platform/upstream/gcc.git
5 years agore PR tree-optimization/88415 (ICE: verify_gimple failed (error: dead STMT in EH...
Richard Biener [Mon, 10 Dec 2018 12:39:07 +0000 (12:39 +0000)]
re PR tree-optimization/88415 (ICE: verify_gimple failed (error: dead STMT in EH table))

2018-12-10  Richard Biener  <rguenther@suse.de>

PR middle-end/88415
* gimple.c (gimple_assign_set_rhs_with_ops): Transfer EH
info to a newly allocated stmt.

* gcc.dg/gomp/pr88415.c: New testcase.

From-SVN: r266951

5 years agolibphobos: Fix modify immutable error on Solaris.
Iain Buclaw [Mon, 10 Dec 2018 12:11:55 +0000 (12:11 +0000)]
libphobos: Fix modify immutable error on Solaris.

Backported from upstream druntime 2.079.

Reviewed-on: https://github.com/dlang/druntime/pull/2089

From-SVN: r266950

5 years agopatchable_function_entry-decl.c: Pass -mcpu=gr6 for Visium and remove other specific...
Eric Botcazou [Mon, 10 Dec 2018 11:16:47 +0000 (11:16 +0000)]
patchable_function_entry-decl.c: Pass -mcpu=gr6 for Visium and remove other specific handling.

* c-c++-common/patchable_function_entry-decl.c: Pass -mcpu=gr6 for
Visium and remove other specific handling.
* c-c++-common/patchable_function_entry-default.c: Likewise.
* c-c++-common/patchable_function_entry-definition.c: Likewise.

From-SVN: r266948

5 years agoAdd missing def of TARGET_VXWORKS_HAVE_CTORS_DTORS for VxWorksAE
Jerome Lambourg [Mon, 10 Dec 2018 10:44:28 +0000 (10:44 +0000)]
Add missing def of TARGET_VXWORKS_HAVE_CTORS_DTORS for VxWorksAE

* config/vxworksae.h (TARGET_VXWORKS_HAVE_CTORS_DTORS): Define.

From-SVN: r266947

5 years agoDon't try to use libgcc-unwind.map with --disable-shared (PR bootstrap/65725)
Fredrik Nyström [Mon, 10 Dec 2018 09:49:02 +0000 (10:49 +0100)]
Don't try to use libgcc-unwind.map with --disable-shared (PR bootstrap/65725)

2018-12-10  Fredrik Nyström  <fredrik@lysator.liu.se>

PR bootstrap/65725
* config/sol2.h: Only use libgcc-unwind.map if
ENABLE_SHARED_LIBGCC.

From-SVN: r266946

5 years agoauto-profile.c (afdo_calculate_branch_prob): Convert profile_count to afdo precision...
Bin Cheng [Mon, 10 Dec 2018 07:02:49 +0000 (07:02 +0000)]
auto-profile.c (afdo_calculate_branch_prob): Convert profile_count to afdo precision quality.

* auto-profile.c (afdo_calculate_branch_prob): Convert profile_count
to afdo precision quality.
(afdo_callsite_hot_enough_for_early_inline): Likewise.
(afdo_annotate_cfg): Set profile_count for exit basic block.

From-SVN: r266943

5 years ago* auto-profile.c (string_table::get_index_by_decl)
Bin Cheng [Mon, 10 Dec 2018 06:54:02 +0000 (06:54 +0000)]
* auto-profile.c (string_table::get_index_by_decl)
(function_instance::get_function_instance_by_decl): Avoid infinite
recursion by using DECL_FROM_INLINE.

From-SVN: r266942

5 years agoDaily bump.
GCC Administrator [Mon, 10 Dec 2018 00:16:47 +0000 (00:16 +0000)]
Daily bump.

From-SVN: r266939

5 years agore PR fortran/88205 (ICE in gfc_wide_strncasecmp, at fortran/scanner.c:249)
Steven G. Kargl [Sun, 9 Dec 2018 23:49:14 +0000 (23:49 +0000)]
re PR fortran/88205 (ICE in gfc_wide_strncasecmp, at fortran/scanner.c:249)

2018-12-09  Steven G. Kargl  <kargl@gcc.gnu.org>

PR fortran/88205
* io.c (gfc_match_open): Move NEWUNIT checks to after STATUS checks.

2018-12-09  Steven G. Kargl  <kargl@gcc.gnu.org>

PR fortran/88205
* gfortran.dg/pr88205.f90: New unit.

From-SVN: r266936

5 years agolibphobos: Add druntime bindings for SPARC/Solaris
Iain Buclaw [Sun, 9 Dec 2018 23:46:45 +0000 (23:46 +0000)]
libphobos: Add druntime bindings for SPARC/Solaris

Backported from upstream druntime master.

Initial patch by Rainer Orth.

Reviewed-on: https://github.com/dlang/druntime/pull/2344

From-SVN: r266935

5 years agoGenerate and scan documentation output in Ddoc tests.
Iain Buclaw [Sun, 9 Dec 2018 19:55:54 +0000 (19:55 +0000)]
Generate and scan documentation output in Ddoc tests.

The tests in gdc.test/compilable/ddoc*.d don't require the module to be
compiled all the way down to object code.  Instead, only compile the
test sources with -fdoc, and scan the generated html content.

gcc/testsuite/ChangeLog:

PR d/88039
* gdc.test/gdc-test.exp (gdc-convert-args): Handle -D.
(dmd2dg): Check generated html in ddoc tests.
(gdc-do-test): Set dg-do-what-default to compile for ddoc tests.

From-SVN: r266933

5 years agoconfig.gcc (hppa*-*-linux*): Add pa/t-pa to tmake_file.
John David Anglin [Sun, 9 Dec 2018 19:41:39 +0000 (19:41 +0000)]
config.gcc (hppa*-*-linux*): Add pa/t-pa to tmake_file.

* config.gcc (hppa*-*-linux*): Add pa/t-pa to tmake_file.  Define
d_target_objs.
* config/pa/pa-protos.h (pa_d_target_versions): New prototype.
* config/pa/pa.h (TARGET_D_CPU_VERSIONS): Define.
* config/pa/pa-d.c: New file.
* config/pa/t-pa: New file.

From-SVN: r266931

5 years agore PR fortran/88206 (ICE in gfc_match_type_spec, at fortran/match.c:2229)
Steven G. Kargl [Sun, 9 Dec 2018 19:10:57 +0000 (19:10 +0000)]
re PR fortran/88206 (ICE in gfc_match_type_spec, at fortran/match.c:2229)

2018-12-09  Steven G. Kargl  <kargl@gcc.gnu.org>

PR fortran/88206
* match.c (gfc_match_type_spec): REAL can be an intrinsic function.

2018-12-09  Steven G. Kargl  <kargl@gcc.gnu.org>

PR fortran/88206
* gfortran.dg/pr88206.f90: New test.

From-SVN: r266930

5 years agore PR libfortran/88411 (Random crashes for ASYNCHRONOUS writes (bad locking?))
Thomas Koenig [Sun, 9 Dec 2018 18:54:47 +0000 (18:54 +0000)]
re PR libfortran/88411 (Random crashes for ASYNCHRONOUS writes (bad locking?))

2018-12-09  Thomas Koenig  <tkoenig@gcc.gnu.org>

PR fortran/88411
* io/transfer.c (dta_transfer_init): Do not treat as an
asynchronous statement unless the statement has
ASYNCHRONOUS="YES".
(st_write_done): Likewise.
(st_read_done): Do not perform async_wait for synchronous I/O
on an async unit.
(st_read_done): Likewise.

2018-12-09  Thomas Koenig  <tkoenig@gcc.gnu.org>

PR fortran/88411
* testsuite/libgomp.fortran/async_io_8.f90: New test.

From-SVN: r266929

5 years ago[multiple changes]
Steven G. Kargl [Sun, 9 Dec 2018 16:32:48 +0000 (16:32 +0000)]
[multiple changes]

2018-12-09  Fritz Reese  <fritzoreese@gmail.com>

PR fortran/88228
* resolve.c (resolve_operator):  Do not call resolve_function.
Break like other cases.

2018-12-09  Steven G. Kargl  <kargl@gcc.gnu.org>

PR fortran/88228
* gfortran.dg/pr88228.f90: New test.

From-SVN: r266926

5 years agoMerge dmd upstream e2fe2687b
Iain Buclaw [Sun, 9 Dec 2018 15:06:22 +0000 (15:06 +0000)]
Merge dmd upstream e2fe2687b

Backports VRP fixes from the D front-end implementation to the C++ port,
and fixes errors reported by ubsan build where the conversion from D
didn't include adjusting integer suffixes from 'UL' to 'ULL'.

Fixes https://gcc.gnu.org/PR88366

Reviewed-on: https://github.com/dlang/dmd/pull/9046

From-SVN: r266925

5 years agoCorrect the reported line number in Fortran combined OpenACC directives
Thomas Schwinge [Sun, 9 Dec 2018 12:49:20 +0000 (13:49 +0100)]
Correct the reported line number in Fortran combined OpenACC directives

gcc/fortran/
* trans-openmp.c (gfc_trans_oacc_combined_directive): Set the
location of combined acc loops.
gcc/testsuite/
* gfortran.dg/goacc/combined-directives-3.f90: New file.

Reviewed-by: Thomas Schwinge <thomas@codesourcery.com>
From-SVN: r266924

5 years agoCorrect the reported line number in C++ combined OpenACC directives
Cesar Philippidis [Sun, 9 Dec 2018 12:48:26 +0000 (04:48 -0800)]
Correct the reported line number in C++ combined OpenACC directives

The C++ FE doesn't set the expr_location of the split acc loop in combined acc
parallel/kernels loop directives. This only happens for with combined
directives, otherwise cp_parser_omp_construct would be responsible for setting
the location. After fixing this bug, I was able to resolve a couple of long
standing diagnostics discrepancies between the C/C++ FEs in the test suite.

gcc/cp/
* parser.c (cp_parser_oacc_kernels_parallel): Adjust EXPR_LOCATION
on the combined acc loop.
gcc/testsuite/
* c-c++-common/goacc/combined-directives-3.c: New test.
* c-c++-common/goacc/loop-2-kernels.c (void K): Adjust test.
* c-c++-common/goacc/loop-2-parallel.c (void P): Adjust test.
* c-c++-common/goacc/loop-3.c (void p2): Adjust test.

Reviewed-by: Thomas Schwinge <thomas@codesourcery.com>
From-SVN: r266923

5 years agoUse existing middle end checking for Fortran OpenACC loop clauses
Thomas Schwinge [Sun, 9 Dec 2018 12:47:58 +0000 (13:47 +0100)]
Use existing middle end checking for Fortran OpenACC loop clauses

Don't duplicate in the Fortran front end what's generically being checked in
the middle end.

gcc/fortran/
* openmp.c (resolve_oacc_loop_blocks): Remove checking of OpenACC
loop clauses.
gcc/testsuite/
* gfortran.dg/goacc/loop-2-kernels.f95: Update.
* gfortran.dg/goacc/loop-2-parallel.f95: Likewise.
* gfortran.dg/goacc/nested-parallelism.f90: Likewise.

From-SVN: r266922

5 years agoSplit up "gfortran.dg/goacc/loop-2.f95"
Thomas Schwinge [Sun, 9 Dec 2018 12:47:47 +0000 (13:47 +0100)]
Split up "gfortran.dg/goacc/loop-2.f95"

gcc/testsuite/
* gfortran.dg/goacc/loop-2.f95: Split into...
* gfortran.dg/goacc/loop-2-kernels-nested.f95: ... this new
file...
* gfortran.dg/goacc/loop-2-kernels-tile.f95: ..., and this new
file...
* gfortran.dg/goacc/loop-2-kernels.f95: ..., and this new file...
* gfortran.dg/goacc/loop-2-parallel-3.f95: ..., and this new
file...
* gfortran.dg/goacc/loop-2-parallel-nested.f95: ..., and this new
file...
* gfortran.dg/goacc/loop-2-parallel-tile.f95: ..., and this new
file...
* gfortran.dg/goacc/loop-2-parallel.f95: ..., and this new file.

From-SVN: r266921

5 years ago[PR88420] Fortran OpenACC "Clause SEQ conflicts with INDEPENDENT"
Thomas Schwinge [Sun, 9 Dec 2018 12:47:35 +0000 (13:47 +0100)]
[PR88420] Fortran OpenACC "Clause SEQ conflicts with INDEPENDENT"

The Fortran front end declares that the OpenACC "Clause SEQ conflicts with
INDEPENDENT".  While that combination doesn't make too much sense indeed, it's
still valid; these are orthogonal concepts.

gcc/fortran/
PR fortran/88420
* openmp.c (resolve_oacc_loop_blocks): Remove "Clause SEQ
conflicts with INDEPENDENT" diagnostic.
gcc/testsuite/
PR fortran/88420
* gfortran.dg/goacc/loop-1-2.f95: Update.
* gfortran.dg/goacc/loop-1.f95: Likewise.

From-SVN: r266920

5 years agoCoalesce host to device transfers in libgomp: not for link pointer
Thomas Schwinge [Sun, 9 Dec 2018 12:47:23 +0000 (13:47 +0100)]
Coalesce host to device transfers in libgomp: not for link pointer

libgomp/
* target.c (gomp_map_vars): Call gomp_copy_host2dev instead of
devicep->host2dev_func.

Co-Authored-By: Jakub Jelinek <jakub@redhat.com>
From-SVN: r266919

5 years agoipa-cp.c (update_profiling_info): Call adjust_for_ipa_scaling for zero profile count.
Bin Cheng [Sun, 9 Dec 2018 06:37:56 +0000 (06:37 +0000)]
ipa-cp.c (update_profiling_info): Call adjust_for_ipa_scaling for zero profile count.

* ipa-cp.c (update_profiling_info): Call adjust_for_ipa_scaling for
zero profile count.

From-SVN: r266918

5 years agore PR fortran/88048 (ICE in check_data_variable, at fortran/resolve.c:15491)
Steven G. Kargl [Sun, 9 Dec 2018 06:09:47 +0000 (06:09 +0000)]
re PR fortran/88048 (ICE in check_data_variable, at fortran/resolve.c:15491)

2018-12-08  Steven G. Kargl  <kargl@gcc.gnu.org>

PR fortran/88048
* resolve.c (check_data_variable): Named constant cannot be a
data object.

2018-12-08  Steven G. Kargl  <kargl@gcc.gnu.org>

PR fortran/88048
* gfortran.dg/pr88048.f90: New test.

From-SVN: r266916

5 years ago20180-12-08 Steven G. Kargl <kargl@gcc.gnu.org>
Steven G. Kargl [Sun, 9 Dec 2018 04:02:44 +0000 (04:02 +0000)]
20180-12-08  Steven G. Kargl  <kargl@gcc.gnu.org>

PR fortran/87945
* decl.c (var_element): Inquiry parameters cannit be data objects.

20180-12-08  Steven G. Kargl  <kargl@gcc.gnu.org>

PR fortran/87945
* gfortran.dg/pr87945_1.f90: New test.
* gfortran.dg/pr87945_2.f90: New test.

From-SVN: r266915

5 years agore PR fortran/88025 (ICE in gfc_apply_init, at fortran/expr.c:4468)
Steven G. Kargl [Sun, 9 Dec 2018 01:02:41 +0000 (01:02 +0000)]
re PR fortran/88025 (ICE in gfc_apply_init, at fortran/expr.c:4468)

2018-12-08  Steven G. Kargl  <kargl@gcc.gnu.org>

PR fortran/88025
* expr.c (gfc_apply_init):  Remove asserts that cannot trigger.
Check for a NULL pointer.

2018-12-08  Steven G. Kargl  <kargl@gcc.gnu.org>

PR fortran/88025
* gfortran.dg/pr88025.f90: New test.

From-SVN: r266913

5 years agoDaily bump.
GCC Administrator [Sun, 9 Dec 2018 00:16:33 +0000 (00:16 +0000)]
Daily bump.

From-SVN: r266912

5 years agoc: Delete a stray line in asm inline
Segher Boessenkool [Sat, 8 Dec 2018 22:31:09 +0000 (23:31 +0100)]
c: Delete a stray line in asm inline

I noticed I accidentally copied a line too many from the "volatile"
handling to the "inline" handling.  This fixes it.

gcc/c/
* c-parser (c_parser_asm_statement) [RID_INLINE]: Delete stray line
setting "quals".

From-SVN: r266909

5 years agore PR fortran/88357 (ICE in parse_associate, at fortran/parse.c:4568)
Steven G. Kargl [Sat, 8 Dec 2018 18:09:05 +0000 (18:09 +0000)]
re PR fortran/88357 (ICE in parse_associate, at fortran/parse.c:4568)

2018-12-08  Steven G. Kargl  <kargl@gcc.gnu.org>

PR fortran/88357
* class.c (insert_component_ref): Check for NULL pointer and
previous error message issued.
* parse.c (parse_associate): Check for NULL pointer.
* resolve.c (resolve_assoc_var): Check for NULL pointer.

2018-12-08  Steven G. Kargl  <kargl@gcc.gnu.org>

* gfortran.dg/pr88357_1.f90: New test.
* gfortran.dg/pr88357_2.f90: New test.

From-SVN: r266908

5 years agore PR fortran/88304 (ICE in use_pointer_in_frame, at tree-nested.c:267)
Jakub Jelinek [Sat, 8 Dec 2018 13:37:49 +0000 (14:37 +0100)]
re PR fortran/88304 (ICE in use_pointer_in_frame, at tree-nested.c:267)

PR fortran/88304
* tree-nested.c (convert_local_reference_stmt): Handle clobbers where
lhs is not a decl normally, don't call use_pointer_in_frame on that
lhs.

* gfortran.fortran-torture/compile/pr88304-2.f90: New test.

From-SVN: r266907

5 years agore PR rtl-optimization/88390 (g++.dg/tree-prof/pr57451.C FAILs)
Eric Botcazou [Sat, 8 Dec 2018 11:07:53 +0000 (11:07 +0000)]
re PR rtl-optimization/88390 (g++.dg/tree-prof/pr57451.C FAILs)

PR rtl-optimization/88390
* dwarf2cfi.c (struct dw_cfi_row): Add window_save field.
(cfi_row_equal_p): Compare it.
(dwarf2out_frame_debug_cfa_window_save): Add FAKE parameter.
If FAKE is false, set window_save of the current row.
(dwarf2out_frame_debug) <REG_CFA_TOGGLE_RA_MANGLE>: Call above
function with FAKE parameter set to true.
<REG_CFA_WINDOW_SAVE>: Likewise but with FAKE parameter set to false.
(change_cfi_row): Emit a DW_CFA_GNU_window_save if necessary.

From-SVN: r266906

5 years agoPass GDCFLAGS and CCASFLAGS to libphobos subdirs
Rainer Orth [Sat, 8 Dec 2018 10:02:19 +0000 (10:02 +0000)]
Pass GDCFLAGS and CCASFLAGS to libphobos subdirs

* Makefile.am (AM_MAKEFLAGS): Pass CCASFLAGS, GDCFLAGS.
* Makefile.in: Regenerate.

From-SVN: r266905

5 years agore PR libgomp/87995 (libgomp.c/../libgomp.c-c++-common/cancel-taskgroup-3.c fails...
Jakub Jelinek [Sat, 8 Dec 2018 08:58:24 +0000 (09:58 +0100)]
re PR libgomp/87995 (libgomp.c/../libgomp.c-c++-common/cancel-taskgroup-3.c fails consistently after r265930)

PR libgomp/87995
* testsuite/libgomp.c-c++-common/cancel-taskgroup-3.c: Require
tls_runtime effective target.
(t): New threadprivate variable.
(main): Set t in threads which execute iterations of the worksharing
loop.  Propagate that to the task after the loop and don't abort
if the current taskgroup hasn't been cancelled.

From-SVN: r266904

5 years agoDaily bump.
GCC Administrator [Sat, 8 Dec 2018 00:16:34 +0000 (00:16 +0000)]
Daily bump.

From-SVN: r266903

5 years agodecl2.c (grokbitfield): Use DECL_SOURCE_LOCATION in error messages about bit-fields...
Paolo Carlini [Fri, 7 Dec 2018 17:48:39 +0000 (17:48 +0000)]
decl2.c (grokbitfield): Use DECL_SOURCE_LOCATION in error messages about bit-fields with function type...

/cp
2018-12-07  Paolo Carlini  <paolo.carlini@oracle.com>

* decl2.c (grokbitfield): Use DECL_SOURCE_LOCATION in error messages
about bit-fields with function type, warn_if_not_aligned type, and
static bit-fields; avoid DECL_NAME for unnamed declarations.

/testsuite
2018-12-07  Paolo Carlini  <paolo.carlini@oracle.com>

* g++.dg/other/bitfield7.C: New.
* g++.dg/parse/bitfield8.C: Likewise.
* g++.dg/parse/bitfield9.C: Likewise.
* g++.dg/pr53037-4.C: Test the locations too.

From-SVN: r266900

5 years agore PR target/87496 (ICE in aggregate_value_p at gcc/function.c:2046)
Peter Bergner [Fri, 7 Dec 2018 17:33:55 +0000 (17:33 +0000)]
re PR target/87496 (ICE in aggregate_value_p at gcc/function.c:2046)

gcc/
PR target/87496
* config/rs6000/rs6000.c (rs6000_option_override_internal): Disallow
-mabi=ieeelongdouble and -mabi=ibmlongdouble without -mlong-double-128.
Do not error for -mabi=ibmlongdouble and no ISA 2.06 support.
* doc/invoke.texi: Document -mabi=ibmlongdouble and -mabi=ieeelongdouble
require -mlong-double-128.

gcc/testsuite/
PR target/87496
* gcc.target/powerpc/pr87496.c: Rename from this...
* gcc.target/powerpc/pr87496-1.c: ...to this.  Update comment.
* gcc.target/powerpc/pr87496-2.c: New test.
* gcc.target/powerpc/pr87496-3.c: New test.

From-SVN: r266899

5 years ago[AArch64][2/2] Add sve_width -moverride tunable
Kyrylo Tkachov [Fri, 7 Dec 2018 17:05:21 +0000 (17:05 +0000)]
[AArch64][2/2] Add sve_width -moverride tunable

On top of the previous patch that implements TARGET_ESTIMATED_POLY_VALUE
and adds an sve_width tuning field to the CPU structs, this patch implements
an -moverride knob to adjust this sve_width field to allow for experimentation.
Again, reminder that this only has an effect when compiling for VLA-SVE that is,
without msve-vector-bits=<foo>. This just adjusts tuning heuristics in the compiler,,
like profitability thresholds for vectorised versioned loops, and others.

It can be used, for example like -moverride=sve_width=256 to set the sve_width
tuning field to 256. Widths outside of the accepted SVE widths [128 - 2048] are rejected
as you'd expect.

    * config/aarch64/aarch64.c (aarch64_tuning_override_functions): Add
    sve_width entry.
    (aarch64_parse_sve_width_string): Define.

    * gcc.target/aarch64/sve/override_sve_width_1.c: New test.

From-SVN: r266898

5 years agogimple-ssa-evrp-analyze.h (class evrp_range_analyzer): Add m_update_global_ranges...
Jeff Law [Fri, 7 Dec 2018 17:02:11 +0000 (10:02 -0700)]
gimple-ssa-evrp-analyze.h (class evrp_range_analyzer): Add m_update_global_ranges member.

* gimple-ssa-evrp-analyze.h (class evrp_range_analyzer): Add
m_update_global_ranges member.  Add corresponding argument to ctor.
* gimple-ssa-evrp-analyze.c
(evrp_range_analyzer::evrp_range_analyzer): Add new argument and
initialize m_update_global_ranges.
(evrp_range_analyzer::set_ssa_range_info): Assert that we are
updating global ranges.
(evrp_range_analyzer::record_ranges_from_incoming_edge): Only
update global ranges if explicitly requested.
(evrp_range_analyzer::record_ranges_from_phis): Similarly.
(evrp_range_analyzer::record_ranges_from_stmt): Similarly.
* gimple-ssa-evrp.c (evrp_dom_walker): Pass new argument to
evrp_range_analyzer ctor.
* gimple-ssa-sprintf.c (sprintf_dom_walker): Similarly.
* tree-ssa-dom.c (dom_opt_dom_walker): Similarly.

* gcc.c-torture/builtins/strnlen.x: New file to filter -Og from
options to test.

From-SVN: r266897

5 years ago[AArch64][1/2] Implement TARGET_ESTIMATED_POLY_VALUE
Kyrylo Tkachov [Fri, 7 Dec 2018 16:48:39 +0000 (16:48 +0000)]
[AArch64][1/2] Implement TARGET_ESTIMATED_POLY_VALUE

The hook TARGET_ESTIMATED_POLY_VALUE allows a target to give an estimate for a poly_int run-time value.
It is used exclusively in tuning decisions, things like estimated loop iterations, probabilities etc.
It is not relied on for correctness.

If we know the SVE width implemented in hardware we can make more more
informed decisions in the implementation of TARGET_ESTIMATED_POLY_VALUE,
even when compiling for VLA vectorisation.

This patch adds an sve_width field to our tuning structs and sets it for
the current CPU tunings.

A new value is introduced to the aarch64_sve_vector_bits_enum enum that indicates
that SVE is not available: SVE_NOT_IMPLEMENTED. I set it to the same value as SVE_SCALABLE
so that parts of the aarch64 backend that follow the pattern:
if (vector_width == SVE_SCALABLE)
  do_vla_friendly_action ()
else
  assume_specific_width_for_correctness ()

continue to work without change, but the CPU tuning structs can use a more
appropriate moniker for indicating the absence of SVE.

This sets sve_width to SVE_NOT_IMPLEMENTED for all cores.
I aim to add an -moverride switch in the next patch that allows a power user to experiment
with different values of it for investigations.

* config/aarch64/aarch64-opts.h (aarch64_sve_vector_bits_enum):
Add SVE_NOT_IMPLEMENTED value.
* config/aarch64/aarch64-protos.h (struct tune_params): Add sve_width
field.
* config/aarch64/aarch64.c (generic_tunings,cortexa35_tunings,
cortexa53_tunings, cortexa57_tunings, cortexa72_tunings,
cortexa73_tunings, exynosm1_tunings, thunderx_tunings,
thunderx_tunings, tsv110_tunings, xgene1_tunings, qdf24xx_tunings,
saphira_tunings, thunderx2t99_tunings, emag_tunings):
Specify sve_width.
(aarch64_estimated_poly_value): Define.
(TARGET_ESTIMATED_POLY_VALUE): Define.

From-SVN: r266896

5 years ago[rs6000] mmintrin.h: fix use of "vector"
Paul A. Clarke [Fri, 7 Dec 2018 16:32:34 +0000 (16:32 +0000)]
[rs6000] mmintrin.h: fix use of "vector"

A recent patch inadvertently added the use of "vector" to mmintrin.h
when all such uses should be "__vector".

[gcc]

2018-12-07  Paul A. Clarke  <pc@us.ibm.com>

PR target/88408
* config/rs6000/mmintrin.h (_mm_packs_pu16): Correctly use "__vector".

From-SVN: r266895

5 years agore PR rtl-optimization/88349 ([MIPS] Redundant store instructions generated start...
Vladimir Makarov [Fri, 7 Dec 2018 16:08:17 +0000 (16:08 +0000)]
re PR rtl-optimization/88349 ([MIPS]  Redundant store instructions generated start with r266385)

2018-12-07  Vladimir Makarov  <vmakarov@redhat.com>

PR rtl-optimization/88349
* ira-costs.c (record_operand_costs): Check bigger reg class on
NO_REGS.

2018-12-07  Vladimir Makarov  <vmakarov@redhat.com>

PR rtl-optimization/88349
* gcc.target/mips/pr88349.c: New.

From-SVN: r266894

5 years agore PR c++/86669 (Complete object constructor clone omits length for a c++11 braced...
Jakub Jelinek [Fri, 7 Dec 2018 15:20:04 +0000 (16:20 +0100)]
re PR c++/86669 (Complete object constructor clone omits length for a c++11 braced initialiser)

PR c++/86669
* call.c (make_temporary_var_for_ref_to_temp): Call pushdecl even for
automatic vars.

* g++.dg/cpp0x/initlist105.C: New test.
* g++.dg/cpp0x/initlist106.C: New test.
* g++.dg/other/pr86669.C: New test.

From-SVN: r266893

5 years ago[AArch64][SVE] Remove unnecessary PTRUEs from integer arithmetic
Richard Sandiford [Fri, 7 Dec 2018 15:03:15 +0000 (15:03 +0000)]
[AArch64][SVE] Remove unnecessary PTRUEs from integer arithmetic

When using the unpredicated immediate forms of MUL, LSL, LSR and ASR,
the rtl patterns would still have the predicate operand we created for
the other forms.  This patch splits the patterns after reload in order
to get rid of the predicate, like we already do for WHILE.

2018-12-07  Richard Sandiford  <richard.sandiford@arm.com>

gcc/
* config/aarch64/aarch64-sve.md (*mul<mode>3, *v<optab><mode>3):
Split the patterns after reload if we don't need the predicate
operand.
(*post_ra_mul<mode>3, *post_ra_v<optab><mode>3): New patterns.

gcc/testsuite/
* gcc.target/aarch64/sve/pred_elim_2.c: New test.

From-SVN: r266892

5 years ago[AArch64][SVE] Remove unnecessary PTRUEs from FP arithmetic
Richard Sandiford [Fri, 7 Dec 2018 15:01:47 +0000 (15:01 +0000)]
[AArch64][SVE] Remove unnecessary PTRUEs from FP arithmetic

When using the unpredicated all-register forms of FADD, FSUB and FMUL,
the rtl patterns would still have the predicate operand we created for
the other forms.  This patch splits the patterns after reload in order
to get rid of the predicate, like we already do for WHILE.

2018-12-07  Richard Sandiford  <richard.sandiford@arm.com>

gcc/
* config/aarch64/iterators.md (SVE_UNPRED_FP_BINARY): New code
iterator.
(sve_fp_op): Handle minus and mult.
* config/aarch64/aarch64-sve.md (*add<mode>3, *sub<mode>3)
(*mul<mode>3): Split the patterns after reload if we don't
need the predicate operand.
(*post_ra_<sve_fp_op><mode>3): New pattern.

gcc/testsuite/
* gcc.target/aarch64/sve/pred_elim_1.c: New test.

From-SVN: r266891

5 years agoruntime: add missing return for non-GNU/Linux version of tgkill
Ian Lance Taylor [Fri, 7 Dec 2018 14:22:55 +0000 (14:22 +0000)]
runtime: add missing return for non-GNU/Linux version of tgkill

    Path from Rainer Orth.

    Reviewed-on: https://go-review.googlesource.com/c/153118

From-SVN: r266890

5 years agoAdd forgotten PR marker.
Rainer Orth [Fri, 7 Dec 2018 13:56:17 +0000 (13:56 +0000)]
Add forgotten PR marker.

From-SVN: r266889

5 years agoBuild gcc.target/i386/ipa-stack-alignment-2.c with -fomit-frame-pointer
Rainer Orth [Fri, 7 Dec 2018 13:01:20 +0000 (13:01 +0000)]
Build gcc.target/i386/ipa-stack-alignment-2.c with -fomit-frame-pointer

* gcc.target/i386/ipa-stack-alignment-2.c: Add
-fomit-frame-pointer to dg-options.

From-SVN: r266888

5 years agoAdd a recursion limit to libiberty's demangling code. The limit is enabled by defaul...
Nick Clifton [Fri, 7 Dec 2018 10:33:30 +0000 (10:33 +0000)]
Add a recursion limit to libiberty's demangling code.  The limit is enabled by default, but can be disabled via a new demangling option.

include * demangle.h (DMGL_NO_RECURSE_LIMIT): Define.
        (DEMANGLE_RECURSION_LIMIT): Define

PR 87681
PR 87675
PR 87636
PR 87350
PR 87335
libiberty * cp-demangle.h (struct d_info): Add recursion_level field.
* cp-demangle.c (d_function_type): Add recursion counter.
If the recursion limit is reached and the check is not disabled,
then return with a failure result.
(cplus_demangle_init_info): Initialise the recursion_level field.
        (d_demangle_callback): If the recursion limit is enabled, check
for a mangled string that is so long that there is not enough
stack space for the local arrays.
        * cplus-dem.c (struct work): Add recursion_level field.
(squangle_mop_up): Set the numb and numk fields to zero.
(work_stuff_copy_to_from): Handle the case where a btypevec or
ktypevec field is NULL.
(demangle_nested_args): Add recursion counter.  If
the recursion limit is not disabled and reached, return with a
failure result.

From-SVN: r266886

5 years agoprofile-count.h (profile_count::oeprator>=): Fix typo by inverting return condition...
Bin Cheng [Fri, 7 Dec 2018 09:49:13 +0000 (09:49 +0000)]
profile-count.h (profile_count::oeprator>=): Fix typo by inverting return condition when *this is precise zero.

* profile-count.h (profile_count::oeprator>=): Fix typo by inverting
return condition when *this is precise zero.

From-SVN: r266885

5 years agoDaily bump.
GCC Administrator [Fri, 7 Dec 2018 00:16:25 +0000 (00:16 +0000)]
Daily bump.

From-SVN: r266884

5 years agore PR target/85593 (GCC on ARM allocates R3 for local variable when calling naked...
Jakub Jelinek [Thu, 6 Dec 2018 23:41:04 +0000 (00:41 +0100)]
re PR target/85593 (GCC on ARM allocates R3 for local variable when calling naked function with O2 optimizations enabled)

PR target/85593
* final.c (rest_of_handle_final): Don't call collect_fn_hard_reg_usage
for functions with naked attribute.

* gcc.target/i386/pr85593.c: New test.

From-SVN: r266881

5 years agore PR rtl-optimization/85770 (ICE: in lra_eliminate, at lra-eliminations.c:1439 with...
Jakub Jelinek [Thu, 6 Dec 2018 23:39:12 +0000 (00:39 +0100)]
re PR rtl-optimization/85770 (ICE: in lra_eliminate, at lra-eliminations.c:1439 with -march=nano-1000)

PR rtl-optimization/85770
* gcc.target/i386/pr85770.c: Require int128 effective target.

From-SVN: r266880

5 years agore PR fortran/88377 (ICE in gfc_omp_clause_copy_ctor, at fortran/trans-openmp.c:614)
Jakub Jelinek [Thu, 6 Dec 2018 23:29:04 +0000 (00:29 +0100)]
re PR fortran/88377 (ICE in gfc_omp_clause_copy_ctor, at fortran/trans-openmp.c:614)

PR fortran/88377
* trans-openmp.c (gfc_omp_clause_default_ctor,
gfc_omp_clause_copy_ctor, gfc_omp_clause_assign_op,
gfc_omp_clause_linear_ctor, gfc_omp_clause_dtor): Only consider
GFC_DECL_GET_SCALAR_ALLOCATABLE vars as scalar allocatables if they
have pointer type.

* gfortran.dg/gomp/pr88377.f90: New test.

From-SVN: r266879

5 years agore PR tree-optimization/88367 (-fno-delete-null-pointer-checks doesn't work properly)
Jakub Jelinek [Thu, 6 Dec 2018 23:28:04 +0000 (00:28 +0100)]
re PR tree-optimization/88367 (-fno-delete-null-pointer-checks doesn't work properly)

PR c/88367
* tree-vrp.c (extract_range_from_binary_expr): For POINTER_PLUS_EXPR
with -fno-delete-null-pointer-checks, set_nonnull only if the pointer
is non-NULL and offset is known to have most significant bit clear.
* vr-values.c (vr_values::vrp_stmt_computes_nonzero): For ADDR_EXPR
of MEM_EXPR, return true if the MEM_EXPR has non-zero offset with
most significant bit clear.  If offset does have most significant bit
set and -fno-delete-null-pointer-checks, don't return true even if
the base pointer is non-NULL.

* gcc.dg/tree-ssa/pr88367.c: New test.

From-SVN: r266878

5 years agore PR c++/87506 (ICE with inherited constexpr constructor with const argument)
Jakub Jelinek [Thu, 6 Dec 2018 23:25:10 +0000 (00:25 +0100)]
re PR c++/87506 (ICE with inherited constexpr constructor with const argument)

PR c++/87506
* constexpr.c (adjust_temp_type): Handle EMPTY_CLASS_EXPR.

* g++.dg/cpp0x/constexpr-87506.C: New test.

From-SVN: r266877

5 years agoclass.c (check_bitfield_decl): In error message about non-integral type print the...
Paolo Carlini [Thu, 6 Dec 2018 23:20:16 +0000 (23:20 +0000)]
class.c (check_bitfield_decl): In error message about non-integral type print the type itself too.

/cp
2018-12-06  Paolo Carlini  <paolo.carlini@oracle.com>

* class.c (check_bitfield_decl): In error message about non-integral
type print the type itself too.
* decl.c (grokdeclarator): Do not ICE on unnamed bit-fields declared
friends; when calling build_decl for a FIELD_DECL possibly pass the
declarator->id_loc.

/testsuite
2018-12-06  Paolo Carlini  <paolo.carlini@oracle.com>

* g++.dg/parse/bitfield7.C: New.
* g++.dg/other/bitfield2.C: Check location and type.
* g++.dg/parse/bitfield1.C: Likewise.
* g++.dg/parse/bitfield2.C: Likewise.

From-SVN: r266876

5 years ago[PR86747] tsubst friend tpl ctxt before looking it up for dupes
Alexandre Oliva [Thu, 6 Dec 2018 23:18:40 +0000 (23:18 +0000)]
[PR86747] tsubst friend tpl ctxt before looking it up for dupes

When a member template is redeclared as a friend, we enter the context
of the member before looking it up, and then we check that the decls
are compatible.  However, when the member template references template
types of the enclosing context, say an enclosing template class, the
compare fails because the friend decl is already tsubsted, whereas the
looked up name isn't.

The problem is that the enclosing context is taken from the friend
declaration before tsubsting it, so we look up in the context of the
generic template instead of that of the tsubsted one we're
specializing.  The solution is to tsubst the enclosing context when
it's a non-namespace scope.

for  gcc/cp/ChangeLog

PR c++/86747
* pt.c (tsubst_friend_class): Enter tsubsted class context.

for  gcc/testsuite/ChangeLog

PR c++/86747
* g++.dg/pr86747.C: New.

From-SVN: r266875

5 years ago[PR86397] resolve nondependent noexcept specs early in C++1[14]
Alexandre Oliva [Thu, 6 Dec 2018 23:18:30 +0000 (23:18 +0000)]
[PR86397] resolve nondependent noexcept specs early in C++1[14]

build_noexcept_spec refrained from resolving nondependent noexcept
expressions when they were not part of the function types (C++ 11 and
14).  This caused problems during mangling: canonical_eh_spec, when
called on the template function type, would find an unresolved but not
explicitly deferred expression, and nothrow_spec_p would reject it.

We could relax the mangling logic to skip canonical_eh_spec, but since
-Wnoexcept-type warns when mangling function names that change as
noexcept specs become part of types and of mangling in C++17, and the
test at mangling time may give incorrect results if the spec is not
resolved, we might as well keep things simple and resolve nondependent
noexcept specs sooner rather than later.  This is what this patch does.

for  gcc/cp/ChangeLog

PR c++/86397
* except.c (build_noexcept_spec): Resolve nondependent
expressions.

for gcc/testsuite/ChangeLog

PR c++/86397
* g++.dg/cpp0x/pr86397-1.C: New.
* g++.dg/cpp0x/pr86397-2.C: New.

From-SVN: r266874

5 years agoavoid null ptr deref in cselib_record_sets
Alexandre Oliva [Thu, 6 Dec 2018 23:18:19 +0000 (23:18 +0000)]
avoid null ptr deref in cselib_record_sets

Jeff Law tells me h8300-elf fails gcc.c-torture/compile/pr49029.c
with -O2 -g -mint32 -mh.  This patch fixes it.

The problem is that strict low part handling in cselib_record_sets
assumes src_elt is not NULL.  That src_elt is taken from a strict low
part set, but it won't always have a src_elt to begin with.  In this
case, it's because src is a volatile MEM; we don't record values for
those.

Although we could fix the problem by testing for a NULL src_elt before
creating the zero extends corresponding to strict low part sets of
formerly const0_rtx REGs, there's no point in recording the additional
set that we won't be able to use anyway.

We could still record that the whole register has a zero-extend of
the value stored in the narrower-mode strict low part of the register,
but is that of any use?  I guess not, but if we find otherwise, we can
change that later.

for  gcc/ChangeLog

* cselib.c (cselib_record_sets): Skip strict low part sets
with NULL src_elt.

From-SVN: r266873

5 years ago[rs6000] Enable x86-compat vector intrinsics testing
Paul A. Clarke [Thu, 6 Dec 2018 22:14:55 +0000 (22:14 +0000)]
[rs6000] Enable x86-compat vector intrinsics testing

The testsuite tests for the compatibility implementations of x86 vector
intrinsics for "powerpc" had been inadvertently made to PASS
without actually running the test code.

This patch removes the code which kept the tests from running the actual
test code.

2018-12-06  Paul A. Clarke  <pc@us.ibm.com>

[gcc/testsuite]

PR target/88316
* gcc.target/powerpc/bmi-check.h: Remove test for
__BUILTIN_CPU_SUPPORTS__, thereby enabling test code to run.
* gcc.target/powerpc/bmi2-check.h: Likewise.
* gcc.target/powerpc/mmx-check.h: Likewise.
* gcc.target/powerpc/sse-check.h: Likewise.
* gcc.target/powerpc/sse2-check.h: Likewise.
* gcc.target/powerpc/sse3-check.h: Likewise.

From-SVN: r266870

5 years ago[rs6000] Fix x86-compat vector intrinsics testcases for BE, 32bit
Paul A. Clarke [Thu, 6 Dec 2018 22:11:01 +0000 (22:11 +0000)]
[rs6000] Fix x86-compat vector intrinsics testcases for BE, 32bit

Fix general endian issues found in the test cases for thecompatibility
implementations of the x86 vector intrinsics.  (The tests had been
inadvertently made to PASS without actually running the test code.
A later patch fixes this issue.)

Additionally, a new <smmintrin.h> is added, as some of the APIs therein are
now used by the test cases.  It is _not_ a complete implementation of the
SSE4 interfaces, only the few "extract" interfaces uses by the tests.

2018-12-06  Paul A. Clarke  <pc@us.ibm.com>

[gcc]

PR target/88316
* config/rs6000/smmintrin.h: New file.
* config.gcc: Add smmintrin.h to extra_headers for powerpc*-*-*.

[gcc/testsuite]

PR target/88316
* gcc.target/powerpc/mmx-packssdw-1.c: Fixes for big-endian.
* gcc.target/powerpc/mmx-packsswb-1.c: Likewise.
* gcc.target/powerpc/mmx-packuswb-1.c: Likewise.
* gcc.target/powerpc/mmx-pmulhw-1.c: Likewise.
* gcc.target/powerpc/sse-cvtpi32x2ps-1.c: Likewise.
* gcc.target/powerpc/sse-cvtpu16ps-1.c: Likewise.
* gcc.target/powerpc/sse-cvtss2si-1.c: Likewise.
* gcc.target/powerpc/sse-cvtss2si-2.c: Likewise.
* gcc.target/powerpc/sse2-pshufhw-1.c: Likewise.
* gcc.target/powerpc/sse2-pshuflw-1.c: Likewise.

From-SVN: r266869

5 years ago[rs6000] x86-compat vector intrinsics fixes for BE, 32bit
Paul A. Clarke [Thu, 6 Dec 2018 22:03:25 +0000 (22:03 +0000)]
[rs6000] x86-compat vector intrinsics fixes for BE, 32bit

Fix general endian and 32-bit mode issues found in the
compatibility implementations of the x86 vector intrinsics when running the
associated test suite tests.  (The tests had been inadvertently made to PASS
without actually running the test code.  A later patch fixes this issue.)

2018-12-03  Paul A. Clarke  <pc@us.ibm.com>

PR target/88316
* config/rs6000/mmintrin.h (_mm_unpackhi_pi8): Fix for big-endian.
(_mm_unpacklo_pi8): Likewise.
(_mm_mulhi_pi16): Likewise.
(_mm_packs_pi16): Fix for big-endian. Use preferred API.
(_mm_packs_pi32): Likewise.
(_mm_packs_pu16): Likewise.
* config/rs6000/xmmintrin.h (_mm_cvtss_si32): Fix for big-endian.
(_mm_cvtss_si64): Likewise.
(_mm_cvtpi32x2_ps): Likewise.
(_mm_shuffle_ps): Likewise.
(_mm_movemask_pi8): Likewise.
(_mm_mulhi_pu16): Likewise.
(_mm_sad_pu8): Likewise.
(_mm_sad_pu8): Likewise.
(_mm_cvtpu16_ps): Fix for big-endian. Use preferred API.
(_mm_cvtpu8_ps): Likewise.
(_mm_movemask_ps): Better #else case for big-endian (no functional
change).
(_mm_shuffle_pi16): Likewise.
* config/rs6000/emmintrin.h (_mm_movemask_pd): Fix for big-endian.
Better #else case for big-endian (no functional change).
(_mm_movemask_epi8): Likewise.
(_mm_shufflehi_epi16): Likewise.
(_mm_shufflelo_epi16): Likewise.
(_mm_shuffle_epi32): Likewise.
(_mm_mul_epu32): Fix for big-endian.
(_mm_bsrli_si128): Likewise.
(_mm_cvtps_pd): Better #else case for big endian.
(_mm_mulhi_epi16): Likewise.
(_mm_mul_epu32): Likewise.
(_mm_slli_si128): Likewise.
(_mm_sll_epi16): Likewise.
(_mm_sll_epi32): Likewise.
(_mm_sra_epi16): Likewise.
(_mm_sra_epi32): Likewise.
(_mm_srl_epi16): Likewise.
(_mm_srl_epi32): Likewise.
(_mm_mulhi_epu16): Likewise.
(_mm_sad_epu8): Likewise.
* config/rs6000/pmmintrin.h (_mm_hadd_ps): Fix for big-endian.
(_mm_sub_ps): Likewise.
* config/rs6000/mmintrin.h (_mm_cmpeq_pi8): Fix for 32-bit mode.
* gcc/config/rs6000/tmmintrin.h (_mm_alignr_epi8): Use ENDIAN
macros consistently (no functional changes).
(_mm_alignr_pi8): Likewise.

From-SVN: r266868

5 years agoPR c++/88136 - -Wdeprecated-copy false positives
Jason Merrill [Thu, 6 Dec 2018 21:17:08 +0000 (16:17 -0500)]
PR c++/88136 - -Wdeprecated-copy false positives

Deprecating the copy operations because the class has a user-provided
destructor turns out to have too many false positives; this patch adjusts
-Wdeprecated-copy to only deprecate if the other copy operation is
user-provided.  To get the earlier behavior, people can explicitly request
it with -Wdeprecated-copy-dtor.

gcc/c-family/
* c.opt (Wdeprecated-copy-dtor): New.
(Wdeprecated-copy): Move to -Wextra.
gcc/cp/
* class.c (classtype_has_depr_implicit_copy): Rename from
classtype_has_user_copy_or_dtor.
* method.c (lazily_declare_fn): Adjust.
* decl2.c (cp_warn_deprecated_use): Refer to -Wdeprecated-copy-dtor
if deprecation is due to a destructor.

From-SVN: r266867

5 years agoDarwin - fix PR c++/87380
Iain Sandoe [Thu, 6 Dec 2018 20:09:26 +0000 (20:09 +0000)]
Darwin - fix PR c++/87380

This is [intentionally] broken C++ ABI, that was catering for a
tool problem that existed in a very old Darwin toolchain.

The issue is not present after Darwin7 (using default system
tools) so confine the fix to that revision or earlier.

2018-12-06  Iain Sandoe  <iain@sandoe.co.uk>

PR c++/87380
* config/darwin.h (TARGET_WEAK_NOT_IN_ARCHIVE_TOC) Remove; use the
default.
* config/rs6000/darwin7.h (TARGET_WEAK_NOT_IN_ARCHIVE_TOC): New.

From-SVN: r266866

5 years agoPR c++/88373 - wrong parse error with ~.
Marek Polacek [Thu, 6 Dec 2018 19:43:17 +0000 (19:43 +0000)]
PR c++/88373 - wrong parse error with ~.

* parser.c (cp_parser_template_name): Check tag_type for
none_type.

* g++.dg/cpp2a/fn-template19.C: New test.

From-SVN: r266865

5 years agoDarwin fix libstdc++ build warnings.
Iain Sandoe [Thu, 6 Dec 2018 19:42:21 +0000 (19:42 +0000)]
Darwin fix libstdc++ build warnings.

GCC does not export construction vtable symbols from shared libraries.

The symbols are marked hidden in the objects; for Darwin that makes
them also external (“private_extern” is Darwin’s hidden) which means
that they show up in the list of possible symbols for export from
libstdc++, and there are sufficiently relaxed match conditions that
they reach the exports list.   When Darwin’s static linker encounters
them it generates a warning that they cannot be exported.  This patch
prunes them from the list of symbols to be considered, thus
eliminating the warnings. No functional Change inended to the library
exports.

2018-12-06  Iain Sandoe  <iain@sandoe.co.uk>

* scripts/make_exports.pl (check names): Don’t try to export
construction vtable symbols.

From-SVN: r266864

5 years agoFix PR libstdc++/64883 Darwin headers use always_inline so don't test that
Jonathan Wakely [Thu, 6 Dec 2018 19:21:32 +0000 (19:21 +0000)]
Fix PR libstdc++/64883 Darwin headers use always_inline so don't test that

    Because darwin system headers use always_inline rather than
    __always_inline__ the libstdc++ test will fail, even if our headers only
    use the reserved form of the attribute. Don't test it on Darwin, and
    assume that testing on other targets will catch any accidental misuses
    in libstdc++ headers.

2018-12-06  Jonathan Wakely  <jwakely@redhat.com>
    Iain Sandoe  <iain@sandoe.co.uk>

            PR libstdc++/64883
            * testsuite/17_intro/headers/c++1998/all_attributes.cc: Don't test
            always_inline on Darwin.
            * testsuite/17_intro/headers/c++2011/all_attributes.cc: Likewise.
            * testsuite/17_intro/headers/c++2014/all_attributes.cc: Likewise.
            * testsuite/17_intro/headers/c++2017/all_attributes.cc: Likewise.
            * testsuite/17_intro/headers/c++2020/all_attributes.cc: Likewise.

Co-Authored-By: Iain Sandoe <iain@sandoe.co.uk>
From-SVN: r266863

5 years agore PR target/88282 (ICE in df_install_refs at gcc/df-scan.c:2379)
Vladimir Makarov [Thu, 6 Dec 2018 18:41:46 +0000 (18:41 +0000)]
re PR target/88282 (ICE in df_install_refs at gcc/df-scan.c:2379)

2018-12-06  Vladimir Makarov  <vmakarov@redhat.com>

PR target/88282
* ira.c (ira_init_register_move_cost): Use info from
hard_regno_mode_ok instead of contains_reg_of_mode.
* ira-costs.c (contains_reg_of_mode): Don't use cost from bigger
hard register class for some fixed hard registers.

From-SVN: r266862

5 years agoRemove bogus test line from vect-over-widen-23.c
Richard Sandiford [Thu, 6 Dec 2018 18:02:09 +0000 (18:02 +0000)]
Remove bogus test line from vect-over-widen-23.c

I'd cut-&-paste vect-over-widen-23.c in an attempt to get the target
reuirements right, but by doing so carried over an unwanted test for
shifts.

2018-12-06  Richard Sandiford  <richard.sandiford@arm.com>

gcc/testsuite/
* gcc.dg/vect/vect-over-widen-23.c: Remove unwanted line.

From-SVN: r266861

5 years agoasm inline
Segher Boessenkool [Thu, 6 Dec 2018 17:56:58 +0000 (18:56 +0100)]
asm inline

The Linux kernel people want a feature that makes GCC pretend some
inline assembler code is tiny (while it would think it is huge), so
that such code will be inlined essentially always instead of
essentially never.

This patch lets you say "asm inline" instead of just "asm", with the
result that that inline assembler is always counted as minimum cost
for inlining.  It implements this for C and C++, making "inline"
another asm-qualifier (supplementing "volatile" and "goto").

* doc/extend.texi (Using Assembly Language with C): Document asm inline.
(Size of an asm): Fix typo.  Document asm inline.
* gimple-pretty-print.c (dump_gimple_asm): Handle asm inline.
* gimple.h (enum gf_mask): Add GF_ASM_INLINE.
(gimple_asm_set_volatile): Fix typo.
(gimple_asm_inline_p): New.
(gimple_asm_set_inline): New.
* gimplify.c (gimplify_asm_expr): Propagate the asm inline flag from
tree to gimple.
* ipa-icf-gimple.c (func_checker::compare_gimple_asm): Compare the
gimple_asm_inline_p flag, too.
* tree-core.h (tree_base): Document that protected_flag is ASM_INLINE_P
in an ASM_EXPR.
* tree-inline.c (estimate_num_insns): If gimple_asm_inline_p return
a minimum size for an asm.
* tree.h (ASM_INLINE_P): New.

gcc/c/
* c-parser.c (c_parser_asm_statement): Detect the inline keyword
after asm.  Pass a flag for it to build_asm_expr.
* c-tree.h (build_asm_expr): Update declaration.
* c-typeck.c (build_asm_stmt): Add is_inline parameter.  Use it to
set ASM_INLINE_P.

gcc/cp/
* cp-tree.h (finish_asm_stmt): Update declaration.
* parser.c (cp_parser_asm_definition): Detect the inline keyword
after asm.  Pass a flag for it to finish_asm_stmt.
* pt.c (tsubst_expr): Pass the ASM_INLINE_P flag to finish_asm_stmt.
* semantics.c (finish_asm_stmt): Add inline_p parameter.  Use it to
set ASM_INLINE_P.

gcc/testsuite/
* c-c++-common/torture/asm-inline.c: New testcase.
* gcc.dg/asm-qual-2.c: Test asm inline, too.

From-SVN: r266860

5 years agoasm qualifiers (PR55681)
Segher Boessenkool [Thu, 6 Dec 2018 17:47:52 +0000 (18:47 +0100)]
asm qualifiers (PR55681)

PR55681 observes that currently only one qualifier is allowed for
inline asm, so that e.g. "volatile asm" is allowed, "const asm" is also
okay (with a warning), but "const volatile asm" gives an error.  Also
"goto" has to be last.

This patch changes things so that only "asm-qualifiers" are allowed,
that is "volatile" and "goto", in any combination, in any order, but
without repetitions.

PR inline-asm/55681
* doc/extend.texi (Basic Asm): Update grammar.
(Extended Asm): Update grammar.

gcc/c/
PR inline-asm/55681
* c-parser.c (c_parser_asm_statement): Update grammar.  Allow any
combination of volatile and goto, in any order, without repetitions.

gcc/cp/
PR inline-asm/55681
* parser.c (cp_parser_asm_definition): Update grammar.  Allow any
combination of volatile and goto, in any order, without repetitions.

gcc/testsuite/
PR inline-asm/55681
* gcc.dg/asm-qual-1.c: Test that "const" and "restrict" are refused.
* gcc.dg/asm-qual-2.c: New test, test that asm-qualifiers are allowed
in any order, but that duplicates are not allowed.

From-SVN: r266859

5 years agore PR testsuite/86540 (pr77445-2.c and ssa-dom-thread-7.c regressions since May 20...
Jeff Law [Thu, 6 Dec 2018 15:35:34 +0000 (08:35 -0700)]
re PR testsuite/86540 (pr77445-2.c and ssa-dom-thread-7.c regressions since May 20, 2018)

PR testsuite/86540
* gcc.dg/tree-ssa/ssa-dom-thread-7.c: Skip the post switch conversion
tests on aarch64.
* gcc.dg/tree-ssa/pr77445-2.c: Similarly.

From-SVN: r266857

5 years agoC++: improvements to diagnostics using %P (more PR c++/85110)
David Malcolm [Thu, 6 Dec 2018 15:25:12 +0000 (15:25 +0000)]
C++: improvements to diagnostics using %P (more PR c++/85110)

This patch is based on grepping the C++ frontend for %P
i.e. diagnostics that refer to a parameter number.  It fixes up
these diagnostics to highlight the pertinent param where appropriate
(and possible), along with various other tweaks, as described in the
ChangeLog.

gcc/cp/ChangeLog:
PR c++/85110
* call.c (conversion_null_warnings): Try to use the location of
the expression for the warnings.  Add notes showing the parameter
of the function decl, where available.
(get_fndecl_argument_location): For implicitly-declared functions,
use the fndecl location rather than that of the param.
(maybe_inform_about_fndecl_for_bogus_argument_init): New function.
(convert_like_real): Use it in various places to avoid repetition.
(complain_about_bad_argument): Likewise.
* cp-tree.h (maybe_inform_about_fndecl_for_bogus_argument_init):
New declaration.
* decl2.c (check_default_args): Put all diagnostics for the fndecl
into a diagnostic group.  Use the location of the parameter when
complaining about parameters with missing default arguments in
preference to that of the fndecl.  Attempt to record the location
of the first parameter with a default argument and emit a note
for the first parameter that's missing one.
* typeck.c (convert_arguments): When complaining about parameters
with incomplete types, attempt to use the location of the
argument.  Where available, add a note showing the pertinent
parameter in the fndecl.
(convert_for_assignment): When complaining about bad conversions
at function calls, use the location of the unstripped argument.
(convert_for_initialization): When checking for bogus references,
add an auto_diagnostic_group, and update the note to use the
location of the pertinent parameter, rather than just the callee.

gcc/testsuite/ChangeLog:
PR c++/85110
* g++.dg/diagnostic/missing-default-args.C: New test.
* g++.dg/diagnostic/param-type-mismatch-3.C: New test.
* g++.dg/diagnostic/param-type-mismatch.C: Add tests for invalid
references and incomplete types.
* g++.dg/warn/Wconversion-null-4.C: New test.

From-SVN: r266856

5 years agoDrop duplicate half of doubly-applied test.
Jeff Law [Thu, 6 Dec 2018 13:46:25 +0000 (06:46 -0700)]
Drop duplicate half of doubly-applied test.

From-SVN: r266854

5 years agoFix for PR78444 by ensuring 128b alignment at call sites.
Iain Sandoe [Thu, 6 Dec 2018 13:36:35 +0000 (13:36 +0000)]
Fix for PR78444 by ensuring 128b alignment at call sites.

2018-12-06  Iain Sandoe  <iain@sandoe.co.uk>

PR target/78444
* config/i386/darwin.h (STACK_BOUNDARY): Remove macro.
* config/i386/i386.c (ix86_compute_frame_layout): Ensure at least 128b
stack alignment in non-leaf functions.

From-SVN: r266853

5 years agore PR target/87598 (Rejects "%a0" with constant)
Jakub Jelinek [Thu, 6 Dec 2018 12:20:09 +0000 (13:20 +0100)]
re PR target/87598 (Rejects "%a0" with constant)

PR target/87598
* config/aarch64/aarch64.c (aarch64_print_address_internal): Don't
call output_operand_lossage on VOIDmode CONST_INTs.  After
output_operand_lossage do return false.

* gcc.target/aarch64/asm-5.c: New test.

From-SVN: r266852

5 years agodf-problems.c (df_rd_local_compute): Use bitmap_release.
Richard Biener [Thu, 6 Dec 2018 11:32:52 +0000 (11:32 +0000)]
df-problems.c (df_rd_local_compute): Use bitmap_release.

2018-12-06  Richard Biener  <rguenther@suse.de>

* df-problems.c (df_rd_local_compute): Use bitmap_release.
(df_live_free): Likewise.
(df_md_local_compute): Likewise.
(df_md_free): Release df_md_scratch bitmap.
* loop-invariant.c (calculate_loop_reg_pressure): Use
bitmap_release.
* sched-deps.c (true_dependency_cache, output_dependency_cache,
anti_dependency_cache, control_dependency_cache,
spec_dependency_cache): Use bitmap instead of bitmap_head *.
* sched-ebb.c (schedule_ebbs_init): Initialize non-GTY
dont_calc_deps as bitmap allocated from obstack not GC.
(schedule_ebbs_finish): Use bitmap_release.
* sched-rgn.c (schedule_insns): Initialize non-GTY
not_in_df as bitmap allocated from obstack not GC.
Use bitmap_release.
* sel-sched.c (_forced_ebb_heads): Remove premature optimization.
(sel_region_init): Allocate forced_ebb_heads.
(sel_region_finish): Free forced_ebb_heads.

From-SVN: r266851

5 years agobitmap.c (bitmap_head::crashme): Define.
Richard Biener [Thu, 6 Dec 2018 11:32:09 +0000 (11:32 +0000)]
bitmap.c (bitmap_head::crashme): Define.

2018-12-06  Richard Biener  <rguenther@suse.de>

* bitmap.c (bitmap_head::crashme): Define.
* bitmap.h (bitmap_head): Add constexpr default constructor
poisoning the obstack member.
(bitmap_head::crashme): Declare.
(bitmap_release): New function clearing a bitmap and poisoning
the obstack member.
* gengtype.c (main): Make it recognize CONSTEXPR.

* lra-constraints.c (lra_inheritance): Use bitmap_release
instead of bitmap_clear.

* ira.c (ira): Work around class-memaccess warning.
* regrename.c (create_new_chain): Likewise.

From-SVN: r266850

5 years agore PR tree-optimization/85726 (div C1 to div C2 match.pd suboptimization)
Jakub Jelinek [Thu, 6 Dec 2018 10:45:47 +0000 (11:45 +0100)]
re PR tree-optimization/85726 (div C1 to div C2 match.pd suboptimization)

PR tree-optimization/85726
* generic-match-head.c (optimize_successive_divisions_p): New function.
* gimple-match-head.c (optimize_successive_divisions_p): Likewise.
* match.pd: Don't combine successive divisions if they aren't exact
and optimize_successive_divisions_p is false.

* gcc.dg/tree-ssa/pr85726-1.c: New test.
* gcc.dg/tree-ssa/pr85726-2.c: New test.
* gcc.dg/tree-ssa/pr85726-3.c: New test.
* gcc.dg/tree-ssa/pr85726-4.c: New test.

From-SVN: r266848

5 years agore PR fortran/88304 (ICE in use_pointer_in_frame, at tree-nested.c:267)
Jakub Jelinek [Thu, 6 Dec 2018 10:28:31 +0000 (11:28 +0100)]
re PR fortran/88304 (ICE in use_pointer_in_frame, at tree-nested.c:267)

PR fortran/88304
* tree-nested.c (convert_nonlocal_reference_stmt): Remove clobbers
for non-local automatic decls.

* gfortran.fortran-torture/compile/pr88304.f90: New test.

From-SVN: r266847

5 years agore PR middle-end/63184 (Fails to simplify comparison)
Richard Biener [Thu, 6 Dec 2018 10:20:39 +0000 (10:20 +0000)]
re PR middle-end/63184 (Fails to simplify comparison)

2018-12-06  Richard Biener  <rguenther@suse.de>

PR middle-end/63184
* c-c++-common/pr19807-2.c: Try link only on x86, add xfailed
optimized dump scanning.
* c-c++-common/pr19807-3.c: Likewise.

From-SVN: r266846

5 years agodecl2.c (grokbitfield): Use DECL_SOURCE_LOCATION in error message...
Paolo Carlini [Thu, 6 Dec 2018 08:23:04 +0000 (08:23 +0000)]
decl2.c (grokbitfield): Use DECL_SOURCE_LOCATION in error message...

/cp
2018-12-06  Paolo Carlini  <paolo.carlini@oracle.com>

* decl2.c (grokbitfield): Use DECL_SOURCE_LOCATION in error message;
print the type too; fix thinko in INDIRECT_TYPE_P use.

/testsuite
2018-12-06  Paolo Carlini  <paolo.carlini@oracle.com>

* g++.dg/parse/bitfield6b.C: New.
* g++.dg/parse/bitfield3.C: Test location and type.

From-SVN: r266841

5 years agotestsuite - handle the -fno-diagnostics-show-line-numbers in compat tests.
Iain Sandoe [Thu, 6 Dec 2018 08:14:57 +0000 (08:14 +0000)]
testsuite - handle the -fno-diagnostics-show-line-numbers in compat tests.

2018-12-06  Iain Sandoe  <iain@sandoe.co.uk>

* lib/c-compat.exp (compat-use-alt-compiler):
Handle -fno-diagnostics-show-line-numbers.
(compat_setup_dfp): Likewise.

From-SVN: r266840

5 years agore PR rtl-optimization/85770 (ICE: in lra_eliminate, at lra-eliminations.c:1439 with...
Jeff Law [Thu, 6 Dec 2018 00:40:08 +0000 (17:40 -0700)]
re PR rtl-optimization/85770 (ICE: in lra_eliminate, at lra-eliminations.c:1439 with -march=nano-1000)

PR rtl-optimization/85770
* gcc.target/i386/pr85770.c: New test.

From-SVN: r266839

5 years agoaix72.h (ASM_DEFAULT_SPEC): Match Power7 processor default.
David Edelsohn [Thu, 6 Dec 2018 00:36:52 +0000 (00:36 +0000)]
aix72.h (ASM_DEFAULT_SPEC): Match Power7 processor default.

        * config/rs6000/aix72.h (ASM_DEFAULT_SPEC): Match Power7 processor
        default.

From-SVN: r266838

5 years agoDaily bump.
GCC Administrator [Thu, 6 Dec 2018 00:16:36 +0000 (00:16 +0000)]
Daily bump.

From-SVN: r266837

5 years agodwarf2out.c (field_byte_offset): Remove #ifdef PCC_BITFIELD_TYPE_MATTERS check.
Uros Bizjak [Thu, 6 Dec 2018 00:14:41 +0000 (01:14 +0100)]
dwarf2out.c (field_byte_offset): Remove #ifdef PCC_BITFIELD_TYPE_MATTERS check.

* dwarf2out.c (field_byte_offset): Remove
#ifdef PCC_BITFIELD_TYPE_MATTERS check.

From-SVN: r266834

5 years agore PR c/87028 (false positive -Wstringop-truncation strncpy with global variable...
Jeff Law [Wed, 5 Dec 2018 23:10:08 +0000 (16:10 -0700)]
re PR c/87028 (false positive -Wstringop-truncation strncpy with global variable source string)

PR c/87028
* calls.c (get_attr_nonstring_decl): Avoid setting *REF to
SSA_NAME_VAR.
* gcc/gimple-low.c (lower_stmt): Fold builtin calls here.
* gimplify (maybe_fold_stmt): Avoid folding builtin calls.

PR c/87028
* c-c++-common/Wstringop-truncation.c: Remove xfails.
* gcc.dg/Wstringop-truncation-5.c: New test.
* gcc.dg/strcmpopt_1.c: Adjust.
* gcc.dg/tree-ssa/pr79697.c: Same.

From-SVN: r266833

5 years agoruntime: add precise stack scan support
Ian Lance Taylor [Wed, 5 Dec 2018 23:09:51 +0000 (23:09 +0000)]
runtime: add precise stack scan support

    This CL adds support of precise stack scan using stack maps to
    the runtime. The stack maps are generated by the compiler (if
    supported). Each safepoint is associated with a (real or dummy)
    landing pad, and its "type info" in the exception table is a
    pointer to the stack map. When a stack is scanned, the stack map
    is found by the stack unwinding code by inspecting the exception
    table (LSDA).

    For precise stack scan we need to unwind the stack. There are
    three cases:

    - If a goroutine is scanning its own stack, it can unwind the
      stack and scan the frames.

    - If a goroutine is scanning another, stopped, goroutine, it
      cannot directly unwind the target stack. We handle this by
      switching (runtime.gogo) to the target g, letting it unwind
      and scan the stack, and switch back.

    - If we are scanning a goroutine that is blocked in a syscall,
      we send a signal to the target goroutine's thread, and let the
      signal handler unwind and scan the stack. Extra care is needed
      as this races with enter/exit syscall.

    Currently this is only implemented on linux.

    Reviewed-on: https://go-review.googlesource.com/c/140518

From-SVN: r266832

5 years agodarwin - add configuration support for 'otool'
Iain Sandoe [Wed, 5 Dec 2018 21:57:00 +0000 (21:57 +0000)]
darwin - add configuration support for 'otool'

2018-12-05  Iain Sandoe  <iain@sandoe.co.uk>

* configure.ac (NCN_STRICT_CHECK_TOOLS): Check otool.
(ACX_CHECK_INSTALLED_TARGET_TOOL): Likewise
(GCC_TARGET_TOOL): Likewise.
* Makefile.tpl (HOST_EXPORTS): Add OTOOL, OTOOL_FOR_TARGET.
(BASE_TARGET_EXPORTS): OTOOL, export OTOOL_FOR_TARGET.
OTOOL, OTOOL_FOR_TARGET: New substitutions.
(EXTRA_HOST_FLAGS, EXTRA_TARGET_FLAGS): Add OTOOL.
* configure: Regenerate.
* Makefile.in: Likewise.

gcc/

* configure.ac (gcc_cv_otool): Set.
* configure: Regenerate.

From-SVN: r266831

5 years agocygming.h (PCC_BITFIELD_TYPE_MATTERS): Remove.
Uros Bizjak [Wed, 5 Dec 2018 20:40:56 +0000 (21:40 +0100)]
cygming.h (PCC_BITFIELD_TYPE_MATTERS): Remove.

* config/i386/cygming.h (PCC_BITFIELD_TYPE_MATTERS): Remove.
(GROUP_BITFIELDS_BY_ALIGN): Ditto.

From-SVN: r266830

5 years agoUse unsigned arithmetic for demoted vector plus/minus/mult (PR 88064)
Richard Sandiford [Wed, 5 Dec 2018 15:53:03 +0000 (15:53 +0000)]
Use unsigned arithmetic for demoted vector plus/minus/mult (PR 88064)

As Jakub pointed out, if we narrow a plus, minus or mult operation based
on the number of bits that consumers need, we have to convert a signed
operation to an unsigned one in order to avoid new undefined behaviour.
This patch does that and generalises vect_convert_input and
vect_recog_over_widening_pattern to cope with the extra casts.
(The changes to both functions are covered by existing tests.)

2018-12-03  Richard Sandiford  <richard.sandiford@arm.com>

gcc/
PR tree-optimization/88064
* tree-vect-patterns.c (vect_convert_input): Convert the result of
an existing cast if it has the right width but the wrong sign.
Do not test the signedness of the required result when
considering whether to split an existing cast; instead split to
a type with the same signedness as the source of the cast, then
convert it to the opposite signedness where necessary.
(vect_recog_over_widening_pattern): Handle sign changes between
the final PLUS_EXPR and the RSHIFT_EXPR.
(vect_recog_average_pattern): Use an unsigned operation when
truncating an addition, subtraction or multiplication.  Cast the
result back to the "real" signedness before promoting.

gcc/testsuite/
PR tree-optimization/88064
* gcc.dg/vect/vect-over-widen-23.c: New test.

From-SVN: r266829

5 years agotestsuite: turn down verbosity of "process-message"
David Malcolm [Wed, 5 Dec 2018 15:16:05 +0000 (15:16 +0000)]
testsuite: turn down verbosity of "process-message"

When debugging a failing test, I typically invoke DejaGnu at
verbosity level 2 (via RUNTESTFLAGS="-v -v dg.exp=something"),
so that DejaGnu prints the command line used to invoke the
compiler; specifically these two sites:
  target.exp "Invoking the compiler as "
  remote.exp "Executing on $hostname"
which are both verbosity level 2.

Unfortunately I run into an O(n^2) issue with logging from
process-message:

  verbose "process-message:\n${dg-messages}" 2

where, as each message each processed, it emits the state
of dg-messages, containing the new message and all messages so far,
leading to exponentially-increasing output at level 2 as more test
messages are added.

This patch papers over the problem by moving the
problematic message to verbosity level 3.

gcc/testsuite/ChangeLog:
* lib/gcc-dg.exp (process-message): Change verbosity level of
"verbose" from 2 to 3.
(dg-locus): Likewise.

From-SVN: r266828

5 years agore PR middle-end/63184 (Fails to simplify comparison)
Richard Biener [Wed, 5 Dec 2018 14:55:59 +0000 (14:55 +0000)]
re PR middle-end/63184 (Fails to simplify comparison)

2018-12-05  Richard Biener  <rguenther@suse.de>

PR middle-end/63184
* c-c++-common/pr19807-2.c: New testcase.
* c-c++-common/pr19807-3.c: Likewise.

From-SVN: r266827

5 years ago* udivmodhi4.c (__udivmodhi4): Fix loop end check.
Paul Koning [Wed, 5 Dec 2018 14:25:11 +0000 (09:25 -0500)]
* udivmodhi4.c (__udivmodhi4): Fix loop end check.

From-SVN: r266826

5 years agoProvide SPARCv9 MAXALIGN in gcc.target/sparc/attr-aligned.c (PR testsuite/88208)
Rainer Orth [Wed, 5 Dec 2018 14:16:08 +0000 (14:16 +0000)]
Provide SPARCv9 MAXALIGN in gcc.target/sparc/attr-aligned.c (PR testsuite/88208)

PR testsuite/88208
* gcc.target/sparc/attr-aligned.c (MAXALIGN) [__sparcv9 ||
__arch64__]: Define.

From-SVN: r266825

5 years agoBuild gcc.target/sparc/20181129-?.c as C99
Rainer Orth [Wed, 5 Dec 2018 14:10:06 +0000 (14:10 +0000)]
Build gcc.target/sparc/20181129-?.c as C99

* gcc.target/sparc/20181129-1.c: Compile with -std=c99.
* gcc.target/sparc/20181129-2.c: Likewise.

From-SVN: r266824

5 years agore PR tree-optimization/86637 (ICE: tree check: expected block, have <invalid tree...
Richard Biener [Wed, 5 Dec 2018 11:36:03 +0000 (11:36 +0000)]
re PR tree-optimization/86637 (ICE: tree check: expected block, have <invalid tree code> in inlining_chain_to_json, at optinfo-emit-json.cc:293)

2018-12-05  Richard Biener  <rguenther@suse.de>

PR tree-optimization/86637
* tree-vectorizer.c (pass_slp_vectorize::execute): Reset
vect_location at the end.

From-SVN: r266821

5 years agore PR tree-optimization/87360 (ICE in remove_redundant_iv_tests at gcc/tree-ssa-loop...
Jakub Jelinek [Wed, 5 Dec 2018 11:05:59 +0000 (12:05 +0100)]
re PR tree-optimization/87360 (ICE in remove_redundant_iv_tests at gcc/tree-ssa-loop-ivcanon.c:571 since r255467)

PR tree-optimization/87360
* gimple-loop-jam.c (tree_loop_unroll_and_jam): On failure to analyze
data dependencies, don't return false, just continue.  Formatting
fixes.
(merge_loop_tree, bb_prevents_fusion_p, unroll_jam_possible_p,
fuse_loops): Formatting fixes.

* g++.dg/opt/pr87360.C: New test.
* gfortran.dg/pr87360.f90: New test.

From-SVN: r266820

5 years agore PR c++/87897 (ICE in maybe_constant_value, at cp/constexpr.c:5255 since r265788)
Jakub Jelinek [Wed, 5 Dec 2018 08:43:12 +0000 (09:43 +0100)]
re PR c++/87897 (ICE in maybe_constant_value, at cp/constexpr.c:5255 since r265788)

PR c++/87897
* g++.dg/init/const13.C: New test.

From-SVN: r266818