gcc.git
5 years agoDaily bump.
GCC Administrator [Wed, 8 May 2019 00:16:19 +0000 (00:16 +0000)]
Daily bump.

From-SVN: r270997

5 years agocompiler: add an option to emit optimization diagnostics
Cherry Zhang [Wed, 8 May 2019 00:14:17 +0000 (00:14 +0000)]
compiler: add an option to emit optimization diagnostics

    Add a -fgo-debug-optimization option to emit optimization
    diagnostics. This can be used for testing optimizations. Apply
    this to the range clear optimizations of maps and arrays.

    Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/170002

gcc/go:

* lang.opt (-fgo-debug-optimization): New option.
* go-c.h (struct go_create_gogo_args): Add debug_optimization
field.
* go-lang.c (go_langhook_init): Set debug_optimization field.
* gccgo.texi (Invoking gccgo): Document -fgo-debug-optimization.

gcc/testsuite:

* go.dg/arrayclear.go: New test.
* go.dg/mapclear.go: New test.

From-SVN: r270993

5 years agoUpdate .po files.
Joseph Myers [Wed, 8 May 2019 00:05:06 +0000 (01:05 +0100)]
Update .po files.

* be.po, da.po, de.po, el.po, es.po, fi.po, fr.po, hr.po, id.po,
ja.po, nl.po, ru.po, sr.po, sv.po, tr.po, uk.po, vi.po, zh_CN.po,
zh_TW.po: Update.

From-SVN: r270991

5 years agoImprove API docs for Filesystem TS and Networking TS
Jonathan Wakely [Tue, 7 May 2019 23:22:01 +0000 (00:22 +0100)]
Improve API docs for Filesystem TS and Networking TS

* include/experimental/bits/fs_path.h: Improve docs.
* include/experimental/bits/net.h: Fix wrong header name in comment.
Do not document implementation details.
* include/experimental/netfwd: Fix doxygen grouping.

From-SVN: r270990

5 years agoImprove API docs for std::pair
Jonathan Wakely [Tue, 7 May 2019 22:46:56 +0000 (23:46 +0100)]
Improve API docs for std::pair

* include/bits/stl_pair.h: Improve docs.
* include/std/tuple: Likewise.

From-SVN: r270989

5 years agoImprove API docs for <chrono> and <ratio>
Jonathan Wakely [Tue, 7 May 2019 22:46:53 +0000 (23:46 +0100)]
Improve API docs for <chrono> and <ratio>

* doc/doxygen/doxygroups.cc (std::literals): Add documentation for
inline namespace.
* include/std/chrono: Improve docs.
* include/std/ratio: Do not document implementation details.
* testsuite/20_util/ratio/cons/cons_overflow_neg.cc: Adjust dg-error
line numbers.
* testsuite/20_util/ratio/operations/ops_overflow_neg.cc: Likewise.

From-SVN: r270988

5 years agoPR libstdc++/89102 implement new common_type rules (P0435R1, P0548R1)
Jonathan Wakely [Tue, 7 May 2019 22:46:39 +0000 (23:46 +0100)]
PR libstdc++/89102 implement new common_type rules (P0435R1, P0548R1)

This change ensures that std::common_type<> is a complete type (LWG
2408), and that std::common_type<T>, std::common_type<cv T1, cv T2>, and
std::common_type<T1, T2, R...> will use program-defined specializations
for std::common_type<T1, T2> (LWG 2465).

The implementation of common_type<T1, T2, R...> is changed to use
void_t, and the specializations for duration and time_point are modified
to also use void_t instead of depending on implementation details of
common_type.

PR libstdc++/89102
* doc/xml/manual/intro.xml: Document DR 2408 and 2465 changes.
* include/std/chrono (__duration_common_type_wrapper): Replace with ...
(__duration_common_type): New helper.
(common_type<chrono::duration<R1, P2>, chrono::duration<R2, P2>>): Use
__duration_common_type.
(__timepoint_common_type_wrapper): Replace with ...
(__timepoint_common_type): New helper.
(common_type<chrono::time_point<C, D2>, chrono::time_point<C, D2>>):
Use __time_point_common_type.
* include/std/type_traits (common_type<>): Define, as per LWG 2408.
(__common_type_impl): If either argument is transformed by decay,
use the common_type of the decayed types.
(__common_type_impl<_Tp, _Up, _Tp, _Up>): If the types are already
decayed, use __do_common_type_impl to get the common_type.
(common_type<_Tp>): Use common_type<_Tp, _Tp>.
(__do_member_type_wrapper, __member_type_wrapper)
(__expanded_common_type_wrapper): Remove.
(__common_type_pack, __common_type_fold): New helpers.
(common_type<_Tp, _Up, _Vp...>): Use new helpers instead of
__member_type_wrapper and __expanded_common_type_wrapper.
* testsuite/20_util/common_type/requirements/explicit_instantiation.cc:
Test zero-length template argument list.
* testsuite/20_util/common_type/requirements/sfinae_friendly_1.cc:
Test single argument cases and argument types that should decay.
* testsuite/20_util/common_type/requirements/sfinae_friendly_2.cc:
Adjust expected error.
* testsuite/20_util/duration/literals/range_neg.cc: Use zero for
dg-error lineno.
* testsuite/20_util/duration/requirements/typedefs_neg1.cc: Likewise.
* testsuite/20_util/duration/requirements/typedefs_neg2.cc: Likewise.
* testsuite/20_util/duration/requirements/typedefs_neg3.cc: Likewise.

From-SVN: r270987

5 years agoPR c++/90171 - reorganize usual_deallocation_fn_p
Jason Merrill [Tue, 7 May 2019 22:37:24 +0000 (18:37 -0400)]
PR c++/90171 - reorganize usual_deallocation_fn_p

When fixing 90171 it struck me as undesirable to have so many separate
functions that all needed to know about the definition of a usual
deallocation function.  So this patch condenses them into one.  I left
destroying_delete_p because it is used by other files as well.

* call.c (struct dealloc_info): New.
(usual_deallocation_fn_p): Take a dealloc_info*.
(aligned_deallocation_fn_p, sized_deallocation_fn_p): Remove.
(build_op_delete_call): Adjust.

From-SVN: r270986

5 years agoPR c++/86485 - -Wmaybe-unused with empty class ?:
Jason Merrill [Tue, 7 May 2019 22:37:19 +0000 (18:37 -0400)]
PR c++/86485 - -Wmaybe-unused with empty class ?:

* typeck.c (build_static_cast_1): Use cp_build_addr_expr.

For GCC 9 I fixed this bug with a patch to gimplify_cond_expr, but this
function was also doing the wrong thing.

Using build_address does not push the ADDR_EXPR down into the arms of a
COND_EXPR, which we need for proper handling of conversion of an lvalue ?:
to another reference type.

From-SVN: r270985

5 years agoA non-type template parm with a placeholder type is type-dependent.
Jason Merrill [Tue, 7 May 2019 22:37:14 +0000 (18:37 -0400)]
A non-type template parm with a placeholder type is type-dependent.

* pt.c (type_dependent_expression_p): A non-type template parm with
a placeholder type is type-dependent.

From-SVN: r270984

5 years agors6000: Remove reload leftovers
Segher Boessenkool [Tue, 7 May 2019 22:25:05 +0000 (00:25 +0200)]
rs6000: Remove reload leftovers

There are a few things left in the rs6000 port that are unused now
that we do not support old reload anymore.  This removes those.

* config/rs6000/rs6000-protos.h (rs6000_legitimize_reload_address_ptr):
Delete declaration.
* config/rs6000/rs6000.c (rs6000_legitimize_reload_address): Delete.
(rs6000_debug_legitimize_reload_address): Delete.
(rs6000_legitimize_reload_address_ptr): Delete.
(rs6000_option_override_internal): Adjust.
(mem_operand_gpr): Adjust comment.
(legitimate_lo_sum_address_p): Ditto.
(rs6000_legitimize_reload_address): Delete.
(rs6000_debug_legitimize_reload_address): Delete.
* config/rs6000/rs6000.h (LEGITIMIZE_RELOAD_ADDRESS): Delete.

From-SVN: r270983

5 years agore PR middle-end/89765 (Multiple problems with vec-insert implementation on PowerPC)
Kelvin Nilsen [Tue, 7 May 2019 21:40:46 +0000 (21:40 +0000)]
re PR middle-end/89765 (Multiple problems with vec-insert implementation on PowerPC)

gcc/ChangeLog:

2019-05-07  Kelvin Nilsen  <kelvin@gcc.gnu.org>

PR target/89765
* config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin):
In handling of ALTIVEC_BUILTIN_VEC_INSERT, use modular arithmetic
to compute vector element selector for both constant and variable
operands.

gcc/testsuite/ChangeLog:

2019-05-07  Kelvin Nilsen  <kelvin@gcc.gnu.org>

PR target/89765
* gcc.target/powerpc/pr89765-mc.c: New test.
* gcc.target/powerpc/vsx-builtin-10c.c: New test.
* gcc.target/powerpc/vsx-builtin-10d.c: New test.
* gcc.target/powerpc/vsx-builtin-11c.c: New test.
* gcc.target/powerpc/vsx-builtin-11d.c: New test.
* gcc.target/powerpc/vsx-builtin-12c.c: New test.
* gcc.target/powerpc/vsx-builtin-12d.c: New test.
* gcc.target/powerpc/vsx-builtin-13c.c: New test.
* gcc.target/powerpc/vsx-builtin-13d.c: New test.
* gcc.target/powerpc/vsx-builtin-14c.c: New test.
* gcc.target/powerpc/vsx-builtin-14d.c: New test.
* gcc.target/powerpc/vsx-builtin-15c.c: New test.
* gcc.target/powerpc/vsx-builtin-15d.c: New test.
* gcc.target/powerpc/vsx-builtin-16c.c: New test.
* gcc.target/powerpc/vsx-builtin-16d.c: New test.
* gcc.target/powerpc/vsx-builtin-17c.c: New test.
* gcc.target/powerpc/vsx-builtin-17d.c: New test.
* gcc.target/powerpc/vsx-builtin-18c.c: New test.
* gcc.target/powerpc/vsx-builtin-18d.c: New test.
* gcc.target/powerpc/vsx-builtin-19c.c: New test.
* gcc.target/powerpc/vsx-builtin-19d.c: New test.
* gcc.target/powerpc/vsx-builtin-20c.c: New test.
* gcc.target/powerpc/vsx-builtin-20d.c: New test.
* gcc.target/powerpc/vsx-builtin-9c.c: New test.
* gcc.target/powerpc/vsx-builtin-9d.c: New test.

From-SVN: r270982

5 years agoi386.md (cvt_mnemonic): New mode attribute.
Uros Bizjak [Tue, 7 May 2019 21:36:42 +0000 (23:36 +0200)]
i386.md (cvt_mnemonic): New mode attribute.

* config/i386/i386.md (cvt_mnemonic): New mode attribute.
(ashr<mode>3_cvt): Merge insn pattern from ashrsi3_cvt and
ashrdi3_cvt using SWI48 mode iterator.

From-SVN: r270981

5 years ago[libcpp] Reimplement mkdeps data structures
Nathan Sidwell [Tue, 7 May 2019 18:13:57 +0000 (18:13 +0000)]
[libcpp] Reimplement mkdeps data structures

https://gcc.gnu.org/ml/gcc-patches/2019-05/msg00323.html
* files.c (_cpp_stack_file): Empty filenames aren't dependencies.
* mkdeps.c (deps_add_dep): Assert not empty.

From-SVN: r270978

5 years agoResign as IA-64 maintainer.
Jim Wilson [Tue, 7 May 2019 17:46:51 +0000 (17:46 +0000)]
Resign as IA-64 maintainer.

* MAINTAINERS: Remove myself as IA-64 maintainer.

From-SVN: r270977

5 years agoThis patch adds support to vectorize sum of abslolute differences (SAD_EXPR)
Alejandro Martinez [Tue, 7 May 2019 16:34:20 +0000 (16:34 +0000)]
This patch adds support to vectorize sum of abslolute differences (SAD_EXPR)
using SVE.

Given this input code:

int
sum_abs (uint8_t *restrict x, uint8_t *restrict y, int n)
{
  int sum = 0;

  for (int i = 0; i < n; i++)
    {
      sum += __builtin_abs (x[i] - y[i]);
    }

  return sum;
}

The resulting SVE code is:

0000000000000000 <sum_abs>:
   0: 7100005f  cmp w2, #0x0
   4: 5400026d  b.le 50 <sum_abs+0x50>
   8: d2800003  mov x3, #0x0                    // #0
   c: 93407c42  sxtw x2, w2
  10: 2538c002  mov z2.b, #0
  14: 25221fe0  whilelo p0.b, xzr, x2
  18: 2538c023  mov z3.b, #1
  1c: 2518e3e1  ptrue p1.b
  20: a4034000  ld1b {z0.b}, p0/z, [x0, x3]
  24: a4034021  ld1b {z1.b}, p0/z, [x1, x3]
  28: 0430e3e3  incb x3
  2c: 0520c021  sel z1.b, p0, z1.b, z0.b
  30: 25221c60  whilelo p0.b, x3, x2
  34: 040d0420  uabd z0.b, p1/m, z0.b, z1.b
  38: 44830402  udot z2.s, z0.b, z3.b
  3c: 54ffff21  b.ne 20 <sum_abs+0x20>  // b.any
  40: 2598e3e0  ptrue p0.s
  44: 04812042  uaddv d2, p0, z2.s
  48: 1e260040  fmov w0, s2
  4c: d65f03c0  ret
  50: 1e2703e2  fmov s2, wzr
  54: 1e260040  fmov w0, s2
  58: d65f03c0  ret

Notice how udot is used inside a fully masked loop.

gcc/Changelog:

2019-05-07  Alejandro Martinez  <alejandro.martinezvicente@arm.com>

* config/aarch64/aarch64-sve.md (<su>abd<mode>_3): New define_expand.
(aarch64_<su>abd<mode>_3): Likewise.
(*aarch64_<su>abd<mode>_3): New define_insn.
(<sur>sad<vsi2qi>): New define_expand.
* config/aarch64/iterators.md: Added MAX_OPP attribute.
* tree-vect-loop.c (use_mask_by_cond_expr_p): Add SAD_EXPR.
(build_vect_cond_expr): Likewise.

gcc/testsuite/Changelog:

2019-05-07  Alejandro Martinez  <alejandro.martinezvicente@arm.com>

* gcc.target/aarch64/sve/sad_1.c: New test for sum of absolute
differences.

From-SVN: r270975

5 years agocfgexpand.c (asm_clobber_reg_is_valid): Reject clobbers outside of accessible_reg_set.
Uros Bizjak [Tue, 7 May 2019 15:42:42 +0000 (17:42 +0200)]
cfgexpand.c (asm_clobber_reg_is_valid): Reject clobbers outside of accessible_reg_set.

* cfgexpand.c (asm_clobber_reg_is_valid): Reject
clobbers outside of accessible_reg_set.
* config/i386/i386.c (ix86_conditional_register_usage):
Disable register sets by clearing corresponding bits in
accessible_reg_set.  Do not set corresponding bits in fixed_regs,
call_used_regs and don't clear corresponding reg_names array members.

testsuite/ChangeLog:

* gcc.target/i386/asm-7.c: New test.
* gcc.target/i386/asm-1.c: Update expected error string.
* gcc.target/i386/pr62120.c: Ditto.

From-SVN: r270955

5 years agoFix incorrect DR numbers in libstdc++ manual
Jonathan Wakely [Tue, 7 May 2019 15:30:52 +0000 (16:30 +0100)]
Fix incorrect DR numbers in libstdc++ manual

* doc/xml/manual/intro.xml: Fix DR 2537 and DR 2566 confusion.

From-SVN: r270954

5 years agoMake allocator propagation more consistent for operator+(basic_string) (P1165R1)
Nina Dinka Ranns [Tue, 7 May 2019 15:30:46 +0000 (15:30 +0000)]
Make allocator propagation more consistent for operator+(basic_string) (P1165R1)

2019-05-01  Nina Dinka Ranns  <dinka.ranns@gmail.com>

Make allocator propagation more consistent for
operator+(basic_string) (P1165R1)
* include/bits/basic_string.h
(operator+(basic_string&&, basic_string&&): Changed resulting
allocator to always be the one from the first parameter.
* include/bits/basic_string.tcc
(operator+(const _CharT*, const basic_string&)): Changed
resulting allocator to be SOCCC on the second parameter's allocator.
(operator+(_CharT, const basic_string&)): Likewise.
* testsuite/21_strings/basic_string/allocator/char/operator_plus.cc:
New.
* testsuite/21_strings/basic_string/allocator/wchar_t/operator_plus.cc:
New.

From-SVN: r270953

5 years agoImprove API docs for std::regex
Jonathan Wakely [Tue, 7 May 2019 15:30:40 +0000 (16:30 +0100)]
Improve API docs for std::regex

* include/bits/regex.h: Improve docs.
* include/bits/regex.tcc: Do not document implementation details.

From-SVN: r270952

5 years agoAdd test for std::hash<std::error_code>
Jonathan Wakely [Tue, 7 May 2019 15:30:29 +0000 (16:30 +0100)]
Add test for std::hash<std::error_code>

Copied from 19_diagnostics/error_condition/hash.cc added recently.

* testsuite/19_diagnostics/error_code/hash.cc: New test.

From-SVN: r270951

5 years agoFix make install-gcc-specs with empty GCC_SPECS_FILES
Rainer Orth [Tue, 7 May 2019 15:19:26 +0000 (15:19 +0000)]
Fix make install-gcc-specs with empty GCC_SPECS_FILES

* gcc-interface/Makefile.in (install-gcc-specs): Use foreach.
Honor DESTDIR.

From-SVN: r270947

5 years agotree-vect-stmts.c (vect_is_simple_cond): When vectype is not specified still compute...
Richard Biener [Tue, 7 May 2019 13:07:48 +0000 (13:07 +0000)]
tree-vect-stmts.c (vect_is_simple_cond): When vectype is not specified still compute a comp_vectype for invariant compares.

2019-05-07  Richard Biener  <rguenther@suse.de>

* tree-vect-stmts.c (vect_is_simple_cond): When vectype is
not specified still compute a comp_vectype for invariant
compares.

From-SVN: r270945

5 years agore PR tree-optimization/90316 (large compile time increase in opt / alias stmt walkin...
Richard Biener [Tue, 7 May 2019 13:03:19 +0000 (13:03 +0000)]
re PR tree-optimization/90316 (large compile time increase in opt / alias stmt walking for Go example)

2019-05-07  Richard Biener  <rguenther@suse.de>

PR tree-optimization/90316
* tree-ssa-pre.c (translate_vuse_through_block): When
same_valid is NULL do not bother to search for a virtual
PHI continuation.
(phi_translate_1): When operands changed we cannot keep
the same value-number so do not bother to ask whether
that's possible from translate_vuse_through_block.

From-SVN: r270944

5 years ago[libcpp] Reimplement mkdeps data structures
Nathan Sidwell [Tue, 7 May 2019 12:39:59 +0000 (12:39 +0000)]
[libcpp] Reimplement mkdeps data structures

https://gcc.gnu.org/ml/gcc-patches/2019-05/msg00293.html
* include/mkdeps.h (deps_write): Add PHONY arg.
(deps_phony_targets): Delete.
* init.c (cpp_finish): Just call deps_write.
* mkdeps.c (struct mkdeps): Add local vector class.  Reimplement
vector handling.
(munge): Munge to static buffer.
(apply_vpath): Adjust vector handling.
(deps_init, deps_free): Use new, delete.
(deps_add_target): Do not munge here.  Record quoting low water mark.
(deps_add_dep): Do not munge here.
(deps_add_vpath): Adjust vector handling.
(make_write_name): New.  Munge on demand here.
(make_write_vec): New.
(deps_phony_targets): Delete.
(make_write): New.
(deps_write): Forward to deps_Write.
(deps_save, deps_restore): Adjust vector handling.

From-SVN: r270943

5 years agoFix bitmap registration of overheads.
Martin Liska [Tue, 7 May 2019 12:15:45 +0000 (14:15 +0200)]
Fix bitmap registration of overheads.

2019-05-07  Martin Liska  <mliska@suse.cz>

* bitmap.c (bitmap_register): Come up with
alloc_descriptor_max_uid and assign it for
a new bitmap.
(register_overhead): Use get_descriptor as
a descriptor.
(release_overhead): New.
(bitmap_elem_to_freelist): Call it.
(bitmap_elt_clear_from): Likewise.
(bitmap_obstack_free): Likewise.
(bitmap_move): Sensitively release memory.
* bitmap.h (struct GTY): Add alloc_descriptor and padding.
(bitmap_initialize): Initialize alloc_descriptor to zero.
* tree-ssa-pre.c (do_hoist_insertion): Use bitmap_move.

From-SVN: r270942

5 years agotree-vect-slp.c (vect_build_slp_tree_2): Bump size whenever we build a SLP node.
Richard Biener [Tue, 7 May 2019 11:47:48 +0000 (11:47 +0000)]
tree-vect-slp.c (vect_build_slp_tree_2): Bump size whenever we build a SLP node.

2019-05-07  Richard Biener  <rguenther@suse.de>

* tree-vect-slp.c (vect_build_slp_tree_2): Bump size whenever
we build a SLP node.  Remove max_size and limiting.
(vect_analyze_slp_instance): Record and dump size of the SLP graph.

From-SVN: r270941

5 years agore PR tree-optimization/90316 (large compile time increase in opt / alias stmt walkin...
Richard Biener [Tue, 7 May 2019 11:17:00 +0000 (11:17 +0000)]
re PR tree-optimization/90316 (large compile time increase in opt / alias stmt walking for Go example)

2019-05-07  Richard Biener  <rguenther@suse.de>

PR tree-optimization/90316
* tree-ssa-alias.h (get_continuation_for_phi): Take walking
limit by reference.
(walk_non_aliased_vuses): Take walking limit argument.
* tree-ssa-alias.c (maybe_skip_until): Take limit and abort
walking if it is reached instead of just counting.
(get_continuation_for_phi): Likewise.
(walk_non_aliased_vuses): Likewise, instead of leaving counter
limiting to the callback.
* tree-ssa-sccvn.c (vn_reference_lookup_2): Adjust.
(vn_reference_lookup_3): Likewise.
(vn_reference_lookup_pieces): Likewise.
(vn_reference_lookup): Likewise.
* tree-ssa-pre.c (translate_vuse_through_block): Limit walking.
* tree-ssa-scopedtables.c (vuse_eq): Adjust.
(avail_exprs_stack::lookup_avail_expr): Likewise.

From-SVN: r270940

5 years agoCheck if Solaris ld supports -z relax=transtls
Rainer Orth [Tue, 7 May 2019 10:38:45 +0000 (10:38 +0000)]
Check if Solaris ld supports -z relax=transtls

* m4/druntime/os.m4 (DRUNTIME_OS_LINK_SPEC): Only use -z
relax=transtls if linker supports it.
* configure.ac (enable_libphobos, LIBPHOBOS_SUPPORTED): Move down.
(x86_64-*-solaris2.* | i?86-*-solaris2.*): Only
mark supported with either gld or ld -z relax=transtls.
* configure: Regenerate.

From-SVN: r270939

5 years agotree-ssa-alias.c (aliasing_component_refs_p): Continue looking for comparaible types...
Jan Hubicka [Tue, 7 May 2019 09:36:05 +0000 (09:36 +0000)]
tree-ssa-alias.c (aliasing_component_refs_p): Continue looking for comparaible types in the second direction even if...

* tree-ssa-alias.c (aliasing_component_refs_p): Continue looking
for comparaible types in the second direction even if first one
hits incomparable type.

From-SVN: r270938

5 years agore PR lto/90369 (error: could not unlink output file)
Richard Biener [Tue, 7 May 2019 08:52:33 +0000 (08:52 +0000)]
re PR lto/90369 (error: could not unlink output file)

2019-05-07  Richard Biener  <rguenther@suse.de>

PR lto/90369
* lto-wrapper.c (debug_objcopy): Use the original filename
including archive offset for the filename used for -save-temps.

From-SVN: r270937

5 years agoFix Solaris bootstrap: lto-common.c, lto-dump.c format mismatches
Rainer Orth [Tue, 7 May 2019 07:44:55 +0000 (07:44 +0000)]
Fix Solaris bootstrap: lto-common.c, lto-dump.c format mismatches

* lto-common.c (lto_file_read): Print section->start as int64_t,
section->len as uint64_t.
* lto-dump.c (symbol_entry::dump): Print sz as uint64_t.

From-SVN: r270935

5 years agoFix a typo in two_value_replacement function
Li Jia He [Tue, 7 May 2019 05:39:01 +0000 (05:39 +0000)]
Fix a typo in two_value_replacement function

GCC revision 267634 implemented two_value_replacement function.
However, a typo occurred during the parameter check, which caused
us to miss some optimizations.

The intent of the code might be to check that the input parameters
are const int and their difference is one.  However, when I read
the code, I found that it is wrong to detect whether an input data
plus one is equal to itself.  This could be a typo.

2019-05-07  Li Jia He  <helijia@linux.ibm.com>

* tree-ssa-phiopt.c (two_value_replacement): Fix a typo in parameter
detection.
* gcc.dg/tree-ssa/pr37508.c: Add the no-ssa-phiopt option to skip phi
optimization.
* gcc.dg/tree-ssa/pr88676-2.c: New testcase.

From-SVN: r270934

5 years agoDaily bump.
GCC Administrator [Tue, 7 May 2019 00:17:00 +0000 (00:17 +0000)]
Daily bump.

From-SVN: r270933

5 years agors6000: Renumber the registers
Segher Boessenkool [Mon, 6 May 2019 23:50:42 +0000 (01:50 +0200)]
rs6000: Renumber the registers

This renumbers the registers.

It moves the VRs to 64..95, right after the GPRs and the FPRS.  This
means that the VSRs (which are aliases to the FPRs and the VRs, in
that order) are consecutive now.

It removes MQ, which has been just a stub for ages (it is a leftover
from RIOS, old POWER).

It moves the CR fields to 100..107, which is a bit easier to read
than the 68..75 is was before.

The rest fills the holes.  It should be easy to move anything else
after this series, so the exact order isn't very important anymore,
we aren't stuck with it if we dislike it.

Many things still want the GPRs to be at 0..31, and some things want
the FPRs at 32..63.  I don't think we'll ever want to change that,
so I left it be.

Small things...  It removes DWARF_FRAME_REGISTERS, it used to save
1000 or so words of memory, but it has been just a handful for a
while, and now it is one.  Some whitespace fixes.  Testing showed one
or two places where register allocation was different (not worse, not
better, just different).

* config/rs6000/rs6000.md (FIRST_ALTIVEC_REGNO, LAST_ALTIVEC_REGNO)
(LR_REGNO, CTR_REGNO, CA_REGNO, ARG_POINTER_REGNUM, CR0_REGNO)
(CR1_REGNO, CR2_REGNO, CR3_REGNO, CR4_REGNO, CR5_REGNO, CR6_REGNO)
(CR7_REGNO, MAX_CR_REGNO, VRSAVE_REGNO, VSCR_REGNO)
(FRAME_POINTER_REGNUM): Change numbering.
* config/rs6000/rs6000.c (rs6000_reg_names): Adjust.
(alt_reg_names): Adjust.
(rs6000_conditional_register_usage): Don't mark hard register 64 as
fixed.
* config/rs6000/rs6000.h (FIRST_PSEUDO_REGISTER): Adjust.
(DWARF_FRAME_REGISTERS): Delete.
(DWARF2_FRAME_REG_OUT): Fix whitespace.
(FIXED_REGISTERS, CALL_USED_REGISTERS, CALL_REALLY_USED_REGISTERS):
Adjust.
(REG_ALLOC_ORDER): Adjust.
(FRAME_POINTER_REGNUM, ARG_POINTER_REGNUM): Adjust.
(REG_CLASS_CONTENTS): Adjust.
(RETURN_ADDR_RTX): Change comment.
(REGNO_OK_FOR_INDEX_P, REGNO_OK_FOR_BASE_P): Use ARG_POINTER_REGNUM
instead of 67.
(REGISTER_NAMES): Adjust.
(ADDITIONAL_REGISTER_NAMES): Adjust.
* config/rs6000/darwin.h (REGISTER_NAMES): Adjust.

From-SVN: r270929

5 years agore PR fortran/90290 (-std=f2008 should reject non-constant stop and error stop codes)
Steven G. Kargl [Mon, 6 May 2019 23:24:32 +0000 (23:24 +0000)]
re PR fortran/90290 (-std=f2008 should reject non-constant stop and error stop codes)

2019-05-06  Steven G. Kargl  <kargl@gcc.gnu.org>

PR fortran/90290
* match.c (gfc_match_stopcode): Check F2008 condition on stop code.

2019-05-06  Steven G. Kargl  <kargl@gcc.gnu.org>

PR fortran/90290
* gfortran.dg/pr90290.f90: New test.

From-SVN: r270928

5 years agors6000: Remove TM regs
Segher Boessenkool [Mon, 6 May 2019 22:04:19 +0000 (00:04 +0200)]
rs6000: Remove TM regs

We do not need to expose the TM registers in debug info.  It isn't
actually useful there, because none of the things that can modify
these registers (other than explicit moves) are marked.

We also do not need the registers for GCC itself internally.  This
patch deletes them.

* config/rs6000/rs6000.md (TFHAR_REGNO, TFIAR_REGNO, TEXASR_REGNO):
Delete.
* config/rs6000/rs6000.h (FIRST_PSEUDO_REGISTER): Adjust.
(DWARF_FRAME_REGISTERS): Adjust.
(FIXED_REGISTERS, CALL_USED_REGISTERS, CALL_REALLY_USED_REGISTERS):
Adjust.
(REG_ALLOC_ORDER): Adjust.
(enum reg_class): Delete SPR_REGS.
(REG_CLASS_NAMES): Delete SPR_REGS.
(REG_CLASS_CONTENTS): Delete SPR_REGS.  Adjust for deleted TM regs.
(REGISTER_NAMES): Adjust.
(ADDITIONAL_REGISTER_NAMES): Adjust.
* config/rs6000/darwin.h (REGISTER_NAMES): Adjust.
* config/rs6000/htm.md (htm_mfspr_<mode>, htm_mtspr_<mode>): Adjust.
* config/rs6000/predicates.md (htm_spr_reg_operand): Delete.
* config/rs6000/rs6000.c (rs6000_init_hard_regno_mode_ok): Adjust.
(htm_spr_regno): Delete.
(htm_expand_builtin): Adjust: the HTM builtins now have one fewer
argument.
(rs6000_dbx_register_number): Adjust.

From-SVN: r270927

5 years agors6000: Delete PRE_GCC3_DWARF_FRAME_REGISTERS
Segher Boessenkool [Mon, 6 May 2019 22:02:24 +0000 (00:02 +0200)]
rs6000: Delete PRE_GCC3_DWARF_FRAME_REGISTERS

We don't need this.

* config/rs6000/rs6000.h (PRE_GCC3_DWARF_FRAME_REGISTERS): Delete.

From-SVN: r270926

5 years agors6000: rs6000_dbx_register_number for fp/ap/mq
Segher Boessenkool [Mon, 6 May 2019 22:01:27 +0000 (00:01 +0200)]
rs6000: rs6000_dbx_register_number for fp/ap/mq

The frame pointer and the argument pointer aren't real registers.  MQ
was a register on old POWER.  All three are still used as arguments to
rs6000_dbx_register_number during initialisation.  If we handle them
explicitly we can do a gcc_unreachable to catch other unexpected
registers.

* config/rs6000/rs6000.c (rs6000_dbx_register_number): Handle
FRAME_POINTER_REGNUM, ARG_POINTER_REGNUM, and 64 (which was MQ).

From-SVN: r270925

5 years agore PR tree-optimization/88709 (Improve store-merging)
Jakub Jelinek [Mon, 6 May 2019 21:50:14 +0000 (23:50 +0200)]
re PR tree-optimization/88709 (Improve store-merging)

PR tree-optimization/88709
PR tree-optimization/90271
* params.def (PARAM_STORE_MERGING_MAX_SIZE): New parameter.
* gimple-ssa-store-merging.c (encode_tree_to_bitpos): Handle
non-clobber CONSTRUCTORs with no elts.  Remove useless tmp_int
variable.
(imm_store_chain_info::coalesce_immediate_stores): Punt if the size
of the store merging group is larger than
PARAM_STORE_MERGING_MAX_SIZE parameter.
(split_group): Add bzero_first argument.  If set, always emit first
the first store which must be = {} of the whole area and then for the
rest of the stores consider all zero bytes as paddings.
(imm_store_chain_info::output_merged_store): Check if first store
is = {} of the whole area and if yes, determine which setting of
bzero_first for split_group gives smaller number of stores.  Adjust
split_group callers.
(lhs_valid_for_store_merging_p): Allow decls.
(rhs_valid_for_store_merging_p): Allow non-clobber CONTRUCTORs with
no elts.
(pass_store_merging::process_store): Likewise.

* gcc.dg/store_merging_26.c: New test.
* gcc.dg/store_merging_27.c: New test.
* gcc.dg/store_merging_28.c: New test.
* gcc.dg/store_merging_29.c: New test.

From-SVN: r270924

5 years agore PR target/89424 (__builtin_vec_ext_v1ti (v, i) results in ICE with variable i...
Kelvin Nilsen [Mon, 6 May 2019 17:00:46 +0000 (17:00 +0000)]
re PR target/89424 (__builtin_vec_ext_v1ti (v, i) results in ICE with variable i (RS6000))

gcc/ChangeLog:

2019-05-06  Kelvin Nilsen  <kelvin@gcc.gnu.org>

PR target/89424
* config/rs6000/rs6000.c (rs6000_expand_vector_extract): Add
handling of V1TImode.

gcc/testsuite/ChangeLog:

2019-05-06  Kelvin Nilsen  <kelvin@gcc.gnu.org>

PR target/89424
* gcc.target/powerpc/pr89424-0.c: New test.
* gcc.target/powerpc/vsx-builtin-13a.c: Define macro PR89424 to
enable testing of newly patched capability.
* gcc.target/powerpc/vsx-builtin-13b.c: Likewise.
* gcc.target/powerpc/vsx-builtin-20a.c: Likewise.
* gcc.target/powerpc/vsx-builtin-20b.c: Likewise.

From-SVN: r270918

5 years agoPR c++/90265 - ICE with generic lambda.
Marek Polacek [Mon, 6 May 2019 16:56:39 +0000 (16:56 +0000)]
PR c++/90265 - ICE with generic lambda.

* pt.c (tsubst_copy_and_build): Use a dedicated variable for the last
element in the vector.

* g++.dg/cpp1y/lambda-generic-90265.C: New test.

From-SVN: r270917

5 years agore PR target/89221 (--enable-frame-pointer does not work as intended)
Uros Bizjak [Mon, 6 May 2019 14:58:57 +0000 (16:58 +0200)]
re PR target/89221 (--enable-frame-pointer does not work as intended)

PR target/89221
* config.gcc (i[34567]86-*-*, x86_64-*-*): Move tests for enable_cld
and enable_frame_pointer ...
* configure.ac: ... here.  Update help strings for
--enable-frame-pointer.
* configure: Regenerate.
* config/i386/i386-options.c (ix86_option_override_internal): Remove
USE_X86_64_FRAME_POINTER define, use USE_IX86_FRAME_POINTER instead.
* config/i386/sol2.h (USE_IX86_FRAME_POINTER): Remove.
(USE_X86_64_FRAME_POINTER): Ditto.

From-SVN: r270914

5 years agoAppend to target_gtfiles in order to fix Darwin bootstrap.
Martin Liska [Mon, 6 May 2019 14:03:27 +0000 (16:03 +0200)]
Append to target_gtfiles in order to fix Darwin bootstrap.

2019-05-06  Martin Liska  <mliska@suse.cz>

* config.gcc: Append to target_gtfiles and fix indentation.

From-SVN: r270913

5 years agore PR tree-optimization/90358 (526.blender_r train run does not finish after r270847...
Richard Biener [Mon, 6 May 2019 13:40:00 +0000 (13:40 +0000)]
re PR tree-optimization/90358 (526.blender_r train run does not finish after r270847 on znver1)

2019-05-06  Richard Biener  <rguenther@suse.de>

PR tree-optimization/90358
* tree-vect-stmts.c (get_group_load_store_type): Properly
detect unused upper half of load.
(vectorizable_load): Likewise.

* gcc.target/i386/pr90358.c: New testcase.

From-SVN: r270911

5 years agoThis reverts commit r270894, as new testcase fails on arm and aarch64, I am sorry...
Li Jia He [Mon, 6 May 2019 13:17:20 +0000 (13:17 +0000)]
This reverts commit r270894, as new testcase fails on arm and aarch64, I am sorry for this.

From-SVN: r270910

5 years agore PR tree-optimization/88828 (Inefficient update of the first element of vector...
Richard Biener [Mon, 6 May 2019 12:43:30 +0000 (12:43 +0000)]
re PR tree-optimization/88828 (Inefficient update of the first element of vector registers)

2019-05-06  Richard Biener  <rguenther@suse.de>

PR tree-optimization/88828
* tree-ssa-forwprop.c (get_bit_field_ref_def): Split out from...
(simplify_vector_constructor): ...here.  Handle constants in
the constructor.

* gcc.target/i386/pr88828-0.c: New testcase.

From-SVN: r270908

5 years agore PR tree-optimization/90328 (Wrong loop distribution with aliasing)
Richard Biener [Mon, 6 May 2019 12:38:35 +0000 (12:38 +0000)]
re PR tree-optimization/90328 (Wrong loop distribution with aliasing)

2019-05-06  Richard Biener  <rguenther@suse.de>

PR tree-optimization/90328
* tree-data-ref.h (dr_may_alias_p): Pass in the actual loop nest.
* tree-data-ref.c (dr_may_alias_p): Check whether the clique
is valid in the loop nest before using it.
(initialize_data_dependence_relation): Adjust.
* graphite-scop-detection.c (build_alias_set): Pass the SCOP enclosing
loop as loop-nest to dr_may_alias_p.

* gcc.dg/torture/pr90328.c: New testcase.

From-SVN: r270906

5 years ago[libcpp] struct deps renaming
Nathan Sidwell [Mon, 6 May 2019 11:34:47 +0000 (11:34 +0000)]
[libcpp] struct deps renaming

https://gcc.gnu.org/ml/gcc-patches/2019-05/msg00199.html
libcpp/
* include/mkdeps.h: Rename struct deps to struct mkdeps.
* mkdeps.c: Likewise.
* include/cpplib.h (cpp_get_deps): Rename return type..
* directives.c (cpp_get_deps): Likewise.
* internal.h (struct cpp_reader): Rename deps field type.

gcc/c-family/
* c-opts.c (handle_defered_opts): Rename struct deps to struc mkdeps.

From-SVN: r270905

5 years agodwarf2out.c (mem_loc_descriptor): Initialize int_mode.
Richard Biener [Mon, 6 May 2019 09:38:09 +0000 (09:38 +0000)]
dwarf2out.c (mem_loc_descriptor): Initialize int_mode.

2019-05-06  Richard Biener  <rguenther@suse.de>

* dwarf2out.c (mem_loc_descriptor): Initialize int_mode.

From-SVN: r270903

5 years agore PR tree-optimization/90316 (large compile time increase in opt / alias stmt walkin...
Richard Biener [Mon, 6 May 2019 08:54:40 +0000 (08:54 +0000)]
re PR tree-optimization/90316 (large compile time increase in opt / alias stmt walking for Go example)

2019-05-06  Richard Biener  <rguenther@suse.de>

PR tree-optimization/90316
* tree-ssa-alias.c (maybe_skip_until): Pass in target BB,
compute target on demand.
(get_continuation_for_phi): Remove code walking stmts to
get to a target virtual operand which could end up being
quadratic.

From-SVN: r270902

5 years agoClean up libgomp GCC 5 legacy support
Thomas Schwinge [Mon, 6 May 2019 08:49:55 +0000 (10:49 +0200)]
Clean up libgomp GCC 5 legacy support

libgomp/
* oacc-parallel.c: Add comments to legacy entry points (GCC 5).

From-SVN: r270901

5 years agore PR testsuite/90331 (New test case gcc.dg/pr87314-1.c fails)
Richard Biener [Mon, 6 May 2019 07:38:26 +0000 (07:38 +0000)]
re PR testsuite/90331 (New test case gcc.dg/pr87314-1.c fails)

2019-05-06  Richard Biener  <rguenther@suse.de>

PR testsuite/90331
* gcc.dg/pr87314-1.c: Align the substring to open up
string merging for targets aligning strings to 8 bytes.

From-SVN: r270900

5 years agoError only when a non-default -mabi is used with sanitizers (PR sanitizer/90312).
Martin Liska [Mon, 6 May 2019 07:35:59 +0000 (09:35 +0200)]
Error only when a non-default -mabi is used with sanitizers (PR sanitizer/90312).

2019-05-06  Martin Liska  <mliska@suse.cz>

PR sanitizer/90312
* config/i386/i386-options.c (ix86_option_override_internal): Error only
when -mabi is selected to a non-default version.
2019-05-06  Martin Liska  <mliska@suse.cz>

PR sanitizer/90312
* gcc.dg/asan/pr87930.c: Run the test only on *linux or *gnu
systems.
* gcc.dg/tsan/pr88017.c: Likewise.

From-SVN: r270899

5 years agoFix GNU coding style in lto-common.c.
Martin Liska [Mon, 6 May 2019 07:23:35 +0000 (09:23 +0200)]
Fix GNU coding style in lto-common.c.

2019-05-06  Martin Liska  <mliska@suse.cz>

* lto-common.c: Update coding style.
* lto.c (materialize_cgraph): Likewise.

From-SVN: r270898

5 years agoAdd lto-dump tool.
Hrishikesh Kulkarni [Mon, 6 May 2019 07:23:25 +0000 (07:23 +0000)]
Add lto-dump tool.

2019-05-06  Hrishikesh Kulkarni  <hrishikeshparag@gmail.com>
    Martin Liska  <mliska@suse.cz>

* Makefile.in: Add lto-dump.texi.
* cgraph.h: Add new functions get_visibility_string and
get_symtab_type_string.
* doc/gcc.texi: Include lto-dump section.
* doc/lto-dump.texi: New file.
* dumpfile.c (dump_switch_p_1): Use parse_dump_option.
(parse_dump_option): Factor out this function.
* dumpfile.h (enum dump_flag): Add new value TDF_ERROR.
(parse_dump_option): Export the function.
* symtab.c (symtab_node::get_visibility_string): New function.
(symtab_node::get_symtab_type_string): Likewise.
2019-05-06  Hrishikesh Kulkarni  <hrishikeshparag@gmail.com>
    Martin Liska  <mliska@suse.cz>

* Make-lang.in: Add lto_dump-related definition.
* config-lang.in: Likewise.
* lang.opt: Add new language LTODump and options related
to LTO dump tool.
* lto-common.c (lto_read_decls): Support type statistics dump.
(lto_file_read): Likewise for object files.
* lto-dump.c: New file.
* lto-lang.c (lto_option_lang_mask): Move from ..
* lto.c (lto_option_lang_mask): .. here.
* lto.h (lto_option_lang_mask): New declaration.

Co-Authored-By: Martin Liska <mliska@suse.cz>
From-SVN: r270897

5 years agoSplit part of functionality from lto.c to lto-common.c.
Martin Liska [Mon, 6 May 2019 07:23:03 +0000 (09:23 +0200)]
Split part of functionality from lto.c to lto-common.c.

2019-05-06  Martin Liska  <mliska@suse.cz>
    Hrishikesh Kulkarni  <hrishikeshparag@gmail.com>

* Make-lang.in: Add lto-common.o.
* config-lang.in: Likewise for gtfiles.
* lto-common.c: New file.
* lto-common.h: New file.
* lto-lang.c: Include lto-common.h.
* lto.c: Move most of the functionality
into lto-common.c file.

Co-Authored-By: Hrishikesh Kulkarni <hrishikeshparag@gmail.com>
From-SVN: r270896

5 years agoSplit i386.c.
Martin Liska [Mon, 6 May 2019 07:18:26 +0000 (09:18 +0200)]
Split i386.c.

2019-05-06  Martin Liska  <mliska@suse.cz>

* config/i386/i386-builtins.c: New file.
* config/i386/i386-builtins.h: New file.
* config/i386/i386-expand.c: New file.
* config/i386/i386-expand.h: New file.
* config/i386/i386-features.c: New file.
* config/i386/i386-features.h: New file.
* config/i386/i386-options.c: New file.
* config/i386/i386-options.h: New file.
* config.gcc: Add new files into extra_objs and
target_gtfiles.
* config/i386/i386.c: Split content of the file
into newly introduced files.
* config/i386/i386.h: Declare common variables
and macros.
* config/i386/t-i386: Define dependencies for new files.

From-SVN: r270895

5 years agoFix a typo in two_value_replacement function
Li Jia He [Mon, 6 May 2019 06:00:03 +0000 (06:00 +0000)]
Fix a typo in two_value_replacement function

GCC revision 267634 implemented two_value_replacement function.
However, a typo occurred during the parameter check, which caused
us to miss some optimizations.

The intent of the code might be to check that the input parameters
are const int and their difference is one.  However, when I read
the code, I found that it is wrong to detect whether an input data
plus one is equal to itself.  This could be a typo.

2019-05-06  Li Jia He  <helijia@linux.ibm.com>

* tree-ssa-phiopt.c (two_value_replacement): Fix a typo in parameter
detection.
* gcc.dg/tree-ssa/pr37508.c: Add the no-ssa-phiopt option to skip phi
optimization.
* gcc.dg/tree-ssa/pr88676-2.c: New testcase.

From-SVN: r270894

5 years agoprinters.py (add_one_template_type_printer): Add type printer for container types...
François Dumont [Mon, 6 May 2019 05:33:23 +0000 (05:33 +0000)]
printers.py (add_one_template_type_printer): Add type printer for container types in std::__debug namespace.

2019-05-06  François Dumont  <fdumont@gcc.gnu.org>

* python/libstdcxx/v6/printers.py (add_one_template_type_printer):
Add type printer for container types in std::__debug namespace.
* testsuite/lib/gdb-test.exp (whatis-regexp-test): New.
(gdb-tests): Use distinct parameters for the type of test and use of
regex.
(gdb-test): Check for regex test even if 'whatis' test.
* testsuite/libstdc++-prettyprinters/80276.cc: Adapt for _GLIBCXX_DEBUG
mode.
* testsuite/libstdc++-prettyprinters/cxx11.cc: Likewise.
* testsuite/libstdc++-prettyprinters/cxx17.cc: Likewise.
* testsuite/libstdc++-prettyprinters/libfundts.cc: Likewise.
* testsuite/libstdc++-prettyprinters/simple.cc: Likewise.
* testsuite/libstdc++-prettyprinters/simple11.cc: Likewise.
* testsuite/libstdc++-prettyprinters/whatis.cc: Likewise.
* testsuite/libstdc++-prettyprinters/whatis2.cc: Likewise.

From-SVN: r270893

5 years agoDaily bump.
GCC Administrator [Mon, 6 May 2019 00:16:16 +0000 (00:16 +0000)]
Daily bump.

From-SVN: r270892

5 years agore PR fortran/90344 (small code that compiles and runs in 7.3.0 but not 7.4.1)
Thomas Koenig [Sun, 5 May 2019 13:53:24 +0000 (13:53 +0000)]
re PR fortran/90344 (small code that compiles and runs in 7.3.0 but not 7.4.1)

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

PR fortran/90344
* gfortran.dg/pr90344.f90: New test.

From-SVN: r270882

5 years agoDaily bump.
GCC Administrator [Sun, 5 May 2019 00:16:19 +0000 (00:16 +0000)]
Daily bump.

From-SVN: r270881

5 years agoFix reference to wrong class in comment
Jonathan Wakely [Sat, 4 May 2019 15:52:52 +0000 (16:52 +0100)]
Fix reference to wrong class in comment

        * include/std/system_error (error_category): Fix comment.

From-SVN: r270877

5 years agoPR libstdc++/90299 make filesystem::absolute overloads consistent
Jonathan Wakely [Sat, 4 May 2019 14:35:33 +0000 (15:35 +0100)]
PR libstdc++/90299 make filesystem::absolute overloads consistent

In this implementation it is an error to pass the empty path to absolute,
because the empty path doesn't represent any file in the filesystem so
the function cannot meet its postcondition.

Currently the absolute(const path&, error_code&) overload reports an
error for the empty path, but using errc::no_such_file_or_directory, and
the other overload does not report an error. This patch makes them
consistntly report an errc::invalid_argument error for the empty path.

PR libstdc++/90299
* src/c++17/fs_ops.cc (absolute(const path&)): Report an error if the
argument is an empty path.
(absolute(const path&, error_code&)): Use invalid_argument as error
code instead of no_such_file_or_directory.
* testsuite/27_io/filesystem/operations/absolute.cc: Check handling
of non-existent paths and empty paths with both overloads of absolute.

From-SVN: r270874

5 years agoImprove API docs for <system_error> header
Jonathan Wakely [Sat, 4 May 2019 14:35:29 +0000 (15:35 +0100)]
Improve API docs for <system_error> header

* include/std/system_error (error_category, error_code)
(error_condition): Improve docs.
* libsupc++/exception: Add missing @addtogroup Doxygen command.
* libsupc++/exception_ptr.h (exception_ptr): Link equality operators
to class documentation. Suppress documentation for implementation
details.
* libsupc++/nested_exception.h (throw_with_nested, rethrow_if_nested):
Suppress documentation for implementation details.

From-SVN: r270873

5 years agoFix std::hash<std::error_condition>
Jonathan Wakely [Sat, 4 May 2019 14:35:25 +0000 (15:35 +0100)]
Fix std::hash<std::error_condition>

The hash value should be based on the identity (i.e. address) of the
error_category member, not its object representation (i.e. underlying
bytes).

* include/std/system_error (error_code): Remove friend declaration
for hash<error_code>.
(hash<error_code>::operator()): Use public member functions to access
value and category.
(hash<error_condition>::operator()): Use address of category, not
its object representation.
* src/c++11/compatibility-c++0x.cc (hash<error_code>::operator()):
Use public member functions to access value and category.
* testsuite/19_diagnostics/error_condition/hash.cc: New test.

From-SVN: r270872

5 years agocrontab: Snapshots from trunk are now GCC 10 related.
Gerald Pfeifer [Sat, 4 May 2019 13:28:49 +0000 (13:28 +0000)]
crontab: Snapshots from trunk are now GCC 10 related.

* crontab: Snapshots from trunk are now GCC 10 related.
Add GCC 9 snapshots from the respective branch.

From-SVN: r270871

5 years agoAdd missing PR libstdc++/90277 reference.
François Dumont [Sat, 4 May 2019 07:41:39 +0000 (07:41 +0000)]
Add missing PR libstdc++/90277 reference.

From-SVN: r270870

5 years agoAdd myself to MAINTAINERS.
Roland Illig [Sat, 4 May 2019 07:38:52 +0000 (07:38 +0000)]
Add myself to MAINTAINERS.

2019-05-04  Roland Illig  <roland.illig@gmx.de>

* MAINTAINERS (Write After Approval): Add myself.

From-SVN: r270869

5 years agohashtable.h (_Hashtable<>::rehash): Review comment.
François Dumont [Sat, 4 May 2019 07:38:46 +0000 (07:38 +0000)]
hashtable.h (_Hashtable<>::rehash): Review comment.

2019-05-04  François Dumont  <fdumont@gcc.gnu.org>

* include/bits/hashtable.h (_Hashtable<>::rehash): Review comment.
* include/bits/hashtable_policy.h
(_Prime_rehash_policy::_M_bkt_for_elements): Use __builtin_ceill.
(_Power2_rehash_policy::_M_bkt_for_elements): Likewise.
(_Power2_rehash_policy::_M_next_bkt): Enforce returning a result not
smaller than input value rather than always greater. Preserve
_M_next_resize if called with 0 input. Use __builtin_floorl.
(_Power2_rehash_policy::_M_need_rehash): Rehash only if number of
elements + number of insertions is greater than _M_next_resize. Start
with 11 buckets if not told otherwise. Use __builtin_floorl.
(_Rehash_base<>::reserve): Use rehash policy _M_bkt_for_elements.
* src/c++11/hashtable_c++0x.cc (_Prime_rehash_policy::_M_next_bkt):
Preserve _M_next_resize if called with 0 input. Use __builtin_floorl.
(_Prime_rehash_policy::_M_need_rehash): Start with 11 buckets if not
told otherwise. Use __builtin_floorl.
* testsuite/23_containers/unordered_set/hash_policy/71181.cc: Adapt test
to also validate _Power2_rehash_policy.
* testsuite/23_containers/unordered_set/hash_policy/power2_rehash.cc:
Adapt.

From-SVN: r270868

5 years agoDaily bump.
GCC Administrator [Sat, 4 May 2019 00:16:16 +0000 (00:16 +0000)]
Daily bump.

From-SVN: r270867

5 years agocompiler: recognize and optimize array range clear
Ian Lance Taylor [Fri, 3 May 2019 21:45:35 +0000 (21:45 +0000)]
compiler: recognize and optimize array range clear

    Recognize

            for i := range a { a[i] = zero }

    for array or slice a, and rewrite it to call memclr, as the gc
    compiler does.

    Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/169398

From-SVN: r270862

5 years agoFix new testcase to not require std::copysign
Jonathan Wakely [Fri, 3 May 2019 19:25:05 +0000 (20:25 +0100)]
Fix new testcase to not require std::copysign

Use __builtin_copysign{,f,l} when std::copysign isn't available.

PR libstdc++/61761
* testsuite/26_numerics/complex/proj.cc: Don't assume <cmath> defines
std::copysign.

From-SVN: r270859

5 years agoAvoid -Woverflow warning in __numeric_limits_integer
Jonathan Wakely [Fri, 3 May 2019 19:13:31 +0000 (20:13 +0100)]
Avoid -Woverflow warning in __numeric_limits_integer

This is the same fix as was done for std::numeric_limits in r183905.

PR libstdc++/52119
* include/ext/numeric_traits.h (__glibcxx_min): Avoid integer
overflow warning with -Wpedantic -Wsystem-headers.

From-SVN: r270858

5 years agoos/user: disable TestGroupIds for AIX
Ian Lance Taylor [Fri, 3 May 2019 17:15:54 +0000 (17:15 +0000)]
os/user: disable TestGroupIds for AIX

    The corresponding Go Toolchain patch is CL 164039

    Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/175079

From-SVN: r270857

5 years ago[arm] PR target/89400 fix thumb1 unaligned access expansion
Richard Earnshaw [Fri, 3 May 2019 13:45:59 +0000 (13:45 +0000)]
[arm] PR target/89400 fix thumb1 unaligned access expansion

Armv6 has support for unaligned accesses to memory.  However, the
thumb1 code patterns were trying to use the 32-bit code constraints.
One failure mode from this was that the patterns are designed to be
compatible with conditional execution and this was then causing an
assert in the compiler.

The unaligned_loadhis pattern is only used for expanding extv, which
in turn is only enabled for systems supporting thumb2.  Given that
there is no simple expansion for a thumb1 sign-extending load (the
instruction has no immediate offset form and requires two registers in
the address) it seems simpler to just disable this for thumb1.

Fixed thusly:

PR target/89400
* config/arm/arm.md (unaligned_loadsi): Add variant for thumb1.
Restrict 'all' variant to 32-bit configurations.
(unaligned_loadhiu): Likewise.
(unaligned_storehi): Likewise.
(unaligned_storesi): Likewise.
(unaligned_loadhis): Disable when compiling for thumb1.

From-SVN: r270853

5 years agoLet ldist ignore clobbers
Marc Glisse [Fri, 3 May 2019 13:41:36 +0000 (15:41 +0200)]
Let ldist ignore clobbers

2019-05-03  Marc Glisse  <marc.glisse@inria.fr>

PR tree-optimization/90269
gcc/
* tree-loop-distribution.c (find_seed_stmts_for_distribution):
Ignore clobbers.

gcc/testsuite/
* g++.dg/tree-ssa/ldist-1.C: New file.

From-SVN: r270852

5 years agoCome up with is_empty for hash_{table,map,set}.
Martin Liska [Fri, 3 May 2019 12:37:22 +0000 (14:37 +0200)]
Come up with is_empty for hash_{table,map,set}.

2019-05-03  Martin Liska  <mliska@suse.cz>

* hash-map.h: Add is_empty function.
* hash-set.h: Likewise.
* hash-table.h: Likewise.
* dwarf2out.c (dwarf2out_finish): Use is_empty instead of
elements () == 0 (and similar usages).
* gimple-ssa-store-merging.c (pass_store_merging::terminate_and_process_all_chains): Likewise.
* gimplify.c (gimplify_bind_expr): Likewise.
(gimplify_switch_expr): Likewise.
* hash-map-tests.c (test_map_of_strings_to_int): Likewise.
* ipa-icf.c (sem_item_optimizer::remove_symtab_node): Likewise.
* postreload-gcse.c (dump_hash_table): Likewise.
(gcse_after_reload_main): Likewise.
* predict.c (combine_predictions_for_bb): Likewise.
* tree-parloops.c (reduction_phi): Likewise.
(separate_decls_in_region): Likewise.
(transform_to_exit_first_loop): Likewise.
(gen_parallel_loop): Likewise.
(gather_scalar_reductions): Likewise.
(try_create_reduction_list): Likewise.
* var-tracking.c (dump_vars): Likewise.
(emit_notes_for_changes): Likewise.
(vt_emit_notes): Likewise.
2019-05-03  Martin Liska  <mliska@suse.cz>

* call.c (build_aggr_conv): Use is_empty instead of
elements () == 0 (and similar usages).
* parser.c (cp_parser_lambda_introducer): Likewise.

From-SVN: r270851

5 years agore PR tree-optimization/90316 (large compile time increase in opt / alias stmt walkin...
Richard Biener [Fri, 3 May 2019 11:21:18 +0000 (11:21 +0000)]
re PR tree-optimization/90316 (large compile time increase in opt / alias stmt walking for Go example)

2019-05-03  Richard Biener  <rguenther@suse.de>

PR tree-optimization/90316
* tree-ssa-pre.c (pass_pre::execute): Re-compute DOM fast queries
before running VN.

From-SVN: r270848

5 years agotree-vect-stmts.c (get_group_load_store_type): Avoid peeling for gaps by loading...
Richard Biener [Fri, 3 May 2019 10:47:21 +0000 (10:47 +0000)]
tree-vect-stmts.c (get_group_load_store_type): Avoid peeling for gaps by loading only lower halves of vectors if possible.

2019-05-03  Richard Biener  <rguenther@suse.de>

* tree-vect-stmts.c (get_group_load_store_type): Avoid
peeling for gaps by loading only lower halves of vectors
if possible.
(vectorizable_load): Likewise.

* gcc.dg/vect/slp-reduc-sad-2.c: New testcase.

From-SVN: r270847

5 years agore PR tree-optimization/89518 (missed optimisation for array address calculations)
Richard Biener [Fri, 3 May 2019 10:46:13 +0000 (10:46 +0000)]
re PR tree-optimization/89518 (missed optimisation for array address calculations)

2019-05-03  Richard Biener  <rguenther@suse.de>

PR middle-end/89518
* match.pd: Add pattern to optimize (A / B) * B + (A % B) to A.

* gcc.dg/pr89518.c: New testcase.

From-SVN: r270846

5 years agore PR tree-optimization/87314 (pointless comparison of malloc result to a string...
Richard Biener [Fri, 3 May 2019 10:44:17 +0000 (10:44 +0000)]
re PR tree-optimization/87314 (pointless comparison of malloc result to a string not eliminated)

2019-05-03  Richard Biener  <rguenther@suse.de>

PR middle-end/87314
* match.pd (cmp (convert1?@2 addr@0) (convert2? addr@1)):
Handle STRING_CST vs DECL or STRING_CST.

* gcc.dg/pr87314-1.c: New testcase.

From-SVN: r270845

5 years agore PR middle-end/88963 (gcc generates terrible code for vectors of 64+ length which...
Richard Biener [Fri, 3 May 2019 10:39:56 +0000 (10:39 +0000)]
re PR middle-end/88963 (gcc generates terrible code for vectors of 64+ length which are not natively supported)

2019-05-03  Richard Biener  <rguenther@suse.de>

PR tree-optimization/88963
* tree-ssa-forwprop.c (pass_forwprop::execute): Rewrite
vector loads feeding only BIT_FIELD_REFs to component
loads.  Rewrite stores fed by CONSTRUCTORs to component
stores.

* gcc.dg/tree-ssa/ssa-fre-31.c: Disable forwprop.
* gcc.target/i386/pr88963-1.c: New testcase.
* gcc.target/i386/pr88963-2.c: Likewise.

From-SVN: r270844

5 years agore PR target/88809 (do not use rep-scasb for inline strlen/memchr)
Dominique d'Humieres [Fri, 3 May 2019 10:00:27 +0000 (12:00 +0200)]
re PR target/88809 (do not use rep-scasb for inline strlen/memchr)

2019-05-03  Dominique d'Humieres  <dominiq@gcc.gnu.org>

PR target/88809
* gcc.target/i386/pr88809.c: Adjust for darwin.
* gcc.target/i386/pr88809-2.c: Adjust for i386 and darwin.

From-SVN: r270843

5 years ago* gennews (files): Add files for GCC 9.
Jakub Jelinek [Fri, 3 May 2019 07:40:10 +0000 (09:40 +0200)]
* gennews (files): Add files for GCC 9.

From-SVN: r270837

5 years agoopts.h (finish_options): Remove lang_mask argument.
Jakub Jelinek [Fri, 3 May 2019 07:33:04 +0000 (09:33 +0200)]
opts.h (finish_options): Remove lang_mask argument.

* opts.h (finish_options): Remove lang_mask argument.
(print_help, help_option_argument): Declare.
* opts.c (print_help): Remove forward declaration, no longer static.
(finish_options): Remove lang_mask argument, don't call print_help
here.
* opts-global.c (decode_options): Adjust finish_option caller, call
print_help here.

From-SVN: r270836

5 years agore PR tree-optimization/90303 (ICE in hash_odr_name with fastcall attribute starting...
Jakub Jelinek [Fri, 3 May 2019 07:32:06 +0000 (09:32 +0200)]
re PR tree-optimization/90303 (ICE in hash_odr_name with fastcall attribute starting with r267359)

PR tree-optimization/90303
* ipa-devirt.c (obj_type_ref_class, get_odr_type): Don't use
TYPE_CANONICAL for TYPE_STRUCTURAL_EQUALITY_P types in !in_lto_p mode.

* g++.target/i386/pr90303.C: New test.

From-SVN: r270835

5 years agoMakefile.am (finclude): Remove.
Jakub Jelinek [Fri, 3 May 2019 07:20:04 +0000 (09:20 +0200)]
Makefile.am (finclude): Remove.

* Makefile.am (finclude): Remove.
* Makefile.in: Regenerated.

From-SVN: r270834

5 years agore PR c++/89698 (Run-time error due to optimization of field access after cast at...
Richard Biener [Fri, 3 May 2019 07:07:28 +0000 (07:07 +0000)]
re PR c++/89698 (Run-time error due to optimization of field access after cast at -Os/-O2 and higher)

2019-05-03  Richard Biener  <rguenther@suse.de>

PR tree-optimization/89698
* gimple-fold.c (canonicalize_constructor_val): Early out
for constants, handle unfolded INTEGER_CSTs as they appear in
C++ virtual table ctors.

* g++.dg/tree-ssa/pr89698.C: New testcase.

From-SVN: r270833

5 years agopasses.c (execute_function_todo): Remove dead code.
Richard Biener [Fri, 3 May 2019 07:05:34 +0000 (07:05 +0000)]
passes.c (execute_function_todo): Remove dead code.

2019-05-03  Richard Biener  <rguenther@suse.de>

* passes.c (execute_function_todo): Remove dead code.

From-SVN: r270832

5 years agoDaily bump.
GCC Administrator [Fri, 3 May 2019 00:16:20 +0000 (00:16 +0000)]
Daily bump.

From-SVN: r270831

5 years agoPR libstdc++/90314 fix non-equivalent declarations of std::swap
Jonathan Wakely [Thu, 2 May 2019 21:23:38 +0000 (22:23 +0100)]
PR libstdc++/90314 fix non-equivalent declarations of std::swap

In order to use the _GLIBCXX_NOEXCEPT_IF macro for an expression
containing commas I enclosed it in parentheses, so the preprocessor
wouldn't treat it as two arguments to the function-like macro. Clang
gives an error because now the noexcept-specifier noexcept((C)) is not
equivalent to the noexcept(C) one on the declaration of swap in
<type_traits>.

Instead of requiring extra parentheses around the expression, redefine
_GLIBCXX_NOEXCEPT_IF as a variadic macro (even though supporting that in
C++98 is a GNU extension).

PR libstdc++/90314
* include/bits/c++config (_GLIBCXX_NOEXCEPT_IF): Use variadic macro.
* include/bits/move.h (swap): Remove extra parentheses.

From-SVN: r270827

5 years agoRemove redundant __constexpr_addressof function
Jonathan Wakely [Thu, 2 May 2019 21:23:35 +0000 (22:23 +0100)]
Remove redundant __constexpr_addressof function

The std::__addressof function is always constexpr, even in C++14, so we
can just use that.

* include/experimental/bits/lfts_config.h: Improve doc markup.
* include/experimental/optional: Improve docs.
(_Has_addressof_mem, _Has_addressof_free, _Has_addressof)
(__constexpr_addressof): Remove.
(optional::operator->()): Use std::__addressof().
* include/std/optional (optional::operator->()): Adjust whitespace.
* testsuite/experimental/optional/constexpr/observers/2.cc: Check
that operator-> is still constexpr with overloaded operator&. Change
to compile-only test.
* testsuite/experimental/optional/constexpr/observers/3.cc: Change to
compile-only test.

From-SVN: r270826

5 years agoImprove smart pointer docs
Jonathan Wakely [Thu, 2 May 2019 21:23:25 +0000 (22:23 +0100)]
Improve smart pointer docs

* include/bits/shared_ptr.h: Improve docs.
* include/bits/shared_ptr_atomic.h: Likewise.
* include/bits/unique_ptr.h: Likewise. Adjust whitespace.

From-SVN: r270825

5 years ago* gcc.pot: Regenerate.
Joseph Myers [Thu, 2 May 2019 20:30:14 +0000 (21:30 +0100)]
* gcc.pot: Regenerate.

From-SVN: r270822

5 years agors6000: Make debug regnums independent of internal ones
Segher Boessenkool [Thu, 2 May 2019 18:46:38 +0000 (20:46 +0200)]
rs6000: Make debug regnums independent of internal ones

Where we use "internal GCC register numbers" in debug info, that
defines an ABI, so we cannot change those numbers.  But we want to
change the internal numbers, and sometimes we do that without
remembering this gotcha anyway; so let's make everything independent
of the internal numbers.

For those registers that are not recognised here (we still have MQ for
example, but also the GCC-internal frame pointer and arg pointer
registers), this just returns the internal register number.  This is a
bit worrying: that number could be the same as that for a register we
validly want to have in debug info.  I first had a gcc_unreachable ()
for that, but this does now work because dwarf2cfi calls
rs6000_dbx_register_number for every internal register.  Then I just
returned 0 for the internal regs, but that causes various regression
tests to fail.  So now I return the internal register number again,
as it was before; but this needs to be fixed.

* config/rs6000/rs6000.c (rs6000_dbx_register_number): Do not use
the internal register number, for any "real" register.

From-SVN: r270820

5 years agors6000: Debug regnums for TM registers
Segher Boessenkool [Thu, 2 May 2019 18:45:22 +0000 (20:45 +0200)]
rs6000: Debug regnums for TM registers

Since GCC 8, we have output incorrect numbers for the transactional
memory registers.

Also, we didn't output the correct DWARF register numbers for those.
The number for sprN is 100+N.

This fixes both these issues.

* config/rs6000/rs6000.c (rs6000_dbx_register_number): Return the
correct numbers for TFHAR, TFIAR, TEXASR.

From-SVN: r270819

5 years agolibphobos: Correct my identity in the last ChangeLog entry
Maciej W. Rozycki [Thu, 2 May 2019 17:03:25 +0000 (17:03 +0000)]
libphobos: Correct my identity in the last ChangeLog entry

From-SVN: r270817

5 years agolibphobos: RISC-V: Fix soft-float build errors with IEEE exception flags
Maciej W. Rozycki [Thu, 2 May 2019 16:38:08 +0000 (16:38 +0000)]
libphobos: RISC-V: Fix soft-float build errors with IEEE exception flags

Fix assembly errors:

.../libphobos/src/std/math.d: Assembler messages:.../libphobos/src/std/math.d:4773: Error: unrecognized opcode `frflags a0'.../libphobos/src/std/math.d:4856: Error: unrecognized opcode `fsflags a5'.../libphobos/src/std/math.d:4856: Error: unrecognized opcode `fsflags a5'.../libphobos/src/std/math.d:4773: Error: unrecognized opcode `frflags a0'.../libphobos/src/std/math.d:5549: Error: unrecognized opcode `fscsr a5'.../libphobos/src/std/math.d:5456: Error: unrecognized opcode `frcsr a5'.../libphobos/src/std/math.d:5456: Error: unrecognized opcode `frcsr a5'.../libphobos/src/std/math.d:5549: Error: unrecognized opcode `fscsr a5'.../libphobos/src/std/math.d:5456: Error: unrecognized opcode `frcsr a5'.../libphobos/src/std/math.d:5549: Error: unrecognized opcode `fscsr a0'.../libphobos/src/std/math.d:5456: Error: unrecognized opcode `frcsr a0'.../libphobos/src/std/math.d:5456: Error: unrecognized opcode `frcsr a0'.../libphobos/src/std/math.d:5549: Error: unrecognized opcode `fscsr s2'make[8]: *** [Makefile:1119: std/math.lo] Error 1

triggered with the RISC-V lp64 multilib in a GCC build configured with
`--enable-multilib --enable-languages=all --target=riscv64-linux-gnu'.
This is due to unconditional explicit use of F extension instructions
within inline assembly, to access IEEE exception flags.  The use of
these instructions is not allowed when building for a soft-float ABI.

Correct the problem by wrapping said inline assembly into a conditional
such that if `D_SoftFloat' is true, then reads from IEEE exception flags
return 0 and writes are ignored instead, complementing r270522
("libphobos: Add D support for RISC-V Linux"), which is an updated
version of <https://gcc.gnu.org/ml/gcc-patches/2019-04/msg00325.html>,
where the problematic code has originated from.

libphobos/ChangeLog:

2019-05-02  Maciej W. Rozycki  <macro@wdc.com>

* std/math.d (IeeeFlags.getIeeeFlags): Handle RISC-V soft-float ABI.
(IeeeFlags.resetIeeeFlags): Likewise.
(FloatingPointControl.getControlState): Likewise.
(FloatingPointControl.setControlState): Likewise.

From-SVN: r270815