1. The upstream sources do not, in general, support the range
of Darwin versions covered by GCC.
In order to support versions back to Darwin17, at least we
provide definitions for missing macro values and ensure that
headers are only conditionally included where they apply.
2. GCC does not support the clang __builtin_os_log_format and
therefore must fall back to older reporting methods.
3. Finally, we address a FIXME (for missing Blocks support)
used implement the search for dyld on macOS >= 13 with the
dyld_shared_cache_iterate_text() interface which requires an
(Apple) Block closure as a parameter.
If the compiler supports blocks (__BLOCKS__ is defined) then we
use the upstream implementation. If not, then we synthesize the
equivalent code-gen manually.
Implements submdspan_canonicalize_slices as described in P3663 and adds
it to the std module.
There's one deviation from the standard. Doesn't (under all
circumstances) require:
0 <= begin[k] <= end[k] <= exts.extent(k)
where the k-th slice range is [begin[k], end[k]). Instead, it requires
that the k-th slice ranges is contained in the k-th extent interval. If
the slice range is empty, then that condition is always satisfied, even if
begin[k] == end[k] > exts.extent(k)
The deviation is that we enforce the above inequality through
preconditions. This is analogous to what the standard requires if
begin[k] is a constant wrapper.
PR libstdc++/110352
libstdc++-v3/ChangeLog:
* include/std/mdspan (submdspan_canonicalize_slices): New
function.
* src/c++23/std.cc.in (submdspan_canonicalize_slices): Add.
* testsuite/23_containers/mdspan/submdspan/submdspan_canonicalize_slices.cc: New test.
* testsuite/23_containers/mdspan/submdspan/submdspan_canonicalize_slices_neg.cc: New test.
Reviewed-by: Tomasz Kamiński <tkaminsk@redhat.com> Signed-off-by: Luc Grosheintz <luc.grosheintz@gmail.com>
Kito Cheng [Tue, 2 Dec 2025 13:40:06 +0000 (06:40 -0700)]
RISC-V: Remove unused placeholder_p parameter from add_function
The placeholder_p parameter of function_builder::add_function is always
passed as false. This was inherited from the AArch64 implementation but
is unnecessary for RISC-V.
Jakub Jelinek [Tue, 2 Dec 2025 13:36:49 +0000 (14:36 +0100)]
c++: Diagnose taking addresses of hard reg vars in their initializers [PR122860]
DECL_HARD_REGISTER is set only in cp_finish_decl together with
set_user_assembler_name. If user attempts to take address of
such a var later, cxx_mark_addressable diagnoses it.
But if as in the following testcase the address is taken in its
initializer, we just ICE during expansion.
The following patch fixes it by emitting errors if TREE_ADDRESABLE
at the point we'd otherwise set DECL_HARD_REGISTER on it.
2025-12-02 Jakub Jelinek <jakub@redhat.com>
PR c++/122860
* decl.cc (make_rtl_for_nonlocal_decl): Diagnose taking address
of a hard register decl in its initializer.
(cp_finish_decl): Likewise.
Jason Merrill [Tue, 2 Dec 2025 12:48:39 +0000 (18:18 +0530)]
c++: alias template parm conv and redecl [PR122171]
Here when substituting BITS into poly_table convert_template_argument adds
an IMPLICIT_CONV_EXPR to represent the conversion to the alias template
parameter. In r16-4115 I extended that to value-dependent arguments as well
as type-dependent, in case the conversion turns out to be narrowing.
tsubst_expr needs the same change so maybe_update_decl_type doesn't
replace the IMPLICIT_CONV_EXPR with a NOP_EXPR.
The do_auto_deduction change is to avoid a regression in nontype-auto21.C
when the first test is changed from uses_template_parms (as it was in
convert_template_argument) to dependent_type_p; this mattered because we
were failing to resolve the auto return type before deducing the auto
non-type parameter type from helper<token>::c. Many other places that call
resolve_nondeduced_context similarly then call mark_single_function.
PR c++/122171
PR c++/112632
gcc/cp/ChangeLog:
* pt.cc (dependent_implict_conv_p): Split out...
(convert_template_argument): ...from here.
(tsubst_expr) [IMPLICIT_CONV_EXPR]: Use it.
(do_auto_deduction): Call mark_single_function.
Nathaniel Shead [Mon, 1 Dec 2025 23:24:01 +0000 (10:24 +1100)]
c++/modules: Remove incorrect is_import check in using-directives
When I wrote this check in r16-5811 I was thinking of checking if the
using-directive was imported, but this just checks if the target
namespace was imported, which is not what we want. We don't build deps
to see if the using-directive itself was imported, so just remove the
check. I haven't been able to come up with a testcase this breaks but
it still seems reasonable to adjust.
Tamar Christina [Tue, 2 Dec 2025 10:55:51 +0000 (10:55 +0000)]
vect: don't hoist conditional loads above their condition [PR122868]
The example in the PR
#include <vector>
std::vector<bool> x, y;
int main() { return x == y; }
now vectorizes but the attributes on std::vector indicate that the vector is
aligned to the natural vector alignment. In C this is equivalent to the
testcase
int f (int a[12], int b[12], int n)
{
a = __builtin_assume_aligned (a, 16);
b = __builtin_assume_aligned (b, 16);
for (int i = 0; i < n; i++)
{
if (b[i] == 0)
return 0;
if (a[0] > b[i])
return 1;
}
return 2;
}
Here the load a[0] is loop invariant, and the vectorizer hoists this out of the
loop into the pre-header. For early break this isn't safe to do as a[0] is
conditionally valid based on the conditions in the block preceding it. As such
we need some guarantee that the load is valid before we can hoist it or the load
needs to be unconditional (e.g. in the loop header block).
Conceptually alignment peeling can provide this guarantee since making it
through the prologue means the invariant value was loaded at least once and so
we know the address is valid. At the moment however there's no real defined
order between how GCC inserts conditions in the pre-header, so having tried to
change the order a few times the load always ends up before the prologue. So
for now I marked it as a missed optimization.
Since we still can hoist invariant loads if in the header, I didn't change
LOOP_VINFO_NO_DATA_DEPENDENCIES since that would be global and instead I
modified the usage site of LOOP_VINFO_NO_DATA_DEPENDENCIES.
PR tree-optimization/122868
* gcc.dg/vect/vect-early-break_140-pr122868_1.c: New test.
* gcc.dg/vect/vect-early-break_140-pr122868_2.c: New test.
* gcc.dg/vect/vect-early-break_140-pr122868_3.c: New test.
* gcc.dg/vect/vect-early-break_140-pr122868_4.c: New test.
Alexandre Oliva [Tue, 2 Dec 2025 02:58:43 +0000 (23:58 -0300)]
hard-reg-set: use ctz for iteration
Simplify the HARD_REG_SET iteration, using ctz and avoiding some
unnecessary operations.
for gcc/ChangeLog
* hard-reg-set.h (hard_reg_set_iter_init): Drop unnecessary
increment of min.
(hard_reg_set_iter_set): Use ctz_hwi, and compute
word-advanced regno from word_no.
(hard_reg_set_iter_next): Only clear the cached LSB.
MIPS: Add Allegrex support for madd/msub instructions.
Tweaking mul_acc_si/mul_sub_si was necessary to fix the implicit
assumption that CPUs with madd/msub support also support MUL3
multiplication. This disables the non-existent alternative if the CPU
does not have proper MUL3 support. This issue exists also for other CPUs
(ie. using -march=mips1 with -mimadd fails).
gcc/ChangeLog:
* config/mips/mips.h (ISA_HAS_MADD_MSUB): Include allegrex.
* config/mips/mips.md: Tweak mul_acc_si/mul_sub_si to make it
work when MUL3 is not available.
gcc/testsuite/ChangeLog:
* gcc.target/mips/madd-10.c: New test.
* gcc.target/mips/maddu-5.c: New test.
* gcc.target/mips/msub-9.c: New test.
* gcc.target/mips/msubu-5.c: New test.
Signed-off-by: David Guillen Fandos <david@davidgf.net>
* config/mips/mips.h (ISA_HAS_WSBW): Defined a new macro.
* config/mips/mips.md (bswapsi2): Add new instruction.
(wsbwsi2): Replace with expand to support both wsbw and wsbh.
gcc/testsuite/ChangeLog:
* gcc.target/mips/bswap-7.c: New test.
Signed-off-by: David Guillen Fandos <david@davidgf.net>
MIPS: Add support for Allegrex min/max instructions
gcc/ChangeLog:
* config/mips/mips.h (ISA_HAS_MIN_MAX): Defined a new macro.
* config/mips/mips.md (sminsi3): Defined a new instruction.
(smaxsi3): Defined a new instruction.
gcc/testsuite/ChangeLog:
* gcc.target/mips/max-1.c: New test.
* gcc.target/mips/min-1.c: New test.
Signed-off-by: David Guillen Fandos <david@davidgf.net>
The MIPS Allegrex CPU is based on MIPS2 with some additional MIPS32r2
instructions and a few novel ones. Support for this CPU was added as of
binutils 2.41.
gcc/ChangeLog:
* config/mips/mips-cpus.def (MIPS_CPU): Added a new CPU.
* config/mips/mips-tables.opt: Regenerated table.
* config/mips/mips.cc: Added cost table for the new CPU.
* config/mips/mips.h (TARGET_ALLEGREX): Defined a new macro.
(TUNE_ALLEGREX): Defined a new macro.
(ISA_HAS_CONDMOVE): Added Allegrex CPU to the list.
(ISA_HAS_LDC1_SDC1): Exclude Allegrex from the list.
(ISA_HAS_COND_TRAP): Exclude Allegrex from the list.
(ISA_HAS_COND_TRAPI): Exclude Allegrex from the list.
(ISA_HAS_CLZ_CLO): Added Allegrex CPU to the list.
(ISA_HAS_ROR): Added Allegrex CPU to the list.
(ISA_HAS_WSBH): Added Allegrex CPU to the list.
(ISA_HAS_SEB_SEH): Added Allegrex CPU to the list.
(ISA_HAS_EXT_INS): Added Allegrex CPU to the list.
(ISA_HAS_XFER_DELAY): Exclude Allegrex from the list.
(ISA_HAS_HILO_INTERLOCKS): Added Allegrex CPU to the list.
* config/mips/mips.md: Added Allegrex CPU as a new processor.
* doc/invoke.texi: Documented Allegrex as a new arch
Signed-off-by: David Guillen Fandos <david@davidgf.net>
Saurabh Jha [Thu, 20 Nov 2025 15:14:24 +0000 (15:14 +0000)]
aarch64: mingw: Implement support for variadic ABI
The aarch64-w64-mingw32 target is different from aarch64-**-linux-gnu
targets with respect to how arguments for variadic functions are
handled. Specifically:
1. Homogeneous Floating-Point Aggregate (HFA) and Homogeneous Vector
Aggregate (HVA) are not handled in a special way. They are handled
like other composite types.
2. SIMD and Floating-Point registers aren't used.
This patch implements these differences for the aarch64-w64-mingw32
target. The new ABI specific functions to be used in target hooks are
declared in aarch64-abi-ms-protos.h and defined aarch64-abi-ms.cc. We
identify whether we are on aarch64-w64-mingw32 by the
TARGET_AARCH64_MS_ABI macro.
gcc/ChangeLog:
* config.gcc: Add new Makefile fragment and new object file.
* config/aarch64/aarch64-builtins.cc
(aarch64_ms_variadic_abi_init_builtins): Initialize builtin
variadic functions for aarch64-w64-mingw32.
* config/aarch64/aarch64-protos.h
(aarch64_ms_variadic_abi_init_builtins): Initialize builtin
variadic functions for aarch64-w64-mingw32.
* config/aarch64/aarch64.cc
(handle_aarch64_vector_pcs_attribute): Add support for
ARM_PCS_MS_VARIADIC.
(aarch64_ms_variadic_abi): Return descriptor to variadic
function call ABI for aarch64-w64-mingw32 target.
(aarch64_fntype_abi): Add support for variadic functions for
aarch64-w64-mingw32 target.
(aarch64_reg_save_mode): Add support for ARM_PCS_MS_VARIADIC.
(num_pcs_arg_regs): Add support for ARM_PCS_MS_VARIADIC.
(get_pcs_arg_reg): Add support for ARM_PCS_MS_VARIADIC.
(aarch64_arg_size): Returns size of argument.
(aarch64_ms_variadic_abi_layout_arg): aarch64-w64-mingw32
specific support for variadic ABI.
(aarch64_layout_arg): Add support for ARM_PCS_MS_VARIADIC.
(aarch64_function_arg): Implement TARGET_FUNCTION_ARG.
(aarch64_function_arg_advance): Add support for
ARM_PCS_MS_VARIADIC.
(aarch64_function_arg_regno_p): Add support for
ARM_PCS_MS_VARIADIC.
(aarch64_init_builtins): Add support for TARGET_AARCH64_MS_ABI.
(aarch64_ms_variadic_abi_build_builtin_va_list): Setup va_list
for aarch64-w64-mingw32.
(aarch64_build_builtin_va_list): Add support for
TARGET_AARCH64_MS_ABI.
(aarch64_ms_variadic_abi_expand_builtin_va_start): Implement
TARGET_BUILD_BUILTIN_VA_START.
(aarch64_setup_incoming_varargs): Implement
TARGET_SETUP_INCOMING_VARARGS.
(aarch64_mangle_type): Implement TARGET_MANGLE_TYPE.
(aarch64_variadic_abi_strict_argument_naming): Implement
TARGET_STRICT_ARGUMENT_NAMING.
* config/aarch64/aarch64.h
(aarch64_frame): Add new field
unaligned_saved_varargs_size.
(enum arm_pcs): Add new enum option
ARM_PCS_MS_VARIADIC.
* config/aarch64/cygming.h
(SUBTARGET_ATTRIBUTE_TABLE): Add support for ms_abi.
* config/mingw/winnt.cc
(aarch64_handle_ms_abi_attribute): Handle ms_abi attribue.
* config/mingw/winnt.h
(aarch64_handle_ms_abi_attribute): Handle ms_abi attribute.
* config/aarch64/aarch64-abi-ms-protos.h:
(aarch64_arg_partial_bytes): Declare.
(aarch64_ms_variadic_abi_canonical_va_list_type): Declare.
(aarch64_ms_variadic_abi_enum_va_list): Declare.
(aarch64_ms_variadic_abi_fn_abi_va_list): Implement
TARGET_FN_ABI_VA_LIST.
* config/aarch64/aarch64-abi-ms.cc:
(aarch64_arg_partial_bytes): Implement TARGET_ARG_PARTIAL_BYTES.
(aarch64_ms_variadic_abi_canonical_va_list_type): Implement
TARGET_CANONICAL_VA_LIST_TYPE.
(aarch64_ms_variadic_abi_enum_va_list): Implement
TARGET_ENUM_VA_LIST_P.
(aarch64_ms_variadic_abi_fn_abi_va_list): Implement
TARGET_FN_ABI_VA_LIST.
* config/aarch64/t-aarch64-mingw: New Makefile fragment.
gcc/testsuite/ChangeLog:
* gcc.target/aarch64/mingw/variadic_hfa.c: New test.
* gcc.target/aarch64/mingw/variadic_hva.c: New test.
* gcc.target/aarch64/mingw/variadic_int.c: New test.
Saurabh Jha [Thu, 9 Oct 2025 15:03:18 +0000 (15:03 +0000)]
aarch64: mingw: Make long double 64 bit
On windows targets, the size of long double is 64 bits. Therefore,
unlike aarch64-**-linux-gnu, where the size of long double is 128
bits, the size of long double for aarch64-w64-mingw32 has to be 64
bits.
This commit makes changes to gcc so that long double is 64 bits for
aarch64-w64-mingw32.
gcc/ChangeLog:
* config/aarch64/aarch64-abi-ms.h
(TARGET_LONG_DOUBLE_128): Set this to 0.
* config/aarch64/aarch64.cc
(aarch64_scalar_mode_supported_p): Make long double 64 bits.
(aarch64_c_mode_for_floating_type): Return true for TFmode.
* config/aarch64/aarch64.h
(TARGET_LONG_DOUBLE_128): Set this to 1.
gcc/testsuite/ChangeLog:
* gcc.target/aarch64/mingw/mingw.exp: New test.
* gcc.target/aarch64/mingw/long_double_size.c: New test.
co-authored-by: Radek Barton <radek.barton@microsoft.com>
co-authored-by: Martin Vejbora <mvejbora@microsoft.com>
Saurabh Jha [Thu, 9 Oct 2025 14:04:45 +0000 (14:04 +0000)]
aarch64: mingw: emit seh_endproc as comment
For mingw targets, there is no way to identify the end of function using
assembly directives right now.
This patch adds such directive as a comment. This is not a real
directive because some other things, like Structured Exception Handling
(SEH), needs to be supported before we can emit a real directive.
However, emitting an end of function marker now will let us modify
check-function-bodies in scanasm.exp, which in turn enables us to write
dg-compile tests for mingw target.
gcc/ChangeLog:
* config/aarch64/aarch64-abi-ms.h
(ASM_COMMENT_START): Specify start of comment.
(ASM_OUTPUT_TYPE_DIRECTIVE): Moved from aarch64-coff.h.
(ASM_DECLARE_FUNCTION_SIZE): Specify end of function as comment.
* config/aarch64/aarch64-coff.h
(ASM_OUTPUT_TYPE_DIRECTIVE): Moved to aarch64-abi-ms.h.
(ASM_DECLARE_FUNCTION_SIZE): Moved to aarch64-abi-ms.h.
Nathaniel Shead [Mon, 1 Dec 2025 13:43:18 +0000 (00:43 +1100)]
c++/modules: Stream DECL_CHAIN for decl_specialization_friend_p functions
r16-5298 attached the owning class for a friend template specialisation
on its DECL_CHAIN. However we don't stream DECL_CHAIN in general to
avoid walking into unrelated entities on the scope chain; this patch
adds a special case for these functions to ensure we don't lose this
information.
Ideally this would occur in trees_{out,in}::core_vals, but we can't
check decl_specialization_friend_p until after DECL_TEMPLATE_INFO has
been streamed, hence the slightly unusual placement.
gcc/cp/ChangeLog:
* module.cc (trees_out::lang_decl_vals): Stream DECL_CHAIN for
decl_specialization_friend_p functions.
(trees_in::lang_decl_vals): Likewise.
gcc/testsuite/ChangeLog:
* g++.dg/modules/friend-12_a.C: New test.
* g++.dg/modules/friend-12_b.C: New test.
Signed-off-by: Nathaniel Shead <nathanieloshead@gmail.com> Reviewed-by: Jason Merrill <jason@redhat.com>
Nathaniel Shead [Mon, 1 Dec 2025 12:32:40 +0000 (23:32 +1100)]
c++/modules: Fix ICE when writing imported using-directive [PR122915]
The crash in the PR is caused because we are attempting to write a
using-directive that we never made a dep for. This should only happen
for imported using-directives, where if we never opened the relevant
namespace in the module purview we don't think there's anything
interesting to discover and so never walk it.
There's actually no reason we need to emit imported using-directives at
all, however, unless they came from a partition, because importers will
be able to get that directive directly from the originating module if it
was going to be visible anyway. And we will always walk and create a
dependency (marked !import_p) for partition decls. So this patch fixes
the ICE by just skipping such cases.
To help validate this the patch also starts setting DECL_MODULE_IMPORT_P
correctly for using-directives.
PR c++/122915
gcc/cp/ChangeLog:
* module.cc (module_state::write_using_directives): Don't emit
imported using-directives.
(module_state::read_using_directives): Rename
add_using_namespace to add_imported_using_namespace.
* name-lookup.cc (add_using_namespace): Handle imported
using-directives.
(add_imported_using_namespace): Rename to match new
functionality.
* name-lookup.h (add_using_namespace): Rename to...
(add_imported_using_namespace): ...this.
gcc/testsuite/ChangeLog:
* g++.dg/modules/namespace-16_a.C: New test.
* g++.dg/modules/namespace-16_b.C: New test.
* g++.dg/modules/namespace-16_c.C: New test.
* g++.dg/modules/namespace-16_d.C: New test.
Signed-off-by: Nathaniel Shead <nathanieloshead@gmail.com> Reviewed-by: Jason Merrill <jason@redhat.com>
Nathaniel Shead [Sun, 30 Nov 2025 03:43:33 +0000 (14:43 +1100)]
c++/modules: Fix assertion in write_function_def for DECL_LOCAL_DECL_P
I hadn't retested r16-5727-g858f3007278337 on trunk before pushing and
I'd missed that it interacts badly with the assertion added by r16-5305-gc38bf35f0c7fa1. This adjusts the assertion to not check OMP
user-defined reductions (as they won't have import_export_decl called on
them anyway).
PR c++/119864
PR c++/122939
gcc/cp/ChangeLog:
* module.cc (trees_out::write_function_def): Don't crash on
OMP used-defined type reduction function definitions.
Signed-off-by: Nathaniel Shead <nathanieloshead@gmail.com> Reviewed-by: Patrick Palka <ppalka@redhat.com>
Peter Bergner [Mon, 1 Dec 2025 23:03:44 +0000 (16:03 -0700)]
[PATCH][PR target/122942] RISC-V: Add zifencei extension to the rva23s64 and rvb23s64 profiles [PR122942]
While verifying our Ascalon extensions list was complete (it isn't, follow-on
patch coming), I noticed that the rva23s64 and rvb23s64 profiles were missing
the mandatory zifencei extension, hence the following patch.
This was bootstrapped and regtested with no regressions.
Ok for trunk?
Peter
The RVA23S64 and RVB23S64 profiles both state that zifencei is a mandatory
extension. Add it to both profiles.
doc: add mention to Algol 68 in the Installing GCC manual
The "Downloading the source" sectio in the "Installing GCC" manual
mentions all the compilers distributed within the GCC source
distribution. This patch adds mentions to the ga68 compiler and its
libga68 run-time library.
Care is taken to describe the Algol 68 compiler as experimental,
according to the stipulations agreed with the SC.
Signed-off-by: Jose E. Marchesi <jemarch@gnu.org>
gcc/ChangeLog
* doc/install.texi (Downloading the source): Mention Algol 68.
Patrick Palka [Mon, 1 Dec 2025 22:08:01 +0000 (17:08 -0500)]
libstdc++: Inconsistent const in flat_map's value_type [PR122921]
flat_map's value_type is pair<key_type, mapped_type>, which we correctly
define within the container but incorrectly within the iterator.
PR libstdc++/122921
libstdc++-v3/ChangeLog:
* include/std/flat_map (_Flat_map_impl::_Iterator::value_type):
Remove const from key_type to make consistent with the
container's value_type.
* testsuite/23_containers/flat_map/1.cc (test09): New test.
* testsuite/23_containers/flat_multimap/1.cc (test09): New test.
Reported-by: Vincent X Reviewed-by: Jonathan Wakely <jwakely@redhat.com>
Christophe Lyon [Wed, 26 Nov 2025 13:03:37 +0000 (13:03 +0000)]
arm: Fix constraints in MVE asrl and lsll patterns [PR122858]
The second alternative for operand 1 needs a new constraint which does
not overlap with Pg, except that it can handle 32 to generate an
optimal mov.
This patch introduces a new Ph constraint which has the [32..255]
range.
This fixes a lot of regressions when running the testsuite for an MVE
target such as -march=armv8.1-m.main+mve.fp+fp.dp -mfloat-abi=hard.
gcc/ChangeLog:
PR target/122858
* config/arm/constraints.md (Ph): New constraint.
* config/arm/mve.md (mve_asrl_imm, mve_lsll_imm): Fix constraints
of operand 1 and handle 32 as special shift amount.
gcc/testsuite/ChangeLog:
PR target/122858
* gcc.target/arm/mve/pr122858.c: New test.
fortran: Honor array constructor type-spec during folding [PR107721]
When an array constructor has an explicit type-spec, all elements must be
converted to that type and character elements must be padded/truncated to
the specified length. This was working for simple cases but failing when:
1. Elements were parenthesized: [integer :: ([1.0])]
2. Constructors were nested: [[integer :: [1.0]]]
3. Character constructors were used with concatenation operators:
[character(16) :: 'a', 'b'] // '|'
4. Nested character constructors with concatenation:
[character(16) :: ['a', 'b']] // '|'
5. Outer constructor without type-spec wrapping inner with type-spec:
[[character(16) :: ['a', 'b']]] // '|'
6. Nested character constructors with different type-specs:
[character(16) :: [character(2) :: 'abcd']]
The root cause was twofold:
First, parenthesized expressions like ([1.0]) create EXPR_OP nodes that were
not being simplified before type conversion in check_constructor_type(),
so type conversion was applied to the EXPR_OP rather than its contents.
Second, character array constructors with explicit type-spec were not being
resolved before CONCAT operations in eval_intrinsic(), so elements retained
their original lengths instead of being padded to the type-spec length.
Additionally, nested array constructors needed their type-spec propagated
from the outer constructor.
The fix adds:
- Simplification of non-constant expressions in check_constructor_type()
before attempting type conversion
- Call to gfc_check_constructor_type() in eval_intrinsic() to ensure
type-spec conversion happens before any operations on array constructors
- Character array constructor resolution before CONCAT operations
- Recursive type-spec propagation for nested array constructors.
When a nested array constructor has its own explicit type-spec, it is
resolved first to enforce its own length (truncation/padding) before
propagating the outer type-spec and resolving again.
- Detection of nested character constructors with explicit type-spec
(via length_from_typespec) when the outer constructor has no type-spec
PR fortran/107721
PR fortran/102417
gcc/fortran/ChangeLog:
* arith.cc (eval_intrinsic): Call gfc_check_constructor_type on
array constructor operands with explicit type-spec to ensure
element type conversion before operations. Resolve character
array constructors before CONCAT operations.
(reduce_binary_ac, reduce_binary_ca, reduce_binary_aa): Preserve
character length info in result arrays.
* array.cc (check_constructor_type): Simplify non-constant
expressions before type checking to handle parenthesized elements.
Handle nested character array constructors with explicit type-spec
when outer constructor has no type-spec.
(gfc_resolve_character_array_constructor): Recursively propagate
type-spec to nested array constructors. If the nested constructor
has an explicit type-spec, resolve it first before propagating
the outer type-spec.
gcc/testsuite/ChangeLog:
* gfortran.dg/array_constructor_typespec_1.f90: New test.
Co-authored-by: Harald Anlauf <anlauf@gcc.gnu.org> Signed-off-by: Christopher Albert <albert@tugraz.at>
Jakub Jelinek [Mon, 1 Dec 2025 16:54:05 +0000 (17:54 +0100)]
c++: Fix ODR regressions caused by P2115R0 changes [PR122905]
The following testcase fails due to ODR warnings starting with the
r16-3233 change (P2115R0 PR120503 implementation).
The problem is that for C++20 we mangle differently the anonymous
union at the class scope from C++17, in C++17 the
unnamed enumeration that has an enumerator as a name for linkage purposes
before it is counted as TYPE_UNNAMED_P in nested_anon_class_index,
but for C++20 it is not, therefore the ODR warning.
While the term defined in https://eel.is/c++draft/dcl.enum#12.sentence-2
is defined for all enum types, its only use in
https://eel.is/c++draft/basic.link#4.5 is solely for enumeration types at
namespace scope, changing anything for those at class scope or block scope
has undesirable ABI consequences.
2025-12-01 Jakub Jelinek <jakub@redhat.com>
PR c++/122905
* decl.cc (enum_with_enumerator_for_linkage_p): Only return true
for namespace scope types.
* g++.dg/lto/pr122905.h: New file.
* g++.dg/lto/pr122905_0.C: New test.
* g++.dg/lto/pr122905_1.C: New test.
Robin Dapp [Wed, 12 Nov 2025 09:17:47 +0000 (10:17 +0100)]
RISC-V: vsetvl: Add null check for fault-first loop [PR122652].
For a fault-first load we store the first instruction that read its VL
result. The loop to do so uses next_nondebug_insn () which returns
nullptr when we are at the end. Check for that before accessing the
next insn.
Tobias Burnus [Mon, 1 Dec 2025 14:52:14 +0000 (15:52 +0100)]
gfortran.texi: Remove spurious @menu entry
Fixed the issue:
gfortran.texi:2542: warning: node up `Unsigned integers' in menu
`Default exponents' and in sectioning `Extensions implemented in
GNU Fortran' differ
There is an 'Unsigned integers' @menu entry under @section level
'Extensions implemented in GNU Fortran', where it should be. But some
spurious '@menu' entry, only with 'Unsigned integers' in it, was under
'@subsection Default exponents' just before the node of the
'@subsection Unsigned integers'. - The latter worked but was bogus
and lead the warning. Hence, it is now gone.
Rainer Orth [Mon, 1 Dec 2025 13:17:32 +0000 (14:17 +0100)]
testsuite: fortran: Fix gfortran.dg/alloc_comp_deep_copy_5.f90 etc. with non-gas/gld [PR122596]
The gfortran.dg/alloc_comp_deep_copy_[56].f90 tests FAIL on Solaris:
FAIL: gfortran.dg/alloc_comp_deep_copy_5.f90 -O0 (test for excess errors)
UNRESOLVED: gfortran.dg/alloc_comp_deep_copy_5.f90 -O0 compilation failed to produce executable
and many more when using either or both the native assembler or linker:
Excess errors:
Assembler:
"", line 1 : Illegal flag (-)
which is due to passing --noexecstack to the assembler. This is wrong in
general unless GNU as is used. Likewise, -Wl,-z,noexecstack has the same
issue.
This is *not* a target issue: the tests PASS as-is when using gas/gld on
Solaris (thus the current { target { ! *-*-darwin* } } is wrong, too).
Instead, the gas and gld effective-target keywords should be used to
control whether the flags are used, which is what this patch does.
Tested on i386-pc-solaris2.11, sparc-sun-solaris2.11,
x86_64-pc-linux-gnu, and x86_64-apple-darwin25.1.0.
Rainer Orth [Mon, 1 Dec 2025 13:04:03 +0000 (14:04 +0100)]
testsuite: xfail g++.dg/gcov/pr16855*.C on Solaris [PR52477,PR81337]
The g++.dg/gcov/pr16855*.C tests FAIL on Solaris:
FAIL: g++.dg/gcov/pr16855.C -std=gnu++17 gcov: 1 failures in line counts, 0 in branch percentages, 0 in condition/decision, 0 in prime-paths, 0 in return percentages, 0 in intermediate format, 0 failed in filters
FAIL: g++.dg/gcov/pr16855.C -std=gnu++17 line 24: is #####:should be 1
FAIL: g++.dg/gcov/pr16855.C -std=gnu++26 gcov: 6 failures in line counts, 0 in branch percentages, 0 in condition/decision, 0 in prime-paths, 0 in return percentages, 0 in intermediate format, 0 failed in filters
FAIL: g++.dg/gcov/pr16855.C -std=gnu++26 line 15: is 3:should be 1
FAIL: g++.dg/gcov/pr16855.C -std=gnu++26 line 22: is 3:should be 1
FAIL: g++.dg/gcov/pr16855.C -std=gnu++26 line 24: is #####:should be 1
FAIL: g++.dg/gcov/pr16855.C -std=gnu++26 line 38: is 3:should be 1
FAIL: g++.dg/gcov/pr16855.C -std=gnu++26 line 44: is 3:should be 1
FAIL: g++.dg/gcov/pr16855.C -std=gnu++26 line 49: is 3:should be 1
Same for -std=gnu++98.
The issue has long been known and the failures generate an excessive
amount of noise. Since the PRs haven't seen any activity in years, this
patch xfail's them.
Tested on i386-pc-solaris2.11, sparc-sun-solaris2.11,
x86_64-pc-linux-gnu, and x86_64-apple-darwin25.1.0.
Jonathan Wakely [Mon, 1 Dec 2025 12:54:50 +0000 (12:54 +0000)]
libstdc++: Use chrono::nanoseconds for __wait_until_impl parameter
Use the chrono::nanoseconds typedef instead of the equivalent
__wait_clock_t::duration typedef, and add a comment explaining why we
use a duration not a time_point.
libstdc++-v3/ChangeLog:
* include/bits/atomic_timed_wait.h (__wait_until_impl): Use
chrono::nanoseconds for parameter.
* src/c++20/atomic.cc (__wait_until_impl): Likewise.
Jonathan Wakely [Thu, 27 Nov 2025 11:40:46 +0000 (11:40 +0000)]
libstdc++: Fix spinloop in atomic timed waiting function [PR122878]
The __spin_until_impl function was presumably intended to just spin for
a short time, then give up and let the caller wait on a futex or
condvar. However, __spin_until_impl will never stop spinning unless
either the value changes or the timeout is reached. This means that when
__spin_until_impl returns, the caller should immediately return (because
either the value we were waiting for has changed, or the timeout
happened). So __wait_until_impl should never block on a futex or
condvar. However, the check for the return value of __spin_until_impl
would only return if the value changed (i.e. !__res._M_timeout). So if
the timeout occurred, it would fall through and block on the
futex/condvar, even though the timeout has already been reached.
This was causing a major performance regression in the timed waiting
functions of std::counting_semaphore.
The simplest fix is to replace __spin_until_impl entirely, just calling
__spin_impl to spin a small, finite number of times, and then return
immediately if either the value changed or the timeout happened. This
ensures that we don't block on the futex/condvar unnecessarily.
Removing __spin_until_impl also has the advantage that we no longer keep
calling steady_clock::now() on every iteration to check for a timeout.
That was also adding significant overhead to the timed waiting
functions.
libstdc++-v3/ChangeLog:
PR libstdc++/122878
* src/c++20/atomic.cc (__spin_until_impl): Remove.
(__wait_until_impl): Use __spin_impl instead of
__spin_until_impl and return if timeout is reached after
spinning.
Reviewed-by: Tomasz Kamiński <tkaminsk@redhat.com>
Jonathan Wakely [Thu, 27 Nov 2025 10:24:39 +0000 (10:24 +0000)]
libstdc++: Refactor futex usage in src/c++20/atomic.cc
The __futex_wait_flags scoped enum doesn't really have any benefit in
this file, because this code is no longer in the <atomic> header and so
we don't need to worry so much about namespace pollution. Just defining
the constants as int (and locally in the functions where they're needed)
avoids needing a static_cast<int> from the enum type.
I also noticed that _GLIBCXX_HAVE_LINUX_FUTEX_PRIVATE was never defined,
which meant we never used the FUTEX_PRIVATE_FLAG to tell the kernel that
all futex ops are process-private.
Also add comments and deleted definitions describing the API expected
for targets that define _GLIBCXX_HAVE_PLATFORM_WAIT.
libstdc++-v3/ChangeLog:
* src/c++20/atomic.cc: Document platform wait API.
(__futex_wait_flags): Remove enumeration type.
(futex_private_flag): Define constant for FUTEX_PRIVATE_FLAG.
(__platform_wait): Use local variables for futex op constants.
(__platform_notify): Likewise.
(__platform_wait_until): Likewise. Adjust parameter types for
consistency with __platform_wait.
Reviewed-by: Tomasz Kamiński <tkaminsk@redhat.com>
Rainer Orth [Mon, 1 Dec 2025 12:36:35 +0000 (13:36 +0100)]
algol68: Fix SPARC build
Algol68 bootstrap currently fails on Solaris/SPARC:
n file included from ./tm_p.h:4,
from gcc/algol68/a68-imports.cc:36:
gcc/config/sparc/sparc-protos.h:46:47: error: use of enum ‘memmodel’ without previous declaration
46 | extern void sparc_emit_membar_for_model (enum memmodel, int, int);
| ^~~~~~~~
tm_p.h needs memmodel.h on some targets, like SPARC.
Andrew Stubbs [Thu, 20 Nov 2025 17:18:13 +0000 (17:18 +0000)]
libgomp, amdgcn: Implement Managed Memory
This patch implements "managed" memory for AMD GCN GPUs in OpenMP. It
builds on the support added to the NVPTX libgomp for CUDA Managed Memory, a
week or two ago.
These features were first posted here a few years ago, as part of a larger
Unified Shared Memory patch series, and then in a slightly changed version just
over a year ago. Hopefully this time the controversial bits have been removed.
Since we do not use HIP we cannot use hipMallocManaged, so this patch attempts
to replicate the same effect by setting the appropriate attributes. This works
on more devices than support proper USM, but still I cannot be sure that the
settings are correct for every device out there (I have tested on gfx900,
gfx906, gfx908, gfx90a, and gfx1100).
The HSA header file update uses the most recent files relicensed for us by AMD,
at the time of the first patch posting. Those files have certainly moved on in
the upstream sources, but I did not ask to get those relicensed.
This commit creates initial gcc/algol68/ChangeLog and
libga68/ChangeLog files with copyright notices and content moved from
the top-level ChangeLog and gcc/ChangeLog.
OpenMP/Fortran: Allow explicit map followed by implicit deep mapping [PR120505]
Consider the following source code, assuming tiles is allocatable:
```
!$omp target enter data map(var%tiles(1)%den1, var%tiles(1)%den2) ! (1)
[...]
!$omp target ! implicitly maps var, which triggers deep mapping of tiles (2)
```
Each omp directive causes a run-time error in libgomp:
(1) libgomp: Mapped array elements must be the same (0x14d729c0 vs 0x14d72a18)
(2) libgomp: Trying to map into device [0x3704ca50..0x3704cb00) object when
[0x3704ca50..0x3704caa8) is already mapped
Regarding (1), the OpenMP spec has the following restriction: "If multiple list
items are explicitly mapped on the same construct and have the same containing
array or have base pointers that share original storage, and if any of the list
items do not have corresponding list items that are present in the device data
environment prior to a task encountering the construct, then the list items must
refer to *the same array elements* of either the containing array or the
implicit array of the base pointers."
Because tiles is allocatable, we cannot prove at compile time that array
elements are the same, so the check is deferred to libgomp. But there the
condition enforcing that all addresses are the same is too strict, so this patch
relaxes it to only check that addresses are sorted in increasing order.
The OpenMP spec allows (2) as long as it is implicit, without extending the
original mapping. So this patch sets the GOMP_MAP_IMPLICIT flag appropriately
on deep maps at compile time to let libgomp know that it is fine.
This patch ensures that such user code is accepted by:
(1) Setting the GOMP_MAP_IMPLICIT flag appropriately on deep maps;
(2) Relaxing the restriction on struct mapping from different containing arrays,
so that the element index need not be the same, instead addresses must be sorted
in increasing order.
This fixes the two errors currently seen when running SPEC HPC clvleaf
benchmark. However, further mapping issues prevent the benchmark from running to
completion.
PR fortran/120505
gcc/ChangeLog:
* omp-low.cc (lower_omp_target): Set GOMP_MAP_IMPLICIT flag.
libgomp/ChangeLog:
* target.c (gomp_map_vars_internal): Allow struct mapping from different
containing array elements as long as adresses are in increasing order.
* testsuite/libgomp.c-c++-common/map-arrayofstruct-2.c: Adjust
dg-output.
* testsuite/libgomp.c-c++-common/map-arrayofstruct-3.c: Likewise.
* testsuite/libgomp.fortran/map-subarray-5.f90: Likewise.
* testsuite/libgomp.fortran/map-subarray-10.f90: New test.
* testsuite/libgomp.fortran/map-subarray-9.f90: New test.
Jakub Jelinek [Mon, 1 Dec 2025 09:44:48 +0000 (10:44 +0100)]
a68: Fix algol68 build on i686-linux
GCC with enabled algol68 fails to build on i686-linux, the error is
../../gcc/algol68/a68-low-multiples.cc:636:31: error: format ‘%ld’ expects argument of type ‘long int’, but argument 2 has type ‘size_t’ {aka ‘unsigned int’} [-Werror=format=]
xasprintf is printf family, so it can't use %zd portably, so the
following patch uses what is used elsewhere, the HOST_SIZE_T_PRINT*
macros with (fmt_size_t) cast - the macros pick the smallest of
%d, %ld and %lld depending on SIZE_MAX, but it could still disagree
on the exact type and cause warnings or for hosts with say 24-bit
size_t it could be even larger, so the cast is needed to handle that.
2025-12-01 Jakub Jelinek <jakub@redhat.com>
* algol68/a68-low-multiples.cc (copy_multiple_dimension_elems): Use
HOST_SIZE_T_PRINT_DEC in xasprintf format string and cast to
fmt_size_t.
Jose E. Marchesi [Sun, 30 Nov 2025 19:42:43 +0000 (20:42 +0100)]
a68: some small a68-diagnostics.cc improvements
This commit fixes a few problems in algol68/a68-diagnostics.cc,
reported by David Malcolm in-list.
gcc/ChangeLog
* algol68/a68-diagnostics.cc (diagnostic): Copypasto "a meek"
should be "a firm". Support printing line number for programs
longer than 9 lines. Use obstack_append_str rather than
obstack_grow.
(obstack_append_str): New function.
Jakub Jelinek [Sun, 30 Nov 2025 14:52:27 +0000 (15:52 +0100)]
c++: Fix error recovery in cp_hide_range_decl [PR122465]
The following testcase shows that range_decl in cp_hide_range_decl is
sometimes also NULL_TREE and not just error_mark_node, and the function
IMHO should treat both the same, not try to hide anything in that case
because it doesn't know what should be hidden. This ICEs during
error recovery since something like cp_hide_range_decl has been introduced
(earlier it wasn't called that way).
The fix tweaks cp_parser_simple_declaration, such that it stores error_mark_node
instead of NULL_TREE into *maybe_range_for_decl in the erroneous cases.
2025-11-30 Jakub Jelinek <jakub@redhat.com>
PR c++/122465
* parser.cc (cp_parser_simple_declaration): Adjust function comment.
Set *maybe_range_for_decl to error_mark_node instead of keeping it
NULL_TREE in error cases or when followed by CPP_COLON.
but which is highly inefficient. This loops has 3 IVs (PR119577), one normal
scalar one, two vector ones, one counting up and one counting down (PR115120)
and has a forced unrolling due to an increase in VF because of the mismatch in
modes between the IVs and the loop body (PR119860).
This patch fixed all three of these issues and we now generate:
which shows that the new scalar IV is efficiently merged with the loop
control one based on IVopts.
To accomplish this the patch reworks how we handle "forced lived inductions"
with regard to vectorization.
Prior to this change when we vectorize a loop with early break any induction
variables would be forced live. Forcing live means that even though the values
aren't used inside the loop we must preserve the values such that when we start
the scalar loop we can pass the correct initial values.
However this had several side-effects:
1. We must be able to vectorize the induction.
2. The induction variable participates in VF determination. This would often
times lead to a higher VF than would have normally been needed. As such the
vector loops become less profitable.
3. IVcannon on constant loop iterations inserts a downward counting IV in
addition to the upwards one in order to support things like doloops.
Normally this duplicate IV is removed by IV opts, but IV doesn't understand
vector inductions. As such we end up with 3 IVs.
This patch fixes all three of these by choosing instead to create a new scalar
IV that's adjusted within the loop and to update all the IV statements outside
the loop by using this new IV.
We re-use vect_update_ivs_after_vectorizer for all exits now and put in a dummy
value representing the IV that is to be generated later.
To do this we delay when we call vect_update_ivs_after_vectorizer until after
the skip_epilogue edge is created and vect_update_ivs_after_vectorizer now
updates all out of loop usages of IVs and not just that in the merge edge to
the scalar loop. This not only generates better code, but negates the need to
fixup the "forced live" scalar IVs later on.
This new scalar IV is then materialized in
vect_update_ivs_after_vectorizer_for_early_breaks. When PFA using masks by
skipping iterations we now roll up the pfa IV into the new scalar IV by
adjusting the first iteration back from start - niters_peel and then take the
MAX <scal_iv, 0> to correctly handle the first iteration.
Because we are now re-using vect_update_ivs_after_vectorizer we have an issue
with UB clamping on non-linear inductions.
At the moment when doing early exit updating I just ignore the possibility of UB
since if the main exit is OK, the early exit is one iteration behind the main
one and so should be ok.
Things however get complicated with PEELED loops.
gcc/ChangeLog:
PR tree-optimization/115120
PR tree-optimization/119577
PR tree-optimization/119860
* tree-vect-loop-manip.cc (vect_can_advance_ivs_p): Check for nonlinear
mult induction and early break.
(vect_update_ivs_after_vectorizer): Support early break exits.
(vect_do_peeling): Support scalar IVs.
* tree-vect-loop.cc (vect_peel_nonlinear_iv_init): Support early break.
(vect_update_nonlinear_iv): use `unsigned_type_for` such that function
works for both vector and scalar types.
(vectorizable_induction, vectorizable_live_operation): Remove vector
early break IV code.
(vect_update_ivs_after_vectorizer_for_early_breaks): New.
(vect_transform_loop): Support new scalar IV for early break.
* tree-vect-slp.cc (vect_analyze_slp): Remove SLP build for early break
IVs.
* tree-vect-stmts.cc (vect_stmt_relevant_p): No longer mark early break
IVs as completely unused rather than used_only_live. They no longer
contribute to the vector loop and so should not be analyzed.
(can_vectorize_live_stmts): Remove vector early vreak IV code.
* tree-vectorizer.h (LOOP_VINFO_EARLY_BRK_NITERS_VAR): New.
(class loop_vec_info): Add early_break_niters_var.
Jose E. Marchesi [Sat, 11 Oct 2025 17:43:02 +0000 (19:43 +0200)]
a68: darwin specific support
This commit:
- Adapts specs in config/darwin.h for libga68.a.
- Amends section processing for non-LTO use in libibery on Darwin.
The initial implementation of the Mach-O simple object code was
mainly targeting LTO cases. The implementation was not suitable for
cases where we are just looking for a regular named section.
Jose E. Marchesi [Sat, 11 Oct 2025 17:42:30 +0000 (19:42 +0200)]
a68: DWARF language codes
This commit makes GCC aware of the DWARF numbers recently allocated
for Algol 68.
For DWARF 5, DW_LANG_Algol68 = 0x44.
For DWARF 6, DW_LNAME_Algol68 = 0x2e
with versioning schema YYYY, starting with 1973 for the original
Revised language. The language extensions we are working on will
be encoded in subsequent versions, 2025 etc.
See https://dwarfstd.org/issues/250304.1.html for more information.
Signed-off-by: Jose E. Marchesi <jemarch@gnu.org>
gcc/ChangeLog
* dwarf2out.cc: Set DW_LANG_Algol68 an DW_LNAME_Algol68.
Jose E. Marchesi [Sat, 11 Oct 2025 17:41:38 +0000 (19:41 +0200)]
a68: documentation
This commit adds a new section to the GCC Internals Manual and also
adds two new manuals.
ga68.texi is The GNU Algol 68 Compiler user manual. It describes how
to use the compiler and all the GNU extensions implemented on top of
the Algol 68 programming language.
ga68-internals.texi is The GNU algol68 Compiler Internals manual. It
describes the implementation of the front-end and it is of interest
primarily for developers.
Signed-off-by: Jose E. Marchesi <jemarch@gnu.org>
gcc/ChangeLog
* algol68/ga68-internals.texi: New file.
* algol68/ga68.texi: Likewise.
Jose E. Marchesi [Sat, 11 Oct 2025 17:58:04 +0000 (19:58 +0200)]
a68: testsuite: revised MC Algol 68 test set
We cannot distribute the MC Test Set with GCC as of now, due to not
clear distribution terms of the stuff. Until this gets clarified with
the CWI (then Mathematical Centrum) a README.mcts file explains how to
manually fetch and install the test set.
projects / thirdparty / gcc.git / log
