Previously I've changed gimple_split_edge to avoid PHI node
re-allocation, but this introduced swapping of PHI arguments
due to the way edge redirection works. This is now a problem
for me and which can be solved with the following approach
reducing the overhead of split_edge even more. We can simply
pretend there are no PHI nodes if we can make sure the
new fallthru will have the same dest_idx as the old edge
into the destination.
2020-10-20 Richard Biener <rguenther@suse.de>
* tree-cfg.c (reinstall_phi_args): Remove.
(gimple_split_edge): Remove PHIs around the edge redirection
to avoid touching them at all.
We were using the wrong loop to figure the latch arg of a
double-reduction PHI. Which isn't a problem in case ->dest_idx
match up with the outer loop edges - but that's of course not guaranteed.
2020-10-20 Richard Biener <rguenther@suse.de>
* tree-vect-loop.c (vectorizable_reduction): Use the correct
loops latch edge for the PHI arg lookup.
This implements support for the OMP_TARGET_OFFLOAD environment variable
introduced in the OpenMP 5.0 standard, which controls how offloading
is handled. It may be set to MANDATORY (abort if offloading cannot be
performed), DISABLED (no offloading to devices) or DEFAULT (offload to
device if possible, fall back to host if not).
2020-10-20 Kwok Cheung Yeung <kcy@codesourcery.com>
libgomp/
* env.c (gomp_target_offload_var): New.
(parse_target_offload): New.
(handle_omp_display_env): Print value of OMP_TARGET_OFFLOAD.
(initialize_env): Parse OMP_TARGET_OFFLOAD.
* libgomp.h (gomp_target_offload_t): New.
(gomp_target_offload_var): New.
* libgomp.texi (OMP_TARGET_OFFLOAD): New section.
* target.c (resolve_device): Generate error if device not found and
offloading is mandatory.
(gomp_target_fallback): Generate error if offloading is mandatory.
(GOMP_target): Add argument in call to gomp_target_fallback.
(GOMP_target_ext): Likewise.
(gomp_target_data_fallback): Generate error if offloading is mandatory.
(GOMP_target_data): Add argument in call to gomp_target_data_fallback.
(GOMP_target_data_ext): Likewise.
(gomp_target_task_fn): Add argument in call to gomp_target_fallback.
(gomp_target_init): Return early if offloading is disabled.
This moves the __noop_coro_frame type, the __noop_coro_fr global
variable, and the __dummy_resume_destroy function from namespace scope,
replacing them with private members of the specialization
coroutine_handle<noop_coroutine_promise>.
The function and variable are also declared inline, so that they
generate no code unless used.
libstdc++-v3/ChangeLog:
PR libstdc++/95917
* include/std/coroutine (__noop_coro_frame): Replace with
noop_coroutine_handle::__frame.
(__dummy_resume_destroy): Define inline in __frame.
(__noop_coro_fr): Replace with noop_coroutine_handle::_S_fr
and define as inline.
* testsuite/18_support/coroutines/95917.cc: New test.
This removes the coroutine_handle<> base class from the primary template
and the noop_coroutine_promise explicit specialization. To preserve the
API various members are added, as they are no longer inherited from the
base class.
I've also tweaked some indentation and formatting, and replaced
subclause numbers from the standard with stable names like
[coroutine.handle.con].
libstdc++-v3/ChangeLog:
* include/std/coroutine (coroutine_handle<_Promise>): Remove
base class. Add constructors, conversions, accessors etc. as
proposed for LWG 3460.
(coroutine_handle<noop_coroutine_promise>): Likewise.
* testsuite/18_support/coroutines/lwg3460.cc: New test.
This fixes an ICE caused by an unrecognizeable insn generated when
compiling gcc.c-torture/execute/pr97386-1.c at -O0.
gcc/ChangeLog:
* config/msp430/msp430.md (andneghi3): Allow general operand for
op1 and update output assembler template.
gcc/ada/
* sem_ch3.adb (Constrain_Decimal, Constrain_Enumeration,
Constrain_Float, Constrain_Integer, Constrain_Ordinary_Fixed):
Refine parameter type from Node_Id to Entity_Id.
gcc/ada/
* sinfo.ads, sinfo.adb: The flag Box_Present can appear in
Iterated_Element_Association nodes.
* sem_aggr.adb (Resolve_Aggregate): Call
Resolve_Container_Aggregate when type of context has
corresponding aspect.
* sem_type.adb (Covers): In Ada_2020 an aggregate is compatible
with a type that carries the corresponding aspect.
* exp_ch3.adb (Make_Controlling_Function_Wrappers): Do not
create declarations and bodies for inherited primitive functions
of null extensions that dispatch on result, when current scope
includes an immediately visible non-overloadable homonym of the
function.
* libgnat/a-cborse.adb, libgnat/a-cborse.ads,
libgnat/a-cbhase.ads, libgnat/a-cbhase.adb,
libgnat/a-cborma.adb, libgnat/a-cborma.ads,
libgnat/a-cbhama.adb, libgnat/a-cbhama.ads,
libgnat/a-cbdlli.adb, libgnat/a-cbdlli.ads,
libgnat/a-convec.ads, libgnat/a-ciorse.ads,
libgnat/a-cihase.ads, libgnat/a-cihase.adb,
libgnat/a-ciorma.ads, libgnat/a-cihama.ads,
libgnat/a-cihama.adb, libgnat/a-cidlli.ads,
libgnat/a-cidlli.adb, libgnat/a-coinve.adb,
libgnat/a-cobove.adb, libgnat/a-cobove.ads,
libgnat/a-convec.adb, libgnat/a-coinve.ads,
libgnat/a-coorse.ads, libgnat/a-cohase.adb,
libgnat/a-cohase.ads, libgnat/a-coorma.ads,
libgnat/a-cohama.adb, libgnat/a-cohama.ads,
libgnat/a-cdlili.ads: Add primitive function Empty for use in
aspect Aggregate, and add corresponding body or expression
function.
gcc/ada/
* aspects.adb (Has_Aspect_Specifications_Flag): Add
N_Parameter_Specification.
* par-ch13.adb (Aspect_Specifications_Present): Also handle case
of an unknown aspect on the last formal parameter (terminated by
a Tok_Right_Paren). Minor reformatting.
* par-ch6.adb (P_Formal_Part): Scan aspects on formal
parameters.
* par.adb: Fix typos.
* sem_ch6.adb (Process_Formals): Add processing of aspects and
in particular Unreferenced aspect for now.
* sinfo.ads: Allow ASPECT_SPECIFICATIONS on a
PARAMETER_SPECIFICATION.
* doc/gnat_rm/implementation_defined_aspects.rst
(Aspect Unreferenced): Update documentation.
* gnat_rm.texi: Regenerate.
gcc/ada/
* sem_util.ads: Declare a new package, Old_Attr_Util, which in
turn declares two more packages, Conditional_Evaluation and
Indirect_Temps. Conditional_Evaluation provides a predicate for
deciding whether a given 'Old attribute reference is eligible
for conditional evaluation and, in the case where it is
eligible, a function that constructs the Boolean-valued
condition that is to be evaluated at run time in deciding
whether to evaluate the attribute prefix. Indirect_Temps
provides support for declaring a temporary which is only
initialized conditionally; more specifically, an access type and
a variable of that type are declared (unconditionally) and then
the variable is (conditionally) initialized with an allocator.
The existence of the access type and the pointer variable is
hidden from clients, except that a predicate,
Is_Access_Type_For_Indirect_Temp, is provided for identifying
such access types. This is needed because we want such an access
type to be treated like a "normal" access type (specifically
with respect to finalization of allocated objects). Other parts
of the compiler treat access types differently if
Comes_From_Source is False, or if the secondary stack storage
pool is used; this predicate is used to disable this special
treatment.
* sem_attr.adb (Uneval_Old_Msg): Improve message text to reflect
Ada202x changes.
(Analyze_Attribute): A previously-illegal 'Old attribute
reference is accepted in Ada2020 if it is eligible for
conditional evaluation.
* sem_res.adb (Valid_Conversion): Do not treat a rewritten 'Old
attribute like other rewrite substitutions. This makes a
difference, for example, in the case where we are generating the
expansion of a membership test of the form "Saooaaat'Old in
Named_Access_Type"; in this case Valid_Conversion needs to
return True (otherwise the expansion will be False - see the
call site in exp_ch4.adb).
* exp_attr.adb (Expand_N_Attribute_Reference): When expanding a
'Old attribute reference, test for the case where the reference
is eligible for conditional evaluation. In that case, use the
new "indirect temporary" mechanism provided by Sem_Util.
* exp_prag.adb
(Expand_Attributes_In_Consequence.Expand_Attributes): If
Sem_Util.Indirect_Temp_Needed indicates that there could be
correctness problems associated with the old expansion scheme
for dealing with 'Old attributes in contract cases consequences,
then we use the new "indirect temporary" mechanism provided by
Sem_Util instead. We do not want to do this unconditionally.
* sem_util.adb: Provide a body for the new Old_Attr_Util
package. Further work is needed in several areas for
correctness:
- The function Is_Repeatedly_Evaluated does not deal with
container aggregates yet.
- The function Is_Known_On_Entry does not deal with interactions
with the Global aspect.
Each area where more work is needed is indicated with a "???"
comment in the code; a more detailed description can be found
there. Some optimization opportunties are similarly indicated
with a "???" comment.
* exp_ch3.adb (Freeze_Type): In deciding whether to generate
expansion for the list controller of an access type, take the
predicate Is_Access_Type_For_Indirect_Temp into account. If the
predicate yields True, then generate the expansion.
* exp_util.adb (Build_Allocate_Deallocate_Proc): We don't
normally finalize allocated objects that are allocated on the
secondary stack. Add an exception to this rule if the predicate
Is_Access_Type_For_Indirect_Temp yields True. As a result of
this exception, we have to deal with the fact that gigi expects
a different parameter profile if we are using the secondary
stack pool; the Pool and Alignment parameters must be omitted in
this case.
gcc/ada/
* sem_ch12.adb (Install_Parents_Of_Generic_Context,
Remove_Parents_Of_Generic_Context): New subprograms.
(Instantiate_Package_Body): Adding assertions to ensure that
installed parents are properly removed.
gcc/ada/
* gnat_cuda.adb (Build_Register_Function_Call): Make procedure
call instead of function, rename to
Build_Register_Procedure_Call.
(Build_CUDA_Init_Proc): Make procedure call instead of function.
gcc/ada/
* errout.adb (Write_Source_Code_Line): Adopt display closer to
GCC format.
(Output_Messages): Deal specially with info messages.
* erroutc.adb (Prescan_Message): Fix bug leading to check
messages being considered as error messages in pretty output
mode.
gcc/ada/
* exp_ch6.adb (Expand_Call_Helper): Properly handle the case
where the condition of a conditional expression has been
optimized out when calculating the value of an extra
accessibility parameter.
gcc/ada/
* sem_aggr.adb (Resolve_Aggregate): Warn on not fully
initialized box aggregate.
* sem_aggr.ads: Fix typo.
* sem_res.adb (Resolve_Actuals): Fix typo in error message
format marking it incorrectly as a continuation message.
* sem_elab.adb (Check_Internal_Call_Continue): Similarly, add
missing primary message in case of a call to an actual generic
subprogram.
* sem_warn.adb (Check_References): Do not warn on read but never
assigned variables if the type is partially initialized.
* libgnat/a-except.ads, libgnat/a-ststun.ads,
libgnat/g-sechas.ads, libgnat/a-cbdlli.ads,
libgnat/a-cfdlli.ads, libgnat/a-cobove.ads,
libgnat/a-cohata.ads, libgnat/a-crbltr.ads,
libgnat/a-cbmutr.ads, libgnat/a-crdlli.ads,
libgnat/a-cbsyqu.ads: Address new warning.
* doc/gnat_ugn/building_executable_programs_with_gnat.rst:
Update doc on -gnatwv.
* gnat_ugn.texi: Regenerate.
gcc/testsuite/
* gnat.dg/opt11.adb: Add new expected warning.
gcc/ada/
* sem_ch8.adb (Check_Constrained_Object): Suppress discriminant
checks when the type has default discriminants and comes from
expansion of a "for of" loop.
gcc/ada/
* atree.ads: Make Default_Node a constant. Remove the
modification of Comes_From_Source, and use a separate flag for
that. Change Sloc to 0; it always overwritten, and never left
as the No_Location value.
(Print_Statistics): Move to spec so we can call it from
gnat1drv.
(Num_Nodes): Rename to clarify that this is approximate.
Correct comment: nodes and entities are never deleted, the count
is never decremented, and this is not used by Xref.
(Initialize): Correct comment: Error_List is not created here.
Other minor naming and comment changes.
* atree.adb (Extend_Node, New_Copy, New_Entity, New_Node):
Streamline these. Simplify and improve efficiency. Move code
from Allocate_Initialize_Node to these, where it can be executed
unconditionally. Take advantage of automatic zeroing of the
Nodes table.
(Allocate_Initialize_Node): Remove this. It was an efficiency
bottleneck, and somewhat complicated, because it was called from
4 places, and had all sorts of conditionals to check where it
was called from. Better to move most of that code to the call
sites, where it can be executed (or not) unconditionally.
(Allocate_New_Node): New procedure to partly replace
Allocate_Initialize_Node (called from just 2 of those 4 places).
(Comes_From_Source_Default): New flag written/read by
Set_Comes_From_Source_Default/Get_Comes_From_Source_Default.
This allows us to make Default_Node into a constant with
all-zeros value.
(Set_Paren_Count_Of_Copy): New procedure to avoid duplicated
code.
(Report): New procedure to encapsulate the call to the reporting
procedure.
(Atree_Private_Part): We now need a body for this package, to
contain package body Nodes.
(Approx_Num_Nodes_And_Entities): Was Num_Nodes. For efficiency,
compute the answer from Nodes.Last. That way we don't need to
increment a counter on every node creation. Other minor naming
and comment changes.
* gnat1drv.adb: Call Atree.Print_Statistics if -gnatd.A switch
was given. Add comment documenting the new order dependency (we
must process the command line before calling Atree.Initialize).
* debug.adb: Document -gnatd.A.
* einfo.adb, sinfo.adb: Remove useless Style_Checks pragmas.
* nlists.ads (Allocate_List_Tables): Inline makes node creation
a little faster.
* nlists.adb (Initialize): Remove local constant E, which didn't
seem to add clarity.
* treepr.adb (Print_Init): Use renamed
Approx_Num_Nodes_And_Entities function.
* types.ads: Change the Low and High bounds as described above.
* types.h: Change Low and High bounds to match types.ads.
* sem_ch8.adb, namet.adb, namet.ads: Move the computation of
Last_Name_Id from sem_ch8 to namet, and correct it to not assume
Name_Ids are positive.
* ali.adb, ali-util.adb, bindo-writers.adb, exp_dist.adb,
fmap.adb, fname-uf.adb, osint.adb: Fix various hash functions to
avoid assuming the various ranges are positive. Note that "mod"
returns a nonnegative result when the second operand is
positive. "rem" can return negative values in that case (in
particular, if the first operand is negative, which it now is).
* switch-c.adb: Allow switch -gnaten to control the value of
Nodes_Size_In_Meg.
* doc/gnat_ugn/building_executable_programs_with_gnat.rst:
Remove traling whitespaces.
* opt.ads (Nodes_Size_In_Meg): New Variable.
gcc/ada/
* exp_util.adb (Remove_Side_Effects): Always generate a renaming
that is handled by the front-end in the case of an indexed or a
selected component whose prefix has a nonstandard representation.
gcc/ada/
* errout.adb: (Error_Msg-Internal): Pass the location for a line
insertion if any in the message.
(Output_Messages: Add display of source code lines if -gnatdF is
set.
(Write_Source_Code_Line): Code clean up.
* erroutc.adb (Prescan_Message): Apply prescan for continuation
lines when -gnatdF is set, and record presence of line
insertion.
* erroutc.ads (Has_Insertion_Line): New global for prescan.
(Error_Msg_Object): Add field to record line insertion if
present.
* errutil.adb (Error_Msg): Pass no location for Insertion_Sloc.
gcc/ada/
* exp_aggr.adb (Expand_Iterated_Component): Reorganize code to
ensure that Loop_Id is properly initialized on all paths, and
remove code duplication.
gcc/ada/
* aspects.ads: Introduce Subprogram_Variant aspect with the
following properties: GNAT-specific, with mandatory expression,
not a representation aspect, never delayed.
* contracts.adb (Expand_Subprogram_Contract): Mention new aspect
in the comment.
(Add_Contract_Item): Support addition of pragma
Subprogram_Variant to N_Contract node.
(Analyze_Entry_Or_Subprogram_Contract): Mention new aspect in
the comment; add pragma Subprogram_Variant to N_Contract node.
(Build_Postconditions_Procedure): Adapt call to
Insert_Before_First_Source_Declaration, which is now reused in
expansion of new aspect.
(Process_Contract_Cases_For): Also process Subprogram_Variant,
which is stored in N_Contract node together with Contract_Cases.
* contracts.ads (Analyze_Entry_Or_Subprogram_Contract): Mention
new aspect in the comment.
(Analyze_Entry_Or_Subprogram_Body_Contract): Likewise.
* einfo.adb (Get_Pragma): Support retrieval of new pragma.
* einfo.ads (Get_Pragma): Likewise.
* exp_ch6.adb (Check_Subprogram_Variant): New routine for
emitting call to check Subprogram_Variant expressions at run
time.
(Expand_Call_Helper): Check Subprogram_Variant expressions at
recursive calls.
* exp_prag.adb (Make_Op): Moved from expansion of pragma
Loop_Variant to Exp_Util, so it is now reused for expansion of
pragma Subprogram_Variant.
(Process_Variant): Adapt call to Make_Op after moving it to
Exp_Util.
(Expand_Pragma_Subprogram_Variant): New routine.
* exp_prag.ads (Expand_Pragma_Subprogram_Variant): Likewise.
* exp_util.adb (Make_Variant_Comparison): Moved from Exp_Prag
(see above).
* exp_util.ads (Make_Variant_Comparison): Likewise.
* inline.adb (Remove_Aspects_And_Pragmas): Handle aspect/pragma
Subprogram_Variant just like similar contracts.
* par-prag.adb (Prag): Likewise.
* sem.adb (Insert_Before_First_Source_Declaration): Moved from
Contracts (see above).
* sem.ads (Insert_Before_First_Source_Declaration): Likewise.
* sem_ch12.adb: Mention new aspect in the comment about
"Implementation of Generic Contracts", just like similar aspects
are mentioned there.
* sem_ch13.adb (Insert_Pragma): Mention new aspect in the
comment, because this routine is now used for Subprogram_Variant
just like for other similar aspects.
(Analyze_Aspect_Specifications): Mention new aspect in comments;
it is handled just like aspect Contract_Cases.
(Check_Aspect_At_Freeze_Point): Do not expect aspect
Subprogram_Variant just like we don't expect aspect
Contract_Cases.
* sem_prag.adb (Ensure_Aggregate_Form): Now also used for pragma
Subprogram_Variant, so update comment.
(Analyze_Pragma): Add initial checks for pragma
Subprogram_Variant.
(Analyze_Subprogram_Variant_In_Decl_Part): New routine with
secondary checks on the new pragma.
(Sig_Flags): Handle references within pragma Subprogram_Variant
expression just like references in similar pragma
Contract_Cases.
(Is_Valid_Assertion_Kind): Handle Subprogram_Variant just like
other similar contracts.
* sem_prag.ads (Analyze_Subprogram_Variant_In_Decl_Part): New
routine.
* sem_res.adb (Same_Or_Aliased_Subprograms): Moved to Sem_Util,
so it can be reused for detection of recursive calls where
Subprogram_Variant needs to be verified.
* sem_util.adb (Is_Subprogram_Contract_Annotation): Handle new
Subprogram_Variant annotation just like other similar
annotations.
(Same_Or_Aliased_Subprograms): Moved from Sem_Res (see above).
* sem_util.ads (Is_Subprogram_Contract_Annotation): Mention new
aspect in the comment.
(Same_Or_Aliased_Subprograms): Moved from Sem_Res (see above).
* sinfo.ads (N_Contract): Document handling of
Subprogram_Variant.
* snames.ads-tmpl: Add name for the internally generated
procedure with checks for Subprogram_Variant expression, name
for the new aspect and new pragma corresponding to aspect
Subprogram_Variant.
I forgot to guard the promotion to external for the case where the
def is in a pattern.
2020-10-20 Richard Biener <rguenther@suse.de>
PR tree-optimization/97496
* tree-vect-slp.c (vect_get_and_check_slp_defs): Guard extern
promotion with not in pattern.
* gcc.dg/vect/bb-slp-pr97496.c: New testcase.
In case the vectorized version of strlen is used, then each memory
access inside the loop is 16-byte aligned. Thus add this kind of
information so that vector alignment hints can later on be emitted.
gcc/ChangeLog:
* config/s390/s390.c (s390_expand_vec_strlen): Add alignment
for memory access inside loop.
Sanity check ranges by comparing just SIGN and PRECISION.
gcc/
PR tree-optimization/97360
* gimple-range.h (range_compatible_p): New.
* gimple-range-gori.cc (is_gimple_logical_p): Use range_compatible_p.
(range_is_either_true_or_false): Ditto.
(gori_compute::outgoing_edge_range_p): Cast result to the correct
type if necessary.
(logical_stmt_cache::cacheable_p): Use range_compatible_p.
* gimple-range.cc (gimple_ranger::calc_stmt): Check range_compatible_p
before casting the range.
(gimple_ranger::range_on_exit): Use range_compatible_p.
(gimple_ranger::range_on_edge): Ditto.
gcc/testsuite/
* gcc.dg/pr97360-2.c: New test.
This is the std::unique_ptr part of P1020R1 (as amended by P1973R1) for
C++20. The std::shared_ptr part still needs to be done.
libstdc++-v3/ChangeLog:
* include/bits/unique_ptr.h (make_unique_for_overwrite): Define
for C++20.
* testsuite/20_util/unique_ptr/creation/array_neg.cc: Remove
unused header. Adjust standard reference.
* testsuite/20_util/unique_ptr/creation/for_overwrite.cc: New test.
* testsuite/20_util/unique_ptr/creation/for_overwrite__neg.cc: New test.
At one stage, use cases were proposed for allowing the promise
type to contain both return_value and return_void. That was
not accepted into C++20, so we should reject it as per the PR.
gcc/cp/ChangeLog:
PR c++/97438
* coroutines.cc (struct coroutine_info): Add a field to
record that we emitted a promise type error.
(coro_promise_type_found_p): Check for the case that the
promise type contains both return_void and return_value.
Emit an error if so, with information about the wrong
type methods.
gcc/testsuite/ChangeLog:
PR c++/97438
* g++.dg/coroutines/pr97438.C: New test.
In PR 97456, IPA-SRA triggers a bug in tree-complex.c where it
converts:
<bb 2>
a$_M_value_21 = COMPLEX_EXPR <ISRA.18_10(D), ISRA.18_10(D)>;
(where ISRA.18 is IPA-SRA created PARM_DECL with DECL_IGNORED_P set,
which is why it only happens with IPA-SRA) into:
<bb 2>
a$_M_value_21 = COMPLEX_EXPR <a$_M_value$real_10(D), a$_M_value$real_10(D)>;
i.e. it replaces two uses of the parameter default-def with two
uninitialized default-defs of a new variable - all in hope to produce
code with better debug info.
This patch fixes it by avoiding the transform when the SSA_NAME to be
replaced is a default definition.
gcc/ChangeLog:
2020-10-19 Martin Jambor <mjambor@suse.cz>
PR tree-optimization/97456
* tree-complex.c (set_component_ssa_name): Do not replace ignored decl
default definitions with new component vars. Reorder if conditions.
gcc/testsuite/ChangeLog:
2020-10-19 Martin Jambor <mjambor@suse.cz>
PR tree-optimization/97456
* gcc.dg/tree-ssa/pr97456.c: New test.
Although the compiler supports the [[no_unique_address]] attribute, it's
not a reserved name prior to C++20, so we can't use it in std::tuple.
Use [[__no_unique_address__]] instead.
libstdc++-v3/ChangeLog:
* include/std/tuple (_Head_base<Idx, Head, true>): Use reserved
form of __no_unique_address__ attribute because
no_unique_address is not reserved prior to C++20.
