platform/upstream/gcc.git
5 years agore PR fortran/90577 (FAIL: gfortran.dg/lrshift_1.f90 with -O(2|3) and -flto)
Harald Anlauf [Fri, 14 Jun 2019 18:41:20 +0000 (18:41 +0000)]
re PR fortran/90577 (FAIL: gfortran.dg/lrshift_1.f90 with -O(2|3) and -flto)

2019-06-14  Harald Anlauf  <anlauf@gmx.de>

PR fortran/90577
PR fortran/90578
* trans-intrinsic.c (gfc_conv_intrinsic_shift): Properly
distinguish logical/arithmetic shifts.
* intrinsic.texi: Update documentation for SHIFTR/SHIFTL/SHIFTA
(Fortran 2008) and LSHIFT/RSHIFT (GNU extensions).

PR fortran/90577
PR fortran/90578
* gfortran.dg/lrshift_1.f90: Adjust testcase.
* gfortran.dg/shiftalr_3.f90: New testcase.

From-SVN: r272309

5 years agoDarwin, opts processing - Initial tidy.
Iain Sandoe [Fri, 14 Jun 2019 18:31:44 +0000 (18:31 +0000)]
Darwin, opts processing - Initial tidy.

This change adds RejectNegative in a number of places it was needed,
reorders the opts into two groups (one general and one driver-only).
We also add a minimal description to each opt, and note some that
are now obsolete.  The only functional change is to permit
-mtarget-linker= as an alias of -mtarget-linker.

gcc/

2019-06-14  Iain Sandoe  <iain@sandoe.co.uk>

* config/darwin.opt: Add RejectNegative where needed, reorder
and add minimal functional descriptions.

From-SVN: r272308

5 years agore PR fortran/89646 (Spurious actual argument might interfere warning)
Steven G. Kargl [Fri, 14 Jun 2019 18:17:00 +0000 (18:17 +0000)]
re PR fortran/89646 (Spurious actual argument might interfere warning)

2019-06-14  Steven G. Kargl  <kargl@gcc.gnu.org>

PR fortran/89646
* dependency.c (gfc_check_argument_var_dependency): Suppress spurious
warnings by comparing variable names.

2019-06-14  Steven G. Kargl  <kargl@gcc.gnu.org>

PR fortran/89646
* gfortran.dg/pr89646.f90: New test.

From-SVN: r272307

5 years agoUpdate preferred_stack_boundary only when expanding function call
H.J. Lu [Fri, 14 Jun 2019 16:24:56 +0000 (16:24 +0000)]
Update preferred_stack_boundary only when expanding function call

locate_and_pad_parm is called when expanding function call from
initialize_argument_information and when generating function body
from assign_parm_find_entry_rtl:

  /* Remember if the outgoing parameter requires extra alignment on the
     calling function side.  */
  if (crtl->stack_alignment_needed < boundary)
    crtl->stack_alignment_needed = boundary;
  if (crtl->preferred_stack_boundary < boundary)
    crtl->preferred_stack_boundary = boundary;

stack_alignment_needed and preferred_stack_boundary should be updated
only when expanding function call, not when generating function body.

Add update_stack_alignment_for_call to update stack alignment when
outgoing parameter is passed in the stack.

gcc/

PR rtl-optimization/90765
* calls.c (update_stack_alignment_for_call): New function.
(expand_call): Call update_stack_alignment_for_call when
outgoing parameter is passed in the stack.
(emit_library_call_value_1): Likewise.
* function.c (locate_and_pad_parm): Don't update
stack_alignment_needed and preferred_stack_boundary.

gcc/testsuite/

PR rtl-optimization/90765
* gcc.target/i386/pr90765-1.c: New test.
* gcc.target/i386/pr90765-2.c: Likewise.

From-SVN: r272296

5 years agoUse __builtin_mul_overflow in xmallocarray
Janne Blomqvist [Fri, 14 Jun 2019 16:05:54 +0000 (19:05 +0300)]
Use __builtin_mul_overflow in xmallocarray

As GCC now provides builtins for doing integer overflow checking, lets
use it when checking for overflow in xmallocarray.

Regtested on x86_64-pc-linux-gnu.

libgfortran/ChangeLog:

2019-06-14  Janne Blomqvist  <jb@gcc.gnu.org>

* runtime/memory.c (SIZE_MAX):Remove macro definition.
(xmallocarray): Use __builtin_mul_overflow.

From-SVN: r272295

5 years agoi386: Update SSE <-> integer move costs
Hongtao Liu [Fri, 14 Jun 2019 15:41:43 +0000 (15:41 +0000)]
i386: Update SSE <-> integer move costs

Since inline_secondary_memory_needed has

  /* ??? This is a lie.  We do have moves between mmx/general, and for
     mmx/sse2.  But by saying we need secondary memory we discourage the
     register allocator from using the mmx registers unless needed.  */
  if (MMX_CLASS_P (class1) != MMX_CLASS_P (class2))
    return true;

moves between MMX and non-MMX units require secondary memory.   There
is no need to check moves between MMX and integer units.

struct processor_costs has:

  const int mmxsse_to_integer;  /* cost of moving mmxsse register to
                                   integer.  */
  const int ssemmx_to_integer;  /* cost of moving integer to mmxsse register. */

This patch also renames mmxsse_to_integer to sse_to_integer and
ssemmx_to_integer to integer_to_sse.

Tested on Linux/x86-64.

PR target/90877
* config/i386/i386-features.c
(dimode_scalar_chain::compute_convert_gain): Replace
mmxsse_to_integer with sse_to_integer.
* config/i386/i386.c (ix86_register_move_cost): Verify that
moves between MMX and non-MMX units require secondary memory.
Correct costs of moves between SSE and integer units.
* config/i386/i386.h (processor_costs): Rename cost of moving
SSE register to integer to sse_to_integer.  Rename cost of
moving integer register to SSE to integer_to_sse.

From-SVN: r272294

5 years agoPR c++/90884 - stray note with -Wctor-dtor-privacy.
Marek Polacek [Fri, 14 Jun 2019 14:49:57 +0000 (14:49 +0000)]
PR c++/90884 - stray note with -Wctor-dtor-privacy.

* class.c (maybe_warn_about_overly_private_class): Guard the call to
inform.

* g++.dg/warn/ctor-dtor-privacy-4.C: New.
* g++.dg/warn/ctor-dtor-privacy-4.h: New.

From-SVN: r272291

5 years ago[NetBSD] Add support for the Arm EABI.
Matt Thomas [Fri, 14 Jun 2019 14:04:20 +0000 (14:04 +0000)]
[NetBSD] Add support for the Arm EABI.

This is a roll-up of a set of changes needed to support the Arm EABI on NetBSD.

2019-06-14  Matt Thomas  <matt@3am-software.com>
    Matthew Green  <mrg@eterna.com.au>
    Nick Hudson  <skrll@netbsd.org>
    Maya Rashish  <coypu@sdf.org>
    Richard Earnshaw  <rearnsha@arm.com>

gcc:

* config.gcc (arm*-*-netbsdelf*) Add support for EABI configuration.
* config.host (arm*-*-netbsd*): Use driver-arm.o on native NetBSD.
* config/arm/netbsd-eabi.h: New file.
* config/arm/netbsd-elf.h (TARGET_OS_CPP_BUILTINS): Undefine before
redefining.
(SUBTARGET_EXTRA_ASM_SPEC): Don't pass -matpcs to the assembler.
* config/netbsd-elf.h (NETBSD_LINK_LD_ELF_SO_SPEC): New define.
(NETBSD_SUBTARGET_EXTRA_SPECS): New define.
(SUBTARGET_EXTRA_SPECS): Define to NETBSD_SUBTARGET_EXTRA_SPECS.

libatomic:
* configure.tgt (arm*): Handle NetBSD in the same way as FreeBSD.

libgcc:
* config.host (arm*-*-netbsdelf*): Add support for EABI configurations.
* config/arm/t-netbsd (LIB1ASMFUNCS): Add some additional assembler
functions to build.
* config/arm/t-netbsd-eabi: New file.

Co-Authored-By: Matthew Green <mrg@eterna.com.au>
Co-Authored-By: Maya Rashish <coypu@sdf.org>
Co-Authored-By: Nick Hudson <skrll@netbsd.org>
Co-Authored-By: Richard Earnshaw <rearnsha@arm.com>
From-SVN: r272290

5 years agoDisable -Wctor-dtor-privacy warnings for some standard types
Jonathan Wakely [Fri, 14 Jun 2019 14:03:20 +0000 (15:03 +0100)]
Disable -Wctor-dtor-privacy warnings for some standard types

* include/experimental/type_traits (experimental::nonesuch): Use
pragma to disable -Wctor-dtor-privacy warnings.
* include/std/type_traits (__is_convertible_helper<From, To, false>)
(__is_nt_convertible_helper<From, To, false>, __nonesuch): Likewise.

From-SVN: r272289

5 years agoAdd __cpp_lib_bind_front macro to <version> header
Jonathan Wakely [Fri, 14 Jun 2019 14:03:16 +0000 (15:03 +0100)]
Add __cpp_lib_bind_front macro to <version> header

* include/std/version (__cpp_lib_bind_front): Add missing macro.

From-SVN: r272288

5 years agoPR c++/85552 - wrong instantiation of dtor for DMI.
Jason Merrill [Fri, 14 Jun 2019 13:22:33 +0000 (09:22 -0400)]
PR c++/85552 - wrong instantiation of dtor for DMI.

The problem here is that when processing direct-initialization of a data
member, we don't need to worry about destruction semantics; that will be
handled in the [cd]tor.  Conveniently, we already have tf_no_cleanup from a
similar fix to new-expressions.

* typeck2.c (digest_nsdmi_init): Set tf_no_cleanup for direct-init.

From-SVN: r272287

5 years agotree-loop-distribution.c (classify_partition): Return whether a reduction appeared...
Richard Biener [Fri, 14 Jun 2019 11:29:44 +0000 (11:29 +0000)]
tree-loop-distribution.c (classify_partition): Return whether a reduction appeared in all partitions and do not stop builtin...

2019-06-14  Richard Biener  <rguenther@suse.de>

* tree-loop-distribution.c (classify_partition): Return
whether a reduction appeared in all partitions and do not
stop builtin detection because of this.
(distribute_loop): Sort a non-builtin partition last if
there's a reduction in all partitions and make sure the
partition prevailing as last is not a builtin.

* gcc.dg/tree-ssa/ldist-26.c: Adjust.

From-SVN: r272284

5 years agore PR ipa/90401 (Missed propagation of by-ref constant argument to callee function)
Feng Xue [Fri, 14 Jun 2019 02:34:48 +0000 (02:34 +0000)]
re PR ipa/90401 (Missed propagation of by-ref constant argument to callee function)

PR ipa/90401

gcc/ChangeLog:

* ipa-prop.c (add_to_agg_contents_list): New function.
(clobber_by_agg_contents_list_p): Likewise.
(extract_mem_content): Likewise.
(get_place_in_agg_contents_list): Delete.
(determine_known_aggregate_parts): Renamed from
determine_locally_known_aggregate_parts.  New parameter
aa_walk_budget_p.

gcc/testsuite/ChangeLog:

* gcc.dg/ipa/ipcp-agg-10.c: New test.

From-SVN: r272282

5 years agoPR tree-optimization/90662 - strlen of a string in a vla plus offset not folded
Martin Sebor [Fri, 14 Jun 2019 02:07:02 +0000 (02:07 +0000)]
PR tree-optimization/90662 - strlen of a string in a vla plus offset not folded

gcc/ChangeLog:

PR tree-optimization/90662
* tree-ssa-strlen.c (get_stridx): Convert fold_build2 operands
to the same type.

gcc/testsuite/ChangeLog:

PR tree-optimization/90662
* gcc.dg/pr90866-2.c: New test.
* gcc.dg/pr90866.c: Ditto.

From-SVN: r272281

5 years agoDaily bump.
GCC Administrator [Fri, 14 Jun 2019 00:16:16 +0000 (00:16 +0000)]
Daily bump.

From-SVN: r272280

5 years agore PR bootstrap/90873 (-Wmaybe-uninitialized warning in gcc/tree-ssa-forwprop.c break...
Jan Hubicka [Thu, 13 Jun 2019 21:56:37 +0000 (23:56 +0200)]
re PR bootstrap/90873 (-Wmaybe-uninitialized warning in gcc/tree-ssa-forwprop.c breaks 32-bit bootstrap)

PR bootstrap/90873
* tree-ssa-alias.c (indirect_ref_may_alias_decl_p): Also check that
dbase is not TARGET_MEM_REF.

From-SVN: r272273

5 years ago* ChangeLog: Fix my last entry.
Uros Bizjak [Thu, 13 Jun 2019 21:37:17 +0000 (23:37 +0200)]
* ChangeLog: Fix my last entry.

From-SVN: r272272

5 years agoi386.md (SWIM1248s): Rename from SWIM1248x.
Uros Bizjak [Thu, 13 Jun 2019 21:29:06 +0000 (23:29 +0200)]
i386.md (SWIM1248s): Rename from SWIM1248x.

* config/i386/i386.md (SWIM1248s): Rename from SWIM1248x.
Update all uses.
(and<mode>3): Use gen_extend_insn instead of indirect functions.
Do not generate DImode extends for 32bit targets.
(and->zext post-reload splitter): Use gen_extend_insn
instead of indirect functions.
(anddi->zext pre-reload splitter): New.

* config/i386/i386-expand.c (ix86_expand_int_sse_cmp):
Use gen_sub3_insn instead of indirect function.
(ix86_expand_ashl_const): Use gen_add2_insn instead of
indirect function.
(ix86_adjust_counter): Ditto.

From-SVN: r272270

5 years agore PR tree-optimization/77820 (A jump threading opportunity with conditionals)
Jiufu Guo [Thu, 13 Jun 2019 18:55:55 +0000 (18:55 +0000)]
re PR tree-optimization/77820 (A jump threading opportunity with conditionals)

PR tree-optimization/77820
* tree-ssa-threadedge.c
(edge_forwards_cmp_to_conditional_jump_through_empty_bb_p): New
function.
(thread_across_edge): Add call to
edge_forwards_cmp_to_conditional_jump_through_empty_bb_p.

PR tree-optimization/77820
* gcc.dg/tree-ssa/phi_on_compare-1.c: New testcase.
* gcc.dg/tree-ssa/phi_on_compare-2.c: New testcase.
* gcc.dg/tree-ssa/phi_on_compare-3.c: New testcase.
* gcc.dg/tree-ssa/phi_on_compare-4.c: New testcase.
* gcc.dg/tree-ssa/split-path-6.c: Update testcase.
* gcc.target/sh/pr51244-20.c: Update testcase.

Co-Authored-By: Lijia He <helijia@linux.ibm.com>
From-SVN: r272261

5 years agoDarwin, Driver - Improve processing of macosx-version-min=
Iain Sandoe [Thu, 13 Jun 2019 18:53:05 +0000 (18:53 +0000)]
Darwin, Driver - Improve processing of macosx-version-min=

For PR target/63810 some improvements were made in the parsing of
the version string at the point it's used to define the built-in
__ENVIRONMENT_MAC_OS_X_VERSION_MIN_REQUIRED__.  This is fine, but the
specs processing also uses the version, and specs version-compare
doesn't like leading zeros on components.  This means that while we
succeed in processing -mmacosx-version-min=010.00002.000099 on compile
lines, it fails for any other line that uses the value as part of a spec
(in particular, link lines fail).

To fix this, we need to apply a bit of clean-up to the version that's
presented to the driver, and push that back into the command line opts.

The value can come from four places:
 1. User-entered on the command line
 2. User-entered as MACOSX_DEPLOYMENT_TARGET= environment var.
 3. Absent those two
   3a For self-hosting systems, look-up from the kernel
   3b For cross-compilers, as a default supplied at configure time.

We apply the clean-up to all 4 (although it shouldn't really be needed
for the cases under 3).

We also supply a test-case that adapts to the target-version of the
system, so that the link requirements are met by the SDK in use (if you
try to link i686-darwin9 on an x86-64-darwin18 SDK, it will fail).

gcc/

2019-06-13  Iain Sandoe  <iain@sandoe.co.uk>

* config/darwin-driver.c (validate_macosx_version_min): New.
(darwin_default_min_version): Cleanup and validate supplied version.
(darwin_driver_init): Likewise and push cleaned version into opts.

gcc/testsuite/

2019-06-13  Iain Sandoe  <iain@sandoe.co.uk>

* gcc.dg/darwin-minversion-link.c: New test.

From-SVN: r272260

5 years agore PR fortran/68544 (ICE trying to pass derived type constructor as a function)
Steven G. Kargl [Thu, 13 Jun 2019 18:40:19 +0000 (18:40 +0000)]
re PR fortran/68544 (ICE trying to pass derived type constructor as a function)

2019-06-13  Steven G. Kargl  <kargl@gcc.gnu.org>

PR fortran/68544
* resolve.c (is_dt_name): New function to compare symbol name against
list of derived types.
(resolve_actual_arglist): Use it to find wrong code.

2019-06-13  Steven G. Kargl  <kargl@gcc.gnu.org>

PR fortran/68544
* gfortran.dg/pr68544.f90: New test.
* gfortran.dg/pr85687.f90: Modify test for new error message.

From-SVN: r272259

5 years agoDarwin, testsuite - Fix pr71694 fail for m32.
Iain Sandoe [Thu, 13 Jun 2019 18:32:32 +0000 (18:32 +0000)]
Darwin, testsuite - Fix pr71694 fail for m32.

This test fails for Darwin m32 because it's scanning for absence
of an instruction that's validly used in PIC code.  Fixed, in this
case, by using non-PIC codegen.

2019-06-13  Iain Sandoe  <iain@sandoe.co.uk>

* g++.dg/pr71694.C: Use non-PIC codegen for Darwin m32.

From-SVN: r272257

5 years agore PR fortran/89344 (uncaught programmer error: polymorphic variable is INTENT(IN...
Steven G. Kargl [Thu, 13 Jun 2019 18:07:53 +0000 (18:07 +0000)]
re PR fortran/89344 (uncaught programmer error:  polymorphic variable is INTENT(IN) but assigned to without error)

2019-06-13  Steven G. Kargl  <kargl@gcc.gnu.org>

PR fortran/89344
* expr.c (gfc_check_vardef_context): Check for INTENT(IN) variable
in SELECT TYPE construct.

2019-06-13  Steven G. Kargl  <kargl@gcc.gnu.org>

PR fortran/89344
 * gfortran.dg/pr89344.f90: New test.

From-SVN: r272255

5 years agore PR fortran/88810 (gcc/fortran/dependency.c:2200: possible cut'n'paste error ?)
Steven G. Kargl [Thu, 13 Jun 2019 17:45:52 +0000 (17:45 +0000)]
re PR fortran/88810 (gcc/fortran/dependency.c:2200: possible cut'n'paste error ?)

2019-06-13  Steven G. Kargl  <kargl@gcc.gnu.org>

PR fortran/88810
* dependency.c (gfc_dep_resolver): Re-arrange code to make the logic
a bit more transparent.  Fix 2 nearby formatting issues.

From-SVN: r272254

5 years agoDarwin, testsuite - Adjust two tests that fail on newer OS versions.
Iain Sandoe [Thu, 13 Jun 2019 17:15:37 +0000 (17:15 +0000)]
Darwin, testsuite - Adjust two tests that fail on newer OS versions.

Newer OS versions (10.14+) do not provide some of the CRTs that are used
for older ones (e.g. 10.5), and thus link tests that specify targeting
such a revision fail.  We retain the testing of the correct defined OS
version number but switch to compile-only testing for these two tests.

2019-06-13  Iain Sandoe  <iain@sandoe.co.uk>

* gcc.dg/darwin-minversion-1.c: Use compile rather than link/run.
* gcc.dg/darwin-minversion-2.c: Likewise.

From-SVN: r272251

5 years agore PR target/90871 (g++.dg/ext/altivec-15.C fails after r272236)
Paolo Carlini [Thu, 13 Jun 2019 17:10:03 +0000 (17:10 +0000)]
re PR target/90871 (g++.dg/ext/altivec-15.C fails after r272236)

2019-06-13  Paolo Carlini  <paolo.carlini@oracle.com>

PR target/90871
* g++.dg/ext/altivec-15.C: Add dg-error directive.

From-SVN: r272250

5 years agotestsuite - Require alias support for pr90760.
Iain Sandoe [Thu, 13 Jun 2019 16:58:35 +0000 (16:58 +0000)]
testsuite - Require alias support for pr90760.

This test fails on Darwin because it requires alias support, so
add the relevant dg-requires.

2019-06-13  Iain Sandoe  <iain@sandoe.co.uk>

* gcc.dg/pr90760.c: Require alias support.

From-SVN: r272248

5 years agore PR tree-optimization/90869 (Non-disambiguated memory accesses)
Jan Hubicka [Thu, 13 Jun 2019 15:00:41 +0000 (17:00 +0200)]
re PR tree-optimization/90869 (Non-disambiguated memory accesses)

PR tree-optimize/90869
* tree-ssa-alias.c (indirect_ref_may_alias_decl_p): Watch for view
converts in MEM_REF referencing decl rather than view converts
from decl type to MEM_REF type.

* g++.dg/tree-ssa/alias-access-path-1.C: New testcase.

From-SVN: r272247

5 years agore PR tree-optimization/90856 (ICE: verify_gimple failed (error: incompatible types...
Richard Biener [Thu, 13 Jun 2019 13:03:13 +0000 (13:03 +0000)]
re PR tree-optimization/90856 (ICE: verify_gimple failed (error: incompatible types in 'PHI' argument 1))

2019-06-13  Richard Biener  <rguenther@suse.de>

PR tree-optimization/90856
* tree-sra.c (build_ref_for_model): Only use
build_reconstructed_reference when address-spaces are the same.

* gcc.target/i386/pr90856.c: New testcase.

From-SVN: r272244

5 years agonvptx.c (nvptx_sese_number, [...]): Don't wrap ei variable name in the declaration...
Jakub Jelinek [Thu, 13 Jun 2019 12:51:39 +0000 (14:51 +0200)]
nvptx.c (nvptx_sese_number, [...]): Don't wrap ei variable name in the declaration in ()s.

* config/nvptx/nvptx.c (nvptx_sese_number, nvptx_sese_pseudo): Don't
wrap ei variable name in the declaration in ()s.
(nvptx_single): Actually use mode_label variable.  Formatting fix.

From-SVN: r272243

5 years agossa-dse-1.C: Don't match exact number of chars of = {} store.
Jakub Jelinek [Thu, 13 Jun 2019 11:29:13 +0000 (13:29 +0200)]
ssa-dse-1.C: Don't match exact number of chars of = {} store.

* g++.dg/tree-ssa/ssa-dse-1.C: Don't match exact number of chars of
= {} store.
* g++.dg/tree-ssa/pr31146.C: Change -fdump-tree-forwprop to
-fdump-tree-forwprop1 in dg-options.  Expect <int[5]> in MEM.

From-SVN: r272240

5 years agotree-vectorizer.h (vect_loop_vectorized_call): Declare.
Richard Biener [Thu, 13 Jun 2019 10:10:05 +0000 (10:10 +0000)]
tree-vectorizer.h (vect_loop_vectorized_call): Declare.

2019-06-13  Richard Biener  <rguenther@suse.de>

* tree-vectorizer.h (vect_loop_vectorized_call): Declare.
* tree-vectorizer.c (vect_loop_vectorized_call): Export and
also return the condition stmt.
* tree-vect-loop-manip.c (vect_loop_versioning): Compute outermost
loop we can version and version that, reusing the loop version
created by if-conversion instead of versioning again.

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

From-SVN: r272239

5 years agoRevamp value_range::may_contain_p.
Aldy Hernandez [Thu, 13 Jun 2019 09:46:00 +0000 (09:46 +0000)]
Revamp value_range::may_contain_p.

From-SVN: r272238

5 years ago[ARC] Add documentation naked, ilink and firq
Claudiu Zissulescu [Thu, 13 Jun 2019 09:35:19 +0000 (11:35 +0200)]
[ARC] Add documentation naked, ilink and firq

gcc/
xxxx-xx-xx  Claudiu Zissulescu  <claziss@synopsys.com>

* doc/extend.texi (ARC Function Attributes): Update info.

From-SVN: r272237

5 years agodecl.c (grokdeclarator): Use id_loc in five additional places in the last part of...
Paolo Carlini [Thu, 13 Jun 2019 08:45:46 +0000 (08:45 +0000)]
decl.c (grokdeclarator): Use id_loc in five additional places in the last part of the function.

/cp
2019-06-13  Paolo Carlini  <paolo.carlini@oracle.com>

* decl.c (grokdeclarator): Use id_loc in five additional places
in the last part of the function.

/testsuite
2019-06-13  Paolo Carlini  <paolo.carlini@oracle.com>

* g++.dg/diagnostic/variably-modified-type-1.C: New.
* g++.dg/cpp0x/alias-decl-1.C: Test the location too.
* g++.dg/other/pr84792-1.C: Likewise.
* g++.dg/other/pr84792-2.C: Likewise.
* g++.dg/parse/error24.C: Likewise.
* g++.dg/parse/error32.C: Likewise.
* g++.dg/parse/error33.C: Likewise.
* g++.dg/parse/saved1.C: Likewise.
* g++.dg/template/operator6.C: Likewise.
* g++.dg/template/pr61745.C: Likewise.
* g++.dg/template/typedef41.C: Likewise.
* g++.old-deja/g++.jason/crash10.C: Likewise.

/cp
2019-06-13  Paolo Carlini  <paolo.carlini@oracle.com>

* decl.c (grokdeclarator): Move further up the declaration of
id_loc, use it immediately, update its value after the loop
over declarator, use it again in the final part of function;
improve locations of error messages about multiple data types
and conflicting specifiers.

/testsuite
2019-06-13  Paolo Carlini  <paolo.carlini@oracle.com>

* g++.dg/diagnostic/conflicting-specifiers-1.C: New.
* g++.dg/diagnostic/two-or-more-data-types-1.C: Likewise.
* g++.dg/parse/error10.C: Adjust location.
* g++.dg/parse/pragma2.C: Likewise.

From-SVN: r272236

5 years agore PR c++/90801 (A recurring hang)
Richard Biener [Thu, 13 Jun 2019 08:41:13 +0000 (08:41 +0000)]
re PR c++/90801 (A recurring hang)

2019-06-13  Richard Biener  <rguenther@suse.de>

PR c++/90801
* typeck2.c (split_nonconstant_init_1): Properly count
num_split_elts, optimize single constructor elt removal.

From-SVN: r272235

5 years agoPR tree-optimization/89713 - Assume loop with an exit is finite
Feng Xue [Thu, 13 Jun 2019 04:17:42 +0000 (04:17 +0000)]
PR tree-optimization/89713 - Assume loop with an exit is finite

gcc/ChangeLog:

        * doc/invoke.texi (-ffinite-loops): Document new option.
        * common.opt (-ffinite-loops): New option.
        * tree-ssa-dce.c (mark_stmt_if_obviously_necessary): Mark
        IFN_GOACC_LOOP calls as necessary.
        * tree-ssa-loop-niter.c (finite_loop_p): Assume loop with an exit
        is finite.
        * omp-offload.c (oacc_xform_loop): Skip lowering if return value of
        IFN_GOACC_LOOP call is not used.
        * opts.c (default_options_table): Enable -ffinite-loops at -O2+.

gcc/testsuite/ChangeLog:

        * g++.dg/tree-ssa/empty-loop.C: New test.
        * gcc.dg/tree-ssa/dce-2.c: New test.
        * gcc.dg/const-1.c: Add -fno-finite-loops option.
        * gcc.dg/graphite/graphite.exp: Likewise.
        * gcc.dg/loop-unswitch-1.c: Likewise.
        * gcc.dg/predict-9.c: Likewise.
        * gcc.dg/pure-2.c: Likewise.
        * gcc.dg/tree-ssa/20040211-1.c: Likewise.
        * gcc.dg/tree-ssa/loop-10.c: Likewise.
        * gcc.dg/tree-ssa/split-path-6.c: Likewise.
        * gcc.dg/tree-ssa/ssa-thread-12.c: Likewise.

libgomp/ChangeLog:

        * testsuite/libgomp.oacc-c-c++-common/pr84955-1.c: New test.

From-SVN: r272234

5 years agore PR target/88838 ([SVE] Use 32-bit WHILELO in LP64 mode)
Kugan Vivekanandarajah [Thu, 13 Jun 2019 03:34:28 +0000 (03:34 +0000)]
re PR target/88838 ([SVE] Use 32-bit WHILELO in LP64 mode)

gcc/ChangeLog:

2019-06-13  Kugan Vivekanandarajah  <kugan.vivekanandarajah@linaro.org>

PR target/88838
* tree-vect-loop-manip.c (vect_set_loop_masks_directly): If the
compare_type is not with Pmode size, we will create an IV with
Pmode size with truncated use (i.e. converted to the correct type).
* tree-vect-loop.c (vect_verify_full_masking): Find IV type.
(vect_iv_limit_for_full_masking): New. Factored out of
vect_set_loop_condition_masked.
* tree-vectorizer.h (LOOP_VINFO_MASK_IV_TYPE): New.
(vect_iv_limit_for_full_masking): Declare.

gcc/testsuite/ChangeLog:

2019-06-13  Kugan Vivekanandarajah  <kugan.vivekanandarajah@linaro.org>

PR target/88838
* gcc.target/aarch64/pr88838.c: New test.
* gcc.target/aarch64/sve/while_1.c: Adjust.

From-SVN: r272233

5 years agore PR target/88834 ([SVE] Poor addressing mode choices for LD2 and ST2)
Kugan Vivekanandarajah [Thu, 13 Jun 2019 03:18:54 +0000 (03:18 +0000)]
re PR target/88834 ([SVE] Poor addressing mode choices for LD2 and ST2)

gcc/ChangeLog:

2019-06-13  Kugan Vivekanandarajah  <kugan.vivekanandarajah@linaro.org>

PR target/88834
* tree-ssa-loop-ivopts.c (get_mem_type_for_internal_fn): Handle
IFN_MASK_LOAD_LANES and IFN_MASK_STORE_LANES.
(get_alias_ptr_type_for_ptr_address): Likewise.
(add_iv_candidate_for_use): Add scaled index candidate if useful.
* tree-ssa-address.c (preferred_mem_scale_factor): New.
* config/aarch64/aarch64.c (aarch64_classify_address): Relax
allow_reg_index_p.

gcc/testsuite/ChangeLog:

2019-06-13  Kugan Vivekanandarajah  <kugan.vivekanandarajah@linaro.org>

PR target/88834
* gcc.target/aarch64/pr88834.c: New test.
* gcc.target/aarch64/sve/struct_vect_1.c: Adjust.
* gcc.target/aarch64/sve/struct_vect_14.c: Likewise.
* gcc.target/aarch64/sve/struct_vect_15.c: Likewise.
* gcc.target/aarch64/sve/struct_vect_16.c: Likewise.
* gcc.target/aarch64/sve/struct_vect_17.c: Likewise.
* gcc.target/aarch64/sve/struct_vect_7.c: Likewise.

From-SVN: r272232

5 years agoiterators.md (ADDSUB): Fix typo in comment.
Kugan Vivekanandarajah [Thu, 13 Jun 2019 03:13:56 +0000 (03:13 +0000)]
iterators.md (ADDSUB): Fix typo in comment.

gcc/ChangeLog:

2019-06-13  Kugan Vivekanandarajah  <kugan.vivekanandarajah@linaro.org>

* config/aarch64/iterators.md (ADDSUB): Fix typo in comment.

From-SVN: r272231

5 years agore PR c++/87410 (internal compiler error: in fold_convert_loc, at fold-const.c:2530)
Marek Polacek [Thu, 13 Jun 2019 01:31:03 +0000 (01:31 +0000)]
re PR c++/87410 (internal compiler error: in fold_convert_loc, at fold-const.c:2530)

PR c++/87410
* g++.dg/cpp1y/pr87410.C: New test.

From-SVN: r272230

5 years agoDaily bump.
GCC Administrator [Thu, 13 Jun 2019 00:16:15 +0000 (00:16 +0000)]
Daily bump.

From-SVN: r272228

5 years agoPR c++/66999 - 'this' captured by reference.
Marek Polacek [Wed, 12 Jun 2019 22:41:35 +0000 (22:41 +0000)]
PR c++/66999 - 'this' captured by reference.

* parser.c (cp_parser_lambda_introducer): Reject `&this'.  Use
cp_lexer_nth_token_is instead of cp_lexer_peek_nth_token.

* g++.dg/cpp0x/lambda/lambda-this21.C: New test.

From-SVN: r272223

5 years agoio.c (check_format): Use G_(...) instead of _(...) for error values...
Jakub Jelinek [Wed, 12 Jun 2019 22:38:10 +0000 (00:38 +0200)]
io.c (check_format): Use G_(...) instead of _(...) for error values...

* io.c (check_format): Use G_(...) instead of _(...) for error values,
append " in format string at %L" to all strings but unexpected_element,
use error as gfc_error formating string instead of
"%s in format string at %L".  Formatting fixes.

From-SVN: r272222

5 years agoPR c++/90825 - endless recursion when evaluating sizeof.
Marek Polacek [Wed, 12 Jun 2019 22:26:54 +0000 (22:26 +0000)]
PR c++/90825 - endless recursion when evaluating sizeof.

PR c++/90832 - endless recursion when evaluating sizeof.
* constexpr.c (cxx_eval_constant_expression): Don't recurse on the
result of fold_sizeof_expr if is returns a SIZEOF_EXPR.
* typeck.c (cxx_sizeof_expr): Only return a SIZEOF_EXPR if the operand
is instantiation-dependent.

* g++.dg/cpp0x/constexpr-sizeof2.C: New test.
* g++.dg/cpp0x/constexpr-sizeof3.C: New test.

From-SVN: r272221

5 years agogfortran.h (gfc_free_dt_list): Remove prototype.
Steven G. Kargl [Wed, 12 Jun 2019 22:18:37 +0000 (22:18 +0000)]
gfortran.h (gfc_free_dt_list): Remove prototype.

2019-06-12  Steven G. Kargl  <kargl@gcc.gnu.org>

* gfortran.h (gfc_free_dt_list): Remove prototype.

From-SVN: r272220

5 years agoTweak tests to avoid fallout from MEM_REF changes on targets that don't enable...
Martin Sebor [Wed, 12 Jun 2019 20:43:27 +0000 (20:43 +0000)]
Tweak tests to avoid fallout from MEM_REF changes on targets that don't enable...

Tweak tests to avoid fallout from MEM_REF changes on targets that don't
enable store merging but achieve the same effect by other means (DSE).

gcc/testsuite?ChangeLog:

PR middle-end/90676
* gcc.dg/tree-ssa/pr30375.c: Adjust and simplify expected test
output some more.
* gcc.dg/tree-ssa/slsr-27.c: Ditto.
* gcc.dg/tree-ssa/slsr-28.c: Ditto.
* gcc.dg/tree-ssa/slsr-29.c: Ditto.
* gcc.dg/tree-ssa/ssa-dse-24.c: Ditto.

From-SVN: r272218

5 years agoPR c++/90736 - bogus error with alignof.
Marek Polacek [Wed, 12 Jun 2019 20:17:36 +0000 (20:17 +0000)]
PR c++/90736 - bogus error with alignof.

* constexpr.c (adjust_temp_type): Use cv_unqualified type.

* g++.dg/cpp0x/alignof5.C: New test.

From-SVN: r272217

5 years agoFix incorrect __cpp_lib_parallel_algorithm macro definitions
Jonathan Wakely [Wed, 12 Jun 2019 20:16:03 +0000 (21:16 +0100)]
Fix incorrect __cpp_lib_parallel_algorithm macro definitions

* include/std/algorithm (__cpp_lib_parallel_algorithm): Fix value.
* include/std/memory (__cpp_lib_parallel_algorithm): Likewise.
* include/std/numeric (__cpp_lib_parallel_algorithm): Likewise.
* testsuite/25_algorithms/pstl/feature_test.cc: New test.

From-SVN: r272216

5 years agoAdd myself to MAINTAINERS
Dimitar Dimitrov [Wed, 12 Jun 2019 20:10:21 +0000 (20:10 +0000)]
Add myself to MAINTAINERS

2019-06-12  Dimitar Dimitrov  <dimitar@dinux.eu>

* MAINTAINERS (Write After Approval): Add myself.

From-SVN: r272215

5 years agotestsuite: Mark testsuite that PRU has different
Dimitar Dimitrov [Wed, 12 Jun 2019 19:52:54 +0000 (19:52 +0000)]
testsuite: Mark testsuite that PRU has different

gcc/testsuite/ChangeLog:

2019-06-12  Dimitar Dimitrov  <dimitar@dinux.eu>

        * gcc.dg/builtin-apply2.c: Skip for PRU.
        * gcc.dg/torture/stackalign/builtin-apply-2.c: Ditto.

From-SVN: r272212

5 years agotestsuite: Mark that PRU uses all function pointer bits
Dimitar Dimitrov [Wed, 12 Jun 2019 19:50:45 +0000 (19:50 +0000)]
testsuite: Mark that PRU uses all function pointer bits

gcc/testsuite/ChangeLog:

2019-06-12  Dimitar Dimitrov  <dimitar@dinux.eu>

* g++.old-deja/g++.abi/ptrmem.C: Add PRU to list.

From-SVN: r272211

5 years agotestsuite: Mark that PRU has one-cycle jumps
Dimitar Dimitrov [Wed, 12 Jun 2019 19:48:59 +0000 (19:48 +0000)]
testsuite: Mark that PRU has one-cycle jumps

gcc/testsuite/ChangeLog:

2019-06-12  Dimitar Dimitrov  <dimitar@dinux.eu>

* gcc.dg/tree-ssa/20040204-1.c: XFAIL on pru.
* gcc.dg/tree-ssa/reassoc-33.c: Ditto.
* gcc.dg/tree-ssa/reassoc-34.c: Ditto.
* gcc.dg/tree-ssa/reassoc-35.c: Ditto.
* gcc.dg/tree-ssa/reassoc-36.c: Ditto.

From-SVN: r272210

5 years agotestsuite: Define PRU stack usage
Dimitar Dimitrov [Wed, 12 Jun 2019 19:47:09 +0000 (19:47 +0000)]
testsuite: Define PRU stack usage

gcc/testsuite/ChangeLog:

2019-06-12  Dimitar Dimitrov  <dimitar@dinux.eu>

* gcc.dg/stack-usage-1.c: Define PRU stack usage.

From-SVN: r272209

5 years agotestsuite: Remove PRU from test cases requiring hosted environment
Dimitar Dimitrov [Wed, 12 Jun 2019 19:44:21 +0000 (19:44 +0000)]
testsuite: Remove PRU from test cases requiring hosted environment

gcc/testsuite/ChangeLog:

2019-06-12  Dimitar Dimitrov  <dimitar@dinux.eu>

* gcc.c-torture/execute/20101011-1.c: Define DO_TEST to 0 for PRU.
* gcc.dg/20020312-2.c: No PIC register for PRU.

From-SVN: r272208

5 years agotestsuite: Add check for unsupported TI ABI PRU features to testsuite
Dimitar Dimitrov [Wed, 12 Jun 2019 19:41:02 +0000 (19:41 +0000)]
testsuite: Add check for unsupported TI ABI PRU features to testsuite

gcc/testsuite/ChangeLog:

2019-06-12  Dimitar Dimitrov  <dimitar@dinux.eu>

* lib/gcc-dg.exp: Filter unsupported features in PRU's TI ABI mode.
* lib/target-utils.exp: Ditto.
* lib/target-supports.exp (check_effective_target_function_pointers,
check_effective_target_large_return_values): New.

From-SVN: r272207

5 years agotestsuite: Add check for overflowed IMEM region to testsuite
Dimitar Dimitrov [Wed, 12 Jun 2019 19:36:34 +0000 (19:36 +0000)]
testsuite: Add check for overflowed IMEM region to testsuite

gcc/testsuite/ChangeLog:

2019-06-12  Dimitar Dimitrov  <dimitar@dinux.eu>

* lib/gcc-dg.exp: Bail on region overflow for tiny targets.
* lib/target-utils.exp: Ditto.
* lib/target-supports.exp: Declare PRU target as tiny.

From-SVN: r272206

5 years agotestsuite: Add PRU tests
Dimitar Dimitrov [Wed, 12 Jun 2019 19:33:31 +0000 (19:33 +0000)]
testsuite: Add PRU tests

gcc/testsuite/ChangeLog:

2019-06-12  Dimitar Dimitrov  <dimitar@dinux.eu>

* gcc.target/pru/abi-arg-struct.c: New test.
* gcc.target/pru/ashiftrt.c: New test.
* gcc.target/pru/builtins-1.c: New test.
* gcc.target/pru/builtins-error.c: New test.
* gcc.target/pru/clearbit.c: New test.
* gcc.target/pru/loop-asm.c: New test.
* gcc.target/pru/loop-dowhile.c: New test.
* gcc.target/pru/loop-hi-1.c: New test.
* gcc.target/pru/loop-hi-2.c: New test.
* gcc.target/pru/loop-qi-1.c: New test.
* gcc.target/pru/loop-qi-2.c: New test.
* gcc.target/pru/loop-short-1.c: New test.
* gcc.target/pru/loop-short-2.c: New test.
* gcc.target/pru/loop-si-1.c: New test.
* gcc.target/pru/loop-si-2.c: New test.
* gcc.target/pru/loop-u8_pcrel_overflow.c: New test.
* gcc.target/pru/loop-ubyte-1.c: New test.
* gcc.target/pru/loop-ubyte-2.c: New test.
* gcc.target/pru/lra-framepointer-fragmentation-1.c: New test.
* gcc.target/pru/lra-framepointer-fragmentation-2.c: New test.
* gcc.target/pru/mabi-ti-1.c: New test.
* gcc.target/pru/mabi-ti-2.c: New test.
* gcc.target/pru/mabi-ti-3.c: New test.
* gcc.target/pru/mabi-ti-4.c: New test.
* gcc.target/pru/mabi-ti-5.c: New test.
* gcc.target/pru/mabi-ti-6.c: New test.
* gcc.target/pru/mabi-ti-7.c: New test.
* gcc.target/pru/pr64366.c: New test.
* gcc.target/pru/pragma-ctable_entry.c: New test.
* gcc.target/pru/pru.exp: New file.
* gcc.target/pru/qbbc-1.c: New test.
* gcc.target/pru/qbbc-2.c: New test.
* gcc.target/pru/qbbc-3.c: New test.
* gcc.target/pru/qbbs-1.c: New test.
* gcc.target/pru/qbbs-2.c: New test.
* gcc.target/pru/setbit.c: New test.
* gcc.target/pru/zero_extend-and-hisi.c: New test.
* gcc.target/pru/zero_extend-and-qihi.c: New test.
* gcc.target/pru/zero_extend-and-qisi.c: New test.
* gcc.target/pru/zero_extend-hisi.c: New test.
* gcc.target/pru/zero_extend-qihi.c: New test.
* gcc.target/pru/zero_extend-qisi.c: New test.
* lib/target-supports.exp: Add PRU to feature filters.

From-SVN: r272205

5 years agoInitial TI PRU libgcc port
Dimitar Dimitrov [Wed, 12 Jun 2019 19:28:51 +0000 (19:28 +0000)]
Initial TI PRU libgcc port

libgcc/ChangeLog:

2019-06-12  Dimitar Dimitrov  <dimitar@dinux.eu>

* config.host: Add PRU target.
* config/pru/asri.c: New file.
* config/pru/eqd.c: New file.
* config/pru/eqf.c: New file.
* config/pru/ged.c: New file.
* config/pru/gef.c: New file.
* config/pru/gtd.c: New file.
* config/pru/gtf.c: New file.
* config/pru/led.c: New file.
* config/pru/lef.c: New file.
* config/pru/lib2bitcountHI.c: New file.
* config/pru/lib2divHI.c: New file.
* config/pru/lib2divQI.c: New file.
* config/pru/lib2divSI.c: New file.
* config/pru/libgcc-eabi.ver: New file.
* config/pru/ltd.c: New file.
* config/pru/ltf.c: New file.
* config/pru/mpyll.S: New file.
* config/pru/pru-abi.h: New file.
* config/pru/pru-asm.h: New file.
* config/pru/pru-divmod.h: New file.
* config/pru/sfp-machine.h: New file.
* config/pru/t-pru: New file.

From-SVN: r272204

5 years agoFix ChangeLog dates for last commit
Dimitar Dimitrov [Wed, 12 Jun 2019 19:26:31 +0000 (19:26 +0000)]
Fix ChangeLog dates for last commit

From-SVN: r272203

5 years agoInitial TI PRU GCC port
Dimitar Dimitrov [Wed, 12 Jun 2019 19:04:24 +0000 (19:04 +0000)]
Initial TI PRU GCC port

ChangeLog:

2019-06-19  Dimitar Dimitrov  <dimitar@dinux.eu>

* configure: Regenerate.
* configure.ac: Add PRU target.

gcc/ChangeLog:

2019-06-19  Dimitar Dimitrov  <dimitar@dinux.eu>

* common/config/pru/pru-common.c: New file.
* config.gcc: Add PRU target.
* config/pru/alu-zext.md: New file.
* config/pru/constraints.md: New file.
* config/pru/predicates.md: New file.
* config/pru/pru-opts.h: New file.
* config/pru/pru-passes.c: New file.
* config/pru/pru-pragma.c: New file.
* config/pru/pru-protos.h: New file.
* config/pru/pru.c: New file.
* config/pru/pru.h: New file.
* config/pru/pru.md: New file.
* config/pru/pru.opt: New file.
* config/pru/t-pru: New file.
* doc/extend.texi: Document PRU pragmas.
* doc/invoke.texi: Document PRU-specific options.
* doc/md.texi: Document PRU asm constraints.

From-SVN: r272202

5 years agore PR fortran/90002 (ICE: free_expr0(): Bad expr type)
Steven G. Kargl [Wed, 12 Jun 2019 18:28:32 +0000 (18:28 +0000)]
re PR fortran/90002 (ICE: free_expr0(): Bad expr type)

2019-06-12  Steven G. Kargl  <kargl@gcc.gnu.org>

PR fortran/90002
* array.c (gfc_free_array_spec): When freeing an array-spec, avoid
an ICE for assumed-shape coarrays

2019-06-12  Steven G. Kargl  <kargl@gcc.gnu.org>

PR fortran/90002
* gfortran.dg/pr90002.f90: New test.

From-SVN: r272201

5 years agoUpdate last ChangeLog entry.
Martin Sebor [Wed, 12 Jun 2019 17:27:41 +0000 (11:27 -0600)]
Update last ChangeLog entry.

From-SVN: r272200

5 years agoPR middle-end/90676 - default GIMPLE dumps lack information
Martin Sebor [Wed, 12 Jun 2019 17:05:35 +0000 (17:05 +0000)]
PR middle-end/90676 - default GIMPLE dumps lack information

gcc/ChangeLog:

PR middle-end/90676
* tree-pretty-print.c (dump_mem_ref): New function.  Include
MEM_REF type in output when different size than operand.
(dump_generic_node): Move code to dump_mem_ref and call it.

gcc/testsuite/ChangeLog:

PR middle-end/90676
* gcc.dg/tree-ssa/dump-6.c: New test.
* g++.dg/tree-ssa/pr19807.C: Adjust expected output.
* g++.dg/tree-ssa/ssa-dse-1.C: Same.
* gcc.dg/store_merging_5.c: Same.
* gcc.dg/tree-prof/stringop-2.c: Same.
* gcc.dg/tree-ssa/pr30375.c: Same.
* gcc.dg/tree-ssa/slsr-27.c: Same.
* gcc.dg/tree-ssa/slsr-28.c: Same.
* gcc.dg/tree-ssa/slsr-29.c: Same.
* gcc.dg/tree-ssa/ssa-dse-24.c: Same.

From-SVN: r272199

5 years agoPR tree-optimization/90662 - strlen of a string in a vla plus offset not folded
Martin Sebor [Wed, 12 Jun 2019 16:33:04 +0000 (16:33 +0000)]
PR tree-optimization/90662 - strlen of a string in a vla plus offset not folded

gcc/ChangeLog:

PR tree-optimization/90662
* tree-ssa-strlen.c (get_stridx): Handle simple VLAs and pointers
to arrays.

gcc/testsuite/ChangeLog:

PR tree-optimization/90662
* gcc.dg/strlenopt-62.c: New test.
* gcc.dg/strlenopt-63.c: New test.
* gcc.dg/strlenopt-64.c: New test.

From-SVN: r272197

5 years ago[nvptx] Assert fork has at most one join in nvptx_find_par
Tom de Vries [Wed, 12 Jun 2019 14:56:18 +0000 (14:56 +0000)]
[nvptx] Assert fork has at most one join in nvptx_find_par

With the test-case of PR90009 we run into a SIGSEGV in nvptx_neuter_pars,
because par->join_insn and par->join_block are NULL.

Detect this problem earlier, by adding an assert in nvptx_find_par that
asserts that a fork cannot be paired with more than one join.

Build and tested on x86_64 with nvptx accelerator enabled.

2019-06-12  Tom de Vries  <tdevries@suse.de>

PR tree-optimization/90009
* config/nvptx/nvptx.c (nvptx_find_par): Assert fork has at most join.

From-SVN: r272189

5 years agoImprove static_assert messages for std::variant
Jonathan Wakely [Wed, 12 Jun 2019 14:52:09 +0000 (15:52 +0100)]
Improve static_assert messages for std::variant

Also fix a warning with -Wunused-parameter -Wsystem-headers.

* include/std/variant (get<T>, get<N>, get_if<N>, get_if<T>)
(variant::emplace): Change static_assert messages from "should be"
to "must be".
(hash<monostate>::operator()): Remove name of unused parameter.

From-SVN: r272188

5 years agoSimplify std::scoped_lock destructor
Jonathan Wakely [Wed, 12 Jun 2019 14:52:06 +0000 (15:52 +0100)]
Simplify std::scoped_lock destructor

* include/std/mutex (scoped_lock::~scoped_lock()): Use fold
expression.

From-SVN: r272187

5 years agoReplace std::to_string for integers with optimized version
Jonathan Wakely [Wed, 12 Jun 2019 14:52:02 +0000 (15:52 +0100)]
Replace std::to_string for integers with optimized version

The std::to_chars functions from C++17 can be used to implement
std::to_string with much better performance than calling snprintf. Only
the __detail::__to_chars_len and __detail::__to_chars_10 functions are
needed for to_string, because it always outputs base 10 representations.

The return type of __detail::__to_chars_10 should not be declared before
C++17, so the function body is extracted into a new function that can be
reused by to_string and __detail::__to_chars_10.

The existing tests for to_chars rely on to_string to check for correct
answers. Now that they use the same code that doesn't actually ensure
correctness, so add new tests for std::to_string that compare against
printf output.

* include/Makefile.am: Add new <bits/charconv.h> header.
* include/Makefile.in: Regenerate.
* include/bits/basic_string.h (to_string(int), to_string(unsigned))
(to_string(long), to_string(unsigned long), to_string(long long))
(to_string(unsigned long long)): Rewrite to use __to_chars_10_impl.
* include/bits/charconv.h: New header.
(__detail::__to_chars_len): Move here from <charconv>.
(__detail::__to_chars_10_impl): New function extracted from
__detail::__to_chars_10.
* include/std/charconv (__cpp_lib_to_chars): Add, but comment out.
(__to_chars_unsigned_type): New class template that reuses
__make_unsigned_selector_base::__select to pick a type.
(__unsigned_least_t): Redefine as __to_chars_unsigned_type<T>::type.
(__detail::__to_chars_len): Move to new header.
(__detail::__to_chars_10): Add inline specifier. Move code doing the
output to __detail::__to_chars_10_impl and call that.
* include/std/version (__cpp_lib_to_chars): Add, but comment out.
* testsuite/21_strings/basic_string/numeric_conversions/char/
to_string.cc: Fix reference in comment. Remove unused variable.
* testsuite/21_strings/basic_string/numeric_conversions/char/
to_string_int.cc: New test.

From-SVN: r272186

5 years agoDisable hash-table sanitization for mem stats maps.
Martin Liska [Wed, 12 Jun 2019 13:07:07 +0000 (15:07 +0200)]
Disable hash-table sanitization for mem stats maps.

2019-06-12  Martin Liska  <mliska@suse.cz>

* ggc-common.c (ggc_prune_overhead_list): Do not sanitize
the created map.
* hash-map.h: Add sanitize_eq_and_hash into ::hash_map.
* mem-stats.h (mem_alloc_description::mem_alloc_description):
Do not sanitize created maps.

From-SVN: r272183

5 years agoRemove value_range_constant_singleton in favor of value_range::singleton_p.
Aldy Hernandez [Wed, 12 Jun 2019 12:53:20 +0000 (12:53 +0000)]
Remove value_range_constant_singleton in favor of value_range::singleton_p.

From-SVN: r272182

5 years agore PR target/90811 ([nvptx] ptxas error on OpenMP offloaded code)
Jakub Jelinek [Wed, 12 Jun 2019 09:31:26 +0000 (11:31 +0200)]
re PR target/90811 ([nvptx] ptxas error on OpenMP offloaded code)

PR target/90811
* cfgexpand.c (align_local_variable): Add really_expand argument,
don't SET_DECL_ALIGN if it is false.
(add_stack_var): Add really_expand argument, pass it through to
align_local_variable.
(expand_one_stack_var_1): Pass true as really_expand to
align_local_variable.
(expand_one_ssa_partition): Pass true as really_expand to
add_stack_var.
(expand_one_var): Pass really_expand through to add_stack_var.

From-SVN: r272181

5 years ago[arm] Implement usadv16qi and ssadv16qi standard names
Przemyslaw Wirkus [Wed, 12 Jun 2019 08:27:59 +0000 (08:27 +0000)]
[arm] Implement usadv16qi and ssadv16qi standard names

This patch implements the usadv16qi and ssadv16qi standard names for arm.

The V16QImode variant is important as it is the most commonly used pattern:
reducing vectors of bytes into an int.
The midend expects the optab to compute the absolute differences of operands 1
and 2 and reduce them while widening along the way up to SImode. So the inputs
are V16QImode and the output is V4SImode.

I've based my solution on Aarch64 usadv16qi and ssadv16qi standard names
current implementation (r260437). This solution emits below sequence of
instructions:

        VABDL.u8        tmp, op1, op2   # op1, op2 lowpart
        VABAL.u8        tmp, op1, op2   # op1, op2 highpart
        VPADAL.u16      op3, tmp

So, for the code:

$ arm-none-linux-gnueabihf-gcc -S -O3 -march=armv8-a+simd -mfpu=auto -mfloat-abi=hard usadv16qi.c -dp

#define N 1024
unsigned char pix1[N];
unsigned char pix2[N];

int
foo (void)
{
  int i_sum = 0;
  int i;
  for (i = 0; i < N; i++)
    i_sum += __builtin_abs (pix1[i] - pix2[i]);
  return i_sum;
}

we now generate on arm:
foo:
        movw    r3, #:lower16:pix2      @ 57    [c=4 l=4]  *arm_movsi_vfp/3
        movt    r3, #:upper16:pix2      @ 58    [c=4 l=4]  *arm_movt/0
        vmov.i32        q9, #0  @ v4si  @ 3     [c=4 l=4]  *neon_movv4si/2
        movw    r2, #:lower16:pix1      @ 59    [c=4 l=4]  *arm_movsi_vfp/3
        movt    r2, #:upper16:pix1      @ 60    [c=4 l=4]  *arm_movt/0
        add     r1, r3, #1024   @ 8     [c=4 l=4]  *arm_addsi3/4
.L2:
        vld1.8  {q11}, [r3]!    @ 11    [c=8 l=4]  *movmisalignv16qi_neon_load
        vld1.8  {q10}, [r2]!    @ 10    [c=8 l=4]  *movmisalignv16qi_neon_load
        cmp     r1, r3  @ 21    [c=4 l=4]  *arm_cmpsi_insn/2
        vabdl.u8        q8, d20, d22    @ 12    [c=8 l=4]  neon_vabdluv8qi
        vabal.u8        q8, d21, d23    @ 15    [c=88 l=4]  neon_vabaluv8qi
        vpadal.u16      q9, q8  @ 16    [c=8 l=4]  neon_vpadaluv8hi
        bne     .L2             @ 22    [c=16 l=4]  arm_cond_branch
        vadd.i32        d18, d18, d19   @ 24    [c=120 l=4]  quad_halves_plusv4si
        vpadd.i32       d18, d18, d18   @ 25    [c=8 l=4]  neon_vpadd_internalv2si
        vmov.32 r0, d18[0]      @ 30    [c=12 l=4]  vec_extractv2sisi/1

instead of:
foo:
        @ args = 0, pretend = 0, frame = 0
        @ frame_needed = 0, uses_anonymous_args = 0
        @ link register save eliminated.
        movw    r3, #:lower16:pix1
        movt    r3, #:upper16:pix1
        vmov.i32        q9, #0  @ v4si
        movw    r2, #:lower16:pix2
        movt    r2, #:upper16:pix2
        add     r1, r3, #1024
.L2:
        vld1.8  {q8}, [r3]!
        vld1.8  {q11}, [r2]!
        vmovl.u8 q10, d16
        cmp     r1, r3
        vmovl.u8 q8, d17
        vmovl.u8 q12, d22
        vmovl.u8 q11, d23
        vsub.i16        q10, q10, q12
        vsub.i16        q8, q8, q11
        vabs.s16        q10, q10
        vabs.s16        q8, q8
        vaddw.s16       q9, q9, d20
        vaddw.s16       q9, q9, d21
        vaddw.s16       q9, q9, d16
        vaddw.s16       q9, q9, d17
        bne     .L2
        vadd.i32        d18, d18, d19
        vpadd.i32       d18, d18, d18
        vmov.32 r0, d18[0]

2019-06-12  Przemyslaw Wirkus  <przemyslaw.wirkus@arm.com>

        * config/arm/iterators.md (VABAL): New int iterator.
        * config/arm/neon.md (<sup>sadv16qi): New define_expand.
        * config/arm/unspecs.md ("unspec"): Define UNSPEC_VABAL_S, UNSPEC_VABAL_U
        values.

        * gcc.target/arm/ssadv16qi.c: New test.
        * gcc.target/arm/usadv16qi.c: Likewise.

From-SVN: r272180

5 years agoRemove wrong assert about single value profiler.
Martin Liska [Wed, 12 Jun 2019 07:17:42 +0000 (09:17 +0200)]
Remove wrong assert about single value profiler.

2019-06-12  Martin Liska  <mliska@suse.cz>

* value-prof.c (stream_out_histogram_value): Only first value
can't be negative.

From-SVN: r272179

5 years agore PR c/90760 (ICE on attributes section and alias in set_section, at symtab.c:1573)
Jakub Jelinek [Wed, 12 Jun 2019 07:14:08 +0000 (09:14 +0200)]
re PR c/90760 (ICE on attributes section and alias in set_section, at symtab.c:1573)

PR c/90760
* symtab.c (symtab_node::set_section): Allow being called on aliases
as long as they aren't analyzed yet.

* gcc.dg/pr90760.c: New test.

From-SVN: r272178

5 years agoDaily bump.
GCC Administrator [Wed, 12 Jun 2019 00:16:15 +0000 (00:16 +0000)]
Daily bump.

From-SVN: r272177

5 years agoremoved extra .com, fixed e-mail.
Matthew Beliveau [Tue, 11 Jun 2019 20:15:43 +0000 (20:15 +0000)]
removed extra .com, fixed e-mail.

From-SVN: r272170

5 years agomips.c (mips_final_postscan_insn): Modify call to `mips_set_text_contents_type' to...
Faraz Shahbazker [Tue, 11 Jun 2019 20:09:57 +0000 (20:09 +0000)]
mips.c (mips_final_postscan_insn): Modify call to `mips_set_text_contents_type' to indicate whether a non-debug...

* config/mips/mips.c (mips_final_postscan_insn): Modify call
to `mips_set_text_contents_type' to indicate whether a
non-debug insn follows.

* gcc.target/mips/data-sym-pool.c: Update expected output.
* gcc.target/mips/data-sym-multi-pool.c: New test.

From-SVN: r272169

5 years ago* MAINTAINERS (Write After Approval): Add myself.
Matthew Beliveau [Tue, 11 Jun 2019 19:31:36 +0000 (19:31 +0000)]
* MAINTAINERS (Write After Approval): Add myself.

From-SVN: r272168

5 years agoDarwin, testsuite - Fix PR 65364 (uninit-19.c).
Iain Sandoe [Tue, 11 Jun 2019 18:22:58 +0000 (18:22 +0000)]
Darwin, testsuite - Fix PR 65364 (uninit-19.c).

This test currently fails on Darwin, because the port inlines fn2 for
both PIC (and non-pic for m32).  Fixed by adjusting the target condition.

2019-06-11  Iain Sandoe  <iain@sandoe.co.uk>

PR testsuite/65364
* gcc.dg/uninit-19.c (fn1): Adjust target condition for Darwin.
(fn2): Likewise.

From-SVN: r272167

5 years agoDo not enable -mpcrel by default at the current time
Michael Meissner [Tue, 11 Jun 2019 17:36:58 +0000 (17:36 +0000)]
Do not enable -mpcrel by default at the current time

From-SVN: r272165

5 years agore PR target/90811 ([nvptx] ptxas error on OpenMP offloaded code)
Jakub Jelinek [Tue, 11 Jun 2019 16:40:10 +0000 (18:40 +0200)]
re PR target/90811 ([nvptx] ptxas error on OpenMP offloaded code)

PR target/90811
* config/nvptx/nvptx.c (nvptx_output_softstack_switch): Use and.b%d
instead of and.u%d.

* testsuite/libgomp.c/pr90811.c: New test.

From-SVN: r272161

5 years agoFix ConstexprIterator requirements tests - No constexpr algorithms!
Edward Smith-Rowland [Tue, 11 Jun 2019 16:29:35 +0000 (16:29 +0000)]
Fix ConstexprIterator requirements tests - No constexpr algorithms!

2019-06-09  Edward Smith-Rowland  <3dw4rd@verizon.net>

Fix ConstexprIterator requirements tests - No constexpr algorithms!
* testsuite/21_strings/basic_string_view/requirements/constexpr_iter.cc:
Replace copy with hand-rolled loop.
* testsuite/23_containers/array/requirements/constexpr_iter.cc:
Ditto.

From-SVN: r272159

5 years agoAllow conversions in X/[ex]4 < Y/[ex]4
Marc Glisse [Tue, 11 Jun 2019 15:22:43 +0000 (17:22 +0200)]
Allow conversions in X/[ex]4 < Y/[ex]4

2019-06-11  Marc Glisse  <marc.glisse@inria.fr>

gcc/
* match.pd (X/[ex]4<Y/[ex]4): Handle conversions.

gcc/testsuite/
* gcc.dg/tree-ssa/cmpexactdiv-5.c: New file.

From-SVN: r272158

5 years agoPR c++/90449 - add -Winaccessible-base option.
Matthew Beliveau [Tue, 11 Jun 2019 15:03:21 +0000 (15:03 +0000)]
PR c++/90449 - add -Winaccessible-base option.

       * doc/invoke.texi (Winaccessible-base): Document.

       * c.opt (Winaccessible-base): Added new option.

       * class.c (warn_about_ambiguous_bases): Changed name to:
       maybe_warn_about_inaccessible_bases.
       (maybe_warn_about_inaccessible_bases):  Implemented new
       Winaccessible-base warning option for both direct and virtual
       base warnings.
       (layout_class_type): Call to warn_about_ambiguous_bases changed to fit
       new name.

        * g++.dg/warn/Winaccessible-base-1.C: New file.
        * g++.dg/warn/Winaccessible-base-2.C: New file.
        * g++.dg/warn/Winaccessible-virtual-base-1.C: New file.
        * g++.dg/warn/Winaccessible-virtual-base-2.C: New file.

From-SVN: r272157

5 years agore PR c++/90801 (A recurring hang)
Richard Biener [Tue, 11 Jun 2019 14:03:41 +0000 (14:03 +0000)]
re PR c++/90801 (A recurring hang)

2019-06-11  Richard Biener  <rguenther@suse.de>

PR c++/90801
* typeck2.c (split_nonconstant_init_1): Avoid ordered remove
from CONSTRUCTOR by marking to remove elements and doing all
of them in a O(n) scan.

From-SVN: r272156

5 years agore PR c++/90810 (Different intrinsics behavior on G++ 7.4 and G++ 8.3)
Jakub Jelinek [Tue, 11 Jun 2019 13:01:40 +0000 (15:01 +0200)]
re PR c++/90810 (Different intrinsics behavior on G++ 7.4 and G++ 8.3)

PR c++/90810
* init.c (constant_value_1): Handle VECTOR_CST DECL_INITIAL for
!DECL_INITIALIZED_BY_CONSTANT_EXPRESSION_P decls like CONSTRUCTOR.

* g++.dg/ext/vector37.C: New test.

From-SVN: r272152

5 years agofold_real_zero_addition_p for vectors
Marc Glisse [Tue, 11 Jun 2019 12:48:26 +0000 (14:48 +0200)]
fold_real_zero_addition_p for vectors

2019-06-11  Marc Glisse  <marc.glisse@inria.fr>

PR tree-optimization/62041
* fold-const.c (fold_real_zero_addition_p): Handle vectors.

From-SVN: r272151

5 years agogdbhooks.py (TreePrinter.to_string): Recognize ggc_free'd memory.
Jason Merrill [Tue, 11 Jun 2019 12:40:08 +0000 (08:40 -0400)]
gdbhooks.py (TreePrinter.to_string): Recognize ggc_free'd memory.

* gdbhooks.py (TreePrinter.to_string): Recognize ggc_free'd memory.

* tree.c (get_tree_code_name): Likewise.
* print-tree.c (print_node): Only briefly print a node with an
invalid code.

From-SVN: r272150

5 years agoImport these changes from the binutils/gdb repository:
Nick Clifton [Tue, 11 Jun 2019 12:05:49 +0000 (12:05 +0000)]
Import these changes from the binutils/gdb repository:

2019-05-28  Nick Alcock  <nick.alcock@oracle.com>

* Makefile.def (dependencies): configure-libctf depends on all-bfd
and all its deps.
* Makefile.in: Regenerated.

2019-05-28  Nick Alcock  <nick.alcock@oracle.com>

* Makefile.def (host_modules): Add libctf.
* Makefile.def (dependencies): Likewise.
libctf depends on zlib, libiberty, and bfd.
* Makefile.in: Regenerated.
* configure.ac (host_libs): Add libctf.
* configure: Regenerated.

From-SVN: r272148

5 years agotoir.cc (pop_label): Only queue labels in a vector.
Richard Biener [Tue, 11 Jun 2019 08:30:47 +0000 (08:30 +0000)]
toir.cc (pop_label): Only queue labels in a vector.

2019-06-11  Richard Biener  <rguenthe@suse.de>

d/90778
* toir.cc (pop_label): Only queue labels in a vector.
(cmp_labels): Label decl comparator.
(pop_binding_level): Pop labels in DECL_UID order to avoid
debug info differences.

From-SVN: r272146

5 years agore PR bootstrap/90819 (-Walloc-size-larger-than error in trans-mem.c)
Jakub Jelinek [Tue, 11 Jun 2019 08:20:02 +0000 (10:20 +0200)]
re PR bootstrap/90819 (-Walloc-size-larger-than error in trans-mem.c)

PR bootstrap/90819
* trans-mem.c (tm_memopt_compute_available): Add assertion
that blocks is not empty.  Formatting fix.

From-SVN: r272145

5 years agoDisable htable sanitization in pt.c (PR c++/87847).
Martin Liska [Tue, 11 Jun 2019 07:55:19 +0000 (09:55 +0200)]
Disable htable sanitization in pt.c (PR c++/87847).

2019-06-11  Martin Liska  <mliska@suse.cz>

PR c++/87847
* hash-table.h: Extend create_gcc, add one parameter
that is passed into hash_table::hash_table.
2019-06-11  Martin Liska  <mliska@suse.cz>

PR c++/87847
* pt.c (init_template_processing): Disable hash table
sanitization for decl_specializations and type_specializations.

From-SVN: r272144

5 years ago* libgcov-merge.c (__gcov_merge_single): Revert previous change.
Jakub Jelinek [Tue, 11 Jun 2019 07:54:17 +0000 (09:54 +0200)]
* libgcov-merge.c (__gcov_merge_single): Revert previous change.

From-SVN: r272143

5 years agoDaily bump.
GCC Administrator [Tue, 11 Jun 2019 00:16:16 +0000 (00:16 +0000)]
Daily bump.

From-SVN: r272141

5 years ago* config/i386/i386.md (absneg fp_reg splitter): Fix operator mode.
Uros Bizjak [Mon, 10 Jun 2019 21:51:42 +0000 (23:51 +0200)]
* config/i386/i386.md (absneg fp_reg splitter): Fix operator mode.

From-SVN: r272134

5 years agocompiler: support inlining functions that use index expressions
Ian Lance Taylor [Mon, 10 Jun 2019 21:34:12 +0000 (21:34 +0000)]
compiler: support inlining functions that use index expressions

    Also move the determine_types pass on an inlined function body to one
    place, rather than doing it ad hoc as needed.

    This adds 79 new inlinable functions in the standard library, such as
    bytes.HasPrefix and bytes.LastIndexByte.

    Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/181261

From-SVN: r272133

5 years agocompiler: make heap expression's write barrier conditional
Ian Lance Taylor [Mon, 10 Jun 2019 21:08:42 +0000 (21:08 +0000)]
compiler: make heap expression's write barrier conditional

    Heap_expression::do_get_backend emits an unconditional write
    barrier if the type has pointers and it is not a stack allocation.
    This CL changes it to use a write barrier for the assignment only
    when write barriers are enabled. While here, also change it to
    call gcWriteBarrier instead of typedmemmove for pointer-shaped
    types.

    For this to work, Function::build needs to be adjusted so that
    Heap_expression::do_get_backend is called when there is a parent
    block.

    Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/181540

From-SVN: r272132

5 years agocompiler: permit inlining functions with labels and goto statements
Ian Lance Taylor [Mon, 10 Jun 2019 21:02:20 +0000 (21:02 +0000)]
compiler: permit inlining functions with labels and goto statements

    This permits inlining functions with for loops and some switches, as
    they are lowered to if and goto statements before exporting them.

    This by itself only adds three new inlinable functions in the standard
    library: sort.Search, context.(*emptyCtx).String, and
    cmd/go/internal/work.(*Builder).disableBuildID.

    Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/181197

From-SVN: r272131

5 years agocompiler: use gcWriteBarrier for pointer-shaped struct/array
Ian Lance Taylor [Mon, 10 Jun 2019 20:35:14 +0000 (20:35 +0000)]
compiler: use gcWriteBarrier for pointer-shaped struct/array

    If a struct/array is pointer-shaped (i.e. having a single field
    that is pointer-shaped), we can use gcWriteBarrier instead of
    typedmemmove for the write barrier.

    Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/181539

From-SVN: r272130