platform/upstream/gcc.git
2 years ago[Ada] Do not issue a warning on a postcondition of True or False
Yannick Moy [Thu, 10 Feb 2022 11:36:52 +0000 (12:36 +0100)]
[Ada] Do not issue a warning on a postcondition of True or False

Do not issue a warning about the postcondition of a function not
mentioning its result when this postcondition is statically True or
False, as this is a specification of non-termination (for value False)
or a hint to SPARK prover for not inlining an expression function (for
value True). In any case, the warning brings no value here.

gcc/ada/

* sem_util.adb (Check_Result_And_Post_State): Exempt trivial
post.

2 years ago[Ada] Remove front-end SJLJ processing
Arnaud Charlet [Sun, 6 Feb 2022 16:06:36 +0000 (16:06 +0000)]
[Ada] Remove front-end SJLJ processing

We no longer use the so called front-end SJLJ exception mechanism, so
get rid of it.

gcc/ada/

* ali.adb, ali.ads, bcheck.adb, exp_ch11.adb, fe.h,
gnat1drv.adb, opt.adb, opt.ads, targparm.adb, targparm.ads,
lib-writ.adb: Get rid of Frontend_Exceptions processing.
* libgnat/system-aix.ads, libgnat/system-darwin-arm.ads,
libgnat/system-darwin-ppc.ads, libgnat/system-darwin-x86.ads,
libgnat/system-djgpp.ads, libgnat/system-dragonfly-x86_64.ads,
libgnat/system-freebsd.ads, libgnat/system-hpux-ia64.ads,
libgnat/system-hpux.ads, libgnat/system-linux-alpha.ads,
libgnat/system-linux-arm.ads, libgnat/system-linux-hppa.ads,
libgnat/system-linux-ia64.ads, libgnat/system-linux-m68k.ads,
libgnat/system-linux-mips.ads, libgnat/system-linux-ppc.ads,
libgnat/system-linux-riscv.ads, libgnat/system-linux-s390.ads,
libgnat/system-linux-sh4.ads, libgnat/system-linux-sparc.ads,
libgnat/system-linux-x86.ads, libgnat/system-lynxos178-ppc.ads,
libgnat/system-lynxos178-x86.ads, libgnat/system-mingw.ads,
libgnat/system-qnx-aarch64.ads, libgnat/system-rtems.ads,
libgnat/system-solaris-sparc.ads,
libgnat/system-solaris-x86.ads,
libgnat/system-vxworks-arm-rtp-smp.ads,
libgnat/system-vxworks-arm-rtp.ads,
libgnat/system-vxworks-arm.ads,
libgnat/system-vxworks-e500-kernel.ads,
libgnat/system-vxworks-e500-rtp-smp.ads,
libgnat/system-vxworks-e500-rtp.ads,
libgnat/system-vxworks-ppc-kernel.ads,
libgnat/system-vxworks-ppc-rtp-smp.ads,
libgnat/system-vxworks-ppc-rtp.ads,
libgnat/system-vxworks-x86-kernel.ads,
libgnat/system-vxworks-x86-rtp-smp.ads,
libgnat/system-vxworks-x86-rtp.ads,
libgnat/system-vxworks7-aarch64-rtp-smp.ads,
libgnat/system-vxworks7-aarch64.ads,
libgnat/system-vxworks7-arm-rtp-smp.ads,
libgnat/system-vxworks7-arm.ads,
libgnat/system-vxworks7-e500-kernel.ads,
libgnat/system-vxworks7-e500-rtp-smp.ads,
libgnat/system-vxworks7-e500-rtp.ads,
libgnat/system-vxworks7-ppc-kernel.ads,
libgnat/system-vxworks7-ppc-rtp-smp.ads,
libgnat/system-vxworks7-ppc-rtp.ads,
libgnat/system-vxworks7-ppc64-kernel.ads,
libgnat/system-vxworks7-ppc64-rtp-smp.ads,
libgnat/system-vxworks7-x86-kernel.ads,
libgnat/system-vxworks7-x86-rtp-smp.ads,
libgnat/system-vxworks7-x86-rtp.ads,
libgnat/system-vxworks7-x86_64-kernel.ads,
libgnat/system-vxworks7-x86_64-rtp-smp.ads: Remove
Frontend_Exceptions line.
* gcc-interface/decl.cc, gcc-interface/trans.cc
(gnat_to_gnu_entity, gnat_to_gnu_subprog_type, gigi,
gnat_to_gnu): Remove Front_End_SJLJ processing and always assume
Back_End_Exceptions.

2 years ago[Ada] Make debug printouts more robust
Bob Duff [Thu, 10 Feb 2022 19:55:32 +0000 (14:55 -0500)]
[Ada] Make debug printouts more robust

This patch improves some debug printouts so that they avoid crashing on
invalid data.

In addition, the relevant code uses Global_Name_Buffer all over the
place. This patch cleans up some of those uses, in particular ones in
the same code as the robustness changes, and code called by that code.

gcc/ada/

* namet.ads, namet.adb (Write_Name_For_Debug): New more-robust
version of Write_Name.
(Destroy_Global_Name_Buffer): New procedure to help detect bugs
related to use of Global_Name_Buffer.  Misc cleanup and comment
improvements. E.g. we don't need to document every detail of
debugging printouts, especially since they can change.
* uname.ads, uname.adb (Write_Unit_Name_For_Debug): New
more-robust version of Write_Unit_Name.
(Get_Unit_Name_String): Pass buffer in, instead of using the
global variable. Misc cleanup. Remove the "special fudge", which
is apparently not needed, and anyway the comment "the %s or %b
has already been eliminated" seems wrong.
(Write_Unit_Name): Call the new version of Get_Unit_Name_String.
* errout.adb (Set_Msg_Insertion_Unit_Name): Call the new version
of Get_Unit_Name_String. We pass the global variable here,
because it's too much trouble to disentangle such uses in
Errout.
* sem_util.ads, sem_util.adb, sem_dist.adb
(Get_Library_Unit_Name): New version of
Get_Library_Unit_Name_String that avoids usage of the global
variable.
* casing.ads, casing.adb, exp_prag.adb, exp_util.adb
(Set_All_Upper_Case): Remove. There is no need for a wrapper
here -- code is clearer without it.
* treepr.adb (Print_Name): Call Write_Name_For_Debug, which
deals with No_Name (etc), rather than duplicating that here.
Note that the call to Get_Name_String was superfluous.
(Tree_Dump): Call Write_Unit_Name_For_Debug instead of
Write_Unit_Name, which crashes if not Is_Valid_Name.
* erroutc.ads: Improve comments.
* erroutc.adb (Set_Msg_Name_Buffer): Call
Destroy_Global_Name_Buffer to detect potential bugs where it
incorrectly looks at the global variable.
* sinput.adb (Write_Location): Call Write_Name_For_Debug instead
of Write_Name, so it won't blow up on invalid data.
* sinput.ads: Improve comments; remove some verbosity.
* libgnat/s-imagef.adb: Fix typo in comment.

2 years ago[Ada] Add ghost code to facilitate proof with SPARK
Yannick Moy [Thu, 10 Feb 2022 10:54:44 +0000 (11:54 +0100)]
[Ada] Add ghost code to facilitate proof with SPARK

Proof of generic units for Long_Long_Long_Unsigned instantiations is
harder for provers, as they have to deal with larger values. Add ghost
code to make the proof easier.

gcc/ada/

* libgnat/s-imageu.adb (Set_Image_Unsigned): Add lemma.
* libgnat/s-valueu.adb (Scan_Raw_Unsigned): Add assertion.

2 years ago[Ada] Remove use of use-clauses in loaded runtime units
Yannick Moy [Thu, 10 Feb 2022 09:31:05 +0000 (10:31 +0100)]
[Ada] Remove use of use-clauses in loaded runtime units

The spec of runtime units that may be loaded by the compiler should not
contain use-clauses, for visibility to be correctly handled.  Remove
use-clauses that were introduced for the ghost big integers unit as part
of the proof of runtime units.

gcc/ada/

* libgnat/s-aridou.ads: Remove use-clause, add renames and
subtypes.
* libgnat/s-exponn.ads: Same.
* libgnat/s-expont.ads: Same.
* libgnat/s-widthu.ads: Same.

2 years ago[Ada] Fix support for ISO-8859-15 and IBM CP 850 encoding
Etienne Servais [Tue, 8 Feb 2022 11:17:48 +0000 (12:17 +0100)]
[Ada] Fix support for ISO-8859-15 and IBM CP 850 encoding

In ISO-8859-15, code for "lower y with diaeresis" is used for the upper
case.  In IBM CP 850, upper and lower o with stroke were missing.

gcc/ada/

* csets.adb (Fold_Latin_9): Fix y with diaeresis.
(Fold_IBM_PC_850): Fix o with stroke.

2 years ago[Ada] Remove redundant guard against Any_String
Piotr Trojanek [Wed, 9 Feb 2022 16:40:48 +0000 (17:40 +0100)]
[Ada] Remove redundant guard against Any_String

A sequence of checks for a valid Boolean argument fails when applied to
Any_String because its component is not of a Boolean type. The explicit
guard was unnecessary; it was only needed when a First_Index applied to
Any_String would crash, but this was fixed soon after this guard was
added.

Cleanup related to recent fixes for resolution of Boolean operators.

gcc/ada/

* sem_type.adb (Valid_Boolean_Arg): Remove redundant guard.

2 years ago[Ada] Accept raise expressions as operands of boolean operators
Piotr Trojanek [Wed, 9 Feb 2022 16:15:27 +0000 (17:15 +0100)]
[Ada] Accept raise expressions as operands of boolean operators

This patch restores the previous behaviour of a recently rewritten
routine Sem_Ch4.Find_Boolean_Types for boolean operators where one of
the operands is a raise-expression, e.g.:

  (raise Program_Error or else (X /= Y))

This change is required for the Entity field of the "or else" operator
to be set. For GNAT it doesn't matter, but GNATprove requires this field
to be set.

The affected call sequence starts at Find_Boolean_Types, goes via
Check_Right_Argument and Check_Boolean_Pair to Valid_Boolean_Arg, which
ultimately guards a call to Add_One_Interp.

gcc/ada/

* sem_type.adb (Valid_Boolean_Arg): Operands of Raise_Type are
valid boolean arguments.

2 years ago[Ada] Plug loophole in Possible_Type_For_Conditional_Expression
Eric Botcazou [Wed, 9 Feb 2022 16:16:09 +0000 (17:16 +0100)]
[Ada] Plug loophole in Possible_Type_For_Conditional_Expression

We need to check that the designated subprograms of access-to-subprogram
types are subtype conformant before registering a common interpretation.

gcc/ada/

* sem_ch4.adb (Possible_Type_For_Conditional_Expression): Add
test for subtype conformance in the cases of
access-to-subprogram types.

2 years ago[Ada] Adapt CodePeer analysis of GNAT to changes in dependencies
Yannick Moy [Tue, 8 Feb 2022 17:04:49 +0000 (18:04 +0100)]
[Ada] Adapt CodePeer analysis of GNAT to changes in dependencies

New dependencies in s-imagei are causing issues for CodePeer analysis
of GNAT source and libs. Updating the setup.

gcc/ada/

* libgnat/s-imagef.adb: Justify false message from CodePeer.

2 years ago[Ada] Revamp analysis of conditional expressions
Eric Botcazou [Mon, 7 Feb 2022 10:44:29 +0000 (11:44 +0100)]
[Ada] Revamp analysis of conditional expressions

The current implementation is skewed toward the first dependent
expression and does not look into the interpretations of the others if
the first one is not overloaded, which can create spurious ambiguities.
And more precise error messages are now given if the types of the
dependent expressions are not compatible.

gcc/ada/

* sem_ch4.adb (Analyze_Case_Expression): Rewrite.
(Analyze_If_Expression): Likewise.
(Possible_Type_For_Conditional_Expression): New function.
* sem_type.adb (Specific_Type): Retur the base type in more
cases.

2 years ago[Ada] Proof of 'Image support for signed integers
Yannick Moy [Fri, 4 Feb 2022 14:20:20 +0000 (15:20 +0100)]
[Ada] Proof of 'Image support for signed integers

Prove System.Image_I, similarly to the proof done for System.Image_U.
The contracts make the connection between the result of Image_Integer,
the available space computed with System.Width_U and the result of
'Value as computed by Value_Integer.

I/O units that now depend on non-pure units are also marked not Pure
anymore.

gcc/ada/

* libgnat/s-imagef.adb: Adapt to new signature of Image_I, by
providing ghost imported subprograms. For now, no contract is
used on these subprograms, as System.Image_F is not proved.
* libgnat/s-imagef.ads: Add modular type Uns as formal
parameter, to use in defining Int_Params for instantiating
Image_I.
* libgnat/s-imagei.adb: Add contracts and ghost code.
* libgnat/s-imagei.ads: Replace Int formal parameter by package
Int_Params, which bundles type Int and Uns with ghost
subprograms.  Add contracts.
* libgnat/s-imfi128.ads: Adapt to new formal of Image_F.
* libgnat/s-imfi32.ads: Adapt to new formal of Image_F.
* libgnat/s-imfi64.ads: Adapt to new formal of Image_F.
* libgnat/s-imgint.ads: Adapt to new formals of Image_I.
* libgnat/s-imglli.ads: Adapt to new formals of Image_I.
* libgnat/s-imgllli.ads: Adapt to new formals of Image_I.
* libgnat/s-valint.ads: Adapt to new formals of Value_I.
* libgnat/s-vallli.ads: Adapt to new formals of Value_I.
* libgnat/s-valllli.ads: Adapt to new formals of Value_I.
* libgnat/s-valuei.adb (Prove_Scan_Only_Decimal_Ghost): New
ghost lemma.
* libgnat/s-valuei.ads: New formal parameters to prove the new
lemma.
* libgnat/s-valuti.ads (Int_Params): Define a generic package to
be used as a trait-like formal parameter in Image_I and other
generics that need to instantiate Image_I.
* libgnat/s-widthu.ads (Big_10): Qualify the 10 literal.

2 years ago[Ada] Rewrite Sem_Ch4.Find_Boolean_Types
Eric Botcazou [Fri, 4 Feb 2022 17:53:43 +0000 (18:53 +0100)]
[Ada] Rewrite Sem_Ch4.Find_Boolean_Types

Using a straight implementation like the one in Find_Arithmetic_Types.

gcc/ada/

* sem_ch4.adb (Find_Arithmetic_Types): Use local variables.
(Find_Boolean_Types): Rewrite modeled on Find_Arithmetic_Types.

2 years ago[Ada] Introduce hardbool Machine_Attribute for Ada
Alexandre Oliva [Fri, 4 Feb 2022 02:00:09 +0000 (23:00 -0300)]
[Ada] Introduce hardbool Machine_Attribute for Ada

Implement and document hardened booleans, from nonstandard boolean types
with representation clauses to the extra validity checking performed on
boolean types annotated with the "hardbool" Machine_Attribute pragma.

gcc/ada/

* doc/gnat_rm/security_hardening_features.rst (Hardened
Booleans): New.
* exp_util.adb (Adjust_Condition): Perform validity checking on
hardbool-annotated types even with -gnatVT.
* gnat_rm.texi: Regenerate.
* gcc-interface/utils.cc (gnat_internal_attribute_table): Ignore
hardbool.

2 years ago[Ada] Fix scope of block in expanded protected entry body
Piotr Trojanek [Fri, 4 Feb 2022 19:56:51 +0000 (20:56 +0100)]
[Ada] Fix scope of block in expanded protected entry body

We expand protected entry body into procedure with a nested block. The
scope of this block is naturally the enclosing procedure. However, the
scope field was wrongly set to the entity of the enclosing procedure
body (i.e. E_Subprogram_Body); now it is set to the entity of the
enclosing procedure spec (i.e. E_Procedure).

In general, the scope field should always points to spec, even for
entities that are syntactically nested within a program unit body. The
scope set to E_Subprogram_Body entity was only accepted because in gigi
we had a protection for that, but this protection is now removed.

This patch fixes a regression in a ravenscar testsuite test (c_rf02_06),
where GNAT was crashing when compiling in configurable runtime mode with
restriction No_Exception_Propagation.

gcc/ada/

* exp_ch9.adb (Build_Protected_Entry): Set scope of the nested
block to spec and not the body of the procedure created for a
protected entry.

2 years ago[Ada] Fix typo in comment
Etienne Servais [Mon, 7 Feb 2022 14:34:01 +0000 (15:34 +0100)]
[Ada] Fix typo in comment

gcc/ada/

* sem_ch3.adb (Find_Type_Of_Object): Remove duplicate "i" in
comment.

2 years ago[Ada] Fix style checking rule for square brackets in Ada 2022 and above
Marc Poulhiès [Fri, 28 Jan 2022 15:55:43 +0000 (16:55 +0100)]
[Ada] Fix style checking rule for square brackets in Ada 2022 and above

The square bracket syntax introduced in Ada 2022 was incorrectly handled
by the style checker and incorrect spacing was enforced. The issue was
in part caused by the wide character syntax (support removed starting
from Ada 2022) that treats the square bracket as a valid identifier
character.

gcc/ada/

* csets.adb (Initialize): Only treat square bracket as valid
identifier character for Ada versions prior to Ada 2022.
* style.ads (Check_Left_Paren): Rename...
(Check_Left_Paren_Square_Bracket): ...to this.
* styleg.adb (Check_Left_Bracket): Rename...
(Check_Left_Paren_Square_Bracket): ...to this.
* styleg.ads (Check_Left_Paren): Rename...
(Check_Left_Paren_Square_Bracket): ...to this.
* scng.adb (Scan): Add check for spacing around left square
bracket and use new name for Check_Left_Paren_Square_Bracket.
* libgnat/a-szmzco.ads (Control_Ranges, Graphic_Ranges)
(Letter_Ranges, Decimal_Digit_Ranges, ISO_646_Ranges)
(Character_Ranges): Fix style (remove extra space).
* libgnat/a-swmwco.ads (Control_Ranges, Graphic_Ranges)
(Letter_Ranges, Decimal_Digit_Ranges, ISO_646_Ranges)
(Character_Ranges): Likewise.
* opt.adb (Set_Config_Switches): Remove [ from Identifier_Char
set.

2 years ago[Ada] Fix remaining asymmetry in Specific_Type
Eric Botcazou [Sun, 6 Feb 2022 14:54:25 +0000 (15:54 +0100)]
[Ada] Fix remaining asymmetry in Specific_Type

gcc/ada/

* sem_type.adb (Specific_Type): Add swapped cases for interfaces.

2 years ago[Ada] Fix typos in syntax for implementation-defined pragmas
Piotr Trojanek [Fri, 4 Feb 2022 08:16:53 +0000 (09:16 +0100)]
[Ada] Fix typos in syntax for implementation-defined pragmas

Documentation cleanup. Spotted while looking at description of
configuration pragmas.

gcc/ada/

* doc/gnat_rm/implementation_defined_pragmas.rst:
(Export_Object, Import_Object, Short_Descriptors): Fix pragma
syntax specification.
* gnat_rm.texi: Regenerate.

2 years ago[Ada] Reuse generic string hash for invocation signatures
Piotr Trojanek [Thu, 3 Feb 2022 18:55:34 +0000 (19:55 +0100)]
[Ada] Reuse generic string hash for invocation signatures

Use a generic string hash routine for hashing of invocation signatures.
The System.String_Hash unit was added to GNAT in 2009 and shouldn't
cause any bootstrap problems these days. To be safe, we don't use the
GNAT.String_Hash renaming, which was added later.

Cleanup related to a new restriction No_Uninitialized_Local_Scalars,
which will also rely on GNAT hash tables. Semantics is unaffected.

gcc/ada/

* ali.adb (Hash): Reuse GNAT.String_Hash.Hash and don't pollute
the Name_Id table.

2 years ago[Ada] Document Aggregate_Individually_Assign as a configuration pragma
Piotr Trojanek [Thu, 3 Feb 2022 22:53:13 +0000 (23:53 +0100)]
[Ada] Document Aggregate_Individually_Assign as a configuration pragma

Pragma Aggregate_Individually_Assign was apparently forgotten to be
listed in the GNAT User's Guide as a configuration pragma.

gcc/ada/

* doc/gnat_ugn/the_gnat_compilation_model.rst (Configuration
Pragmas): Add Aggregate_Individually_Assign; otherwise the list
is complete except for some obsoleted pragmas, which most likely
are intentionally omitted.
* gnat_ugn.texi: Regenerate.

2 years agolibgomp nvptx plugin: Split 'PLUGIN_NVPTX_DYNAMIC' into 'PLUGIN_NVPTX_INCLUDE_SYSTEM_...
Thomas Schwinge [Thu, 7 Apr 2022 21:10:16 +0000 (23:10 +0200)]
libgomp nvptx plugin: Split 'PLUGIN_NVPTX_DYNAMIC' into 'PLUGIN_NVPTX_INCLUDE_SYSTEM_CUDA_H' and 'PLUGIN_NVPTX_LINK_LIBCUDA'

Including the GCC-shipped 'include/cuda/cuda.h' vs. system <cuda.h> and
'dlopen'ing the CUDA Driver library vs. linking it are separate concerns.

libgomp/
* plugin/Makefrag.am: Handle 'PLUGIN_NVPTX_DYNAMIC'.
* plugin/configfrag.ac (PLUGIN_NVPTX_DYNAMIC): Change
'AC_DEFINE_UNQUOTED' into 'AM_CONDITIONAL'.
* plugin/plugin-nvptx.c: Split 'PLUGIN_NVPTX_DYNAMIC' into
'PLUGIN_NVPTX_INCLUDE_SYSTEM_CUDA_H' and
'PLUGIN_NVPTX_LINK_LIBCUDA'.
* Makefile.in: Regenerate.
* config.h.in: Likewise.
* configure: Likewise.

2 years agolibgomp plugins: Don't 'AC_SUBST' and 'AC_DEFINE_UNQUOTED' for 'PLUGIN_GCN', 'PLUGIN_...
Thomas Schwinge [Thu, 5 May 2022 21:10:23 +0000 (23:10 +0200)]
libgomp plugins: Don't 'AC_SUBST' and 'AC_DEFINE_UNQUOTED' for 'PLUGIN_GCN', 'PLUGIN_NVPTX'

Nothing ever used these.

libgomp/
* plugin/configfrag.ac: Don't 'AC_SUBST' and 'AC_DEFINE_UNQUOTED'
for 'PLUGIN_GCN', 'PLUGIN_NVPTX'.
* Makefile.in: Regenerate.
* config.h.in: Likewise.
* configure: Likewise.
* testsuite/Makefile.in: Likewise.

2 years agotree-optimization/105562 - avoid uninit diagnostic with better FRE
Richard Biener [Thu, 12 May 2022 10:13:29 +0000 (12:13 +0200)]
tree-optimization/105562 - avoid uninit diagnostic with better FRE

We can avoid some uninit diagnostics by making FRE disambiguate
against CLOBBERs since any aliasing there would invoke undefined
behavior for a read we are looking up.

2022-05-12  Richard Biener  <rguenther@suse.de>

PR tree-optimization/105562
* tree-ssa-sccvn.cc (vn_reference_lookup_3): Disambiguate
against all CLOBBER defs if there's not an obvious must-alias
and we are not doing redundant store elimination.
(vn_walk_cb_data::redundant_store_removal_p): New field.
(vn_reference_lookup_pieces): Initialize it.
(vn_reference_lookup): Add argument to specify if we are
doing redundant store removal.
(eliminate_dom_walker::eliminate_stmt): Specify we do.
* tree-ssa-sccvn.h (vn_reference_lookup): Adjust.

* g++.dg/warn/uninit-pr105562.C: New testcase.

2 years agoc++: Support module language-decl semantics
Nathan Sidwell [Mon, 9 May 2022 11:51:32 +0000 (04:51 -0700)]
c++: Support module language-decl semantics

In modules purview, one can attach a declaration to the global module
(i.e. not the named module in whence the declaration appears), using a
language declaration:

  export module Foo;
  extern "C++" void *operator new (std::size_t);

This implements those semantics.

gcc/cp/
* parser.cc (cp_parser_linkage_specification): Implement
global module attachment semantics.
gcc/testsuite/
* g++.dg/modules/lang-3_a.C: New.
* g++.dg/modules/lang-3_b.C: New.
* g++.dg/modules/lang-3_c.C: New.

2 years agoi386: Add combine splitter to transform pxor/pcmpeqb/pmovmskb/cmp 0xffff to ptest.
Haochen Jiang [Tue, 8 Feb 2022 02:51:26 +0000 (10:51 +0800)]
i386: Add combine splitter to transform pxor/pcmpeqb/pmovmskb/cmp 0xffff to ptest.

gcc/ChangeLog:

PR target/104371
* config/i386/sse.md (vi1avx2const): New define_mode_attr.
(pxor/pcmpeqb/pmovmskb/cmp 0xffff to ptest splitter):
New define_split pattern.

gcc/testsuite/ChangeLog:

PR target/104371
* gcc.target/i386/pr104371-1.c: New test.
* gcc.target/i386/pr104371-2.c: Ditto.

2 years agoopenmp: Add omp_all_memory support (C/C++ only so far)
Jakub Jelinek [Thu, 12 May 2022 06:31:20 +0000 (08:31 +0200)]
openmp: Add omp_all_memory support (C/C++ only so far)

The ugly part is that OpenMP 5.1 made omp_all_memory a reserved identifier
which isn't allowed to be used anywhere but in the depend clause, this is
against how everything else has been handled in OpenMP so far (where
some identifiers could have special meaning in some OpenMP clauses or
pragmas but not elsewhere).
The patch handles it by making it a conditional keyword (for -fopenmp
only) and emitting a better diagnostics when it is used in a primary
expression.  Having a nicer diagnostics when e.g. trying to do
int omp_all_memory;
or
int *omp_all_memory[10];
etc. would mean changing too many spots and hooking into name lookups
to reject declaring any such symbols would be too ugly and I'm afraid
there are way too many spots where one can introduce a name
(variables, functions, namespaces, struct, enum, enumerators, template
arguments, ...).

Otherwise, the handling is quite simple, normal depend clauses lower
into addresses of variables being handed over to the library, for
omp_all_memory I'm using NULL pointers.  omp_all_memory can only be
used with inout or out depend kinds and means that a task is dependent
on all previously created sibling tasks that have any dependency (of
any depend kind) and that any later created sibling tasks will be
dependent on it if they have any dependency.

2022-05-12  Jakub Jelinek  <jakub@redhat.com>

gcc/
* gimplify.cc (gimplify_omp_depend): Don't build_fold_addr_expr
if null_pointer_node.
(gimplify_scan_omp_clauses): Likewise.
* tree-pretty-print.cc (dump_omp_clause): Print null_pointer_node
as omp_all_memory.
gcc/c-family/
* c-common.h (enum rid): Add RID_OMP_ALL_MEMORY.
* c-omp.cc (c_finish_omp_depobj): Don't build_fold_addr_expr
if null_pointer_node.
gcc/c/
* c-parser.cc (c_parse_init): Register omp_all_memory as keyword
if flag_openmp.
(c_parser_postfix_expression): Diagnose uses of omp_all_memory
in postfix expressions.
(c_parser_omp_variable_list): Handle omp_all_memory in depend
clause.
* c-typeck.cc (c_finish_omp_clauses): Handle omp_all_memory
keyword in depend clause as null_pointer_node, diagnose invalid
uses.
gcc/cp/
* lex.cc (init_reswords): Register omp_all_memory as keyword
if flag_openmp.
* parser.cc (cp_parser_primary_expression): Diagnose uses of
omp_all_memory in postfix expressions.
(cp_parser_omp_var_list_no_open): Handle omp_all_memory in depend
clause.
* semantics.cc (finish_omp_clauses): Handle omp_all_memory
keyword in depend clause as null_pointer_node, diagnose invalid
uses.
* pt.cc (tsubst_omp_clause_decl): Pass through omp_all_memory.
gcc/testsuite/
* c-c++-common/gomp/all-memory-1.c: New test.
* c-c++-common/gomp/all-memory-2.c: New test.
* c-c++-common/gomp/all-memory-3.c: New test.
* g++.dg/gomp/all-memory-1.C: New test.
* g++.dg/gomp/all-memory-2.C: New test.
libgomp/
* libgomp.h (struct gomp_task): Add depend_all_memory member.
* task.c (gomp_init_task): Initialize depend_all_memory.
(gomp_task_handle_depend): Handle omp_all_memory.
(gomp_task_run_post_handle_depend_hash): Clear
parent->depend_all_memory if equal to current task.
(gomp_task_maybe_wait_for_dependencies): Handle omp_all_memory.
* testsuite/libgomp.c-c++-common/depend-1.c: New test.
* testsuite/libgomp.c-c++-common/depend-2.c: New test.
* testsuite/libgomp.c-c++-common/depend-3.c: New test.

2 years agotestsuite: Update Wconversion testcase check type.
jiawei [Thu, 5 May 2022 10:33:20 +0000 (18:33 +0800)]
testsuite: Update Wconversion testcase check type.

Some compiler target like arm-linux\riscv\power\s390x\xtensa-gcc handle
char as unsigned char, then there are no warnings occur and got FAIL cases.
Just change the type char into explicit signed char to keep the feature
consistency.

gcc/testsuite/ChangeLog:

* c-c++-common/Wconversion-1.c: Update type.

2 years agoDaily bump.
GCC Administrator [Thu, 12 May 2022 00:16:36 +0000 (00:16 +0000)]
Daily bump.

2 years agoanalyzer: fix memory leaks
David Malcolm [Wed, 11 May 2022 21:24:42 +0000 (17:24 -0400)]
analyzer: fix memory leaks

These leaks all relate to logging within -fdump-analyzer[-stderr]
or are one-time leaks; seen with valgrind.

gcc/analyzer/ChangeLog:
* checker-path.cc (state_change_event::get_desc): Call maybe_free
on label_text temporaries.
* diagnostic-manager.cc
(diagnostic_manager::prune_for_sm_diagnostic): Likewise.
* engine.cc (exploded_graph::~exploded_graph): Fix leak of
m_per_point_data and m_per_call_string_data values.  Simplify
cleanup of m_per_function_stats and m_per_point_data values.
(feasibility_state::maybe_update_for_edge): Fix leak of result of
superedge::get_description.
* region-model-manager.cc
(region_model_manager::~region_model_manager): Move cleanup of
m_setjmp_values to match the ordering of the fields within
region_model_manager.  Fix leak of values within
m_repeated_values_map, m_bits_within_values_map,
m_asm_output_values_map, and m_const_fn_result_values_map.

Signed-off-by: David Malcolm <dmalcolm@redhat.com>
2 years agotree: introduce range adaptor for TREE_VEC
Patrick Palka [Wed, 11 May 2022 20:14:59 +0000 (16:14 -0400)]
tree: introduce range adaptor for TREE_VEC

This patch implements a simple tree wrapper, named tree_vec_range, which
lets us idiomatically loop over all the elements of a TREE_VEC using a
C++11 range-based for loop:

  // v is a TREE_VEC
  for (tree e : tree_vec_range (v))
    ...

This is similar to the existing tree-based range adaptors ovl_range and
lkp_range added to the C++ FE in r12-340-g3307b9a07a3c51.

This patch also converts some existing loops over TREE_VEC within the
C++ FE to use tree_vec_range / range-for.

gcc/cp/ChangeLog:

* constraint.cc (tsubst_parameter_mapping): Convert loop over
TREE_VEC into a range-based for loop using tree_vec_range.
* pt.cc (iterative_hash_template_arg): Likewise.
(template_parms_level_to_args): Likewise.
(deducible_template_args): Likewise.
(check_undeduced_parms): Likewise.
(dependent_type_p_r): Likewise.
(value_dependent_expression_p) <case NONTYPE_ARGUMENT_PACK>:
Likewise.
(dependent_template_arg_p): Likewise.
* tree.cc (cp_walk_subtrees) <case NONTYPE_ARGUMENT_PACK>:
Likewise.

gcc/ChangeLog:

* tree.h (TREE_VEC_BEGIN): Define.
(TREE_VEC_END): Correct 'length' member access.
(class tree_vec_range): Define.

2 years agoc++: lambda template in requires [PR105541]
Jason Merrill [Wed, 11 May 2022 18:53:26 +0000 (14:53 -0400)]
c++: lambda template in requires [PR105541]

Since the patch for PR103408, the template parameters for the lambda in this
test have level 1 instead of 2, and we were treating null template args as 1
level of arguments, so tsubst_template_parms decided it had nothing to do.
Fixed by distinguishing between <> and no args at all, which is what we have
in our "substitution" in a requires-expression.

PR c++/105541

gcc/cp/ChangeLog:

* cp-tree.h (TMPL_ARGS_DEPTH): 0 for null args.
* parser.cc (cp_parser_enclosed_template_argument_list):
Use 0-length TREE_VEC for <>.

gcc/testsuite/ChangeLog:

* g++.dg/cpp2a/lambda-requires1.C: New test.

2 years agoc++: fix arm-eabi bootstrap [PR105567]
Jason Merrill [Wed, 11 May 2022 13:44:57 +0000 (09:44 -0400)]
c++: fix arm-eabi bootstrap [PR105567]

Since my r13-112, in the template we were changing 'return' to 'return this'
on cdtor_returns_this targets, and then getting confused by that when
instantiating.  So only make that change at instantiation time.

PR bootstrap/105567

gcc/cp/ChangeLog:

* typeck.cc (check_return_expr): Don't mess with ctor return value
while parsing a template.

gcc/testsuite/ChangeLog:

* g++.dg/template/ctor10.C: New test.

2 years agoFortran: fix error recovery on invalid array section
Harald Anlauf [Tue, 10 May 2022 21:41:57 +0000 (23:41 +0200)]
Fortran: fix error recovery on invalid array section

gcc/fortran/ChangeLog:

PR fortran/105230
* expr.cc (find_array_section): Correct logic to avoid NULL
pointer dereference on invalid array section.

gcc/testsuite/ChangeLog:

PR fortran/105230
* gfortran.dg/pr105230.f90: New test.

Co-authored-by: Steven G. Kargl <kargl@gcc.gnu.org>
2 years agors6000: Remove <Fv>
Segher Boessenkool [Tue, 10 May 2022 20:45:13 +0000 (20:45 +0000)]
rs6000: Remove <Fv>

The <Fv> mode iterator always expands to "wa".

2022-05-11  Segher Boessenkool  <segher@kernel.crashing.org>

* config/rs6000/rs6000.md: Use wa instead of <Fv>.

2 years agors6000: Remove <Ff>
Segher Boessenkool [Tue, 10 May 2022 20:35:56 +0000 (20:35 +0000)]
rs6000: Remove <Ff>

The <Ff> mode iterator can just use "d" always.

2022-05-11  Segher Boessenkool  <segher@kernel.crashing.org>

* config/rs6000/rs6000.md: Use d instead of <Ff>.

2 years agors6000: Delete RS6000_CONSTRAINT_f
Segher Boessenkool [Tue, 10 May 2022 18:47:31 +0000 (18:47 +0000)]
rs6000: Delete RS6000_CONSTRAINT_f

Constraint "f" is always exactly the same as constraint "d", so we do
not need RS6000_CONSTRAINT_f.

2022-05-11  Segher Boessenkool  <segher@kernel.crashing.org>

* config/rs6000/constraints.md (register_constraint "f"): Use
RS6000_CONSTRAINT_d.
* config/rs6000/rs6000.cc (rs6000_debug_reg_global): Do not handle
RS6000_CONSTRAINT_f.
(rs6000_init_hard_regno_mode_ok): Ditto.  Reorder and simplify a bit.
* config/rs6000/rs6000.h (r6000_reg_class_enum): Delete
RS6000_CONSTRAINT_d.

2 years agoUse variadic template functions for the gimple_build forwarders
Richard Biener [Wed, 11 May 2022 12:53:37 +0000 (14:53 +0200)]
Use variadic template functions for the gimple_build forwarders

The following consolidates the forwarders to gimple_build with
UNKNOWN_LOCATION by using two variadic template functions.  This
makes the code visually less disturbing and is in preparation
of adding more forwarders with different stmt insertion schemes.

2022-05-11  Richard Biener  <rguenther@suse.de>

* gimple-fold.h (gimple_build): Use variadic template
functions for the gimple_build API forwarders without
location_t argument.

2 years agomiddle-end/70090: Register __bdos for sanitizers if necessary
Siddhesh Poyarekar [Wed, 11 May 2022 13:43:09 +0000 (19:13 +0530)]
middle-end/70090: Register __bdos for sanitizers if necessary

The asan initializer registers __builtin_object_size for languages that
don't have it, e.g. Fortran.  Register __builtin_dynamic_object_size too
(we need both because __builtin_dynamic_object_size computation may
involve generating __builtin_object_size as a fallback) so that
gfortran.dg/ubsan/bind-c-intent-out-2.f90 does not crash anymore.

gcc/ChangeLog:

PR middle-end/70090
* asan.cc (initialize_sanitizer_builtins): Register
__builtin_dynamic_object_size if necessary.

Signed-off-by: Siddhesh Poyarekar <siddhesh@gotplt.org>
2 years ago[c++] Add module attachment
Nathan Sidwell [Mon, 9 May 2022 11:47:14 +0000 (04:47 -0700)]
[c++] Add module attachment

This adds module attachment as a distinct flag to 'in module purview'.
A declaration may have neither or both (as before), but can also have
just the 'in [named-module] purview', which was previously not
representable.

This new state allows some cleanup of redeclarations (particularly the
builtins), which was a little warty.  Some other internal APIs get
similarly clarified.

gcc/cp/
* cp-tree.h (DECL_MODULE_ATTACH_P): New.
(struct lang_decl_base): Add module_attach_p flag.
* decl.cc (duplicate_decls): Rework module redeclaration
checking.
* module.cc (trees_out::lang_decl_bools): Write attach flag.
(trees_in::lang_decl_bools): ... and read it back.
(trees_out::decl_value): Rework module attachment handling.
(trees_in::decl_value): Rename local var to reflect meaning.
(trees_in::key_mergeable): Likewise.
(get_originating_module): Use DECL_MODULE_ATTACH_P.  No need
to special-case mangling.
(module_may_redeclare): Reimplement.
(set_originating_module): Deal with attachment.
* name-lookup.cc (maybe_record_mergeable_decl): Deal with
attachment.
(mergeable_namespace_slots): Likewise.
(do_nonmember_using_decl): Likewise.
* name-lookup.h (mergeable_namespace_slots): Adjust parm
meaning.
* ptree.cc (cxx_print_decl): Adjust purview & attach printing.

2 years agortl-optimization/105559 - avoid quadratic behavior in delete_insn_and_edges
Richard Biener [Wed, 11 May 2022 11:34:37 +0000 (13:34 +0200)]
rtl-optimization/105559 - avoid quadratic behavior in delete_insn_and_edges

When the insn to delete is a debug insn there's no point in figuring
out whether it might be the last real insn and thus we have to purge
dead edges.

2022-05-11  Richard Biener  <rguenther@suse.de>

PR rtl-optimization/105559
* cfgrtl.cc (delete_insn_and_edges): Only perform search to BB_END
for non-debug insns.

2 years agoImplement some of fold_binary_op_with_conditional_arg in match.pd
Richard Biener [Wed, 11 May 2022 09:53:53 +0000 (11:53 +0200)]
Implement some of fold_binary_op_with_conditional_arg in match.pd

The following allows (c != 0 ? 0 : 100) != 0 to be simplified as
c != 0 as fold_binary_op_with_conditional_arg would have done
via forwprop and GENERIC folding.  Likewise it allows to combine
(a != 0) != 0 directly via match.pd instead of only via
forwprop and again fold_binary_op_with_conditional_arg.

The patterns do not fully implement all cases of
fold_binary_op_with_conditional_arg, some aspects like
"any of the operands simplify" cannot currently be expressed.

2022-05-11  Richard Biener  <rguenther@suse.de>

* generic-match-head.cc: Include tree-eh.h.
* match.pd ((cond ...) cmp X): New simplification inspired
by fold_binary_op_with_conditional_arg.
(eq/ne (cmp ...) true/false): Likewise.

* gcc.dg/tree-ssa/pr61839_1.c: Adjust.
* gcc.dg/tree-ssa/vrp24.c: Likewise.

2 years agolibgomp: Remove unused '--with-hsa-runtime', '--with-hsa-runtime-include', '--with...
Thomas Schwinge [Wed, 6 Apr 2022 10:26:13 +0000 (12:26 +0200)]
libgomp: Remove unused '--with-hsa-runtime', '--with-hsa-runtime-include', '--with-hsa-runtime-lib'

With recent commit 2e309a4eff80e55b53d32d26926a2a94eabfea21 "libgomp testsuite:
Don't amend 'LD_LIBRARY_PATH' for system-provided HSA Runtime library",
and commit d6adba307508c75f1ccb2121eb1a43c9ab1d4056 "libgomp GCN plugin:
 Clean up unused references to system-provided HSA Runtime library", the last
uses of '--with-hsa-runtime' etc. are gone.

gcc/
* doc/install.texi: Don't document '--with-hsa-runtime',
'--with-hsa-runtime-include', '--with-hsa-runtime-lib'.
libgomp/
* plugin/configfrag.ac: Remove '--with-hsa-runtime',
'--with-hsa-runtime-include', '--with-hsa-runtime-lib' processing.
* Makefile.in: Regenerate.
* configure: Likewise.
* testsuite/Makefile.in: Likewise.

2 years agolibgomp GCN plugin: Clean up always-empty 'PLUGIN_GCN_CPPFLAGS', 'PLUGIN_GCN_LDFLAGS'
Thomas Schwinge [Wed, 6 Apr 2022 10:15:28 +0000 (12:15 +0200)]
libgomp GCN plugin: Clean up always-empty 'PLUGIN_GCN_CPPFLAGS', 'PLUGIN_GCN_LDFLAGS'

After recent commit d6adba307508c75f1ccb2121eb1a43c9ab1d4056
"libgomp GCN plugin: Clean up unused references to system-provided HSA Runtime
library", these aren't set anymore.

libgomp/
* plugin/Makefrag.am (libgomp_plugin_gcn_la_CPPFLAGS): Don't
consider 'PLUGIN_GCN_CPPFLAGS'.
(libgomp_plugin_gcn_la_LDFLAGS): Don't consider
'PLUGIN_GCN_LDFLAGS'.
* plugin/configfrag.ac (PLUGIN_GCN_CPPFLAGS, PLUGIN_GCN_LDFLAGS):
Remove.
* Makefile.in: Regenerate.
* configure: Likewise.
* testsuite/Makefile.in: Likewise.

2 years agolibgomp GCN plugin: Clean up unused references to system-provided HSA Runtime library
Thomas Schwinge [Wed, 6 Apr 2022 09:31:45 +0000 (11:31 +0200)]
libgomp GCN plugin: Clean up unused references to system-provided HSA Runtime library

This is only active if GCC is 'configure'd with '--with-hsa-runtime=[...]' or
'--with-hsa-runtime-include=[...]', '--with-hsa-runtime-lib=[...]' -- which
nobody really is doing, as far as I can tell.

Originally changed for the libgomp HSA plugin in
commit b8d89b03db5f212919e4571671ebb4f5f8b1e19d (r242749)
"Remove build dependence on HSA run-time", and later propagated into the GCN
plugin, these are no longer built against system-provided HSA Runtime library.
Instead, unconditionally built against the GCC-shipped 'include/hsa*.h' header
files, and at run time does 'dlopen("libhsa-runtime64.so.1")'.  It thus doesn't
make sense to consider references to system-provided HSA Runtime library during
libgomp GCN plugin build.

libgomp/
* plugin/configfrag.ac (HSA_RUNTIME_CPPFLAGS)
(HSA_RUNTIME_LDFLAGS): Remove.
* configure: Regenerate.

2 years agolibgomp testsuite: Don't amend 'LD_LIBRARY_PATH' for system-provided HSA Runtime...
Thomas Schwinge [Wed, 6 Apr 2022 08:39:56 +0000 (10:39 +0200)]
libgomp testsuite: Don't amend 'LD_LIBRARY_PATH' for system-provided HSA Runtime library

This is only active if GCC is 'configure'd with '--with-hsa-runtime=[...]' or
'--with-hsa-runtime-lib=[...]' -- which nobody really is doing, as far as I can
tell.

'libgomp/testsuite/lib/libgomp.exp:libgomp_init' states:

    # For build-tree testing, also consider the library paths used for builing.
    # For installed testing, we assume all that to be provided in the sysroot.
    if { $blddir != "" } {
        [...]
        global hsa_runtime_lib
        if { $hsa_runtime_lib != "" } {
            append always_ld_library_path ":$hsa_runtime_lib"
        }
    }

However, the libgomp GCN plugin is unconditionally built against the
GCC-shipped 'include/hsa*.h' header files, and at run time does
'dlopen("libhsa-runtime64.so.1")', so there is no system-provided HSA Runtime
library "used for builing".  It thus doesn't make sense to amend
'LD_LIBRARY_PATH' for system-provided HSA Runtime library.

libgomp/
* testsuite/lib/libgomp.exp (libgomp_init): Don't
'append always_ld_library_path ":$hsa_runtime_lib"'.
* testsuite/libgomp-test-support.exp.in (hsa_runtime_lib): Don't set.

2 years agodocs: document --with-zstd
Martin Liska [Wed, 11 May 2022 11:21:26 +0000 (13:21 +0200)]
docs: document --with-zstd

PR other/105527

gcc/ChangeLog:

* doc/install.texi: Document the configure option --with-zstd.

2 years agoi386: simplify cpu_feature handling
Martin Liska [Wed, 15 Dec 2021 09:54:23 +0000 (10:54 +0100)]
i386: simplify cpu_feature handling

The patch removes unneeded loops for cpu_features2 and CONVERT_EXPR
that can be simplified with NOP_EXPR.

gcc/ChangeLog:

* common/config/i386/cpuinfo.h (has_cpu_feature): Directly
compute index in cpu_features2.
(set_cpu_feature): Likewise.
* config/i386/i386-builtins.cc (fold_builtin_cpu): Also remove
loop for cpu_features2 and use NOP_EXPRs.

2 years ago[Ada] Proof of 'Image support for unsigned integers
Yannick Moy [Tue, 1 Feb 2022 11:33:54 +0000 (12:33 +0100)]
[Ada] Proof of 'Image support for unsigned integers

Prove System.Image_U, making the connection with the space available in
the string as computed with System.Width_U and the functions that
support the other direction of 'Value in System.Value_U.

The units that support 'Image cannot be marked Pure anymore, as they now
depend on non-pure units.

gcc/ada/

* libgnat/s-imaged.ads: Remove Pure.
* libgnat/s-imagef.ads: Remove Pure.
* libgnat/s-imager.ads: Remove Pure.
* libgnat/s-imageu.adb: Add ghost code.
* libgnat/s-imageu.ads: Add contracts.
* libgnat/s-imde128.ads: Remove Pure.
* libgnat/s-imde32.ads: Remove Pure.
* libgnat/s-imde64.ads: Remove Pure.
* libgnat/s-imfi128.ads: Remove Pure.
* libgnat/s-imfi32.ads: Remove Pure.
* libgnat/s-imfi64.ads: Remove Pure.
* libgnat/s-imgflt.ads: Remove Pure.
* libgnat/s-imglfl.ads: Remove Pure.
* libgnat/s-imgllf.ads: Remove Pure.
* libgnat/s-imglllu.ads: Instantiate with ghost subprograms.
* libgnat/s-imgllu.ads: Instantiate with ghost subprograms.
* libgnat/s-imgrea.ads: Remove Pure.
* libgnat/s-imguns.ads: Instantiate with ghost subprograms.
* libgnat/s-imguti.ads: Remove Pure.
* libgnat/s-valueu.adb (Prove_Iter_Scan_Based_Number_Ghost,
Prove_Scan_Only_Decimal_Ghost): New lemmas.
* libgnat/s-valueu.ads (Uns_Option): Do not make type ghost to
be able to use it as formal in instantiations.
(Only_Decimal_Ghost): New ghost query.
(Prove_Iter_Scan_Based_Number_Ghost,
Prove_Scan_Only_Decimal_Ghost): New lemmas.
* libgnat/s-widlllu.ads: Adapt to changes in Width_U.
* libgnat/s-widllu.ads: Adapt to changes in Width_U.
* libgnat/s-widthu.adb: Change generic function in generic
package in order to complete the postcondition. Tighten the
upper bound on the result by 1.
* libgnat/s-widthu.ads: Same.
* libgnat/s-widuns.ads: Adapt to changes in Width_U.
* gcc-interface/Make-lang.in: Add dependencies on a-nubinu,
a-numeri.ads and a-widuns.ads.

2 years ago[Ada] Sort Detect_Blocking alphabetically among names and pragmas
Piotr Trojanek [Fri, 4 Feb 2022 12:34:52 +0000 (13:34 +0100)]
[Ada] Sort Detect_Blocking alphabetically among names and pragmas

In GNAT User's Guide section about configuration pragmas the
Detect_Blocking has been recently moved to an alphabetic order.  This
patch moves the its Name_Id and Pragma_Is in the source code to match
this order.

gcc/ada/

* snames.ads-tmpl (Detect_Blocking): Move after
Default_Storage_Pool.

2 years ago[Ada] Fix Compile_Time_(Error|Warning) as non-configuration pragmas
Piotr Trojanek [Thu, 3 Feb 2022 23:22:41 +0000 (00:22 +0100)]
[Ada] Fix Compile_Time_(Error|Warning) as non-configuration pragmas

Uses of Compile_Time_Error and Compile_Time_Warning as configuration
pragmas were detected with a custom check in semantic analysis. Now they
are detected with an existing general check in parser.

gcc/ada/

* sem_prag.adb (Analyze_Pragma): Remove custom check for
Compile_Time_Error and Compile_Time_Warning pragmas.
* snames.ads-tmpl (Compile_Time_Error, Compile_Time_Warning):
Move beyond Last_Configuration_Pragma_Name.

2 years ago[Ada] Correctly reject record aggregate using brackets
Marc Poulhiès [Mon, 31 Jan 2022 14:54:41 +0000 (15:54 +0100)]
[Ada] Correctly reject record aggregate using brackets

Ada 2022 bracket syntax is only allowed for arrays and containers, not
records. This change correctly rejects record aggregates using brackets
along with few cases of brackets being used instead of parentheses.

gcc/ada/

* libgnat/a-direct.adb (Start_Search_Internal): Use parentheses
for record aggregate.
* libgnat/a-stbubo.adb (Get_UTF_8): Likewise.
* libgnat/a-stbuun.adb (Get_UTF_8): Likewise.
* libgnat/a-stzbou.ads (Null_Bounded_Wide_Wide_String):
Likewise.
* libgnat/a-swmwco.ads (Lower_Case_Mapping, Basic_Mapping):
Likewise.
* libgnat/a-szmzco.ads (Lower_Case_Mapping, Basic_Mapping):
Likewise.
* libgnat/s-statxd.adb (Fields): Likewise.
* libgnat/s-stausa.adb (Initialize): Likewise.
* sem_aggr.adb (Resolve_Record_Aggregate): Reject aggregates
using brackets.

2 years ago[Ada] Fix unbalanced paren in documentation marker for GNAT Studio
Piotr Trojanek [Fri, 4 Feb 2022 08:47:27 +0000 (09:47 +0100)]
[Ada] Fix unbalanced paren in documentation marker for GNAT Studio

Unbalanced parens can be found when editing in emacs using the
"check-parens" command. Offending occurrences must be examined manually,
because few of them are intentional.

Minor cleanup of typos in documentation.

gcc/ada/

* doc/gnat_rm/implementation_defined_pragmas.rst,
doc/gnat_rm/implementation_of_specific_ada_features.rst,
doc/gnat_rm/representation_clauses_and_pragmas.rst,
doc/gnat_rm/the_gnat_library.rst,
doc/gnat_rm/the_implementation_of_standard_i_o.rst,
doc/gnat_ugn/building_executable_programs_with_gnat.rst,
doc/gnat_ugn/gnat_and_program_execution.rst,
doc/gnat_ugn/platform_specific_information.rst: Fix unbalanced
parentheses and brackets.
* sem_prag.adb (pragma Obsolescent): Likewise.
* gnat_rm.texi, gnat_ugn.texi: Regenerate.

2 years ago[Ada] New No_Local_Tagged_Types restriction
Marc Poulhiès [Tue, 1 Feb 2022 11:20:51 +0000 (12:20 +0100)]
[Ada] New No_Local_Tagged_Types restriction

Add new configuration restriction No_Local_Tagged_Types. This new
restriction ensures at compile time that all tagged types are only
declared at the library level.

gcc/ada/

* libgnat/s-rident.ads (Restriction_Id): Add
No_Local_Tagged_Types, update comment for adding new GNAT
specific restriction.
* restrict.ads (Implementation_Restriction): Add
No_Local_Tagged_Types.
* sem_ch3.adb (Analyze_Full_Type_Declaration): Add check for
tagged type at library level.
* doc/gnat_rm/standard_and_implementation_defined_restrictions.rst:
Add No_Local_Tagged_Types description.
* gnat_rm.texi: Regenerate.

2 years ago[Ada] Fix spurious ambiguity for if_expression containing operator
Eric Botcazou [Wed, 2 Feb 2022 20:04:18 +0000 (21:04 +0100)]
[Ada] Fix spurious ambiguity for if_expression containing operator

It comes from a discrepancy in the analysis of binary operators, between
the predefined ones for which we register an interpretation with the base
type and the user-defined ones for which we register an interpretation
with the subtype directly.

This is harmless in almost all cases because the two interpretations do not
differ by this type property only, but is problematic when the operation is
the dependent expression of a conditional expression, because it causes two
interpretations of the conditional expression to be registered, which this
time differ by this type property only, thus making them impossible to be
disambiguated later.

gcc/ada/

* sem_ch4.adb (Analyze_User_Defined_Binary_Op): Use the base type
in the interpretation of the operator.

2 years ago[Ada] Fix style of body header boxes
Piotr Trojanek [Thu, 3 Feb 2022 20:28:44 +0000 (21:28 +0100)]
[Ada] Fix style of body header boxes

gcc/ada/

* sem_case.adb (Free_Value_Sets): Fix style in body header box.
* sem_ch12.adb (Build_Subprogram_Wrappers): Likewise.
* sem_util.adb (Statically_Denotes_Entity): Likewise.

2 years ago[Ada] Do not create useless itype in Constrain_Access
Etienne Servais [Wed, 2 Feb 2022 14:18:29 +0000 (15:18 +0100)]
[Ada] Do not create useless itype in Constrain_Access

In the case of a constrained access definition for a record component we
are calling create_itype twice the former not being updated.  This leads
to a malformed node that crashes -gnatG when predicates are activated.
Instead of creating a default Itype for Desig_Subtype, create it with
the correct scope in each case.

gcc/ada/

* sem_ch3.adb (Constrain_Access): Call Desig_Subtype in each if
branch to avoid calling it twice.

2 years ago[Ada] Detect infinite loops with operators in exit conditions
Piotr Trojanek [Wed, 2 Feb 2022 16:19:59 +0000 (17:19 +0100)]
[Ada] Detect infinite loops with operators in exit conditions

To warn about infinite loops we detect variables referenced in loop exit
conditions. We handle references within boolean operators, i.e.
comparison and negation, which are likely to appear at the top level of
the condition (e.g. "X > 0"). However, we can easily handle all
operators, because they are likely to appear inside the condition (e.g.
"abs (X) > 0.0").

Cleanup related to a new restriction No_Uninitialized_Local_Scalars.

gcc/ada/

* sem_warn.adb (Find_Var): Detect all operators; replace
"condition" to "expression" in comments, because when this
routine is called recursively it no longer examines the
condition.
(Is_Suspicious_Function_Name): Reduce scope of a local variable
to avoid shadowing with a parameter of a nested
Substring_Present function.

2 years ago[Ada] Adapt proof of System.Arith_Double after update of Z3
Yannick Moy [Wed, 2 Feb 2022 11:52:36 +0000 (12:52 +0100)]
[Ada] Adapt proof of System.Arith_Double after update of Z3

Update to version 4.8.14 of prover Z3 requires minor adjustments of the
ghost code to add necessary intermediate assertions that drive the
automatic proof.

gcc/ada/

* libgnat/s-aridou.adb (Double_Divide, Scaled_Divide): Add
intermediate assertions.

2 years ago[Ada] Update categorization of implementation restrictions
Piotr Trojanek [Tue, 1 Feb 2022 14:21:01 +0000 (15:21 +0100)]
[Ada] Update categorization of implementation restrictions

Some of the restrictions were initially implemented as specific to GNAT,
but now they are part of Ada; likewise, some of recently added
restrictions were considered to be part of Ada, but are not yet in the
standard. This patch updates their categorization based on Ada 202x
Draft 32.

Only a No_Implementation_Restrictions restriction, which is specific to
GNAT, is affected.

Cleanup related to a new restriction No_Uninitialized_Local_Scalars.

gcc/ada/

* libgnat/s-rident.ads (Implementation_Restriction): Remove
No_Dynamic_Attachment, No_Implementation_Attributes,
No_Implementation_Pragmas, No_Local_Protected_Objects,
No_Protected_Type_Allocators, No_Relative_Delay,
No_Requeue_Statements, No_Select_Statements,
No_Task_Termination, Pure_Barriers and Simple_Barriers; add
No_Dynamic_Sized_Objects,
No_Implicit_Protected_Object_Allocations,
No_Implicit_Task_Allocations, No_Task_At_Interrupt_Priority and
Static_Dispatch_Tables; sort alphabetically.
* restrict.ads (Restriction_Id): Change Pure_Barriers and
No_Task_Termination from GNAT to Ada restrictions.

2 years ago[Ada] Fix "S p e c" and "B o d y" file headers
Piotr Trojanek [Tue, 1 Feb 2022 10:40:23 +0000 (11:40 +0100)]
[Ada] Fix "S p e c" and "B o d y" file headers

Fix Spec/Body headers in .adb/.ads files, respectively; all violations
detected with grep and manually filtered because of .tmpl files.

gcc/ada/

* exp_atag.adb, lib-util.ads, libgnat/g-decstr.adb,
libgnat/g-exptty.adb, libgnat/g-socpol.adb,
libgnat/s-strhas.adb: Fix headers.

2 years ago[Ada] Add SUSE 32bit dependency
Joao Azevedo [Mon, 31 Jan 2022 12:02:42 +0000 (12:02 +0000)]
[Ada] Add SUSE 32bit dependency

On SUSE 15, glibc-locale-base-32bit is needed to run 32bit versions
of Libadalang tools.

gcc/ada/

* doc/gnat_ugn/platform_specific_information.rst: Add
glibc-locale-base-32bit as a dependency in SUSE distributions.
* gnat_ugn.texi: Regenerate.

2 years ago[Ada] Fix formatting glitches in GNAT User's Guide
Piotr Trojanek [Sun, 30 Jan 2022 17:41:16 +0000 (18:41 +0100)]
[Ada] Fix formatting glitches in GNAT User's Guide

Only enhancement of formatting.

gcc/ada/

* doc/gnat_ugn/gnat_and_program_execution.rst,
doc/gnat_ugn/gnat_utility_programs.rst,
doc/gnat_ugn/inline_assembler.rst: Fix typos and formatting
glitches.
* gnat_ugn.texi: Regenerate.

2 years ago[Ada] Use pygments for Ada code examples of elaboration control
Piotr Trojanek [Sun, 30 Jan 2022 17:39:39 +0000 (18:39 +0100)]
[Ada] Use pygments for Ada code examples of elaboration control

Only enhancement of formatting.

gcc/ada/

* doc/gnat_ugn/elaboration_order_handling_in_gnat.rst: Change
blocks from plain code to Ada.

2 years ago[Ada] Excess finalization on assignment with target name symbol
Gary Dismukes [Fri, 28 Jan 2022 01:17:03 +0000 (20:17 -0500)]
[Ada] Excess finalization on assignment with target name symbol

In cases where the Ada 2022 feature of target name symbols (@) is used
and the evaluation of the name is side-effect free, the compiler creates
a temporary object to hold the value of the target object for use as
the value of "@" symbols in the right-hand side expression. In the case
where the target's type is controlled, or has controlled subcomponents,
this can result in extra finalization calls (for the temporary object).
The correction is to bypass the use of a temp and fall back on the more
general approach of creating and calling a procedure with an in-out
parameter for performing the assignment.

gcc/ada/

* exp_ch5.adb (Expand_Assign_With_Target_Names): Bypass the temp
object approach in the case where the type of the LHS is
controlled or has controlled subcomponents (Needs_Finalization
is True), and use the procedure approach instead.

2 years ago[Ada] fix crash on Secondary_Stack_Size with discriminant
Bob Duff [Fri, 28 Jan 2022 19:40:27 +0000 (14:40 -0500)]
[Ada] fix crash on Secondary_Stack_Size with discriminant

This patch fixes a crash caused by specifying the Secondary_Stack_Size
aspect of a task type as the value of a discriminant of the task type,
and then declaring a record component whose type is the task type,
constrained to a discriminant of the record.

gcc/ada/

* sem_ch3.adb (Get_Discr_Value): Copy the result. Otherwise, the
"tree" can be malformed (a DAG, not a tree) because
Get_Discr_Value could be returning some subtree already attached
to the tree.

2 years ago[Ada] Reset Reachable field when mutating label into loop entity
Piotr Trojanek [Fri, 28 Jan 2022 17:12:18 +0000 (18:12 +0100)]
[Ada] Reset Reachable field when mutating label into loop entity

An entity flag Reachable now only applies to E_Label entities and needs
to be explicitly reset when mutating labels into loop entities. Only
needed to prevent cascaded errors when compiling a malicious ACATS test
with -gnatq (try semantics, even if parse errors).

Cleanup related to detection of uninitialized scalars with GOTO
statements.

gcc/ada/

* sem_ch5.adb (Analyze_Statements): Only clear Reachable flag on
proper label entities.

2 years ago[Ada] Fix layout of tables in PDF version of GNAT RM
Piotr Trojanek [Fri, 28 Jan 2022 15:02:43 +0000 (16:02 +0100)]
[Ada] Fix layout of tables in PDF version of GNAT RM

Tables with small left columns and large right columns had colliding
text in the PDF version of GNAT RM.

Spotted while adding description of pragma Ada_2022.

gcc/ada/

* doc/gnat_rm/implementation_defined_characteristics.rst: Add
tabularcolums directive, as explained in documentation of Sphinx
and tabular LaTeX package.

2 years ago[Ada] Fix layout in description of aspects and pragmas
Piotr Trojanek [Fri, 28 Jan 2022 14:27:47 +0000 (15:27 +0100)]
[Ada] Fix layout in description of aspects and pragmas

Remove extra whitespace in examples of pragmas and aspects.

Spotted while adding description of pragma Ada_2022.

gcc/ada/

* doc/gnat_rm/implementation_defined_aspects.rst,
doc/gnat_rm/implementation_defined_pragmas.rst: Remove extra
whitespace.
* gnat_rm.texi: Regenerate.

2 years ago[Ada] Fix markup in description of implementation-defined characteristics
Piotr Trojanek [Fri, 28 Jan 2022 14:02:11 +0000 (15:02 +0100)]
[Ada] Fix markup in description of implementation-defined characteristics

Fix formatting of description lists, i.e. continuation lines are now
indented like their preceding lines.

Spotted while adding description of pragma Ada_2022.

gcc/ada/

* doc/gnat_rm/implementation_defined_characteristics.rst: Remove
extra indent in continuation lines; indent both descriptions of
Max_Picture_Length.
* gnat_rm.texi: Regenerate.

2 years ago[Ada] Fix markup in description of pragma Eliminate
Piotr Trojanek [Fri, 28 Jan 2022 13:16:09 +0000 (14:16 +0100)]
[Ada] Fix markup in description of pragma Eliminate

Fix formatting of a nested bullet lists. Spotted while adding
description of pragma Ada_2022.

gcc/ada/

* doc/gnat_rm/implementation_defined_pragmas.rst (Pragma
Eliminate): Fix markup.
* gnat_rm.texi: Regenerate.

2 years ago[Ada] Suppress warning in g-sthcso for non-unix platforms
Johannes Kliemann [Fri, 28 Jan 2022 14:54:21 +0000 (14:54 +0000)]
[Ada] Suppress warning in g-sthcso for non-unix platforms

gcc/ada/

* libgnat/g-sthcso.adb: Suppress warning about unneeded use type
clause.

2 years ago[Ada] Add guard for making only legal labels unreachable
Piotr Trojanek [Fri, 28 Jan 2022 09:24:04 +0000 (10:24 +0100)]
[Ada] Add guard for making only legal labels unreachable

An entity flag Reachable now only applies to E_Label entities. We had an
appropriate guard for setting this flag, but not for clearing.

Cleanup related to detection of uninitialized scalars with GOTO
statements.

gcc/ada/

* sem_ch5.adb (Analyze_Statements): Only clear Reachable flag on
proper label entities.

2 years ago[Ada] Document pragma Ada_2022
Piotr Trojanek [Thu, 27 Jan 2022 21:03:58 +0000 (22:03 +0100)]
[Ada] Document pragma Ada_2022

Pragma Ada_2022 is similar to existing pragma Ada_2012 and similarly
deserves to be documented.

gcc/ada/

* doc/gnat_rm/implementation_defined_pragmas.rst (Pragma
Ada_2022): Copy description from pragma Ada 2012 and adapt.
* doc/gnat_ugn/the_gnat_compilation_model.rst (Configuration
Pragmas): List pragma Ada 2022.
* gnat_rm.texi, gnat_ugn.texi: Regenerate.

2 years ago[Ada] Properly reject unsupported address specifications
Steve Baird [Tue, 25 Jan 2022 00:46:58 +0000 (16:46 -0800)]
[Ada] Properly reject unsupported address specifications

In the case of an object declaration with an indefinite nominal subtype
(roughly speaking, that's an object that takes its bounds,
discriminants, and/or tag from its explicit initial value), GNAT does
not support address specifications unless the size of the object is
known at compile time.  In some cases, such unsupported address
specifications were not properly rejected. This could lead to either an
internal error during compilation or (in the class-wide case) to a
warning accompanied by raising Program_Error at run time.

gcc/ada/

* sem_ch13.adb (Analyze_Attribute_Definition_Clause): Replace
the existing check for an address specification for an object of
a class-wide type with a more general check which rejects either
the class-wide case or the case where the FE would (if the
address specification were accepted) build a malformed
tree (specifically, an object renaming declaration with a
specified address). In the case where the check fails, reject
the construct at compile time instead of generating an
unconditional raise of Program_Error.
* doc/gnat_rm/representation_clauses_and_pragmas.rst: Update
documentation to reflect these changes.
* gnat_rm.texi: Regenerate.

2 years ago[Ada] Move Reachable flag to E_Label entities
Piotr Trojanek [Thu, 27 Jan 2022 11:59:58 +0000 (12:59 +0100)]
[Ada] Move Reachable flag to E_Label entities

An entity flag Reachable is described as "defined in labels". It is only
read and written for entities of kind E_Label (the code has necessary
guards for that). There no need for this flag to be specified for all
entities in the generated AST.

Cleanup related to detection of uninitialized scalars with GOTO
statements.

gcc/ada/

* gen_il-gen-gen_entities.adb (Reachable): Move from
Root_Entity_Type to E_Label.

2 years ago[Ada] Set_Is_Known_Valid only if Safe_To_Capture_Value
Alexandre Oliva [Wed, 29 Dec 2021 07:10:45 +0000 (04:10 -0300)]
[Ada] Set_Is_Known_Valid only if Safe_To_Capture_Value

Library-level variables with initializers could have Is_Known_Valid
set when analyzing their definition, and the flag would only be
cleared when analyzing a statement that assigned to them. Procedures
and functions analyzed before the flag got cleared could skip validity
checking for the corresponding variable. This patch fixes this
problem: we no longer set Is_Known_Valid when analyzing initializers
of library-level variables,and use the same Safe_To_Capture_Value
predicate that prevents assignments from recording known-valid states.

This causes any variable with an initialization value, that would have
had its initializer value used as its known constant value if the use
is analyzed before any assignment to the variable, to no longer be
regarded as holding a constant value. Some might turn out to have a
constant value, after all, but we don't know that yet: we can only
tell after analyzing every subprogram that could possibly assign to
it. At the points where Safe_To_Capture_Value calls are introduced,
Never_Set_In_Source does not yet hold its final value.

gcc/ada/

* exp_ch3.adb (Expand_N_Object_Declaration): Guard
Set_Is_Known_Valid with Safe_To_Capture_Value.

2 years ago[Ada] Avoid crash for -gnatR -gnatc
Bob Duff [Thu, 27 Jan 2022 16:13:41 +0000 (11:13 -0500)]
[Ada] Avoid crash for -gnatR -gnatc

If the -gnatR -gnatc are both given, then the compiler crashes.  This
patch fixes that, and avoids printing the uncomputed sizes and
alignments that were causing the crash. (Previous versions of the
compiler printed incorrect values in such cases.)

gcc/ada/

* repinfo.adb (List_Object_Info): Do not try to print values
that have not been computed (and so are No_Uint).
(Rep_Not_Constant): Reverse sense and change name to
Compile_Time_Known_Rep. This makes the code at call sites a
little more readable. Simplify code to a single return
statement.

2 years agobootstrap/105551 - restore nvptx build
Richard Biener [Wed, 11 May 2022 08:47:34 +0000 (10:47 +0200)]
bootstrap/105551 - restore nvptx build

The following makes sure to disable var-tracking if only
dwarf2-line debuginfo is present.

2022-05-11  Richard Biener  <rguenther@suse.de>

PR bootstrap/105551
* opts.cc (finish_options): Also disable var-tracking if
!DWARF2_DEBUGGING_INFO.

2 years agoOptimize movzwl + vmovd/vmovq to vmovw.
liuhongt [Tue, 29 Mar 2022 01:21:21 +0000 (09:21 +0800)]
Optimize movzwl + vmovd/vmovq to vmovw.

Similarly optimize movl + vmovq to vmovd.

gcc/ChangeLog:

PR target/104915
* config/i386/sse.md (*vec_set<mode>_0_zero_extendhi): New
pre_reload define_insn_and_split.
(*vec_setv2di_0_zero_extendhi_1): Ditto.
(*vec_set<mode>_0_zero_extendsi): Ditto.
(*vec_setv2di_0_zero_extendsi_1): Ditto.
(ssewvecmode): New mode attr.
(ssewvecmodelower): Ditto.
(ssepackmodelower): Ditto.

gcc/testsuite/ChangeLog:

* gcc.target/i386/pr104915-vmovd.c: New test.
* gcc.target/i386/pr104915-vmovw.c: New test.

2 years agors6000: Fix constraint v with rs6000_constraints[RS6000_CONSTRAINT_v]
Kewen Lin [Wed, 11 May 2022 06:18:30 +0000 (01:18 -0500)]
rs6000: Fix constraint v with rs6000_constraints[RS6000_CONSTRAINT_v]

Currently we hardcode ALTIVEC_REGS for register constraint v and
RS6000_CONSTRAINT_v looks confusing, this is to fix it with
rs6000_constraints[RS6000_CONSTRAINT_v] to align with some other
existing register constraints with RS6000_CONSTRAINT_*.

gcc/ChangeLog:

* config/rs6000/constraints.md (register constraint v): Use
rs6000_constraints[RS6000_CONSTRAINT_v] instead of ALTIVEC_REGS.

2 years agoopts: do not allow Separate+Joined ending with =
Martin Liska [Tue, 26 Apr 2022 09:41:09 +0000 (11:41 +0200)]
opts: do not allow Separate+Joined ending with =

Do not allow strange option format:
-msmall-data-limit= 11.

For -output-pch we need to use Separate syntax as lang spec
rules automatically append a space.

PR target/105355

gcc/ChangeLog:

* config/riscv/riscv.opt: Remove Separate from
-msmall-data-limit=.
* optc-gen.awk: Report error for the described situation.
* gcc.cc: Use Separate syntax.
* opts.cc (gen_command_line_string): Change option name.

gcc/c-family/ChangeLog:

* c-opts.cc (c_common_handle_option): Change option name.
* c.opt: Remove Joined and use Separate option.

gcc/cp/ChangeLog:

* lang-specs.h: Use Separate syntax.

gcc/objc/ChangeLog:

* lang-specs.h: Use Separate syntax.

gcc/objcp/ChangeLog:

* lang-specs.h: Use Separate syntax.

2 years agoMAINTAINERS: Remove myself from DCO section
Kewen Lin [Wed, 11 May 2022 03:18:24 +0000 (22:18 -0500)]
MAINTAINERS: Remove myself from DCO section

ChangeLog:

* MAINTAINERS: Remove myself from DCO section.

2 years agoThis patch skips constant folding for fmin/max when either argument is sNaN. Accordin...
Haochen Gui [Mon, 9 May 2022 09:34:23 +0000 (17:34 +0800)]
This patch skips constant folding for fmin/max when either argument is sNaN. According to C standard, fmin(sNaN, sNaN)= qNaN, fmin(sNaN, NaN) = qNaN.

gcc/
PR target/105414
* match.pd (minmax): Skip constant folding for fmin/fmax when both
arguments are sNaN or one is sNaN and another is NaN.

gcc/testsuite/
PR target/105414
* gcc.dg/pr105414.c: New.

2 years agoDaily bump.
GCC Administrator [Wed, 11 May 2022 00:16:40 +0000 (00:16 +0000)]
Daily bump.

2 years agoUpdate gcc hr.po
Joseph Myers [Tue, 10 May 2022 20:22:22 +0000 (20:22 +0000)]
Update gcc hr.po

* hr.po: Update.

2 years agoFortran: check TEAM arguments to coarray intrinsics
Harald Anlauf [Mon, 9 May 2022 20:14:21 +0000 (22:14 +0200)]
Fortran: check TEAM arguments to coarray intrinsics

TEAM arguments to coarray intrinsics must be scalar expressions of type
TEAM_TYPE of intrinsic module ISO_FORTRAN_ENV.

gcc/fortran/ChangeLog:

PR fortran/105526
* resolve.cc (check_team): New.
(gfc_resolve_code): Add checks for arguments to coarray intrinsics
FORM TEAM, CHANGE TEAM, and SYNC TEAM.

gcc/testsuite/ChangeLog:

PR fortran/105526
* gfortran.dg/coarray_50.f90: New test.

2 years agox86: Add .note.GNU-stack section only for Linux
H.J. Lu [Fri, 6 May 2022 17:55:53 +0000 (10:55 -0700)]
x86: Add .note.GNU-stack section only for Linux

Add .note.GNU-stack section only for Linux since it may not be supported
on non-Linux OSes.  __ELF__ isn't checked since these tests can only run
on Linux/x86 ELF systems.

PR target/105472
* gcc.target/i386/iamcu/asm-support.S: Add .note.GNU-stack section
only for Linux.
* gcc.target/x86_64/abi/asm-support.S: Likewise.
* gcc.target/x86_64/abi/avx/asm-support.S: Likewise.
* gcc.target/x86_64/abi/avx512f/asm-support.S: Likewise.
* gcc.target/x86_64/abi/avx512fp16/asm-support.S: Likewise.
* gcc.target/x86_64/abi/avx512fp16/m256h/asm-support.S: Likewise.
* gcc.target/x86_64/abi/avx512fp16/m512h/asm-support.S: Likewise.
* gcc.target/x86_64/abi/ms-sysv/do-test.S: Likewise.

2 years agoisel: Fix up gimple_expand_vec_set_expr [PR105528]
Jakub Jelinek [Tue, 10 May 2022 16:34:08 +0000 (18:34 +0200)]
isel: Fix up gimple_expand_vec_set_expr [PR105528]

The following testcase ICEs (and only without -g), because we don't replace
one VEC_COND_EXPR with .VCOND* call.
We don't do that because gimple_expand_vec_set_expr adds some stmts before
*gsi and then uses gsi_remove to remove it.  gsi_remove moves the iterator
to the next stmt and in the caller we then do gsi_next before looking at
another stmt, which means we can skip processing of one stmt, which in this
case happened to be a VEC_COND_EXPR but with -g is some debug stmt in
between.  As we always emit some stmts before it, it is easy to update the
iterator to the last stmt emitted there, so that caller continues really
with the next stmt.

2022-05-10  Jakub Jelinek  <jakub@redhat.com>

PR tree-optimization/105528
* gimple-isel.cc (gimple_expand_vec_set_expr): After gsi_remove
set *gsi to gsi_for_stmt (ass_stmt).  Fix up function comment.

* gcc.dg/pr105528.c: New test.

2 years agodoc: Fix mode iterator example
Segher Boessenkool [Tue, 10 May 2022 16:00:05 +0000 (16:00 +0000)]
doc: Fix mode iterator example

The example missed the mode condition in the replacement text.

2022-05-10  Segher Boessenkool  <segher@kernel.crashing.org>

* doc/md.texi (Defining Mode Iterators): Correct example replacement
text.

2 years agoc++: Remove SET_PACK_EXPANSION_PATTERN / SET_ARGUMENT_PACK_ARGS
Patrick Palka [Tue, 10 May 2022 15:52:29 +0000 (11:52 -0400)]
c++: Remove SET_PACK_EXPANSION_PATTERN / SET_ARGUMENT_PACK_ARGS

Unlike in C, in C++ the conditional operator yields an lvalue if both
branches are lvalues, so these setter macros are unnecessary; we can
just assign to PACK_EXPANSION_PATTERN and ARGUMENT_PACK_ARGS directly.

gcc/cp/ChangeLog:

* coroutines.cc (instantiate_coro_traits): Adjust accordingly.
* cp-tree.def: Remove mention of SET_PACK_EXPANSION_PATTERN.
* cp-tree.h (SET_PACK_EXPANSION_PATTERN): Remove.
(SET_ARGUMENT_PACK_ARGS): Remove.
* module.cc (trees_in::tree_node): Adjust accordingly.
* parser.cc (make_char_string_pack): Likewise.
(make_string_pack): Likewise.
* pt.cc (make_pack_expansion): Likewise.
(template_parm_to_arg): Likewise.
(coerce_template_parameter_pack): Likewise.
(extract_fnparm_pack): Likewise.
(extract_locals_r): Likewise.
(make_argument_pack): Likewise.
(tsubst_argument_pack): Likewise.
(lookup_init_capture_pack): Likewise.
(type_unification_real): Likewise.
(unify_pack_expansion): Likewise.
(tsubst_initializer_list): Likewise.

2 years agoc++: Harden *_PACK_EXPANSION and *_ARGUMENT_PACK macros
Patrick Palka [Tue, 10 May 2022 15:52:21 +0000 (11:52 -0400)]
c++: Harden *_PACK_EXPANSION and *_ARGUMENT_PACK macros

This makes the accessor macros for TYPE_/EXPR_PACK_EXPANSION
and TYPE_/NONTYPE_ARGUMENT_PACK check the tree code of the argument.

gcc/cp/ChangeLog:

* cp-tree.h (PACK_EXPANSION_CHECK): Define.
(PACK_EXPANSION_PATTERN): Use PACK_EXPANSION_CHECK.
(SET_PACK_EXPANSION_PATTERN): Likewise.
(PACK_EXPANSION_PARAMETER_PACKS): Likewise.
(PACK_EXPANSION_EXTRA_ARGS): Likewise.
(PACK_EXPANSION_LOCAL_P): Likewise.
(PACK_EXPANSION_SIZEOF_P): Likewise.
(PACK_EXPANSION_AUTO_P): Likewise.
(PACK_EXPANSION_FORCE_EXTRA_ARGS_P): Likewise.
(ARGUMENT_PACK_CHECK): Define.
(ARGUMENT_PACK_ARGS): Use ARGUMENT_PACK_CHECK.
(SET_ARGUMENT_PACK_ARGS): Likewise.
* parser.cc (cp_parser_sizeof_pack): Check for error_mark_node
before setting PACK_EXPANSION_SIZEOF_P.

2 years agolibstdc++: Add <spanstream> declarations to <iosfwd> [PR105284]
Jonathan Wakely [Tue, 10 May 2022 12:06:47 +0000 (13:06 +0100)]
libstdc++: Add <spanstream> declarations to <iosfwd> [PR105284]

libstdc++-v3/ChangeLog:

PR libstdc++/105284
* include/std/iosfwd: Add declarations for <spanstream> class
templates and typedefs.
* include/std/spanstream (basic_spanbuf, basic_ispanstream)
(basic_ospanstream, basic_spanstream): Remove default template
arguments.
* testsuite/27_io/headers/iosfwd/synopsis.cc: Add <spanstream>
declarations.
* testsuite/27_io/headers/iosfwd/types.cc: Check <spanstream>
default arguments.

2 years agolibstdc++: Add <syncstream> declarations to <iosfwd> [PR105284]
Jonathan Wakely [Tue, 10 May 2022 12:03:14 +0000 (13:03 +0100)]
libstdc++: Add <syncstream> declarations to <iosfwd> [PR105284]

libstdc++-v3/ChangeLog:

PR libstdc++/105284
* include/std/iosfwd: Add declarations for <synstream> class
templates and typedefs.
* include/std/syncstream (basic_syncbuf, basic_osyncstream):
Remove default template arguments.
* testsuite/27_io/headers/iosfwd/synopsis.cc: New test.
* testsuite/27_io/headers/iosfwd/types.cc: New test.

2 years agolibiberty: fix wrong replacent in comments
Martin Liska [Tue, 10 May 2022 15:36:28 +0000 (17:36 +0200)]
libiberty: fix wrong replacent in comments

libiberty/ChangeLog:

* regex.c: Restore comments.

2 years agolibiberty: fix type in allocation
Martin Liska [Tue, 10 May 2022 15:31:24 +0000 (17:31 +0200)]
libiberty: fix type in allocation

The allocation function alloc_f is called for nsize
items where each is of type void *.

libiberty/ChangeLog:

* hashtab.c (htab_empty): Use void * type instead of void **.
(htab_expand): Likewise.

Co-Authored-By: Alan Modra <amodra@gmail.com>
2 years agolibiberty: fix bad replacement.
Martin Liska [Tue, 10 May 2022 15:00:34 +0000 (17:00 +0200)]
libiberty: fix bad replacement.

libiberty/ChangeLog:

* random.c: Remove 'define PTR'.

2 years agoRemove non-ANSI C macros in ansidecl.h.
Martin Liska [Tue, 10 May 2022 07:47:08 +0000 (09:47 +0200)]
Remove non-ANSI C macros in ansidecl.h.

include/ChangeLog:

* ansidecl.h (PTR): Remove.
(const): Likewise.
(volatile): Likewise.
(signed): Likewise.

2 years agolibiberty: stop using PTR macro
Martin Liska [Tue, 10 May 2022 14:00:53 +0000 (16:00 +0200)]
libiberty: stop using PTR macro

include/ChangeLog:

* hashtab.h (HTAB_EMPTY_ENTRY): Use void * instead PTR.
(HTAB_DELETED_ENTRY): Likewise.

libiberty/ChangeLog:

* alloca.c (C_alloca): Use void * instead PTR.
* calloc.c (malloc): Likewise.
(bzero): Likewise.
(calloc): Likewise.
* hashtab.c (find_empty_slot_for_expand): Likewise.
(eq_pointer): Likewise.
(htab_create_alloc_ex): Likewise.
(htab_create_typed_alloc): Likewise.
(htab_set_functions_ex): Likewise.
(htab_delete): Likewise.
(htab_empty): Likewise.
(htab_expand): Likewise.
(htab_find_with_hash): Likewise.
(htab_find): Likewise.
(htab_find_slot_with_hash): Likewise.
(htab_find_slot): Likewise.
(htab_remove_elt): Likewise.
(htab_remove_elt_with_hash): Likewise.
(htab_clear_slot): Likewise.
(htab_traverse_noresize): Likewise.
(htab_traverse): Likewise.
(htab_hash_string): Likewise.
(iterative_hash): Likewise.
(hash_pointer): Likewise.
* memchr.c (memchr): Likewise.
* memcmp.c (memcmp): Likewise.
* memcpy.c (memcpy): Likewise.
* memmove.c (memmove): Likewise.
* mempcpy.c (memcpy): Likewise.
(mempcpy): Likewise.
* memset.c (memset): Likewise.
* objalloc.c (malloc): Likewise.
(free): Likewise.
(objalloc_create): Likewise.
(_objalloc_alloc): Likewise.
(objalloc_free_block): Likewise.
* random.c (PTR): Likewise.
(void): Likewise.
(initstate): Likewise.
(setstate): Likewise.
* regex.c: Likewise.
* spaces.c (malloc): Likewise.
(free): Likewise.
* stpcpy.c (memcpy): Likewise.
* strdup.c (malloc): Likewise.
(memcpy): Likewise.
* strerror.c (malloc): Likewise.
(memset): Likewise.
* strndup.c (malloc): Likewise.
(memcpy): Likewise.
* strsignal.c (malloc): Likewise.
(memset): Likewise.
* vasprintf.c (malloc): Likewise.
* vprintf-support.c: Likewise.
* xatexit.c (malloc): Likewise.
* xmalloc.c (xmalloc): Likewise.
(xcalloc): Likewise.
(xrealloc): Likewise.
* xmemdup.c (xmemdup): Likewise.

2 years agoRISC-V: Provide `fmin'/`fmax' RTL patterns
Maciej W. Rozycki [Tue, 10 May 2022 14:02:31 +0000 (15:02 +0100)]
RISC-V: Provide `fmin'/`fmax' RTL patterns

As at r2.2 of the RISC-V ISA specification[1] (equivalent to version 2.0
of the "F" and "D" standard architecture extensions for single-precision
and double-precision floating-point respectively) the FMIN and FMAX
machine instructions fully match our requirement for the `fminM3' and
`fmaxM3' standard RTL patterns:

"For FMIN and FMAX, if at least one input is a signaling NaN, or if both
inputs are quiet NaNs, the result is the canonical NaN.  If one operand
is a quiet NaN and the other is not a NaN, the result is the non-NaN
operand."

suitably for the IEEE 754-2008 `minNum' and `maxNum' operations.

However we only define `sminM3' and `smaxM3' standard RTL patterns to
produce the FMIN and FMAX machine instructions, which in turn causes the
`__builtin_fmin' and `__builtin_fmax' family of intrinsics to emit the
corresponding libcalls rather than the relevant machine instructions.
This is according to earlier revisions of the RISC-V ISA specification,
which we however do not support anymore, as from commit 4b81528241ca
("RISC-V: Support version controling for ISA standard extensions").

As from r20190608 of the RISC-V ISA specification (equivalent to version
2.2 of the "F" and "D" standard ISA extensions for single-precision and
double-precision floating-point respectively) the definition of the FMIN
and FMAX machine instructions has been updated[2]:

"Defined the signed-zero behavior of FMIN.fmt and FMAX.fmt, and changed
their behavior on signaling-NaN inputs to conform to the minimumNumber
and maximumNumber operations in the proposed IEEE 754-201x
specification."

and specifically[3]:

"Floating-point minimum-number and maximum-number instructions FMIN.S
and FMAX.S write, respectively, the smaller or larger of rs1 and rs2 to
rd.  For the purposes of these instructions only, the value -0.0 is
considered to be less than the value +0.0.  If both inputs are NaNs, the
result is the canonical NaN.  If only one operand is a NaN, the result
is the non-NaN operand.  Signaling NaN inputs set the invalid operation
exception flag, even when the result is not NaN."

Consequently for forwards compatibility with r20190608+ hardware we
cannot use the FMIN and FMAX machine instructions unconditionally even
where the ISA level of r2.2 has been specified with the `-misa-spec=2.2'
option where operation would be different between ISA revisions, that
is the handling of signaling NaN inputs.

Therefore provide new `fmin<mode>3' and `fmax<mode>3' patterns removing
the need to emit libcalls with the `__builtin_fmin' and `__builtin_fmax'
family of intrinsics, however limit them to where `-fno-signaling-nans'
is in effect, deferring to other code generation strategies otherwise as
applicable.  Use newly-defined UNSPECs as the operation codes so that
the patterns are only ever used if referred to by their names, as there
is no RTL expression defined for the IEEE 754-2008 `minNum' and `maxNum'
operations.

References:

[1] "The RISC-V Instruction Set Manual, Volume I: User-Level ISA",
    Document Version 2.2, May 7, 2017, Section 8.3 "NaN Generation and
    Propagation", p. 48

[1] "The RISC-V Instruction Set Manual, Volume I: Unprivileged ISA",
    Document Version 20190608-Base-Ratified, June 8, 2019, "Preface",
    p. ii

[2] same, Section 11.6 "Single-Precision Floating-Point Computational
    Instructions", p. 66

gcc/
* config/riscv/riscv.md (UNSPEC_FMIN, UNSPEC_FMAX): New
constants.
(fmin<mode>3, fmax<mode>3): New insns.

gcc/testsuite/
* gcc.target/riscv/fmax-snan.c: New test.
* gcc.target/riscv/fmax.c: New test.
* gcc.target/riscv/fmaxf-snan.c: New test.
* gcc.target/riscv/fmaxf.c: New test.
* gcc.target/riscv/fmin-snan.c: New test.
* gcc.target/riscv/fmin.c: New test.
* gcc.target/riscv/fminf-snan.c: New test.
* gcc.target/riscv/fminf.c: New test.
* gcc.target/riscv/smax-ieee.c: New test.
* gcc.target/riscv/smax.c: New test.
* gcc.target/riscv/smaxf-ieee.c: New test.
* gcc.target/riscv/smaxf.c: New test.
* gcc.target/riscv/smin-ieee.c: New test.
* gcc.target/riscv/smin.c: New test.
* gcc.target/riscv/sminf-ieee.c: New test.
* gcc.target/riscv/sminf.c: New test.