
call to "bar".
* gcc.target/i386/pr66819-4.c: Likewise.
* gcc.target/i386/pr82662.c
* gcc.target/i386/ptwrite2.c: Make the checks look for the specific
destination register, don't try the m32 test on m64 targets.

From-SVN: r271591

Darwin, x86, testsuite - Adjust test labels.

A couple of the i386.exp tests fail because the label spelling used
is different between ELF and Mach-O targets. Adjusted here.

2019-05-24  Iain Sandoe  <iain@sandoe.co.uk>

* gcc.target/i386/pr67985-2.c: Adjust label checks for
Darwin.
* gcc.target/i386/pr77881.c: Likewise.

From-SVN: r271590

Darwin, x86, testsuite - Disable tests that cannot pass.

A few of the i386.exp target tests cannot pass on Darwin
because either the port doesn't support a feature, or the
ABI demands sufficiently different codegen that matching
the output with scan-asms would require a different test.

Disabling those tests here.

2019-05-24  Iain Sandoe  <iain@sandoe.co.uk>

* gcc.target/i386/falign-functions-2.c: Skip for Darwin.
* gcc.target/i386/pr70738-7.c: Likewise.
* gcc.target/i386/pr24414.c: Likewise.

From-SVN: r271589

Daily bump.

From-SVN: r271588

Fix testsuite bugs

One of the static assertions in 20_util/function_objects/invoke/1.cc was
wrong, but didn't fail because by default it was compiled with
-std=gnu++14 which didn't use that static assertion. Split out the C++17
parts to a new file that always runs with -std=gnu++17, so those checks
are always done.

The 23_containers/unordered_set/allocator/ext_ptr.cc test is supposed to
be a run-time test but was unintentionally compile-only.

* testsuite/20_util/function_objects/invoke/1.cc: Move C++17-specific
tests to ...
* testsuite/20_util/function_objects/invoke/3.cc: New test.
* testsuite/23_containers/unordered_set/allocator/ext_ptr.cc: Change
"compile" test to "run".

From-SVN: r271584

LWG 2996 add rvalue overloads for shared_ptr aliasing and casting

* doc/xml/manual/intro.xml: Document LWG DR 2996 change.
* doc/html/*: Regenerate.
* include/bits/shared_ptr.h (shared_ptr(shared_ptr&&, T*)): Add
rvalue aliasing constructor.
(static_pointer_cast, const_pointer, dynamic_pointer_cast)
(reinterpret_pointer_cast): Add overloads taking rvalues.
* include/bits/shared_ptr_base.h (__shared_ptr(__shared_ptr&&, T*)):
Add rvalue aliasing constructor.
* testsuite/20_util/shared_ptr/casts/1.cc: Change "compile" test to
"run" and check return values as well as types.
* testsuite/20_util/shared_ptr/casts/reinterpret.cc: Likewise.
* testsuite/20_util/shared_ptr/casts/rval.cc: New test.
* testsuite/20_util/shared_ptr/cons/alias-rval.cc: New test.
* testsuite/20_util/shared_ptr/cons/alias.cc: Remove unused return
values.

From-SVN: r271583

LWG 2921 remove packaged_task constructors taking allocators

* doc/xml/manual/evolution.xml: Document LWG DR 2921 change.
* doc/xml/manual/intro.xml: Likewise.
* include/std/future (__create_task_state): Add default arguments
to make providing an allocator optional.
(packaged_task::packaged_task(F&&)): Call __create_task_state directly
instead of delegating to another constructor.
(packaged_task::packaged_task(allocator_arg_t, const A&, ...)): Do not
define allocator-extended constructors for C++17 and later.
* testsuite/30_threads/packaged_task/cons/alloc.cc: Only run test for
C++11 and C++14.
* testsuite/30_threads/packaged_task/cons/alloc2.cc: Likewise.
* testsuite/30_threads/packaged_task/cons/alloc_min.cc: Likewise.
* testsuite/30_threads/packaged_task/uses_allocator.cc: Likewise.

From-SVN: r271582

darwin, x86 - disable '-mfentry' for the port.

This can't work without special support because placing
the call directly after the function label breaks the
ABI requirement that call sites are 16byte aligned.

gcc/

2019-05-23  Iain Sandoe  <iain@sandoe.co.uk>

       * config/i386/darwin.h: Reject -mfentry*.
       * doc/sourcebuild.texi: Document mfentry target support.

gcc/testsuite/

2019-05-23  Iain Sandoe  <iain@sandoe.co.uk>

       * lib/target-supports.exp (check_effective_target_mfentry): New.
       * gcc.target/i386/fentry-override.c: Require effective target mfentry.
       * gcc/testsuite/gcc.target/i386/fentry.c: Likewise
       * gcc.target/i386/fentryname1.c: Likewise
       * gcc.target/i386/fentryname2.c: Likewise
       * gcc.target/i386/fentryname3.c: Likewise
       * gcc.target/i386/nop-mcount.c: Likewise
       * gcc.target/i386/pr82699-2.c: Likewise
       * gcc.target/i386/pr82699-4.c: Likewise
       * gcc.target/i386/pr82699-5.c: Likewise
       * gcc.target/i386/pr82699-6.c: Likewise
       * gcc.target/i386/returninst1.c: Likewise
       * gcc.target/i386/returninst2.c: Likewise
       * gcc.target/i386/returninst3.c : Likewise

From-SVN: r271580

rs6000.c (rs6000_global_entry_point_needed_p): Rename to rs6000_global_entry_point_prologue_needed_p.

[gcc]

2019-05-22  Bill Schmidt  <wschmidt@linux.ibm.com>

* config/rs6000/rs6000.c (rs6000_global_entry_point_needed_p):
Rename to rs6000_global_entry_point_prologue_needed_p.  Return
false for PC-relative functions.
(rs6000_output_function_prologue): Change called function name to
rs6000_global_entry_point_prologue_needed_p.  Emit ".localentry
name,1" for PC-relative functions.
(rs6000_elf_declare_function_name): Change called function name to
rs6000_global_entry_point_prologue_needed_p.

[gcc/testsuite]

2019-05-22  Bill Schmidt  <wschmidt@linux.ibm.com>

* gcc.target/powerpc/localentry-1.c: New file.

From-SVN: r271577

re PR target/90552 (attribute((optimize(3))) not overriding -Os)

PR target/90552
* config/i386/i386.c (gen_rtx_cost):
Use ix86_tune_cost instead of ix86_cost.

testsuite/ChangeLog:

PR target/90552
* gcc.target/i386/pr90552.c: New test.

From-SVN: r271576

rs6000-cpus.def (ISA_FUTURE_MASKS_SERVER): Add OPTION_MASK_PCREL.

2019-05-22  Bill Schmidt  <wschmidt@linux.ibm.com>
    Michael Meissner  <meissner@linux.ibm.com>
    Segher Boessenkool  <segher@kernel.crashing.org>

* config/rs6000/rs6000-cpus.def (ISA_FUTURE_MASKS_SERVER): Add
OPTION_MASK_PCREL.
(POWERPC_MASKS): Add OPTION_MASK_PCREL.
* config/rs6000/rs6000-protos.h (rs6000_pcrel_p): New prototype.
(rs6000_fndecl_pcrel_p): Likewise.
* config/rs6000/rs6000.c (rs6000_option_override_internal): Report
error if -mpcrel is requested without -mcpu=future.
(rs6000_opt_masks): Add entry for pcrel.
(rs6000_fndecl_pcrel_p): New function.
(rs6000_pcrel_p): Likewise.
* config/rs6000/rs6000.opt (mpcrel): New option.
* doc/invoke.texi: Document -mpcrel and -mno-pcrel.

Co-Authored-By: Michael Meissner <meissner@linux.ibm.com>
Co-Authored-By: Segher Boessenkool <segher@kernel.crashing.org>
From-SVN: r271575

From what I understand of the libstdc++/83237 thread at
<https://gcc.gnu.org/ml/gcc-patches/2017-12/msg00573.html>, the
high numbers are not arbitrary, so it seems wrong to try
lowering them, or we'd just waste cycles testing nothing, or
worse, ending up with a bogus error indication.  Better to just
plain disable this part of the test for simulator targets; I
assume the results should be the same on any IEEE-float target,
i.e. no target-specific things going on here that'd raise a need
to cover it everywhere.

With this part of the test disabled, I saw the test finishing in
(time) "124.74s user" where it was before "1120.26s user"
running the cris-elf-run simulator on a "i7-4770K CPU @ 3.50GHz"
host.  Most certainly that indidates that the remainder of the
test is still too much for *some* host+simulator combos, but I'm
happy with the runtime lowered to 1/5 of the timeout (10
minutes) on this particular combination, and I'd think this
fixes timeouts for many other simulator combos too.

This construct (disabling or lowering limits for simulators) is
used elsewhere in the libstdc++ test-suite and in particular the
SIMULATOR_TEST macro is used in the testsuite machinery (though
AFAICT not in testDiscreteDist).

        * testsuite/26_numerics/random/poisson_distribution/operators/values.cc:
        Don't run the libstdc++/83237 part on simulator targets.

From-SVN: r271574

fmt_f_default_field_width_1.f90: Hide REAL(16) behind __GFC_REAL_16__.

2019-05-23  Mark Eggleston  <mark.eggleston@codethink.com>

* gfortran.dg/fmt_f_default_field_width_1.f90: Hide REAL(16) behind
__GFC_REAL_16__.  Add -cpp to dg-options.
* gfortran.dg/fmt_f_default_field_width_2.f90: Ditto.
* gfortran.dg/fmt_f_default_field_width_3.f90: Ditto.
* gfortran.dg/fmt_g_default_field_width_1.f90: Ditto.
* gfortran.dg/fmt_g_default_field_width_2.f90: Ditto.
* gfortran.dg/fmt_g_default_field_width_3.f90: Ditto.

From-SVN: r271573

re PR tree-optimization/90576 (SPEC CPU2006 450.soplex miscompiled with -Os -flto after r271413)

PR tree-optimization/90576
* tree-ssa-alias.c (compare_sizes): Remove dead calls to
poly_int_tree_p.
(aliasing_component_refs_p): Fix three way size compare conditional;
give up earlier in case we can not decide on equivalence.

Co-Authored-By: Martin Liska <mliska@suse.cz>
From-SVN: r271572

Add missing feature test macro to C++17 status table

* doc/xml/manual/status_cxx2017.xml: Add feature test macro for
P0040R3.
* doc/html/*: Regenerate.

From-SVN: r271571

Add infrastructure to support -mcpu=future to represent a future
architecture level, as yet unnamed.

[gcc]

2019-05-22  Bill Schmidt  <wschmidt@linux.ibm.com>
    Michael Meissner  <meissner@linux.ibm.com>
    Segher Boessenkool  <segher@kernel.crashing.org>

* config.gcc: Add future cpu.
* config/rs6000/driver-rs6000.c (asm_names): Add future cpu.
* config/rs6000/rs6000-cpus.def (ISA_FUTURE_MASKS_SERVER): New
#define.
(POWERPC_MASKS): Add OPTION_MASK_FUTURE.
(RS6000_CPU): New instantiation for future cpu.
* config/rs6000/rs6000-opts.h (enum processor_type): Add
PROCESSOR_FUTURE.
* config/rs6000/rs6000-string.c (expand_compare_loop): Treat
PROCESSOR_FUTURE like PROCESSOR_POWER9 for now.
* config/rs6000/rs6000-tables.opt: Regenerate.
* config/rs6000/rs6000.c (rs6000_option_override_internal): Treat
PROCESSOR_FUTURE similarly to PROCESSOR_POWER9 for now.
(rs6000_machine_from_flags): Handle future cpu.
(rs6000_reassociation_width): Treat PROCESSOR_FUTURE like
PROCESSOR_POWER9 for now.
(rs6000_adjust_cost): Likewise.
(rs6000_issue_rate): Likewise.
(rs6000_register_move_cost): Likewise.
(rs6000_opt_mask): Add entry for future.
* config/rs6000/rs6000.h (ASM_CPU_SPEC): Add future cpu.
(MASK_FUTURE): New #define.
* config/rs6000/rs6000.md (define_attr "cpu"): Add future cpu.
* config/rs6000/rs6000.opt (mfuture): New target option.
* doc/invoke.texi (mcpu): Add future cpu.

[gcc/testsuite]

2019-05-22  Bill Schmidt  <wschmidt@linux.ibm.com>

* gcc.target/powerpc/cpu-future.c: New test.

Co-Authored-By: Michael Meissner <meissner@linux.ibm.com>
Co-Authored-By: Segher Boessenkool <segher@kernel.crashing.org>
From-SVN: r271567

Make any_cast compare typeinfo as well as function pointers

It's possible for the function pointer comparison to fail even though
the type is correct, because the function could be defined multiple
times with different addresses when shared libraries are in use.

Retain the function pointer check for the common case where the check
succeeds, but compare typeinfo (if RTTI is enabled) if the first check
fails.

* include/experimental/any (__any_caster): Use RTTI if comparing
addresses fails, to support non-unique addresses in shared libraries.
* include/std/any (__any_caster): Likewise.

From-SVN: r271557

PR libstdc++/90220 fix experimental::any_cast for non-object types

This corresponds to the fixes done for std::any_cast, but has to be done
without if-constexpr. The dummy specialization of _Manager_internal<_Op>
is used to avoid instantiating the real _Manager_internal<T>::_S_manage
function just to compare its address.

PR libstdc++/90220
* include/experimental/any (__any_caster): Constrain to only be
callable for object types. Use remove_cv_t instead of decay_t.
If the type decays or isn't copy constructible, compare the manager
function to a dummy specialization.
(__any_caster): Add overload constrained for non-object types.
(any::_Manager_internal<_Op>): Add dummy specialization.
* testsuite/experimental/any/misc/any_cast.cc: Test function types
and array types.

From-SVN: r271556

Do not use tree_to_wide_ref that point to a temporary (PR c++/90587).

2019-05-23  Martin Liska  <mliska@suse.cz>

PR c++/90587
* tree-ssa-uninit.c (value_sat_pred_p): The result of &
operation points to a temporary (pointed via tree_to_wide_ref)
that is out of scope after the &.

From-SVN: r271555

PR c++/90592 add missing word "scope" to __func__ docs

PR c++/90592
* doc/extend.texi (Function Names): Add missing word.

From-SVN: r271554

re PR tree-optimization/88440 (size optimization of memcpy-like code)

2019-05-23  Richard Biener  <rguenther@suse.de>

PR tree-optimization/88440
* opts.c (default_options_table): Enable -ftree-loop-distribute-patterns
at -O[2s]+.
* tree-loop-distribution.c (generate_memset_builtin): Fold the
generated call.
(generate_memcpy_builtin): Likewise.
(distribute_loop): Pass in whether to only distribute patterns.
(prepare_perfect_loop_nest): Also allow size optimization.
(pass_loop_distribution::execute): When optimizing a loop
nest for size allow pattern replacement.

* gcc.dg/tree-ssa/ldist-37.c: New testcase.
* gcc.dg/tree-ssa/ldist-38.c: Likewise.
* gcc.dg/vect/vect.exp: Add -fno-tree-loop-distribute-patterns.
* gcc.dg/tree-ssa/ldist-37.c: Adjust.
* gcc.dg/tree-ssa/ldist-38.c: Likewise.
* g++.dg/tree-ssa/pr78847.C: Likewise.
* gcc.dg/autopar/pr39500-1.c: Likewise.
* gcc.dg/autopar/reduc-1char.c: Likewise.
* gcc.dg/autopar/reduc-7.c: Likewise.
* gcc.dg/tree-ssa/ivopts-lt-2.c: Likewise.
* gcc.dg/tree-ssa/ivopts-lt.c: Likewise.
* gcc.dg/tree-ssa/predcom-dse-1.c: Likewise.
* gcc.dg/tree-ssa/predcom-dse-2.c: Likewise.
* gcc.dg/tree-ssa/predcom-dse-3.c: Likewise.
* gcc.dg/tree-ssa/predcom-dse-4.c: Likewise.
* gcc.dg/tree-ssa/prefetch-7.c: Likewise.
* gcc.dg/tree-ssa/prefetch-8.c: Likewise.
* gcc.dg/tree-ssa/prefetch-9.c: Likewise.
* gcc.dg/tree-ssa/scev-11.c: Likewise.
* gcc.dg/vect/costmodel/i386/costmodel-vect-31.c: Likewise.
* gcc.dg/vect/costmodel/i386/costmodel-vect-33.c: Likewise.
* gcc.dg/vect/costmodel/x86_64/costmodel-vect-31.c: Likewise.
* gcc.dg/vect/costmodel/x86_64/costmodel-vect-33.c: Likewise.
* gcc.target/i386/pr30970.c: Likewise.
* gcc.target/i386/vect-double-1.c: Likewise.
* gcc.target/i386/vect-double-2.c: Likewise.
* gcc.dg/tree-ssa/gen-vect-2.c: Likewise.
* gcc.dg/tree-ssa/gen-vect-26.c: Likewise.
* gcc.dg/tree-ssa/gen-vect-28.c: Likewise.
* gcc.dg/tree-ssa/gen-vect-32.c: Likewise.
* gfortran.dg/vect/vect-5.f90: Likewise.
* gfortran.dg/vect/vect-8.f90: Likewise.

From-SVN: r271553

re PR target/90568 (stack protector should use cmp or sub, not xor, to allow macro-fusion on x86)

PR target/90568
* config/i386/i386.md (stack_protect_test_<mode>): Use sub instead
of xor.

From-SVN: r271552

* c-ada-spec.c (compare_node): Compare the DECL_UIDs as a last resort.

From-SVN: r271549

Do not instrument static target_expr for use-after-scope (PR sanitizer/90570).

2019-05-23  Martin Liska  <mliska@suse.cz>

PR sanitizer/90570
* gimplify.c (gimplify_target_expr): Skip TREE_STATIC target
expression similarly to gimplify_decl_expr.
2019-05-23  Martin Liska  <mliska@suse.cz>

PR sanitizer/90570
* g++.dg/asan/pr90570.C: New test.

From-SVN: r271548

* gnat.dg/opt78.ad[sb]: New test.

From-SVN: r271545

x86, testsuite - update fuse-caller-save tests.

These tests had started to XPASS on pic targets where
the codegen is now as expected.

gcc/testsuite/

2019-05-23  Iain Sandoe  <iain@sandoe.co.uk>

PR rtl-optimisation/64895
* gcc.target/i386/fuse-caller-save-rec.c: Remove XFAILs.
* gcc.target/i386/fuse-caller-save.c: Likewise.
* gcc.target/i386/fuse-caller-save-xmm.c: Adjust tests for
PIC cases, remove XFAILs.

From-SVN: r271544

Add whitespace between operator and operand

* cp-tree.h (CP_AGGREGATE_TYPE_P): Fix whitespace.

From-SVN: r271543

Remove conflict marker from ChangeLog

From-SVN: r271542

Remove resolved TODO comment

* init.c (std_placement_new_fn_p): Remove outdated TODO comment that
was resolved by r254694.

From-SVN: r271541

[PR90510] Adjust 'brig.dg/test/gimple/packed.hsail'

... for r271463 "Fix PR90510, VEC_PERM -> BIT_INSERT folding".

gcc/testsuite/
PR middle-end/90510
* brig.dg/test/gimple/packed.hsail: Adjust.

From-SVN: r271540

cse.c (cse_dump_path): s/dump_file/f.

2019-05-23  Prathamesh Kulkarni  <prathamesh.kulkarni@linaro.org>

* cse.c (cse_dump_path): s/dump_file/f.

From-SVN: r271539

* g++.dg/cpp2a/is-constant-evaluated8.C: New test.

From-SVN: r271538

There was a regression for gfortran.dg/fmt_en.f90 for cris-elf that on...

There was a regression for gfortran.dg/fmt_en.f90 for cris-elf
that on inspection was due to it having acquired a truncation
call through the runtime.  I updated that and the new tests that
had "Fortran runtime error: required ftruncate or chsize support
not present" messages in gfortran.log, ran past cris-elf and
committed as obvious.  See also
<https://gcc.gnu.org/ml/gcc-patches/2008-05/msg00975.html> from
which I copy-pasted most of this message.  (Yep, 11 years ago.)

gcc/testsuite:
* gfortran.dg/dec_io_1.f90, gfortran.dg/dtio_1.f90,
gfortran.dg/dtio_12.f90, gfortran.dg/fmt_en.f90,
gfortran.dg/namelist_89.f90: Gate test on effective_target
fd_truncate.

From-SVN: r271536

Bulletproof -fdiagnostics-format=json against bad locations (PR c++/90462)

PR c++/90462 reports an ICE with -fdiagnostics-format=json when
attempting to serialize a malformed location to JSON.

The compound location_t in question has meaningful "caret" and "start"
locations, but has UNKNOWN_LOCATION for its "finish" location,
leading to a NULL pointer dereference when attempting to build a JSON
string for the filename.

This patch bulletproofs the JSON output so that attempts to write
a JSON object for a location with a NULL file will lead to an object
with no "file" key, and attempts to write a compound location with
UNKNOWN_LOCATION for its start or finish will lead to the corresponding
JSON child object being omitted.

This patch also adds a json::object::get member function, for self-testing
the above.

gcc/ChangeLog:
PR c++/90462
* diagnostic-format-json.cc: Include "selftest.h".
(json_from_expanded_location): Only add "file" key for non-NULL
file strings.
(json_from_location_range): Don't add "start" and "finish"
children if they are UNKNOWN_LOCATION.
(selftest::test_unknown_location): New selftest.
(selftest::test_bad_endpoints): New selftest.
(selftest::diagnostic_format_json_cc_tests): New function.
* json.cc (json::object::get): New function.
(selftest::test_object_get): New selftest.
(selftest::json_cc_tests): Call it.
* json.h (json::object::get): New decl.
* selftest-run-tests.c (selftest::run_tests): Call
selftest::diagnostic_format_json_cc_tests.
* selftest.h (selftest::diagnostic_format_json_cc_tests): New
decl.

gcc/testsuite/ChangeLog:
PR c++/90462
* g++.dg/pr90462.C: New test.

From-SVN: r271535

Daily bump.

From-SVN: r271534

udlit-char-template-neg.C: Expect the error on a different line.

* g++.dg/cpp1y/udlit-char-template-neg.C: Expect the error on a
different line.  Check the column number too.

From-SVN: r271530

PR libstdc++/90557 fix path assignment that alters source

PR libstdc++/90557
* src/c++17/fs_path.cc (path::_List::operator=(const _List&)): Fix
reversed arguments to uninitialized_copy_n.
* testsuite/27_io/filesystem/path/assign/copy.cc: Check that source
is unchanged by copy assignment.
* testsuite/util/testsuite_fs.h (compare_paths): Use std::equal to
compare path components.

From-SVN: r271527

Add support for constructors and destuctors on GCN

2019-05-22  Kwok Cheung Yeung  <kcy@codesourcery.com>
            Andrew Stubbs  <amd@codesourcery.com>

gcc/
* config.gcc (gcc_cv_initfini_array): Set for AMD GCN.
* config/gcn/gcn-run.c (init_array_kernel, fini_array_kernel): New.
(kernel): Rename to...
(main_kernel): ... this.
(load_image): Load _init_array and _fini_array kernels.
(run): Add argument for kernel to run.
(main): Run init_array_kernel before main_kernel, and
fini_array_kernel after.
* config/gcn/gcn.c (gcn_handle_amdgpu_hsa_kernel_attribute): Allow
amdgpu_hsa_kernel attribute on functions.
(gcn_disable_constructors): Delete.
(TARGET_ASM_CONSTRUCTOR, TARGET_ASM_DESTRUCTOR): Delete.
* config/gcn/crt0.c (size_t): Define.
(_init_array, _fini_array): New.
(__preinit_array_start, __preinit_array_end,
__init_array_start, __init_array_end,
__fini_array_start, __fini_array_end): Declare weak references.

Co-Authored-By: Andrew Stubbs <ams@codesourcery.com>
From-SVN: r271526

Fix trampoline execution failures on GCN5.

2019-05-22  Andrew Stubbs  <ams@codesourcery.com>

gcc/
* config/gcn/gcn.c (gcn_trampoline_init): Call "sorry" on GCN5.

From-SVN: r271525

gimplify.c (gimplify_cond_expr): Don't check TREE_ADDRESSABLE.

* gimplify.c (gimplify_cond_expr): Don't check TREE_ADDRESSABLE.

The front end shouldn't produce a GENERIC COND_EXPR of TREE_ADDRESSABLE
type.

From-SVN: r271524

PR c++/20408 - unnecessary code for empty struct.

Here initializing the argument from a TARGET_EXPR isn't an empty class
copy even though the type is !TREE_ADDRESSABLE, so we should check
simple_empty_class_p.

* call.c (build_call_a): Use simple_empty_class_p.

From-SVN: r271523

PR libstdc++/77691 fix resource_adaptor failures due to max_align_t bugs

Remove the hardcoded whitelist of allocators expected to return memory
aligned to alignof(max_align_t), because that doesn't work when the
platform's malloc() and GCC's max_align_t do not agree what the largest
fundamental alignment is. It's also sub-optimal for user-defined
allocators that return memory suitable for any fundamental alignment.

Instead use a hardcoded list of alignments that are definitely supported
by the platform malloc, and use a copy of the allocator rebound to a POD
type with the requested alignment. Only allocate an oversized
buffer to use with std::align for alignments larger than any of the
hardcoded values.

For 32-bit Solaris x86 do not include alignof(max_align_t) in the
hardcoded values.

PR libstdc++/77691
* include/experimental/memory_resource: Add system header pragma.
(__resource_adaptor_common::__guaranteed_alignment): Remove.
(__resource_adaptor_common::_Types)
(__resource_adaptor_common::__new_list)
(__resource_adaptor_common::_New_list)
(__resource_adaptor_common::_Alignments)
(__resource_adaptor_common::_Fund_align_types): New utilities for
creating a list of types with fundamental alignments.
(__resource_adaptor_imp::do_allocate): Call new _M_allocate function.
(__resource_adaptor_imp::do_deallocate): Call new _M_deallocate
function.
(__resource_adaptor_imp::_M_allocate): New function that first tries
to use an allocator rebound to a type with a fundamental alignment.
(__resource_adaptor_imp::_M_deallocate): Likewise for deallocation.
* testsuite/experimental/memory_resource/new_delete_resource.cc:
Adjust expected allocation sizes.
* testsuite/experimental/memory_resource/resource_adaptor.cc: Remove
xfail.

From-SVN: r271522

PR c++/86485 - simple_empty_class_p

Yet another tweak that would have fixed this bug: we should treat INIT_EXPR
and MODIFY_EXPR differently for determining whether this is a simple empty
class copy, since a TARGET_EXPR on the RHS is direct initialization if
INIT_EXPR but copy if MODIFY_EXPR.

* cp-gimplify.c (simple_empty_class_p): Also true for MODIFY_EXPR.

From-SVN: r271521

vect-signbitf.c (dg-final): Improve scan-assembler-not string to avoid false matching on 32bit targets.

* gcc.target/i386/vect-signbitf.c (dg-final): Improve
scan-assembler-not string to avoid false matching on 32bit targets.

From-SVN: r271520

x86: Don't allocate stack frame nor align stack if not needed

get_frame_size () returns used stack slots during compilation, which
may be optimized out later.  This patch does the followings:

1. Add stack_frame_required to machine_function to indicate that the
function needs a stack frame.
2. Change ix86_find_max_used_stack_alignment to set stack_frame_required.
3. Always call ix86_find_max_used_stack_alignment to check if stack
frame is needed.

Tested on i686 and x86-64 with

--with-arch=native --with-cpu=native

Tested on AVX512 machine configured with

--with-arch=native --with-cpu=native

gcc/

PR target/88483
* config/i386/i386-options.c (ix86_init_machine_status): Set
stack_frame_required to true.
* config/i386/i386.c (ix86_get_frame_size): New function.
(ix86_frame_pointer_required): Replace get_frame_size with
ix86_get_frame_size.
(ix86_compute_frame_layout): Likewise.
(ix86_find_max_used_stack_alignment): Changed to void.  Set
stack_frame_required.
(ix86_finalize_stack_frame_flags): Always call
ix86_find_max_used_stack_alignment.  Replace get_frame_size with
ix86_get_frame_size.
* config/i386/i386.h (machine_function): Add stack_frame_required.

gcc/testsuite/

PR target/88483
* gcc.target/i386/stackalign/pr88483-1.c: New test.
* gcc.target/i386/stackalign/pr88483-2.c: Likewise.

From-SVN: r271517

sse.md (sse_cvtpi2ps): Use TARGET_MMX in insn condition.

* config/i386/sse.md (sse_cvtpi2ps): Use TARGET_MMX in insn condition.

From-SVN: r271515

[aarch64] Introduce flags for SVE2.

This patch adds support in the compiler for the architecture feature
flags that binutils will use to enable/disable the new "Future
Architecture Technologies" feature Scalable Vector Extension V2 (SVE2)
announced at Linaro Connect this week.

The "sve2" extension that enables the core sve2 instructions.
This also enables the sve extension, since sve is a requirement of sve2.

Extra optional sve2 features are the bitperm, sm4, aes, and sha3 extensions.
These are all given extra feature flags, "bitperm", "sve2-sm4",
"sve2-aes", and "sve2-sha3" respectively.
The sm4, aes, and sha3 extensions are explicitly marked as sve2
extensions to distinguish them from the corresponding NEON extensions.

When introducing macros to denote these new features we have gone past
what a 32 bit value can represent which means we need to change the type
of those variables working with these feature flags to ensure they use
64 bit quantities.

Tested with bootstrap on aarch64-none-linux-gnu and manually seeing that
-march=armv8-a+typo prints out the expected flags while using the new
feature flags does not complain about a missing flag (until reaching the
assembler).

gcc/ChangeLog:

2019-05-22  Matthew Malcomson  <matthew.malcomson@arm.com>

* common/config/aarch64/aarch64-common.c
(struct aarch64_option_extension, struct processor_name_to_arch,
struct arch_to_arch_name, aarch64_parse_extension, opt_ext_cmp,
aarch64_contains_opt,
aarch64_get_extension_string_for_isa_flags): Change type of
variables storing flags to uint64_t.
* config/aarch64/aarch64-option-extensions.def (sve2, sve2-sm4,
sve2-aes, sve2-sha3, bitperm): New optional SVE2 extension flags.
* config/aarch64/aarch64.c (struct processor,
aarch64_parse_arch, aarch64_parse_cpu, aarch64_validate_mcpu,
aarch64_validate_march, aarch64_override_options,
aarch64_option_print, aarch64_handle_attr_isa_flags,
aarch64_declare_function_name, aarch64_start_file): Make flag
variables uint64_t.
* config/aarch64/aarch64.h (AARCH64_FL_SVE2, AARCH64_FL_SVE2_AES,
AARCH64_FL_SVE2_SM4, AARCH64_FL_SVE2_SHA3,
AARCH64_FL_SVE2_BITPERM): New macro feature flags.
* config/aarch64/aarch64.opt (aarch64_isa_flags): Make uint64_t.
* config/aarch64/driver-aarch64.c
(struct aarch64_arch_extension, struct aarch64_core_data,
struct aarch64_arch_driver_info, host_detect_local_cpu): Make
flag variables uint64_t.
* doc/invoke.texi: Add documentation for new arguments.

From-SVN: r271514

Fix new coarray failures.

2019-05-22  Andrew Stubbs  <ams@codesourcery.com>

gcc/testsuite/
* gfortran.dg/coarray_lock_7.f90: Fix output patterns.

From-SVN: r271513

Fix fortran size_type_node parameter passing.

2019-05-22  Andrew Stubbs  <ams@codesourcery.com>

gcc/fortran/
* trans-stmt.c (gfc_trans_critical): Use size_type_node for
gfor_fndecl_caf_lock and gfor_fndecl_caf_unlock calls.
(gfc_trans_allocate): Use size_type_node for gfor_fndecl_caf_sync_all
call.

From-SVN: r271512

fortran/89100: Default widths with -fdec-format-defaults

gcc/fortran ChangeLog:

2019-05-22  Jeff Law  <law@redhat.com>
            Mark Eggleston  <mark.eggleston@codethink.com>

        PR fortran/89100
        * gfortran.texi: Add Default widths for F, G and I format
        descriptors to Extensions section.
        * invoke.texi: Add -fdec-format-defaults
        * io.c (check_format): Use default widths for i, f and g when
        flag_dec_format_defaults is enabled.
        * lang.opt: Add new option.
        * options.c (set_dec_flags): Add SET_BITFLAG for
        flag_dec_format_defaults.

gcc/testsuite ChangeLog:

2019-05-22  Mark Eggleston  <mark.eggleston@codethink.com>

        PR fortran/89100
        * gfortran.dg/fmt_f_default_field_width_1.f90: New test.
        * gfortran.dg/fmt_f_default_field_width_2.f90: New test.
        * gfortran.dg/fmt_f_default_field_width_3.f90: New test.
        * gfortran.dg/fmt_g_default_field_width_1.f90: New test.
        * gfortran.dg/fmt_g_default_field_width_2.f90: New test.
        * gfortran.dg/fmt_g_default_field_width_3.f90: New test.
        * gfortran.dg/fmt_i_default_field_width_1.f90: New test.
        * gfortran.dg/fmt_i_default_field_width_2.f90: New test.
        * gfortran.dg/fmt_i_default_field_width_3.f90: New test.

libgfortran ChangeLog:

2019-05-22  Jeff Law  <law@redhat.com>

        PR fortran/89100
        * io/format.c (parse_format_list): set default width when the
        IOPARM_DT_DEC_EXT flag is set for i, f and g.
        * io/io.h: add default_width_for_integer, default_width_for_float
        and default_precision_for_float.
        * io/write.c (write_boz): extra parameter giving length of data
        corresponding to the type's kind.
        (write_b): pass data length as extra parameter in calls to
        write_boz.
        (write_o): pass data length as extra parameter in calls to
        write_boz.
        (write_z): pass data length as extra parameter in calls to
        write_boz.
        (size_from_kind): also set size is default width is set.
        * io/write_float.def (build_float_string): new paramter inserted
        before result parameter. If default width use values passed
        instead of the values in fnode.
        (FORMAT_FLOAT): macro modified to check for default width and
        calls to build_float_string to pass in default width.
        (get_float_string): set width and precision to defaults when
        needed.

From-SVN: r271511

alias.c (ao_ref_from_mem): Move stack-slot sharing rewrite ...

2019-05-22  Richard Biener  <rguenther@suse.de>

* alias.c (ao_ref_from_mem): Move stack-slot sharing
rewrite ...
* emit-rtl.c (set_mem_attributes_minus_bitpos): ... here.

From-SVN: r271510

Do not use quotes in tests (PR testsuite/90564).

2019-05-22  Martin Liska  <mliska@suse.cz>

PR testsuite/90564
* gcc.target/powerpc/pr80315-1.c: Remove usage of quotes.
* gcc.target/powerpc/pr80315-2.c: Likewise.
* gcc.target/powerpc/pr80315-3.c: Likewise.
* gcc.target/powerpc/pr80315-4.c: Likewise.

From-SVN: r271509

Fix wrong escaping of brackets (PR testsuite/90564).

2019-05-22  Martin Liska  <mliska@suse.cz>

PR testsuite/90564
* gcc.target/powerpc/pr80315-1.c: Add one extra \ to escape
brackets.
* gcc.target/powerpc/pr80315-2.c: Likewise.
* gcc.target/powerpc/pr80315-3.c: Likewise.
* gcc.target/powerpc/pr80315-4.c: Likewise.

From-SVN: r271508

decltype-pr66548.C: Reinstate correct version.

2019-05-22  Paolo Carlini  <paolo.carlini@oracle.com>

      * g++.dg/cpp0x/decltype-pr66548.C: Reinstate correct version.

From-SVN: r271507

decltype-pr66548.C: Remove, mistakenly added.

2019-05-22  Paolo Carlini  <paolo.carlini@oracle.com>

      * g++.dg/cpp0x/decltype-pr66548.C: Remove, mistakenly added.

From-SVN: r271506

darwin, testsuite fix PR27221

The test can't succeed on 32b powerpc Darwin since the
ABI demands a different result to the tested one.

2019-05-22  Iain Sandoe  <iain@sandoe.co.uk>

PR testsuite/27221
* g++.dg/ext/alignof2.C: XFAIL for 32bit Darwin.

From-SVN: r271505

Strip target_clones in copy attribute (PR lto/90500).

2019-05-22  Martin Liska  <mliska@suse.cz>

PR lto/90500
* c-attribs.c (handle_copy_attribute): Do not copy
target_clones attribute.
2019-05-22  Martin Liska  <mliska@suse.cz>

PR lto/90500
* gcc.target/i386/pr90500-1.c: Make the test-case valid
now.
2019-05-22  Martin Liska  <mliska@suse.cz>

PR lto/90500
* doc/extend.texi: Document the change.

From-SVN: r271504

re PR tree-optimization/90450 (Hash function in gather_mem_refs_stmt does not match with mem_ref_hasher::equal)

2019-05-22  Richard Biener  <rguenther@suse.de>

PR tree-optimization/90450
* tree-ssa-loop-im.c (struct im_mem_ref): Add ref_decomposed.
(mem_ref_hasher::equal): Check it.
(mem_ref_alloc): Initialize it.
(gather_mem_refs_stmt): Set it.

From-SVN: r271503

gimple-fold.c (arith_code_with_undefined_signed_overflow): Add ABS_EXPR.

2019-05-22  Richard Biener  <rguenther@suse.de>

* gimple-fold.c (arith_code_with_undefined_signed_overflow):
Add ABS_EXPR.
(rewrite_to_defined_overflow): Handle rewriting ABS_EXPR
as ABSU_EXPR.

* gcc.dg/tree-ssa/ssa-lim-13.c: New testcase.

From-SVN: r271501

[RS6000] Don't pass -many to the assembler

I'd like to remove -many from the options passed by default to the
assembler, on the grounds that a gcc bug in instruction selection (eg.
emitting a power9 insn for -mcpu=power8) is better found at assembly
time than run time.

For now, just do this when --enable-checking or gcc is not a release.

This patch also emits .machine assembler directives for ELF targets
when functions are compiled for different cpus via attributes or
pragmas.  That's necessary when the initial -m<cpu> option passed to
the assembler doesn't enable the superset of all opcodes emitted, as
seen by a failure of gcc.target/powerpc/clone2.c without .machine
when building gcc for power8.

rs6000_machine_from_flags deliberately uses ISA_2_4_MASKS rather than
ISA_2_2_MASKS for power5 because "friz" and other similar instructions
enabled by gcc with TARGET_FPRND are enabled in gas by "-mpower5".
(gas -mpower5 supports power5+ too.)  rs6000-cpus.def puts
OPTION_MASK_FPRND in ISA_2_4_MASKS, so ISA_2_4_MASKS is the one to use
in deciding to pass "-mpower5" to gas.

O3-pr70130.c also failed on an earlier version of this patch (when
only testing one ISA bit to determine .machine).  This is a test for a
power7 vector bug, but on power8 hw check_vect_support_and_set_flags
passes -mpower8-vector which means the test isn't exercising the
original bug exactly.  I reckon that is wrong, and similary for other
vector testcases that ask for a specific cpu.  I've fixed it here by
explicitly passing -mno-power8-vector and similar vector options.

* config/rs6000/rs6000.h (ASM_OPT_ANY): Define.
(ASM_CPU_SPEC): Conditionally add -many.
* config/rs6000/rs6000.c (rs6000_machine): New static var.
(rs6000_machine_from_flags, emit_asm_machine): New functions..
(rs6000_file_start): ..extracted from here, and modified to
test all ISA bits.
(rs6000_output_function_prologue): Emit .machine as necessary.
* testsuite/gcc.target/powerpc/ppc32-abi-dfp-1.c: Don't use
power mnemonics.
* testsuite/gcc.dg/vect/O3-pr70130.c: Disable default options
added by check_vect_support_and_set_flags.
* testsuite/gcc.dg/vect/pr48765.c: Likewise.
* testsuite/gfortran.dg/vect/pr45714-b.f: Likewise.

From-SVN: r271500

re PR middle-end/90553 (Register allocation allocates post-incremented address-load of call to call-clobbered register)

PR middle-end/90553
* gcc.dg/torture/pr90553.c: New test.

From-SVN: r271499

re PR middle-end/90553 (Register allocation allocates post-incremented address-load of call to call-clobbered register)

PR middle-end/90553
* ira-lives.c (process_bb_node_lives): Consider defs
for a call insn to be die before the call, not after.

From-SVN: r271498

In <https://gcc.gnu.org/ml/gcc-patches/2019-05/msg01417.html>...

In <https://gcc.gnu.org/ml/gcc-patches/2019-05/msg01417.html>:
...a change introduced in r94104 where BITS_PER_WORD was chosenperhaps because we expect register-sized writes into this area.

See <https://gcc.gnu.org/ml/gcc-patches/2005-01/msg01564.html>
for the r94104 change.

* function.c (assign_parm_setup_block): Raise alignment of
stacked parameter only for STRICT_ALIGNMENT targets.

From-SVN: r271497

Daily bump.

From-SVN: r271496

parser.c (cp_parser_template_declaration_after_parameters): Use DECL_SOURCE_LOCATION in literal operator template errors.

/cp
2019-05-21  Paolo Carlini  <paolo.carlini@oracle.com>

* parser.c (cp_parser_template_declaration_after_parameters): Use
DECL_SOURCE_LOCATION in literal operator template errors.

/testsuite
2019-05-21  Paolo Carlini  <paolo.carlini@oracle.com>

* g++.dg/cpp0x/udlit-tmpl-arg-neg2.C: Check locations too.
* g++.dg/cpp0x/udlit-tmpl-parms-neg.C: Likewise.

From-SVN: r271492

re PR c++/67184 (Missed optimization with C++11 final specifier)

/cp
2019-05-21  Paolo Carlini  <paolo.carlini@oracle.com>

PR c++/67184
PR c++/69445
* call.c (build_over_call): Devirtualize when the final overrider
comes from the base.

/testsuite
2019-05-21  Paolo Carlini  <paolo.carlini@oracle.com>

PR c++/67184
PR c++/69445
* g++.dg/other/final3.C: New.
* g++.dg/other/final4.C: Likewise.
* g++.dg/other/final5.C: Likewise.

From-SVN: r271491

re PR c++/67184 (Missed optimization with C++11 final specifier)

/cp
2019-05-21  Paolo Carlini  <paolo.carlini@oracle.com>

PR c++/67184
PR c++/69445
* call.c (build_over_call): Devirtualize when the final overrider
comes from the base.

/testsuite
2019-05-21  Paolo Carlini  <paolo.carlini@oracle.com>

PR c++/67184
PR c++/69445
* g++.dg/other/final3.C: New.
* g++.dg/other/final4.C: Likewise.
* g++.dg/other/final5.C: Likewise.

From-SVN: r271490