Before the P1169R4 changes, operator () of a lambda was
always a method, so it was fine to pass method_p = 1 unconditionally,
but it isn't always the case, so this patch adds a check for whether
it is a method or nor.
2023-01-25 Jakub Jelinek <jakub@redhat.com>
PR c++/108525
* mangle.cc (write_closure_type_name): Don't assume all
lambda operator() fns are methods.
* g++.dg/cpp23/static-operator-call5.C: New test.
gcc/testsuite/ChangeLog:
* gcc.target/arm/mve/intrinsics/vqrdmlashq_n_s16.c: Use
check-function-bodies instead of scan-assembler checks. Use
extern "C" for C++ testing.
* gcc.target/arm/mve/intrinsics/vqrdmlashq_n_s32.c: Likewise.
* gcc.target/arm/mve/intrinsics/vqrdmlashq_n_s8.c: Likewise.
The following fixes not converging iteration in value-numbering of
PHI nodes when we use an equivalence to prove the PHI node is
degenerate. We have to avoid the situation where we oscillate
between the two equivalent values because the result is fed back
via a backedge.
PR tree-optimization/108523
* tree-ssa-sccvn.cc (visit_phi): Avoid using the exclusive
backedge value for the result when using predication to
prove equivalence.
Andreas noticed that when I removed lto_read_all_file_options I
failed to update the internals manual which refers to it. The
following attempts to reflect the current situation.
* doc/lto.texi (Command line options): Reword and update reference
to removed lto_read_all_file_options.
The addition of TARGET_CSSC meant that we wouldn't generate SVE
UQDEC instructions unless +cssc was also enabled.
Fixes:
- gcc.target/aarch64/sve/slp_4.c
- gcc.target/aarch64/sve/slp_10.c
- gcc.target/aarch64/sve/while_4.c
gcc/
* config/aarch64/aarch64.md (umax<mode>3): Separate the CNT and CSSC
tests.
The sizeless-*.c tests contained (deliberately) invalid constructors
that had two errors. The first error now suppresses the second error,
but the second error was the main focus of the test. This patch
therefore rewrites it into a different form.
gcc/testsuite/
* gcc.target/aarch64/sve/acle/general-c/sizeless-1.c: Avoid
"initializer element is not constant" error.
* gcc.target/aarch64/sve/acle/general-c/sizeless-2.c: Likewise.
This patch adds aarch64 to the list of vect_long_long targets. Regression tested
on aarch64-none-linux-gnu.
gcc/testsuite/ChangeLog:
* lib/target-supports.exp (check_effective_target_vect_long_long): Add
aarch64 to list of targets supporting long long vectorization.
The first of the following testcases is miscompiled on powerpc64-linux -O2
-m64 at least, the latter at least on x86_64-linux -m32/-m64.
Since GCC 11 store-merging has a separate string_concatenation mode which
turns stores into setting a MEM_REF from a STRING_CST.
This mode is triggered if at least one of the to be merged stores
is a STRING_CST store and either the first store (to earliest address)
is that STRING_CST store or the first store is 8-bit INTEGER_CST store
and then there are some rules when to turn that mode off or not merge
further stores into it.
The problem with these 2 testcases is that the actual implementation
relies on start/width of the store to be at byte boundaries, as it
simply creates a char array, MEM_REF can be only on byte boundaries
and the char array too, plus obviously STRING_CST as well.
But as can be easily seen in the second testcase, nothing verifies this,
while the first store has to be a STRING_CST (which will be aligned)
or 8-bit INTEGER_CST, that 8-bit INTEGER_CST store could be a bitfield
store, nothing verifies any stores in between whether they actually are
8-bit and aligned, the only major requirement is that all the stores
are consecutive.
For GCC 14 I think we should reconsider this, simply treat STRING_CST
stores during the merging like INTEGER_CST stores and deal with it only
during split_group where we can create multiple parts, this part
would be a normal store, this part would be STRING_CST store, this part
another normal store etc. But that is quite a lot of work, the following
patch just disables the string_concatenate mode if boundaries aren't byte
aligned in the spot where we disable it if it is too short too.
If that happens, we'll just try to do the merging using normal 1/2/4/8 etc.
byte stores as usually with RMW masking for any bits that shouldn't be
touched or punt if we end up with too many stores compared to the original.
Note, an original STRING_CST store will count as one store in that case,
something we might want to reconsider later too (but, after all, CONSTRUCTOR
stores (aka zeroing) already have the same problem, they can be large and
expensive and we still count them as one store).
2023-01-25 Jakub Jelinek <jakub@redhat.com>
PR tree-optimization/108498
* gimple-ssa-store-merging.cc (class store_operand_info):
End coment with full stop rather than comma.
(split_group): Likewise.
(merged_store_group::apply_stores): Clear string_concatenation if
start or end aren't on a byte boundary.
* gcc.c-torture/execute/pr108498-1.c: New test.
* gcc.c-torture/execute/pr108498-2.c: New test.
Use the offset in TREE_OPERAND(component_ref, 2) when available instead
of DECL_FIELD_OFFSET when trying to compute offset for a COMPONENT_REF.
Co-authored-by: Jakub Jelinek <jakub@redhat.com>
gcc/ChangeLog:
PR tree-optimization/108522
* tree-object-size.cc (compute_object_offset): Use
TREE_OPERAND(ref, 2) for COMPONENT_REF when available.
gcc/testsuite/ChangeLog:
PR tree-optimization/108522
* gcc.dg/builtin-dynamic-object-size-0.c
(test_dynarray_struct_member): New test.
(main): Call it.
Signed-off-by: Siddhesh Poyarekar <siddhesh@gotplt.org>
On some systems /etc/localtime is a tzfile, not a symlink to one. We
cannot use it to determine the current time zone in that case. See if
/etc/sysconfig/clock sets the variable DEFAULT_TIMEZONE instead.
libstdc++-v3/ChangeLog:
PR libstdc++/108530
* src/c++20/tzdb.cc (current_zone): Look for DEFAULT_TIMEZONE in
/etc/sysconfig/clock.
In order for std::make_from_tuple to work with tuple-like types, the
overloads of std::get for those types must have been declared before the
definition of std::make_from_tuple. That means we need to include the
definition of std::ranges::subrange in <tuple>.
The definitions of std::pair and its overloads of std::get are already
included in <tuple>. We provide forward declarations of std::array and
its std::get overloads in <tuple>. We could just declare subrange
without defining it, and give ranges::get a non-deduced return type,
like so:
namespace ranges
{
enum class subrange_kind : bool { unsized, sized};
template<input_or_output_iterator I, sentinel_for<I> S,
subrange_kind K>
requires (K == subrange_kind::sized || !sized_sentinel_for<S, I>)
class subrange;
template<size_t _Num, class _It, class _Sent, subrange_kind _Kind>
requires (_Num < 2)
constexpr __conditional_t<_Num == 0, _It, _Sent>
get(const subrange<_It, _Sent, _Kind>& __r);
template<size_t _Num, class _It, class _Sent, subrange_kind _Kind>
requires (_Num < 2)
constexpr __conditional_t<_Num == 0, _It, _Sent>
get(subrange<_It, _Sent, _Kind>&& __r)
}
using ranges::get;
It is a bit late in the GCC 13 dev cycle to do this, so just include the
right headers for now.
Also add the dangling check to std::make_from_tuple added by P2255.
libstdc++-v3/ChangeLog:
PR libstdc++/102301
* include/bits/ranges_base.h: Include <bits/stl_iterator.h> for
std::make_reverse_iterator.
* include/std/tuple: Include <bits/ranges_util.h> for subrange.
(make_from_tuple): Add static assertion from P2255 to diagnose
dangling references.
* testsuite/20_util/tuple/make_from_tuple/dangling_ref.cc: New test.
* testsuite/20_util/tuple/make_from_tuple/tuple_like.cc: New test.
The test fails on targets where size_t is not unsigned long
due to extra diagnostics.
As the testcase is tested in C++98 too, I'm not using decltype (sizeof 0)
but __SIZE_TYPE__.
2023-01-25 Jakub Jelinek <jakub@redhat.com>
PR c++/107329
PR testsuite/108533
* g++.dg/init/new51.C (size_t): New typedef.
(RexxClass::operator new, RexxClass::operator delete): Use size_t
instead of unsigned long.
Since r11-2095 we pass flags to cp_lexer_get_preprocessor_token, and
cp_lexer_new_main passes C_LEX_STRING_NO_JOIN when lexing most of the
translation unit, but doesn't do that for the very first token; as a
result, if the first token is a string literal, we try to join strings and
get confused if that encounters a pragma.
PR c++/108504
gcc/cp/ChangeLog:
* parser.cc (cp_lexer_new_main): Pass C_LEX_STRING_NO_JOIN for first
token, too.
gcc/testsuite/ChangeLog:
* g++.dg/ext/pragma1.C: New test.
tsubst_lambda_expr uses build_memfn_type to build a METHOD_TYPE for the new
lamba op(). This is not what we want for a C++23 static op(), but since we
also use that METHOD_TYPE to communicate the closure type down to
tsubst_function_decl, let's wait and turn it back at that point.
PR c++/108526
gcc/cp/ChangeLog:
* pt.cc (tsubst_function_decl): Handle static lambda.
gcc/testsuite/ChangeLog:
* g++.dg/cpp23/static-operator-call5.C: New test.
In the previously posted patch
"xtensa: Make complex hard register clobber elimination more robust and accurate",
the check code for insns that refer to the [DS]Cmode hard register before
it is overwritten after it is clobbered is incomplete. Fortunately such
insns are seldom emitted, so it didn't matter.
This patch fixes that for the sake of completeness.
gcc/ChangeLog:
* config/xtensa/xtensa.md:
Fix exit from loops detecting references before overwriting in the
split pattern.
LRA does elimination but not always checks insn constraints in this case.
This results in LRA failure for PDP11 target whose addition is only 2-op insn.
The same might happen for other analogous targets. The patch fixes this problem.
PR rtl-optimization/108388
gcc/ChangeLog:
* lra-constraints.cc (get_hard_regno): Remove final_p arg. Always
do elimination but only for hard register.
(operands_match_p, uses_hard_regs_p, process_alt_operands): Adjust
calls of get_hard_regno.
gcc/testsuite/ChangeLog:
* gcc.target/pdp11/pdp11.exp: New.
* gcc.target/pdp11/pr108388.c: New.
gcc/fortran/ChangeLog:
PR fortran/108529
* simplify.cc (simplify_transformation): Do not try to simplify
transformational intrinsic when the ARRAY argument has a NULL shape.
gcc/testsuite/ChangeLog:
PR fortran/108529
* gfortran.dg/pr108529.f90: New test.
In the context of D the interpretation of S390, S390X, and SystemZ is a
bit fuzzy. The wording S390X was wrongly deprecated in favour of
SystemZ by commit
3b50a4c3fa
Thus, SystemZ is used for 64-bit targets, now, and S390 for 31-bit
targets. However, in TARGET_D_CPU_VERSIONS depending on TARGET_ZARCH we
set the CPU version to SystemZ. This is also the case if compiled for
31-bit targets leading to the following error:
libphobos/libdruntime/core/sys/posix/sys/stat.d:967:13: error: static assert: '96u == 144u' is false
967 | static assert(stat_t.sizeof == 144);
| ^
Thus in order to keep this patch simple I went for keeping SystemZ for
64-bit targets and S390, as usual, for 31-bit targets and dropped the
distinction between ESA and z/Architecture.
gcc/ChangeLog:
* config/s390/s390-d.cc (s390_d_target_versions): Fix detection
of CPU version.
This patch allows a subsequent patch to turn on/off the consuming
of C comments.
gcc/m2/ChangeLog:
* m2.flex (cpreprocessor): Add temporary variable
which is initialized to 0.
(commentCLevel): New variable.
(endOfCComment): New function.
(splitSlashStar): New function to split /* into / and *
tokens.
(COMMENTC): New flex state.
("/*"): New rule to test whether we should treat /*
as a single token or as two tokens.
(<COMMENTC>.): New rule to skip a character.
(<COMMENTC>\n.*): New rule to consume the line.
(<COMMENTC>"*/"): New rule to call endOfCComment.
Signed-off-by: Gaius Mulley <gaiusmod2@gmail.com>
RTco is a definition for C module and therefore there is no
RTco.mod. The RTco.cc uses RTentity and the import in RTco.def
ensures that cc1gm2 can build a graph of all dependencies
should -fscaffold-static be used.
gcc/m2/ChangeLog:
* gm2-libs-iso/RTco.def: Import RTentity.
Declare RTco as a definition for C module.
Signed-off-by: Gaius Mulley <gaiusmod2@gmail.com>
This patch adds the memory operand of MVE masked stores as input operands to
mimic the 'partial' writes, to prevent erroneous write-after-write
optimizations as described in the PR.
gcc/ChangeLog:
PR target/108177
* config/arm/mve.md (mve_vstrbq_p_<supf><mode>, mve_vstrhq_p_fv8hf,
mve_vstrhq_p_<supf><mode>, mve_vstrwq_p_<supf>v4si): Add memory operand
as input operand.
gcc/testsuite/ChangeLog:
* gcc.target/arm/mve/pr108177-1-run.c: New test.
* gcc.target/arm/mve/pr108177-1.c: New test.
* gcc.target/arm/mve/pr108177-10-run.c: New test.
* gcc.target/arm/mve/pr108177-10.c: New test.
* gcc.target/arm/mve/pr108177-11-run.c: New test.
* gcc.target/arm/mve/pr108177-11.c: New test.
* gcc.target/arm/mve/pr108177-12-run.c: New test.
* gcc.target/arm/mve/pr108177-12.c: New test.
* gcc.target/arm/mve/pr108177-13-run.c: New test.
* gcc.target/arm/mve/pr108177-13.c: New test.
* gcc.target/arm/mve/pr108177-14-run.c: New test.
* gcc.target/arm/mve/pr108177-14.c: New test.
* gcc.target/arm/mve/pr108177-2-run.c: New test.
* gcc.target/arm/mve/pr108177-2.c: New test.
* gcc.target/arm/mve/pr108177-3-run.c: New test.
* gcc.target/arm/mve/pr108177-3.c: New test.
* gcc.target/arm/mve/pr108177-4-run.c: New test.
* gcc.target/arm/mve/pr108177-4.c: New test.
* gcc.target/arm/mve/pr108177-5-run.c: New test.
* gcc.target/arm/mve/pr108177-5.c: New test.
* gcc.target/arm/mve/pr108177-6-run.c: New test.
* gcc.target/arm/mve/pr108177-6.c: New test.
* gcc.target/arm/mve/pr108177-7-run.c: New test.
* gcc.target/arm/mve/pr108177-7.c: New test.
* gcc.target/arm/mve/pr108177-8-run.c: New test.
* gcc.target/arm/mve/pr108177-8.c: New test.
* gcc.target/arm/mve/pr108177-9-run.c: New test.
* gcc.target/arm/mve/pr108177-9.c: New test.
* gcc.target/arm/mve/pr108177-main.x: New test include.
* gcc.target/arm/mve/pr108177.x: New test include.
The SYSROOT_SUFFIX_SPEC works even when multilib is disabled.
So when build no-multilib glibc toolchain and the options are
not same as MULTILIB_DEFAULTS, the sysroot will specify wrong
because the libc will not be installed as such.
This bug causes glibc regression test error:
https://sourceware.org/pipermail/libc-testresults/2023q1/010706.html
The error is:
/scratch/jmyers/glibc-bot/install/compilers/csky-linux-gnuabiv2/csky-glibc-linux-gnuabiv2/bin/ld: cannot find -lc: No such file or directory
gcc/
* config.gcc(csky-*-linux*): Define CSKY_ENABLE_MULTILIB
and only include 'csky/t-csky-linux' when enable multilib.
* config/csky/csky-linux-elf.h(SYSROOT_SUFFIX_SPEC): Don't
define it when disable multilib.
CFG cleanup computes dominators before the loop over blocks looking
for merging opportunities. That computes also the fast-query DFS
numbers but that's a bit pointless since any CFG cleanup will invalidate
them immediately (they are re-computed before fixing up loops).
The following avoids this and fixes the SIGSEGV due to the deep
recursion in assign_dfs_numbers after inlining very many small
functions.
PR tree-optimization/108500
* dominance.h (calculate_dominance_info): Add parameter
to indicate fast-query compute, defaulted to true.
* dominance.cc (calculate_dominance_info): Honor
fast-query compute parameter.
* tree-cfgcleanup.cc (cleanup_tree_cfg_noloop): Do
not compute the dominator fast-query DFS numbers.
Fix an obvious copy-and-paste error where ptr1 was used instead of ptr2.
This bug caused the dump file produced by -fdump-ipa-inline-details to
not correctly show the difference in target options when a function
could not be inlined due to a target option mismatch.
gcc/ChangeLog:
PR bootstrap/90543
* optc-save-gen.awk: Fix copy-and-paste error.
Signed-off-by: Eric Biggers <ebiggers@google.com>
As reported by Andrew Pinski, structured bindings (with the exception
of the ones using std::tuple_{size,element} and get which are really
standalone variables in addition to the binding one) also use
DECL_VALUE_EXPR and needs the same treatment in static initializers.
On Sun, Jan 22, 2023 at 07:19:07PM -0500, Jason Merrill wrote:
> Though, actually, why not instead fix expand_expr_real_1 (and staticp) to
> look through DECL_VALUE_EXPR?
Doing it when emitting the initializers seems to be too late to me,
we in various spots try to put parts of the static var DECL_INITIAL expressions
into the IL, or e.g. for varpool purposes remember which vars are referenced
there.
This patch moves it to record_reference, which is called from varpool_node::analyze
and so about the same time as gimplification of the bodies which also
replaces DECL_VALUE_EXPRs.
2023-01-24 Jakub Jelinek <jakub@redhat.com>
PR c++/108474
* cgraphbuild.cc: Include gimplify.h.
(record_reference): Replace VAR_DECLs with DECL_HAS_VALUE_EXPR_P with
their corresponding DECL_VALUE_EXPR expressions after unsharing.
* cp-gimplify.cc (cp_fold_r): Revert 2023-01-19 changes.
* g++.dg/cpp1z/decomp57.C: New test.
* g++.dg/cpp1z/decomp58.C: New test.
