Similar to 37e65643d3 ("testsuite/101269: fix testcase when used with
-m32"), RISC-V needs to be told not to put symbols in the
sdata/srodata/sbss sections.
gcc/testsuite/ChangeLog
* gcc.dg/debug/btf/btf-datasec-1.c: Don't use small data on RISC-V.
Re-using some common things like EQ_EXPR and other relationals made
certain things easier, but complicated debugging and added extra overhead
when accessing lookup tables. With forthcoming additional relation types,
it makes more sense to simple have a distinct relation kind.
* gimple-range-fold.cc (fold_using_range::range_of_phi): Use new VREL_*
enumerated values.
* gimple-range-path.cc (maybe_register_phi_relation): Ditto.
* range-op.cc (*::lhs_op1_relation): Return relation_kind, and use
new VREL enumerated values.
(*::lhs_op2_relation): Ditto.
(*::op1_op2_relation): Ditto.
(*::fold_range): Use new VREL enumerated values.
(minus_op1_op2_relation_effect): Ditto.
(range_relational_tests): Ditto.
* range-op.h (fold_range, op1_range, op2_range): Use VREL_VARYING.
(lhs_op1_relation, lhs_op2_relation, op1_op2_relation): Return
relation_kind.
(*_op1_op2_relation): Return relation_kind.
(relop_early_resolve): Use VREL_UNDEFINED.
* value-query.cc (range_query::query_relation): Use VREL_VARYING.
* value-relation.cc (VREL_LAST): Change enumerated value.
(vrel_range_assert): Delete.
(print_relation): Remove range assert.
(rr_negate_table): Adjust table to use new enumerated values..
(relation_negate): Remove range assert.
(rr_swap_table): Adjust.
(relation_swap): Remove range assert.
(rr_intersect_table): Adjust.
(relation_intersect): Remove range assert.
(rr_union_table): Adjust.
(relation_union): Remove range assert.
(rr_transitive_table): Adjust.
(relation_transitive): Remove range assert.
(equiv_oracle::query_relation): Use new VREL enumerated values.
(equiv_oracle::register_relation): Ditto.
(relation_oracle::register_stmt): Ditto.
(dom_oracle::set_one_relation): Ditto.
(dom_oracle::register_transitives): Ditto.
(dom_oracle::query_relation): Ditto.
(path_oracle::register_relation): Ditto.
(path_oracle::query_relation): Ditto.
* value-relation.h (enum relation_kind_t): New relation_kind.
(*_op1_op2_relation): Adjust prototypes.
Union_ returns a boolean indicating if the operation changes the range.
Also optimize the common single-pair UNION single-pair case.
* gimple-range-edge.cc (calc_switch_ranges): Check union return value.
* value-range.cc (irange::legacy_verbose_union_): Add return value.
(irange::irange_single_pair_union): New.
(irange::irange_union): Add return value.
* value-range.h (class irange): Adjust prototypes.
Return true if the intersection of ranges changed the original value.
Speed up the case when there is no change by calling an efficient
contains routine.
* value-range.cc (irange::legacy_verbose_intersect): Add return value.
(irange::irange_contains_p): New.
(irange::irange_intersect): Add return value.
* value-range.h (class irange): Adjust prototypes.
The "is_current" status is returned by parameter, but was being returned by the
function as well instead of true if NAME had a global range, and FALSE
if it did not.
* gimple-range-cache.cc (ranger_cache::get_global_range): Return the
had_global value instead.
We use the relation between op1 and op2 to help fold a statement, but
it was not provided to the lhs_op1_relation and lhs_op2_relation routines
to determine if is also creates a relation between the LHS and either operand.
gcc/
PR tree-optimization/104547
* gimple-range-fold.cc (fold_using_range::range_of_range_op): Add
the op1/op2 relation to the relation call.
* range-op.cc (*::lhs_op1_relation): Add param.
(*::lhs_op2_relation): Ditto.
(operator_minus::lhs_op1_relation): New.
(range_relational_tests): Add relation param.
* range-op.h (lhs_op1_relation, lhs_op2_relation): Adjust prototype.
gcc/testsuite/
* g++.dg/pr104547.C: New.
Internal-linkage entity mangling is entirely implementation defined --
there's no ABI issue. Let's not mangle in any module attachment to
them, it makes the symbols unnecessarily longer.
gcc/cp/
* mangle.cc (maybe_write_module): Check external linkage.
gcc/testsuite/
* g++.dg/modules/mod-sym-4.C: New.
VRP currently searches the ssa_name list for globals to exported after it
finishes running. Recent changes have VRP calling a side-effect routine for
each stmt during the walk. This change simply exports globals as they are
calculated the final time during the walk.
* gimple-range.cc (gimple_ranger::register_side_effects): First check
if the DEF should be exported as a global.
* tree-vrp.cc (rvrp_folder::pre_fold_bb): Process PHI side effects,
which will export globals.
(execute_ranger_vrp): Remove call to export_global_ranges.
Add modes to range_from_dom such that we can simply query, or adjust the
cache and deal with multiple predecessor blocks.
* gimple-range-cache.cc (ranger_cache::ranger_cache): Start with
worlist truncated.
(ranger_cache::entry_range): Add rfd_mode parameter.
(ranger_cache::exit_range): Ditto.
(ranger_cache::edge_range): New. Incorporate from range_on_edge.
(ranger_cache::range_of_expr): Adjust call to entry_range.
(ranger_cache::range_on_edge): Split to edge_range and call.
(ranger_cache::fill_block_cache): Always invoke range_from_dom.
(ranger_cache::range_from_dom): Make reentrant, add search mode, handle
mutiple predecessors.
(ranger_cache::update_to_nonnull): Adjust call to exit_range.
* gimple-range-cache.h (ranger_cache): Add enum rfd_mode. Adjust
prototypes.
libstdc++-v3/ChangeLog:
* include/bits/ostream_insert.h: Mark helper functions as
undocumented by Doxygen.
* include/bits/stl_algo.h: Use markdown for formatting and mark
helper functions as undocumented.
* include/bits/stl_numeric.h: Likewise.
* include/bits/stl_pair.h (pair): Add @headerfile.
Add @headerfile and @since tags. Improve grouping of non-member
functions via @relates tags.
Mark the std::pair base class of std::sub_match as undocumented, so that
the docs don't show all the related non-member functions are part of the
sub_match API. Use a new macro to re-add the data members for Doxygen
only.
libstdc++-v3/ChangeLog:
* doc/doxygen/user.cfg.in (PREDEFINED): Define macro
_GLIBCXX_DOXYGEN_ONLY to expand its argument.
* include/bits/c++config (_GLIBCXX_DOXYGEN_ONLY): Define.
* include/bits/regex.h: Improve doxygen docs.
* include/bits/regex_constants.h: Likewise.
* include/bits/regex_error.h: Likewise.
libstdc++-v3/ChangeLog:
* include/std/atomic: Suppress doxygen docs for
implementation details.
* include/bits/atomic_base.h: Likewise.
* include/bits/shared_ptr_atomic.h: Use markdown. Fix grouping
so that std::atomic is not added to the pointer abstractions
group.
Use macros to open and close the inline namespace _V2 that is used for
ABI versioning of individual components such as chrono::system_clock.
This allows the namespace to be hidden in the docs generated by Doxygen,
so that we document std::foo instead of std::_V2::foo.
This also makes it easy to remove that namespace entirely for the
gnu-versioned-namespace build, where everything is already versioned as
std::__8 and there are no backwards compatibility guarantees.
libstdc++-v3/ChangeLog:
* doc/doxygen/user.cfg.in (PREDEFINED): Expand new macros to
nothing.
* include/bits/c++config (_GLIBCXX_BEGIN_INLINE_ABI_NAMESPACE)
(_GLIBCXX_END_INLINE_ABI_NAMESPACE): Define new macros.
* include/bits/algorithmfwd.h (_V2::__rotate): Use new macros
for the namespace.
* include/bits/chrono.h (chrono::_V2::system_clock): Likewise.
* include/bits/stl_algo.h (_V2::__rotate): Likewise.
* include/std/condition_variable (_V2::condition_variable_any):
Likewise.
* include/std/system_error (_V2::error_category): Likewise.
Before Doxygen version 1.9.2 this option is broken (see
https://github.com/doxygen/doxygen/issues/8638 for more details) and
classes are not added to the correct groups by @ingroup and @addtogroup.
Also remove the obsolete CLASS_DIAGRAMS option that causes a warning.
libstdc++-v3/ChangeLog:
* doc/doxygen/user.cfg.in (GROUP_NESTED_COMPOUNDS): Set to NO.
(CLASS_DIAGRAMS): Remove obsolete option.
Reverse iteration over blocks, in gimple-harden-conditionals.cc, was
supposed to avoid visiting blocks introduced by hardening and
introducing further reversed conditionals and traps for them, but
newly-created blocks may be inserted before the current block, as
shown by the PR105455 testcase.
Use an auto_sbitmap to gather preexisting blocks that need visiting.
for gcc/ChangeLog
* gimple-harden-conditionals.cc: Include sbitmap.h.
(pass_harden_conditional_branches::execute): Skip new blocks.
(pass_harden_compares::execute): Likewise.
The compiler is allowed to assume it can access String bounds, such as
the prefix passed to Get_External_Name, even in circumstances in which
the prefix is not going to be used and has_prefix is false, so, from
the C side, we have to build a proper String_Template for the
String_Pointer.
for gcc/ada/ChangeLog
* gcc-interface/decl.cc (is_cplusplus_method): Build proper
String for Get_External_Name.
Test for the validity checking performed on nonstandard booleans
annotated with the "hardbool" Machine_Attribute pragma.
for gcc/testsuite/ChangeLog
* gnat.dg/hardbool.ads: New.
* gnat.dg/hardbool.adb: New.
Vector operations in MVE must be aligned to the element size, so if we
are asked for a misaligned move in a wider mode we must recast it to a
form suitable for the known alignment (larger elements have better
address offset ranges, so there is some advantage to using wider
element sizes if possible). Whilst fixing this, also rework the
predicates used for validating operands - the Neon predicates are
not right for MVE.
gcc/ChangeLog:
PR target/105463
* config/arm/mve.md (*movmisalign<mode>_mve_store): Use
mve_memory_operand.
(*movmisalign<mode>_mve_load): Likewise.
* config/arm/vec-common.md (movmisalign<mode>): Convert to generator
form...
(@movmisalign<mode>): ... thus. Use generic predicates and then
rework operands if they are not valid. For MVE rework to a
narrower element size if the alignment is not high enough.
There are a couple of issues with the mve_vector_mem_operand function.
Firstly, SP is permitted as a register provided there is no write-back
operation. Secondly, there were some cases where 'strict' was not
being applied when checking which registers had been used.
gcc/ChangeLog:
* config/arm/arm.cc (mve_vector_mem_operand): Allow SP_REGNUM
when there is no write-back. Fix use when strict is true.
These patterns were deprecated since GCC 4.8.
gcc/ChangeLog:
* config/xtensa/xtensa.md (extvsi, extvsi_internal, extzvsi,
extzvsi_internal): Rename from extv, extv_internal, extzv and
extzv_internal, respectively.
Most cases of VIEW_CONVERT_EXPRs involving reverse scalar storage order are
disqualified for SRA because they are storage_order_barrier_p, but you can
still have a VIEW_CONVERT_EXPR to a regular composite type being applied to
a component of a record type with reverse scalar storage order.
In this case the bypass for !useless_type_conversion_p in sra_modify_assign,
albeit already heavily guarded, triggers and may generate wrong code, so the
patch makes sure that it does only when the SSO is the same on both side.
gcc/
* tree-sra.cc (sra_modify_assign): Check that scalar storage order
is the same on the LHS and RHS before rewriting one with the model
of the other.
gcc/testsuite/
* gnat.dg/sso17.adb: New test.
This is a return convention mismatch coming from a discrepancy of the
Returns_By_Ref flag for the inherited function.
gcc/ada/
* sem_ch3.adb (Derive_Subprogram): For a function, also copy the
Returns_By_Ref flag from the parent.
After a compilation unit containing a Designated_Storage_Model aspect
was processed by the compiler front end, the aspect's expression was
seen as not having been analyzed. In some cases this can lead to the
Storage_Model_Object function not returning the entity of the associated
model object (such as when the model object is specified using an
expanded name). This was due to a second copy of the aspect's expression
being made in Check_Aspect_At_Freeze_Point, and that copy wasn't
analyzed, even though the earlier version of the expression prior to
that copy had been. (The copy is needed for proper checking of Ada's
rules for ensuring that certain aspect expressions resolve the same way
at the freeze point and later at the end of the declarations.) This is
corrected by doing analysis of the new aspect expression within
Check_Aspect_At_Freeze_Point, as is already done in that procedure in
the case of several other aspects.
gcc/ada/
* sem_ch13.adb (Check_Aspect_At_Freeze_Point): Analyze the
expression of the aspect at this point, like other aspects that
denote simple entities. Ensures that function
Storage_Model_Object will have an entity to retrieve from the
aspect expression. Also, move comment about aspects that don't
require delay analysis in front of last alternative of the case
statement, where it's supposed to be.
gcc/ada/
* contracts.adb (Build_Unique_Name): New subprogram.
(Make_Class_Precondition_Subps): Use Build_Unique_Name to
generate the names of the call helpers and the name of indirect
call wrappers.
* freeze.adb (Needs_Wrapper): Remove dead code.
(Check_Inherited_Conditions): Defer building helpers and ICW
until all the dispatch table wrappers have been built and
analyzed. Required to ensure uniqueness in their names because
when building these wrappers for overlapped subprograms their
homonym number is not definite until they have been analyzed.
If we're not going to install a signal handler for SIGSEGV then we do
not need an alternate stack to handle this signal (and e.g. stack
overflows).
gcc/ada/
* libgnarl/s-taprop__linux.adb (Initialize): Do not use an
alternate stack if no handler for SEGV is installed.
Having realized that noreturn calls of __builtin_return are special,
and other noreturn calls don't get edges to the exit block, I've
realized the consequences of the logic to insert checking before
noreturn and tail calls were not quite what I'd expected before.
Specifically, noreturn calls other than __builtin_return don't get any
checking whatsoever. Moreover, optional tail calls are only detected
long after hardcfr; the logic should work for must-tail calls, but
there doesn't seem to be a way to test it.
Documentation has been simplified so as to remove mention of these
possibilities, that can't really be relied on.
gcc/ada/
* doc/gnat_rm/security_hardening_features.rst (Control Flow
Redundancy): Drop mentions of noreturn and tail calls.
* gnat_rm.texi: Regenerate.
Compiling CUDA code requires compiling code for the host (= CPU) and for
the device (= GPU). Device code is embedded into the host code and must
be registered with the CUDA runtime by the host.
The original approach we took for registering CUDA kernels was to
generate the registration-code on a unit basis, i.e. each unit took care
of registering its own kernels. Unfortunately, this makes linking
kernels and device functions that belong to different units much harder.
We thus rework this approach in order to have GNAT generate kernel names
in ALI files. The binder reads the ALI files and generates kernel
registration code for each of the kernels found in ALI files.
gcc/ada/
* ali.adb: Introduce new 'K' line in ALI files, used to
represent CUDA kernel entries.
* ali.ads: Create new CUDA_Kernels table, which contains entries
of type CUDA_Kernel_Record. Each CUDA_Kernel_Record corresponds
to a K line in an ali file.
* bindgen.adb: Introduce new Gen_CUDA_Init procedure in the
binder, which generates CUDA kernel registration code.
* gnat_cuda.adb: Move Get_CUDA_Kernels spec to package spec to
make it available to bindgen.adb.
* gnat_cuda.ads: Likewise.
* lib-writ.adb: Introduce new Output_CUDA_Symbols procedure,
which generates one 'K' line in the ALI file per visible CUDA
kernel.
* opt.ads: Introduce Enable_CUDA_Expansion option, triggered by
using the -gnatd_c flag.
* switch-b.adb: Likewise.
* switch-c.adb: Likewise.
Create a scope for the Ada 2022 iterated component association loops. In
the case of elements needing finalization, the late expansion would
crash on references to the loop variable within the loop body.
gcc/ada/
* exp_aggr.adb (Gen_Loop): Create scope for loop variable of
iterated components.
As tampering checks do not exist in formal hashed sets and maps, remove
the machinery for such checks in the version of generic key and node
operations for formal sets/maps. Update comments as well.
gcc/ada/
* libgnat/a-chtgfk.adb (Checked_Equivalent_Keys, Checked_Index):
Remove useless functions.
(Delete_Key_Sans_Free, Find, Generic_Conditional_Insert): Adapt
to removal of wrapper functions.
* libgnat/a-chtgfk.ads (Checked_Equivalent_Keys, Checked_Index):
Remove useless functions.
* libgnat/a-chtgfo.adb (Checked_Index): Remove useless function.
(Clear): Delete code commented out regarding Busy and Lock
management.
(Delete_Node_At_Index): Delete unused procedure.
(Delete_Node_Sans_Free, Free, Generic_Read, Next): Adapt to
removal of wrapper functions.
* libgnat/a-chtgfo.ads (Checked_Index): Remove useless function.
(Delete_Node_At_Index): Delete unused procedure.
Remove violations of SPARK rules, to prepare for the proof of hashed
sets and maps:
- Make the type of hash tables not tagged, so that it will be possible
to mark the type of nodes as having relaxed initialization.
- Remove comparison of addresses as check or optimization: as a check,
it is not necessary in SPARK as the caller cannot pass in aliased
parameters in such cases; as an optimization, it is not possible in
SPARK code.
- Avoid aliasing when inserting a node in the hash table. The code for
insertion in sets and maps was explicitly aliasing a global for the
container and a parameter for its hash table component, both being
written. Rewrite the code to pass only the hash table as parameter.
- Insert constants for subtype constraints, which cannot depend on
variables in SPARK code.
- Use procedures instead of functions when side-effects are expected.
- Rename variables whose value is only written through calls and not
read, using Unused prefix, so that flow analysis does not warn about
it.
gcc/ada/
* libgnat/a-cfhama.adb (Generic_Allocate): Retype to avoid
aliasing.
(Assign, Move): Remove address comparison.
(Include): Insert constants for subtype constraints.
(Insert): Rewrite to avoid aliasing and function with side-effects.
* libgnat/a-cfhase.adb (Generic_Allocate): Retype to avoid
aliasing.
(Assign, Move): Remove address comparison.
(Difference, Intersection, Is_Subset, Overlap,
Symmetric_Difference, Union): Remove address comparison. Insert
constants for subtype constraints.
(Insert): Rewrite to avoid aliasing and function with
side-effects.
* libgnat/a-chtgfk.adb (Checked_Equivalent_Keys, Checked_Index,
Delete_Key_Sans_Free, Find, Generic_Replace_Element, Index):
Type for hash tables not tagged anymore.
(Generic_Conditional_Insert): New_Node generic formal is a
procedure taking the hash table as first parameter now, to avoid
aliasing in the caller.
* libgnat/a-chtgfk.ads: Same.
* libgnat/a-chtgfo.adb (Checked_Index, Clear,
Delete_Node_At_Index, Delete_Node_Sans_Free, First, Free,
Generic_Allocate, Generic_Iteration, Generic_Read,
Generic_Write, Index, Next): Type for hash tables not tagged
anymore.
(Generic_Equal): Removed tagged. Remove address comparison.
* libgnat/a-chtgfo.ads: Same.
* libgnat/a-cohata.ads (Hash_Table_Type): Remove tagged.
Formal hashed sets and maps are now using a different type of hash table
that does not require controlled types. This is possible because formal
containers do not rely on tampering checks, as cursors do not hold a
pointer to a single matching container, but are logical indexes that can
be used with multiple containers.
For simplicity of the package organisation, and symmetry with other
versions of containers, add a formal version of hashed types in the
shared Ada.Containers.Hash_Tables unit. This unit should be specialized
to only include the formal version of hash tables for runtimes that
do not support controlled types (so that the dependency on
Ada.Containers.Helpers can be removed).
The new versions of units for generic keys and operations are modified
versions of the units for bounded containers, with all tampering checks
removed (they were no-op in formal containers). A declaration of
constant Checks is added at the start of the body, as these units cannot
use the version in Ada.Containers.Helpers.
gcc/ada/
* Makefile.rtl: Add new files.
* libgnat/a-cfhama.adb: Use formal version of hash tables.
* libgnat/a-cfhama.ads: Same.
* libgnat/a-cfhase.adb: Same.
* libgnat/a-cfhase.ads: Same.
* libgnat/a-chtgfk.adb: New unit for formal containers, modified
version of Generic_Bounded_Keys.
* libgnat/a-chtgfk.ads: Same.
* libgnat/a-chtgfo.adb: New unit for formal containers, modified
version of Generic_Bounded_Operations.
* libgnat/a-chtgfo.ads: Same.
* libgnat/a-cohata.ads (Generic_Formal_Hash_Table_Types): Third
version of the type for hash tables, equivalent to the bounded
version without tampering checks.
Compiling with and without assertions enabled the name of some generated
symbols differ; this is an issue when using pre-built libraries.
gcc/ada/
* freeze.adb (Check_Inherited_Conditions): Dispatch table
wrappers must be placed in the list of entities of their scope
at the same place of their wrapped primitive. This is required
for private types since these wrappers are built when their full
tagged type declaration is frozen but they may override a
primitive defined in the public part of the package (and it is
important to maintain the wrapper in the list of public entities
of the package to ensure their correct visibility).
The GNSA_ROOT environment variable can be used to indicate to the
compiler how to find the runtime directory.
gcc/ada/
* osint.ads, osint.adb (Relocate_Path): If the GNSA_ROOT
environment variable is set, we use that as the prefix, instead
of computing the prefix from the executable location.
Default initialization of a record object is required to initialize any
components that "require late initialization" after other components.
This includes the case of a type extension; "late initialization"
components of the parent type are required to be initialized after
non-late-init extension components. This is implemented by generalizing
the use of an existing init proc parameter. Previously, the init proc
for a tagged type took a Boolean parameter indicating whether or not to
initialize the Tag component. With this change, this parameter can now
take on any of four values indicating whether to perform
1) Full initialization (including the tag component).
2) Full initialization except for the tag component.
3) Early (non-tag) initialization only.
4) Late (non-tag) initialization only.
With this change, the init proc for a type extension has the option of
performing the early and late portions of the parent's initialization
via two separate calls to the parent type's init proc.
gcc/ada/
* exp_ch3.ads (Build_Intialization_Call): Add new formal
parameter, Init_Control_Actual, with default value. Clients
outside of package Exp_Ch3 are unaffected.
* exp_ch3.adb (Initialization_Control): new package; support for
this 4-valued parameter. The existing Requires_Late_Init
function is moved into this new package.
(Build_Initialization_Call): Add new formal parameter for
subprogram body, use this new formal parameter in generating an
init proc call.
(Build_Record_Init_Proc): Replace Set_Tag Boolean formal
parameter with 4-valued Init_Control_Formal. Wrap if-statements
with appropriate conditions around tag initialization, early
initialization, and late initialization statements.
* exp_util.adb (Build_Task_Image_Decl): Avoid problem with
duplicate declarations when an init proc for a type extension
calls the parent type's init proc twice.
The compiler currently rejects iterating on an interface type derived
from an iterator type.
See Ada RM 5.5.1(6/3) and 5.5.2(3/3).
gcc/ada/
* sem_ch5.adb (Analyze_Iterator_Specification): Fix Typ in the
case of a class-wide derived iterator.
Consistently use >= operator in both the code and contracts of
function/procedure Overwrite, to facilitate proof, instead of the strict
inequality > sometimes, as only New_Item remains in the result in the
case of equal size too.
gcc/ada/
* libgnat/a-strbou.ads (Overwrite): Switch to >= operator in
contracts.
* libgnat/a-strsup.adb (Super_Overwrite): Switch to >= operator
in code of procedure (function already uses it).
* libgnat/a-strsup.ads (Super_Overwrite): Switch to >= operator
in contracts.
