[Ada] Additional warnings on overlapping actuals of composite types

2020-06-11  Ed Schonberg  <schonberg@adacore.com>

gcc/ada/

* sem_warn.adb (Warn_On_Overlapping_Actuals): Simplify code,
remove inner predicate Is_Covered_Formal, preserve warning for
two overlapping composite types when only one is writable, and
for two overlapping and writable elementary types.

[Ada] Allow specifying volatility refinement aspects for types

2020-06-11  Steve Baird  <baird@adacore.com>

gcc/ada/

* contracts.adb (Add_Contract_Item): Support specifying
volatility refinement aspects for types.
(Analyze_Contracts): Add call to Analyze_Type_Contract in the
case of a contract for a type.
(Freeze_Contracts): Add call to Analyze_Type_Contract in the
case of a contract for a type.
(Check_Type_Or_Object_External_Properties): A new procedure
which performs the work that needs to be done for both object
declarations and types.
(Analyze_Object_Contract): Add a call to
Check_Type_Or_Object_External_Properties and remove the code in
this procedure which did much of the work that is now performed
by that call.
(Analyze_Type_Contract): Implement this new routine as nothing
more than a call to Check_Type_Or_Object_External_Properties.
* contracts.ads: Update comment for Add_Contract_To_Item because
types can have contracts.  Follow (questionable) precedent and
declare new routine Analyze_Type_Contract as visible (following
example of Analyze_Object_Contract), despite the fact that it is
never called from outside of the package where it is declared.
* einfo.adb (Contract, Set_Contract): Id argument can be a type;
support this case.
(Write_Field34_Name): Field name is "contract" for a type.
* einfo.ads: Update comment describing Contract attribute.
* sem_ch3.adb (Build_Derived_Numeric_Type): Is_Volatile should
return same answer for all subtypes of a given type. Thus, when
building the base type for something like type Volatile_1_To_10
is range 1 .. 10 with Volatile; that basetype should be marked
as being volatile.
(Access_Type_Declaration): Add SPARK-specific legality check
that the designated type of an access type shall be compatible
with respect to volatility with the access type.
* sem_ch12.adb (Check_Shared_Variable_Control_Aspects): Add
SPARK-specific legality check that an actual type parameter in
an instantiation shall be compatible with respect to volatility
with the corresponding formal type.
* sem_ch13.adb (Analyze_Aspect_Specifications): Perform checks
for aspect specs for the 4 volatility refinement aspects that
were already being performed for all language-defined aspects.
* sem_prag.adb (Analyze_External_Property_In_Decl_Part,
Analyze_Pragma): External properties (other than No_Caching) may
be specified for a type, including a generic formal type.
* sem_util.ads: Declare new subprograms - Async_Readers_Enabled,
Async_Writers_Enabled, Effective_Reads, Effective_Writes, and
Check_Volatility_Compatibility.
* sem_util.adb (Async_Readers_Enabled, Async_Writers_Enabled,
Effective_Reads, Effective_Writes): Initial implementation of
new functions for querying aspect values.
(Check_Volatility_Compatibility): New procedure intended for use
in checking all SPARK legality rules of the form "<> shall be
compatible with respect to volatility with <>".
(Has_Enabled_Property): Update comment because Item_Id can be a
type.  Change name of nested Variable_Has_Enabled_Property
function to Type_Or_Variable_Has_Enabled_Property; add a
parameter to that function because recursion may be needed,
e.g., in the case of a derived typ).  Cope with the case where
the argument to Has_Enabled_Property is a type.

[Ada] Avoid "others => <>" association in resolved record aggregates

2020-06-11  Piotr Trojanek  <trojanek@adacore.com>

gcc/ada/

* sem_aggr.adb (Add_Association): Add assertion about the formal
parameters.
(Propagate_Discriminants): Always add an explicit component
association, so that an "others => <>" association is never
needed.

[Ada] Put_Image attribute

2020-06-11  Bob Duff  <duff@adacore.com>

gcc/ada/

* exp_put_image.adb (Build_Elementary_Put_Image_Call): If the
underlying type is real, call Put_Image_Unknown.
(Build_Unknown_Put_Image_Call): Pass the type name to
Put_Image_Unknown.
* libgnat/s-putima.ads, libgnat/s-putima.adb
(Put_Image_Unknown): Add Type_Name parameter.  Remove
overly-detailed documentation of what it does; better to leave
it open.

[Ada] Skip unnecessary iterations over constraint expressions

2020-06-11  Piotr Trojanek  <trojanek@adacore.com>

gcc/ada/

* sem_ch3.adb (Build_Constrained_Array_Type,
Build_Constrained_Discriminated_Type): Skip unnecessary loop
iterations.

[Ada] Move duplicated routines for building itypes to Sem_Util

2020-06-11  Piotr Trojanek  <trojanek@adacore.com>

gcc/ada/

* sem_aggr.adb (Build_Constrained_Itype): Move to Sem_Util.
* sem_ch3.adb (Build_Subtype, Inherit_Predicate_Flags): Move...
* sem_util.adb (Build_Subtype): Here.  Add parameters for
references to objects previously declared in enclosing scopes.
(Inherit_Predicate_Flags): And here, because it is called by
Build_Subtype.
* sem_util.ads (Build_Overriding_Spec): Reorder alphabetically.
(Build_Subtype): Moved from Sem_Ch3; comments updated.
(Build_Constrained_Itype): Moved from Sem_Aggr; comments
updated.

[Ada] Refine type for sorting case-choices tables

2020-06-11  Piotr Trojanek  <trojanek@adacore.com>

gcc/ada/

* sem_aggr.adb (Case_Table_Type): Change index type from Nat to
Pos.

[Ada] Create constrained itypes for nested record aggregates

2020-06-11  Piotr Trojanek  <trojanek@adacore.com>

gcc/ada/

* sem_aggr.adb (Build_Constrained_Itype): Previously a declare
block, now a separate procedure; the only change is that now
New_Assoc_List might include components and an others clause,
which we ignore (while we deal with discriminants exactly as we
did before); extend a ??? comment about how this routine is
different from the Build_Subtype
(Resolve_Record_Aggregate): Create a constrained itype not just
for the outermost record aggregate, but for its inner record
aggregates as well.

[Ada] Consolidate handling of implicit dereferences into semantic analysis

2020-06-11  Eric Botcazou  <ebotcazou@adacore.com>

gcc/ada/

* checks.adb (Build_Discriminant_Checks): Build an explicit
dereference when the type is an access type.
* exp_atag.adb (Build_CW_Membership): Add explicit dereferences.
(Build_Get_Access_Level): Likewise.
(Build_Get_Alignment): Likewise.
(Build_Inherit_Prims): Likewise.
(Build_Get_Transportable): Likewise.
(Build_Set_Size_Function): Likewise.
* exp_ch3.adb (Build_Offset_To_Top_Function): Likewise.
* exp_ch4.adb (Expand_Allocator_Expression): Likewise.
(Expand_N_Indexed_Component ): Remove code dealing with implicit
dereferences.
(Expand_N_Selected_Component): Likewise.
(Expand_N_Slice): Likewise.
* exp_ch9.adb (Add_Formal_Renamings): Add explicit dereference.
(Expand_Accept_Declarations): Likewise.
(Build_Simple_Entry_Call): Remove code dealing with implicit
dereferences.
(Expand_N_Requeue_Statement): Likewise.
* exp_disp.adb (Expand_Dispatching_Call): Build an explicit
dereference when the controlling type is an access type.
* exp_spark.adb (Expand_SPARK_N_Selected_Component): Delete.
(Expand_SPARK_N_Slice_Or_Indexed_Component): Likewise.
(Expand_SPARK): Do not call them.
* sem_ch4.adb (Process_Implicit_Dereference_Prefix): Delete.
(Process_Indexed_Component): Call Implicitly_Designated_Type
to get the designated type for an implicit dereference.
(Analyze_Overloaded_Selected_Component): Do not insert an
explicit dereference here.
(Analyze_Selected_Component): Likewise.
(Analyze_Slice): Call Implicitly_Designated_Type to get the
designated type for an implicit dereference.
* sem_ch8.adb (Has_Components): New predicate extracted from...
(Is_Appropriate_For_Record): ...this.  Delete.
(Is_Appropriate_For_Entry_Prefix): Likewise.
(Analyze_Renamed_Entry): Deal with implicit dereferences.
(Find_Selected_Component): Do not insert an explicit dereference
here.  Call Implicitly_Designated_Type to get the designated type
for an implicit dereference.  Call Has_Components, Is_Task_Type
and Is_Protected_Type directly.  Adjust test for error.
* sem_res.adb (Resolve_Implicit_Dereference): New procedure.
(Resolve_Call): Call Resolve_Indexed_Component last.
(Resolve_Entry): Call Resolve_Implicit_Dereference on the prefix.
(Resolve_Indexed_Component): Call Implicitly_Designated_Type to
get the designated type for an implicit dereference and
Resolve_Implicit_Dereference on the prefix at the end.
(Resolve_Selected_Component): Likewise.
(Resolve_Slice): Likewise.  Do not apply access checks here.
* sem_util.ads (Implicitly_Designated_Type): Declare.
* sem_util.adb (Copy_And_Maybe_Dereference): Simplify.
(Implicitly_Designated_Type): New function.
(Object_Access_Level): Fix typo.
* sem_warn.adb (Check_Unset_Reference): Test Comes_From_Source
on the original node.

[Ada] Remove a dubious optimization for Object Specific Data dispatching

2020-06-11  Piotr Trojanek  <trojanek@adacore.com>

gcc/ada/

* exp_disp.adb: Minor reformatting.
* exp_aggr.adb (Is_Static_Dispatch_Table_Aggregate): Recognize
aggregates of the Ada.Tags.Object_Specific_Data type as static.
* sem_aggr.adb (Check_Static_Discriminated_Subtype): Deconstruct
and do not call it from Build_Constrained_Itype.

[Ada] Make Object Specific Dispatch tables constant

2020-06-11  Piotr Trojanek  <trojanek@adacore.com>

gcc/ada/

* exp_disp.adb (Make_Secondary_DT): Internally generated OSD
tables are now constant.

[Ada] Remove useless code in Backend_Processing_Possible

2020-06-11  Piotr Trojanek  <trojanek@adacore.com>

gcc/ada/

* exp_aggr.adb (Backend_Processing_Possible): Remove useless
call.

[Ada] Fix punctuation in comment of Building_Static_Dispatch_Tables

2020-06-11  Piotr Trojanek  <trojanek@adacore.com>

gcc/ada/

* opt.ads (Building_Static_Dispatch_Tables): Fix punctuation in
comment.

[Ada] AI12-0356 Root_Storage_Pool_With_Subpools & Preelaborable_Init

2020-06-11  Arnaud Charlet  <charlet@adacore.com>

gcc/ada/

* libgnat/s-stposu.ads (Root_Storage_Pool_With_Subpools,
Root_Subpool): Mark with Preelaborable_Initialization.

[Ada] Crash on dispatching conditional entry call

2020-06-11  Javier Miranda  <miranda@adacore.com>

gcc/ada/

* exp_ch9.adb (Expand_N_Conditional_Entry_Call): Replace call to
New_Copy_List by calls to the new routine
New_Copy_Separate_List.
* sem_util.ads (New_Copy_Separate_List, New_Copy_Separate_Tree):
New routines.

* sem_util.adb (New_Copy_Separate_List, New_Copy_Separate_Tree):
New routines.
(New_Copy_Tree): Extend the machinery that detects syntactic
nodes to handle lists of indentifiers with field More_Ids;
otherwise such nodes are erroneously handled as semantic nodes.
Copy aspect specifications attached to nodes.
* sem_ch12.adb (Copy_Generic_Node): Protect reading attribute
Etype.

[Ada] Fix unnesting crash with Predicate_Failure/no pred

2020-06-11  Bob Duff  <duff@adacore.com>

gcc/ada/

* sem_ch13.adb (Analyze_Aspect_Specifications): Do not set the
Has_Predicates flag when the Predicate_Failure aspect is seen.
It is legal (but pointless) to use this aspect without a
predicate.  If we set the flag, we generate a half-baked
Predicate procedure, and if that procedure is nested, it causes
unnesting to crash.

[Ada] Put_Image attribute

2020-06-11  Bob Duff  <duff@adacore.com>

gcc/ada/

* exp_put_image.adb (Build_Record_Put_Image_Procedure): Remove
special processing of protected types, because those are handled
by Build_Protected_Put_Image_Call.
(Enable_Put_Image): Use the switch -gnatd_z to control enabling
of Put_Image. Disable Put_Image for types in Remote_Types
packages.
* debug.adb: Document -gnatd_z switch.
* exp_imgv.adb, libgnat/a-stteou.ads, opt.ads: Minor cleanups.

[Ada] Simplify iteration over formal parameters for aliasing error

2020-06-11  Piotr Trojanek  <trojanek@adacore.com>

gcc/ada/

* sem_warn.adb (Warn_On_Overlapping_Actuals): Add label to the
outer loop and use it in the exit statement.

[Ada] Generate predicate checks for on assignments in records

2020-06-11  Ghjuvan Lacambre  <lacambre@adacore.com>

gcc/ada/

* exp_ch3.adb (Build_Assignment): Generate predicate check if
subtype mark has predicate.

[Ada] Missing accessibility error on object in type conversion

2020-06-11  Justin Squirek  <squirek@adacore.com>

gcc/ada/

* sem_util.adb (Expand_N_Attribute_Reference): Use original
nodes where required to avoid looking at the expanded tree.

[Ada] Remove aspects that were commented out

2020-06-11  Arnaud Charlet  <charlet@adacore.com>

gcc/ada/

* libgnat/a-catizo.ads, libgnat/a-nbnbin.ads,
libgnat/a-nbnbre.ads, libgnat/a-nubinu.ads,
libgnat/s-aoinar.ads, libgnat/s-aomoar.ads,
libgnat/s-aotase.ads, libgnat/s-stopoo.ads: Remove aspects that
we will not implement.

[Ada] Add fallback on Integer_Arithmetic

2020-06-11  Arnaud Charlet  <charlet@adacore.com>

gcc/ada/

* libgnat/s-aoinar.adb (Atomic_Fetch_And_Add,
Atomic_Fetch_And_Subtract): Add fallback using
compare-and-exchange, in case the integer type does not map to a
machine type.

asan: fix RTX emission for ilp32

gcc/ChangeLog:

PR sanitizer/95634
* asan.c (asan_emit_stack_protection): Fix emission for ilp32
by using Pmode instead of ptr_mode.

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

vect: Rename things related to rgroup_masks

Power supports vector memory access with length (in bytes) instructions.
Like existing fully masking for SVE, it is another approach to vectorize
the loop using partially-populated vectors.

As Richard Sandiford pointed out, we can rename the rgroup struct
rgroup_masks to rgroup_controls, rename its members mask_type to type,
masks to controls to be more generic.

Besides, this patch also renames some functions like vect_set_loop_mask
to vect_set_loop_control, release_vec_loop_masks to
release_vec_loop_controls, vect_set_loop_masks_directly to
vect_set_loop_controls_directly.

Bootstrapped/regtested on aarch64-linux-gnu.

gcc/ChangeLog:

* tree-vect-loop-manip.c (vect_set_loop_mask): Renamed to ...
(vect_set_loop_control): ... this.
(vect_maybe_permute_loop_masks): Rename rgroup_masks related things.
(vect_set_loop_masks_directly): Renamed to ...
(vect_set_loop_controls_directly): ... this.  Also rename some
variables with ctrl instead of mask.  Rename vect_set_loop_mask to
vect_set_loop_control.
(vect_set_loop_condition_masked): Rename rgroup_masks related things.
Also rename some variables with ctrl instead of mask.
* tree-vect-loop.c (release_vec_loop_masks): Renamed to ...
(release_vec_loop_controls): ... this.  Rename rgroup_masks related
things.
(_loop_vec_info::~_loop_vec_info): Rename release_vec_loop_masks to
release_vec_loop_controls.
(can_produce_all_loop_masks_p): Rename rgroup_masks related things.
(vect_get_max_nscalars_per_iter): Likewise.
(vect_estimate_min_profitable_iters): Likewise.
(vect_record_loop_mask): Likewise.
(vect_get_loop_mask): Likewise.
* tree-vectorizer.h (struct rgroup_masks): Renamed to ...
(struct rgroup_controls): ... this.  Also rename mask_type
to type and rename masks to controls.

vect: Rename fully_masked_p to using_partial_vectors_p

Power supports vector memory access with length (in bytes) instructions.
Like existing fully masking for SVE, it is another approach to vectorize
the loop using partially-populated vectors.

As Richard Sandiford suggested, this patch is to update the existing
fully_masked_p field to using_partial_vectors_p.  Introduce one macro
LOOP_VINFO_USING_PARTIAL_VECTORS_P for partial vectorization checking
usage, update the LOOP_VINFO_FULLY_MASKED_P with
LOOP_VINFO_USING_PARTIAL_VECTORS_P && !masks.is_empty() and still use
it for mask-based partial vectors approach specific checks.

Bootstrapped/regtested on aarch64-linux-gnu.

gcc/ChangeLog:

* tree-vect-loop-manip.c (vect_set_loop_condition): Rename
LOOP_VINFO_FULLY_MASKED_P to LOOP_VINFO_USING_PARTIAL_VECTORS_P.
(vect_gen_vector_loop_niters): Likewise.
(vect_do_peeling): Likewise.
* tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Rename
fully_masked_p to using_partial_vectors_p.
(vect_analyze_loop_costing): Rename LOOP_VINFO_FULLY_MASKED_P to
LOOP_VINFO_USING_PARTIAL_VECTORS_P.
(determine_peel_for_niter): Likewise.
(vect_estimate_min_profitable_iters): Likewise.
(vect_transform_loop): Likewise.
* tree-vectorizer.h (LOOP_VINFO_FULLY_MASKED_P): Updated.
(LOOP_VINFO_USING_PARTIAL_VECTORS_P): New macro.

vect: Rename can_fully_mask_p to can_use_partial_vectors_p

Power supports vector memory access with length (in bytes) instructions.
Like existing fully masking for SVE, it is another approach to vectorize
the loop using partially-populated vectors.

As Richard Sandiford pointed out, we should extend the existing flag
can_fully_mask_p to be more generic, to indicate whether we have
any chances with partial vectors for this loop.  So this patch
is to rename this flag to can_use_partial_vectors_p to be more
meaningful, also rename the macro LOOP_VINFO_CAN_FULLY_MASK_P
to LOOP_VINFO_CAN_USE_PARTIAL_VECTORS_P.

Bootstrapped/regtested on aarch64-linux-gnu.

gcc/ChangeLog:

* tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Rename
can_fully_mask_p to can_use_partial_vectors_p.
(vect_analyze_loop_2): Rename LOOP_VINFO_CAN_FULLY_MASK_P to
LOOP_VINFO_CAN_USE_PARTIAL_VECTORS_P.  Rename saved_can_fully_mask_p
to saved_can_use_partial_vectors_p.
(vectorizable_reduction): Rename LOOP_VINFO_CAN_FULLY_MASK_P to
LOOP_VINFO_CAN_USE_PARTIAL_VECTORS_P.
(vectorizable_live_operation): Likewise.
* tree-vect-stmts.c (permute_vec_elements): Likewise.
(check_load_store_masking): Likewise.
(vectorizable_operation): Likewise.
(vectorizable_store): Likewise.
(vectorizable_load): Likewise.
(vectorizable_condition): Likewise.
* tree-vectorizer.h (LOOP_VINFO_CAN_FULLY_MASK_P): Renamed to ...
(LOOP_VINFO_CAN_USE_PARTIAL_VECTORS_P): ... this.
(_loop_vec_info): Rename can_fully_mask_p to can_use_partial_vectors_p.

prepare-commit-hook: Use gcc-config.diff-file.

contrib/ChangeLog:

* prepare-commit-msg: Replace ENV variable with a git config
value.

contrib: Avoid redundant 'git diff' in prepare-commit-msg hook

contrib/ChangeLog:

* prepare-commit-msg: Use 'tee' to save the diff to a file
instead of running 'git diff' twice.

Fix -Wformat-diag in options-save.c

The patch removes bunch of warnings:

options-save.c:12004:29: warning: unquoted identifier or keyword ‘global_options’ in format [-Wformat-diag]
12004 |     internal_error ("Error: global_options are modified in local context\n");

gcc/ChangeLog:

* optc-save-gen.awk: Quote error string.

slim up mem exprs to avoid line breaks in -fverbose-asm

An asm operand with a "VIEW_CONVERT_EXPR<struct {
  [...]
}>" will output the definition of the struct as asm code.  Oops.

Enable TDF_SLIM in print_mem_expr to avoid such line breaks.

for  gcc/ChangeLog

* print-rtl.c (print_mem_expr): Enable TDF_SLIM in dump_flags.

RISC-V: Unify the output asm pattern between gpr_save and gpr_restore pattern.

gcc/ChangeLog:

* config/riscv/riscv-protos.h (riscv_output_gpr_save): Remove.
* config/riscv/riscv-sr.c (riscv_sr_match_prologue): Update
value.
* config/riscv/riscv.c (riscv_output_gpr_save): Remove.
* config/riscv/riscv.md (gpr_save): Update output asm pattern.

RISC-V: Describe correct USEs for gpr_save pattern [PR95252]

 - Verified on rv32emc/rv32gc/rv64gc bare-metal target and rv32gc/rv64gc
   linux target with qemu.

gcc/ChangeLog:

* config/riscv/predicates.md (gpr_save_operation): New.
* config/riscv/riscv-protos.h (riscv_gen_gpr_save_insn): New.
(riscv_gpr_save_operation_p): Ditto.
* config/riscv/riscv-sr.c (riscv_remove_unneeded_save_restore_calls):
Ignore USEs for gpr_save patter.
* config/riscv/riscv.c (gpr_save_reg_order): New.
(riscv_expand_prologue): Use riscv_gen_gpr_save_insn to gen gpr_save.
(riscv_gen_gpr_save_insn): New.
(riscv_gpr_save_operation_p): Ditto.
* config/riscv/riscv.md (S3_REGNUM): New.
(S4_REGNUM): Ditto.
(S5_REGNUM): Ditto.
(S6_REGNUM): Ditto.
(S7_REGNUM): Ditto.
(S8_REGNUM): Ditto.
(S9_REGNUM): Ditto.
(S10_REGNUM): Ditto.
(S11_REGNUM): Ditto.
(gpr_save): Model USEs correctly.

gcc/testsuite/ChangeLog:

* gcc.target/riscv/pr95252.c: New.

Daily bump.

libstdc++: Fix some ranges algos optimizations [PR95578]

ranges::copy and a number of other ranges algorithms have unwrapping
optimizations for iterators of type __normal_iterator, move_iterator and
reverse_iterator.  But in the checks that guard these optimizations we
currently only test that the iterator of the iterator/sentinel pair has
the appropriate type before proceeding with the corresponding
optimization, and do not also test the sentinel type.

This breaks the testcase in this PR because this testcase constructs via
range adaptors a range whose begin() is a __normal_iterator and whose
end() is a custom sentinel type, and then performs ranges::copy on it.
From there we bogusly perform the __normal_iterator unwrapping
optimization on this iterator/sentinel pair, which immediately leads to
a constraint failure since the custom sentinel type does not model
sentinel_for<int*>.

This patch fixes this issue by refining each of the problematic checks
to also test that the iterator and sentinel types are the same before
applying the corresponding unwrapping optimization.  Along the way, some
code simplifications are made.

libstdc++-v3/ChangeLog:

PR libstdc++/95578
* include/bits/ranges_algo.h (__lexicographical_compare_fn):
Also check that the iterator and sentinel have the same type before
applying the unwrapping optimization for __normal_iterator.
Split the check into two, one for the first iterator/sentinel
pair and another for second iterator/sentinel pair.  Remove uses
of __niter_base, and remove uses of std::move on a
__normal_iterator.
* include/bits/ranges_algobase.h (__equal_fn): Likewise.
(__copy_or_move): Likewise.  Perform similar adjustments for
the reverse_iterator and move_iterator optimizations.  Inline
the checks into the if-constexprs, and use using-declarations to
make them less visually noisy.  Remove uses of __niter_wrap.
(__copy_or_move_backward): Likewise.
* testsuite/25_algorithms/copy/95578.cc: New test.
* testsuite/25_algorithms/copy_backward/95578.cc: New test.
* testsuite/25_algorithms/equal/95578.cc: New test.
* testsuite/25_algorithms/lexicographical_compare/95578.cc: New test.
* testsuite/25_algorithms/move/95578.cc: New test.
* testsuite/25_algorithms/move_backward/95578.cc: New test.

[PR51447] restore the global reg var before returning from main

A runtime system might legitimately hold in rbx a value expected to be
preserved across the call to main, but its use as a global register
variable stops main from preserving it.

for  gcc/testsuite/ChangeLog

PR rtl-optimization/51447
* gcc.c-torture/execute/pr51447.c (main): Preserve call-saved
register.

coroutines: Make call argument handling more robust [PR95440]

build_new_method_call is supposed to be able to handle a null
arguments list pointer (when the method has no parms).  There
were a couple of places where uses of the argument list pointer
were not defended against NULL.

gcc/cp/ChangeLog:

PR c++/95440
* call.c (add_candidates): Use vec_safe_length() for
testing the arguments list.
(build_new_method_call_1): Use vec_safe_is_empty() when
checking for an empty args list.

gcc/testsuite/ChangeLog:

PR c++/95440
* g++.dg/coroutines/pr95440.C: New test.

PR middle-end/95353 - spurious -Wstringop-overflow writing to a trailing array plus offset

Also resolves:
PR middle-end/92939 - missing -Wstringop-overflow on negative index from the end of array

gcc/ChangeLog:

PR middle-end/95353
PR middle-end/92939
* builtins.c (inform_access): New function.
(check_access): Call it.  Add argument.
(addr_decl_size): Remove.
(get_range): New function.
(compute_objsize): New overload.  Only use compute_builtin_object_size
with raw memory function.
(check_memop_access): Pass new argument to compute_objsize and
check_access.
(expand_builtin_memchr, expand_builtin_strcat): Same.
(expand_builtin_strcpy, expand_builtin_stpcpy_1): Same.
(expand_builtin_stpncpy, check_strncat_sizes): Same.
(expand_builtin_strncat, expand_builtin_strncpy): Same.
(expand_builtin_memcmp): Same.
* builtins.h (check_nul_terminated_array): Declare extern.
(check_access): Add argument.
(struct access_ref, struct access_data): New structs.
* gimple-ssa-warn-restrict.c (clamp_offset): New helper.
(builtin_access::overlap): Call it.
* tree-object-size.c (decl_init_size): Declare extern.
(addr_object_size): Correct offset computation.
* tree-object-size.h (decl_init_size): Declare.
* tree-ssa-strlen.c (handle_integral_assign): Remove a call
to maybe_warn_overflow when assigning to an SSA_NAME.

gcc/testsuite/ChangeLog:

PR middle-end/95353
PR middle-end/92939
* c-c++-common/Wstringop-truncation.c: Remove an xfail.
* gcc.dg/Warray-bounds-46.c: Remove a bogus warning.
* gcc.dg/Wrestrict-9.c: Disable -Wstringop-overflow.
* gcc.dg/Wstringop-overflow-12.c: Remove xfails.
* gcc.dg/Wstringop-overflow-28.c: Same.
* gcc.dg/builtin-stringop-chk-4.c: Same.
* gcc.dg/builtin-stringop-chk-5.c: Same.
* gcc.dg/builtin-stringop-chk-8.c: Same.
* gcc.dg/strlenopt-74.c: Avoid buffer overflow.
* gcc.dg/Wstringop-overflow-34.c: New test.
* gcc.dg/Wstringop-overflow-35.c: New test.
* gcc.dg/Wstringop-overflow-36.c: New test.
* gcc.dg/Wstringop-overflow-37.c: New test.
* gcc.dg/Wstringop-overflow-38.c: New test.

libstdc++: Extend memcmp optimization in std::lexicographical_compare

Make the memcmp optimization work for std::deque iterators and safe
iterators.

Co-authored-by: Jonathan Wakely <jwakely@redhat.com>
libstdc++-v3/ChangeLog:

2020-06-08  François Dumont  <fdumont@gcc.gnu.org>
    Jonathan Wakely  <jwakely@redhat.com>

* include/bits/deque.tcc (__lex_cmp_dit): New.
(__lexicographical_compare_aux1): Define overloads for deque
iterators.
* include/bits/stl_algobase.h (__lexicographical_compare::__3way):
New static member function.
(__lexicographical_compare<true>::__3way): Likewise.
(__lexicographical_compare<true>::__lc): Use __3way.
(__lexicographical_compare_aux): Rename to
__lexicographical_compare_aux1 and declare overloads for deque
iterators.
(__lexicographical_compare_aux): Define new forwarding function
that calls __lexicographical_compare_aux1 and declare new overloads
for safe iterators.
(lexicographical_compare): Do not use __niter_base on
parameters.
* include/debug/safe_iterator.tcc
(__lexicographical_compare_aux): Define overloads for safe
iterators.
* testsuite/25_algorithms/lexicographical_compare/1.cc: Add
checks with random access iterators.
* testsuite/25_algorithms/lexicographical_compare/deque_iterators/1.cc:
New test.

avoid stmt-info allocation for debug stmts

The following avoids allocating stmt info structs for debug stmts.

2020-06-10  Richard Biener  <rguenther@suse.de>

* tree-vect-loop.c (vect_determine_vectorization_factor):
Skip debug stmts.
(_loop_vec_info::_loop_vec_info): Likewise.
(vect_update_vf_for_slp): Likewise.
(vect_analyze_loop_operations): Likewise.
(update_epilogue_loop_vinfo): Likewise.
* tree-vect-patterns.c (vect_determine_precisions): Likewise.
(vect_pattern_recog): Likewise.
* tree-vect-slp.c (vect_detect_hybrid_slp): Likewise.
(_bb_vec_info::_bb_vec_info): Likewise.
* tree-vect-stmts.c (vect_mark_stmts_to_be_vectorized):
Likewise.

tree-optimization/95576 - fix compare-debug issue with SLP vectorization

The following avoids leading debug stmts in BB vectorizer regions.

2020-06-10  Richard Biener  <rguenther@suse.de>

PR tree-optimization/95576
* tree-vect-slp.c (vect_slp_bb): Skip leading debug stmts.

* g++.dg/vect/pr95576.cc: New testcase.

aarch64: Fix an ICE in register_tuple_type [PR95523]

When registering the tuple type in register_tuple_type, the
TYPE_ALIGN (tuple_type) will be changed by -fpack-struct=n. We need to
maintain natural alignment in handle_arm_sve_h.

2020-06-10  Haijian Zhang  <z.zhanghaijian@huawei.com>

gcc/
PR target/95523
* config/aarch64/aarch64-sve-builtins.h
(sve_switcher::m_old_maximum_field_alignment): New member.
* config/aarch64/aarch64-sve-builtins.cc
(sve_switcher::sve_switcher): Save maximum_field_alignment in
m_old_maximum_field_alignment and clear maximum_field_alignment.
(sve_switcher::~sve_switcher): Restore maximum_field_alignment.

gcc/testsuite/
PR target/95523
* gcc.target/aarch64/sve/pr95523.c: New test.

[Ada] AI12-0364 Add a modular atomic arithmetic package

2020-06-10  Arnaud Charlet  <charlet@adacore.com>

gcc/ada/

* libgnat/s-aomoar.ads, libgnat/s-aomoar.adb: New files.
* libgnat/s-atopar.ads: Move...
* libgnat/s-aoinar.ads: Here.
* libgnat/s-atopar.adb: Move...
* libgnat/s-aoinar.adb: Here.
* impunit.adb: Update list of runtime files.
* Makefile.rtl (GNATRTL_NONTASKING_OBJS=): Adjust.

gcc-changelog: fix parse_git_name_status for renames.

Renamed files are listed in the following format:

M gcc/ada/Makefile.rtl
M gcc/ada/impunit.adb
R097 gcc/ada/libgnat/s-atopar.adb gcc/ada/libgnat/s-aoinar.adb
R095 gcc/ada/libgnat/s-atopar.ads gcc/ada/libgnat/s-aoinar.ads
A gcc/ada/libgnat/s-aomoar.adb
A gcc/ada/libgnat/s-aomoar.ads

So 'R' is followed by a percentage number.

contrib/ChangeLog:

* gcc-changelog/git_commit.py: Fix renamed files in
parse_git_name_status.
* gcc-changelog/test_email.py: Add test for it.

c++: Fix ICE with delayed parsing of noexcept-specifier [PR95562]

Here we ICE because a DEFERRED_PARSE expression leaked to tsubst_copy.
We create these expressions for deferred noexcept-specifiers in
cp_parser_save_noexcept; they are supposed to be re-parsed in
cp_parser_late_noexcept_specifier.  In this case we never got around
to re-parsing it because the noexcept-specifier was attached to a
pointer to a function, not to a function declaration.  But we should
not have delayed the parsing here in the first place; we already
avoid delaying the parsing for alias-decls, typedefs, and friend
function declarations.  (Clang++ also doesn't delay the parsing
for pointers to function.)

gcc/cp/ChangeLog:

PR c++/95562
* parser.c (cp_parser_direct_declarator): Clear
CP_PARSER_FLAGS_DELAY_NOEXCEPT if the declarator kind is not
cdk_id.

gcc/testsuite/ChangeLog:

PR c++/95562
* g++.dg/cpp0x/noexcept60.C: New test.

[Ada] AI12-0311 New checks for language-defined units

2020-06-10  Arnaud Charlet  <charlet@adacore.com>

gcc/ada/

* snames.ads-tmpl (Name_Characters_Assertion_Check,
Name_Containers_Assertion_Check,
Name_Interfaces_Assertion_Check, Name_IO_Assertion_Check,
Name_Numerics_Assertion_Check, Name_Strings_Assertion_Check,
Name_System_Assertion_Check): New constants.
* types.ads (Characters_Assertion_Check,
Containers_Assertion_Check, Interfaces_Assertion_Check,
IO_Assertion_Check, Numerics_Assertion_Check,
Strings_Assertion_Check, System_Assertion_Check): New constants.
(All_Checks): Update accordingly.

[Ada] Don't build equivalent record aggregate if type has predicates

2020-06-10  Ghjuvan Lacambre  <lacambre@adacore.com>

gcc/ada/

* exp_ch3.adb (Build_Equivalent_Record_Aggregate): Return Empty
if Etype of record component has predicates.

[Ada] Ada 202x AI12-0192 "requires late initialization"

2020-06-10  Arnaud Charlet  <charlet@adacore.com>

gcc/ada/

* exp_ch3.adb (Build_Init_Statements): Implement the notion of
"require late initialization".

[Ada] Add missing Sloc on new explicit dereferences

2020-06-10  Eric Botcazou  <ebotcazou@adacore.com>

gcc/ada/

* sem_util.adb (Copy_And_Maybe_Dereference): Temporarily copy
the parent node of the original tree when dereferencing.

[Ada] Additional warnings on overlapping actuals of composite types

2020-06-10  Ed Schonberg  <schonberg@adacore.com>

gcc/ada/

* sem_warn.adb (Warn_On_Overlapping_Actuals): Add a warning when
two actuals in a call overlap, both are composite types that may
be passed by reference, and only one of them is writable.

[Ada] Implement AI12-0162 Memberships and Unchecked_Unions

2020-06-10  Eric Botcazou  <ebotcazou@adacore.com>

gcc/ada/

* exp_ch4.adb (Expand_N_In): Use an expression with actions to
insert the PE raise statement for the Unchecked_Union case.

[Ada] Remove obsolete code in Resolve_Call

2020-06-10  Eric Botcazou  <ebotcazou@adacore.com>

gcc/ada/

* sem_ch4.adb (Analyze_Call): Use idiomatic condition.
* sem_res.adb (Resolve_Call): Remove obsolete code.

[Ada] Insert explicit dereferences when building actual subtype

2020-06-10  Eric Botcazou  <ebotcazou@adacore.com>

gcc/ada/

* sem_util.adb (Copy_And_Maybe_Dereference): New function.
(Build_Access_Record_Constraint): Use it to copy the prefix.
(Build_Actual_Array_Constraint): Likewise.
(Build_Actual_Record_Constraint): Likewise.

[Ada] Disable unwanted warnings in Assertion_Policy(Ignore) mode

2020-06-10  Bob Duff  <duff@adacore.com>

gcc/ada/

* sem_prag.adb (Invariant): Remove the pragma removing code.  It
doesn't work to remove the pragma, because various flags are set
during Build_Invariant_Procedure_Declaration and
Build_Invariant_Procedure_Body that need to be set to avoid the
spurious warnings.
* exp_util.adb (Make_Invariant_Call): Avoid calling the
invariant-checking procedure if the body is empty. This is an
optimization.

[Ada] Fix gnatmetric switches description

2020-06-10  Vasiliy Fofanov  <fofanov@adacore.com>

gcc/ada/

* doc/gnat_ugn/gnat_utility_programs.rst: Fix gnatmetric
switches description.

[Ada] Update headers

2020-06-10  Arnaud Charlet  <charlet@adacore.com>

gcc/ada/

* repinfo-input.ads, repinfo-input.adb, repinfo.adb,
repinfo.ads: Update header.

[Ada] Improve code generated for dynamic discriminated aggregate

2020-06-10  Eric Botcazou  <ebotcazou@adacore.com>

gcc/ada/

* exp_aggr.adb (In_Place_Assign_OK): Do not necessarily return
false for a type with discriminants.
(Convert_To_Assignments): Use Parent_Node and Parent_Kind more
consistently.  In the in-place assignment case, first apply a
discriminant check if need be, and be prepared for a rewritten
aggregate as a result.

[Ada] Remove more references to ASIS

2020-06-10  Arnaud Charlet  <charlet@adacore.com>

gcc/ada/

* atree.adb, contracts.adb, debug.adb, freeze.adb,
repinfo-input.adb, repinfo.adb, sem_attr.adb, sem_ch10.adb,
sem_ch13.adb, sem_ch3.adb, sem_ch5.adb, sem_ch6.adb,
sem_ch8.adb, sem_ch9.adb, sem_disp.adb, sem_eval.adb,
sem_prag.adb: Remove more references to ASIS.

[Ada] Fix typo in exception message

2020-06-10  Arnaud Charlet  <charlet@adacore.com>

gcc/ada/

* libgnat/s-secsta.adb (Round_Up): Fix typo in exception
message.

[Ada] Fix minor typo in comment of membership choice parsing

2020-06-10  Piotr Trojanek  <trojanek@adacore.com>

gcc/ada/

* par-ch4.adb (P_Membership_Test): Fix typo in a grammar rule.

[Ada] Revert workaround for expansion of Enum_Rep in GNATprove mode

2020-06-10  Piotr Trojanek  <trojanek@adacore.com>

gcc/ada/

* exp_spark.adb (Expand_SPARK_N_Attribute_Reference): Remove
expansion of First and Last attributes.

[Ada] Fold Enum_Rep attribute in evaluation and not in expansion

2020-06-10  Piotr Trojanek  <trojanek@adacore.com>

gcc/ada/

* exp_attr.adb (Expand_N_Attribute_Reference): Remove folding
for Enum_Rep attribute.
* exp_spark.adb (Expand_SPARK_N_Attribute_Reference): Remove
duplicated code for folding Enum_Rep attribute.
* sem_attr.adb (Eval_Attribute): Relax condition for folding
Enum_Rep attribute; previously dead code is now executed when
the attribute prefix is an enumeration literal; refine type in
processing of Enum_Val.

[Ada] Classwide controlled obj not dispatching

2020-06-10  Javier Miranda  <miranda@adacore.com>

gcc/ada/

* sem_ch3.adb (Analyze_Declarations): Adjust the machinery that
takes care of late body overriding of initialize, adjust,
finalize.  Remove ASIS mode code.

[Ada] Ada_2020 AI12-0220: Pre/Postconditions on Access_To_Subprogram types

2020-06-10  Ed Schonberg  <schonberg@adacore.com>

gcc/ada/

* einfo.ads (Access_Subprogram_Wrapper): New attribute of
Subprogram_Type entities. Denotes subprogram constructed for
Access_To_Subprogram types that include pre- and postconditions.
* einfo.adb: Subprogram bodies for Access_Subprogram_Wrapper.
* exp_ch6.adb (Expand_Call): An indirect call through an
Access_To_subprogram that includes contracts is rewritten as a
call to the corresponding Access_ ubprogram_Wrapper. Handle
derived types that inherit contract from parent.
* sem_prag.adb (Build_Access_Subprogram_Wrapper): Build
subprogram declaration for subprogram that incorporates the
contracts of an Access_To_Subprogram type declaration. Build
corresponding body and attach it to freeze actions for type.
* sem_util.ads, sem_util.adb (Is_Access_Subprogram_Wrapper):
Utility that uses signature of the subprogram to determine
whether it is a generated wrapper for an Access_To_Subprogram
type.

[Ada] Fix assertion failure on functions with contracts

2020-06-10  Ghjuvan Lacambre  <lacambre@adacore.com>

gcc/ada/

* par-ch6.adb (P_Subprogram): Make sure the specification
belongs to a procedure.

[Ada] Simplify detection of static membership choices

2020-06-10  Piotr Trojanek  <trojanek@adacore.com>

gcc/ada/

* sem_ch13.adb (All_Membership_Choices_Static): Assert an AST
property documented in sinfo.ads and simplify an excessive
condition.

[Ada] Minor fix style and typos in comments

2020-06-10  Piotr Trojanek  <trojanek@adacore.com>

gcc/ada/

* exp_attr.adb (Expand_N_Attribute_Reference): Fix a copy-paste
mistake in comment.
* sem_res.adb (Flag_Effectively_Volatile_Objects): Fix a type in
the SPARK RM rule number.
* exp_ch4.adb, sem_util.adb: Fix style in single line comments.

[Ada] Remove unreferenced GNATprove utility routine Get_Low_Bound

2020-06-10  Piotr Trojanek  <trojanek@adacore.com>

gcc/ada/

* sem_aux.ads, sem_aux.adb (Get_Low_Bound): Remove.

[Ada] Remove Determine_License

2020-06-10  Arnaud Charlet  <charlet@adacore.com>

gcc/ada/

* scn.adb (Determine_License): Remove.

[Ada] Incorrect accessibility checks on functions calls

2020-06-10  Justin Squirek  <squirek@adacore.com>

gcc/ada/

* exp_ch3.adb (Expand_N_Object_Declaration): Add condition to
handle processing of objects initialized by a call to a function
return an anonymous access type.
* exp_ch6.adb, exp_ch6.ads
(Has_Unconstrained_Access_Discriminants): Moved to sem_util.adb
(Needs_Result_Accessibility_Level): Moved to sem_util.adb
* sem_util.adb, sem_util.ads
(Has_Unconstrained_Access_Discriminants): Moved from exp_ch6.adb
(Needs_Result_Accessibility_Level): Moved from exp_ch6.adb
* sem_res.adb (Valid_Conversion): Add condition for the special
case where the operand of a conversion is the result of an
anonymous access type

[Ada] Reject illegal bodies for null procedures

2020-06-10  Ghjuvan Lacambre  <lacambre@adacore.com>

gcc/ada/

* par-ch6.adb (P_Subprogram): Reject duplicate subprogram
declarations.

[Ada] Fix incorrect insertion of post-call actions in if-expression

2020-06-10  Eric Botcazou  <ebotcazou@adacore.com>

gcc/ada/

* exp_ch6.adb (Insert_Post_Call_Actions): Deal with the context
of an if-expression and with a call written in prefixed notation.

[Ada] Minor reformatting and some typo fixes

2020-06-10  Gary Dismukes  <dismukes@adacore.com>

gcc/ada/

* einfo.ads: Minor reformatting of a comment.
* exp_aggr.adb: Minor reformatting and a grammar correction.
* exp_attr.adb: Minor reformatting and a typo fix in some
comments.
* sem_ch12.adb: Fix three typos in comments.

Make {SLP_TREE,STMT_VINFO}_VEC_STMTS a vector of gimple *

This makes {SLP_TREE,STMT_VINFO}_VEC_STMTS a vector of gimple * and
not allocate a stmt_vec_info for vectorizer generated stmts since
this is now possible after removing the only use which was chaining
of vector stmts via STMT_VINFO_RELATED_STMT.

This also removes all stmt_vec_info allocations done for vector
stmts, the remaining ones are for stmts in the scalar IL and for
patterns which are not part of the IL.  Thus after this the stmt
UIDs inside a basic-block are suitable for dominance checking
if you ignore (or lazy-fill) UIDs of zero of the vector stmts
inserted during transform.  This property is ensured by a new
flag set when pattern analysis is complete.

2020-06-10  Richard Biener  <rguenther@suse.de>

* tree-vectorizer.h (_slp_tree::vec_stmts): Make it a vector
of gimple * stmts.
(_stmt_vec_info::vec_stmts): Likewise.
(vec_info::stmt_vec_info_ro): New flag.
(vect_finish_replace_stmt): Adjust declaration.
(vect_finish_stmt_generation): Likewise.
(vectorizable_induction): Likewise.
(vect_transform_reduction): Likewise.
(vectorizable_lc_phi): Likewise.
* tree-vect-data-refs.c (vect_create_data_ref_ptr): Do not
allocate stmt infos for increments.
(vect_record_grouped_load_vectors): Adjust.
* tree-vect-loop.c (vect_create_epilog_for_reduction): Likewise.
(vectorize_fold_left_reduction): Likewise.
(vect_transform_reduction): Likewise.
(vect_transform_cycle_phi): Likewise.
(vectorizable_lc_phi): Likewise.
(vectorizable_induction): Likewise.
(vectorizable_live_operation): Likewise.
(vect_transform_loop): Likewise.
* tree-vect-patterns.c (vect_pattern_recog): Set stmt_vec_info_ro.
* tree-vect-slp.c (vect_get_slp_vect_def): Adjust.
(vect_get_slp_defs): Likewise.
(vect_transform_slp_perm_load): Likewise.
(vect_schedule_slp_instance): Likewise.
(vectorize_slp_instance_root_stmt): Likewise.
* tree-vect-stmts.c (vect_get_vec_defs_for_operand): Likewise.
(vect_finish_stmt_generation_1): Do not allocate a stmt info.
(vect_finish_replace_stmt): Do not return anything.
(vect_finish_stmt_generation): Likewise.
(vect_build_gather_load_calls): Adjust.
(vectorizable_bswap): Likewise.
(vectorizable_call): Likewise.
(vectorizable_simd_clone_call): Likewise.
(vect_create_vectorized_demotion_stmts): Likewise.
(vectorizable_conversion): Likewise.
(vectorizable_assignment): Likewise.
(vectorizable_shift): Likewise.
(vectorizable_operation): Likewise.
(vectorizable_scan_store): Likewise.
(vectorizable_store): Likewise.
(vectorizable_load): Likewise.
(vectorizable_condition): Likewise.
(vectorizable_comparison): Likewise.
(vect_transform_stmt): Likewise.
* tree-vectorizer.c (vec_info::vec_info): Initialize
stmt_vec_info_ro.
(vec_info::replace_stmt): Copy over stmt UID rather than
unsetting/setting a stmt info allocating a new UID.
(vec_info::set_vinfo_for_stmt): Assert !stmt_vec_info_ro.

Merge evrp uses of substitute_and_fold_engine into the engine itself.

This patch merges the evrp uses of the substitute and fold engine into
the engine itself, at least the parts that can be re-used by other
engine uses.  It also adds a context parameter to get_value() for
further use.

gcc/
* gimple-loop-versioning.cc (loop_versioning::name_prop::get_value):
Add stmt parameter.
* gimple-ssa-evrp.c (class evrp_folder): New.
(class evrp_dom_walker): Remove.
(execute_early_vrp): Use evrp_folder instead of evrp_dom_walker.
* tree-ssa-ccp.c (ccp_folder::get_value): Add stmt parameter.
* tree-ssa-copy.c (copy_folder::get_value): Same.
* tree-ssa-propagate.c (substitute_and_fold_engine::replace_uses_in):
Pass stmt to get_value.
(substitute_and_fold_engine::replace_phi_args_in): Same.
(substitute_and_fold_dom_walker::after_dom_children): Call
post_fold_bb.
(substitute_and_fold_dom_walker::foreach_new_stmt_in_bb): New.
(substitute_and_fold_dom_walker::propagate_into_phi_args): New.
(substitute_and_fold_dom_walker::before_dom_children): Adjust to
call virtual functions for folding, pre_folding, and post folding.
Call get_value with PHI.  Tweak dump.
* tree-ssa-propagate.h (class substitute_and_fold_engine):
New argument to get_value.
New virtual function pre_fold_bb.
New virtual function post_fold_bb.
New virtual function pre_fold_stmt.
New virtual function post_new_stmt.
New function propagate_into_phi_args.
* tree-vrp.c (vrp_folder::get_value): Add stmt argument.
* vr-values.c (vr_values::extract_range_from_stmt): Adjust dump
output.
(vr_values::fold_cond): New.
(vr_values::simplify_cond_using_ranges_1): Call fold_cond.
* vr-values.h (class vr_values): Add
simplify_cond_using_ranges_when_edge_is_known.
gcc/testsuite/
* gcc.dg/tree-ssa/ssa-dse-30.c: Adjust test for folding of
memmove happening later.

Add missing store in emission of asan_stack_free.

gcc/ChangeLog:

2020-05-19  Martin Liska  <mliska@suse.cz>

PR sanitizer/94910
* asan.c (asan_emit_stack_protection): Emit
also **SavedFlagPtr(FakeStack, class_id) = 0 in order to release
a stack frame.

AArch64: Adjust costing of by element MUL to be the same as SAME3 MUL.

The cost model is currently treating multiplication by element as being more
expensive than 3 same multiplication.  This means that if the value is on the
SIMD side we add an unneeded DUP.  If the value is on the genreg side we use the
more expensive DUP instead of fmov.

This patch corrects the costs such that the two multiplies are costed the same
which allows us to generate

        fmul    v3.4s, v3.4s, v0.s[0]

instead of

        dup     v0.4s, v0.s[0]
        fmul    v3.4s, v3.4s, v0.4s

gcc/ChangeLog:

* config/aarch64/aarch64.c (aarch64_rtx_mult_cost): Adjust costs for mul.

gcc/testsuite/ChangeLog:

* gcc.target/aarch64/asimd-mull-elem.c: New test.

Introduce STMT_VINFO_VEC_STMTS

This gets rid of the linked list of STMT_VINFO_VECT_STMT and
STMT_VINFO_RELATED_STMT in preparation for vectorized stmts no
longer needing a stmt_vec_info (just for this chaining).  This
has ripple-down effects in all places we gather vectorized
defs.  For this new interfaces are introduced and used
throughout vectorization, simplifying code in a lot of places
and merging it with the SLP way of gathering vectorized
operands.  There is vect_get_vec_defs as the new recommended
unified interface and vect_get_vec_defs_for_operand as one
for non-SLP operation.  I've resorted to keep the structure
of the code the same where using vect_get_vec_defs would have
been too disruptive for this already large patch.

2020-06-10  Richard Biener  <rguenther@suse.de>

* tree-vect-data-refs.c (vect_vfa_access_size): Adjust.
(vect_record_grouped_load_vectors): Likewise.
* tree-vect-loop.c (vect_create_epilog_for_reduction): Likewise.
(vectorize_fold_left_reduction): Likewise.
(vect_transform_reduction): Likewise.
(vect_transform_cycle_phi): Likewise.
(vectorizable_lc_phi): Likewise.
(vectorizable_induction): Likewise.
(vectorizable_live_operation): Likewise.
(vect_transform_loop): Likewise.
* tree-vect-slp.c (vect_get_slp_defs): New function, split out
from overload.
* tree-vect-stmts.c (vect_get_vec_def_for_operand_1): Remove.
(vect_get_vec_def_for_operand): Likewise.
(vect_get_vec_def_for_stmt_copy): Likewise.
(vect_get_vec_defs_for_stmt_copy): Likewise.
(vect_get_vec_defs_for_operand): New function.
(vect_get_vec_defs): Likewise.
(vect_build_gather_load_calls): Adjust.
(vect_get_gather_scatter_ops): Likewise.
(vectorizable_bswap): Likewise.
(vectorizable_call): Likewise.
(vectorizable_simd_clone_call): Likewise.
(vect_get_loop_based_defs): Remove.
(vect_create_vectorized_demotion_stmts): Adjust.
(vectorizable_conversion): Likewise.
(vectorizable_assignment): Likewise.
(vectorizable_shift): Likewise.
(vectorizable_operation): Likewise.
(vectorizable_scan_store): Likewise.
(vectorizable_store): Likewise.
(vectorizable_load): Likewise.
(vectorizable_condition): Likewise.
(vectorizable_comparison): Likewise.
(vect_transform_stmt): Adjust and remove no longer applicable
sanity checks.
* tree-vectorizer.c (vec_info::new_stmt_vec_info): Initialize
STMT_VINFO_VEC_STMTS.
(vec_info::free_stmt_vec_info): Relase it.
* tree-vectorizer.h (_stmt_vec_info::vectorized_stmt): Remove.
(_stmt_vec_info::vec_stmts): Add.
(STMT_VINFO_VEC_STMT): Remove.
(STMT_VINFO_VEC_STMTS): New.
(vect_get_vec_def_for_operand_1): Remove.
(vect_get_vec_def_for_operand): Likewise.
(vect_get_vec_defs_for_stmt_copy): Likewise.
(vect_get_vec_def_for_stmt_copy): Likewise.
(vect_get_vec_defs): New overloads.
(vect_get_vec_defs_for_operand): New.
(vect_get_slp_defs): Declare.

tmmark: verify_ssa failed [PR95569]

This patch fixes a latent bug exposed by eb72dc663e9070b281be83a80f6f838a3a878822.
See the discussion on the bug for details.

Bootstrapped and regtested on aarch64-linux-gnu. No new fails introduced.

2020-06-10  Qian Chao  <qianchao9@huawei.com>

gcc/ChangeLog:

PR tree-optimization/95569
* trans-mem.c (expand_assign_tm): Ensure that rtmp is marked TREE_ADDRESSABLE.

gcc/testsuite/ChangeLog:

PR tree-optimization/95569
* gcc.dg/tm/pr95569.c: New test.

Add gcc_assert that &global_options are not dirty modified.

gcc/ChangeLog:

2020-03-20  Martin Liska  <mliska@suse.cz>

PR tree-optimization/92860
* optc-save-gen.awk: Generate new function cl_optimization_compare.
* opth-gen.awk: Generate declaration of the function.

gcc/c-family/ChangeLog:

2020-03-20  Martin Liska  <mliska@suse.cz>

PR tree-optimization/92860
* c-attribs.c (handle_optimize_attribute):
Save global options and compare it after parsing of function
attribute.
* c-pragma.c (opt_stack::saved_global_options): New field.
(handle_pragma_push_options): Save global_options.
(handle_pragma_pop_options): Compare them after pop.

gcc-changelog: add more Review by prefixes.

contrib/ChangeLog:

* gcc-changelog/git_commit.py: Add more supported prefixes.
* gcc-changelog/test_email.py: Fix one flake8 error.
* gcc-changelog/test_patches.txt: Add tested for newly
supported prefixes.

gcc-changelog: Improve git_commit.py diagnostics

This changes some error messages to be more self-consistent and to fix
some grammar.

contrib/ChangeLog:

* gcc-changelog/git_commit.py (GitCommit.parse_changelog):
Improve error strings.
* gcc-changelog/test_email.py: Update expected errors.

Daily bump.

libstdc++: Define converting assignment operator for std::move_iterator

As clarified by LWG 3265, std::move_iterator is supposed to have an
assignment operator that converts from a different specialization of
std::move_iterator, which performs an assignment. That has always been
missing from libstdc++, so assigning a different type actually performs
a converting construction, then an assignment. This is non-conforming
for the (fairly contrived) case where the converting assignment is
well-formed but the converting construction is not.

* include/bits/stl_iterator.h (move_iterator::operator=): Define.
* testsuite/24_iterators/move_iterator/dr3265.cc: New test.

libstdc++: Define std::bad_optional_access constructor as defaulted

The standard requires that std::bad_optional_access' default
constructor has a non-throwing exception specification.

* include/std/optional (bad_optional_access): Define default
constructor and destructor as defaulted.
* testsuite/20_util/optional/bad_access.cc: New test.

PowerPC: Add future hwcap2 bits

This patch adds support for the two new HWCAP2 fields used by the
__builtin_cpu_supports function.  It adds support in the target_clones
attribute for -mcpu=future.

The two new __builtin_cpu_supports tests are:
__builtin_cpu_supports ("isa_3_1")
__builtin_cpu_supports ("mma")

The bits used are the bits that the Linux kernel engineers will be using for
these new features.

gcc/
2020-06-09  Michael Meissner  <meissner@linux.ibm.com>

* config/rs6000/ppc-auxv.h (PPC_PLATFORM_FUTURE): Allocate
'future' PowerPC platform.
(PPC_FEATURE2_ARCH_3_1): New HWCAP2 bit for ISA 3.1.
(PPC_FEATURE2_MMA): New HWCAP2 bit for MMA.
* config/rs6000/rs6000-call.c (cpu_supports_info): Add ISA 3.1 and
MMA HWCAP2 bits.
* config/rs6000/rs6000.c (CLONE_ISA_3_1): New clone support.
(rs6000_clone_map): Add 'future' system target_clones support.

testsuite/
2020-06-09  Michael Meissner  <meissner@linux.ibm.com>

* gcc.target/powerpc/clone3.c: New test for using 'future' with
the target_clones attribute.

PowerPC: Add future hwcap2 bits

This patch adds support for the two new HWCAP2 fields used by the
__builtin_cpu_supports function.  It adds support in the target_clones
attribute for -mcpu=future.

The two new __builtin_cpu_supports tests are:
__builtin_cpu_supports ("isa_3_1")
__builtin_cpu_supports ("mma")

The bits used are the bits that the Linux kernel engineers will be using for
these new features.

testsuite/
2020-06-05  Michael Meissner  <meissner@linux.ibm.com>

* gcc.target/powerpc/clone3.c: New test for using 'future' with
the target_clones attribute.

PowerPC: Add future hwcap2 bits

This patch adds support for the two new HWCAP2 fields used by the
__builtin_cpu_supports function.  It adds support in the target_clones
attribute for -mcpu=future.

The two new __builtin_cpu_supports tests are:
__builtin_cpu_supports ("isa_3_1")
__builtin_cpu_supports ("mma")

The bits used are the bits that the Linux kernel engineers will be using for
these new features.

gcc/
2020-06-05  Michael Meissner  <meissner@linux.ibm.com>

* config/rs6000/ppc-auxv.h (PPC_PLATFORM_FUTURE): Allocate
'future' PowerPC platform.
(PPC_FEATURE2_ARCH_3_1): New HWCAP2 bit for ISA 3.1.
(PPC_FEATURE2_MMA): New HWCAP2 bit for MMA.
* config/rs6000/rs6000-call.c (cpu_supports_info): Add ISA 3.1 and
MMA HWCAP2 bits.
* config/rs6000/rs6000.c (CLONE_ISA_3_1): New clone support.
(rs6000_clone_map): Add 'future' system target_clones support.

PowerPC: Add pcrel/prefixed target supports.

2020-06-09  Michael Meissner  <meissner@linux.ibm.com>

* lib/target-supports.exp (check_effective_target_powerpc_pcrel):
New.
(check_effective_target_powerpc_prefixed_addr): New.

When specifying a non-system prefix with --with-zstd, the build fails because the header and library cannot be found (see https://gcc.gnu.org/bugzilla/show_bug.cgi?id=95005).

The attached patch fixes the problem and is what we use in Spack to
make GCC build with zstd support.

gcc/
* Makefile.in (ZSTD_INC): Define.
(ZSTD_LIB): Include ZSTD_LDFLAGS.
(CFLAGS-lto-compress.o): Add ZSTD_INC.
* configure.ac (ZSTD_CPPFLAGS, ZSTD_LDFLAGS): New variables for
AC_SUBST.
* configure: Rebuilt.

asan: fix wrong options for a test-case.

gcc/testsuite/ChangeLog:

* c-c++-common/asan/inline-kernel.c: Add missing
-fno-sanitize-address option.

coroutines: Ensure distinct DTOR trees [PR95137].

Part of the PR notes that there are UBSAN fails for the coroutines
test suite.  These are primarily related to the use of the same DTOR
tree in the two edges from the await block.  Fixed by building a new
tree for each.

gcc/cp/ChangeLog:

PR c++/95137
* coroutines.cc (expand_one_await_expression): Build separate
DTOR trees for the awaitable object on the destroy and resume
paths.

gcc-changelog: Use non-zero exit status on error

Also add comment explaining what the script does.

contrib/ChangeLog:

* gcc-changelog/git_email.py: Set exit status on error.

tree-inline: Fix VLA handling [PR95552]

The problem in this testcase comes from cloning the constructor into
complete and base variants.  When we clone the body the first time,
walk_tree_1 calls copy_tree_body_r on the type of the artificial TYPE_DECL
we made for the VLA type without calling it on the decl itself, so we
overwrite the type of the TYPE_DECL without copying the decl first.

This has been broken since we started inserting a TYPE_DECL for anonymous
VLAs in r7-457.

This patch fixes walk_tree_1 to call the function on the TYPE_DECL, as we do
for other decls of a DECL_EXPR.

gcc/ChangeLog:

PR c++/95552
* tree.c (walk_tree_1): Call func on the TYPE_DECL of a DECL_EXPR.

gcc/testsuite/ChangeLog:

PR c++/95552
* g++.dg/ext/vla23.C: New test.

tsan: Add optional support for distinguishing volatiles

Add support to optionally emit different instrumentation for accesses to
volatile variables. While the default TSAN runtime likely will never
require this feature, other runtimes for different environments that
have subtly different memory models or assumptions may require
distinguishing volatiles.

One such environment are OS kernels, where volatile is still used in
various places, and often declare volatile to be appropriate even in
multi-threaded contexts. One such example is the Linux kernel, which
implements various synchronization primitives using volatile
(READ_ONCE(), WRITE_ONCE()).

Here the Kernel Concurrency Sanitizer (KCSAN), is a runtime that uses
TSAN instrumentation but otherwise implements a very different approach
to race detection from TSAN:

https://github.com/google/ktsan/wiki/KCSAN

Due to recent changes in requirements by the Linux kernel, KCSAN
requires that the compiler supports tsan-distinguish-volatile (among
several new requirements):

https://lore.kernel.org/lkml/20200521142047.169334-7-elver@google.com/

gcc/
* params.opt: Define --param=tsan-distinguish-volatile=[0,1].
* sanitizer.def (BUILT_IN_TSAN_VOLATILE_READ1): Define new
builtin for volatile instrumentation of reads/writes.
(BUILT_IN_TSAN_VOLATILE_READ2): Likewise.
(BUILT_IN_TSAN_VOLATILE_READ4): Likewise.
(BUILT_IN_TSAN_VOLATILE_READ8): Likewise.
(BUILT_IN_TSAN_VOLATILE_READ16): Likewise.
(BUILT_IN_TSAN_VOLATILE_WRITE1): Likewise.
(BUILT_IN_TSAN_VOLATILE_WRITE2): Likewise.
(BUILT_IN_TSAN_VOLATILE_WRITE4): Likewise.
(BUILT_IN_TSAN_VOLATILE_WRITE8): Likewise.
(BUILT_IN_TSAN_VOLATILE_WRITE16): Likewise.
* tsan.c (get_memory_access_decl): Argument if access is
volatile. If param tsan-distinguish-volatile is non-zero, and
access if volatile, return volatile instrumentation decl.
(instrument_expr): Check if access is volatile.

gcc/testsuite/
* c-c++-common/tsan/volatile.c: New test.

PowerPC: PowerPC tests: Add prefixed/pcrel tests.

testsuite/
2020-06-09  Michael Meissner  <meissner@linux.ibm.com>

* lib/target-supports.exp (check_effective_target_powerpc_pcrel):
New.
(check_effective_target_powerpc_prefixed_addr): New.

Remove dead code

This removes dead code left over from the reduction vectorization
refactoring last year.

2020-06-09  Richard Biener  <rguenther@suse.de>

* tree-vect-loop.c (vectorizable_induction): Remove dead code.

d: Merge upstream dmd 13d67c575.

Reviewed-on: https://github.com/dlang/dmd/pull/11224

gcc/d/ChangeLog:

* dmd/MERGE: Merge upstream dmd 13d67c575.
* d-builtins.cc (build_frontend_type): Update call to
TypeVector::create.
* d-frontend.cc (Global::_init): Move setting of errorLimit to ...
* d-lang.cc (d_init_options): ... here.  Update for new field
location of errorLimit.
(d_post_options): Likewise.
* d-port.cc (Port::readwordLE): Update signature.
(Port::readwordBE): Likewise.
(Port::readlongLE): Likewise.
(Port::readlongBE): Likewise.
* decl.cc (get_symbol_decl): Update for new field types.

c++: Tweak predeclare_vla.

We only need to predeclare a VLA type if it's wrapped in a pointer type;
otherwise gimplify_type_sizes will handle it.

gcc/cp/ChangeLog:

PR c++/95552
* cp-gimplify.c (predeclare_vla): Only predeclare a VLA if it's
wrapped in a pointer type.

openmp: ensure variables in offload table are streamed out (PRs 94848 + 95551)

gcc/ChangeLog:

* omp-offload.c (add_decls_addresses_to_decl_constructor,
omp_finish_file): With in_lto_p, stream out all offload-table
items even if the symtab_node does not exist.

Remove dead code

This removes dead code that was left over from the reduction
vectorization refactoring last year.

2020-06-09  Richard Biener  <rguenther@suse.de>

* tree-vect-stmts.c (vect_transform_stmt): Remove dead code.