pass_return_slot::execute has:
/* Ignore internal functions without direct optabs,
those are expanded specially and aggregate_value_p
on their result might result in undesirable warnings
with some backends. */
&& (!gimple_call_internal_p (stmt)
|| direct_internal_fn_p (gimple_call_internal_fn (stmt)))
&& aggregate_value_p (TREE_TYPE (gimple_call_lhs (stmt)),
gimple_call_fndecl (stmt)))
But what the comment says applies to directly-mapped internal functions
too, since they're only used if the target supports them without a
libcall.
This was triggering an ICE on the attached testcase. The svld3 call
is folded to an IFN_LOAD_LANES, which returns an array of vectors with
VNx48QImode. Since no such return type can exist in C, the target hook
was complaining about an unexpected use of SVE modes. (And we want to
keep asserting for that, so that we don't accidentally define an ABI for
an unexpected corner case.)
2020-01-09 Richard Sandiford <richard.sandiford@arm.com>
gcc/
* tree-nrv.c (pass_return_slot::execute): Handle all internal
functions the same way, rather than singling out those that
aren't mapped directly to optabs.
gcc/testsuite/
* gcc.target/aarch64/sve/acle/general/nrv_1.c: New test.
From-SVN: r280048
One problem with adding an N-bit vector extension to an existing
architecture is to decide how N-bit vectors should be passed to
functions and returned from functions. Allowing all N-bit vector
types to be passed in registers breaks backwards compatibility,
since N-bit vectors could be used (and emulated) before the vector
extension was added. But always passing N-bit vectors on the
stack would be inefficient for things like vector libm functions.
For SVE we took the compromise position of predefining new SVE vector
types that are distinct from all existing vector types, including
GNU-style vectors. The new types are passed and returned in an
efficient way while existing vector types are passed and returned
in the traditional way. In the right circumstances, the two types
are inter-convertible.
The SVE types are created using:
vectype = build_distinct_type_copy (vectype);
SET_TYPE_STRUCTURAL_EQUALITY (vectype);
TYPE_ARTIFICIAL (vectype) = 1;
The C frontend maintains this distinction, using VIEW_CONVERT_EXPR
to convert from one type to the other. However, the distinction can
be lost during gimple, which treats two vector types with the same
mode, number of elements, and element type as equivalent. And for
most targets that's the right thing to do.
This patch therefore adds a hook that lets the target choose
whether such vector types are indeed equivalent.
Note that the new tests fail for -mabi=ilp32 in the same way as other
ACLE-based tests. I'm still planning to fix that as a follow-on.
2020-01-09 Richard Sandiford <richard.sandiford@arm.com>
gcc/
* target.def (compatible_vector_types_p): New target hook.
* hooks.h (hook_bool_const_tree_const_tree_true): Declare.
* hooks.c (hook_bool_const_tree_const_tree_true): New function.
* doc/tm.texi.in (TARGET_COMPATIBLE_VECTOR_TYPES_P): New hook.
* doc/tm.texi: Regenerate.
* gimple-expr.c: Include target.h.
(useless_type_conversion_p): Use targetm.compatible_vector_types_p.
* config/aarch64/aarch64.c (aarch64_compatible_vector_types_p): New
function.
(TARGET_COMPATIBLE_VECTOR_TYPES_P): Define.
* config/aarch64/aarch64-sve-builtins.cc (gimple_folder::convert_pred):
Use the original predicate if it already has a suitable type.
gcc/testsuite/
* gcc.target/aarch64/sve/pcs/gnu_vectors_1.c: New test.
* gcc.target/aarch64/sve/pcs/gnu_vectors_2.c: Likewise.
From-SVN: r280047
This prevents the vtables and RTTI from being emitted in every object
file that uses memory_resource and monotonic_buffer_resource.
Objects compiled by GCC 9.1 or 9.2 will contain inline definitions of
the destructors, vtable and RTTI, but this is harmless. The inline
definitions have identical effects to the ones that are now defined in
libstdc++.so so it doesn't matter if the inline ones are used instead of
calling the symbols exported from the runtime library.
PR libstdc++/93208
* config/abi/pre/gnu.ver: Add new exports.
* include/std/memory_resource (memory_resource::~memory_resource()):
Do not define inline.
(monotonic_buffer_resource::~monotonic_buffer_resource()): Likewise.
* src/c++17/memory_resource.cc (memory_resource::~memory_resource()):
Define.
(monotonic_buffer_resource::~monotonic_buffer_resource()): Define.
* testsuite/20_util/monotonic_buffer_resource/93208.cc: New test.
From-SVN: r280044
2020-01-09 Martin Jambor <mjambor@suse.cz>
* cgraph.h (cgraph_edge): Make remove, set_call_stmt, make_direct,
resolve_speculation and redirect_call_stmt_to_callee static. Change
return type of set_call_stmt to cgraph_edge *.
* auto-profile.c (afdo_indirect_call): Adjust call to
redirect_call_stmt_to_callee.
* cgraph.c (cgraph_edge::set_call_stmt): Make return cgraph-edge *,
make the this pointer explicit, adjust self-recursive calls and the
call top make_direct. Return the resulting edge.
(cgraph_edge::remove): Make this pointer explicit.
(cgraph_edge::resolve_speculation): Likewise, adjust call to remove.
(cgraph_edge::make_direct): Likewise, adjust call to
resolve_speculation.
(cgraph_edge::redirect_call_stmt_to_callee): Likewise, also adjust
call to set_call_stmt.
(cgraph_update_edges_for_call_stmt_node): Update call to
set_call_stmt and remove.
* cgraphclones.c (cgraph_node::set_call_stmt_including_clones):
Renamed edge to master_edge. Adjusted calls to set_call_stmt.
(cgraph_node::create_edge_including_clones): Moved "first" definition
of edge to the block where it was used. Adjusted calls to
set_call_stmt.
(cgraph_node::remove_symbol_and_inline_clones): Adjust call to
cgraph_edge::remove.
* cgraphunit.c (walk_polymorphic_call_targets): Adjusted calls to
make_direct and redirect_call_stmt_to_callee.
* ipa-fnsummary.c (redirect_to_unreachable): Adjust calls to
resolve_speculation and make_direct.
* ipa-inline-transform.c (inline_transform): Adjust call to
redirect_call_stmt_to_callee.
(check_speculations_1):: Adjust call to resolve_speculation.
* ipa-inline.c (resolve_noninline_speculation): Adjust call to
resolve-speculation.
(inline_small_functions): Adjust call to resolve_speculation.
(ipa_inline): Likewise.
* ipa-prop.c (ipa_make_edge_direct_to_target): Adjust call to
make_direct.
* ipa-visibility.c (function_and_variable_visibility): Make iteration
safe with regards to edge removal, adjust calls to
redirect_call_stmt_to_callee.
* ipa.c (walk_polymorphic_call_targets): Adjust calls to make_direct
and redirect_call_stmt_to_callee.
* multiple_target.c (create_dispatcher_calls): Adjust call to
redirect_call_stmt_to_callee
(redirect_to_specific_clone): Likewise.
* tree-cfgcleanup.c (delete_unreachable_blocks_update_callgraph):
Adjust calls to cgraph_edge::remove.
* tree-inline.c (copy_bb): Adjust call to set_call_stmt.
(redirect_all_calls): Adjust call to redirect_call_stmt_to_callee.
(expand_call_inline): Adjust call to cgraph_edge::remove.
From-SVN: r280043
PR target/93141
* config/i386/i386.md (subv<mode>4): Use SWIDWI iterator instead of
SWI. Use <general_hilo_operand> instead of <general_operand>. Use
CONST_SCALAR_INT_P instead of CONST_INT_P.
(*subv<mode>4_1): Rename to ...
(subv<mode>4_1): ... this.
(*subv<dwi>4_doubleword, *addv<dwi>4_doubleword_1): New
define_insn_and_split patterns.
(*subv<mode>4_overflow_1, *addv<mode>4_overflow_2): New define_insn
patterns.
* gcc.target/i386/pr93141-1.c: Add tests with constants that have MSB
of the low half of the constant set.
* gcc.target/i386/pr93141-2.c: New test.
From-SVN: r280029
This patch adds a class auto_delete_vec<T>, a subclass of auto_vec <T *>
that deletes all of its elements on destruction; it's used in many
places in the analyzer patch kit.
This is a crude way for a vec to "own" the objects it points to
and clean up automatically (essentially a workaround for not being able
to use unique_ptr, due to C++98).
gcc/ChangeLog:
* vec.c (class selftest::count_dtor): New class.
(selftest::test_auto_delete_vec): New test.
(selftest::vec_c_tests): Call it.
* vec.h (class auto_delete_vec): New class template.
(auto_delete_vec<T>::~auto_delete_vec): New dtor.
From-SVN: r280027
Musl and lld don't support TLS copy relocs, and don't want to add support
for this feature which is unique to RISC-V. Only GNU ld and glibc support
them. In the pasbi discussion, people have pointed out various problems
with using them, so we are deprecating them. There doesn't seem to be an
ABI break from dropping them so this patch modifies gcc to stop creating
them. I'm using an ifdef for now in case a problem turns up and the code
has to be re-enabled. The plan is to add an initial to local exec
relaxation as a replacement, though this has not been defined or
implemented yet.
gcc/
* config/riscv/riscv.c (riscv_legitimize_tls_address): Ifdef out
use of TLS_MODEL_LOCAL_EXEC when not pic.
From-SVN: r280025
Previously it didn't matter whether we looked through a TARGET_EXPR in
constexpr evaluation, but now that we have constexpr destructors it does.
On IRC I mentioned the idea of clearing TARGET_EXPR_CLEANUP in
digest_nsdmi_init, but since this initialization is expressed by an
INIT_EXPR, it's better to handle all INIT_EXPR, not just those for a member
initializer.
* constexpr.c (cxx_eval_store_expression): Look through TARGET_EXPR
when not preevaluating.
From-SVN: r280018
Since we switched to doing constexpr evaluation on pre-GENERIC trees,
we don't have to handle DECL_BY_REFERENCE.
* constexpr.c (cxx_eval_call_expression): Remove DECL_BY_REFERENCE
support.
From-SVN: r280017
This commit makes "make selftest-valgrind" clean by fixing this leak:
4 bytes in 1 blocks are definitely lost in loss record 1 of 734
at 0x483AB1A: calloc (vg_replace_malloc.c:762)
by 0x261DBE0: xcalloc (xmalloc.c:162)
by 0x2538C46: selftest::test_map_of_strings_to_int() (hash-map-tests.c:87)
by 0x253ABD2: selftest::hash_map_tests_c_tests() (hash-map-tests.c:307)
by 0x24A885B: selftest::run_tests() (selftest-run-tests.c:65)
by 0x1373D80: toplev::run_self_tests() (toplev.c:2339)
by 0x1373FA7: toplev::main(int, char**) (toplev.c:2421)
by 0x2550EFF: main (main.c:39)
gcc/ChangeLog:
* hash-map-tests.c (selftest::test_map_of_strings_to_int): Fix
memory leak.
From-SVN: r280015
When recursing into a directory, any errors that occur while removing a
directory entry are ignored, because the subsequent increment of the
directory iterator clears the error_code object.
This fixes that bug by checking the result of each recursive operation
before incrementing. This is a change in observable behaviour, because
previously other directory entries would still be removed even if one
(or more) couldn't be removed due to errors. Now the operation stops on
the first error, which is what the code intended to do all along. The
standard doesn't specify what happens in this case (because the order
that the entries are processed is unspecified anyway).
It also improves the error reporting so that the name of the file that
could not be removed is included in the filesystem_error exception. This
is done by introducing a new helper type for reporting errors with
additional context and a new function that uses that type. Then the
overload of std::filesystem::remove_all that throws an exception can use
the new function to ensure any exception contains the additional
information.
For std::experimental::filesystem::remove_all just fix the bug where
errors are ignored.
PR libstdc++/93201
* src/c++17/fs_ops.cc (do_remove_all): New function implementing more
detailed error reporting for remove_all. Check result of recursive
call before incrementing iterator.
(remove_all(const path&), remove_all(const path&, error_code&)): Use
do_remove_all.
* src/filesystem/ops.cc (remove_all(const path&, error_code&)): Check
result of recursive call before incrementing iterator.
* testsuite/27_io/filesystem/operations/remove_all.cc: Check errors
are reported correctly.
* testsuite/experimental/filesystem/operations/remove_all.cc: Likewise.
From-SVN: r280014
PR target/93187
* config/i386/i386.md (*stack_protect_set_2_<mode> peephole2,
*stack_protect_set_3 peephole2): Also check that the second
insns source is general_operand.
* g++.dg/opt/pr93187.C: New test.
From-SVN: r280012
PR target/93174
* config/i386/i386.md (addcarry<mode>_0): Use nonimmediate_operand
predicate for output operand instead of register_operand.
(addcarry<mode>, addcarry<mode>_1): Likewise. Add alternative with
memory destination and non-memory operands[2].
* gcc.c-torture/compile/pr93174.c: New test.
From-SVN: r280011
2020-01-08 Richard Biener <rguenther@suse.de>
PR middle-end/93199
* tree-eh.c (sink_clobbers): Update virtual operands for
the first and last stmt only. Add a dry-run capability.
(pass_lower_eh_dispatch::execute): Perform clobber sinking
after CFG manipulations and in RPO order to catch all
secondary opportunities reliably.
From-SVN: r280006
gcc/
Add -nodevicespecs option for avr.
PR target/93182
* config/avr/avr.opt (-nodevicespecs): New driver option.
* config/avr/driver-avr.c (avr_devicespecs_file): Only issue
"-specs=device-specs/..." if that option is not set.
* doc/invoke.texi (AVR Options) <-nodevicespecs>: Document.
From-SVN: r279995
When only the rmprofile multilibs are built, compiling for armv7-a
should select the generic v7 multilibs. This used to work before +sec
and +mp were added to the architecture options but it was broken by
that update. This patch fixes those variants and adds some tests to
ensure that they remain fixed ;-)
PR target/93188
* config/arm/t-multilib (MULTILIB_MATCHES): Add rules to match
armv7-a{+mp,+sec,+mp+sec} to appropriate armv7 multilib variants
when only building rm-profile multilibs.
* gcc.target/arm/multilib.exp: Add new tests for rm-profile only.
From-SVN: r279993
2020-01-07 Thomas Rodgers <trodgers@redhat.com>
* include/std/condition_variable
(condition_variable_any::wait_on): Rename to match current draft
standard.
(condition_variable_any::wait_on_until): Likewise.
(condition_variable_any::wait_on_for): Likewise.
* testsuite/30_threads/condition_variable_any/stop_token/wait_on.cc:
Adjust tests to account for renamed methods.
From-SVN: r279988
We need to revert one line of code change from r279942 due to
performance degression.
gcc/ChangeLog:
2020-01-08 Luo Xiong Hu <luoxhu@linux.ibm.com>
PR middle-end/93189
* ipa-inline.c (caller_growth_limits): Restore the AND.
From-SVN: r279986
In the gc compiler, for slicing an array, its AST has an implicit
address operation node. There isn't such node in the gofrontend
AST. During the escape analysis, we create a fake node to mimic
the gc compiler's behavior. For the fake node, the loopdepth was
not tracked correctly, causing miscompilation. Since this is an
address operation, do the same thing as we do for the address
operator.
Fixesgolang/go#36404.
Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/213643
From-SVN: r279984
Use specific panic functions instead, which are mostly already in the
runtime package.
Also correct "defer nil" to panic when we execute the defer, rather
than throw when we queue it.
Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/213642
From-SVN: r279979
