Dimitar Dimitrov [Sun, 22 Mar 2026 09:12:09 +0000 (11:12 +0200)]
testsuite: Require int32plus effective target for pr122925.c
The test case logic relies on the constant literal -4 being promoted to
32-bit int. But on AVR and other targets with 16-bit int types, the
literal is promoted to 16-bit int, which changes the expression value
and ultimately failing the test:
FAIL: gcc.dg/pr122925.c execution test
Fix by requiring an effective target with at least 32-bit ints.
Eric Botcazou [Tue, 31 Mar 2026 16:42:19 +0000 (18:42 +0200)]
Fix internal error on non-byte-sized reference in GIMPLE DSE
This is a regression present on all the active branches and a sibling of:
https://gcc.gnu.org/pipermail/gcc-patches/2024-February/646628.html
during GIMPLE pass: dse
+===========================GNAT BUG DETECTED==============================+
| 16.0.1 20260331 (experimental) [master r16-8354-gbf6989de817] GCC error:|
| in exact_div, at poly-int.h:2179 |
| Error detected around opt107.adb:26:11 |
| Compiling opt107.adb
At least one caller of compute_trims, namely maybe_trim_constructor_store,
expects ref->size to be a multiple of a byte for trimming, and that is most
probably the case for others in practice, from a cursory reading. Therefore
the patch adds the same test on ref->size as the one added to ref->offset.
gcc/
* tree-ssa-dse.cc (compute_trims): Bail out if ref->size is not
byte aligned either.
gcc/testsuite/
* gnat.dg/opt107.adb: New test.
* gnat.dg/opt107_pkg.ads: New helper.
The gcc.dg/analyzer/null-deref-pr123981-1.c test FAILs on 64-bit x86
targets with -m32:
FAIL: gcc.dg/analyzer/null-deref-pr123981-1.c (test for excess errors)
gcc.dg/analyzer/null-deref-pr123981-1.c:10:3: error: implicit declaration of function '_writegsbase_u64' [-Wimplicit-function-declaration]
gcc.dg/analyzer/null-deref-pr123981-1.c:10:20: warning: cast from pointer to integer of different size [-Wpointer-to-int-cast]
As always, restricting x86 tests by target triplet is wrong: instead of
target x86_64-*-* this needs to be x86 && lp64.
Tested on amd64-pc-solaris2.11, i386-pc-solaris2.11, and
x86_64-pc-linux-gnu.
Mikael Morin [Tue, 31 Mar 2026 13:34:29 +0000 (15:34 +0200)]
fortran: Give up reference factoring in case of shared tree [PR124661]
When we are about to create a variable to factor a subreference of an
array descriptor, check whether that subreference is shared with some of
the preliminary code and abort the factoring in that case.
In the example from the PR, the preliminary code contained bounds
checking code and the replacement of the subreferences by fresh
variables was causing those variables to be used before they were
defined in that bounds checking code.
PR fortran/121185
PR fortran/124661
gcc/fortran/ChangeLog:
* trans-array.cc (maybe_save_ref): New wrapper function around
save ref.
(set_factored_descriptor_value): Use the new wrapper function.
Add argument PRELIMINARY_CODE.
(gfc_conv_ss_descriptor): Update caller.
H.J. Lu [Sun, 29 Mar 2026 21:30:28 +0000 (14:30 -0700)]
x86: Call ix86_access_stack_p only for larger alignment
ix86_access_stack_p can be quite expensive. Don't check symbolic constant
load. Instead, call ix86_access_stack_p only if an INSN needs alignment
> stack alignment. This reduces the compile time of PR target/124684 from
185 seconds to 24 seconds. The compile time of PR target/124165 test is
unchanged.
PR target/124165
PR target/124684
* config/i386/i386.cc (ix86_symbolic_const_load_p_1): Renamed to
...
(ix86_need_alignment_p_1): This.
(ix86_symbolic_const_load_p): Renamed to ...
(ix86_need_alignment_p): This.
(ix86_find_max_used_stack_alignment): Call ix86_access_stack_p
only if the INSN needs alignment > stack alignment.
This patch incorporates changes to this file in Binutils
since October 2025. This file includes the new system
registers in the 2025 architecture extensions.
gcc/ChangeLog:
* config/aarch64/aarch64-sys-regs.def: Copy from Binutils.
Eric Botcazou [Tue, 31 Mar 2026 07:50:11 +0000 (09:50 +0200)]
Ada: Fix compiler crash on nested reduce attribute
This plugs a loophole in the resolution of reduction expressions, which
currently cannot be resolved when their immediate context is ambiguous,
for example in the case of a reduction expression nested in another one.
gcc/ada/
* sem_attr.adb (Analyze_Attribute) <Attribute_Reduce>: Rewrite the
analysis of a non-overloaded reducer. Moreover, when both actuals
are overloaded, compute the set of possible interpretations.
(Resolve_Attribute) <Attribute_Reduce>: Streamline the processing.
Jakub Jelinek [Tue, 31 Mar 2026 08:16:14 +0000 (10:16 +0200)]
c++: Fix up eval_data_member_spec
Ivan Lazaric reported privately a bug in eval_data_member_spec,
which I think can't be easily tested.
This is preceded by
if (!cpp_translate_string (parse_in, &istr, &ostr,
j == 2 ? CPP_STRING : CPP_UTF8STRING,
true))
so for j == 2 it does reverse conversion from -fnarrow-character-set= to
SOURCE_CHARSET (typically UTF-8) and otherwise (j == 1) from
UTF-8 to SOURCE_CHARSET, but the messages of exceptions were reversed.
Now, unless SOURCE_CHARSET is UTF-EBCDIC on s390x-tpf (dunno if that works
at all, but in theory could) and there is some character only in UTF-8 and
not in UTF-EBCDIC, or there is some narrow character set supported by iconv
which has characters not represented in UTF-8 (am not aware of those),
then cpp_translate_string shouldn't fail, so the code in there is just
for a theoretical case.
2026-03-31 Jakub Jelinek <jakub@redhat.com>
* reflect.cc (eval_data_member_spec): Fix up check for diagnostics
when reverse conversion fails.
Alexandre Oliva [Tue, 31 Mar 2026 00:58:19 +0000 (21:58 -0300)]
testsuite: x86: move alias-checks.C into vect
g++.dg/alias-checks.C requires vect_double outside vect.exp, where
-msse2 would have got enabled, but that effective_target is not
supposed to be used in such circumstances: it passes even when SSE2 is
disabled. Without enabling SSE2, vectors of doubles aren't really
available, so we don't get the expected vectorizations and the test
fails.
Instead of adding machine-specific options to alias-checks.C, I'm
moving it under g++.dg/vect/ and renaming it so that it gets, and
doesn't override, the default vector options for each platform.
Alexandre Oliva [Tue, 31 Mar 2026 00:58:01 +0000 (21:58 -0300)]
testsuite: x86: bump unroll insn limit for pr83403-*.c
Like rv32 and other targets, pr83403-[12].c need to bump
max-completely-peeled-insns to get the loops unrolled on x86 without
sse2 so as to enable the expected store motions.
for gcc/testsuite/ChangeLog
* gcc.dg/tree-ssa/pr83403-1.c: Bump
max-completely-peeled-insns on x86.
* gcc.dg/tree-ssa/pr83403-2.c: Likewise.
Alexandre Oliva [Tue, 31 Mar 2026 00:57:56 +0000 (21:57 -0300)]
testsuite: select -mfpmath=sse for pr32964.c on x86
When using the 387 FPU, we don't perform the expected optimization of
unions to floats. Setting -fexcess-precision=standard. doesn't fix it
for x86, but selecting -mfpmath=sse and forcing sse enabled does.
for gcc/testsuite/ChangeLog
* gcc.dg/tree-ssa/pr32964.c: Select the SSE FPU on x86.
Alexandre Oliva [Tue, 31 Mar 2026 00:57:48 +0000 (21:57 -0300)]
testsuite: drop loop optimizations in fstack-protector-strong.c
gcc.dg/fstack-protector-strong.c used to expect one stack_chk_fail per
function. When vectorizer improvements caused foo2 to get vectorized
to the point of enabling the loop and the automatic array-containing
data structure to be optimized away entirely, it no longer required
stack protection, and the expected pattern count was decreased.
However, on platforms that don't or can't vectorize the loop, it isn't
optimized away, and then we find the wrong pattern count.
Reverse the decrease and disable optimizations of the loop that could
cause it to be optimized out, so that we get the originally-expected
stack protection everywhere.
Alexandre Oliva [Tue, 31 Mar 2026 00:57:44 +0000 (21:57 -0300)]
testsuite: x86: pr115863-1.c should include stdint.h
stdint-gcc.h isn't necessarily installed by GCC or available from C
libraries. VxWorks ports don't generally have such a header
available. Include stdint.h instead.
Alexandre Oliva [Tue, 31 Mar 2026 00:57:34 +0000 (21:57 -0300)]
testsuite: analyzer: drop stderr from deref-before-check-2.c
On some VxWorks targets, stderr is #defined to a function call, and
this confuses the analyzer in test_1, so we skip the warning. That
appears to reveal a bug in the analyzer, but it doesn't seem to be
what the test was meant to test, so I'm dropping this unintended
feature from this testcase and filing PR124672 with a testcase to
reproduce it elsewhere.
for gcc/testsuite/ChangeLog
* c-c++-common/analyzer/deref-before-check-2.c (test_1):
Drop stderr.
H.J. Lu [Fri, 27 Jun 2025 22:27:25 +0000 (06:27 +0800)]
Adjust check for addressable misaligned stack argument
Adjust alignment check forcing a local copy if the argument on stack is
not aligned to its type and the current alignment, which is determined
by the backend, is less than MAX_SUPPORTED_STACK_ALIGNMENT when the
parameter has its address taken.
For x86, ignore argument passed on stack when updating stack alignment
since caller is responsible to align the outgoing stack for arguments
passed on stack.
gcc/
PR target/120839
PR middle-end/124671
* function.cc (assign_parm_adjust_stack_rtl): Get the parameter
as argument. Adjust alignment check forcing a local copy.
(assign_parms): Adjust.
* config/i386/i386.cc (ix86_argument_passed_on_stack_p): New.
(ix86_update_stack_alignment): Ignore argument passed on stack.
Jerry DeLisle [Mon, 30 Mar 2026 17:40:43 +0000 (10:40 -0700)]
fortran: Fix several issues with INQUIRE and unconnected units.
PR libfortran/124543
libgfortran/ChangeLog:
* io/inquire.c (inquire_via_unit): Fix return value for
NAMED. Add check for ACTION. Fix return values for
ENCODING, PENDING, READ, WRITE, READWRITE.
(inquire_via_filename): Add checks for ACTION, DELIM,
ASYNCHRONOUS, and PENDING. Fix return value for
ENCODING.
* io/unix.c (inquire_access): Change return value from no
to unknown.
Vineet Gupta [Mon, 30 Mar 2026 20:15:04 +0000 (13:15 -0700)]
bpf: md: Enable zero_extend{hi,qi}di2 to generate wN regs ...
... by adjusting the asm template reg string to 'W', similar to commit
gbd275e81812c for zero_extendsidi2. However unlike that change, this is
not a bug fix: zero_extendsidi2 relies on wN reg move to achieve the
32->64 semantics. A {8,16}->64 extension needs to be emulated anyways
by masking out extra bits so the reg used is not important.
However wN reg based codegen is still preferable in general:
- It is easier on verifier as it has to track fewer bits and avoids
corner case false positives.
- Better native codegen as upper bits are guaranteed to be zero and can
leverage target ISA mechanisms to achieve that mostly for free.
- Better native codegen on 32-bit targets which need to use 2 regs for
rN regs.
gcc/ChangeLog:
* config/bpf/bpf.md (zero_extendhidi2): Add alternates for
alu32 and use 'W' for reg operands.
(zero_extendqidi2): Ditto.
gcc/testsuite/ChangeLog:
* gcc.target/bpf/zero-ext.c: Add tests for {HI,QI}mode values.
Piyush Raj [Tue, 24 Feb 2026 20:17:54 +0000 (01:47 +0530)]
testsuite: bpf: Add initial BPF runtime tests
This patch adds runtime tests for BPF target.
These tests are based on existing compile-time tests and depends on
bpf.exp baseboard
To run the testsuite:
make check-gcc RUNTESTFLAGS="--target_board=bpf-vmtest bpf-torture.exp KERNEL_VERSION=6.15 LOG_LEVEL=ERROR"
Options:
KERNEL_VERSION (default: 6.15)
LOG_LEVEL (default: ERROR)
Note: VMTEST_DIR must be set in the environment.
See the bpf-vmtest-tool README for details.
gcc/testsuite/ChangeLog:
* gcc.target/bpf/torture/bpf-torture.exp: New testsuite.
* gcc.target/bpf/torture/invalid-memory-access.c: New test.
* gcc.target/bpf/torture/memcpy.c: New test.
* gcc.target/bpf/torture/memmove.c: New test.
* gcc.target/bpf/torture/memset.c: New test.
* gcc.target/bpf/torture/naked.c: New test.
* gcc.target/bpf/torture/nop.c: New test.
* gcc.target/bpf/torture/trace_openat.c: New test.
Richard Biener [Mon, 30 Mar 2026 12:30:45 +0000 (14:30 +0200)]
tree-optimization/124692 - update stmt before folding
The following makes sure SSA operands are up-to-date before folding.
Esp. when replace_uses_by is invoked from SCEV cprop which now has
ranger enabled we can otherwise end up ICEing where purpoted SSA
names now are constants. This follows what forwprop does.
PR tree-optimization/124692
* tree-cfg.cc (replace_uses_by): Call update_stmt after
substitution and before folding.
fortran: Fix free-form mixed OpenACC/OpenMP continuation state
Clear the opposite directive flag only for fresh free-form OpenMP/OpenACC
sentinel lines so continued directives do not leak OpenACC/OpenMP state into
the next independent directive.
PR fortran/108382
gcc/fortran/ChangeLog:
* scanner.cc (skip_free_oacc_sentinel): Clear openmp_flag only for
fresh OpenACC directive lines.
(skip_free_omp_sentinel): Clear openacc_flag only for fresh OpenMP
directive lines.
gcc/testsuite/ChangeLog:
* gfortran.dg/goacc/omp-108382.f90: New test.
Signed-off-by: Christopher Albert <albert@tugraz.at>
The values for chip IDs and the LITTLE.big variants have been confirmed
in the XNU sources (xnu/osfmk/arm/cpuid.h).
gcc/ChangeLog:
* config/aarch64/aarch64-cores.def (AARCH64_CORE): Add Apple M5
cores.
* config/aarch64/aarch64-tune.md: Regenerate.
* doc/invoke.texi: Add apple-m5 core to the ones listed
for arch and tune selections.
Jeff Law [Sun, 29 Mar 2026 23:36:25 +0000 (17:36 -0600)]
[RISC-V][PR target/124674] Define OImode in the usual way
This testcase tries to call build_nonstandard_integer_type for a 192-bit
precision type. This eventually faults because we don't have a suitably wide
mode.
Somewhat surprisingly, the targets are responsible for defining OImode. So
this adds it to the RISC-V port as well which trivially fixes the ICE. The
test is limited to rv64 as rv32 doesn't support int128. That could have been
handled with a target-supports selector just as easily.
Tested on riscv32-elf and riscv64-elf. Both the BPI and Pioneer are
experiencing NVME drive failures, so no bootstraps until I figure that out.
Waiting on the pre-commit tester to do its thing before pushing
Paul Thomas [Sat, 28 Mar 2026 09:08:54 +0000 (09:08 +0000)]
Fortran: Fix use-after-free in gfc_fixup_sibling_symbols [PR95879]
2026-03-28 Paul Thomas <pault@gcc.gnu.org>
gcc/fortran
PR fortran/95879
* parse.cc (fixup_st_func_formals): New function to update
statement function formal argument lists referencing a symbol
about to be freed.
(gfc_fixup_sibling_symbols): Call fixup_st_func_formals before
gfc_release_symbol.
gcc/testsuite
PR fortran/95879
* gfortran.dg/pr95879.f90: New test.
Alexandre Oliva [Wed, 25 Mar 2026 18:05:49 +0000 (15:05 -0300)]
testsuite: aarch64: skip erratas_opt_*.c on vxworks
VxWorks involves linking oddities in both kernel and rtp modes that
prevent the testcases for finding the expected outputs in the
erratas_opt_*.c testcases. One of the modes doesn't even perform
final linking, so it's kind of hopeless. Just skip them.
Alexandre Oliva [Wed, 25 Mar 2026 18:05:14 +0000 (15:05 -0300)]
testsuite: negated triplets need blank separators
!a-b-c is not the same as ! a-b-c. In !a-b-c, the '!' is part of the
triplet that istarget attempts to match. Fix all occurrences, except
for gcc.dg/vect/vect-93.c, that is taken care of separately.
Alexandre Oliva [Wed, 25 Mar 2026 18:04:53 +0000 (15:04 -0300)]
libgcc: vxworks: work around warnings promoted to errors on aarch64
In RTP mode, defining TLS_DELETE_HOOK_ARG as NULL AKA 0 elicits a
warning, that casting it to TASK_ID avoids.
objLibP.h defines an inline function that, depending on the
configuration, will have unused parameters. Allow warnings about
them.
vxTas calls are supposed to take __vx_tas_t, but upon passing it that
on aarch64, we get warnings about discarding volatile qualifiers, that
not even explicit type casts avoid. So, allow the warnings.
for libgcc/ChangeLog
* config/gthr-vxworks-tls.c (TLS_DELETE_HOOK_ARG): Type-cast
NULL constant to avoid warning.
* config/t-vxworks (LIBGCC2_CFLAGS): Tolerate unused-parameter
and discarded-qualifiers errors.
Alexandre Oliva [Mon, 23 Mar 2026 01:32:55 +0000 (22:32 -0300)]
libstdc++: simd: allow x86 -mlong-double-64
On a target that defaults to -mlong-double-64 -msse,
standard_abi_usable.cc and other tests fail to compile because of the
assert that rejects long double. As on ppc, we can trivially make
things work for -mlong-double-64.
for libstdc++-v3/ChangeLog
* include/experimental/bits/simd.h (__intrinsic_type): Accept
64-bit long doubles on x86 SSE.
which is just a 64-bit copy and doesn't clear the upper 32-bits as
semantically required by the pattern.
The issue is pattern's asm template missing ‘w’/'W' causing bpf_print_register ()
to only emit 'r' regs. Using ‘W’ in template generates correct code.
| w0 = w0
Note that 'W specifier ignores the RTL mode and unconditionally emits
the 'w' reg which might feel like creating a potential future bug.
However for this insn, it is correct/completely safe as follows:
- 'W1' for src operand: src operand has a SI mode in pattern.
- 'W0' for dst: Technically dst is DI, but the upper 32-bits of output
have to be 0 due to zero-extension, thus a 'w' reg is OK.
For selftests this is a wash, except for some spurious changes.
PR target/124670
gcc/ChangeLog:
* config/bpf/bpf.md (zero_extendsidi2): Use 'W' in asm template.
Paul Thomas [Fri, 27 Mar 2026 22:47:41 +0000 (22:47 +0000)]
Fortran: Fix ICE in interface mapping with assumed rank [PR102619]
2026-03-27 Paul Thomas <pault@gcc.gnu.org>
gcc/fortran
PR fortran/102619
* trans-expr.cc (gfc_get_interface_mapping_array): Add argument
'assumed_rank_formal', which if true returns the descriptor in
'ifm'.
(gfc_add_interface_mapping): Detect an assumed rank formal arg
of a non-intrinsic function to set 'assumed_rank_formal' and
pass the descriptor to gfc_get_interface_mapping_array.
gcc/testsuite/
PR fortran/102619
* gfortran.dg/pr102619.f90: New test.
Eric Botcazou [Fri, 27 Mar 2026 20:09:53 +0000 (21:09 +0100)]
Ada: Lift restriction on structural generic instantiation of subprograms
When an entity of a library-level package is used as actual in an attempted
structural instantiation of a library-level generic unit in the package, an
error is given since the structural instance is a library unit with circular
semantic dependence on the package.
This change lifts the restriction for generic subprograms by demoting the
structural instance into a local instance in this case, which means that
several instances might end up being present in a single partition, unless
the optimizer is run on the entire partition, for example by means of LTO.
gcc/ada/
* sem_ch12.adb (Analyze_Structural_Associations): Create a local
instance instead of a structural instance when a local entity of
a library-level package is used in a subprogram instantiation.
Eric Botcazou [Fri, 27 Mar 2026 17:38:09 +0000 (18:38 +0100)]
Ada: Fix overriding indicator wrongly rejected on protected procedure
The case itself is fairly pathological (and very likely not human made), but
it's a clear incentive to get rid of an old kludge in generic instantiation,
whereby the full view of a type is forced on nodes in the generic unit even
though only the partial declaration is visible to them.
gcc/ada/
PR ada/124596
* sem_ch12.ads (Check_Private_View): Move around.
* sem_ch12.adb (Check_Private_View): Retrieve the partial view
by means of a call to Incomplete_Or_Partial_View.
(Save_Global_References.Set_Global_Type): Do not force the full
view of a type when only the partial declaration is visible.
* sem_res.adb (Resolve_Actuals.Insert_Default): Remove obsolete
code coping with the above kludge.
gcc/testsuite/
* gnat.dg/generic_inst21.adb: New test.
Tobias Burnus [Fri, 27 Mar 2026 16:54:00 +0000 (17:54 +0100)]
libgomp: fix omp_target_is_present and omp_get_mapped_ptr
There were a few minor issues with the two routines, partially because of
not handling corner cases and partially some clarifications are only in
newer versions of the spec.
In particular, for omp_target_is_present
* NULL pointer aren't regarded as present
* For (unified-)shared memory, claiming that something has always corresponding
storage is wrong - it mostly never has. (but it is omp_target_is_accessible).
* Even with shared memory, 'declare target' usually has device memory. For
'link' it is made to point to the host, i.e. it is not mapped, all others
are still mapped. (With 'requires self_mapping', 'enter' should also not be
mapped (and turned internally to 'link'), only 'local' needs to be mapped.)
For omp_get_mapped_ptr
* For NULL we can return NULL early also for devices.
* For shared memory, we shouldn't touch link (it is not counting as mapped);
hence return NULL for it.
The documentation was updated add some missing cross references as the more
useful ones were missing. Additionally, the description for the two modified
routines has been updated.
libgomp/ChangeLog:
* target.c (omp_target_is_present, omp_get_mapped_ptr): Update handling
for nullptr and shared-memory devices.
* libgomp.texi (omp_target_is_present, omp_get_mapped_ptr): Update
description, add see-also @refs.
(omp_target_is_accessible, omp_target_associate_ptr): Add see-also
@refs.
* testsuite/libgomp.c/omp_target_is_present.c: New test.
* testsuite/libgomp.c/omp_target_is_present-2.c: New test.
Andrew Pinski [Thu, 26 Mar 2026 20:21:01 +0000 (13:21 -0700)]
regcprop: Return early in maybe_mode_change for unorder modes [PR124649]
Like r16-727-g2ec5082dd24cef but the call to partial_subreg_p happens
in a later place, maybe_mode_change.
For this example we have VNx4QImode and DImode which are not ordered.
Bootstrapped and tested on aarch64-linux-gnu.
PR rtl-optimization/124649
gcc/ChangeLog:
* regcprop.cc (maybe_mode_change): Return early
for unordered modes.
gcc/testsuite/ChangeLog:
* gcc.dg/torture/pr124649-1.c: New test.
Signed-off-by: Andrew Pinski <andrew.pinski@oss.qualcomm.com>
Paul Thomas [Fri, 27 Mar 2026 10:50:39 +0000 (10:50 +0000)]
Fortran: Fix ICE with allocation of scalar pointer entity [PR114021]
2026-03-27 Paul Thomas <pault@gcc.gnu.org>
gcc/fortran
PR fortran/114021
* symbol.cc (gfc_get_unique_symtree): If the namespace argument
is NULL, allocate a new symtree and provide it with the unique
name.
* trans-expr.cc (trans_scalar_assign): In the deep copy of a
derived type with allocatable components, fix the rhs value if
it is not a constant or a variable.
* trans-stmt.cc (gfc_trans_allocate): Do not deallocate
allocatable components of a source that is not a variable and
is a pointer. If the DECL_NAME or its IDENTIFIER_POINTER are
null,use gfc_get_unique_symtree with NULL namespace to obtain a
symtree for the assignment.
gcc/testsuite/
PR fortran/114021
* gfortran.dg/pr114021.f90: New test.
Robin Dapp [Tue, 24 Mar 2026 09:58:14 +0000 (10:58 +0100)]
RISC-V: Allow all vector modes during builtin registration. [PR124613]
In r16-7312-gecc37444062b40 we allowed all vector modes for the
any_target hook. Since then we would ICE in gcc.target/riscv/pr122051.c
as emit_move_multi_word would choose a fractional vector mode.
This patch disallows fractional vector modes for xtheadvector in
riscv_vector_mode_supported_p but makes an exception for builtin
registration (through a global variable). During registration we
need to have all modes available in order to maintain the registration
order for LTO streaming.
PR target/124613
gcc/ChangeLog:
* config/riscv/riscv-vector-builtins.cc (rvv_switcher::rvv_switcher):
Add riscv_registering_builtins.
(rvv_switcher::~rvv_switcher): Set riscv_registering_builtins to
false.
* config/riscv/riscv.cc (riscv_vector_mode_supported_p): Use
riscv_registering_builtins.
* config/riscv/riscv.h: Declare.
We can attempt to add DW_AT_artificial because of two reasons,
because of DECL_ARTIFICIAL on the decl and because of the
artificial function attribute which is attempted when marking
a decl DIE for inlining. The following guards the latter,
avoiding an ICE with checking enabled.
PR debug/124644
* dwarf2out.cc (dwarf2out_abstract_function): Only
add DW_AT_artificial if the decl isn't DECL_ARTIFICIAL.
Paul Thomas [Thu, 26 Mar 2026 18:50:13 +0000 (18:50 +0000)]
Fortran: Minor PDT cleanup and fix in gfc_simplify_exp [PR115315]
2026-03-26 Paul Thomas <pault@gcc.gnu.org>
gcc/fortran
PR fortran/115315
* decl.cc (insert_parameter_exprs): Make strcmp condition more
concise.
(gfc_get_pdt_instance): Use gf_replace_expr where possible and
use return value of gfc_simplify_expr. Correct error in which
params->expr was being simplified instead of c2->initializer.
* expr.cc (gfc_simplify_expr): If the substring 'start' value
is less than zero, it is clearly out of range and so return
false.
gcc/testsuite/
PR fortran/115315
* gfortran.dg/pdt_90.f03: New test.
Jakub Jelinek [Thu, 26 Mar 2026 17:11:06 +0000 (18:11 +0100)]
testsuite: Fix up another tcl 9 compatibility problem
TCL 9 removed case command which has been deprecated before.
https://core.tcl-lang.org/tcl/wiki?name=Migrating+scripts+to+Tcl+9
says "Replace case with switch" and switch has been working in TCL 8
just fine.
2026-03-26 Jakub Jelinek <jakub@redhat.com>
* lib/mike-gcc.exp (postbase): Use switch instead of case for
TCL 9 compatibility.
* lib/mike-g++.exp (postbase): Likewise.
Michal Jires [Thu, 5 Mar 2026 09:03:21 +0000 (10:03 +0100)]
lto/124289 - Correctly handle toplevel asm with -flto-partition=cache
This patch adds create_asm_partitions to cache partitioning
to prevent issues with non-renameable symbols (while partition
joining) and static asm symbols (while partition splitting).
All other relevant partitionings use create_asm_partitions.
This was not used in cache partitioning, because toplevel asm
could be in principle special handled in cache partitioning
with marginally better results, but I never implemented it.
lto/124289
gcc/lto/ChangeLog:
* lto-partition.cc (enum map1to1_content): New.
(map_1_to_1): Use map1to1_content.
(lto_1_to_1_map): Likewise.
(create_asm_partitions): Likewise.
(lto_max_map): Likewise.
(lto_cache_map): Use create_asm_partitions.
gcc/testsuite/ChangeLog:
* gcc.dg/lto/toplevel-extended-asm-2_0.c: Add padding to asm label.
* gcc.dg/lto/toplevel-extended-asm-2_1.c: Add padding to asm label.
Marek Polacek [Thu, 19 Mar 2026 17:10:38 +0000 (13:10 -0400)]
c++/reflection: ICE with substitute and undeduced auto [PR123613]
In substitute7.C we have
template <typename T, auto ... Vs>
constexpr auto construct_from = T{Vs...}; // #1
struct Inner {};
struct Outer { Inner m; };
constexpr auto r = substitute(^^construct_from,
{ ^^Outer, ^^construct_from<Inner> });
which crashes because the auto in #1 hadn't been deduced when we
called eval_can_substitute -> lookup_template_variable when processing
the substitute call. We can call mark_used to resolve this, because
it has:
if (undeduced_auto_decl (decl))
maybe_instantiate_decl (decl);
and to make the json-parser.C test work, we have to do something
similar in eval_substitute, otherwise we crash due to another
undeduced auto.
PR c++/123613
gcc/cp/ChangeLog:
* reflect.cc (get_reflection): Call mark_used.
(eval_can_substitute): Don't resolve_nondeduced_context here.
(eval_substitute): Call lookup_and_finish_template_variable instead
of lookup_template_variable and finish_template_variable.
gcc/testsuite/ChangeLog:
* g++.dg/reflect/json-parser.C: New test.
* g++.dg/reflect/substitute7.C: New test.
* g++.dg/reflect/test.json: New test.
Harald Anlauf [Tue, 24 Mar 2026 21:04:29 +0000 (22:04 +0100)]
Fortran: fix rank/shape check in interface checking [PR124567]
PR fortran/124567
gcc/fortran/ChangeLog:
* interface.cc (gfc_check_dummy_characteristics): Split shape check
into a separate check for rank and a check for shape, taking into
account a corner case where the ambiguity between deferred shape
and assumed shape has not been fully resolved at the time of
checking.
gcc/testsuite/ChangeLog:
* gfortran.dg/pr124567.f90: New test.
* gfortran.dg/proc_decl_30.f90: Likewise.
Richard Biener [Wed, 25 Mar 2026 13:40:59 +0000 (14:40 +0100)]
gcov-profile/121074 - hold onto gcov file for less time
The following fixes the issue of two open gcov files at the
same time, one from -ftest-coverage, opened/closed by
coverage_init/finish and one from -fauto-profile, attempted
to be opened by pass_ipa_auto_profile. The solution is
to open the coverage files only during pass_ipa_tree_profile.
PR gcov-profile/121074
* coverage.h (coverage_init_file): Declare.
(coverage_finish_file): Likewise.
* coverage.cc (coverage_init_file): New function, split
out actual file opening and writing from ...
(coverage_init): ... here.
(coverage_finish_file): Likewise for file closing, from ...
(coverage_finish): ... here.
* tree-profile.cc (tree_profiling): Call coverage_init_file
and coverage_finish_file here.
Richard Biener [Wed, 25 Mar 2026 08:40:56 +0000 (09:40 +0100)]
tree-optimization/124627 - inlining and elided debug stmts
When we fixup a noreturn call during inlining we can end up eliding
debug stmts, so we have to make sure to not re-instantiate SSA
operands on those later when processing all debug stmts.
PR tree-optimization/124627
* tree-inline.cc (copy_debug_stmts): Only copy debug
stmts that are still in the IL.
Alexandre Oliva [Wed, 25 Mar 2026 09:45:47 +0000 (06:45 -0300)]
vxworks: fix gthr visibility issues
xtreme-header-8.C fails on VxWorks because various TU-local gthr
functions defined as static inline are referenced from libstdc++
symbols with global visibility.
C++ modules require those functions to be non-static inline, and other
gthr implementations adopt always_inline in C++.
Follow this practice in gthr-vxworks.h as well.
for libgcc/ChangeLog
* config/gthr-vxworks.h (__GTHREAD_ALWAYS_INLINE,
__GTHREAD_INLINE): Copy from gthr-posix.h. Replace static
inline with __GTHREAD_INLINE.
Alexandre Oliva [Wed, 25 Mar 2026 09:45:35 +0000 (06:45 -0300)]
[testsuite] [vxworks] add -gno-strict-dwarf to pr111409.c
The expected macro debug information is not issued with
-gstrict-dwarf, and ports such as vxworks default to that. Allow
non-strict dwarf for the test.
Alexandre Oliva [Wed, 25 Mar 2026 09:45:25 +0000 (06:45 -0300)]
testsuite: vect: slp-mask-store-1 accesses past the end of flags
The loop invokes undefined behavior: the flags array runs 0..31, but
so does the iterator i, thus the accesses to pairs therein run 0..63.
If we're lucky we find all zeros and don't mess with anything else,
but if we find any nonzero out-of-range flags, we'll end up scribbling
out of range onto x, that runs 0..15.
Surely the iterator wasn't meant to go up to 32. If it goes up to 16,
as proposed herein, we only access flags in range, and since its
second half is all zeros, we only access x in range as well.
30_threads/condition_variable_any/stop_token/wait_on.cc's
test_wait_until occasionally fails on vxworks under very high load, in
a way that suggests wait_until times out before the main thread
requests it to stop. Extend the timeouts to make more room for the
stop request.
for libstdc++-v3/ChangeLog
* testsuite/30_threads/condition_variable_any/stop_token/wait_on.cc
(test_wait_until): Extend the timeout for a stop request.
Alexandre Oliva [Wed, 25 Mar 2026 09:45:09 +0000 (06:45 -0300)]
[testsuite] [ppc] expect vectorization in gen-vect-11c.c with lp64
The first loop in main gets stores "vectorized" on powerpc64 into
full-word stores, even without any vector instruction support, so the
test's expectation of no loop vectorization is not met.
for gcc/testsuite/ChangeLog
* gcc.dg/tree-ssa/gen-vect-11c.c: xfail the test for no
vectorization on powerpc*-*-* && lp64.
Alexandre Oliva [Wed, 25 Mar 2026 09:44:55 +0000 (06:44 -0300)]
c++: arm -Wdeprecated fails for ctors
ARM requires C++ constructors to return the 'this' pointer. Setting
that up disables warnings at the ctor location for Wuse_after_free.
Unfortunately, the nowarn_spec_t mapping maps both Wuse_after_free and
Wdeprecated_declaration to the same NW_OTHER catchall group, so we
don't get deprecation warnings for constructors, and libstdc++'s
20_util/pair/cons/99957.cc fails.
We may want to consider also explicitly mapping Wdeprecated and
Wdeprecated_declaration to another group, but this minimal change
appears to be enough.
for gcc/ChangeLog
* gcc-diagnostic-spec.cc (nowarn_spec_t::nowarn_spec_t): Map
OPT_Wuse_after_free like OPT_Wuse_after_free_.
Paul Thomas [Wed, 25 Mar 2026 09:28:15 +0000 (09:28 +0000)]
Fortran: Invalid error in bounds check of elemental dim [PR119273]
2026-03-25 Paul Thomas <pault@gcc.gnu.org>
gcc/fortran
PR fortran/119273
* trans-array.cc (expr_contains_impure_fcn): New function.
(gfc_expr_contains_impure_fcn): New function calling above fcn.
(array_bound_check_elemental): Add indexse pre-block to se pre.
Warn if the index expression contains a function not declared
to be pure.
gcc/testsuite/
PR fortran/119273
* gfortran.dg/pr119273.f90: New test.
Rainer Orth [Wed, 25 Mar 2026 08:44:54 +0000 (09:44 +0100)]
i386: Fix gcc.target/i386/pr61599-1.c on Solaris/x86
The gcc.target/i386/pr61599-1.c test has been FAILing on Solaris/x86
with the native assembler for a long time:
FAIL: gcc.target/i386/pr61599-1.c (test for excess errors)
UNRESOLVED: gcc.target/i386/pr61599-1.c compilation failed to produce executable
Excess errors:
ld: fatal: relocation error: R_AMD64_PC32: file gcc/amd64/crtbegin.o: symbol completed.0: value 0xc01004b9 does not fit
ld: fatal: relocation error: R_AMD64_PC32: file gcc/amd64/crtbegin.o: symbol completed.0: value 0xc01004a7 does not fit
Looking closer, pr61599-1.o is 3.1 GB large with the native assembler.
Comparing the section headers between as and gas, one sees the difference:
The use of SHT_PROGBITS by as is due to gcc not emitting @nobits for the
.lbss sections, so as uses the default section type:
.section .lbss.a,"aw"
For .bss, we get this instead:
.section .bss.a,"aw",@nobits
as usually doesn't care about section names, but relies on the compiler
to correctly emit section types and flags.
This happens because x86_64_elf_section_type_flags first calls
default_section_type_flags, which doesn't know about .lbss and sets
SECTION_NOTYPE. When default_elf_asm_named_section later emits the
.section directive, @nobits is omitted even though SECTION_BSS has been
added by then.
To work around this, this patch clears SECTION_NOTYPE when SECTION_BSS
is set.
Similarly, the current assembler output relies on the assembler to set
the SHF_X86_64_LARGE section flag based on the section name. While gcc
emits the necessary flag letter ('l' with gas, 'h' with Solaris as) for
plain .lbss sections, it fails to do so for .lbss.*.
This patch adresses this by moving the gas definition of
MACH_DEP_SECTION_ASM_FLAG to i386/x86_64.h so it's picked up by all
64-bit x86 targets. It also extends ix86_in_large_data_p to also
consider .lbss.* etc. sections as large data sections, not just .lbss,
matching x86_64_elf_section_type_flags.
pr61599-1.o is identical without and with the section letter and section
type specifiec.
Bootstrapped without regressions on i386-pc-solaris2.11,
amd64-pc-solaris2.11, x86_64-pc-linux-gnu, and i686-pc-linux-gnu.
gcc:
* config/i386/i386.cc (ix86_in_large_data_p): Check for .lbss etc.
(x86_64_elf_section_type_flags): Clear SECTION_NOTYPE for .lbss etc.
* config/i386/x86-64.h (MACH_DEP_SECTION_ASM_FLAG): Define.
* config/i386/sol2.h [HAVE_SOLARIS_AS]
(MACH_DEP_SECTION_ASM_FLAG): Redefine as 'h'.
gcc/testsuite:
* gcc.target/i386/pr61599-1.c (dg-options): Add -save-temps.
(scan-assembler-times): Check for @nobits.
[PR124572, LRA]: Deal with generation of reload insns during elimination
It was assumed that elimination in LRA does not generate new reload
insns. In the testcase the elimination of SFP subreg generates reload
insn of the subreg and this insn is added at end of RTL code. The insn
is also skipped for necessary processing. This results in ICE. The patch
checks creation of reloads insn during elimination, insert them in the right
place, and add them for later processing.
gcc/ChangeLog:
PR rtl-optimization/124572
* lra-eliminations.cc (lra_eliminate): Push new reload insns for
eliminations in insns.
projects / thirdparty / gcc.git / log
