Because uses-allocator construction is invariably done with a const
lvalue the __uses_alloc helper should use a const lvalue for the
is_constructible checks. Otherwise, it can detect that the type can be
constructed from an rvalue, and then an error happens when a const
lvalue is passed to the constructor instead.
Prior to LWG DR 2586 scoped_allocator_adaptor incorrectly used an rvalue
type in the is_constructible check and then used a non-const lvalue for
the actual construction. The other components using uses-allocator
construction (tuple and polymorphic_allocator) have always done so with
a const lvalue allocator, although the use of __use_alloc in our
implementation meant they behaved the same as scoped_allocator_adaptor
and incorrectly used rvalues for the is_constructible checks.
In C++20 the P0591R4 changes mean that all uses-allocator construction
is defined in terms of the new uses_allocator_construction_args
functions, which always use a const lvalue allocator.
The changes in this patch ensure that the __use_alloc helper correctly
matches the requirements in the standard, consistently using a const
lvalue allocator for the is_constructible checks and the actual
constructor arguments.
* doc/xml/manual/intro.xml: Document LWG 2586 status.
* include/bits/uses_allocator.h (__uses_alloc): Use const lvalue
allocator type in is_constructible checks.
* testsuite/20_util/scoped_allocator/69293_neg.cc: Adjust dg-error.
* testsuite/20_util/scoped_allocator/dr2586.cc: New test.
* testsuite/20_util/tuple/cons/allocators.cc: Add test using
problematic type from LWG 2586 discussion.
* testsuite/20_util/uses_allocator/69293_neg.cc: Adjust dg-error.
* testsuite/20_util/uses_allocator/cons_neg.cc: Likewise.
From-SVN: r268882
When this testcase was introduced it failed to account for the possibility of
targets that do not support arm mode or that do not support the ldrd/strd
instructions.
This patch accounts for both of these by adding some
dg-require-effective-target lines to the testcase.
This patch also adds a new effective-target procedure to check a target
supports ldrd/strd.
This patch also adds a new effective-target procedure to check a target
supports arm ldrd/strd.
The check uses the 'r' constraint to ensure SP is not used so that it will work
for thumb mode code generation as well as arm mode.
Tested by running this testcase with cross compilers using "-march=armv5t",
"-mcpu=cortex-m3", "-mcpu-arm7tdmi", "-mcpu=cortex-a9 -march=armv5t" for both
arm-none-eabi and arm-none-linux-gnueabihf.
Also ran this testcase with `make check` natively.
gcc/testsuite/ChangeLog:
2019-02-14 Matthew Malcomson <matthew.malcomson@arm.com>
* gcc.dg/rtl/arm/ldrd-peepholes.c: Restrict testcase.
* lib/target-supports.exp: Add procedure to check for ldrd.
From-SVN: r268881
This change is safe to make now (in stage 4), because the constructors
are currently incorrect and unusable (unless the supplied container
already contains a heap, in which case the new make_heap calls are
redundant but harmless).
* doc/xml/manual/intro.xml: Document LWG 2537 status.
* include/bits/stl_queue.h
(priority_queue(const Compare&, const Container&, const Alloc&))
(priority_queue(const Compare&, Container&&, const Alloc&)): Call
make_heap.
* testsuite/23_containers/priority_queue/dr2537.cc: New test.
From-SVN: r268878
Although there is no good use for stack<int, deque<double>> or similar
types with a mismatched value_type, it's possible somebody is doing that
and getting away with it currently. This patch only enforces the new
requirement for C++17 and later. During stage 1 we should consider
enforcing it for C++11 and C++14.
* doc/xml/manual/intro.xml: Document LWG 2566 status.
* include/bits/stl_queue.h (queue, priority_queue): Add static
assertions to enforce LWG 2566 requirement on value_type.
* include/bits/stl_stack.h (stack): Likewise.
From-SVN: r268877
The OpenACC 'resolve_oacc_nested_loops' function duplicates most code of the
OpenMP 'resolve_omp_do', but didn't include the PR60127 "ICE with OpenMP and DO
CONCURRENT" (trunk r210331) changes. (Probably the two functions should be
unified?)
The Fortran DO CONCURRENT construct is a way to tell the compiler that loop
iterations don't have any interdependencies -- which is information that would
very well be suitable for OpenACC/OpenMP loops. There are some "details"
however, see the discussion/references in PR60127, so for the time being, make
this a compile-time error instead of an ICE.
gcc/fortran/
* openmp.c (resolve_oacc_nested_loops): Error on do concurrent
loops.
gcc/testsuite/
* gfortran.dg/goacc/loop-3-2.f95: Error on do concurrent loops.
* gfortran.dg/goacc/loop-3.f95: Likewise.
* gfortran.dg/goacc/pr72715.f90: New test.
Reviewed-by: Thomas Schwinge <thomas@codesourcery.com>
From-SVN: r268875
2019-02-14 Martin Liska <mliska@suse.cz>
PR rtl-optimization/89242
* dce.c (delete_unmarked_insns): Call free_dominance_info we
process a transformation.
2019-02-14 Martin Liska <mliska@suse.cz>
PR rtl-optimization/89242
* g++.dg/pr89242.C: New test.
From-SVN: r268873
This DR was already resolved for GCC 7.1 by the implementation of DR
2192, but we didn't have an explicit test for the behaviour that 2735
guarantees.
* doc/xml/manual/intro.xml: Document LWG 2735 status.
* include/bits/std_abs.h: Add comment about LWG 2735.
* testsuite/26_numerics/headers/cstdlib/dr2735.cc: New test.
From-SVN: r268867
gcc/:
* optc-save-gen.awk: Set var_opt_hash for initial optimizations
and set current index for other optimizations.
gcc/testsuite/:
* gcc.dg/func-attr-1.c: New test.
From-SVN: r268860
Clang defines the __cpp_impl_destroying_delete macro unconditionally, so
that the feature is supported whenever the library type is defined. This
is incompatible with the current definition in libstdc++ because we use
constexpr and inline variables, which will give an error for older -std
modes.
This patch defines the destroying_delete_t type and destroying_delete
variable independently of the __cpp_impl_destroying_delete macro, but
only for C++2a (because the names aren't reserved for previous
standards). The __cpp_lib_destroying_delete macro is only defined when
both the library type and compiler macro are defined (i.e. when the type
can actually be used as intended).
PR libstdc++/89345
* include/std/version [__cpp_impl_destroying_delete]
(__cpp_lib_destroying_delete): Only define for C++2a and later.
* libsupc++/new [__cpp_impl_destroying_delete]
(__cpp_lib_destroying_delete): Likewise.
(destroying_delete_t, destroying_delete): Likewise, but define even
when __cpp_impl_destroying_delete is not defined.
* testsuite/18_support/destroying_delete.cc: New test.
From-SVN: r268856
It's too risky to reuse the type field for USING_DECL_SCOPE.
Language-independent parts of the compiler, such as location and
non-lvalue wrappers, happily take the TREE_TYPE of a USING_DECL as if
it was a type rather than an unrelated scope.
For better or worse, USING_DECLs use the non-common struct so we can
use the otherwise unused result field. Adjust fallout, from uses of
TREE_TYPE that were supposed to be USING_DECL_SCOPE, to other
accidental uses of TREE_TYPE of a USING_DECL.
for gcc/cp/ChangeLog
PR c++/86379
* cp-tree.h (USING_DECL_SCOPE): Use result rather than type.
* name-lookup.c (strip_using_decl): Use USING_DECL_SCOPE.
* search.c (protected_accessible_p): Follow USING_DECL_DECLS.
(shared_member_p): Likewise.
(lookup_member): Likewise.
* decl.c (grok_special_member_properties): Skip USING_DECLs.
* semantics.c (finish_omp_declare_simd_methods): Likewise.
(finish_qualified_id_expr): Do not call shared_member_p with
a dependent expr.
for gcc/testsuite/ChangeLog
PR c++/86379
* g++.dg/cpp0x/pr86379.C: New.
From-SVN: r268851
A lambda capture variable initialized with a lambda expr taking more
than one parameter got us confused.
The first problem was that the parameter list was cut short during
tsubsting because we tsubsted it with cp_unevaluated_operand. We
reset it right after, to tsubst the function body, so I've moved the
reset up so that it's in effect while processing the parameters as
well.
The second problem was that the lambda expr appeared twice, once in a
decltype that gave the capture variable its type, and once in its
initializer. This caused us to instantiate two separate lambda exprs
and closure types, and then to flag that the lambda expr in the
initializer could not be converted to the unrelated closure type
determined for the capture variable. Recording the tsubsted expr in
the local specialization map, and retrieving it for reuse fixed it.
However, that required some care to avoid reusing the lambda expr
across different indices in pack expansions.
for gcc/cp/ChangeLog
PR c++/87322
* pt.c (tsubst_lambda_expr): Avoid duplicate tsubsting.
Move cp_evaluated resetting before signature tsubsting.
(gen_elem_of_pack_expansion_instantiation): Separate local
specializations per index.
for gcc/testsuite/ChangeLog
PR c++/87322
* g++.dg/cpp1y/pr87322.C: New.
* g++.dg/cpp0x/lambda/lambda-variadic5.C: Test that we
instantiate the expected number of lambda functions.
From-SVN: r268850
This patch fixes an ICE in the Thumb-1 LDM peepholer. Thumb-1 LDMs
always update the base register except if the base is loaded.
The current implementation rejects LDMs where the base is not dead,
however this doesn't exclude the case where the base is loaded as
well as dead. Fix this by explicitly checking whether the base is
loaded. Also enable LDMs which load the first register.
gcc/
PR target/89190
* config/arm/arm.c (ldm_stm_operation_p) Set
addr_reg_in_reglist correctly for first register.
(load_multiple_sequence): Remove dead base check.
(gen_ldm_seq): Correctly set write_back for Thumb-1.
testsuite/
PR target/89190
* gcc.target/arm/pr89190.c: New test.
From-SVN: r268848
PR c++/89036 reports an ICE due to this assertion failing
1136 /* A class should never have more than one destructor. */
1137 gcc_assert (!current_fns || via_using || !DECL_DESTRUCTOR_P (method));
on this template with a pair of dtors, with
mutually exclusive "requires" clauses:
template<typename T>
struct Y {
~Y() requires(true) = default;
~Y() requires(false) {}
};
Nathan introduced this assertion as part of:
ca9219bf18c68a001d62ecb981bc9176b0feaf12 (aka r251340):
2017-08-24 Nathan Sidwell <nathan@acm.org>
Conversion operators kept on single overload set
which, amongst other changes to add_method had this:
/* A class should never have more than one destructor. */
- if (current_fns && DECL_MAYBE_IN_CHARGE_DESTRUCTOR_P (method))
- return false;
+ gcc_assert (!current_fns || !DECL_DESTRUCTOR_P (method));
The following patch drops the assertion (I already had to generalize
the assertion in r268041 to fix PR c++/88699).
gcc/cp/ChangeLog:
PR c++/89036
* class.c (add_method): Drop destructor assertion.
gcc/testsuite/ChangeLog:
PR c++/89036
* g++.dg/concepts/pr89036.C: New test.
From-SVN: r268847
On AArch64 aarch64_classify_address has a case for when it's non-strict
that will allow it to accept any byte offset from a reg when validating
an address in a given addressing mode.
This because reload would later make the address valid. SVE however requires
the address always be valid, but currently allows any address when a MEM +
offset is used. This causes an ICE as nothing later forces the address to be
legitimate.
The patch forces aarch64_emit_sve_pred_move via expand_insn to ensure that
the addressing mode is valid for any loads/stores it creates, which follows
the SVE way of handling address classifications.
gcc/ChangeLog:
PR target/88847
* config/aarch64/aarch64-sve.md (*pred_mov<mode>, pred_mov<mode>):
Expose as @aarch64_pred_mov.
* config/aarch64/aarch64.c (aarch64_classify_address):
Use expand_insn which legitimizes operands.
gcc/testsuite/ChangeLog:
PR target/88847
* gcc.target/aarch64/sve/pr88847.c: New test.
From-SVN: r268845
2019-02-13 Jakub Jelinek <jakub@redhat.com>
PR middle-end/89303
* tree-ssa-structalias.c (set_uids_in_ptset): Or in vi->is_heap_var
into pt->vars_contains_escaped_heap instead of setting
pt->vars_contains_escaped_heap to it.
2019-02-13 Jonathan Wakely <jwakely@redhat.com>
Jakub Jelinek <jakub@redhat.com>
PR middle-end/89303
* g++.dg/torture/pr89303.C: New test.
From-SVN: r268843
2019-02-13 Martin Liska <mliska@suse.cz>
PR fortran/88649
* resolve.c (resolve_operator): Initialize 't' right
after function entry. Skip switch (e->value.op.op)
for -fdec operands that become function calls.
From-SVN: r268842
PR middle-end/89281
* optabs.c (prepare_cmp_insn): Use UINTVAL (size) instead of
INTVAL (size), compare it to GET_MODE_MASK instead of
1 << GET_MODE_BITSIZE.
From-SVN: r268841
PR target/89290
* config/i386/predicates.md (x86_64_immediate_operand): Allow
TLS UNSPECs offsetted by signed 32-bit CONST_INT even with
-mcmodel=large.
* gcc.target/i386/pr89290.c: New test.
From-SVN: r268837
In the gcc.backtrace module, either one of LibBacktrace or
UnwindBacktrace will always be defined. Giving UnwindBacktrace a higher
precedence over the libc backtrace as the default handler because the
latter depends on a rt.backtrace module that is not compiled in.
libphobos/ChangeLog:
* libdruntime/core/runtime.d (defaultTraceHandler): Give
UnwindBacktrace handler precedence over backtrace.
From-SVN: r268836
2019-02-13 Martin Liska <mliska@suse.cz>
PR lto/88858
* cfgrtl.c (remove_barriers_from_footer): New function.
(try_redirect_by_replacing_jump): Use it.
(cfg_layout_redirect_edge_and_branch): Likewise.
From-SVN: r268835
The 5 new builtins vec_sbox_be, vec_cipher_be, vec_cipherlast_be, vec_ncipher_be
and vec_ncipherlast_be only support vector unsigned char type parameters.
Add new instruction crypto_vsbox_<mode> and crypto_<CR_insn>_<mode> to handle
them accordingly, where the new mode CR_vqdi can be expanded to vector unsigned
long long for none _be postfix builtins or vector unsigned char for _be postfix
builtins.
gcc/ChangeLog
2019-02-13 Xiong Hu Luo <luoxhu@linux.vnet.ibm.com>
* config/rs6000/altivec.h (vec_sbox_be, vec_cipher_be,
vec_cipherlast_be, vec_ncipher_be, vec_ncipherlast_be): New #defines.
* config/rs6000/crypto.md (CR_vqdi): New define_mode_iterator.
(crypto_vsbox_<mode>, crypto_<CR_insn>_<mode>): New define_insns.
* config/rs6000/rs6000-builtin.def (VSBOX_BE): New BU_CRYPTO_1.
(VCIPHER_BE, VCIPHERLAST_BE, VNCIPHER_BE, VNCIPHERLAST_BE):
New BU_CRYPTO_2.
* config/rs6000/rs6000.c (builtin_function_type)
<CRYPTO_BUILTIN_VSBOX_BE, CRYPTO_BUILTIN_VCIPHER_BE,
CRYPTO_BUILTIN_VCIPHERLAST_BE, CRYPTO_BUILTIN_VNCIPHER_BE,
CRYPTO_BUILTIN_VNCIPHERLAST_BE>: New switch options.
* doc/extend.texi (vec_sbox_be, vec_cipher_be, vec_cipherlast_be,
vec_ncipher_be, vec_ncipherlast_be): New builtin functions.
gcc/testsuite/ChangeLog
2019-02-13 Xiong Hu Luo <luoxhu@linux.vnet.ibm.com>
* gcc.target/powerpc/crypto-builtin-1.c
(crypto1_be, crypto2_be, crypto3_be, crypto4_be, crypto5_be):
New testcases.
From-SVN: r268834
In this PR, we were unnecessarily rejecting a constexpr initializer_list
with no elements. This seems like a fairly useless degenerate case, but it
makes sense to avoid allocating an underlying array at all if there are no
elements and instead use a null pointer, like the initializer_list default
constructor.
If the (automatic storage duration) list does have initializer elements, we
continue to reject the declaration, because the initializer_list ends up
referring to an automatic storage duration temporary array, which is not a
suitable constant initializer. If we make it static, it should be OK
because we refer to a static array. The second hunk fixes that case. It
also means we won't diagnose some real errors in templates, but those
diagnostics aren't required, and we'll get them when the template is
instantiated.
* call.c (convert_like_real) [ck_list]: Don't allocate a temporary
array for an empty list.
* typeck2.c (store_init_value): Don't use cxx_constant_init in a
template.
From-SVN: r268827
i386 backend has
INT_MODE (OI, 32);
INT_MODE (XI, 64);
So, XI_MODE represents 64 INTEGER bytes = 64 * 8 = 512 bit operation,
in case of const_1, all 512 bits set.
We can load zeros with narrower instruction, (e.g. 256 bit by inherent
zeroing of highpart in case of 128 bit xor), so TImode in this case.
Some targets prefer V4SF mode, so they will emit float xorps for zeroing
Then the introduction of AVX512F fubared everything by overloading the
meaning of insn mode.
How should we use INSN mode, MODE_XI, in standard_sse_constant_opcode
and patterns which use standard_sse_constant_opcode? 2 options:
1. MODE_XI should only used to check if EXT_REX_SSE_REG_P is true
in any register operand. The operand size must be determined by operand
itself , not by MODE_XI. The operand encoding size should be determined
by the operand size, EXT_REX_SSE_REG_P and AVX512VL.
2. MODE_XI should be used to determine the operand encoding size.
EXT_REX_SSE_REG_P and AVX512VL should be checked for encoding
instructions.
gcc/
PR target/89229
* config/i386/i386.md (*movoi_internal_avx): Revert revision
268678 and revision 268657.
(*movti_internal): Likewise.
gcc/testsuite/
PR target/89229
* gcc.target/i386/pr89229-1.c: New test.
From-SVN: r268811
The following insn:
(insn (set (reg:DI %r2)
(sign_extend:DI (mem:SI
(const:DI (plus:DI (symbol_ref:DI ("*.LC0"))
(const_int 16)))))))
is correctly recognized by LRA as RIL alternative of extendsidi2
define_insn. However, when recognition runs after LRA, it returns RXY
alternative, which is incorrect, since the offset 16 points past the
end of of *.LC0 literal pool entry. Such addresses are normally
rejected by s390_decompose_address ().
This inconsistency confuses annotate_constant_pool_refs: the selected
alternative makes it proceed with annotation, only to find that the
annotated address is invalid, causing ICE.
This patch fixes the root cause, namely, that s390_check_qrst_address ()
behaves differently during and after LRA.
gcc/ChangeLog:
2019-02-12 Ilya Leoshkevich <iii@linux.ibm.com>
PR target/89233
* config/s390/s390.c (s390_decompose_address): Update comment.
(s390_check_qrst_address): Reject invalid address forms after
LRA.
gcc/testsuite/ChangeLog:
2019-02-12 Ilya Leoshkevich <iii@linux.ibm.com>
PR target/89233
* gcc.target/s390/pr89233.c: New test.
From-SVN: r268798
The call to bsearch in dwarf_lookup_pc can have NULL as base argument when
the nmemb argument is 0. The base argument is required to be pointing to the
initial member of an array of nmemb objects. It is not specified what
constitutes a valid pointer to an array of 0 objects, but glibc declares base
with attribute non-null, so the NULL will trigger a sanitizer runtime error.
Fix this by only calling bsearch if nmemb != 0.
2019-02-12 Tom de Vries <tdevries@suse.de>
PR libbacktrace/81983
* dwarf.c (dwarf_lookup_pc): Don't call bsearch if nmemb == 0.
From-SVN: r268796
2019-02-12 Martin Liska <mliska@suse.cz>
PR lto/88876
* ipa-pure-const.c (propagate_pure_const): Revert hunk as
we need default values of funct_state for a function that
is not optimized.
From-SVN: r268795
When a node is removed from a splay tree, the splay tree was
not using the function splay_tree_delete_key_fn to release the key.
This was causing a leak, fixed by Tom Tromey.
This patch fixes another key leak, that happens when a key equal to
a key already present is inserted. In such a case, we have to release
the old KEY.
Note that this is based on the assumption that the caller always
allocates a new KEY when doing an insert.
Also, clarify the documentation about when the release functions are
called.
2019-02-11 Philippe Waroquiers <philippe.waroquiers@skynet.be>
* splay-tree.h (splay_tree_delete_key_fn): Update comment.
(splay_tree_delete_value_fn): Likewise.
libiberty/ChangeLog
2019-02-11 Philippe Waroquiers <philippe.waroquiers@skynet.be>
* splay-tree.c (splay_tree_insert): Also release old KEY in case
of insertion of a key equal to an already present key.
(splay_tree_new_typed_alloc): Update comment.
From-SVN: r268793
* asan.c (asan_expand_mark_ifn): Take into account the alignment of
the object to pick the size of stores on strict-alignment platforms.
* config/sparc/sparc.md (*movsi_insn): Minor tweak.
(*movdi_insn_sp32): Likewise.
(*movdi_insn_sp64): Likewise.
From-SVN: r268792
PR lto/88777
* cgraphunit.c (analyze_functions): Clear READONLY flag for external
types that needs constructiong.
* tree.h (may_be_aliased): Do not check TYPE_NEEDS_CONSTRUCTING.
From-SVN: r268791
2019-02-12 Richard Biener <rguenther@suse.de>
PR tree-optimization/89253
* tree-ssa-loop-split.c (tree_ssa_split_loops): Check we can
duplicate the loop.
* gfortran.dg/pr89253.f: New testcase.
From-SVN: r268790
PR lto/88147 reports an assertion failure due to a bogus location_t value
when adding a line to a pre-existing line map, when there's a large
difference between the two line numbers.
For some "large differences", this leads to a location_t value that exceeds
LINE_MAP_MAX_LOCATION, in which case linemap_line_start returns 0. This
isn't ideal, but at least should lead to safe degradation of location
information.
However, if the difference is very large, it's possible for the line
number offset (relative to the start of the map) to be sufficiently large
that overflow occurs when left-shifted by the column-bits, and hence
the check against the LINE_MAP_MAX_LOCATION limit fails, leading to
a seemingly-valid location_t value, but encoding the wrong location. This
triggers the assertion failure:
linemap_assert (SOURCE_LINE (map, r) == to_line);
The fix (thanks to Martin) is to check for overflow when determining
whether to reuse an existing map, and to not reuse it if it would occur.
gcc/ChangeLog: David Malcolm <dmalcolm@redhat.com>
PR lto/88147
* input.c (selftest::test_line_offset_overflow): New selftest.
(selftest::input_c_tests): Call it.
libcpp/ChangeLog: Martin Liska <mliska@suse.cz>
PR lto/88147
* line-map.c (linemap_line_start): Don't reuse the existing line
map if the line offset is sufficiently large to cause overflow
when computing location_t values.
From-SVN: r268789
