[multiple changes]

2017-04-25  Arnaud Charlet  <charlet@adacore.com>

* a-cfinve.ads, a-cofove.ads (Empty_Vector): add Global contract.

2017-04-25  Justin Squirek  <squirek@adacore.com>

* sem_ch3.adb (Analyze_Declarations): Minor
correction to comments, move out large conditional and scope
traversal into a predicate.
(Uses_Unseen_Lib_Unit_Priv): Predicate function made from extracted
logic.

2017-04-25  Ed Schonberg  <schonberg@adacore.com>

* sem_ch4.adb (Analyze_Selected_Component): Refine analysis
of prefix whose type is a current instance of a synchronized
type. If the prefix is an object this is an external call (or
requeue) that can only access public operations of the object. The
previous predicate was too restrictive, and did not allow public
protected operations, only task entries.

2017-04-25  Hristian Kirtchev  <kirtchev@adacore.com>

* sem_ch5.adb, fname.adb: Minor reformatting.

2017-04-25  Hristian Kirtchev  <kirtchev@adacore.com>

* einfo.adb (Is_Anonymous_Access_Type): New routine.
* einfo.ads Update the placement of
E_Anonymous_Access_Subprogram_Type along with all subtypes that
mention the ekind.
(Is_Anonymous_Access_Type): New routine.
* exp_ch7.adb (Allows_Finalization_Master): Do not generate a
master for an access type subject to pragma No_Heap_Finalization.
* exp_util.adb (Build_Allocate_Deallocate_Proc): An object being
allocated or deallocated does not finalization actions if the
associated access type is subject to pragma No_Heap_Finalization.
* opt.ads Add new global variable No_Heap_Finalization_Pragma.
* par-prag.adb Pragma No_Heap_Finalization does not need special
processing from the parser.
* sem_ch6.adb (Check_Return_Subtype_Indication): Remove ancient
??? comments. Use the new predicate Is_Anonymous_Access_Type.
* sem_prag.adb Add an entry in table Sig_Flags for pragma
No_Heap_Finalization.
(Analyze_Pragma): Add processing for
pragma No_Heap_Finalization. Update various error messages to
use Duplication_Error.
* sem_util.ads, sem_util.adb (No_Heap_Finalization): New routine.
* snames.ads-tmpl: Add new predefined name No_Heap_Finalization
and corresponding pragma id.

From-SVN: r247156

[multiple changes]

2017-04-25  Bob Duff  <duff@adacore.com>

* freeze.adb (Freeze_Record_Type): Use the
underlying type of the component type to determine whether it's
elementary. For representation clause purposes, a private type
should behave the same as its full type.
* fname.ads, fname.adb (Is_Predefined_File_Name):
Make sure things like "system.ali" are recognized as predefined.

2017-04-25  Javier Miranda  <miranda@adacore.com>

* debug.adb: Update documentation of -gnatd.6.

2017-04-25  Ed Schonberg  <schonberg@adacore.com>

* sem_ch5.adb (Preanalyze_Range): Handle properly an Ada2012
element iterator when the name is an overloaded function call,
one of whose interpretations yields an array.

From-SVN: r247155

[multiple changes]

2017-04-25  Bob Duff  <duff@adacore.com>

* uname.ads, uname.adb (Is_Predefined_Unit_Name,
Is_Internal_Unit_Name): New functions for operating on unit
names, as opposed to file names. There's some duplicated code
with fname.adb, which is unfortunate, but it seems like we don't
want to add dependencies here.
* fname-uf.adb (Get_File_Name): Change Is_Predefined_File_Name
to Is_Predefined_Unit_Name; the former was wrong, because Uname
is not a file name at all.
* fname.ads, fname.adb: Document the fact that
Is_Predefined_File_Name and Is_Internal_File_Name can be called
for ALI files, and fix the code so it works properly for ALI
files. E.g. these should return True for "system.ali".

2017-04-25  Justin Squirek  <squirek@adacore.com>

* exp_util.adb (Add_Invariant): Removed,
code moved to Add_Invariant_Check, Add_Inherited_Invariant,
and Add_Own_Invariant. (Add_Invariant_Check): Used
for adding runtime checks from any kind of invariant.
(Add_Inherited_Invariant): Generates invariant checks for
class-wide invariants (Add_Interface_Invariants): Removed, code
moved to Build_Invariant_Procedure_Body (Add_Own_Invariant):
Create a types own invariant procedure (Add_Parent_Invariants):
Removed, code moved to Build_Invariant_Procedure_Body
(Build_Invariant_Procedure_Body): Add refactored calls
and integrated code from Add_Parent_Invariants and
Add_Interface_Invariants.
(Process_Type): Removed, the
relavant code was inlined into both Add_Own_Invariant and
Add_Inherited_Invariant.

From-SVN: r247154

[multiple changes]

2017-04-25  Hristian Kirtchev  <kirtchev@adacore.com>

* make.adb, par-ch2.adb, sem_util.adb, scans.ads, sem_ch8.adb,
scn.adb, osint.adb, fname.adb: Minor reformatting.

2017-04-25  Pascal Obry  <obry@adacore.com>

* s-taprop-mingw.adb: Do not check for CloseHandle in
Finalize_TCB.

From-SVN: r247153

[multiple changes]

2017-04-25  Hristian Kirtchev  <kirtchev@adacore.com>

* sem_util.adb (Check_Part_Of_Reference):
Continue to examine the context if the reference appears within
an expression function.

2017-04-25  Justin Squirek  <squirek@adacore.com>

* exp_ch7.adb, exp_ch7.ads Remove Build_Invariant_Procedure_Body
and Build_Invariant_Procedure_Declaration.
* exp_util.ads, exp_util.adb Add Build_Invariant_Procedure_Body
and Build_Invariant_Procedure_Declaration from exp_ch7
and break-out Is_Untagged_Private_Derivation from
Build_Invariant_Procedure_Body.
(Replace_Type_References):
Append an extra parameter to allow for dispatching replacements
and add the corrasponding logic.
(Type_Invariant): Remove
Replace_Typ_Refs and replace its references with calls to
Replace_Type_References.
* sem_ch3.adb, sem_prag.adb: Remove with and use of exp_ch7.

From-SVN: r247152

[multiple changes]

2017-04-25  Bob Duff  <duff@adacore.com>

* sem_util.ads, sem_util.adb (Should_Ignore_Pragma): New function
that returns True when appropriate.
* par-prag.adb, exp_prag.adb, sem_prag.adb: Do not ignore pragmas
when compiling predefined files.
* fname.ads, fname.adb (Is_Predefined_File_Name): Fix bug:
"gnat.adc" should not be considered a predefined file name.
That required (or at least encouraged) a lot of cleanup of global
variable usage. We shouldn't be communicating information via
the global name buffer.
* bindgen.adb, errout.adb, fname-uf.adb, lib-load.adb, make.adb,
* restrict.adb, sem_ch10.adb, sem_ch6.adb, sem_ch8.adb: Changes
required by the above-mentioned cleanup.

2017-04-25  Ed Schonberg  <schonberg@adacore.com>

* osint.adb (Find_File): Handle properly a request for a
configuration file whose name is a directory.

From-SVN: r247151

[multiple changes]

2017-04-25  Hristian Kirtchev  <kirtchev@adacore.com>

* sem_attr.adb, sem_ch5.adb: Minor reformatting.

2017-04-25  Bob Duff  <duff@adacore.com>

* types.ads: Minor: Fix '???' comment.
* sem_ch8.adb: Minor comment fix.

2017-04-25  Bob Duff  <duff@adacore.com>

* sem_prag.adb: Remove suspicious uses of Name_Buf.
* stringt.ads, stringt.adb, exp_dbug.adb, sem_dim.adb: Remove
Add_String_To_Name_Buffer, to avoid using the global Name_Buf.
Add String_To_Name with no side effects.

2017-04-25  Justin Squirek  <squirek@adacore.com>

* sem_ch3.adb (Analyze_Declarations): Add
additional condition for edge case.

2017-04-25  Bob Duff  <duff@adacore.com>

* par-ch2.adb, scans.ads, scn.adb: Do not give an error for
reserved words inside pragmas. This is necessary to allow the
pragma name Interface to be used in pragma Ignore_Pragma.
* par.adb: Minor comment fix.

2017-04-25  Javier Miranda  <miranda@adacore.com>

* a-tags.ads, a-tags.adb (Type_Is_Abstract): Renamed as Is_Abstract.
* rtsfind.ads (RE_Type_Is_Abstract): Renamed as Is_Abstract.
* exp_disp.adb (Make_DT): Update occurrences of RE_Type_Is_Abstract.
* exp_intr.adb (Expand_Dispatching_Constructor_Call): Update
occurrences of RE_Type_Is_Abstract

2017-04-25  Hristian Kirtchev  <kirtchev@adacore.com>

* exp_util.adb (Build_Chain): Account for ancestor
subtypes while traversing the derivation chain.

From-SVN: r247150

[multiple changes]

2017-04-25  Ed Schonberg  <schonberg@adacore.com>

* sem_attr.adb: minor reformatting.

2017-04-25  Doug Rupp  <rupp@adacore.com>

* sigtramp-vxworks-target.inc [PPC64]: Add a .localentry pseudo-op.

2017-04-25  Ed Schonberg  <schonberg@adacore.com>

* sem_ch5.adb (Analyze_Assignment): Reset Full_Analysis flag on
the first pass over an assignment statement with target names,
to prevent the generation of subtypes (such as discriminated
record components)that may carry the target name outside of the
tree for the assignment. The subtypes will be generated when
the assignment is reanalyzed in full.
(Analyze_Target_Name): Handle properly class-wide types.

From-SVN: r247149

[multiple changes]

2017-04-25  Hristian Kirtchev  <kirtchev@adacore.com>

* elists.ads, elists.adb (Prepend_Unique_Elmt): New routine.
* exp_ch3.adb (Freeze_Type): Signal the DIC body is created for
the purposes of freezing.
* exp_util.adb Update the documentation and structure of the
type map used in class-wide semantics of assertion expressions.
(Add_Inherited_Tagged_DIC): There is really no need to preanalyze
and resolve the triaged expression because all substitutions
refer to the proper entities.  Update the replacement of
references.
(Build_DIC_Procedure_Body): Add formal parameter
For_Freeze. Add local variable Build_Body. Inherited DIC pragmas
are now only processed when freezing occurs.  Build a body only
when one is needed.
(Entity_Hash): Removed.
(Map_Types): New routine.
(Replace_Object_And_Primitive_References): Removed.
(Replace_References): New routine.
(Replace_Type_References): Moved to the library level of Exp_Util.
(Type_Map_Hash): New routine.
(Update_Primitives_Mapping): Update the mapping call.
(Update_Primitives_Mapping_Of_Types): Removed.
* exp_util.ads (Build_DIC_Procedure_Body): Add formal
parameter For_Freeze and update the comment on usage.
(Map_Types): New routine.
(Replace_References): New routine.
(Replace_Type_References): Moved to the library level of Exp_Util.
(Update_Primitives_Mapping_Of_Types): Removed.
* sem_ch7.adb (Preserve_Full_Attributes): Propagate the DIC
properties of the private type to the full view in case the full
view derives from a parent type and inherits a DIC pragma.
* sem_prag.adb (Analyze_Pragma): Guard against a case where a
DIC pragma is placed at the top of a declarative region.

2017-04-25  Arnaud Charlet  <charlet@adacore.com trojanek>

* a-tasatt.adb: Complete previous change and use an unsigned
int to avoid overflow checks.

2017-04-25  Ed Schonberg  <schonberg@adacore.com>

* sem_attr.adb (Analyze_Attribute, case 'Access): Specialize
the error message when the attribute reference is an actual in
a call to a subprogram inherited from a generic formal type with
unknown discriminants, which makes the subprogram and its formal
parameters intrinsic (see RM 6.3.1 (8) and (13)).

From-SVN: r247148

[multiple changes]

2017-04-25  Hristian Kirtchev  <kirtchev@adacore.com>

* sem_aggr.adb, inline.adb, einfo.adb, einfo.ads, scng.adb,
sem_prag.adb: Minor reformatting.

2017-04-25  Bob Duff  <duff@adacore.com>

* sem_attr.adb (Type_Key): Add code in the
recursive Compute_Type_Key to protect against fetching the source
code for Standard, in case a component of the type is declared
in Standard. There was already code to do this for the original
type, but not for its components.

From-SVN: r247147

[multiple changes]

2017-04-25  Javier Miranda  <miranda@adacore.com>

* exp_ch3.adb (Build_Initialization_Call): Handle
subtypes of private types when searching for the underlying full
view of a private type.

2017-04-25  Javier Miranda  <miranda@adacore.com>

* sem_res.adb (Set_Mixed_Mode_Operand): A universal
real conditional expression can appear in a fixed-type context
and must be resolved with that context to facilitate the code
generation to the backend.

2017-04-25  Ed Schonberg  <schonberg@adacore.com>

* einfo.adb, einfo.ads (Body_Needed_For_Inlining): New flag,
to indicate whether during inline processing, when some unit U1
appears in the context of a unit U2 compiled for instantiation
or inlining purposes, the body of U1 needs to be compiled as well.
* sem_prag.adb (Process_Inline): Set Body_Needed_For_Inlining if
context is a package declaration.
* sem_ch12.adb (Analyze_Generic_Subprogram_Declaration,
Analyze_Generic_Package_Declaration): ditto.
* inline.adb (Analyze_Inlined_Bodies): Check
Body_Needed_For_Inlining.

2017-04-25  Ed Schonberg  <schonberg@adacore.com>

* par.adb (Current_Assign_Node): Global variable use to record
the presence of a target_name in the right hand side of the
assignment being parsed.
* par-ch4.adb (P_Name): If the name is a target_name, mark the
enclosing assignment node accordingly.
* par-ch5.adb (P_Assignment_Statement): Set Current_Assign_Node
appropriately.
* sem_ch5.adb (Analyze_Assignment): Disable expansion before
analyzing RHS if the statement has target_names.
* sem_aggr.adb (Resolve_Iterated_Component_Association): Handle
properly choices that are subtype marks.
* exp_ch5.adb: Code cleanup.

2017-04-25  Bob Duff  <duff@adacore.com>

* s-memory.adb: Add a comment regarding efficiency.
* atree.adb: Fix the assertion, and combine 2 assertions into one,
"the source has an extension if and only if the destination does."
* sem_ch3.adb, sem_ch13.adb: Address ??? comments.

2017-04-25  Arnaud Charlet  <charlet@adacore.com trojanek>

* a-tasatt.adb (Set_Value): Fix handling of 32bits -> 64bits
conversion.

2017-04-25  Doug Rupp  <rupp@adacore.com>

* init.c (__gnat_error_handler) [vxworks]: Turn on sigtramp
handling for ppc64-vx7.
* sigtramp-vxworks-target.inc
[SIGTRAMP_BODY]: Add section for ppc64-vx7.

From-SVN: r247146

New file.

From-SVN: r247145

uintp.adb (Most_Sig_2_Digits): In case Direct (Right)...

2017-04-25  Bob Duff  <duff@adacore.com>

* uintp.adb (Most_Sig_2_Digits): In case Direct (Right), fetch
Direct_Val (Right), instead of the incorrect Direct_Val (Left).
(UI_GCD): Remove ??? comment involving possible efficiency
improvements. This just isn't important after all these years.
Also minor cleanup.
* uintp.ads: Minor cleanup.

From-SVN: r247144

exp_util.adb, [...]: Revert previous changes.

2017-04-25  Hristian Kirtchev  <kirtchev@adacore.com>

* exp_util.adb, exp_util.ads, sem_ch7.adb, sem_prag.adb, exp_ch3.adb:
Revert previous changes.
* scng.adb: Minor reformatting.
* s-stratt.ads: Fix unbalanced parens in comment.

From-SVN: r247143

[multiple changes]

2017-04-25  Hristian Kirtchev  <kirtchev@adacore.com>

* sem_ch3.adb, exp_util.adb, sem_prag.adb, freeze.adb, sem_util.adb:
Minor reformatting.

2017-04-25  Ed Schonberg  <schonberg@adacore.com>

* scng.adb (Scan): Handle '@' appropriately.
* sem_ch5.adb: Code cleanup.

From-SVN: r247142

[multiple changes]

2017-04-25  Ed Schonberg  <schonberg@adacore.com>

* freeze.adb (Check_Expression_Function): Do not check for the
use of deferred constants if the freezing of the expression
function is triggered by its generated body, rather than a
premature use.

2017-04-25  Javier Miranda  <miranda@adacore.com>

* exp_attr.adb (Rewrite_Stream_Proc_Call): Handle
subtypes of private types when performing the view conversion.

2017-04-25  Hristian Kirtchev  <kirtchev@adacore.com>

* exp_ch3.adb (Freeze_Type): Signal the DIC body is created for
the purposes of freezing.
* exp_util.adb Update the documentation and structure of the
type map used in class-wide semantics of assertion expressions.
(Add_Inherited_Tagged_DIC): There is really no need to preanalyze
and resolve the triaged expression because all substitutions
refer to the proper entities.  Update the replacement of
references.
(Build_DIC_Procedure_Body): Add formal parameter
For_Freeze. Add local variable Build_Body. Inherited DIC pragmas
are now only processed when freezing occurs.  Build a body only
when one is needed.
(Entity_Hash): Removed.
(Map_Types): New routine.
(Replace_Object_And_Primitive_References): Removed.
(Replace_References): New routine.
(Replace_Type_References): Moved to the library level of Exp_Util.
(Type_Map_Hash): New routine.
(Update_Primitives_Mapping): Update the mapping call.
(Update_Primitives_Mapping_Of_Types): Removed.
* exp_util.ads (Build_DIC_Procedure_Body): Add formal
parameter For_Freeze and update the comment on usage.
(Map_Types): New routine.
(Replace_References): New routine.
(Replace_Type_References): Moved to the library level of Exp_Util.
(Update_Primitives_Mapping_Of_Types): Removed.
* sem_ch7.adb (Preserve_Full_Attributes): Propagate the DIC
properties of the private type to the full view in case the full
view derives from a parent type and inherits a DIC pragma.
* sem_prag.adb (Analyze_Pragma): Guard against a case where a
DIC pragma is placed at the top of a declarative region.

From-SVN: r247141

[multiple changes]

2017-04-25  Tristan Gingold  <gingold@adacore.com>

* s-mmap.ads (Data): Add pragma Inline.

2017-04-25  Hristian Kirtchev  <kirtchev@adacore.com>

* checks.adb (Insert_Valid_Check): Do not use
a renaming to alias a volatile name because this will lead to
multiple evaluations of the volatile name. Use a constant to
capture the value instead.

2017-04-25  Doug Rupp  <rupp@adacore.com>

* init.c [VxWorks Section]: Disable sigtramp for ppc64-vx7.

2017-04-25  Ed Schonberg  <schonberg@adacore.com>

* exp_util.adb, exp_util.ads (Build_Class_Wide_Expression):
Add out parameter to indicate to caller that a wrapper must
be constructed for an inherited primitive whose inherited
pre/postcondition has called to overridden primitives.
* freeze.adb (Check_Inherited_Conditions): Build wrapper body
for inherited primitive that requires it.
* sem_disp.adb (Check_Dispatching_Operation): Such wrappers are
legal primitive operations and belong to the list of bodies
generated after the freeze point of a type.
* sem_prag.adb (Build_Pragma_Check_Equivalent): Use new signature
of Build_Class_Wide_Expression.
* sem_util.adb, sem_util.ads (Build_Overriding_Spec): New procedure
to construct the specification of the wrapper subprogram created
for an inherited operation.

From-SVN: r247140

[multiple changes]

2017-04-25  Bob Duff  <duff@adacore.com>

* s-osinte-linux.ads (pthread_mutexattr_setprotocol,
pthread_mutexattr_setprioceiling): Add new interfaces for these
pthread operations.
* s-taprop-linux.adb (Initialize_Lock, Initialize_TCB): Set
protocols as appropriate for Locking_Policy 'C' and 'I'.
* s-taprop-posix.adb: Minor reformatting to make it more similar
to s-taprop-linux.adb.

2017-04-25  Ed Schonberg  <schonberg@adacore.com>

* sem_ch3.adb (Get_Discriminant_Value, Search_Derivation_Levels):
Handle properly a multi- level derivation involving both renamed
and constrained parent discriminants, when the type to be
constrained has fewer discriminants that the ultimate ancestor.

2017-04-25  Bob Duff  <duff@adacore.com>

* sem_util.adb (Is_Object_Reference): In the
case of N_Explicit_Dereference, return False if it came from a
conditional expression.

2017-04-25  Bob Duff  <duff@adacore.com>

* par-ch4.adb (P_Case_Expression): If a semicolon
is followed by "when", assume that ";" was meant to be ",".

From-SVN: r247139

Minor reformatting.

From-SVN: r247138

sem_ch9.adb, [...]: Minor reformatting and typo fixes.

2017-04-25  Gary Dismukes  <dismukes@adacore.com>

* sem_ch9.adb, sem_ch10.adb, sem_util.adb: Minor reformatting and typo
fixes.

From-SVN: r247137

[multiple changes]

2017-04-25  Arnaud Charlet  <charlet@adacore.com>

* rtsfind.ads (SPARK_Implicit_Load): New procedure for forced
loading of an entity.
* rtsfind.adb (SPARK_Implicit_Load): Body with a pattern
previously repeated in the analysis.
* sem_ch9.adb (Analyze_Protected_Type_Declaration): use new
procedure SPARK_Implicit_Load. (Analyze_Task_Type_Declaration):
use new procedure SPARK_Implicit_Load.
* sem_ch10.adb (Analyze_Compilation_Unit): Use new procedure
SPARK_Implicit_Load.

2017-04-25  Javier Miranda  <miranda@adacore.com>

* sem_util.adb (New_Copy_Tree): By default
copying of defining identifiers is prohibited because
this would introduce an entirely new entity into the
tree. This patch introduces an exception to this general
rule: the declaration of constants and variables located in
Expression_With_Action nodes.
(Copy_Itype_With_Replacement): Renamed as Copy_Entity_With_Replacement.
(In_Map): New subprogram.
(Visit_Entity): New subprogram.
(Visit_Node): Handle EWA_Level,
EWA_Inner_Scope_Level, and take care of defining entities defined
in Expression_With_Action nodes.

2017-04-25  Thomas Quinot  <quinot@adacore.com>

* exp_ch6.adb: minor comment edit.
* sinfo.ads, sinfo.adb: New Null_Statement attribute for null
procedure specifications that come from source.
* par-ch6.adb (P_Subprogram, case of a null procedure): Set new
Null_Statement attribute.
* par_sco.adb (Traverse_Declarations_Or_Statements): For a null
procedure, generate statement SCO for the generated NULL statement.
* sem_ch6.adb (Analyze_Null_Procedure): Use null statement from
parser, if available.

From-SVN: r247136

Add missing changelog entries for last commit.

From-SVN: r247135

S/390: PR80464: Split MEM->GPR vector moves

We do this already for TImode values but it was missing for vector
modes.

gcc/ChangeLog:

2017-04-25  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>

PR target/80464
* config/s390/vector.md: Split MEM->GPR vector moves for
non-s_operand addresses.

gcc/testsuite/ChangeLog:

2017-04-25  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>

PR target/80464
* gfortran.fortran-torture/compile/pr80464.f90: New test.

From-SVN: r247134

S/390: PR79895: Fix TImode constant handling

The P constraint letter is supposed to match every constant which is
acceptable during reload.  However, constraints do not appear to be
able to handle const_wide_int yet.  It works with predicates so the
alternative is modelled with a new predicate now.

gcc/ChangeLog:

2017-04-25  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>

PR target/79895
* config/s390/predicates.md (reload_const_wide_int_operand): New
predicate.
* config/s390/s390.md ("movti"): Remove d/P alternative.
("movti_bigconst"): New pattern definition.

gcc/testsuite/ChangeLog:

2017-04-25  Andreas Krebbel  <krebbel@linux.vnet.ibm.com>

PR target/79895
* gcc.target/s390/pr79895.c: New test.

From-SVN: r247133

S/390: PR80080: Optimize atomic patterns.

The attached patch optimizes the atomic_exchange and atomic_compare
patterns on s390 and s390x (mostly limited to SImode and DImode).
Among general optimizaation, the changes fix most of the problems
reported in PR 80080:
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=80080

gcc/ChangeLog:

2017-04-25  Dominik Vogt  <vogt@linux.vnet.ibm.com>

PR target/80080
* s390-protos.h (s390_expand_cs_hqi): Removed.
(s390_expand_cs, s390_expand_atomic_exchange_tdsi): New prototypes.
* config/s390/s390.c (s390_emit_compare_and_swap): Handle all integer
modes as well as CCZ1mode and CCZmode.
(s390_expand_atomic_exchange_tdsi, s390_expand_atomic): Adapt to new
signature of s390_emit_compare_and_swap.
(s390_expand_cs_hqi): Likewise, make static.
(s390_expand_cs_tdsi): Generate an explicit compare before trying
compare-and-swap, in some cases.
(s390_expand_cs): Wrapper function.
(s390_expand_atomic_exchange_tdsi): New backend specific expander for
atomic_exchange.
(s390_match_ccmode_set): Allow CCZmode <-> CCZ1 mode.
* config/s390/s390.md ("atomic_compare_and_swap<mode>"): Merge the
patterns for small and large integers.  Forbid symref memory operands.
Move expander to s390.c.  Require cc register.
("atomic_compare_and_swap<DGPR:mode><CCZZ1:mode>_internal")
("*atomic_compare_and_swap<TDI:mode><CCZZ1:mode>_1")
("*atomic_compare_and_swapdi<CCZZ1:mode>_2")
("*atomic_compare_and_swapsi<CCZZ1:mode>_3"): Use s_operand to forbid
symref memory operands.  Remove CC mode and call s390_match_ccmode
instead.
("atomic_exchange<mode>"): Allow and implement all integer modes.

gcc/testsuite/ChangeLog:

2017-04-25  Dominik Vogt  <vogt@linux.vnet.ibm.com>

PR target/80080
* gcc.target/s390/md/atomic_compare_exchange-1.c: New test.
* gcc.target/s390/md/atomic_compare_exchange-1.inc: New test.
* gcc.target/s390/md/atomic_exchange-1.inc: New test.

From-SVN: r247132

S/390: Load and test peephole.

gcc/ChangeLog:

2017-04-25  Dominik Vogt  <vogt@linux.vnet.ibm.com>

* config/s390/s390.md (define_peephole2): New peephole to help
combining the load-and-test pattern with volatile memory.

From-SVN: r247131

S/390: Use load-on-condition in cstorecc4

gcc/ChangeLog:

2017-04-25  Dominik Vogt  <vogt@linux.vnet.ibm.com>

* config/s390/s390.md ("cstorecc4"): Use load-on-condition and deal
with CCZmode for TARGET_Z196.

From-SVN: r247130

re PR rtl-optimization/80501 (Wrong code w/ a signed char, a shift, and a conversion to int)

PR rtl-optimization/80501
* combine.c (make_compound_operation_int): Set subreg_code to SET
even for AND with mask of the sign bit of mode.

* gcc.c-torture/execute/pr80501.c: New test.

From-SVN: r247128

re PR rtl-optimization/80500 (ICE: internal consistency failure (error: invalid rtl sharing found in the insn))

PR rtl-optimization/80500
* loop-unroll.c (combine_var_copies_in_loop_exit): Call copy_rtx on
sum's initial value.

* gcc.dg/pr80500.c: New test.

From-SVN: r247126

thunderx2t99.md (thunderx2t99_crc): New Reservation.

* config/aarch64/thunderx2t99.md (thunderx2t99_crc): New Reservation.

Co-Authored-By: Naveen H.S <Naveen.Hurugalawadi@cavium.com>
From-SVN: r247125

thunderx2t99.md (thunderx2t99_aes): New Reservation.

* config/aarch64/thunderx2t99.md (thunderx2t99_aes): New Reservation.
(thunderx2t99_sha): New Reservation.

Co-Authored-By: Naveen H.S <Naveen.Hurugalawadi@cavium.com>
From-SVN: r247124

Let tree_single_nonzero_warnv_p use range information

2017-04-25  Marc Glisse  <marc.glisse@inria.fr>

gcc/
* fold-const.c (tree_single_nonzero_warnv_p): Handle SSA_NAME.

gcc/testsuite/
* gcc.dg/tree-ssa/cmpmul-1.c: New file.
* gcc.dg/Wstrict-overflow-18.c: Xfail.

From-SVN: r247123

aarch64-simd.md (aarch64_simd_vec_set<mode>): Fix type for 1-element load.

* config/aarch64/aarch64-simd.md (aarch64_simd_vec_set<mode>): Fix
type for 1-element load.

Co-Authored-By: Naveen H.S <Naveen.Hurugalawadi@cavium.com>
From-SVN: r247122

Daily bump.

From-SVN: r247121

re PR fortran/80121 (Memory leak with derived-type intent(out) argument)

2017-04-22  Janus Weil  <janus@gcc.gnu.org>

PR fortran/80121
* gfortran.dg/intent_out_9.f90: Remove deprecated cleanup-tree-dump
directive.

From-SVN: r247115

Regenerate .pot files.

gcc/po:
* gcc.pot: Regenerate.

libcpp/po:
* cpplib.pot: Regenerate.

From-SVN: r247113

types.rst (gccjit::type::get_const): Remove comment.

gcc_jit_type_get_aligned

gcc/jit/ChangeLog:
* docs/cp/topics/types.rst (gccjit::type::get_const): Remove
comment.
(gccjit::type::get_aligned): Add.
* docs/topics/compatibility.rst: Add LIBGCCJIT_ABI_7.
* docs/topics/types.rst: Add gcc_jit_type_get_aligned.
* docs/_build/texinfo/libgccjit.texi: Regenerate.
* jit-playback.c (gcc::jit::playback::type::get_aligned): New
method.
* jit-playback.h (gcc::jit::playback::type::get_aligned): New
method.
* jit-recording.c: Within namespace gcc::jit::recording...
(type::get_aligned): New method.
(memento_of_get_aligned::replay_into): New method.
(memento_of_get_aligned::make_debug_string): New method.
(memento_of_get_aligned::write_reproducer): New method.
* jit-recording.h: Within namespace gcc::jit::recording...
(type::get_aligned): New method.
(type::accepts_writes_from): Strip off qualifications from
this when comparing pointer equality.
(decorated_type): New subclass of type, subsuming the
commonality between memento_of_get_const and
memento_of_get_volatile.
(memento_of_get_const): Make a subclass of decorated_type,
rather than type.
(memento_of_get_volatile): Likewise.
(memento_of_get_aligned): Likewise.
* libgccjit++.h: Within namespace gccjit...
(type::get_const): New method.
(type::get_aligned): New method.
* libgccjit.c (gcc_jit_type_get_aligned): New function.
* libgccjit.h (gcc_jit_type_get_aligned): New decl.
* libgccjit.map (LIBGCCJIT_ABI_7): New
(gcc_jit_type_get_aligned): Add.

gcc/testsuite/ChangeLog:
* jit.dg/all-non-failing-tests.h: Add test-alignment.c.
* jit.dg/test-alignment.c: New test case.
* jit.dg/test-alignment.cc: New test case.
* jit.dg/test-error-gcc_jit_type_get_aligned-non-power-of-two.c:
New test case.

From-SVN: r247111

decl.c (grokdeclarator): Use %qT instead of %<%T%> in

        * decl.c (grokdeclarator): Use %qT instead of %<%T%> in
        * diagnostics.
        (start_enum): Likewise.
        (build_enumerator): Likewise. Use %qE instead of plain %E.
        * parser.c (cp_parser_mem_initializer_list): Use %qD instead of
        %<%D%> in diagnostics.
        (cp_parser_elaborated_type_specifier): Likewise.
        * pt.c (make_pack_expansion): Use %qT and %qE instead of
        %<%T%> and %<%E%> in diagnostics.
        (tsubst_pack_expansion): Likewise.

From-SVN: r247110

Fix location of sizeof/alignof (PR c++/80016)

PR c++/80016 reports an issue with bizarre underlined range
for a complicated expression.

The root cause for the incorrect *starting* location of that range
is that alignof and sizeof expressions currently have
start == finish == caret at the opening parenthesis of the
expression.

This patch fixes this by generating appropriate start and finish
locations for alignof and sizeof expressions.

gcc/cp/ChangeLog:
PR c++/80016
* parser.c (cp_parser_unary_expression):  Generate a location
range for alignof and sizeof expressions.

gcc/testsuite/ChangeLog:
PR c++/80016
* g++.dg/plugin/diagnostic-test-expressions-1.C (test_sizeof): New
test function.
(test_alignof): New test function.

From-SVN: r247108

X /[ex] 4 < Y /[ex] 4

2017-04-24  Marc Glisse  <marc.glisse@inria.fr>

gcc/
* match.pd (X/[ex]C CMP Y/[ex]C): New transformation.

gcc/testsuite/
* gcc.dg/tree-ssa/cmpexactdiv-2.c: New file.

From-SVN: r247107

parser.c (cp_parser_cv_qualifier_seq_opt): Add fix-it info to error message.

        * parser.c (cp_parser_cv_qualifier_seq_opt): Add fix-it info to
        error message.
        (cp_parser_virt_specifier_seq_opt): Likewise.
        (set_and_check_decl_spec_loc): Likewise twice.

        * g++.dg/diagnostic/duplicate1.C: New test.
        * g++.dg/cpp0x/duplicate1.C: New test.

From-SVN: r247105

[PR 80293] Dont totally-scalarize char arrays

2017-04-24  Martin Jambor  <mjambor@suse.cz>

PR tree-optimization/80293
* tree-sra.c (scalarizable_type_p): New parameter const_decl, make
char arrays not totally scalarizable if it is false.
(analyze_all_variable_accesses): Pass correct value in the new
parameter.  Add a statistics counter.

testsuite/
* g++.dg/tree-ssa/pr80293.C: New test.

From-SVN: r247104

PR libstdc++/80506 fix constant used in condition

PR libstdc++/80506
* include/bits/random.tcc (gamma_distribution::operator()): Fix magic
number used in loop condition.

From-SVN: r247099

Fix (PR middle-end/79931)

2017-04-24  Jan Hubicka  <hubicka@ucw.cz>

PR middle-end/79931
* ipa-devirt.c (dump_possible_polymorphic_call_targets): Fix ICE.
2017-04-24  Martin Liska  <mliska@suse.cz>

PR middle-end/79931
* g++.dg/ipa/pr79931.C: New test.

From-SVN: r247097

re PR fortran/80494 (ICE in wide_int_to_tree)

2017-04-24  Richard Biener  <rguenther@suse.de>

PR tree-optimization/80494
* tree-scalar-evolution.c (analyze_scalar_evolution_1): Bail
out for complex types.

* gfortran.dg/pr80494.f90: New testcase.

From-SVN: r247095

PR libstdc++/80504 qualify calls to avoid ADL

PR libstdc++/80504
* include/bits/refwrap.h (ref, cref): Qualify calls.
* testsuite/20_util/reference_wrapper/80504.cc: New test.

From-SVN: r247094

PR libstdc++/80493 fix invalid exception specification

PR libstdc++/80493
* include/experimental/optional (optional::swap): Fix exception
specification.

From-SVN: r247093

tree-ssa-sccvn.h (run_scc_vn): Adjust prototype.

2017-04-24  Richard Biener  <rguenther@suse.de>

* tree-ssa-sccvn.h (run_scc_vn): Adjust prototype.
* tree-ssa-sccvn.c (print_scc): Print SCC size.
(extract_and_process_scc_for_name): Never fail but drop SCC to varying.
(DFS): Adjust and never fail.
(sccvn_dom_walker::fail): Remove.
(sccvn_dom_walker::before_dom_children): Adjust.
(run_scc_vn): Likewise and never fail.
* tree-ssa-pre.c (pass_pre::execute): Adjust.
(pass_fre::execute): Likewise.

From-SVN: r247092

re PR tree-optimization/79725 (Sinking opportunity missed if blocked by dead stmt)

2017-04-24  Richard Biener  <rguenther@suse.de>

PR tree-optimization/79725
* tree-ssa-sink.c (statement_sink_location): Return whether
failure reason was zero uses.  Move that check later.
(sink_code_in_bb): Deal with zero uses by removing the stmt
if possible.

* gcc.dg/tree-ssa/ssa-sink-15.c: New testcase.

From-SVN: r247091

re PR c++/2972 (-Wuninitialized could warn about uninitialized member variable usage in constructors)

2017-04-24  Richard Biener  <rguenther@suse.de>

PR c++/2972
* tree-ssa-uninit.c (warn_uninitialized_vars): Handle some
pointer-based references.

* g++.dg/warn/Wuninitialized-10.C: New testcase.

From-SVN: r247090

re PR bootstrap/79814 (pass-instances.def:36:14: error: ‘*<unknown>.gcc::pass_manager::pass_warn_unused_result_1’ is used uninitialized in this function)

2017-04-24  Richard Biener  <rguenther@suse.de>

PR bootstrap/79814
* pass_manager.h (pass_manager::operator new): Remove.
(pass_manager::operator delete): Likewise.
* passes.c (pass_manager::operator new): Remove.
(pass_manager::operator delete): Likewise.
(pass_manager::pass_manager): Zero individual pass members.

From-SVN: r247089

Daily bump.

From-SVN: r247088

re PR fortran/80484 (Three syntax errors involving derived-type I/O)

2017-04-23  Jerry DeLisle  <jvdelisle@gcc.gnu.org>

PR fortran/80484
* io.c (format_lex): Check for '/' and set token to FMT_SLASH.
(check_format): Move FMT_DT checking code to data_desc section.
* module.c (gfc_match_use): Include the case of INTERFACE_DTIO.

PR fortran/80484
* gfortran.dg/dtio_29.f03: New test.

From-SVN: r247084

re PR fortran/80121 (Memory leak with derived-type intent(out) argument)

2017-04-22  Janus Weil  <janus@gcc.gnu.org>

PR fortran/80121
* trans-types.c (gfc_conv_procedure_call): Deallocate the components
of allocatable intent(out) arguments.

2017-04-22  Janus Weil  <janus@gcc.gnu.org>

PR fortran/80121
* gfortran.dg/intent_out_9.f90: New test case.

From-SVN: r247083

re PR target/70799 (STV pass does not convert DImode shifts)

PR target/70799
* config/i386/i386.c (dimode_scalar_to_vector_candidate_p)
<case ASHIFT, case LSHIFTRT>: Also consider variable shifts.
Check "XEXP (src, 1)" operand here.
<case PLUS, case MINUS, case IOR, case XOR, case AND>:
Check "XEXP (src, 1)" operand here.
(dimode_scalar_chain::make_vector_copies): Detect count register
of a shift instruction.  Zero extend count register from QImode
to DImode to satisfy vector shift pattern count operand predicate.
Substitute vector shift count operand with a DImode copy.
(dimode_scalar_chain::convert_reg): Ditto, zero-extend from
vector register.

testsuite/ChangeLog:

PR target/70799
* gcc.target/i186/pr70799-4.c: New test.

From-SVN: r247082

Daily bump.

From-SVN: r247081

Daily bump.

From-SVN: r247075

re PR fortran/80392 ([OOP] ICE with allocatable polymorphic function result in a procedure pointer component)

2017-04-21  Janus Weil  <janus@gcc.gnu.org>

PR fortran/80392
* trans-types.c (gfc_get_derived_type): Prevent an infinite loop when
building a derived type that includes a procedure pointer component
with a polymorphic result.

2017-04-21  Janus Weil  <janus@gcc.gnu.org>

PR fortran/80392
* gfortran.dg/proc_ptr_comp_49.f90: New test case.

From-SVN: r247069

PR c++/80179 - ICE with initialized flexible array member.

* constexpr.c (verify_ctor_sanity): Handle flexible array members.

From-SVN: r247067

i386.md (*extzvqi_mem_rex64): Move above *extzv<mode>.

* config/i386/i386.md (*extzvqi_mem_rex64): Move above *extzv<mode>.
Remove UNSPEC_NOREX_MEM tag.  Update corresponding peephole2 pattern.
(*insvqi_1_mem_rex64): Move above insv<mode>_1.  Remove
UNSPEC_NOREX_MEM tag.  Update corresponding peephole2 pattern.
(UNSPEC_NOREX_MEM): Remove definition.

From-SVN: r247066

pr79804.c: Add additional dg-error directive.

* gcc.target/i386/pr79804.c: Add additional dg-error directive.

From-SVN: r247065

PR libstdc++/80316 make promise::set_value throw no_state error

PR libstdc++/80316
* include/std/future (_State_baseV2::_Setter::operator()): Remove
_S_check calls that are done after the pointer to the shared state is
already dereferenced.
(_State_baseV2::_Setter<_Res, void>): Define specialization for void
as partial specialization so it can be defined within the definition
of _State_baseV2.
(_State_baseV2::__setter): Call _S_check.
(_State_baseV2::__setter(promise<void>*)): Add overload for use by
promise<void>::set_value and promise<void>::set_value_at_thread_exit.
(promise<T>, promise<T&>, promise<void>): Make _State a friend.
(_State_baseV2::_Setter<void, void>): Remove explicit specialization.
(promise<void>::set_value, promise<void>::set_value_at_thread_exit):
Use new __setter overload.
* testsuite/30_threads/promise/members/at_thread_exit2.cc: New test.
* testsuite/30_threads/promise/members/set_exception.cc: Test
promise<T&> and promise<void> specializations.
* testsuite/30_threads/promise/members/set_exception2.cc: Likewise.
Test for no_state error condition.
* testsuite/30_threads/promise/members/set_value2.cc: Likewise.

From-SVN: r247064

Avoid deprecated warnings about std::auto_ptr in library headers

* include/backward/auto_ptr.h: Ignore deprecated warnings from use
of auto_ptr.
* include/bits/shared_ptr.h: Likewise.
* include/bits/shared_ptr_base.h: Likewise.
* include/bits/unique_ptr.h: Likewise.
* libstdc++-v3/testsuite/20_util/default_delete/48631_neg.cc: Adjust
dg-error lineno.
* libstdc++-v3/testsuite/20_util/default_delete/void_neg.cc: Likewise.
* libstdc++-v3/testsuite/20_util/unique_ptr/assign/48635_neg.cc:
Likewise.
* libstdc++-v3/testsuite/20_util/unique_ptr/cons/cv_qual_neg.cc:
Likewise.

From-SVN: r247063

re PR tree-optimization/79547 (duplicate strlen calls with same argument not folded)

2017-04-21  Richard Biener  <rguenther@suse.de>

PR tree-optimization/79547
* tree-ssa-structalias.c (find_func_aliases_for_builtin_call):
Handle strlen, strcmp, strncmp, strcasecmp, strncasecmp, memcmp,
bcmp, strspn, strcspn, __builtin_object_size and __builtin_constant_p
without any constraints.

* gcc.dg/tree-ssa/strlen-2.c: New testcase.

From-SVN: r247062

re PR tree-optimization/78847 (pointer arithmetic from c++ ranged-based for loop not optimized)

2017-04-21  Richard Biener  <rguenther@suse.de>

PR tree-optimization/78847
* fold-const.c (split_tree): Handle POINTER_PLUS_EXPR.

* g++.dg/tree-ssa/pr78847.C: New testcase.

From-SVN: r247061

* es.po, fr.po: Update.

From-SVN: r247059

libiberty: Limit demangler maximum d_print_comp recursion call depth.

The fix for PR demangler/70909 and 67264 (endless demangler recursion)
catches when a demangle_component is printed in a cycle. But that doesn't
protect the call stack blowing up from non-cyclic nested types printed
recursively through d_print_comp. This can happen by a (very) long mangled
string that simply creates a very deep pointer or qualifier chain. Limit
the recursive d_print_comp call depth for a d_print_info to 1K nested
types.

libiberty/ChangeLog:

        * cp-demangle.c (MAX_RECURSION_COUNT): New constant.
        (struct d_print_info): Add recursion field.
        (d_print_init): Initialize recursion.
        (d_print_comp): Check and update d_print_info recursion depth.

From-SVN: r247056

libiberty: Always return NULL if d_add_substitution fails.

d_add_substitution can fail for various reasons, like when the subs array
is full. If d_add_substitution fails d_substitution should return NULL
early and not try to continue. Every other call of d_add_substitution
is handled in the same way.

libiberty/ChangeLog:

* cp-demangle.c (d_substitution): Return NULL if d_add_substitution
fails.

From-SVN: r247055

libiberty: Don't update and remove did_subs field from demangler structs.

The d_info field did_subs was used for estimating the string output
size. It was no longer used when the malloc-less callback interface
was introduced in 2007 (svn r121305). But the field was still updated.
When backtracking was introduced in 2013 (svn r205292) did_subs was
also added to the d_info_checkpoint struct. But except for updating
the field it was still not used.

Since it is never used just stop updating the field and remove it
from the two structs.

libiberty/ChangeLog:

* cp-demangle.h (struct d_info): Remove did_subs field.
* cp-demangle.c (struct d_info_checkpoint): Likewise.
(d_template_param): Don't update did_subs.
(d_substitution): Likewise.
(d_checkpoint): Don't assign did_subs.
(d_backtrack): Likewise.
(cplus_demangle_init_info): Don't initialize did_subs.

From-SVN: r247054

tree.h (build_qualified_type): Annotate with CXX_MEM_STAT_INFO.

2017-04-21  Richard Biener  <rguenther@suse.de>

* tree.h (build_qualified_type): Annotate with CXX_MEM_STAT_INFO.
(build_distinct_type_copy): Likewise.
(build_variant_type_copy): Likewise.
* tree.c (build_qualified_type): Pass down mem-stat info.
(build_distinct_type_copy): Likewise.
(build_variant_type_copy): Likewise.

cp/
* cp-tree.h (copy_decl): Annotate with CXX_MEM_STAT_INFO.
(copy_type): Likewise.
* lex.c (copy_decl): Pass down mem-stat info.
(copy_type): Likewise.

From-SVN: r247053

re PR c/80468 (ICE on invalid AVX512 code with -m32)

PR c/80468
* c-decl.c (finish_declspecs) <case cts_int_n>: If int_n_idx is not
enabled, set specs->type to integer_type_node.

* gcc.dg/pr80468.c: New test.

From-SVN: r247052

Add test-case (PR tree-optimization/66278).

2017-04-21  Martin Liska  <mliska@suse.cz>

PR tree-optimization/66278
* gcc.dg/vect/pr66278.c: New test.

From-SVN: r247051

re PR tree-optimization/80237 (float to double conversion is not optimized away)

2017-04-21  Richard Biener  <rguenther@suse.de>

PR tree-optimization/80237
* tree-ssa-pre.c (find_leader_in_sets): Add third set argument,
defaulted to NULL.
(phi_translate_1): Also allow a leader in AVAIL_OUT of pred
for a simplified result.

* gcc.dg/tree-ssa/tailcall-9.c: New testcase.
* gcc.dg/tree-ssa/ldist-pr45948.c: Remove undefined behavior,
adjust expected optimizations.

From-SVN: r247049

tree-ssa-loop-ivcanon.c (constant_after_peeling): Do not require sth as strict as a simple_iv but a chrec without symbols and an...

2016-04-21  Richard Biener  <rguenther@suse.de>

* tree-ssa-loop-ivcanon.c (constant_after_peeling): Do not require
sth as strict as a simple_iv but a chrec without symbols and an
operand defined in the loop we are peeling (and not some subloop).
(propagate_constants_for_unrolling): Propagate all constants.

* gcc.dg/vect/no-scevccp-outer-13.c: Adjust to prevent unrolling
of inner loops.
* gcc.dg/vect/no-scevccp-outer-7.c: Likewise.
* gcc.dg/vect/vect-104.c: Likewise.

From-SVN: r247048

re PR tree-optimization/80426 (wrong manipulation of range based on INT_MIN)

PR tree-optimization/80426
* gcc.c-torture/execute/20170429-1.c: Rename into...
* gcc.c-torture/execute/20170419-1.c: ...this.

From-SVN: r247046

Daily bump.

From-SVN: r247045

re PR target/79804 (ICE in print_reg, at config/i386/i386.c:17637)

PR target/79804
* config/i386/i386.c (print_reg): Remove assert for disalowed
regno values, call output_operand_lossage instead.

testsuite/ChangeLog:

PR target/79804
* gcc.target/i386/pr79804.c: New test.

From-SVN: r247037

re PR target/78090 ([x86_64]: GCC allows integer register for inter unit conversion under -mtune-ctrl=^inter_unit_conversions .)

PR target/78090
* config/i386/constraints.md (Yc): New register constraint.
* config/i386/i386.md (*float<SWI48:mode><MODEF:mode>2_mixed):
Use Yc constraint for alternative 2 of operand 0.  Remove
preferred_for_speed attribute.

testsuite/ChangeLog:

PR target/78090
* gcc.target/i386/conversion-2.c: Remove obsolete test.

From-SVN: r247036

* g++/opt/pr80436.C: XFAIL on AIX.

From-SVN: r247034

PR c++/80473 allow suppressing notes about over-aligned new

gcc/cp:

PR c++/80473
* init.c (build_new_1): Suppress notes about over-aligned new when
the warning is suppressed.

gcc/testsuite:

PR c++/80473
* g++.dg/diagnostic/pr80473.C: New test.

From-SVN: r247033

omp-low: fix lastprivate/linear lowering for SIMT

gcc/
* omp-low.c (lower_lastprivate_clauses): Correct handling of linear and
lastprivate clauses in SIMT case.

libgomp/
* testsuite/libgomp.c/target-36.c: New testcase.

From-SVN: r247029

c.opt (Wextra-semi): New C++ warning flag.

        * c.opt (Wextra-semi): New C++ warning flag.
        * doc/invoke.texi (-Wextra-semi): Document new warning option.
        * parser.c (cp_parser_member_declaration): Add warning with fixit
        information for extra semicolon after in-class function definition.
        * g++.dg/warn/Wextra-semi.C: New test.

From-SVN: r247028

PR libstdc++/68397 std::tr1::expint fails ... long double arguments.

2017-04-20  Edward Smith-Rowland  <3dw4rd@verizon.net>

PR libstdc++/68397 std::tr1::expint fails ... long double arguments.
* include/tr1/exp_integral.tcc: Increase iteration limits.
* testsuite/tr1/5_numerical_facilities/special_functions/15_expint/
pr68397.cc: New test.
* testsuite/special_functions/14_expint/pr68397.cc: New test.

From-SVN: r247027

re PR target/57796 (AVX2 gather vectorization: code bloat and reduction of performance)

2017-04-20  Richard Biener  <rguenther@suse.de>

PR tree-optimization/57796
* tree-vect-stmts.c (vect_model_store_cost): Cost scatters
as N scalar stores.
(vect_model_load_cost): Cost gathers as N scalar loads.

From-SVN: r247026

ggc-page.c (ggc_allocated_p): Rename to ...

2017-04-20  Richard Biener  <rguenther@suse.de>

* ggc-page.c (ggc_allocated_p): Rename to ...
(safe_lookup_page_table_entry): ... this and return the lookup
result.
(gt_ggc_m_S): Use safe_lookup_page_table_entry.

From-SVN: r247025

re PR debug/80453 (another compare-debug failure)

2017-04-20  Richard Biener  <rguenther@suse.de>

PR tree-optimization/80453
* tree-ssa-sccvn.h (struct vn_phi_s): Add cclhs and ccrhs members.
* tree-ssa-sccvn.c (cond_stmts_equal_p): Use recorded lhs and rhs
from the conditions.
(vn_phi_eq): Pass them down.
(vn_phi_lookup): Record them.
(vn_phi_insert): Likewise.

From-SVN: r247024

MIPS: Prevent buffer overrun in uninitialised variable fix

gcc/
* config/mips/mips.c (mips_expand_vec_perm_const): Re-fix
uninitialized variable warning to avoid buffer overrun.

From-SVN: r247022

PR79862 check macro for type aliases that depend on <stdint.h>

PR libstdc++/79862
* include/std/atomic [!_GLIBCXX_USE_C99_STDINT_TR1] (atomic_int8_t)
(atomic_uint8_t, atomic_int16_t, atomic_uint16_t, atomic_int32_t,
(atomic_uint32_t, atomic_int64_t, atomic_uint64_t)
(atomic_int_least8_t, atomic_uint_least8_t, atomic_int_least16_t)
(atomic_uint_least16_t, atomic_int_least32_t, atomic_uint_least32_t)
(atomic_int_least64_t, atomic_uint_least64_t, atomic_int_fast8_t)
(atomic_uint_fast8_t, atomic_int_fast16_t, atomic_uint_fast16_t)
(atomic_int_fast32_t, atomic_uint_fast32_t, atomic_int_fast64_t)
(atomic_uint_fast64_t, atomic_intmax_t, atomic_uintmax_t): Don't
define.

From-SVN: r247021

* crontab: Enable snapshots from gcc-7-branch.

From-SVN: r247020

doc: mention handling of {0} in -Wmissing-field-initializers (PR 71250)

PR other/71250
* doc/invoke.texi (-Wmissing-field-initializers): Mention that warning
is suppressed for '{ 0 }' in C.

From-SVN: r247018

* BASE-VER: Set to 8.0.0.

From-SVN: r247017

[ARM] Fix type for .init_array.* and .fini_array.* sections

2017-04-20  Thomas Preud'homme  <thomas.preudhomme@arm.com>

    gcc/
    * config/arm/arm.c (arm_elf_asm_cdtor): Create non-default
    priority .init_array and .fini_array section with SECTION_NOTYPE
    flag.

From-SVN: r247015

re PR middle-end/80423 (GC related -fcompare-debug failure)

PR middle-end/80423
* tree.h (build_array_type): Add typeless_storage default argument.
* tree.c (type_cache_hasher::equal): Also compare
TYPE_TYPELESS_STORAGE flag for ARRAY_TYPEs.
(build_array_type): Add typeless_storage argument, set
TYPE_TYPELESS_STORAGE to it, if shared also hash it, and pass to
recursive call.
(build_nonshared_array_type): Adjust build_array_type_1 caller.
(build_array_type): Likewise.  Add typeless_storage argument.
c-family/
* c-common.c (complete_array_type): Preserve TYPE_TYPELESS_STORAGE.
cp/
* tree.c (build_cplus_array_type): Call build_array_type
with the intended TYPE_TYPELESS_STORAGE flag value, instead
of calling build_array_type and modifying later TYPE_TYPELESS_STORAGE
on the shared type.
testsuite/
* g++.dg/other/pr80423.C: New test.

From-SVN: r247014

Skip pr80170.c for non-ptr32plus targets

The testcase uses 32 bit or bigger pointer offsets, and
that triggers an ptrofftype_p assert in tree.c.

gcc/testsuite

2017-04-19  Senthil Kumar Selvaraj  <senthil_kumar.selvaraj@atmel.com>

* gcc.dg/pr80170.c: Require ptr32plus.

From-SVN: r247011

Daily bump.

From-SVN: r247010

re PR tree-optimization/80426 (wrong manipulation of range based on INT_MIN)

PR tree-optimization/80426
* tree-vrp.c (extract_range_from_binary_expr_1): For an additive
operation on symbolic operands, also compute the overflow for the
invariant part when the operation degenerates into a negation.

PR tree-optimization/80426
* gcc.c-torture/execute/20170419-1.c: New test.

Co-Authored-By: Jakub Jelinek <jakub@redhat.com>
From-SVN: r247007

re PR bootstrap/77661 (--enable-maintainer-mode causes in-tree-build of MPC to fail)

2017-04-19  Thomas Koenig  <tkoenig@gcc.gnu.org>
Tobias Burnus  <tobias.burnus@physik.fu-berlin.de>

PR bootstrap/77661
* Makefile.def: Don't pass --enable-maintainer-mode on to an
in-tree build MPC.
* Makefile.in: Regenerate.

Co-Authored-By: Tobias Burnus <tobias.burnus@physik.fu-berlin.de>
From-SVN: r247006

Rationalise ABI links in libstdc++ manual

* doc/xml/manual/abi.xml: Rephrase one of the references to the
Itanium C++ ABI.
* doc/xml/manual/test.xml: Document DejaGnu 1.5.3 requirement.
* doc/html/*: Regenerate.

From-SVN: r247003

re PR debug/80461 (ICE in modified_type_die, at dwarf2out.c:12566)

PR debug/80461
* dwarf2out.c (modified_type_die, gen_type_die_with_usage):
Check for t with zero TYPE_QUALS_NO_ADDR_SPACE.

* g++.dg/debug/pr80461.C: New test.

From-SVN: r247002

re PR c++/80459 (c-c++-common/opaque-vector.c FAILs)

PR c++/80459
* c-c++-common/opaque-vector.c (SIZEOF_MAXINT): Define.
(f): Don't test long double vectors if __SIZEOF_LONG_DOUBLE__
is not power of 2, or is more than 16 or more than SIZEOF_MAXINT.

From-SVN: r247001