]> git.ipfire.org Git - thirdparty/gcc.git/log
thirdparty/gcc.git
3 years ago[Ada] Extend optimization to True/False prefixed with Standard
Piotr Trojanek [Fri, 15 May 2020 18:27:18 +0000 (20:27 +0200)] 
[Ada] Extend optimization to True/False prefixed with Standard

gcc/ada/

* exp_ch5.adb (Expand_N_If_Statement): Detect True/False
prefixed with Standard.

3 years ago[Ada] Check predicates for subtypes of private types
Bob Duff [Sat, 16 May 2020 15:59:50 +0000 (11:59 -0400)] 
[Ada] Check predicates for subtypes of private types

gcc/ada/

* sem_ch13.adb (Analyze_Aspect_Specifications): Add freeze node
for the Underlying_Full_View if it exists. The freeze node is
what triggers the generation of the predicate function.
* freeze.adb: Minor reformatting.

3 years ago[Ada] Small tweak to Narrow_Large_Operation procedure
Eric Botcazou [Sat, 16 May 2020 13:42:04 +0000 (15:42 +0200)] 
[Ada] Small tweak to Narrow_Large_Operation procedure

gcc/ada/

* exp_ch4.adb (Narrow_Large_Operation): Use the base type instead
of the first subtype of standard integer types as narrower type.

3 years agoexpr: Fix REDUCE_BIT_FIELD for constants [PR95694]
Richard Sandiford [Wed, 8 Jul 2020 14:01:14 +0000 (15:01 +0100)] 
expr: Fix REDUCE_BIT_FIELD for constants [PR95694]

This is yet another PR caused by constant integer rtxes not storing
a mode.  We were calling REDUCE_BIT_FIELD on a constant integer that
didn't fit in poly_int64, and then tripped the as_a<scalar_int_mode>
assert on VOIDmode.

AFAICT REDUCE_BIT_FIELD is always passed rtxes that have TYPE_MODE
(rather than some other mode) and it just fills in the redundant
sign bits of that TYPE_MODE value.  So it should be safe to get
the mode from the type instead of the rtx.  The patch does that
and asserts that the modes agree, where information is available.

That on its own is enough to fix the bug, but we might as well
extend the folding case to all constant integers, not just those
that fit poly_int64.

gcc/
PR middle-end/95694
* expr.c (expand_expr_real_2): Get the mode from the type rather
than the rtx, and assert that it is consistent with the mode of
the rtx (where known).  Optimize all constant integers, not just
those that can be represented in poly_int64.

gcc/testsuite/
PR middle-end/95694
* gcc.dg/pr95694.c: New test.

3 years agoc++: Better diagnostic for decltype(auto) in C++11 [PR96103]
Marek Polacek [Tue, 7 Jul 2020 21:09:42 +0000 (17:09 -0400)] 
c++: Better diagnostic for decltype(auto) in C++11 [PR96103]

If you try to use decltype(auto) in C++11, we emit obscure

  error: expected primary-expression before 'auto'

giving the user no hint as to what's wrong.  This patch improves that
diagnostic.  Since we've been giving an error, I'm also using error().

gcc/cp/ChangeLog:

PR c++/96103
* parser.c (cp_parser_decltype): Print error about using decltype(auto)
in C++11.  Check that the token following "auto" is ")".

gcc/testsuite/ChangeLog:

PR c++/96103
* g++.dg/cpp0x/decltype77.C: New test.

3 years agors6000: Add len_load/len_store optab support
Kewen Lin [Wed, 8 Jul 2020 07:44:02 +0000 (02:44 -0500)] 
rs6000: Add len_load/len_store optab support

Define rs6000 specific len_load/len_store for the LEN_LOAD/LEN_STORE
internal function's expansion.  As the doc description of the
len_load/len_store optab shows, we should use V16QI to wrap those
available vector modes.

gcc/ChangeLog:

* config/rs6000/vsx.md (len_load_v16qi): New define_expand.
(len_store_v16qi): Likewise.

3 years agoIFN/optabs: Support vector load/store with length
Kewen Lin [Wed, 8 Jul 2020 07:33:03 +0000 (02:33 -0500)] 
IFN/optabs: Support vector load/store with length

This patch is to add the internal function and optabs support for
vector load/store with length.

For the vector load/store with length optab, the length item would
be measured in lanes by default.  For the targets which support
length measured in bytes like Power, they should only define VnQI
modes to wrap the other same size vector modes.  If the length is
larger than total lane/byte count of the given mode, the behavior
is undefined.  For the remaining lanes/bytes which isn't specified
by length, they would be taken as undefined value.

gcc/ChangeLog:

* doc/md.texi (len_load_@var{m}): Document.
(len_store_@var{m}): Likewise.
* internal-fn.c (len_load_direct): New macro.
(len_store_direct): Likewise.
(expand_len_load_optab_fn): Likewise.
(expand_len_store_optab_fn): Likewise.
(direct_len_load_optab_supported_p): Likewise.
(direct_len_store_optab_supported_p): Likewise.
(expand_mask_load_optab_fn): New macro.  Original renamed to ...
(expand_partial_load_optab_fn): ... here.  Add handlings for
len_load_optab.
(expand_mask_store_optab_fn): New macro.  Original renamed to ...
(expand_partial_store_optab_fn): ... here. Add handlings for
len_store_optab.
(internal_load_fn_p): Handle IFN_LEN_LOAD.
(internal_store_fn_p): Handle IFN_LEN_STORE.
(internal_fn_stored_value_index): Handle IFN_LEN_STORE.
* internal-fn.def (LEN_LOAD): New internal function.
(LEN_STORE): Likewise.
* optabs.def (len_load_optab, len_store_optab): New optab.

3 years agolibstdc++: Fix [multi]map/[multi]set move constructors noexcept qualification
François Dumont [Fri, 3 Jul 2020 06:13:19 +0000 (08:13 +0200)] 
libstdc++: Fix [multi]map/[multi]set move constructors noexcept qualification

Container move constructors shall not consider their allocator move
constructor qualification.

libstdc++-v3/ChangeLog:

* include/bits/stl_tree.h (_Rb_tree_impl(_Rb_tree_impl&&)): Add noexcept
qualification based only on _Compare one.
* testsuite/23_containers/map/cons/noexcept_move_construct.cc: Add
static asserts.
* testsuite/23_containers/multimap/cons/noexcept_move_construct.cc:
Likewise.
* testsuite/23_containers/multiset/cons/noexcept_move_construct.cc:
Likewise.
* testsuite/23_containers/set/cons/noexcept_move_construct.cc: Likewise.

3 years agolibbacktrace: test linker support for DWARF 5
Ian Lance Taylor [Wed, 8 Jul 2020 03:42:21 +0000 (20:42 -0700)] 
libbacktrace: test linker support for DWARF 5

On AIX, the compiler supports DWARF 5 but the linker does not.

2020-07-07 Clement Chigot <clement.chigot@atos.net>

* configure.ac: Test linker support for DWARF5
* configure: Regenerate

3 years agoDaily bump.
GCC Administrator [Wed, 8 Jul 2020 00:16:40 +0000 (00:16 +0000)] 
Daily bump.

3 years agoc++: Add new test [PR92427]
Marek Polacek [Tue, 7 Jul 2020 23:28:04 +0000 (19:28 -0400)] 
c++: Add new test [PR92427]

Fixed in r10-5578.

PR c++/92427
* g++.dg/ext/flexary37.C: New test.

3 years agolibstdc++: Whitespace changes to remove spaces before tabs
Jonathan Wakely [Tue, 7 Jul 2020 22:39:44 +0000 (23:39 +0100)] 
libstdc++: Whitespace changes to remove spaces before tabs

libstdc++-v3/ChangeLog:

* include/std/limits: Whitespace changes.

3 years agolibstdc++: Replace __int_limits with __numeric_traits_integer
Jonathan Wakely [Tue, 7 Jul 2020 22:26:38 +0000 (23:26 +0100)] 
libstdc++: Replace __int_limits with __numeric_traits_integer

I recently added std::__detail::__int_limits as a lightweight
alternative to std::numeric_limits, forgetting that the values it
provides (digits, min and max) are already provided by
__gnu_cxx::__numeric_traits.

This change adds __int_traits as an alias for __numeric_traits_integer.
This avoids instantiating __numeric_traits to decide whether to use
__numeric_traits_integer or __numeric_traits_floating. Then all uses of
__int_limits can be replaced with __int_traits, and __int_limits can be
removed.

libstdc++-v3/ChangeLog:

* include/Makefile.am: Remove bits/int_limits.h.
* include/Makefile.in: Regenerate.
* include/bits/int_limits.h: Removed.
* include/bits/parse_numbers.h (_Select_int_base): Replace
__int_limits with __int_traits.
* include/bits/range_access.h (_SSize::operator()): Likewise.
* include/ext/numeric_traits.h (__numeric_traits_integer): Add
static assertion.
(__int_traits): New alias template.
* include/std/bit (__rotl, __rotr, __countl_zero, __countl_one)
(__countr_zero, __countr_one, __popcount, __bit_ceil)
(__bit_floor, __bit_width) Replace __int_limits with
__int_traits.
* include/std/charconv (__to_chars_8, __from_chars_binary)
(__from_chars_alpha_to_num, from_chars): Likewise.
* include/std/memory_resource (polymorphic_allocator::allocate)
(polymorphic_allocator::allocate_object): Likewise.
* include/std/string_view (basic_string_view::_S_compare):
Likewise.
* include/std/utility (cmp_equal, cmp_less, in_range): Likewise.

3 years agolibstdc++: Whitespace changes to keep lines below 80 columns
Jonathan Wakely [Tue, 7 Jul 2020 22:26:38 +0000 (23:26 +0100)] 
libstdc++: Whitespace changes to keep lines below 80 columns

Also remove leading spaces that are followed by tabs.

libstdc++-v3/ChangeLog:

* include/std/limits: Whitespace changes.

3 years agoc++: wrong pretty printing of nested type [PR95303]
Patrick Palka [Tue, 7 Jul 2020 20:33:12 +0000 (16:33 -0400)] 
c++: wrong pretty printing of nested type [PR95303]

In the testcase below, we pretty print the nested type A<int>::B as
A<int>::B<int> because we don't check whether B is itself a class
template before printing the innermost set of template arguments from
B's TEMPLATE_INFO (which in this case belong to A).  This patch fixes
this by checking PRIMARY_TEMPLATE_P beforehand.

gcc/cp/ChangeLog:

PR c++/95303
* cxx-pretty-print.c (pp_cxx_unqualified_id): Check
PRIMARY_TEMPLATE_P before printing the innermost template
arguments.

gcc/testsuite/ChangeLog:

PR c++/95303
* g++.dg/concepts/diagnostic14.C: New test.

3 years agors6000: fix power10_hw test
Aaron Sawdey [Tue, 7 Jul 2020 17:48:56 +0000 (12:48 -0500)] 
rs6000: fix power10_hw test

The code snippet for this test was returning 1 if power10
instructions executed correctly. It should return 0 if the
test passes.

* lib/target-supports.exp (check_power10_hw_available):
Return 0 for passing test.

3 years agotestsuite: missed testcase
Nathan Sidwell [Tue, 7 Jul 2020 18:43:58 +0000 (11:43 -0700)] 
testsuite: missed testcase

I discovered I'd missed applying a testcase when fixing up the EOF
token location a while back.

gcc/testsuite/
* c-c++-common/cpp/pragma-eof.c: New

3 years agopreprocessor: Better line info for <builtin> & <command-line>
Nathan Sidwell [Tue, 7 Jul 2020 18:28:59 +0000 (11:28 -0700)] 
preprocessor: Better line info for <builtin> & <command-line>

With C++ module header units it becomes important to distinguish
between macros defined in forced headers (& commandline & builtins)
from those defined in the header file being processed.  We weren't
making that easy because we treated the builtins and command-line
locations somewhat file-like, with incrementing line numbers, and
showing them as included from line 1 of the main file.  This patch does
3 things:

0) extend the idiom that 'line 0' of a file means 'the file as a whole'

1) builtins and command-line macros are shown as-if included from line zero.

2) when emitting preprocessed output we keep resetting the line number
so that re-reading that preprocessed output will get the same set of
locations for the command line etc.

For instance the new c-c++-common/cpp/line-2.c test, now emits

In file included from <command-line>:
./line-2.h:4:2: error: #error wrong
    4 | #error wrong
      |  ^~~~~
line-2.c:3:11: error: macro "bill" passed 1 arguments, but takes just 0
    3 | int bill(1);
      |           ^
In file included from <command-line>:
./line-2.h:3: note: macro "bill" defined here
    3 | #define bill() 2
      |

Before it told you about including from <command-line>:31.

the preprocessed output looks like:
...

(There's a new optimization in do_line_marker to stop each of these
line markers causing a new line map.  We can simply rewind the
location, and keep using the same line map.)

libcpp/
* directives.c (do_linemarker): Optimize rewinding to line zero.
* files.c (_cpp_stack_file): Start on line zero when about to inject
headers.
(cpp_push_include, cpp_push_default_include): Use highest_line as
the location.
* include/cpplib.h (cpp_read_main_file): Add injecting parm.
* init.c (cpp_read_main_file): Likewise, inform _cpp_stack_file.
* internal.h (enum include_type): Add IT_MAIN_INJECT.
gcc/c-family/
* c-opts.c (c_common_post_options): Add 'injecting' arg to
cpp_read_main_file.
(c_finish_options): Add linemap_line_start calls for builtin and cmd
maps.  Force token position to line_table's highest line.
* c-ppoutput.c (print_line_1): Refactor, print line zero.
(cb_define): Always increment source line.
gcc/testsuite/
* c-c++-common/cpp/line-2.c: New.
* c-c++-common/cpp/line-2.h: New.
* c-c++-common/cpp/line-3.c: New.
* c-c++-common/cpp/line-4.c: New.
* c-c++-common/cpp/line-4.h: New.

3 years agoAvoid printing informational notes when -Wmismatched-tags is suppressed in system...
Martin Sebor [Tue, 7 Jul 2020 17:23:50 +0000 (11:23 -0600)] 
Avoid printing informational notes when -Wmismatched-tags is suppressed in system headers (PR c++/96063)

Related:
PR c++/96063 - mismatched-tags warnings in stdlib headers

gcc/cp/ChangeLog:

PR c++/96063
* parser.c (class_decl_loc_t::diag_mismatched_tags): Print notes only
if warning_at returns nonzero.

gcc/testsuite/ChangeLog:

PR c++/96063
* g++.dg/warn/Wmismatched-tags-7.C: New test.
* g++.dg/warn/Wmismatched-tags-8.C: New test.

3 years agoAarch64: Change costs for TX2 to expose more vectorization opportunities
Anton Youdkevitch [Tue, 26 May 2020 11:23:04 +0000 (04:23 -0700)] 
Aarch64: Change costs for TX2 to expose more vectorization opportunities

Make the costs such that they do not exaclty reflect
the actual instructions costs from the manual but make
the codegen emit the code we want it to.

gcc/ChangeLog:

* config/aarch64/aarch64.c (thunderx2t99_regmove_cost,
thunderx2t99_vector_cost): Likewise.

3 years agofix detection of negative step DR groups
Richard Biener [Tue, 7 Jul 2020 11:57:40 +0000 (13:57 +0200)] 
fix detection of negative step DR groups

This fixes a condition that caused all negative step DR groups to
be detected as single element interleaving.  Such groups are
rejected by interleaving vectorization but miscompiled by SLP
which is fixed by forcing VMAT_STRIDED_SLP for now.

2020-07-07  Richard Biener  <rguenther@suse.de>

* tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Fix
group overlap condition to allow negative step DR groups.
* tree-vect-stmts.c (get_group_load_store_type): For
multi element SLP groups force VMAT_STRIDED_SLP when the step
is negative.

* gcc.dg/vect/slp-47.c: New testcase.
* gcc.dg/vect/slp-48.c: Likewise.

3 years ago[Ada] Small adjustment in the handling of alignment for arrays
Eric Botcazou [Fri, 15 May 2020 21:15:38 +0000 (23:15 +0200)] 
[Ada] Small adjustment in the handling of alignment for arrays

gcc/ada/

* freeze.adb (Freeze_Array_Type): Use Ctyp more consistently.
Move code setting the alignment in the non-packed case to...
* layout.adb (Layout_Type): ...here.

3 years ago[Ada] Avoid crash on node extension
Bob Duff [Fri, 15 May 2020 15:40:50 +0000 (11:40 -0400)] 
[Ada] Avoid crash on node extension

gcc/ada/

* treepr.adb (Print_Node): Add code to test Is_Extension.

3 years ago[Ada] Assert failure on invalid Variable_Indexing
Arnaud Charlet [Fri, 15 May 2020 09:41:10 +0000 (05:41 -0400)] 
[Ada] Assert failure on invalid Variable_Indexing

gcc/ada/

* sem_ch4.adb (Try_Container_Indexing): Add ??? comment. Protect
against malformed tree in case of errors.

3 years ago[Ada] Use Is_Generic_Subprogram where possible
Piotr Trojanek [Thu, 14 May 2020 19:56:29 +0000 (21:56 +0200)] 
[Ada] Use Is_Generic_Subprogram where possible

gcc/ada/

* einfo.adb, sem_ch10.adb, sem_ch12.adb, sem_ch8.adb: Use
Is_Generic_Subprogram.

3 years ago[Ada] Fix typo in code to get a better error message
Piotr Trojanek [Thu, 14 May 2020 20:08:09 +0000 (22:08 +0200)] 
[Ada] Fix typo in code to get a better error message

gcc/ada/

* sem_ch8.adb (Analyze_Use_Package): Replace low-level,
error-prone Ekind_In tests with high-level Is_Generic_Subprogram
and Is_Subprogram.

3 years ago[Ada] ACATS 4.1J - B854003 - Subtype conformance check missed #2
Arnaud Charlet [Thu, 14 May 2020 12:33:15 +0000 (08:33 -0400)] 
[Ada] ACATS 4.1J - B854003 - Subtype conformance check missed #2

gcc/ada/

* sem_ch6.adb (Check_Formal_Subprogram_Conformance): New
subprogram to handle checking without systematically emitting an
error.
(Check_Conformance): Update call to
Check_Formal_Subprogram_Conformance and fix handling of Conforms
and Errmsg parameters.

3 years ago[Ada] ACATS 4.1K - C452003
Arnaud Charlet [Wed, 13 May 2020 08:41:03 +0000 (04:41 -0400)] 
[Ada] ACATS 4.1K - C452003

gcc/ada/

* exp_ch4.adb (Tagged_Membership): Generate a call to
CW_Membership instead of using Build_CW_Membership.
(Expand_N_In): Remove wrong handling of null access types and
corresponding comment.
* exp_intr.adb (Expand_Dispatching_Constructor_Call): Generate a
call to CW_Membership instead of using Build_CW_Membership.
* rtsfind.ads: Add CW_Membership.
* exp_atag.ads, exp_atag.adb (Build_CW_Membership): Removed.
* einfo.ads: Fix typo.
* libgnat/a-tags.ads, libgnat/a-tags.adb (CW_Membership): Moved
back to spec.

3 years ago[Ada] Errors in handling of access_to_subprogram contracts
Ed Schonberg [Thu, 14 May 2020 15:06:54 +0000 (11:06 -0400)] 
[Ada] Errors in handling of access_to_subprogram contracts

gcc/ada/

* exp_ch3.adb (Build_Access_Subprogram_Wrapper_Body): Create
proper subprogram specification for body, using names in the
subprogram declaration but distinct entities.
* exp_ch6.adb (Expand_Call): If this is an indirect call
involving a subprogram wrapper, insert pointer parameter in list
of actuals with a parameter association, not as a positional
parameter.

3 years ago[Ada] Hang on conditional expression as actual
Justin Squirek [Wed, 13 May 2020 22:59:29 +0000 (18:59 -0400)] 
[Ada] Hang on conditional expression as actual

gcc/ada/

* exp_ch6.adb (Expand_Branch): Verify the original node is a
conditional expression before recursing further.
(Insert_Level_Assign): Transform assertion into an explicit
raise.

3 years ago[Ada] Simplify statically known Max_Size_In_Storage_Elements attribute in more cases
Steve Baird [Wed, 13 May 2020 17:38:35 +0000 (10:38 -0700)] 
[Ada] Simplify statically known Max_Size_In_Storage_Elements attribute in more cases

gcc/ada/

* sem_attr.adb (Eval_Attribute): Generalize static evaluation of
Size attribute references to also handle
Max_Size_In_Storage_Elements references.

3 years ago[Ada] Ada2020: AI12-0198 potentially unevaluated components of arrays
Javier Miranda [Wed, 13 May 2020 10:24:59 +0000 (06:24 -0400)] 
[Ada] Ada2020: AI12-0198 potentially unevaluated components of arrays

gcc/ada/

* sem_util.adb (Is_Potentially_Unevaluated): Code cleanup.

3 years ago[Ada] Move generation of range checks for entry families to expander
Eric Botcazou [Sun, 10 May 2020 20:02:44 +0000 (22:02 +0200)] 
[Ada] Move generation of range checks for entry families to expander

gcc/ada/

* checks.ads (Expander Routines): Update the description of the
Do_Range_Check mechanism.
* checks.adb (Selected_Range_Checks): Fix typo.
* exp_ch9.adb: Add with and use clause for Checks.
(Actual_Index_Expression): Generate a range check if requested.
(Entry_Index_Expression): Likewise.
* sem_attr.adb (Resolve_Attribute) <Attribute_Count>: Call
Apply_Scalar_Range_Check instead of Apply_Range_Check.
* sem_ch9.adb (Analyze_Accept_Statement): Likewise.
* sem_res.adb (Resolve_Entry): Likewise, after having set the
actual index type on the prefix of the indexed component.
(Resolve_Indexed_Component): Remove useless conditional construct.

3 years ago[Ada] Freeze aspect expression relocated to pragma Precondition
Piotr Trojanek [Tue, 12 May 2020 19:17:08 +0000 (21:17 +0200)] 
[Ada] Freeze aspect expression relocated to pragma Precondition

gcc/ada/

* contracts.adb (Analyze_Entry_Or_Subprogram_Contract,
Process_Preconditions_For): Freeze expression that has been
relocated to pragma Precondition, not the expression which is
still in the aspect.

3 years ago[Ada] ACATS 4.1H - BC60005 - null exclusion matching for formal subprograms
Arnaud Charlet [Sun, 10 May 2020 20:36:00 +0000 (16:36 -0400)] 
[Ada] ACATS 4.1H - BC60005 - null exclusion matching for formal subprograms

gcc/ada/

* sem_ch6.adb (Check_Conformance): Remove unnecessary (and
wrong) code.
* sem_ch8.adb (Check_Null_Exclusion): Post error at proper
location.  Introduce new helper Null_Exclusion_Mismatch and fix
implementation wrt formal subprograms used in generic bodies.
(Analyze_Subprogram_Renaming): Fix missing setting of
Error_Msg_Sloc.
(Analyze_Object_Renaming): Replace "in Anonymous_Access_Kind" by
Is_Anonymous_Access_Type.
* sem_util.adb (Has_Null_Exclusion): Fix handling of
N_Parameter_Specification.
* sem_ch12.adb (Instantiate_Object): Replace "in
Anonymous_Access_Kind" by Is_Anonymous_Access_Type.

3 years ago[Ada] Handle explicit dereferences in expression functions
Ed Schonberg [Tue, 12 May 2020 12:20:22 +0000 (08:20 -0400)] 
[Ada] Handle explicit dereferences in expression functions

gcc/ada/

* freeze.adb (Freeze_Expr_Types): Freeze the designated type of
the explicit dereference.

3 years ago[Ada] Ada2020: AI12-0198 potentially unevaluated components of arrays
Javier Miranda [Tue, 12 May 2020 15:01:16 +0000 (11:01 -0400)] 
[Ada] Ada2020: AI12-0198 potentially unevaluated components of arrays

gcc/ada/

* sem_util.adb (Is_Potentially_Unevaluated): Protect reading
attribute Etype.

3 years ago[Ada] Use pragma Unsuppress in Time_IO
Bob Duff [Tue, 12 May 2020 15:07:36 +0000 (11:07 -0400)] 
[Ada] Use pragma Unsuppress in Time_IO

gcc/ada/

* libgnat/g-catiio.adb (Value, Parse_ISO_8601): Unsuppress
checks, and don't rely on 'Valid.

3 years ago[Ada] Add support for XDR streaming in the default runtime
Arnaud Charlet [Tue, 12 May 2020 14:35:12 +0000 (16:35 +0200)] 
[Ada] Add support for XDR streaming in the default runtime

gcc/ada/

* Makefile.rtl: Remove dead code.
* doc/gnat_ugn/building_executable_programs_with_gnat.rst:
Document gnatbind -xdr switch.
* gnat_ugn.texi: Regenerate.

3 years ago[Ada] Fix documentation of -gnatVa switch
Piotr Trojanek [Mon, 11 May 2020 21:48:34 +0000 (23:48 +0200)] 
[Ada] Fix documentation of -gnatVa switch

gcc/ada/

* doc/gnat_ugn/building_executable_programs_with_gnat.rst
(Validity Checking): Add "p" to the list of switches enabled by
-gnatVa.
* gnat_ugn.texi: Regenerate.

3 years ago[Ada] Ensure No_Specification_Of_Aspect forbids pragmas and repr. clauses
Ghjuvan Lacambre [Mon, 11 May 2020 11:51:13 +0000 (13:51 +0200)] 
[Ada] Ensure No_Specification_Of_Aspect forbids pragmas and repr. clauses

gcc/ada/

* sem_ch13.adb (Analyze_Attribute_Definition_Clause): Add call
to Check_Restriction_No_Specification_Of_Aspect.
* sem_prag.adb (Analyze_Pragma): Likewise.
* restrict.ads (Check_Restriction_No_Specification_Of_Aspect):
Mention possible new node kinds in documentation.
* restrict.adb (Check_Restriction_No_Specification_Of_Aspect):
Retrieve aspect id from different fields if given node is an
N_Pragma or an N_Attribute_Definition_Clause.

3 years ago[Ada] Checking type invariants on in params of procedures, not functions (AI12-0044)
Gary Dismukes [Mon, 11 May 2020 14:34:26 +0000 (10:34 -0400)] 
[Ada] Checking type invariants on in params of procedures, not functions (AI12-0044)

gcc/ada/

* contracts.adb (Add_Invariant_And_Predicate_Checks): Relax the
condition for doing invariant checks so that in-mode parameters
of procedures are also checked (required by AI05-0289, and
restricted to procedures by AI12-0044). This is done in a
procedure's nested postconditions procedure.
* exp_ch6.adb (Expand_Actuals): Also perform postcall invariant
checks for in parameters of procedures (but not functions).
Moved invariant-checking code to end of Expand_Actuals
(including the nested function Is_Public_Subp).

3 years ago[Ada] Change local object from variable to constant
Piotr Trojanek [Sun, 10 May 2020 19:56:42 +0000 (21:56 +0200)] 
[Ada] Change local object from variable to constant

gcc/ada/

* sem_aggr.adb (Resolve_Delta_Array_Aggregate): Make Index_Type
a constant.

3 years ago[Ada] Fix expansion of delta aggregates with slices
Piotr Trojanek [Sun, 10 May 2020 19:12:33 +0000 (21:12 +0200)] 
[Ada] Fix expansion of delta aggregates with slices

gcc/ada/

* exp_aggr.adb (Expand_N_Delta_Aggregate): Use type of the delta
base expression for the anonymous object of the delta aggregate.

3 years ago[Ada] Ada2020: AI12-0198 potentially unevaluated components of arrays
Javier Miranda [Sun, 10 May 2020 11:27:02 +0000 (07:27 -0400)] 
[Ada] Ada2020: AI12-0198 potentially unevaluated components of arrays

gcc/ada/

* sem_util.ads (Interval_Lists): Reordering routine.
* sem_util.adb (Interval_Lists): Reordering routines to keep
them alphabetically ordered.

3 years ago[Ada] Remove extra checks for non-static context from resolving 'Update
Piotr Trojanek [Mon, 11 May 2020 16:21:29 +0000 (18:21 +0200)] 
[Ada] Remove extra checks for non-static context from resolving 'Update

gcc/ada/

* sem_attr.adb (Resolve_Attribute): Do not call
Check_Non_Static_Context.

3 years ago[Ada] Spurious elaboration warnings with -gnatc and null procedures
Arnaud Charlet [Sat, 9 May 2020 17:37:13 +0000 (13:37 -0400)] 
[Ada] Spurious elaboration warnings with -gnatc and null procedures

gcc/ada/

* sem_elab.adb (Is_Guaranteed_ABE): Take into account null
procedures.

3 years ago[Ada] Minor cleanup with Is_Assignable
Piotr Trojanek [Sun, 10 May 2020 14:16:30 +0000 (16:16 +0200)] 
[Ada] Minor cleanup with Is_Assignable

gcc/ada/

* sem_ch5.adb (Analyze_Assignment): Reuse Is_Assignable.

3 years ago[Ada] Set range checks for for 'Update on arrays in GNATprove expansion
Piotr Trojanek [Sat, 9 May 2020 20:30:56 +0000 (22:30 +0200)] 
[Ada] Set range checks for for 'Update on arrays in GNATprove expansion

gcc/ada/

* exp_spark.adb (Expand_SPARK_N_Attribute_Reference): Add scalar
range checks for 'Update on arrays just like for 'Update on
records.
* sem_attr.adb (Analyze_Array_Component_Update): Do not set
range checks for single-dimensional arrays.
(Resolve_Attribute): Do not set range checks for both single-
and multi- dimensional arrays.

3 years ago[Ada] Style fixes in the description of Analysis-Resolution-Expansion
Piotr Trojanek [Sun, 10 May 2020 19:55:27 +0000 (21:55 +0200)] 
[Ada] Style fixes in the description of Analysis-Resolution-Expansion

gcc/ada/

* sem.ads (Sem): Fix description.

3 years ago[Ada] Allow aspect Relaxed_Initialization on protected entries
Piotr Trojanek [Sun, 10 May 2020 21:16:43 +0000 (23:16 +0200)] 
[Ada] Allow aspect Relaxed_Initialization on protected entries

gcc/ada/

* sem_ch13.adb (Analyze_Relaxed_Parameter): Fix for protected
entries.

3 years agoFix a typo in doc/generic.texi
Qian Jianhua [Tue, 7 Jul 2020 09:12:07 +0000 (10:12 +0100)] 
Fix a typo in doc/generic.texi

gcc/
* doc/generic.texi: Fix typo.

3 years agoPass correct parameters to c_parser_do_statement.
Kaipeng Zhou [Tue, 7 Jul 2020 09:07:28 +0000 (10:07 +0100)] 
Pass correct parameters to c_parser_do_statement.

gcc/c/ChangeLog:

* c-parser.c (c_parser_statement_after_labels): Pass correct
parameters to c_parser_do_statement.

3 years agolibstdc++: Add PR number to ChangeLog entry and fix typos
Jonathan Wakely [Tue, 7 Jul 2020 07:23:12 +0000 (08:23 +0100)] 
libstdc++: Add PR number to ChangeLog entry and fix typos

3 years agoaccept <case> and [cond] in ChangeLog
Alexandre Oliva [Tue, 7 Jul 2020 07:02:01 +0000 (09:02 +0200)] 
accept <case> and [cond] in ChangeLog

Only '(' and ':' currently terminate file lists in ChangeLog entries
in the ChangeLog parser.  This rules out such legitimate entries as:

* filename <CASE>:
* filename [COND]:

This patch extends the ChangeLog parser to recognize these forms.

for  contrib/ChangeLog

* gcc-changelog/git_commit.py: Support CASE and COND.
* gcc-changelog/test_patches.txt: Add test.
* gcc-changelog/test_email.py: Add test.

Co-Authored-By: Martin Liska <mliska@suse.cz>
3 years agofix LTO streaming order dependence on randomness
Richard Biener [Mon, 6 Jul 2020 09:30:53 +0000 (11:30 +0200)] 
fix LTO streaming order dependence on randomness

This fixes the sorting of to copy symbols in lto_output introduced
with GCC 10 to not depend on the actual values of the randomness
we append to LTO section names but instead on the order they appear
in the unsorted array.

This fixed observed debug info differences due to tree merging
prevailing different early debug pointers.

2020-07-06  Richard Biener  <rguenther@suse.de>

* lto-streamer-out.c (cmp_symbol_files): Use the computed
order map to sort symbols from the same sub-file together.
(lto_output): Compute a map of sub-file to an order number
it appears in the symbol output array.

3 years agoDaily bump.
GCC Administrator [Tue, 7 Jul 2020 00:16:40 +0000 (00:16 +0000)] 
Daily bump.

3 years agoExclude calls to variadic lambda stubs from -Wnonnull checking (PR c++/95984).
Martin Sebor [Mon, 6 Jul 2020 21:23:37 +0000 (15:23 -0600)] 
Exclude calls to variadic lambda stubs from -Wnonnull checking (PR c++/95984).

Resolves:
PR c++/95984 - Internal compiler error: Error reporting routines re-entered in -Wnonnull on a variadic lamnda
PR c++/96021 - missing -Wnonnull passing nullptr to a nonnull variadic lambda

gcc/c-family/ChangeLog:

PR c++/95984
* c-common.c (check_function_nonnull): Avoid checking syntesized calls
to stub lambda objects with null this pointer.
(check_nonnull_arg): Handle C++ nullptr.

gcc/cp/ChangeLog:

PR c++/95984
* call.c (build_over_call): Check calls only when tf_warning is set.

gcc/testsuite/ChangeLog:

PR c++/95984
* g++.dg/warn/Wnonnull6.C: New test.

3 years agolibstdc++: Cleanup whitespace and type trait usage in <optional>
Jonathan Wakely [Mon, 6 Jul 2020 20:54:12 +0000 (21:54 +0100)] 
libstdc++: Cleanup whitespace and type trait usage in <optional>

This makes the formatting in <optional> consistent and also removes
redundant && tokens from template arguments for traits like
is_constructible and is_convertible.

libstdc++-v3/ChangeLog:

* include/std/optional (_Optional_payload_base, _Optional_base)
(optional, __optional_hash_call_base): Adjust whitespace and
other formatting. Remove redundant && tokens on template
arguments to type traits.

3 years agolibstdc++: Constrain std::make_optional
Jonathan Wakely [Mon, 6 Jul 2020 20:54:12 +0000 (21:54 +0100)] 
libstdc++: Constrain std::make_optional

The standard rquires that std::make_optional is constrained similarly to
the std::optional constructors, which our implementation fails to do.

As a conforming extension this also adds a noexcept-specifier to each
std::make_optional overload.

libstdc++-v3/ChangeLog:

* include/std/optional (make_optional): Add enable_if
constraints and noexcept-specifier to each overload.
* testsuite/20_util/optional/make_optional-2.cc: New test.

3 years agolibstdc++: Add noexcept to std::optional initialization (PR 96036)
Jonathan Wakely [Mon, 6 Jul 2020 20:54:12 +0000 (21:54 +0100)] 
libstdc++: Add noexcept to std::optional initialization (PR 96036)

libstdc++-v3/ChangeLog:

PR libstdc++/96036
* include/std/optional (optional): Add noexcept-specifier to
every constructor, assignment operator, emplace function and
dereference operator.
* testsuite/20_util/optional/assignment/noexcept.cc: New test.
* testsuite/20_util/optional/cons/noexcept.cc: New test.

3 years agoc++: Always use pushdecl for exception library helpers
Nathan Sidwell [Mon, 6 Jul 2020 16:54:40 +0000 (09:54 -0700)] 
c++: Always use pushdecl for exception library helpers

The ABI exception helpers like __throw were being created by first
looking for them, and then adding if not found.  Primarily because
libitm wasn't declaring them with the correct exception specifiers.  I
fixed libitm a while back, so let's just use push_library_fn and let
the symbol table machinery deal with duplicates.  push_library_fn was
making the assumtion there wasn't already a decl available, by always
returning the new decl.  Bad things would happen if there was a
duplicate, because duplicate_decls explicitly gcc_frees the new decl.
Fixed by having it return whatever pushdecl returns.

gcc/cp/
* decl.c (push_library_fn): Return the decl pushdecl_toplevel returns.
* except.c (verify_library_fn): Replace with ...
(declare_library_fn_1): ... this fn.  Always push the fn.
(declare_library_fn): Call it.
(build_throw): Call declare_library_fn_1.
gcc/testsuite/
* g++.dg/eh/builtin10.C: Adjust expected errors.
* g++.dg/eh/builtin11.C: Likewise.
* g++.dg/eh/builtin5.C: Likewise.
* g++.dg/eh/builtin6.C: Likewise.
* g++.dg/eh/builtin7.C: Likewise.
* g++.dg/eh/builtin9.C: Likewise.
* g++.dg/parse/crash55.C: Likewise.

3 years agoPR fortran/95980 - ICE on using sync images with -fcheck=bounds
Harald Anlauf [Mon, 6 Jul 2020 16:58:23 +0000 (18:58 +0200)] 
PR fortran/95980 - ICE on using sync images with -fcheck=bounds

In SELECT TYPE, the argument may be an incorrectly specified unlimited
polymorphic variable.  Avoid a NULL pointer dereference for clean error
recovery.

gcc/fortran/
PR fortran/95980
* match.c (copy_ts_from_selector_to_associate, build_class_sym):
Distinguish between unlimited polymorphic and ordinary variables
to avoid NULL pointer dereference.
* resolve.c (resolve_select_type):
Distinguish between unlimited polymorphic and ordinary variables
to avoid NULL pointer dereference.

3 years agoPR fortran/95709 - ICE in gfc_resolve_code, at fortran/resolve.c:11807
Harald Anlauf [Mon, 6 Jul 2020 16:52:39 +0000 (18:52 +0200)] 
PR fortran/95709 - ICE in gfc_resolve_code, at fortran/resolve.c:11807

The legacy "assigned GOTO" accepts only scalar integer variables.
Check for proper arguments.

gcc/fortran/
PR fortran/95709
* resolve.c (gfc_resolve_code): Check for valid arguments to
assigned GOTO.

3 years agolibstdc++: Fix -Wmismatched-tags warnings (PR 96063)
Jonathan Wakely [Mon, 6 Jul 2020 15:19:23 +0000 (16:19 +0100)] 
libstdc++: Fix -Wmismatched-tags warnings (PR 96063)

libstdc++-v3/ChangeLog:

* include/bits/fs_dir.h: Use consistent tag in class-head.
* include/bits/localefwd.h: Likwise.
* include/bits/regex.h: Likwise.
* include/bits/stl_map.h: Likwise.
* include/bits/stl_multimap.h: Likwise.
* include/bits/stl_multiset.h: Likwise.
* include/bits/stl_set.h: Likwise.
* include/std/complex: Likwise.
* include/std/functional: Likwise.
* include/std/future: Likwise.
* include/std/system_error: Likwise.
* include/std/thread: Likwise.
* include/std/tuple: Likwise.
* include/std/type_traits: Likwise.
* include/std/valarray: Likwise.

3 years agoc++: Allow empty-declaration in C++11 and later (PR 96068)
Jonathan Wakely [Mon, 6 Jul 2020 14:58:33 +0000 (15:58 +0100)] 
c++: Allow empty-declaration in C++11 and later (PR 96068)

Since C++11 a semim-colon on its own at namespace scope is not invalid,
so do not give a pedantic diagnostic about it.

gcc/cp/ChangeLog:

PR c++/96068
* parser.c (cp_parser_toplevel_declaration): Only do pedwarn for
empty-declaration in C++98.

gcc/testsuite/ChangeLog:

* g++.old-deja/g++.bugs/900404_04.C: Add c++98_only selector to
dg-error for extra ';'.
* g++.old-deja/g++.law/missed-error2.C: Likewise.

3 years agotree-optimization/96075 - fix bogus misalignment calculation
Richard Biener [Mon, 6 Jul 2020 14:26:50 +0000 (16:26 +0200)] 
tree-optimization/96075 - fix bogus misalignment calculation

This fixes bogus misalignment calculation for negative steps
since an assertion a previous comment indicated no longer holds:

      /* DR_STEP(dr) is the same as -TYPE_SIZE of the scalar type,
         otherwise we wouldn't be here.  */

Thus the following replaces DR_STEP by -TYPE_SIZE.

2020-07-06  Richard Biener  <rguenther@suse.de>

PR tree-optimization/96075
* tree-vect-data-refs.c (vect_compute_data_ref_alignment): Use
TYPE_SIZE_UNIT of the vector component type instead of DR_STEP
for the misalignment calculation for negative step.

* gcc.dg/vect/slp-46.c: New testcase.

3 years ago[Ada] Avoid forced loading of System.Priority in CodePeer mode
Piotr Trojanek [Sun, 10 May 2020 08:03:35 +0000 (10:03 +0200)] 
[Ada] Avoid forced loading of System.Priority in CodePeer mode

gcc/ada/

* checks.adb (Apply_Scalar_Range_Check): Use Is_RTE.

3 years ago[Ada] Time_IO.Image: Allow printing the time zone
Bob Duff [Sun, 10 May 2020 15:29:27 +0000 (11:29 -0400)] 
[Ada] Time_IO.Image: Allow printing the time zone

gcc/ada/

* libgnat/g-catiio.ads: Document newly supported format.  Add
ISO_Time constant, for convenience.
* libgnat/g-catiio.adb (Image_Helper): New helper function to do
all the formatting work, called by the two exported Image
functions.  Add support for "%:::z" here. Add a Time_Zone
parameter used by the "%:::z" processing. This parameter is not
used for the actual time zone computations; local time is always
used for that, for ease of implementation reasons.  It would
make sense to use Append throughout this function, but that's a
cleanup for another day.
(Image): Modify these to pass the local time zone, or the
specified time zone, as appropriate.

3 years ago[Ada] Time_IO.Value: Allow subseconds and time zones together
Bob Duff [Sun, 10 May 2020 00:23:58 +0000 (20:23 -0400)] 
[Ada] Time_IO.Value: Allow subseconds and time zones together

gcc/ada/

* libgnat/g-catiio.ads: Change the regular expression that
documents the allowed format to match what ISO-8601 allows.
* libgnat/g-catiio.adb (Scan_Subsecond): Rewrite so it doesn't
assume the subsecond comes last.
(Parse_ISO_8601): Parse an optional subsecond, followed by an
optional time zone, rather than making these alternatives to
each other.

3 years ago[Ada] Clean up GNAT.Calendar.Time_IO.Value
Bob Duff [Sat, 9 May 2020 17:09:37 +0000 (13:09 -0400)] 
[Ada] Clean up GNAT.Calendar.Time_IO.Value

gcc/ada/

* libgnat/g-catiio.adb (Parse_ISO_8601): Minor cleanups:

Give some objects clearer names.
Make some objects more local to where they are used.

Remove some validity checks that can't fail, because some of
the variables tested have been moved so they're not visible here.
Anyway, Wrong_Syntax is about errors in the input, not data
validity.

Use Time_Zone_Seen instead of Local_Sign = ' ' to determine
that a time zone indication was seen.

We don't need to distinguish two different kinds of
syntax error (End_Of_Source_Reached and Wrong_Syntax),
so use Wrong_Syntax always.

Remove comment, "Certain scanning scenarios may handle
this exception on their own."; there are no such scenarios.

3 years ago[Ada] Implement Time_IO.Image with Time_Zone
Bob Duff [Fri, 8 May 2020 22:37:50 +0000 (18:37 -0400)] 
[Ada] Implement Time_IO.Image with Time_Zone

gcc/ada/

* libgnat/g-catiio.ads, libgnat/g-catiio.adb (Image): New
function. It might seem like the local-time Image should be
implemented in terms of the Time_Zone Image, but that would be
far harder to implement, so we do it the other way around.

3 years ago[Ada] Use Stream_Element_Arrays internally for secure hash computations
Thomas Quinot [Thu, 7 May 2020 14:08:03 +0000 (16:08 +0200)] 
[Ada] Use Stream_Element_Arrays internally for secure hash computations

gcc/ada/

* libgnat/g-sechas.ads, libgnat/g-sechas.adb: Refactor to use
Stream_Element_Array as the internal buffer type.
* libgnat/g-shshco.adb: Adjust to use Stream_Element_Offset
instead of Integer as the index in the internal state buffer.

3 years ago[Ada] Various typo fixes and reformatting of comments
Gary Dismukes [Thu, 7 May 2020 18:58:33 +0000 (14:58 -0400)] 
[Ada] Various typo fixes and reformatting of comments

gcc/ada/

* doc/gnat_rm/implementation_defined_pragmas.rst, errout.ads,
erroutc.adb, exp_ch4.adb, exp_ch6.adb, freeze.adb: Comment
rewording/reformatting/typo fixes.  Replace "ie." with "that is"
in comment; "can not" -> "cannot", and remove an extraneous
underscore in another comment.
* gnat_rm.texi, gnat_ugn.texi: Regenerate.

3 years ago[Ada] Ada2020: AI12-0198 potentially unevaluated array components
Javier Miranda [Tue, 5 May 2020 17:22:47 +0000 (13:22 -0400)] 
[Ada] Ada2020: AI12-0198 potentially unevaluated array components

gcc/ada/

* sem_util.ads (Interval_Lists.Aggregate_Intervals): New
subprogram.
* sem_util.adb (Has_Null_Others_Choice,
Non_Static_Or_Null_Range, Interval_Lists.Aggregate_Intervals):
New subprograms.
(Is_Potentially_Unevaluated): Adding support to detect
potentially unevaluated components of array aggregates.

3 years ago[Ada] Correct time zone in GNAT.Calendar.Time_IO.Value
Bob Duff [Fri, 8 May 2020 13:55:36 +0000 (09:55 -0400)] 
[Ada] Correct time zone in GNAT.Calendar.Time_IO.Value

gcc/ada/

* libgnat/g-catiio.adb (Parse_ISO_8601): New name for
Parse_ISO_8861_UTC.  8601 is the correct ISO standard number.
Also, "UTC" was confusing. All Time values are represented in
UTC, but the ISO 8601 date strings include a time zone.

If a time zone was specified, call
Ada.Calendar.Formatting.Time_Of instead of
GNAT.Calendar.Time_Of, because the latter adjusts to the current
time zone, whereas we want to use (just) the time zone specified
in the ISO string.  This allows us to pass Time_Zone instead to
Time_Of, instead of adjusting by Local_Disp by hand.

If no time zone was specified, call GNAT.Calendar.Time_Of as
before.

Use expanded names to clarify which Time_Of is being called.
Remove redundant comment, and move nonredundant part of the
commment to the spec.
(Value): Minor: use "not in" instead of "or else".
* libgnat/g-catiio.ads: Comment moved here. Correct the ISO
standard number.
* libgnat/g-calend.adb: Add ??? comments.
* libgnat/a-calend.ads, libgnat/a-calend.adb: Update obsolete
comments regarding the representation of type Time. Move the
information about the epoch (year 2150) to the spec, and avoid
uttering "2150" more than once.
* libgnat/a-catizo.ads (Time_Offset): Add comment.

3 years ago[Ada] Crash on derived type with rep. clause and renamed discriminant
Ed Schonberg [Fri, 8 May 2020 13:27:24 +0000 (09:27 -0400)] 
[Ada] Crash on derived type with rep. clause and renamed discriminant

gcc/ada/

* exp_ch5.adb (Find_Component): If the target type is a derived
record type and the required component is a discriminant that is
renamed in the derived type declaration, use the name of the
original discriminant to locate the intended target component.

3 years ago[Ada] Hang on generic declaration with ambiguous formal package
Justin Squirek [Thu, 7 May 2020 11:26:59 +0000 (07:26 -0400)] 
[Ada] Hang on generic declaration with ambiguous formal package

gcc/ada/

* sem_ch12.adb (Analyze_Associations): Add check for errors on
the generic formal before continuing with instantiation.

3 years ago[Ada] AI12-0366 Changes to Big_Integer and Big_Real
Arnaud Charlet [Thu, 7 May 2020 07:09:18 +0000 (03:09 -0400)] 
[Ada] AI12-0366 Changes to Big_Integer and Big_Real

gcc/ada/

* libgnat/a-nbnbin.ads, libgnat/a-nbnbin.adb,
libgnat/a-nbnbin__gmp.adb: Use more Valid_Big_Integer.

3 years ago[Ada] Fix typo in comment of Get_Index_Subtype
Piotr Trojanek [Thu, 7 May 2020 13:52:20 +0000 (15:52 +0200)] 
[Ada] Fix typo in comment of Get_Index_Subtype

gcc/ada/

* exp_util.ads (Get_Index_Subtype): Fix duplicate "Last".

3 years ago[Ada] There is no Expressions field in N_Delta_Aggregate node
Ed Schonberg [Thu, 7 May 2020 14:07:17 +0000 (10:07 -0400)] 
[Ada] There is no Expressions field in N_Delta_Aggregate node

gcc/ada/

* sinfo.ads: Indicate that Field 1 of an N_Delta_Aggregate is
unused. Previously it was erroneously labelled as holding an
Expressions list, in analogy with other aggregate constructs,
but there are no Expressions attached to this node syntactically
or semantically.

3 years ago[Ada] Fix evaluation of Enum_Rep applied to imported constant
Piotr Trojanek [Thu, 7 May 2020 12:08:07 +0000 (14:08 +0200)] 
[Ada] Fix evaluation of Enum_Rep applied to imported constant

gcc/ada/

* sem_attr.adb (Eval_Attribute): Check if constant has an
initialization expression.

3 years ago[Ada] Set range checks flag on 'Update for GNATprove in expansion
Piotr Trojanek [Wed, 29 Apr 2020 20:15:16 +0000 (22:15 +0200)] 
[Ada] Set range checks flag on 'Update for GNATprove in expansion

gcc/ada/

* exp_spark.adb (Expand_SPARK_N_Attribute_Reference): Apply
scalar range checks.
* sem_attr.adb (Resolve_Attribute): Do not set scalar range
checks when resolving attribute Update.

3 years ago[Ada] Predicates and the current instance of a subtype (AI12-0068)
Gary Dismukes [Wed, 6 May 2020 01:10:25 +0000 (21:10 -0400)] 
[Ada] Predicates and the current instance of a subtype (AI12-0068)

gcc/ada/

* sem_attr.adb (Analyze_Attribute, Attribute_Constrained): Issue
a warning if the attribute prefix is a current instance
reference within an aspect of a type or subtype.
(Address_Checks): Replace test of Is_Object (Ent) with
Is_Object_Reference (P) so that testing for current instances
will be done.
(Eval_Attribute): Add test for current instance reference, to
ensure that we still fold array attributes when current
instances are involved, since value prefixes are allowed for
array attributes, and will now be excluded by
Is_Object_Reference.
* sem_util.ads (Is_Current_Instance_Reference_In_Type_Aspect):
New exported query function.
* sem_util.adb (Is_Object_Reference): Return False for the case
where N is a current instance reference within an
aspect_specification of a type or subtype (basically if the
reference occurs within a predicate, invariant, or DIC aspect
expression).
(Is_Current_Instance_Reference_In_Type_Aspect): New function
that tests whether a node is a reference to a current instance
formal of a predicate, invariant, or
Default_Initial_Condition (DIC) subprogram.

3 years ago[Ada] Visibility error with aggregate in Declare_Expression
Ed Schonberg [Wed, 6 May 2020 16:03:27 +0000 (12:03 -0400)] 
[Ada] Visibility error with aggregate in Declare_Expression

gcc/ada/

* sem_res.adb (Resolve_Declare_Expression): New subprogram, to
install the scope holding local declarations of the expression,
before completing its resolution.
(Resolve): For an Expression_With_Actions that comes from a
source occurrence of a Declare_Expression, call new subprogram.

3 years ago[Ada] Simplify implicit loading of Tasking_State in GNATprove_Mode
Piotr Trojanek [Wed, 6 May 2020 15:19:56 +0000 (17:19 +0200)] 
[Ada] Simplify implicit loading of Tasking_State in GNATprove_Mode

gcc/ada/

* sem_attr.adb (Analyze_Attribute): Reuse SPARK_Implicit_Load.

3 years ago[Ada] Crash when an exception handler is executed with -gnatdk
Piotr Trojanek [Wed, 6 May 2020 16:40:22 +0000 (18:40 +0200)] 
[Ada] Crash when an exception handler is executed with -gnatdk

gcc/ada/

* sem_ch5.adb (Analyze_Loop_Parameter_Specification): Propagate
exception when switch -gnatdk is used and no previous errors are
present.
* sem_eval.adb (Compile_Time_Known_Value, Is_In_Range):
Likewise.
* sem_warn.adb (Operand_Has_Warnings_Suppressed): Likewise.

3 years ago[Ada] Refine use of access types in unnesting
Richard Kenner [Mon, 4 May 2020 13:47:30 +0000 (09:47 -0400)] 
[Ada] Refine use of access types in unnesting

gcc/ada/

* exp_unst.adb (Needs_Fat_Pointer): Don't check for formal.
(Unnest_Subprogram): Use 'Unchecked_Access instead of 'Access
when populating activation record.

3 years ago[Ada] Fix bad interaction between Unchecked_Conversion and limited type
Eric Botcazou [Tue, 5 May 2020 20:54:29 +0000 (22:54 +0200)] 
[Ada] Fix bad interaction between Unchecked_Conversion and limited type

gcc/ada/

* exp_ch6.adb (Add_Simple_Call_By_Copy_Code): Rename parameter
from Bit_Packed_Array to Force.
(Expand_Actuals): Do not apply BIP processing if the subprogram
is intrinsic.  Adjust calls to Add_Simple_Call_By_Copy_Code and
add one for In parameters whose actual is a CPP constructor call.

3 years ago[Ada] Remove obsolete code in Eval_Attribute
Eric Botcazou [Tue, 5 May 2020 14:09:33 +0000 (16:09 +0200)] 
[Ada] Remove obsolete code in Eval_Attribute

gcc/ada/

* sem_attr.adb (Eval_Attribute): Do not apply range checks in
the case of 'Pos applied to an integer type here.

3 years ago[Ada] Crash on declaration with aggregate for unchecked union
Ed Schonberg [Tue, 5 May 2020 00:05:03 +0000 (20:05 -0400)] 
[Ada] Crash on declaration with aggregate for unchecked union

gcc/ada/

* sem_ch3.adb (Analyze_Object_Declaration): If the type is an
Unchecked_Union, and the expression is an aggregate. complete
the analysis and resolution of the aggregate, and treat like a
regular object declaration, instead of as a renaming
declarattion.

3 years ago[Ada] Small cleanup throughout Exp_Ch9
Eric Botcazou [Tue, 5 May 2020 08:54:56 +0000 (10:54 +0200)] 
[Ada] Small cleanup throughout Exp_Ch9

gcc/ada/

* exp_ch9.adb (Is_Potentially_Large_Family): Add documentation.
(Actual_Index_Expression): Use Entry_Index_Type.
(Build_Entry_Count_Expression): Likewise.
(Build_Find_Body_Index): Likewise.
(Collect_Entry_Families): Likewise. Use directly the bounds of
the index type to find out whether the family is large.
(Entry_Index_Expression): Likewise.

3 years ago[Ada] Remove special case for processing null range in GNATprove
Piotr Trojanek [Mon, 4 May 2020 17:14:19 +0000 (19:14 +0200)] 
[Ada] Remove special case for processing null range in GNATprove

gcc/ada/

* sem_res.adb (Resolve_Range): Resolve both low and high bounds
with the range type.

3 years ago[Ada] AI12-0376 Relax RM 13.1(10) rule wrt primitive operations
Arnaud Charlet [Wed, 1 Apr 2020 10:29:53 +0000 (06:29 -0400)] 
[Ada] AI12-0376 Relax RM 13.1(10) rule wrt primitive operations

gcc/ada/

* aspects.ads (Is_Representation_Aspect):
Default_Component_Value is a representation aspect.
* sem_ch13.adb (Check_Aspect_Too_Late, Rep_Item_Too_Late): Relax
RM 13.1(10) rule wrt primitive operations for Ada 202x.

3 years ago[Ada] Improve run-time performance for large initialized allocators
Eric Botcazou [Mon, 4 May 2020 17:56:28 +0000 (19:56 +0200)] 
[Ada] Improve run-time performance for large initialized allocators

gcc/ada/

* exp_aggr.adb (Aggr_Assignment_OK_For_Backend): Move to library
level and use a new predicate Is_OK_Aggregate to recognize the
aggregates suitable for direct assignment by the back-end.
(Convert_Array_Aggr_In_Allocator): If neither in CodePeer mode nor
generating C code, generate a direct assignment instead of further
expanding if Aggr_Assignment_OK_For_Backend returns true.

3 years ago[Ada] Wording problems with predicates (AI12-0099)
Gary Dismukes [Sun, 3 May 2020 22:45:37 +0000 (18:45 -0400)] 
[Ada] Wording problems with predicates (AI12-0099)

gcc/ada/

* sem_aux.adb: Add a with clause for Nlists.
(Nearest_Ancestor): Test for the case of concurrent
types (testing for both Is_Concurrent_Type and
Is_Concurrent_Record_Type), and return the first ancestor in the
Interfaces list if present (otherwise will return Empty if no
interfaces).
* sem_ch13.adb (Build_Predicate_Functions): Add a ??? comment
about missing handling for adding predicates when they can be
inherited from multiple progenitors.

3 years ago[Ada] Default_Initial_Condition check done after extended return with init
Arnaud Charlet [Mon, 4 May 2020 08:56:52 +0000 (04:56 -0400)] 
[Ada] Default_Initial_Condition check done after extended return with init

gcc/ada/

* par-ch6.adb (P_Return_Object_Declaration): Set
Has_Init_Expression flag.

3 years agonvptx: Add support for vadd.add and vsub.add instructions.
Roger Sayle [Mon, 6 Jul 2020 06:46:52 +0000 (07:46 +0100)] 
nvptx: Add support for vadd.add and vsub.add instructions.

The following patch adds support for three-input addition instructions to
the nvptx backend.  The PTX ISA's "vadd.u32.u32.u32.add d, a, b, c"
instruction effectively implements 32-bit d = a+b+c, and the
"vsub.u32.u32.u32 d,a,b,c" instruction that provides 32-bit d = (a-b)+c.
The hope is that these mnemonics help ptxas generate the low-level
hardware's IADD3 instruction.

2020-07-06  Roger Sayle  <roger@nextmovesoftware.com>

gcc/ChangeLog:
* config/nvptx/nvptx.md (*vadd_addsi4): New instruction.
(*vsub_addsi4): New instruction.

gcc/testsuite/ChangeLog:
* gcc.target/nvptx/vadd_add.c: New test.
* gcc.target/nvptx/vsub_add.c: New test.

3 years agocris: New peephole2 movulsr + test-case.
Hans-Peter Nilsson [Mon, 6 Jul 2020 00:04:48 +0000 (02:04 +0200)] 
cris: New peephole2 movulsr + test-case.

Combine likes to change a zero-extension / and + shift as seen
in the test-case source to a logical shift followed by an and of
the shifted mask, like:
 lsrq 1,r0
 and.d 0x7f,r0

This was observed in the hot loop of coremark crcu16 and crcu32,
when doing other changes affecting instruction selection.  While
fixable by other means (like instruction costs or combine
patches), I wanted to break this out from those "other means".
The similarity to extant peephole optimizations is not
deliberate.

I noticed some paths to other peephole2 test-cases have changed
due to moves and renaming, so I updated them.

gcc:
* config/cris/cris.md (movulsr): New peephole2.

gcc/testsuite:
* gcc.target/cris/peep2-movulsr.c: New test.

3 years agocris: Correct gcc_assert for atomic_fetch_op pattern
Hans-Peter Nilsson [Sun, 5 Jul 2020 23:51:42 +0000 (01:51 +0200)] 
cris: Correct gcc_assert for atomic_fetch_op pattern

Yet another misnumbering of operands: the asserted non-overlap
would be the only benign operands overlap.  "Suddenly" exposed
by g++.dg/cpp0x/pr81325.C when testing unrelated changes
affecting register allocation.

To wit, operands 2 and 1 are the only ones that are safe for
overlap, it's only that it doesn't seem to make much sense to
write the address of the atomic data as the atomic data.

gcc:
* config/cris/sync.md ("cris_atomic_fetch_<atomic_op_name><mode>_1"):
Correct gcc_assert of overlapping operands.