Richard Biener [Wed, 13 May 2026 12:14:19 +0000 (14:14 +0200)]
Split out ix86_vector_cd_cost
This implements costing of vector construction and decomposition
to a custom (possibly vector) element type to be used for vectorizer
costing of the corresponding operations when dealing with
VMAT_STRIDED_SLP loads and stores.
* config/i386/i386.cc (ix86_vector_cd_cost): New function,
enhanced and split out from ...
(ix86_default_vector_cost): ... here.
Richard Biener [Wed, 17 Jun 2026 11:19:56 +0000 (13:19 +0200)]
Add vec_deconstruct costing kind
The following adds vec_decostruct to replace nunits * vec_to_scalar
which allows for more precise costing whenever vectorization
needs to decompose a vector to pieces, like for emulated gather/scatter
but also for strided [SLP] loads/stores.
This requires changes across targets, both for the different kind
but also for the difference in expected count. For now this adjusts
them all to handle vec_deconstruct like vec_construct to avoid ICEing.
Richard Biener [Wed, 13 May 2026 12:22:01 +0000 (14:22 +0200)]
Also record ls element type for costing
I've realized that on x86 we get vector types as vector composition
element types, so just recording the ls_type isn't enough to
realize we're constructing a V4SI from V2SI elements. The following
adds such a field.
* tree-vectorizer.h (vect_load_store_data::ls_eltype): New field.
* tree-vect-stmts.cc (vectorizable_load): Record ltype
as ls_eltype for VMAT_STRIDED_SLP.
(vectorizable_store): Likewise.
Eric Botcazou [Fri, 5 Jun 2026 19:55:29 +0000 (21:55 +0200)]
ada: Fix missing accessibility check for allocator with access discriminant
This implements the last part of the RM 4.8(10.1) rule about allocators
for types with access discriminants. The implementation is modeled on
that of returns for types with access discriminants, given that the set
of rules applying to them is the same (RM 3.10.2(12-12.4)).
The change also streamlines the latter by applying the RM 3.10.2(12.4)
rule in Apply_Accessibility_Check_For_Discriminated_Return.
Piotr Trojanek [Wed, 6 May 2026 14:48:46 +0000 (16:48 +0200)]
ada: Support new Modifies contract for SPARK
Basic support for a new contract. References to SPARK RM will be completed once
the contract is documented there. Legality checks for modified objects being
outputs need to be completed as well (or implemented in GNATprove).
All the code is copied and adapted from the Subprogram_Variant aspect, just
like with other contracts for SPARK, but with notable differences: this
contract requires a dedicated parsing and has no expansion. Currently it is
only supported as an aspect, despite having a corresponding pragma (which is
required to handle it like other contracts).
gcc/ada/ChangeLog:
* aspects.ads (Aspect_Id, Implementation_Defined_Aspect,
Aspect_Argument, Is_Representation_Aspect, Aspect_Names, Aspect_Delay):
Add new aspect and its basic characteristics.
* contracts.adb (Add_Contract_Item,
Analyze_Entry_Or_Subprogram_Contract,
Analyze_Entry_Or_Subprogram_Body_Contract,
Analyze_Subprogram_Body_Stub_Contract): Handle new aspect.
* contracts.ads (Add_Contract_Item,
Analyze_Entry_Or_Subprogram_Contract,
Analyze_Entry_Or_Subprogram_Body_Contract,
Analyze_Subprogram_Body_Stub_Contract): Likewise.
* doc/gnat_rm/implementation_defined_aspects.rst (Aspect Modifies):
Document new aspect.
* doc/gnat_rm/implementation_defined_pragmas.rst (Pragma Modifies):
Document corresponding pragma.
* einfo-utils.adb
(Get_Pragma): Handle new pragma
* einfo-utils.ads
(Get_Pragma): Likewise.
* inline.adb (Remove_Aspects_And_Pragmas): Handle new pragma.
* par-ch13.adb (P_Modifies_Specification): Parse special aspect syntax.
(Get_Aspect_Specifications): Parse new aspect.
* par-prag.adb (Prag): Temporarily accept new pragma.
* sem_ch12.adb (Implementation of Generic Contracts): Update list of
supported contracts.
* sem_ch13.adb (Analyze_One_Aspect, Analyze_Aspect_Specifications,
Check_Aspect_At_Freeze_Point): Analyze new aspect.
* sem_ch8.adb (In_Abstract_View_Pragma): Abstract states can appear in
new contract.
* sem_prag.adb (Contract_Freeze_Error, Ensure_Aggregate_Form): Update
list of supported contracts.
(Analyze_Modifies_In_Decl_Part, Analyze_Pragma): Analyze new pragma.
(Sig_Flags): Support new pragma.
* sem_prag.ads (Aspect_Specifying_Pragma, Assertion_Expression_Pragma,
Pragma_Significant_To_Subprograms, Find_Related_Declaration_Or_Body):
Handle new pragma.
(Analyze_Modifies_In_Decl_Part): Analyze new pragma.
* sem_util.adb, sem_util.ads (Is_Subprogram_Contract_Annotation):
Handle new contract.
* sinfo.ads (Is_Generic_Contract_Pragma, Contract): New contract is
a generic contract.
* snames.ads-tmpl: New contract name and new pragma.
* gnat_rm.texi: Regenerate.
Eric Botcazou [Thu, 4 Jun 2026 13:45:53 +0000 (15:45 +0200)]
ada: Fix missing dynamic accessibility checks for subpool access type
The checks are those prescribed by the RM 13.11.4(24-27) rules, which are
part of dynamic semantics but can be applied statically for the most part.
While the current implementation is incomplete, the fix is straightforward
and mostly boils down to redoing the associated static accessibility checks
of RM 13.11.4(22-23) without too much duplication in the implementation.
gcc/ada/ChangeLog:
* sem_util.ads (Get_Pool_Object_Or_Dereference): New function.
* sem_util.adb (Get_Pool_Object_Or_Dereference): Likewise, mostly
taken from...
* sem_ch13.adb (Analyze_Attribute_Definition_Clause) <Object_From>:
...here.
<Attribute_Storage_Pool>: Call Get_Pool_Object_Or_Dereference.
* exp_ch3.adb (Freeze_Type): Add missing checks of RM 13.11.4(24-27)
and apply them to all regular pools not statically descendants from
Root_Storage_Pool_With_Subpools when its unit has been loaded.
Eric Botcazou [Wed, 3 Jun 2026 21:16:41 +0000 (23:16 +0200)]
ada: Remove obsolete two-pronged implementation of static accessibility checks
For type conversions in instance bodies. It's barely maintainable and the
raise statements are directly inserted for these checks in other contexts.
gcc/ada/ChangeLog:
* exp_ch4.adb (Expand_N_Type_Conversion): Do not apply again static
accessibility checks in instance bodies.
* sem_res.adb (Valid_Conversion): Insert the raise statements for
the failure of static accessibility checks in instance bodies.
Before this patch, In_Package_Body had two different meanings depending
on whether it was applied to a package or to some other entity. This
extracts the second meaning to a new field, Declared_In_Package_Body, and
ties the maintenance of this field to the Scope field, which actually
becomes a thin wrapper over a new internal Gen_IL field.
A new Gen_IL contract is added to help enforce that In_Package_Body is
now only used with package entities.
One might wonder why the conceptually similar In_Private_Part field is
not given the same treatment. One reason is that it has combined the two
different meanings for much longer than In_Package_Body has, so it's much
more work to categorize the call sites. One other reason is that a change
I have in my pipeline is made possible by just this change to
In_Package_Body.
Also update all call sites that previously used Is_Package_Body_Entity to
use the new Declared_In_Package_Body field, remove Is_Package_Body_Entity,
and make print_node display Scope instead of the internal
Scope_Gen_IL_Private field.
gcc/ada/ChangeLog:
* gen_il-fields.ads (Declared_In_Package_Body): New field.
(Scope): Rename into...
(Scope_Gen_IL_Private): ...this.
* einfo-utils.ads (Scope, Set_Scope): New subprograms.
* einfo-utils.adb (Set_Scope): New subprogram.
* einfo.ads (In_Package_Body): Update documentation.
(Declared_In_Package_Body): Add documentation.
(Is_Package_Body_Entity): Remove.
* fe.h (Scope): New macro.
* gen_il-gen-gen_entities.adb (Gen_Entities): New field.
New contract. Remove Sm (Is_Package_Body_Entity).
* gen_il-gen-gen_nodes.adb (Gen_Nodes): Adapt to renaming.
* gen_il-gen.adb (Compute_Field_Offsets): Adapt to renaming.
* gen_il-internals.adb (Image): Adapt to renaming.
* lib-load.adb, lib.adb: Add with clauses.
* sem_ch3.adb (Analyze_Object_Declaration): Remove field update.
* sem_ch6.adb (Enter_Overloaded_Entity): Likewise.
* sem_util.adb (Collect_Primitive_Operations): Use new field.
(Enter_Name): Replace Is_Package_Body_Entity with
Declared_In_Package_Body.
* treepr.adb (Image): Add F_Scope_Gen_IL_Private alternative
that returns Scope.
* exp_dbug.adb (Is_BNPE): Replace Is_Package_Body_Entity with
Declared_In_Package_Body.
* sem_ch12.adb (Check_Generic_Actuals): Likewise.
* sem_warn.adb (Generic_Package_Spec_Entity): Likewise.
* sem_ch7.adb (Analyze_Package_Body_Helper): Replace
Set_Is_Package_Body_Entity with Set_Declared_In_Package_Body.
Viljar Indus [Wed, 3 Jun 2026 12:15:56 +0000 (15:15 +0300)]
ada: Use correct warning char for unconditional elab warning
Warning messages from Check_Internal_Call_Continue were
not actually conditional on the -gnatwl flag and were
emitted always.
Use the switchless warning character for those messages
and refactor Output_Calls in order to handle all of the
possible warning characters in the continuation messages.
gcc/ada/ChangeLog:
* sem_elab.adb (Output_Calls): Apply correct warning
characeter for each continuation message.
(Check_Internal_Call_Continue): Use regular warning
character in error messages.
Eric Botcazou [Wed, 3 Jun 2026 10:50:58 +0000 (12:50 +0200)]
ada: Fix missing error on formal with access discriminant in allocator and -gnatc
The problem is that the relevant part of the RM 4.8(5.3) rule is implemented
in the expander (the other part being implemented in the analyzer), so this
merges the former with the latter and streamlines the result in the process.
This also tweaks the Accessibility_Level function to make it more robust in
the presence of errors, as well as adjusts the new error message given by
the Check_Return_Construct_Accessibility procedure, which was a mouthful
and not really in keeping with other error messages about accessibility.
gcc/ada/ChangeLog:
* accessibility.adb (Accessibility_Level): Also deal with SAOOAAATs
without extra accessibility object alongside formal parameters.
(Check_Return_Construct_Accessibility): Streamline the error message
and adjust it to expression functions.
* exp_ch4.adb (Expand_Allocator_Expression): Do not perform a static
accessibility check here.
* sem_res.adb (Resolve_Allocator): Streamline the implementation of
the static accessibility check of RM 4.8(5.3), use a more consistent
error message, adjust it to instantiation bodies, and specifically
deal with formal parameters.
ada: Make subprogram insertion code easier to understand
This extracts part of the statement sequence of the
Build_And_Insert_Type_Attr_Subp procedure into a helper function. The
intent is to make it easier to tell when the objects declared inside the
procedure are mutated.
gcc/ada/ChangeLog:
* exp_attr.adb (Skip_Non_Source_Subps): Move before local object
declarations and add formals.
(Build): Move before local object declarations.
(Find_Insertion_Point_For_Ancestor): New local function.
(Build_And_Insert_Type_Attr_Subp): Use new function.
Piotr Trojanek [Tue, 26 May 2026 09:20:20 +0000 (11:20 +0200)]
ada: Fix missing context for class-wide contract expressions
The copies of class-wide pre- and postconditions are now attached to the
original pragma, so that legality checks, like the one for 'Old, can find the
original context of where those expressions appeared.
gcc/ada/ChangeLog:
* sem_attr.adb (Uneval_Old_Msg): Attribute Old never occurs outside of
a pragma.
* sem_ch13.adb (Analyze_One_Aspect): Attach expression copy of a
class-wide aspect to the corresponding pragma.
* sem_prag.adb (Analyze_Pre_Post_Condition): Likewise for a class-wide
pragma.
Eric Botcazou [Sat, 30 May 2026 19:11:44 +0000 (21:11 +0200)]
ada: Fix assertion failure on access to uninitialized package entity
It occurs during the analysis of a generic package declared in a structural
instance of a parent generic package, on accessing the uninitialized entity
of the current semantic unit. But this access is unnecessary because what
happens within an instance does not affect the current semantic unit here.
The change also adds the missing call to Unit_Requires_Body already present
in the generic subprogram case to the generic package case.
gcc/ada/ChangeLog:
* sem_ch12.adb (Analyze_Generic_Package_Declaration): Do not set the
Body_Needed_For_Inlining flag on the current semantic unit if the
package is declared within an instance or does not require a body.
(Analyze_Generic_Subprogram_Declaration): Similarly, do not set the
Body_Needed_For_Inlining flag on the current semantic unit if the
subprogram is declared within an instance.
Javier Miranda [Thu, 28 May 2026 16:58:09 +0000 (16:58 +0000)]
ada: Crash on allocator for tagged object with constructor
This patch fixed a compiler crash when allocating a tagged record
types through class-wide access type.
gcc/ada/ChangeLog:
* exp_ch4.adb (Expand_N_Allocator): For a class-wide designated
type with a constructor initializer, add a type conversion to
the specific type to avoid a dispatching call (since contructors
are not dispatching primitives).
* sem_ch13.adb (Analyze_Aspect_Specifications)
<Aspect_Initialize>: Avoid reporting a spurious error.
Eric Botcazou [Mon, 25 May 2026 09:22:10 +0000 (11:22 +0200)]
ada: Fix missing error on nonaliased formal with access discriminants in return
The construct must fail the static accessibility check of 6.5(5.9) because
the accessibility level of the anonymous access type of the discriminants
is that of the invocation of the function, which is too deep for a return.
The change also gets rid of the implementation of this check in the routine
Apply_Accessibility_Check_For_Anonymous_Return because 1) this routine is
meant for the corresponding dynamic accessibility check 6.5(21) and, thus,
only invoked when expansion is active, and 2) the check is implemented in
other routines, e.g. Check_Return_Construct_Accessibility for discriminants.
The change also implements the 6.8(5) rule, which is the counterpart of the
6.5(5.9) rule for access discriminants in expression functions.
gcc/ada/ChangeLog:
* accessibility.ads (Check_Return_Construct_Accessibility): Document
that it also implements the 6.8(5) rule.
* accessibility.adb (Accessibility_Message): Remove dead return.
(Apply_Accessibility_Check_For_Anonymous_Return): Do not apply the
static check of 6.5(5.9) here.
(Check_Return_Construct_Accessibility): Do not use "return object"
in the error message. Do not reject synthesized return statements
when the function comes from source to handle expression funtions.
Robustify processing and explicitly deal with formal parameters.
Eric Botcazou [Fri, 29 May 2026 06:34:39 +0000 (08:34 +0200)]
ada: Fix dangling reference with Bounded_Indefinite_Holders and Empty_Holder
The issue is that the finalization of an empty container does not deallocate
its subpool, thus creating a dangling reference to it from the global pool.
gcc/ada/ChangeLog:
* libgnat/a-cbinho.adb (Clear): Deallocate the subpool even if the
container is empty.
Viljar Indus [Thu, 28 May 2026 12:26:12 +0000 (15:26 +0300)]
ada: Fix diagnostics inteneded as continuations
The diagnostic messages fixed here were using the | insertion
character instead of the \ used for continuations. Even though
they were clearly intended to supplement the previous message.
Javier Miranda [Tue, 26 May 2026 16:06:49 +0000 (16:06 +0000)]
ada: Improve error reported for wrong constructor call
This patch improves the error reported when the invoked constructor is
not available. This is a common error because constructors are not
inherited; this means that a constructor for a given tagged type may
not exist for its derived types. Fix also a crash in the initialization
of a limited record type.
gcc/ada/ChangeLog:
* sem_attr.adb (Analyze_Attribute): Report an error when the number
of actual parameters for a constructor do not match any available
constructor.
* exp_ch6.adb (Make_Init_Proc_Call): Add the _Init_Level accessibility
level formal when building a call to the init proc of a limited record
type.
(Prepend_Constructor_Procedure_Prologue): Restrict the initialization
of the _parent component to cases where a Super aspect is present
or a parameterless parent constructor exists.
Marek Polacek [Fri, 12 Jun 2026 15:44:15 +0000 (11:44 -0400)]
c++/reflection: ICE in get_reflection on INDIRECT_REF [PR125759]
We crash here in outer_var_p on:
/* These should have been stripped or otherwise handled by the caller. */
gcc_checking_assert (!REFERENCE_REF_P (decl));
because we never stripped the INDIRECT_REF in get_reflection. I don't
think stripping it only for the outer_automatic_var_p call would be
right so I'm doing it at the top of the function.
PR c++/125759
gcc/cp/ChangeLog:
* reflect.cc (get_reflection): Do STRIP_REFERENCE_REF.
Wang Jinghao [Wed, 17 Jun 2026 15:17:47 +0000 (11:17 -0400)]
c++/reflection: Handle stale parameter decarations in eval_is_explicit_object_parameter
eval_is_explicit_object_parameter() determines whether a parameter is
an explicit object parameter by checking whether the passed-in tree
node is the first parameter in its parameter chain. As described in
the comment for maybe_update_function_parm(), if the sequence is
declaration -> reflection -> definition, the old PARM_DECL will be
compared against the new PARM_DECL, causing the check to produce an
incorrect result.
Calling maybe_update_function_parm() before the comparison updates
the old reflection to the new PARM_DECL in the chain, thereby
avoiding this error.
gcc/cp/ChangeLog:
* reflect.cc (eval_is_explicit_object_parameter): Call
maybe_update_function_parm before checking the parameter
against DECL_ARGUMENTS.
gcc/testsuite/ChangeLog:
* g++.dg/reflect/is_explicit_object_parameter2.C: New test.
Signed-off-by: Wang Jinghao <zheng.xianyuwang@gmail.com> Reviewed-by: Patrick Palka <ppalka@redhat.com>
Antoni Boucher [Fri, 5 Sep 2025 14:32:40 +0000 (10:32 -0400)]
libgccjit: Add gcc_jit_type_is_floating_point
gcc/jit/ChangeLog:
* docs/topics/compatibility.rst (LIBGCCJIT_ABI_38): New ABI tag.
* docs/topics/types.rst: Document gcc_jit_type_is_floating_point.
* libgccjit.cc (gcc_jit_type_is_floating_point): New function.
* libgccjit.h (gcc_jit_type_is_floating_point): New function.
* libgccjit.map: New function.
* libgccjit.exports: New function.
gcc/testsuite/ChangeLog:
* jit.dg/test-reflection.c: Add test for
gcc_jit_type_is_floating_point.
Co-authored-by: Robert Zakrzewski <robert.zakrzewski1@stellantis.com>
Marek Polacek [Fri, 5 Jun 2026 17:42:06 +0000 (13:42 -0400)]
c++: explicit instantiation and noexcept-specifier [PR125613]
This patch implements this part of [except.spec]: In an explicit
instantiation a noexcept-specifier may be specified, but is not required.
If a noexcept-specifier is specified in an explicit instantiation, the
exception specification shall be the same as the exception specification
of all other declarations of that function.
But we are not checking this, and are the only compiler that accepts:
Marek Polacek [Fri, 5 Jun 2026 21:25:36 +0000 (17:25 -0400)]
libstdc++: adjust explicit inst of std::basic_string::data()
r15-2650 removed noexcept from non-const std::basic_string::data()
in bits/cow_string.h, but the explicit instantiations in string-inst.cc
kept their noexcept. [except.spec] says that in an explicit instantiation
a noexcept-specifier may be specified, but is not required, but when it
is specified, it must be the same as the exception specification of all
other declarations of that function. So I think we should just remove it.
Wilco Dijkstra [Wed, 17 Jun 2026 13:32:25 +0000 (13:32 +0000)]
libffi: Update LOCAL_PATCHES
Add commit:
commit 94c5f47d21aa5f144a690189222cc08b86a00505 (origin/master, origin/HEAD)
Author: Muhammad Kamran <muhammad.kamran@arm.com>
Date: Wed Jun 17 13:25:31 2026 +0000
libffi: Mark AArch64 BTI and PAC with build attributes
This is an adapted backport of the BTI/PAC parts of upstream libffi commit 547449d705258ea490aa8787dc084c5b38dbe722, which emits AArch64 feature build
attributes when assembler support is available and keeps the existing GNU
property note as the fallback.
GCC's current libffi copy does not include the upstream GCS support, so
this backport only covers the existing BTI and PAC feature marking.
Muhammad Kamran [Wed, 17 Jun 2026 13:25:31 +0000 (13:25 +0000)]
libffi: Mark AArch64 BTI and PAC with build attributes
This is an adapted backport of the BTI/PAC parts of upstream libffi commit 547449d705258ea490aa8787dc084c5b38dbe722, which emits AArch64 feature build
attributes when assembler support is available and keeps the existing GNU
property note as the fallback.
GCC's current libffi copy does not include the upstream GCS support, so
this backport only covers the existing BTI and PAC feature marking.
libffi/ChangeLog:
* src/aarch64/internal.h (AARCH64_POINTER_AUTH): Rename from
GNU_PROPERTY_AARCH64_POINTER_AUTH.
* src/aarch64/sysv.S (AARCH64_BTI): Rename from
GNU_PROPERTY_AARCH64_BTI.
(GNU_PROPERTY): New macro.
(FEATURE_1_AND_MARK): Likewise.
Emit BTI and PAC feature build attributes when
__ARM_BUILDATTR64_FV is defined.
Muhammad Kamran [Wed, 17 Jun 2026 13:23:56 +0000 (13:23 +0000)]
libitm: Mark AArch64 asm features with build attributes
Emit AArch64 build attributes for BTI and PAC from libitm's hand-written
AArch64 assembly when __ARM_BUILDATTR64_FV is available. Keep the
existing GNU property note emission as the fallback for toolchains that do
not support the new attribute form.
libitm/ChangeLog:
* config/aarch64/sjlj.S (FEATURE_1_AND_MARK): Define. Use
AArch64 build attributes when __ARM_BUILDATTR64_FV is available,
otherwise emit a GNU property note.
Muhammad Kamran [Wed, 17 Jun 2026 13:23:50 +0000 (13:23 +0000)]
libatomic: Mark AArch64 asm features with build attributes
Use AArch64 build attributes to record BTI, PAC and GCS feature bits when
__ARM_BUILDATTR64_FV is available. Fall back to the existing GNU property
note emission otherwise.
libatomic/ChangeLog:
* config/linux/aarch64/atomic_16.S (FEATURE_1_AND_MARK): Define.
Use AArch64 build attributes when __ARM_BUILDATTR64_FV is
available, otherwise emit a GNU property note.
Muhammad Kamran [Wed, 17 Jun 2026 13:23:42 +0000 (13:23 +0000)]
libgcc: Mark AArch64 asm features with build attributes
Emit AArch64 build attributes for BTI, PAC and GCS when
__ARM_BUILDATTR64_FV is available. Keep the existing GNU property note
emission as the fallback for toolchains that do not support the new
attribute form.
libgcc/ChangeLog:
* config/aarch64/aarch64-asm.h (FEATURE_1_AND_MARK): Define.
Use AArch64 build attributes when __ARM_BUILDATTR64_FV is
available, otherwise emit a GNU property note.
Tamar Christina [Wed, 17 Jun 2026 11:09:22 +0000 (12:09 +0100)]
vect: Fix early break in PEELED cases [PR125804]
When a loop is LOOP_VINFO_EARLY_BREAKS_VECT_PEELED && ! LOOP_VINFO_EARLY_BRK_NEEDS_EPILOG
then we still need the epilog loop for the main latch edge.
The early exits themselves when LOOP_VINFO_EARLY_BRK_NEEDS_EPILOG don't need to
go to scalar code.
Allowing bound_epilog to still be set to VF when
LOOP_VINFO_EARLY_BREAKS_VECT_PEELED we get to the desired effect that the early
exits do not require an epilog when !LOOP_VINFO_EARLY_BRK_NEEDS_EPILOG but
the main edge still does if LOOP_VINFO_EARLY_BREAKS_VECT_PEELED.
gcc/ChangeLog:
PR tree-optimization/125804
* tree-vect-loop-manip.cc (vect_do_peeling): Have
LOOP_VINFO_EARLY_BREAKS_VECT_PEELED require epilog.
gcc/testsuite/ChangeLog:
PR tree-optimization/125804
* gcc.dg/vect/vect-early-break_144-pr125804.c: New test.
Tobias Burnus [Wed, 17 Jun 2026 10:16:55 +0000 (12:16 +0200)]
Fortran/OpenMP: Rename declare-mapper struct members
The usm->mapper_id vs. usm->usm->mapper_id and
also the usm->usm itself was a bit confusing.
Hence, this is now:
usm->requested_mapper_id
usm->resolved_usm
where the latter has
usm->resolved_usm->mapper_id
Hereby, usm->requested_mapper_id is set for a map/to/from
clause such as 'map(mapper(my_name), to: x' - while
the resolved_usm points to an object that has been
created by 'declare mapper'.
gcc/fortran/ChangeLog:
* gfortran.h (struct gfc_omp_udm): Add comment.
(struct gfc_omp_namelist_udm): Likewise; rename members
mapper_id to requested_mapper_id and usm to resolved_usm.
* module.cc (load_omp_udms, write_omp_udm): Update accordingly.
* openmp.cc (gfc_match_omp_clauses, resolve_omp_clauses): Likewise.
* trans-openmp.cc (gfc_trans_omp_clauses): Likewise.
Dhruv Chawla [Tue, 16 Jun 2026 12:05:39 +0000 (12:05 +0000)]
lto: Fix streaming of loop->can_be_parallel [PR125717]
The can_be_parallel flag was not being streamed out, so any frontend
setting the flag (via annot_expr_parallel_kind) would not observe it doing
anything when LTO was enabled.
Zhongjie Guo [Wed, 17 Jun 2026 03:29:15 +0000 (03:29 +0000)]
i386: Adjust c86-4g-m7 512-bit memory costs
c86-4g-m7 is a split-regs AVX512 target. A 512-bit memory
operation is implemented as two 256-bit halves, so the vectorizer cost
model should not make 512-bit loads and stores almost as cheap as
256-bit ones.
The old c86_4g_m7_cost values made 512-bit loads/stores cost 12/12,
close to or equal to the 256-bit 10/12 costs. This can make 64-byte
vectorization win in the loop body cost comparison even when 32-byte
vectors avoid extra reduction epilogue work.
Set the 512-bit load/store and unaligned load/store costs to twice the
256-bit costs. This removes the artificial 64-byte body-cost advantage;
for dot-product style reduction loops, the reduction epilogue cost can
then make 32-byte vectorization preferable.
Compared with gcc-trunk without this tuning, local SPEC2006/SPEC2017
testing shows improvements in several vector-width sensitive workloads.
SPEC2006 1-copy fp_speed improved by 2.32%, including 436.cactusADM
+12.39%, 433.milc +7.33%, and 459.GemsFDTD +4.76%. SPEC2017 32-copy
fprate improved by 0.40%, with 526.blender_r improving by 2.80%.
gcc/ChangeLog:
* config/i386/x86-tune-costs.h (c86_4g_m7_cost): Increase
512-bit load/store and unaligned load/store costs.
gcc/testsuite/ChangeLog:
* gcc.target/i386/c86-4g-m7-vect-load-cost-reduc.c: New test.
Xin Liu [Wed, 17 Jun 2026 03:28:39 +0000 (03:28 +0000)]
i386: Fix some c86-4g-m7 reservations
Fix several existing c86-4g-m7 scheduling reservation issues.
The fixes correct decode unit selection, branch and call execution units,
missing store resources for store or load/store forms, and several FPU
pipeline resource descriptions. They also rename a few reservations so the
template names better match the instructions they cover, and simplify
duplicate memory attribute checks.
gcc/ChangeLog:
* config/i386/c86-4g-m7.md (c86_4g_m7_imov_xchg): Adjust
reservation units.
(c86_4g_m7_imov_xchg_load): Ditto.
(c86_4g_m7_call): Ditto.
(c86_4g_m7_branch): Ditto.
(c86_4g_m7_branch_load): Ditto.
(c86_4g_m7_fp_spc_direct): Add missing store unit.
(c86_4g_m7_sse_pinsr_reg): Adjust reservation units.
(c86_4g_m7_avx512_insertx_ymm): Ditto.
(c86_4g_m7_avx512_insertx_ymem): Ditto.
(c86_4g_m7_avx512_insertx_zxmm): Ditto.
(c86_4g_m7_avx512_insertx_zxmem): Ditto.
(c86_4g_m7_avx512_abs_load): Add missing store unit.
(c86_4g_m7_avx_sign): Use combined FPU reservation.
(c86_4g_m7_avx_sign_load): Ditto.
(c86_4g_m7_avx_aes): Ditto.
(c86_4g_m7_avx_aes_load): Ditto.
(c86_4g_m7_extr_load): Rename to ...
(c86_4g_m7_extr_store): ... this and restrict to store memory.
(c86_4g_m7_avx_imul): Use combined FPU reservation.
(c86_4g_m7_avx_imul_mem): Ditto.
(c86_4g_m7_avx512_vpmovx_y_load): Add missing store unit.
(c86_4g_m7_avx_vpmovx_xx_load): Ditto.
(c86_4g_m7_avx512_sseadd_maxmin_xy): Rename to ...
(c86_4g_m7_avx512_sseadd_maxmin): ... this and simplify
memory attribute check.
(c86_4g_m7_avx512_sseadd_maxmin_xy_load): Rename to ...
(c86_4g_m7_avx512_sseadd_maxmin_load): ... this and simplify
memory attribute check.
(c86_4g_m7_avx512_sseadd_xy): Rename to ...
(c86_4g_m7_avx512_sseadd): ... this.
(c86_4g_m7_avx512_sseadd_xy_load): Rename to ...
(c86_4g_m7_avx512_sseadd_load): ... this.
(c86_4g_m7_sse_sseiadd_sadbw): Use combined FPU reservation.
(c86_4g_m7_sse_sseiadd_sadbw_mem): Ditto.
(c86_4g_m7_avx512_ssecmp_vp_z): Adjust reservation units.
(c86_4g_m7_avx512_ssecmp_vp_z_load): Ditto.
(c86_4g_m7_avx512_ssecmp_test_load): Ditto.
(c86_4g_m7_avx512_mskmov_k_m): Adjust latency.
Robin Dapp [Wed, 17 Jun 2026 04:40:34 +0000 (22:40 -0600)]
[PATCH] RISC-V: Fix more scalar mode_idx instances [PR125478].
Hi,
This is another case of PR123022 and PR116149 where we query a scalar source
operand instead of a vector operand, leading to a wrong AVL during avlprop.
This patch moves viwalu, vfwalu, viwmul, and vfwmul to the proper
bucket. I wonder why I didn't do that the last two times but it seems
to be the correct choice now 🙂
Regtested on rv64gcv_zvl512b and waiting for the CI.
Regards
Robin
PR target/125478
gcc/ChangeLog:
* config/riscv/vector.md: Set widen-alu mode_idx to 3.
gcc/testsuite/ChangeLog:
* gcc.target/riscv/rvv/autovec/pr125478.c: New test.
Jeff Law [Wed, 17 Jun 2026 02:34:49 +0000 (20:34 -0600)]
[committed][RISC-V] Tighten and xfail test for pr120811
So the test for pr120811 has actually been failing for several months, but the
scan-asm regexp was too general in what it accepted and as a result we weren't
aware it was failing. What's a bit weird, is it looks like post-reload combine
just never tries the relevant insns, though it does look like a viable
post-reload combine opportunity.
Then changes in the last day or so scrambled the output even more to the point
where the faulty scan-asm wouldn't match and the regression has come to light.
This adds the missing escapes to the scan-asm. I've confirmed it works with
the old (desired) code generation and it fails with the change from a couple
months ago which seems to be preventing the post-reload combine. It xfails the
test as well.
I'll get a bug opened to track the performance regression.
gcc/testsuite
* gcc.target/riscv/pr120811.c: Add missing escapes and xfail.
Andrew MacLeod [Thu, 11 Jun 2026 17:16:13 +0000 (13:16 -0400)]
Pointer global ranges of nonzero are no longer invariant.
TO preserve processing, the cache did not track pointer ranges any more
once it was set to non-zero. With prange now tracking pointers, any
conditional may provide points-to info to a nonzero range, so we
must track these values now.
* gimple-range-cache.cc (ranger_cache::set_global_range): Nonzero
pointer ranges are no longer invariant.
Andrew MacLeod [Mon, 1 Jun 2026 18:28:33 +0000 (14:28 -0400)]
Switch VRP to use prange.
Use the points-to info in prange instead olf the value-range-equiv side
table in VRP. Adjust testcases:
- PTA causes less mem*_chk routines to be generated.
- Gimple fold is causing issues by not properly handling pointer_plus.
See PR 123160. XFAILing until resolved.
- Adding noipa prevents VRP from propagating IPA info causing a failure.
gcc/
* tree-vrp.cc (rvrp_folder::value_of_expr): Use prange PTA info.
(value_on_edge): Likewise.
Jose E. Marchesi [Tue, 16 Jun 2026 20:52:14 +0000 (22:52 +0200)]
a68: fix deduplication of in-MOIF modes
Of course I got the in-MOIF deduplication of modes wrong due to a
stupid thinko. This patch fixes the thinko and also adds a little
extra sanity check.
Signed-off-by: Jose E. Marchesi <jemarch@gnu.org>
gcc/algol68/ChangeLog
* a68-imports.cc (a68_replace_equivalent_mode): Get a moif and
check extracts.
(a68_open_packet): Fix deduplication of in-moif modes and move
extract replacement code to a68_replace_equivalent_mode.
* a68-low-moids.cc (a68_lower_moids): Check that all known modes
have an associated ctype as part of the sanity checks.
H.J. Lu [Fri, 8 May 2026 04:20:02 +0000 (12:20 +0800)]
SSP: Check UINTPTR_TYPE to get uintptr_t type
default_stack_protect_guard calls
lang_hooks.types.type_for_mode (ptr_mode, 1);
to get an integer type for __stack_chk_guard which is declared as a
global symbol of type uintptr_t. For 32-bit systems, uintptr_t may
be either unsigned int or unsigned long int. On 32-bit Darwin, we get
$ cat /tmp/x.c
__UINTPTR_TYPE__ __stack_chk_guard = 0x1000;
$ ./xgcc -B./ -S /tmp/x.c -m32
/tmp/x.c:1:18: error: conflicting types for ‘__stack_chk_guard’; have ‘long unsigned int’
1 | __UINTPTR_TYPE__ __stack_chk_guard = 0x1000;
| ^~~~~~~~~~~~~~~~~
cc1: note: previous declaration of ‘__stack_chk_guard’ with type ‘unsigned int’
$
since lang_hooks.types.type_for_mode returns unsigned int while Darwin's
uintptr_t is unsigned long int. Update default_stack_protect_guard to
call unsigned_integer_tree_node_for_type with UINTPTR_TYPE to get unsigned
integer type for uintptr_t instead.
gcc/
PR c/125226
* targhooks.cc (default_stack_protect_guard): If UINTPTR_TYPE
isn't NULL, call unsigned_integer_tree_node_for_type with
UINTPTR_TYPE to get unsigned integer type for uintptr_t.
* tree.cc (unsigned_integer_tree_node_for_type): New function.
(build_common_tree_nodes): Call unsigned_integer_tree_node with
SIZE_TYPE to get unsigned integer type for size_t.
* tree.h (unsigned_integer_tree_node_for_type): New prototype.
Robert Dubner [Tue, 16 Jun 2026 18:26:24 +0000 (14:26 -0400)]
cobol: Simplify tree type determination; eliminate file-static variables.
Use tree_type_from_field() and tree_type_from_refer() to replace a number
of similar routines. Replace a number of specific file-static
variable definitions with non-specific variable definitions.
Repair calculation of DIVIDE ... GIVING REMAINDER where parameters are
negative.
* gmath.cc (multiply_int256_by_int64): Comment.
(int256_as_decimal): Handle negative values.
(__gg__dividef45): Repair remainder logic when numeric-display
parameters are negative.
gcc/testsuite/ChangeLog:
* cobol.dg/group2/DIVIDE_binary-long_giving_remainder.cob: New test.
* cobol.dg/group2/DIVIDE_binary-long_giving_remainder.out: New test.
* cobol.dg/group2/DIVIDE_numeric-display_giving_remainder.cob: New test.
* cobol.dg/group2/DIVIDE_numeric-display_giving_remainder.out: New test.
* cobol.dg/group2/GCC_125616_RT3601_-_IBM_incorrect_DISPLAY_of_COMP-1_COMP-2_float.cob:
New test.
* cobol.dg/group2/PR39_RT3573_-_Parser_issue_w_concat._source_lines.cob: New test.
Jakub Jelinek [Tue, 16 Jun 2026 20:24:57 +0000 (22:24 +0200)]
c++, libcpp: Add -std=c++2[9d] and -std=gnu++2[9d] options
The following patch adds a new -std= option arguments for C++29 and
changes __cplusplus value for C++26 from 202400L to 202603L.
2026-06-16 Jakub Jelinek <jakub@redhat.com>
gcc/
* doc/invoke.texi (-std=c++2d, -std=c++29, -std=gnu++2d,
-std=gnu++29): Document.
(-std=c++2c, -std=c++26, -std=gnu++2c, -std=gnu++26): Tweak so that
26 comes first and is meant as the supported option.
* doc/cpp.texi: Adjust expected __cplusplus value for -std=c++26,
document value for -std=c++29.
* dwarf2out.cc (highest_c_language): Handle C++29.
(gen_compile_unit_die): Likewise. Adjust lversion for C++26.
gcc/c-family/
* c-common.h (enum cxx_dialect): Add cxx29.
* c-opts.cc (set_std_cxx29): New function.
(c_common_handle_option): Handle -std=c++2d, -std=c++29, -std=gnu++2d
and -std=gnu++29.
* c.opt (std=c++2d, std=c++29, std=gnu++2d, std=gnu++29): New.
(std=c++2c, std=c++26, std=gnu++2c, std=gnu++26): Move Undocumented
to the 2c cases, tweak description.
gcc/testsuite/
* g++.dg/cpp26/cplusplus.C: Expect __cplusplus == 202603L.
* g++.dg/cpp29/cplusplus.C: New test.
* g++.dg/cpp29/feat-cxx29.C: New test.
* lib/target-supports.exp (check_effective_target_c++26): Also
true for C++29.
(check_effective_target_c++26_down, check_effective_target_c++29_only,
check_effective_target_c++29): New procedures.
* lib/g++-dg.exp (g++-std-flags): Replace 26 with 29 in the list
and add 26 to the end.
libcpp/
* include/cpplib.h (enum c_lang): Add CLK_GNUCXX29 and CLK_CXX29.
* init.cc (lang_defaults): Add CLK_GNUCXX29 and CLK_CXX29 row.
(cpp_init_builtins): Change __cplusplus value for C++26 from 202400L
to 202603L and set __cplusplus value for C++29 to 202700L.
James K. Lowden [Tue, 16 Jun 2026 15:15:46 +0000 (11:15 -0400)]
cobol: Use cdf namespace in CDF Bison parser output.
Fixes PR 119215 LTO error by removing enumerated types and other
names from the global namespace, and by deleting use of YYLTYPE. A
single location type cbl_loc_t is used by both parsers.
Also fixes PR 122466 using conditional compilation for non-POSIX
symbols.
Require Bison 3.8.2 for C++ output. The generated C++ uses
iostreams to format error messages. The cdf.y file includes the
iostream header and does not use any GCC header files.
gcc/cobol/ChangeLog:
* Make-lang.in: Report Bison version.
* cbldiag.h (defined): Define possibly missing macros.
(ATTRIBUTE_GCOBOL_DIAG): Define if missing.
(ATTRIBUTE_PRINTF_1): Same.
(ATTRIBUTE_PRINTF_3): Same.
(yyerror): Remove.
(struct YYLTYPE): Remove.
(enum cbl_gcobol_feature_t): Relocate from symbols.h.
(YYLTYPE_IS_DECLARED): Remove.
(YYLTYPE_IS_TRIVIAL): Remove.
(cobol_location): Relocate.
(cobol_gcobol_feature_set): Declare.
(struct YDFLTYPE): Remove.
(enum cbl_call_convention_t): Relocate from symbols.h.
(YDFLTYPE_IS_DECLARED): Remove.
(YDFLTYPE_IS_TRIVIAL): Remove.
(current_call_convention): Declare.
(cdf_push): Declare.
(cdf_push_call_convention): Declare.
(cdf_push_current_tokens): Declare.
(cdf_push_dictionary): Declare.
(cdf_push_enabled_exceptions): Declare.
(cdf_push_source_format): Declare.
(cdf_pop): Declare.
(cdf_pop_call_convention): Declare.
(cdf_pop_current_tokens): Declare.
(cdf_pop_dictionary): Declare.
(cdf_pop_source_format): Declare.
(cdf_pop_enabled_exceptions): Declare.
(current_program_index): Declare.
(struct cbl_loc_t): Derive from cbl_loc_base_t.
(struct cbl_loc_base_t): Define.
(cbl_err): Declare.
(cbl_errx): Declare.
(error_msg): Use cbl_loc_t.
(warn_msg): Same.
(cbl_unimplemented_at): Same.
(gcc_location_set): Same.
* cdf.y: Require Bison 3.8.2 and generate C++ in cdf namespace.
* cdfval.h (struct YDFLTYPE): Remove.
(struct cbl_loc_t): Forward declaration.
(struct cdfval_base_t): User-defined conversion from derived.
* copybook.h (gcc_assert): Use assert(3) within cdf.y.
(gcc_unreachable): Declare within cdf.y.
(class copybook_elem_t): Use cbl_loc_t.
(CTOUPPER): Use toupper(3) in uppername_t helper.
(TOUPPER): Same.
(class copybook_t): Use cbl_loc_t.
* exceptg.h (struct cbl_label_t): Declare cbl_label_t.
* gcobc: Support -fno-ec.
* gcobol.1: Reword -fsyntax-only slightly.
* genapi.cc (parser_label_label): Use cobol_location().
* lang.opt: Add comment in re lang.opt.urls.
* lexio.cc (struct replacing_term_t): Use cbl_loc_t.
(location_in): Same.
(parse_copy_directive): Same.
* lexio.h (struct filespan_t): Same.
* messages.cc (cbl_message): Same.
* parse.y: Same, and propagate location variously.
* parse_ante.h (current_data_section_set): Use cbl_loc_t.
(namcpy): Same.
(reject_refmod): Same.
(require_pointer): Same.
(require_integer): Same.
(ast_op): Same.
(perform_tgt_set): Same.
(label_add): Same.
(paragraph_reference): Same.
(tee_up_name): Same.
(ast_inspect): Same.
(ast_enter_section): Same.
(ast_enter_paragraph): Same.
(prototype_add): Same.
(verify_args): Same.
(subscript_dimension_error): Same.
(literal_subscripts_valid): Same.
(literal_refmod_valid): Same.
(struct cbl_fieldloc_t): Remove.
(intrinsic_call_1): Use cbl_loc_t.
(symbol_find): Same.
(valid_redefine): Same.
(field_add): Same.
(field_type_update): Same.
(field_capacity_error): Same.
(field_alloc): Same.
(file_add): Same.
(alphabet_add): Same.
(set_real_from_capacity): Same.
(procedure_division_ready): Same.
(file_section_fd_set): Same.
(ast_call): Same.
(field_binary_usage): Same.
(ast_end_program): Same.
(cobol_location): Same.
(location_set): Same.
(statement_begin): Same.
(ast_first_statement): Same.
* scan.l: Qualify tokens with new cdf namespace.
* scan_ante.h (ydfparse): Now static.
(cdf_context): Declare.
(ydfltype_of): Remove.
(update_location): Use cbl_loc_t.
(reset_location): Same.
(YY_USER_INIT): Same.
(class picture_t): Same.
* scan_post.h (ydfparse): Wrapper for cdf_parser::parse() method.
(ydfchar): Lookahead helper.
(ydfdebug): Same.
(run_cdf): Clearer debug messages.
(struct pending_token_t): Renamed to recent_token_t.
(struct recent_token_t): As above.
(PENDING): Renames to RECENT.
(RECENT): As above.
(next_token): Removed.
(recent_tokens_t): Capture abandoned lookahead tokens.
(prelex): Use recent_tokens queue.
(yylex): Drop normal parsing idea.
* symbols.cc (symbol_field_location): Use cbl_loc_t.
(symbol_alphabet): Same.
(cbl_alphabet_t::cbl_alphabet_t): Same.
(cbl_alphabet_t::assign): Same.
(cbl_alphabet_t::also): Same.
(symbol_temporary_location): Same.
(cbl_field_t::encode): Same.
* symbols.h (enum cbl_gcobol_feature_t): Remove.
(cobol_gcobol_feature_set): Remove.
(struct cbl_field_t): Use cbl_loc_t.
(struct cbl_refer_t): Same.
(struct cbl_alphabet_t): Same.
(struct cbl_perform_tgt_t): Same.
(struct cbl_nameloc_t): Same.
(class name_queue_t): Same.
(tee_up_name): Same.
(symbol_field_location): Same.
(enum cbl_call_convention_t): Remove
(class current_tokens_t): Use cbl_loc_t.
(current_call_convention): Same.
(gcc_location_set): Same.
* util.cc (class cdf_directives_t): Use cbl_loc_t.
(cdf_unreachable): Define as gcc_unreachable.
(cdf_literalize): Do not handle location.
(cdf_file): New function.
(cdf_file_index): Same.
(cdf_file_name): Same.
(cdf_add_field): Same.
(cbl_field_t::encode_numeric): Remove unused parameter.
(cbl_field_t::report_invalid_initial_value): Use cbl_loc_t.
(match_proc): Namespace for local prototype-verification functions.
(DUMP_PROCEDURE_CALLS): Guard macro for debug function.
(procedure_calls_dump): New function to show uses of PERFORM.
(gcc_location_set_impl): Use cbl_loc_t.
(gcc_location_set): Same.
(class temp_loc_t): Same.
(error_msg): Same.
(warn_msg): Same.
(ydfdebug): Same.
(cobol_set_debugging): Same.
(cbl_unimplemented_at): Same.
* util.h (cbl_err): Remove declaration.
(cbl_errx): Same.
(cdf_push): Same.
(cdf_push_call_convention): Same.
(cdf_push_current_tokens): Same.
(cdf_push_dictionary): Same.
(cdf_push_enabled_exceptions): Same.
(cdf_push_source_format): Same.
(cdf_pop): Same.
(cdf_pop_call_convention): Same.
(cdf_pop_current_tokens): Same.
(cdf_pop_dictionary): Same.
(cdf_pop_source_format): Same.
(cdf_pop_enabled_exceptions): Same.
libgcobol/ChangeLog:
* posix/shim/open.cc (defined): Honor _GNU_SOURCE and _POSIX_C_SOURCE.
Marek Polacek [Fri, 12 Jun 2026 19:46:10 +0000 (15:46 -0400)]
c++: bogus error with meta::members_of [PR125770]
mark_used produces the
use of built-in parameter pack '__integer_pack' outside of a template
error even when called with tf_none (in this case, from
resolve_type_of_reflected_decl).
I played with the idea of skipping all fndecl_built_in_p functions
in resolve_type_of_reflected_decl but it doesn't seem to make any
reasonable difference on compile time.
PR c++/125770
gcc/cp/ChangeLog:
* decl2.cc (mark_used): Check complain & tf_error before giving
an error.
Marek Polacek [Mon, 1 Jun 2026 21:32:38 +0000 (17:32 -0400)]
c++: ICE in tsubst_expr with comma expr [PR125539]
Here we crash because a TARGET_EXPR gets into tsubst_expr with something
like:
template<typename>
constexpr static A a = (A{}, A{});
Pre r14-4796, when we had tsubst_copy, we didn't crash because we had
an early exit when substituting with args=NULL_TREE. tsubst_expr
deliberately doesn't have that early exit.
Note that
template<typename>
constexpr static A a = A{};
works because expand_aggr_init_1 sees a COMPOUND_LITERAL_P and does the
early exit without calling expand_default_init. But with a COMPOUND_EXPR
we don't take that path.
The TARGET_EXPR is created via check_initializer -> build_aggr_init_full_exprs
-> build_aggr_init -> expand_aggr_init_1 -> expand_default_init -> ocp_convert
-> build_cplus_new -> build_target_expr. I tried adjusting the big
and ugly check in check_initializer before build_aggr_init_full_exprs
but that didn't work out. I also tried avoiding the call to ocp_convert
but that broke some reflection tests.
We can fix this by calling perform_implicit_conversion in a template to
create an IMPLICIT_CONV_EXPR, then build_cplus_new won't create a TARGET_EXPR.
PR c++/125539
gcc/cp/ChangeLog:
* cvt.cc (ocp_convert): In a template, always call
perform_implicit_conversion. Pass flags to
perform_implicit_conversion_flags.
* decl.cc (check_initializer): Remove a call to
build_implicit_conv_flags.
This patch fixes a mistake in gather/scatter discovery. In the third
"phase" we check for a larger offset type for the needed scaling but
fail to let the vectorizer know it. The patch just sets
supported_offset_vectype and also skips costing an instruction in case
the necessary conversion is a nop.
Bootstrapped and regtested on x86, power10, and aarch64.
Regtested on riscv64.
Regards
Robin
PR tree-optimization/125516
gcc/ChangeLog:
* tree-vect-data-refs.cc (vect_gather_scatter_fn_p): Set
supported_offset_vectype.
* tree-vect-stmts.cc (vectorizable_store): Skip nop conversions
when costing scatters.
(vectorizable_load): Ditto for gathers.
gcc/testsuite/ChangeLog:
* gcc.target/riscv/rvv/autovec/pr125516.c: New test.
Abhishek Kaushik [Tue, 16 Jun 2026 13:43:50 +0000 (07:43 -0600)]
[PATCH] tree-optimization: Query ranger on edge for niter bound expressions
determine_value_range only queried ranger when VAR was an SSA_NAME.
However, number_of_iterations_exit_assumptions calls
expand_simple_operations on the IV bases before asking for the range.
This can expose simple GENERIC expressions, including conversions, even
when the original value had an SSA_NAME with useful range information.
Ranger can analyze such expressions at a program point or edge, so query
it for integral bound expressions on the loop preheader edge rather than
restricting the query to SSA_NAMEs. This lets niter analysis recover the
narrower range for value-preserving conversions such as uint8_t to int,
while still handling non-value-preserving conversions and wrapping
expressions conservatively.
The new test covers a direct converted uint8_t bound, a masked uint16_t
bound whose useful range comes from ranger, a guarded uint16_t bound
whose range is context-sensitive, a non-value-preserving uint8_t to
int8_t conversion, and a wrapping unsigned expression.
Bootstrapped and regression tested on aarch64-unknown-linux-gnu.
gcc/ChangeLog:
* tree-ssa-loop-niter.cc (determine_value_range): Query ranger
for integral expressions, not only SSA_NAMEs. Query ranges on the
loop preheader edge.
gcc/testsuite/ChangeLog:
* gcc.target/aarch64/sve2/niter-convert-range.c: New test.
Kyrylo Tkachov [Tue, 16 Jun 2026 10:16:42 +0000 (12:16 +0200)]
vect: consult the alias oracle for unanalyzable BB SLP dependences
vect_slp_analyze_data_ref_dependence conservatively reported a dependence
whenever the classical (affine) data-dependence test returned chrec_dont_know,
e.g. when one of the accesses has a non-affine or runtime array subscript. In
the BB SLP region check this is overly pessimistic: the unanalyzable subscript
says nothing about whether the two references can actually alias, and the alias
oracle can frequently still prove they cannot (distinct restrict parameters,
distinct non-escaping objects, and so on). When that happens a perfectly good
SLP group is torn down. The motivating case is the deal.II
VectorizedArray<double,N> reciprocal in SPEC CPU 2026 766.femflow_r.
The store-sink and load-hoist walkers already fall back to the alias oracle
(stmt_may_clobber_ref_p_1 / ref_maybe_used_by_stmt_p) for statements that have
no single recorded data reference. Extend that fallback to the chrec_dont_know
case: vect_slp_analyze_data_ref_dependence now returns a three-way result
(chrec_known when the references are provably independent, chrec_dont_know when
the affine test cannot analyze them, and the dependence otherwise) so each
caller can tell "unknown" apart from "dependent", and on "unknown" runs the same
oracle query it already uses for the no-data-reference case, with the TBAA
setting appropriate to what is being moved: no TBAA when sinking a store (a
moving store may change the dynamic type), TBAA when hoisting a load.
On the new gcc.dg/vect/bb-slp-dep-oracle.c the 8-lane reciprocal group is torn
down and emitted scalar without the patch and vectorizes to four vector
divides with it.
Bootstrapped and tested on aarch64-none-linux-gnu.
* tree-vect-data-refs.cc (vect_slp_analyze_data_ref_dependence):
Return a three-way tree result (chrec_known when independent,
chrec_dont_know when the affine test cannot analyze the pair, the
dependence otherwise) instead of a bool.
(vect_slp_analyze_store_dependences): Resort to the alias oracle on
an unknown dependence as well as on a missing data reference; a
store is being moved so do not use TBAA.
(vect_slp_analyze_load_dependences): Likewise on the load-hoist
paths, using TBAA as a load is being hoisted; also record that the
ao_ref has been initialized in check_hoist.
Robin Dapp [Tue, 16 Jun 2026 12:54:48 +0000 (06:54 -0600)]
[PATCH] RISC-V: Fix splitter [PR125670].
Hi,
In the PR we ICE during vsetvl, expecting a register in the VL operand
slot which only contains an immediate 4. Non-VLMAX insns with immediate
length have a NULL_RTX in that slot.
However, during a split, we erroneously use operand[5] instead of
operand[6]. operand[5] is the mask policy and happened to be "1".
"1" indicates a VLMAX insn in the avl_type operand. This caused the
wrong turn in vsetvl.
The patch just corrects the operand number.
Regtested on rv64gcv_zvl512b. Going to wait for the CI.
Regards
Robin
PR target/125670
gcc/ChangeLog:
* config/riscv/autovec-opt.md: Use avl_type operand number.
gcc/testsuite/ChangeLog:
* gcc.target/riscv/rvv/autovec/pr125670.c: New test.
Rainer Orth [Tue, 16 Jun 2026 12:04:04 +0000 (14:04 +0200)]
testsuite: Limit shared memory use of coarray tests [PR125584]
As discussed in PR testsuite/125584, the gfortran.dg/coarray tests with
-lcaf_shmem require excessive amounts of backing store for their shared
memory images: for 64-bit tests, each test uses 4 GB images.
On Solaris, the mapped files reside in /tmp/.LIBRT/SHM, which is tmpfs.
During parallel testing, several such tests can run in parallel, leading
to excessive VM use, in particular on targets like Solaris that don't
use lazy allocation. This is unacceptable since the testsuite defaults
need to be safe for all targets.
The tests PASS just fine when limiting GFORTRAN_SHARED_MEMORY_SIZE to 2M
instead. To allow for testing with either the default or a different
size, setting GCC_TEST_RUN_EXPENSIVE to a non-empty string allows to
override this.
Tested on amd64-pc-solaris2.11 and x86_64-pc-linux-gnu.
Richard Biener [Thu, 11 Jun 2026 13:51:02 +0000 (15:51 +0200)]
tree-optimization/125730 - avoid losing track of base in IVOPTs
The PR highlights several places where IVOPTs gets lost in tracking
an appropriate base to use for a TARGET_MEM_REF. The following
addresses the initial place, enough to fix the bug, but already
showing difficulties in avoiding fallout. The first place is
alloc_iv where we convert the nice pointer IV to an unsigned integer
before applying affine canonicalization.
Removing that resolves the PR and with the fold_plusminus_mult_expr
change tests without regressions on x86_64.
PR tree-optimization/125730
* tree-ssa-loop-ivopts.cc (alloc_iv): Do not convert pointer
IVs to unsigned before canonicalizing.
Richard Biener [Mon, 15 Jun 2026 09:29:08 +0000 (11:29 +0200)]
Improve fold_plusminus_mult_expr for 64bit and larger types
The following enhances fold_plusminus_mult_expr to catch
a * -4U + b * 4 and factor it as (a * -1u + b) * 4. This
does not work currently because we're using HOST_WIDE_INT
arithmetic. Switch that to wide_int, which makes the
folding apply more consistently.
For gcc.dg/loop-versioning-13.c the heuristics in
gimple-loop-versioning.cc get confused as they fail to
truncate some computations. I did not inverstigate further.
* fold-const.cc (fold_plusminus_mult_expr): Use
wide_int for the case of two INTEGER_CST multiplicands.
* gcc.dg/loop-versioning-13.c: XFAIL one transfor for ilp32.
* gcc.dg/pr109393.c: Remove XFAIL for ilp32.
Kyrylo Tkachov [Tue, 16 Jun 2026 07:58:28 +0000 (00:58 -0700)]
aarch64: Fix wrong code for high-64-zero Advanced SIMD constants [PR125794]
r17-1491-gf152cf1734f808 (PR113926) taught aarch64_simd_valid_imm to
materialize a 128-bit Advanced SIMD MOV constant whose high 64 bits are
zero with a 64-bit MOVI/FMOV, which zeroes the upper half of the
register. It records this with simd_immediate_info::width == 64
(output_width).
However, when the low 64 bits are not themselves a valid Advanced SIMD
(MOVI/MVNI/FMOV) immediate, the function fell through to the SVE
immediate forms (aarch64_sve_valid_immediate). Those use a replicating
"mov zN.<T>, #imm", which sets the whole vector, including the high 64
bits that were required to be zero, to the repeated low-64-bit value.
For e.g. the V4SI constant { 0, 1, 0, 0 } this emitted
instead of the intended { 0, 1, 0, 0 }, producing wrong code.
Fix it by not falling through to the SVE forms when output_width is set:
such a constant must be formed by a 64-bit Advanced SIMD MOVI/FMOV
(handled by the Advanced SIMD and floating-point paths just above) or
not at all, in which case the caller materializes it some other way
(e.g. a literal-pool load), which is the pre-r17-1491 behavior for these
constants.
The PR113926 optimization is unaffected: it only applies when the
Advanced SIMD or floating-point path accepts the low 64 bits, and those
still return true before the new check.
Bootstrapped and regression-tested on aarch64-linux-gnu.
Pushing to trunk.
PR target/125794
* config/aarch64/aarch64.cc (aarch64_simd_valid_imm): Do not fall
through to the replicating SVE immediate forms for a 128-bit
Advanced SIMD constant whose high 64 bits are zero (output_width
!= 0).
gcc/testsuite/ChangeLog:
PR target/125794
* gcc.target/aarch64/sve/pr125794.c: New test.
There are certain patterns that are not recognized by the method
optimize_spaceship. For example,
a == b ? 0 : (a > b) : 1 : -1;
is rightly recognized as spaceship operator. But
a <= b ? -(a < b) : 1 or
a >= b ? (a > b) : -1
is not being currently recognized.
This patch recognizes such patterns and chooses to emit the spaceship
optab if target supports it, which improves code-generation for such
targets.
gcc/ChangeLog:
PR tree-optimization/59429
* match.pd: New match patterns to recognize spaceship variants.
* tree-ssa-math-opts.cc (gimple_spaceship): Match function declaration.
(match_spaceship): New function to recognize spaceship given phi node.
(math_opts_dom_walker::after_dom_children): Add match_spaceship check.
gcc/testsuite/ChangeLog:
PR tree-optimization/59429
* lib/target-supports.exp (check_effective_target_spaceship): Add new
proc for spaceship optab. x86, aarch64 and s390 included.
* gcc.dg/spaceship_int_variants.c: New test.
* gcc.dg/spaceship_uint_variants.c: New test.
* gcc.dg/spaceship_mixed_variants.c: New test.
Zhongyao Chen [Fri, 12 Jun 2026 03:59:56 +0000 (11:59 +0800)]
RISC-V: Adjust testcase asm check for vx-[5|6]-i[8|16].c
After commit 9f8409f2e2c, SLP discovery can retry swapped operands for
commutative parents before falling back to an external scalar.
These tests can be vectorized again, so update asm check.
Roger Sayle [Mon, 15 Jun 2026 19:09:55 +0000 (20:09 +0100)]
i386: Tweak cost of SSE fabs/fneg in ix86_insn_cost.
This patch fixes a poor interaction between the splitters for SSE
floating point abs/neg in the i386 backend, and the late-combine pass.
Before reload, these patterns exist as a PARALLEL containing the USE
of a value (pseudo) holding the sign-bit. Currently late-combine
propagates this sign-bit mask from the constant pool, changing the
USE of a REG to the USE of a MEM. This unCSE is reasonable if this
MEM is used only once, but less than optimal if this MEM is accessed
many times.
The problem is that this USE doesn't currently have a cost in
ix86_insn_cost, so propagating this load from memory into the USE
makes if free (to combine's profitable replacement calculation).
This patch improve things by providing a nominal cost for USEs of
MEM.
As an example, consider the following function:
float x, y, z;
void foo()
{
x = -x;
y = -y;
z = -z;
}
Currently with -O2 GCC generates three loads from the constant pool:
Note this is one more instruction, but code size is smaller and
the total cost (as calculated by the i386 backend) is lower.
For a single neg/abs the memory address is still propagated.
2026-06-15 Roger Sayle <roger@nextmovesoftware.com>
gcc/ChangeLog
* config/i386/i386.cc (ix86_insn_cost): Add a suitable penalty
for USE of a MEM in a PARALLEL (for *<absneg>[sd]f2_1 splitter).
gcc/testsuite/ChangeLog
* gcc.target/i386/fabsneg-2.c: New test case.
When a color group spans the whole register file (size == 32), as can
happen for a heavily unrolled, vectorized loop, "1U << 32" is undefined
and evaluates to 1 on AArch64 hosts, so the expression sets no bits at all.
The 32 FPRs of the group are therefore not recorded as allocated.
Subsequent colors (and broaden_colors) then reuse those registers, which
breaks the invariant that distinct colors receive disjoint FPRs.
In PR125795 this let the loop-invariant TBL permute index, which is live
across the whole loop, share v28 with the LD2 tuple destinations, so the
index was clobbered mid-loop and the loop produced wrong results.
Fix this by using a 64-bit shift base: unsigned long long is at least
64 bits on every host, so "1ULL << 32" is well-defined.
best + size <= 32 is guaranteed by the candidate search, which the patch
also asserts, so the result still fits in the 32-bit m_allocated_fprs
When the full-width group can no longer be hidden, allocate_colors correctly
fails to find a register for the other color and the region is left to the
real register allocator, matching -mearly-ra=none.
Bootstrapped and tested on aarch64-none-linux-gnu.
Pushing to trunk and later to the branches after testing.
Jose E. Marchesi [Mon, 15 Jun 2026 14:02:57 +0000 (16:02 +0200)]
a68: handle duplicated modes in module imports
In principle all the modes emitted by the module exporter are
deduplicated. However, in certain cases in which the same unions
result from unraveling, duplicates may occur.
This patch makes a68_open_packet to deduplicate duplicated modes from
the same moif.
Signed-off-by: Jose E. Marchesi <jemarch@gnu.org>
gcc/algol68/ChangeLog
* a68-imports.cc (INCLUDE_VECTOR): Define.
(a68_open_packet): Deduplicate within-moif modes while importing.
gcc/testsuite/ChangeLog
* algol68/execute/modules/module26.a68: New file.
* algol68/execute/modules/program-26.a68: New test.
[PATCH] match: For nonnegative X and Y, relax condition on X % Y < Y to true [PR125737]
From ae75421fd6c7d50e5b1e9aafea2ae3cbcd4ebc1c Mon Sep 17 00:00:00 2001
From: Kael Andrew Alonzo Franco <kaelfandrew@gmail.com>
Date: Sun, 14 Jun 2026 06:28:01 -0400
Subject: [PATCH] match: For nonnegative X and Y, relax condition on X % Y < Y to true [PR125737]
tree_expr_nonnegative_p covers TYPE_UNSIGNED (type) or when X and Y are known to be nonnegative.
Bootstrapped and tested on x86_64-pc-linux-gnu
PR tree-optimization/125737
gcc/ChangeLog:
PR tree-optimization/125737
* match.pd: Use tree_expr_nonnegative_p for X % Y < Y to true.
gcc/testsuite/ChangeLog:
PR tree-optimization/125737
* gcc.dg/pr125737.c: New test.
Andrew Pinski [Fri, 12 Jun 2026 18:40:35 +0000 (11:40 -0700)]
phiopt: reorganize factoring/cselim-limited for phiopt
This is in preparation for adding factoring out loads
for phiopt were we want to loop over all 3 factoring
cases if one of them made a change so that a load elimination
might allow for a store elimination.
This moves the cs-elim limited loop into the factoring out
operation into the same loop.
Bootstrapped and tested on x86_64-linux-gnu.
gcc/ChangeLog:
* tree-ssa-phiopt.cc (factor_out_all): New function.
(pass_phiopt::execute): Call factor_out_all isntead
of factor_out_conditional_operation and
cond_if_else_store_replacement_limited.
Signed-off-by: Andrew Pinski <andrew.pinski@oss.qualcomm.com>
store_integral_bit_field: Graceful fallback when SUBREG narrowing fails [PR123754]
The multi-word narrowing path in store_integral_bit_field uses
simplify_gen_subreg followed by gcc_assert (op0). The symmetric path
in extract_integral_bit_field was switched to force_subreg, but the
store side was deliberately left on simplify_gen_subreg because op0
is an lvalue. When the subreg simplification fails (e.g. a vector
op0 punned through an int mode whose word-aligned subregs are rejected
by validate_subreg, as happens for V8SI on -mbig-endian aarch64),
the assert fires.
The avoid-store-forwarding pass (-favoid-store-forwarding) triggers
this: it routes such a vector op0 through store_integral_bit_field.
Replace the assert with a graceful fallback to store_split_bit_field,
mirroring the cross-word branch immediately above. No change for
inputs where the narrowing succeeds.
Tested on AArch64, x86-64 and PowerPC BE.
PR rtl-optimization/123754
gcc/ChangeLog:
* expmed.cc (store_integral_bit_field): When the SUBREG
narrowing fails, defer to store_split_bit_field instead of
asserting.
gcc/testsuite/ChangeLog:
* gcc.target/aarch64/pr123754.c: New test.
* gcc.target/aarch64/avoid-store-forwarding-be-2.c: New test.
Pan Li [Thu, 4 Jun 2026 14:47:49 +0000 (22:47 +0800)]
RISC-V: Add testcase for unsigned scalar SAT_MUL form 12
The form 12 of unsigned scalar SAT_MUL has supported from
the previous change. Thus, add the test cases to make sure
it works well.
gcc/testsuite/ChangeLog:
* gcc.target/riscv/sat/sat_arith.h: Add test helper macros.
* gcc.target/riscv/sat/sat_u_mul-13-u16.c: New test.
* gcc.target/riscv/sat/sat_u_mul-13-u32.c: New test.
* gcc.target/riscv/sat/sat_u_mul-13-u64.c: New test.
* gcc.target/riscv/sat/sat_u_mul-13-u8.c: New test.
* gcc.target/riscv/sat/sat_u_mul-run-13-u16.c: New test.
* gcc.target/riscv/sat/sat_u_mul-run-13-u32.c: New test.
* gcc.target/riscv/sat/sat_u_mul-run-13-u64.c: New test.
* gcc.target/riscv/sat/sat_u_mul-run-13-u8.c: New test.
RISC-V: Run smart multilib match even when generic matcher picked a dir
The generic textual matcher in gcc.cc:set_multilib_dir does not
understand RISC-V arch supersetting and treats MULTILIB_DEFAULTS
entries as if they were on the command line via default_arg(). When
the user passes a -march= that is a superset of one of MULTILIB_OPTIONS'
arches but does not textually match, default_arg can rescue the wrong
entry and pick a multilib that is not the closest match.
riscv_compute_multilib used to early-return whenever multilib_dir was
already set, accepting that incorrect generic pick. Drop the early
return and run the match-score-based selection unconditionally.
There is no need to fall back to the generic-matched multilib_dir
after the smart matcher runs: the default "." multilib is parsed into
multilib_infos with the compiler's default arch/abi, so the smart
matcher handles every case the generic matcher can reach. If it
still returns NULL the request is genuinely incompatible with all
configured multilibs and riscv_multi_lib_check fires the proper
"Cannot find suitable multilib" diagnostic instead of silently
linking against incompatible default-arch libraries.
With the pre-fix driver, "-march=rv64g_zba_zcmp_zcmt -mabi=lp64f"
selects the rv64gc multilib (textual default rescue); after the fix
the smart matcher correctly picks the rv64g_zcmp_zcmt multilib.
gcc/ChangeLog:
* common/config/riscv/riscv-common.cc (riscv_select_multilib):
Don't set riscv_no_matched_multi_lib here; let the caller own
the flag.
(riscv_compute_multilib): Drop the early return that accepted
the generic-matched multilib_dir; always run the smart matcher
and set riscv_no_matched_multi_lib when it finds no candidate.
Filip Kastl [Mon, 15 Jun 2026 12:20:58 +0000 (14:20 +0200)]
toplev: Ask for 128MB stack when compiled with ASAN [PR124206]
64MB stack is not enough for running
gcc/testsuite/gcc.c-torture/compile/limits-exprparen.c with an
ASAN-instrumented GCC. Ask for more stack if GCC was compiled with ASAN
instrumentation.
PR sanitizer/124206
gcc/ChangeLog:
* gcc.cc (driver::global_initializations): Ask for 128MB stack
instead of just 64MB when __SANITIZE_ADDRESS__ is defined.
* toplev.cc (toplev::main): Ditto.
Rainer Orth [Mon, 15 Jun 2026 12:18:59 +0000 (14:18 +0200)]
libgcc: Fix _mcount on 32-bit Solaris/x86 [PR38239]
Profiling on 32-bit Solaris/x86 has been broken since
Save call-clobbered registers in _mcount on 32-bit Solaris/x86 (PR target/38239)
https://gcc.gnu.org/pipermail/gcc-patches/2016-March/444175.html
This was only recently noticed when setting up a Solaris/i386 binutils
buildbot.
Since internal_mcount is a regular function on Solaris, the selfpc and
frompcindex args need to pushed to the stack. Besides, the patch fixes
a couple of warnings in gmon.c
Bootstrapped withoug regressions on i386-pc-solaris2.11. With this
patch, the binutils gprof tests PASS.
ld: fatal: relocation error: R_SPARC_32: file c_lto_toplevel-extended-asm-1_1.o: symbol .text (section): value 0x100001340 does not fit
The same error has been present with gld all along, but was only
recently introduced in Solaris ld as discussed in binutils PR ld/25802.
It doesn't occur on Linux/sparc64 which uses a different text start
address.
Tested on sparcv9-sun-solaris2.11, sparc64-unknown-linux-gnu, and
x86_64-pc-linux-gnu.
Richard Biener [Mon, 15 Jun 2026 08:40:10 +0000 (10:40 +0200)]
Fix VEC_COND_EXPR matching with inverted compare
The following fixes detecting of VEC_COND_EXPR <cmp, {0,..}, {-1,...}>
which we recognize in ovce_extract_ops by inverting 'cmp'. But
after checking that the false value is {-1,...} we then continue
verifying it is also {0,...} which it of course is not. Fixed
by checking the true value in that case.
* tree-ssa-reassoc.cc (ovce_extract_ops): Fixup
false value matching for the inverted comparison case.
In 2-byte loops, don't force scratch regs into constraint "w".
Use a C function for asm output. It will emit SBIW if possible.
gcc/
* config/avr/avr-protos.h (avr_out_delay_loop): New proto.
* config/avr/avr.cc (avr_out_delay_loop): New function.
(avr_adjust_insn_length) [ADJUST_LEN_DELAY_LOOP]: Handle case.
(avr_expand_delay_cycles): Overhaul. Allow loop counts of
zero; they represent a power of 2.
* config/avr/avr.md (adjust_len) [delay_loop]: Add.
(*delay_cycles_1): Use avr_out_delay_loop for asm out.
(*delay_cycles_4): Same.
(*delay_cycles_3): Same.
(*delay_cycles_2): Same. Relax constraints to "d".
Use two QImode scratch regs instead of one HImode one.
Jose E. Marchesi [Mon, 15 Jun 2026 07:38:57 +0000 (09:38 +0200)]
a68: do not pub non-publicized indicants in exports
Even non-publicized modes were being added to the module interfaces.
This was because for all other kind of taxes (identifiers, operator
indicants, etc) the PUBLICIZED tax attribute is set at taxes
collection time, but for mode indicants the flag shall be set at
extraction time.
Signed-off-by: Jose E. Marchesi <jemarch@gnu.org>
gcc/algol68/ChangeLog
* a68-parser-extract.cc (a68_extract_indicants): Set PUBLICIZED in
the new tags for mode indicants.
* a68-parser.cc (a68_new_tag): Initialize PUBLICIZED to false.
Andrew Pinski [Sun, 14 Jun 2026 20:19:47 +0000 (13:19 -0700)]
match: Fix up `(~x) >> (type)x` pattern for truncation [PR125790]
I missed this during the review and when I suggest adding support
for the cast. But a truncation of the shifter operand the value
could be defined.
Since the front-end adds a cast to unsigned int, we need to split
pr125707.c into two and xfail the long case and change it to
`long long` so it would xfail for ilp32 [and llp64il32] targets.
PR tree-optimization/125790
gcc/ChangeLog:
* match.pd (`(~x)>>x`): Reject truncation of shifter.
gcc/testsuite/ChangeLog:
* gcc.dg/pr125707.c: Move the long over to pr125707-1.c.
* gcc.dg/pr125707-1.c: New test; xfailed.
* gcc.dg/pr125790-1.c: New test.
Signed-off-by: Andrew Pinski <andrew.pinski@oss.qualcomm.com>
Monk Chiang [Wed, 21 Jan 2026 03:29:13 +0000 (11:29 +0800)]
RISC-V: Ensure Zicfilp lpad 4-byte alignment
Add LABEL_ALIGN to align non-local goto target labels, removing the
explicit gen_lpad_align () call. Extend gpr_save to emit the full
CFI guard (.p2align 2, .option push/norelax/norvc, call, .option pop,
lpad 0) when Zicfilp is active. Add lpad_align before the thunk
entry lpad.
gcc/ChangeLog:
* config/riscv/riscv-zicfilp.cc (rest_of_insert_landing_pad):
Remove gen_lpad_align before non-local goto labels; LABEL_ALIGN
now handles their alignment. Simplify gpr_save handling: the
pattern now outputs the full CFI sequence itself.
* config/riscv/riscv.cc (riscv_output_mi_thunk): Add
gen_lpad_align before entry lpad.
* config/riscv/riscv.h (LABEL_ALIGN): New macro; ensures 4-byte
alignment for Zicfilp non-local goto target labels.
* config/riscv/riscv.md (gpr_save): When is_zicfilp_p, output
.p2align 2, .option push/norelax/norvc, call, .option pop, lpad 0.
gcc/testsuite/ChangeLog:
* g++.target/riscv/zicfilp-thunk.C: New test.
* gcc.target/riscv/zicfilp-func-entry.c: New test.
* gcc.target/riscv/zicfilp-gpr-save.c: New test.
* gcc.target/riscv/zicfilp-nonlocal-goto.c: New test.
Monk Chiang [Wed, 10 Jun 2026 01:34:44 +0000 (18:34 -0700)]
RISC-V: Add Zicfilp LPAD protection for setjmp and indirect_return
Add call-site LPAD insertion for two cases:
1. setjmp / __attribute__((returns_twice)) calls, which may return a
second time via longjmp.
2. A new "indirect_return" attribute for functions that may return to
an unexpected address.
Detection uses riscv_call_needs_lpad_p() at expand time. When needed,
call_internal_cfi / call_value_internal_cfi emit .p2align 2,
.option push/norelax/norvc, the call, .option pop, and lpad 0 as a
single insn. This prevents the assembler (c.jal) or linker (jal
relaxation) from shifting the return address off the lpad.
Indirect calls to returns_twice or indirect_return functions are not
covered.
Changes in v4:
- Revert the v2 change, make indirect_return a type attribute.
Changes in v3:
- Fix coding style in riscv.cc
Changes in v2:
- Change indirect_return from a type attribute to a decl attribute,
consistent with returns_twice.
- Add length attributes to call_internal_cfi and call_value_internal_cfi
to reflect the multi-instruction sequence.
- Copy the explaining comment to call_value_internal_cfi.
gcc/ChangeLog:
* config/riscv/riscv-protos.h (riscv_call_needs_lpad_p): Declare.
* config/riscv/riscv.cc (riscv_gnu_attributes): Register new
indirect_return attribute for function declarations.
(riscv_call_needs_lpad_p): New function.
* config/riscv/riscv.md (call_internal_cfi): New insn pattern with
length attribute.
(call_value_internal_cfi): Likewise for call-with-return-value,
with comment and length attribute.
(define_expand "call"): Emit call_internal_cfi when
riscv_call_needs_lpad_p returns true.
(define_expand "call_value"): Likewise.
gcc/testsuite/ChangeLog:
* gcc.target/riscv/zicfilp-indirect-return.c: New test.
* gcc.target/riscv/zicfilp-setjmp.c: New test.
projects / thirdparty / gcc.git / log
