In DWARF4 and earlier, static data members were represented as DW_TAG_member and the
pruning code wouldn't prune those, but in DWARF5 they are represented as DW_TAG_variable
with the class parent and the pruning code prunes those by default unless they are
referenced from a separate definition without the class parent (out of class definition).
C++17 inline vars have the definitions in the class though and even before if the static
data member isn't ODR used, it doesn't need to be defined, so we could just never describe
those static data members in the debug info.
This change stops the pruning of DW_TAG_variable with DW_AT_const_value attribute
with a class parent for -gdwarf-5 and later.
This fixes
-FAIL: g++.dg/debug/dwarf2/constexpr-var-1.C scan-assembler-times DW_AT_const_expr 2
-FAIL: libstdc++-prettyprinters/80276.cc whatis p4
-FAIL: libstdc++-prettyprinters/80276.cc whatis p4
-FAIL: libstdc++-prettyprinters/libfundts.cc print as
-FAIL: libstdc++-prettyprinters/libfundts.cc print as
-FAIL: libstdc++-prettyprinters/libfundts.cc print os
-FAIL: libstdc++-prettyprinters/libfundts.cc print os
2021-02-10 Jakub Jelinek <jakub@redhat.com>
PR debug/98755
* dwarf2out.c (prune_unused_types_walk): Mark DW_TAG_variable DIEs
at class scope for DWARF5+.
This patch adds some XFAILs for PR98979 until the patch to fix them has
been approved. See:
https://gcc.gnu.org/pipermail/gcc-patches/2021-February/564711.html
gcc/testsuite/
PR fortran/98979
* gfortran.dg/goacc/array-with-dt-2.f90: Add expected errors.
* gfortran.dg/goacc/derived-chartypes-1.f90: Skip ICEing test.
* gfortran.dg/goacc/derived-chartypes-2.f90: Likewise.
libgomp/
PR fortran/98979
* testsuite/libgomp.oacc-fortran/array-stride-dt-1.f90: Add expected
errors.
This patch reverts the non-testsuite parts of commit
9a4d32f85c which cause ICEs without the
yet-to-be-approved patch here:
https://gcc.gnu.org/pipermail/gcc-patches/2021-February/564711.html
gcc/fortran/
PR fortran/98979
* openmp.c (resolve_omp_clauses): Omit OpenACC update in
contiguity check and stride-specified error.
Add unordered containers heterogeneous lookup member functions find, count, contains and
equal_range in C++20. Those members are considered for overload resolution only if hash and
equal functors used to instantiate the container have a nested is_transparent type.
libstdc++-v3/ChangeLog:
* include/bits/stl_tree.h
(__has_is_transparent, __has_is_transparent_t): Move...
* include/bits/stl_function.h: ...here.
* include/bits/hashtable_policy.h (_Hash_code_base<>::_M_hash_code_tr): New..
(_Hashtable_base<>::_M_equals_tr): New.
* include/bits/hashtable.h (_Hashtable<>::_M_find_tr, _Hashtable<>::_M_count_tr,
_Hashtable<>::_M_equal_range_tr): New member function templates to perform
heterogeneous lookup.
(_Hashtable<>::_M_find_before_node_tr): New.
(_Hashtable<>::_M_find_node_tr): New.
* include/bits/unordered_map.h (unordered_map::find<>, unordered_map::count<>,
unordered_map::contains<>, unordered_map::equal_range<>): New member function
templates to perform heterogeneous lookup.
(unordered_multimap::find<>, unordered_multimap::count<>,
unordered_multimap::contains<>, unordered_multimap::equal_range<>): Likewise.
* include/bits/unordered_set.h (unordered_set::find<>, unordered_set::count<>,
unordered_set::contains<>, unordered_set::equal_range<>): Likewise.
(unordered_multiset::find<>, unordered_multiset::count<>,
unordered_multiset::contains<>, unordered_multiset::equal_range<>): Likewise.
* include/debug/unordered_map
(unordered_map::find<>, unordered_map::equal_range<>): Likewise.
(unordered_multimap::find<>, unordered_multimap::equal_range<>): Likewise.
* include/debug/unordered_set
(unordered_set::find<>, unordered_set::equal_range<>): Likewise.
(unordered_multiset::find<>, unordered_multiset::equal_range<>): Likewise.
* testsuite/23_containers/unordered_map/operations/1.cc: New test.
* testsuite/23_containers/unordered_multimap/operations/1.cc: New test.
* testsuite/23_containers/unordered_multiset/operations/1.cc: New test.
* testsuite/23_containers/unordered_set/operations/1.cc: New test.
PR analyzer/98575 describes an unexpected -Wanalyzer-malloc-leak false
positive from gcc.dg/analyzer/pr94851-1.c on glibc < 2.28.
The issue is that a getchar call gets inlined into a call to _IO_getc,
and "_IO_getc" is not in the set of FILE * functions the analyzer
"knows about". This exposes a bug in memory leak detection on code
paths in which an unknown function has been called.
The memory leak bug is fixed in the prior commit, but for good
measure this patch special-cases the "_IO_"-prefixed names in glibc
so that the analyzer can reuse its knowledge about the unprefixed
variants.
gcc/analyzer/ChangeLog:
PR analyzer/98575
* sm-file.cc (is_file_using_fn_p): Support "_IO_"-prefixed
variants.
gcc/testsuite/ChangeLog:
PR analyzer/98575
* gcc.dg/analyzer/file-1.c (test_5): New.
* gcc.dg/analyzer/file-3.c: New test.
PR analyzer/98575 describes an unexpected -Wanalyzer-malloc-leak false
positive from gcc.dg/analyzer/pr94851-1.c on glibc < 2.28.
The issue is that a getchar call gets inlined into a call to _IO_getc,
and "_IO_getc" is not in the set of FILE * functions the analyzer
"knows about". This leads to a global pointer
struct buf *curbp;
being treated as UNKNOWN after the call to _IO_getc. Later when a
malloced pointer is written to curbp->b_amark, the write is discarded
(since curbp is unknown) without noting that the pointer has escaped,
and so the pointer is erroneously treated as leaking when the function
returns.
This patch updates the handling of *UNKNOWN to treat pointers written
to them as having escaped, fixing the false positive.
The patch stops the leak warning in gcc.dg/analyzer/explode-1.c.
After merging states at the join-point after the first switch, pp has
UNKNOWN value, and so *pp is a write through UNKNOWN, which with this
patch is now treated as escaping - despite the fact that all possible
values for *pp are on the stack. There doesn't seem to be a good way
to fix this, and the testcase is an artifically constructed one, so the
patch simply removes the dg-warning directive.
gcc/analyzer/ChangeLog:
PR analyzer/98575
* store.cc (store::set_value): Treat a pointer written to *UNKNOWN
as having escaped.
gcc/testsuite/ChangeLog:
PR analyzer/98575
* gcc.dg/analyzer/explode-1.c: Remove expected leak warning.
* gcc.dg/analyzer/pr94851-2.c: New test.
* gcc.dg/analyzer/pr98575-1.c: New test.
This is the miscompilation of Python at -O2 on HP-PA/Linux present
on the mainline and 10 branch, caused by the presence of a call to
__builtin_unreachable () in the middle of a heavily branchy code,
which confuses the reorg pass.
gcc/
PR rtl-optimization/96015
* reorg.c (skip_consecutive_labels): Minor comment tweaks.
(relax_delay_slots): When deleting a jump to the next active
instruction over a barrier, first delete the barrier if the
jump is the only way to reach the target label.
This patch introduces a vect.mul RTX cost and decouples the vector
multiplication costing from the scalar one.
gcc/ChangeLog:
2021-02-09 Andre Vieira <andre.simoesdiasvieira@arm.com>
* config/aarch64/aarch64-cost-tables.h: Add entries for vect.mul.
* config/aarch64/aarch64.c (aarch64_rtx_mult_cost): Use vect.mul for
vector multiplies and vect.alu for SSRA.
* config/arm/aarch-common-protos.h (struct vector_cost_table): Define
vect.mul cost field.
* config/arm/aarch-cost-tables.h: Add entries for vect.mul.
* config/arm/arm.c: Likewise.
gcc/testsuite/ChangeLog:
2021-02-09 Andre Vieira <andre.simoesdiasvieira@arm.com>
* gcc.target/aarch64/asimd-mul-to-shl-sub.c: New test.
Add tests for vpaddq_* Neon intrinsics. Since these intrinsics are
only supported for AArch64, these tests are restricted to only run on
AArch64 targets.
gcc/testsuite/ChangeLog:
2021-02-09 Jonathan Wright <jonathan.wright@arm.com>
* gcc.target/aarch64/advsimd-intrinsics/vpXXXq.inc:
New test template.
* gcc.target/aarch64/advsimd-intrinsics/vpaddq.c: New test.
libstdc++-v3/ChangeLog:
* testsuite/27_io/filesystem/operations/remove_all.cc: Remove
test directory after making it writable again.
* testsuite/experimental/filesystem/operations/remove_all.cc:
Likewise.
The most recent reimplementation of module loading initialization
changed the behaviour of setting an import's location, and broke some
partition handling.
PR c++/98944
gcc/cp/
* module.cc (module_state::is_rooted): Rename to ...
(module_state::has_location): ... here. Adjust callers.
(module_state::read_partitions): Adjust validity check.
Don't overwrite a known location.
gcc/testsuite/
* g++.dg/modules/pr98944_a.C: New.
* g++.dg/modules/pr98944_b.C: New.
* g++.dg/modules/pr98944_c.C: New.
* g++.dg/modules/pr98944_d.C: New.
The coroutine_handle<void>::from_address(void*) version is already
noexcept, and they do the same thing. Make them consistent.
libstdc++-v3/ChangeLog:
PR libstdc++/99021
* include/std/coroutine (coroutine_handle<P>::from_address): Add
noexcept.
The patch fixes build issues occurring if build parameter
"--enable-cstdio=stdio_pure" is specified and no unistd.h is
present in the environment.
libstdc++-v3/ChangeLog:
* include/ext/stdio_sync_filebuf.h: Remove unused <unistd.h>.
* src/c++17/fs_ops.cc (fs::permissions): Qualify mode_t.
This fixes the currently O(region-size) unwinding of avail info
to be O(unwind-size) by tracking a linked-list stack of pushed
avails. This reduces the compile-time spent in complete unrolling
for WRF.
2021-02-09 Richard Biener <rguenther@suse.de>
PR tree-optimization/98863
* tree-ssa-sccvn.h (vn_avail::next_undo): Add.
* tree-ssa-sccvn.c (last_pushed_avail): New global.
(rpo_elim::eliminate_push_avail): Chain pushed avails.
(unwind_state::avail_top): Add.
(do_unwind): Rewrite unwinding of avail entries.
(do_rpo_vn): Initialize last_pushed_avail and
avail_top of the undo state.
In the PR there are several possibilities how to improve _M_disjunct at
least in certain cases so that the compiler can figure out at least in some
cases where __s is provably disjunct from _M_data() ... _M_data() + this->size()
but it is probably GCC 12 material.
The false positive warning is on this particular copy, which is done for
non-disjunct pointers when __len2 > __len1 and the __s >= __p + __len1,
i.e. __s used to point to the characters moved through _S_move a few lines earlier
by __len2 - __len1 characters up to make space. That is why the
_S_copy source is __s + __len2 - __len1. Unfortunately, when the compiler
can't prove objects are disjunct, that copying from __s + __len2 - __len1
of __len2 characters can very well mean accessing characters the source
object (if it is not disjunct) provably can't have.
The following patch works around that by making the _S_copy be a __p based
pointer instead of __s based pointer.
__s + __len2 - __len1
and
__p + (__s - __p) + (__len2 - __len1)
have the same value and the latter may seem to be uselessly longer,
but it seems at least currently in GIMPLE we keep it that way and so that is
what the warning code during expansion will see, and only actually
optimize it to __s + __len2 - __len1 during RTL when we lose information
on what is a pointer and what is a mere offset with the same mode.
So, in the end we emit exactly the same assembly, just without the false
positive warning.
2021-02-09 Jakub Jelinek <jakub@redhat.com>
PR middle-end/98465
* include/bits/basic_string.tcc (basic_string::_M_replace): When __s
points to the characters moved by earlier _S_move, compute the source
address using expression based on the __p pointer rather than __s
pointer.
* g++.dg/warn/Wstringop-overread-1.C: New test.
The print_generic_expr_to_str function ends with
return xstrdup (...); and therefore expects the caller to free
the argument.
The following patch does that after it has been copied.
Instead of doing const_cast to cast away const char * to char *,
because the code uses s0 and s1 in so few places, I chose just
to change the types of the two variables so that const_cast
is not needed. After all, it is a heap allocated string that
this function owns and so if it wanted, it could change it too.
2021-02-09 Jakub Jelinek <jakub@redhat.com>
PR middle-end/99004
* calls.c (maybe_warn_rdwr_sizes): Change s0 and s1 type from
const char * to char * and free those pointers after use.
This works around a SLP graph partitioning or cost collecting issue
by being more forgiving in vect_bb_vectorization_profitable_p.
2021-02-09 Richard Biener <rguenther@suse.de>
PR tree-optimization/99017
* tree-vect-slp.c (vect_bb_vectorization_profitable_p): Allow
zero vector cost entries.
Normally, an explicit instantiation means we want to write out the
instantiation. But not for a consteval function.
gcc/cp/ChangeLog:
PR c++/96905
* pt.c (mark_decl_instantiated): Exit early if consteval.
gcc/testsuite/ChangeLog:
PR c++/96905
* g++.dg/cpp2a/consteval-expinst1.C: New test.
Here, in the thunk returned from the captureless lambda conversion to
pointer-to-function, we try to pass through invisible reference parameters
by reference, without doing a copy. The empty class copy optimization was
messing that up.
gcc/cp/ChangeLog:
PR c++/98326
PR c++/20408
* cp-gimplify.c (simple_empty_class_p): Don't touch an invisiref
parm.
gcc/testsuite/ChangeLog:
PR c++/98326
* g++.dg/cpp1y/lambda-generic-empty1.C: New test.
My second patch for 97566 omits nested CONSTRUCTORs for empty fields, but we
do want them for empty union members.
gcc/cp/ChangeLog:
PR c++/98994
PR c++/97566
* constexpr.c (cxx_eval_store_expression): Only skip empty fields in
RECORD_TYPE.
gcc/testsuite/ChangeLog:
PR c++/98994
* g++.dg/cpp2a/no_unique_address12.C: New test.
The next piece of 98531 is that in some cases we need to create a
cleanup function to do the work (when the object is an array, or we're
using regular atexit). We were not pushing that function's decl
anywhere (not giving it a context) so streaming it failed.
This is a partial fix. You'll notice we're naming these from a per-TU
counter. I've captured that in PR98893.
gcc/cp/
* decl.c (start_cleanup_fn): Push function into
namespace.
gcc/testsuite/
* g++.dg/modules/pr98531-2.h: New.
* g++.dg/modules/pr98531-2_a.H: New.
* g++.dg/modules/pr98531-2_b.C: New.
* g++.dg/modules/pr98531-3.h: New.
* g++.dg/modules/pr98531-3_a.H: New.
* g++.dg/modules/pr98531-3_b.C: New.
The compiler's use of lazily-declared library functions must insert
said functions into a symbol table, so that they can be correctly
merged across TUs at the module-level. We have too many different
ways of declaring such library functions. This fixes __cxa_atexit (or
its system-specific variations), pushing (or merging) the decl into
the appropriate namespace. Because we're pushing a lazy builtin,
check_redeclaration_exception_specification needed a tweak to allow a
such a builtin's eh spec to differ from what the user may have already
declared. (I suspect no all headers declare atexit as noexcept.)
We can't test the -fno-use-cxa-atexit path with modules, as that
requires a followup patch to a closely related piece (which also
affects cxa_atexit targets in other circumstances).
PR c++/98531
gcc/cp/
* cp-tree.h (push_abi_namespace, pop_abi_namespace): Declare.
* decl.c (push_abi_namespace, pop_abi_namespace): Moved
from rtti.c, add default namespace arg.
(check_redeclaration_exception_specification): Allow a lazy
builtin's eh spec to differ from an lready-declared user
declaration.
(declare_global_var): Use push/pop_abi_namespace.
(get_atexit_node): Push the fndecl into a namespace.
* rtti.c (push_abi_namespace, pop_abi_namespace): Moved to
decl.c.
gcc/testsuite/
* g++.dg/modules/pr98531-1.h: New.
* g++.dg/modules/pr98531-1_a.H: New.
* g++.dg/modules/pr98531-1_b.C: New.
* g++.dg/abi/pr98531-1.C: New.
* g++.dg/abi/pr98531-2.C: New.
* g++.dg/abi/pr98531-3.C: New.
* g++.dg/abi/pr98531-4.C: New.
This fixes up the nvectors parameter passed to vect_get_loop_mask in
vectorizable_condition after the STMT_VINFO_VEC_STMTS rework.
gcc/ChangeLog:
2021-02-08 Andre Vieira <andre.simoesdiasvieira@arm.com>
PR middle-end/98974
* tree-vect-stmts.c (vectorizable_condition): Remove shadow vec_num
parameter in vectorizable_condition.
gcc/testsuite/ChangeLog:
2021-02-08 Andre Vieira <andre.simoesdiasvieira@arm.com>
PR middle-end/98974
* gfortran.dg/pr98974.F90: New test.
This implements walking of VECTOR_CST elements in walk_tree, mimicing
the walk of COMPLEX_CST elements. Without this free-lang-data fails
to see some types in case they are only refered to via tree constants
used only as VECTOR_CST elements.
2021-02-08 Richard Biener <rguenther@suse.de>
PR lto/96591
* tree.c (walk_tree_1): Walk VECTOR_CST elements.
* g++.dg/lto/pr96591_0.C: New testcase.
gcc/ChangeLog:
PR lto/98971
* cfgexpand.c (pass_expand::execute): Parse per-function option
flag_patchable_function_entry and use it.
* common.opt: Remove function_entry_patch_area_size and
function_entry_patch_area_start global variables.
* opts.c (parse_and_check_patch_area): New function.
(common_handle_option): Use it.
* opts.h (parse_and_check_patch_area): New function.
* toplev.c (process_options): Parse and use
function_entry_patch_area_size.
Instead, the full name of these codes are explicitly given in
intrinsics.def, to make it clear what these values map to.
gcc/d/ChangeLog:
* d-tree.h (DEF_D_INTRINSIC): Don't insert INTRINSIC_ into the
intrinsic code name.
* intrinsics.cc (DEF_D_INTRINSIC): Don't insert INTRISIC_ and
BUILT_IN_ into the intrinsic and built-in code names.
* intrinsics.def: Explicitly use full intrinsic and built-in
codes in all definitions.
Since most of volatile is deprecated in C++20, we are required to warn
for compound assignments to volatile variables and so on. But here we
have
volatile int x, y, z;
(b ? x : y) = 1;
and we shouldn't warn, because simple assignments like x = 24; should
not provoke the warning when they are a discarded-value expression.
We warn here because when ?: is used as an lvalue, we transform it in
cp_build_modify_expr/COND_EXPR from (a ? b : c) = rhs to
(a ? (b = rhs) : (c = rhs))
and build_conditional_expr then calls mark_lvalue_use for the new
artificial assignments, which then evokes the warning. The calls
to mark_lvalue_use were added in r160289 to suppress warnings in
Wunused-var-10.c, but looks like they're no longer needed.
To warn on
(b ? (x = 2) : y) = 1;
(b ? x : (y = 5)) = 1;
I've tweaked a check in mark_use/MODIFY_EXPR.
I'd argue this is a regression because GCC 9 doesn't warn.
gcc/cp/ChangeLog:
PR c++/98947
* call.c (build_conditional_expr_1): Don't call mark_lvalue_use
on arg2/arg3.
* expr.c (mark_use) <case MODIFY_EXPR>: Don't check read_p when
issuing the -Wvolatile warning. Only set TREE_THIS_VOLATILE if
a warning was emitted.
gcc/testsuite/ChangeLog:
PR c++/98947
* g++.dg/cpp2a/volatile5.C: New test.
Here we ICE in finish_nonmember_using_decl -> lookup_using_decl ->
... -> find_namespace_slot because "name" is not an IDENTIFIER_NODE.
It is a BIT_NOT_EXPR because this broken test uses
using E::~E; // SCOPE::NAME
A using-decl can't refer to a destructor, and lookup_using_decl already
checks that in the class member case. But in C++17, we do the "enum
scope is the enclosing scope" block, and so scope gets set to ::, and
we go into the NAMESPACE_DECL block. In C++20 we don't do it, we go
to the ENUMERAL_TYPE block.
I resorted to hoisting the check along with a diagnostic tweak: we
don't want to print "::::~E names destructor".
gcc/cp/ChangeLog:
PR c++/96462
* name-lookup.c (lookup_using_decl): Hoist the destructor check.
gcc/testsuite/ChangeLog:
PR c++/96462
* g++.dg/cpp2a/using-enum-8.C: New test.
We used to check all unknown input files, even when passing them to a
compiler. But that caused problems. However, not erroring out on
non-existent would-be-linker inputs confuses configure machinery that
probes the compiler to see if it accepts various inputs. This
restores the access check for things that are thought to be linker
input files, when we're not linking. (If we are linking, we presume
the linker will error out on its own accord.)
PR driver/98943
gcc/
* gcc.c (driver::maybe_run_linker): Check for input file
accessibility if not linking.
gcc/testsuite/
* c-c++-common/pr98943.c: New.
The following attempts to account for the fact that BB vectorization
regions now can span multiple loop levels and that an unprofitable
inner loop vectorization shouldn't be offsetted by a profitable
outer loop vectorization to make it overall profitable.
For now I've implemented a heuristic based on the premise that
vectorization should be profitable even if loops may not be entered
or if they iterate any number of times. Especially the first
assumption then requires that stmts directly belonging to loop A
need to be costed separately from stmts belonging to another loop
which also simplifies the implementation.
On x86 the added testcase has in the outer loop
t.c:38:20: note: Cost model analysis for part in loop 1:
Vector cost: 56
Scalar cost: 192
and the inner loop
t.c:38:20: note: Cost model analysis for part in loop 2:
Vector cost: 132
Scalar cost: 48
and thus the vectorization is considered not profitable
(note the same would happen in case the 2nd cost were for
a loop outer to the 1st costing).
Future enhancements may consider static knowledge of whether
a loop is always entered which would allow some inefficiency
in the vectorization of its loop header. Likewise stmts only
reachable from a loop exit can be treated this way.
2021-02-05 Richard Biener <rguenther@suse.de>
PR tree-optimization/98855
* tree-vectorizer.h (add_stmt_cost): New overload.
* tree-vect-slp.c (li_cost_vec_cmp): New.
(vect_bb_slp_scalar_cost): Cost individual loop regions
separately. Account for the scalar instance root stmt.
* g++.dg/vect/slp-pr98855.cc: New testcase.
As written in the PR, TARGET_QIMODE_MATH was meant to be set for all
tunings and it was the case for GCC <= 7, but as the number of
PROCESSOR_* enumerators grew, some AMD tunings (which are at the end
of the list) over time got enumerators with values >= 32 and
TARGET_QIMODE_MATH became disabled for them, in GCC 8 for 2
tunings, in GCC 9 for 7 tunings, in GCC 10 for 8 tunings, and
on the trunk for 11 tunings.
The following patch fixes it by using uhwis rather than uints
and gives them also symbolic names.
2021-02-05 Jakub Jelinek <jakub@redhat.com>
PR target/98957
* config/i386/i386-options.c (m_NONE, m_ALL): Define.
* config/i386/x86-tune.def (X86_TUNE_BRANCH_PREDICTION_HINTS,
X86_TUNE_PROMOTE_QI_REGS): Use m_NONE instead of 0U.
(X86_TUNE_QIMODE_MATH): Use m_ALL instead of ~0U.
We ICE on the following testcase, for incomplete array a on auto [b] { a }; without
giving any kind of diagnostics, with auto [c] = a; during error-recovery.
The problem is that we get too far through check_initializer and e.g.
store_init_value -> constexpr stuff can't deal with incomplete array types.
As the type of the structured binding artificial variable is always deduced,
I think it is easiest to diagnose this early, even if they have array types
we'll need their deduced type to be complete rather than just its element
type.
2021-02-05 Jakub Jelinek <jakub@redhat.com>
PR c++/97878
* decl.c (check_array_initializer): For structured bindings, require
the array type to be complete.
* g++.dg/cpp1z/decomp54.C: New test.
Similar to the vget_low* intrinsics we should just use a proper vec_select rather than
going through V2DI subregs.
gcc/ChangeLog:
* config/aarch64/aarch64-simd-builtins.def (get_high): Define builtin.
* config/aarch64/aarch64-simd.md (aarch64_get_high<mode>): Define.
* config/aarch64/arm_neon.h (__GET_HIGH): Delete.
(vget_high_f16): Reimplement using new builtin.
(vget_high_f32): Likewise.
(vget_high_f64): Likewise.
(vget_high_p8): Likewise.
(vget_high_p16): Likewise.
(vget_high_p64): Likewise.
(vget_high_s8): Likewise.
(vget_high_s16): Likewise.
(vget_high_s32): Likewise.
(vget_high_s64): Likewise.
(vget_high_u8): Likewise.
(vget_high_u16): Likewise.
(vget_high_u32): Likewise.
(vget_high_u64): Likewise.
We can do better on the vget_low* intrinsics.
Currently they reinterpret their argument into a V2DI vector and extract the low "lane",
reinterpreting that back into the shorter vector.
This is functionally correct and generates a sequence of subregs and a vec_select that, by itself,
gets optimised away eventually.
However it's bad when we want to use the result in a other SIMD operations.
Then the subreg-vec_select-subreg combo blocks many combine patterns.
This patch reimplements them to emit a proper low vec_select from the start.
It generates much cleaner RTL and allows for more aggressive combinations, particularly
with the patterns that Jonathan has been pushing lately.
gcc/ChangeLog:
* config/aarch64/aarch64-simd-builtins.def (get_low): Define builtin.
* config/aarch64/aarch64-simd.md (aarch64_get_low<mode>): Define.
* config/aarch64/arm_neon.h (__GET_LOW): Delete.
(vget_low_f16): Reimplement using new builtin.
(vget_low_f32): Likewise.
(vget_low_f64): Likewise.
(vget_low_p8): Likewise.
(vget_low_p16): Likewise.
(vget_low_p64): Likewise.
(vget_low_s8): Likewise.
(vget_low_s16): Likewise.
(vget_low_s32): Likewise.
(vget_low_s64): Likewise.
(vget_low_u8): Likewise.
(vget_low_u16): Likewise.
(vget_low_u32): Likewise.
(vget_low_u64): Likewise.
- Multi-lib only check the default argument is appeared in the multilib
flag list, but we didn't check the case that the flag is required but
default argument didn't provide.
- For example riscv*-rtems has a multilib set:
rv32imafd/ilp32d;@march=rv32imafd@mabi=ilp32d
And when we set the default argument of arch to rv32imafdc and ABI to
ilp32d, gcc will check the ilp32d is matched, but it didn't check
rv32imafd is not provided by default arguments, so it should not
reuse default library.
- The side effcet of this patch is csky-elf target will increase the
number of multilib from 54 to 59, but it's because genmultilib didn't
know the default argument for multilib, so I think it could be
improved in future.
gcc/ChangeLog:
* gcc.c (print_multilib_info): Check all required argument is provided
by default arg.
avx512vl-pr92686-vpcmp-{1,2,intelasm-1}.c are used to guard code
generation of integer mask comparison, but for vector comparison to
vector dest, integer mask comparison is disliked, so delete these
useless tests.
gcc/ChangeLog:
PR target/98537
* config/i386/i386-expand.c (ix86_expand_sse_cmp): Don't
generate integer mask comparison for 128/256-bits vector when
op_true/op_false is NULL_RTX or CONSTM1_RTX/CONST0_RTX. Also
delete redundant !maskcmp condition.
(ix86_expand_int_vec_cmp): Ditto but no redundant deletion
here.
(ix86_expand_sse_movcc): Delete definition of maskcmp, add the
condition directly to if (maskcmp), add extra check for
cmpmode, it should be MODE_INT.
(ix86_expand_fp_vec_cmp): Pass NULL to ix86_expand_sse_cmp's
parameters op_true/op_false.
(ix86_use_mask_cmp_p): New.
gcc/testsuite/ChangeLog:
PR target/98537
* g++.target/i386/avx512bw-pr98537-1.C: New test.
* g++.target/i386/avx512vl-pr98537-1.C: New test.
* g++.target/i386/avx512vl-pr98537-2.C: New test.
* gcc.target/i386/avx512vl-pr88547-1.c: Adjust testcase,
integer mask comparison should not be generated.
* gcc.target/i386/avx512vl-pr92686-vpcmp-1.c: Remove.
* gcc.target/i386/avx512vl-pr92686-vpcmp-2.c: Ditto.
* gcc.target/i386/avx512vl-pr92686-vpcmp-intelasm-1.c: Ditto.
