gcc/ada/
* libgnat/s-valuti.ads (Scan_Natural_Ghost): Split body from
spec and put it into private part, so that GNATprove can pick it
both when analysing the unit and its clients.
gcc/ada/
* freeze.adb (Freeze_Entity): Replace First_Entity/Next_Entity
with First_Component/Next_Component; remove condition with Ekind
equal to E_Component.
* sem_ch13.adb (Check_Record_Representation_Clause): Likewise
for component-or-discriminant.
* sem_util.adb (Is_Fully_Initialized_Type): Likewise; rename Ent
to a more specific Comp.
* sem_warn.adb (Check_References): Likewise.
gcc/ada/
* sem_warn.adb (Warn_On_Unassigned_Out_Parameter): Move inner
loop at the beginning of subprogram, so it is executed only
once; fix order in the "add an ad hoc" phrase.
gcc/ada/
* libgnat/i-c.adb: Add ghost code.
(C_Length_Ghost): New ghost functions to query the C length of a
string.
(To_Ada): Insert constant Count_Cst where needed to comply with
SPARK. Homogeneize code between variants for char, wchar_t,
char16_t and char32_t. Use char16_nul and char32_nul
systematically instead of their value. Fix the type of index To
to be Integer instead of Positive, to avoid a possible range
check failure on an empty Target. Insert an exit statement to
avoid a possible overflow failure when the last index in Target
is Natural'Last (possibly on a small string as well).
* libgnat/i-c.ads: Add contracts.
(C_Length_Ghost): New ghost functions to query the C length of a
string.
* libgnat/s-os_lib.adb: Remove pragma Compiler_Unit_Warning
causing a spurious error during compilation of GNAT, as this
pragma is not needed anymore now that we bootstrap (stage1) with
the base compiler runtime.
gcc/ada/
* libgnat/s-valboo.adb (First_Non_Space_Ghost): Move to
utilities.
(Value_Boolean): Prefix call to First_Non_Space_Ghost.
* libgnat/s-valboo.ads (First_Non_Space_Ghost): Move to
utilities.
(Is_Boolean_Image_Ghost, Value_Boolean): Prefix call to
First_Non_Space_Ghost.
* libgnat/s-valuer.adb (Scan_Raw_Real): Adapt to change of
function Scan_Exponent to procedure.
* libgnat/s-valueu.adb (Scan_Raw_Unsigned): Adapt to change of
function Scan_Exponent to procedure.
* libgnat/s-valuti.adb (First_Non_Space_Ghost): Function moved
here.
(Last_Number_Ghost): New ghost query function.
(Scan_Exponent): Change function with side-effects into
procedure, to mark in SPARK. Prove procedure wrt contract.
Change type of local P to avoid possible range check failure (it
is not known whether this can be activated by callers).
(Scan_Plus_Sign, Scan_Sign): Change type of local P to avoid
possible range check failure. Add loop invariants and assertions
for proof.
(Scan_Trailing_Blanks): Add loop invariant.
(Scan_Underscore): Remove SPARK_Mode Off.
* libgnat/s-valuti.ads (First_Non_Space_Ghost): Function moved
here.
(Last_Number_Ghost, Only_Number_Ghost, Is_Natural_Format_Ghost,
Scan_Natural_Ghost): New ghost query functions.
(Scan_Plus_Sign, Scan_Sign, Scan_Exponent, Scan_Trailing_Blanks,
Scan_Underscore): Add functional contracts.
gcc/ada/
* libgnat/s-imgboo.adb: Mark in SPARK.
* libgnat/s-imgboo.ads: Mark in SPARK. Change from Pure to
Preelaborate unit in order to be able to depend on
System.Val_Bool.
(Image_Boolean): Functionally specify the result of the
procedure by calling System.Val_Bool.Value_Boolean on the
result.
* libgnat/s-valboo.adb: Mark in SPARK.
(First_Non_Space_Ghost): New ghost function.
(Value_Boolean): Change type of L and F to avoid possible range
check failure on empty Str.
* libgnat/s-valboo.ads: Mark in SPARK. Duplicate with-clause
from body in the spec to be able to call
System.Val_Util.Only_Space_Ghost in the contract.
(First_Non_Space_Ghost): New ghost function computing the first
non-space character in a string.
(Is_Boolean_Image_Ghost): New ghost function computing whether a
string is the image of a boolean value.
(Value_Boolean): Add in precondition the conditions to avoid
raising Constraint_Error. This precondition is never executed,
and only used in proof, thanks to the use of pragma
Assertion_Policy. Given that precondition, the postcondition can
simply check the first non-space character to decide whether
True or False is read.
* libgnat/s-valuti.adb: Mark in SPARK, but use SPARK_Mode Off on
all subprograms not yet proved.
(Bad_Value): Annotate expected exception.
(Normalize_String): Rewrite to avoid possible overflow when
incrementing F in the first loop. Add loop invariants.
* libgnat/s-valuti.ads: Mark in SPARK.
(Bad_Value): Add Depends contract to avoid warning on unused S.
(Only_Space_Ghost): New ghost function to query if string has
only space in the specified range.
(Normalize_String): Add functional contract.
(Scan_Exponent): Mark spec as not in SPARK as this function has
side-effects.
__builtin_smul_overflow can be unavailable for some C++ compilers.
Add long long multiplication as backup for overflow processing.
gcc/ChangeLog:
PR rtl-optimization/103437
* ira-color.c (setup_allocno_priorities): Use long long
multiplication as backup for overflow processing.
This is the library fix for PR103520 that also prevents the garbage
collector from releasing live memory. However this requires that the
host compiler has been patched with this fix, so the GC will remain
disabled in the D front-end for now until enough time has passed for
this to have trickled down into enough releases.
libphobos/ChangeLog:
* libdruntime/core/thread/osthread.d (callWithStackShell): Push all
callee-save registers on the stack for AArch64 and ARM.
This optimizes right shift rounding narrow instructions to
rounding add narrow high where one vector is 0 when the shift amount is half
that of the original input type.
i.e.
uint32x4_t foo (uint64x2_t a, uint64x2_t b)
{
return vrshrn_high_n_u64 (vrshrn_n_u64 (a, 32), b, 32);
}
now generates:
foo:
movi v3.4s, 0
raddhn v0.2s, v2.2d, v3.2d
raddhn2 v0.4s, v2.2d, v3.2d
instead of:
foo:
rshrn v0.2s, v0.2d, 32
rshrn2 v0.4s, v1.2d, 32
ret
On Arm cores this is an improvement in both latency and throughput.
Because a vector zero is needed I created a new method
aarch64_gen_shareable_zero that creates zeros using V4SI and then takes a subreg
of the zero to the desired type. This allows CSE to share all the zero
constants.
gcc/ChangeLog:
* config/aarch64/aarch64-protos.h (aarch64_gen_shareable_zero): New.
* config/aarch64/aarch64-simd.md (aarch64_rshrn<mode>,
aarch64_rshrn2<mode>): Generate rounding half-ing add when appropriate.
* config/aarch64/aarch64.c (aarch64_gen_shareable_zero): New.
gcc/testsuite/ChangeLog:
* gcc.target/aarch64/advsimd-intrinsics/shrn-1.c: New test.
* gcc.target/aarch64/advsimd-intrinsics/shrn-2.c: New test.
* gcc.target/aarch64/advsimd-intrinsics/shrn-3.c: New test.
* gcc.target/aarch64/advsimd-intrinsics/shrn-4.c: New test.
We process overflows in cost calculations but for huge functions
priority calculation can overflow as priority can be bigger the cost
used for it. The patch fixes the problem.
gcc/ChangeLog:
PR rtl-optimization/103437
* ira-color.c (setup_allocno_priorities): Process multiplication
overflow.
This updates the testcase bic-bitmask-18.c to seach for " = 0;" as an expression
so it doesn't match any other partial expressions.
gcc/testsuite/ChangeLog:
PR testsuite/103479
* gcc.dg/bic-bitmask-18.c: Update regexpr to expect = 0;.
Here we issue a bogus:
error: '(0 ? fake_tuple_size_v<int> : fake_tuple_size_v<int>)' is not a constant expression
because cxx_constant_value in expand_integer_pack gets
*(0 ? VIEW_CONVERT_EXPR<const int>(fake_tuple_size_v) : VIEW_CONVERT_EXPR<const int>(fake_tuple_size_v))
which is a REFERENCE_REF_P and we evaluate its operand to 3, so we end
up with *3 and that fails. Sounds like we need to get rid of the
REFERENCE_REF_P then. That is what tsubst_copy_and_build/INDIRECT_REF
will do:
if (REFERENCE_REF_P (t))
{
/* A type conversion to reference type will be enclosed in
such an indirect ref, but the substitution of the cast
will have also added such an indirect ref. */
r = convert_from_reference (r);
}
so I think it's reasonable to call instantiate_non_dependent_expr_sfinae.
PR c++/94490
gcc/cp/ChangeLog:
* pt.c (expand_integer_pack): Call
instantiate_non_dependent_expr_sfinae.
gcc/testsuite/ChangeLog:
* g++.dg/ext/integer-pack5.C: New test.
Here we crash when issuing the "constraint C has type T, not bool"
error, because pp_cxx_parameter_mapping wasn't prepared to see an
anonymous template parameter. With this patch we print
error: constraint 'auto(<lambda>) [with <unnamed> = 0]' has type '<lambda()>', not 'bool'
The "<unnamed>" is what this patch adds.
PR c++/103408
gcc/cp/ChangeLog:
* cxx-pretty-print.c (pp_cxx_parameter_mapping): Print "<unnamed>"
rather than crash on an unnamed template parameter.
gcc/testsuite/ChangeLog:
* g++.dg/cpp23/concepts-err1.C: New test.
This avoids using VLA types to initalize a register with
-ftrivial-auto-var-init in some cases.
2021-12-02 Richard Biener <rguenther@suse.de>
PR middle-end/103271
* internal-fn.c (expand_DEFERRED_INIT): When the base
of the LHS is a decl with matching constant size use
that as the initialization target instead of an
eventual VLA typed one.
This makes sure to always use a truth type to build the gather
mask argument even when the target builtin prototype in the end
wants a float vector.
2021-12-02 Richard Biener <rguenther@suse.de>
PR tree-optimization/103527
* tree-vect-stmts.c (vect_build_gather_load_calls): Always
use a truth type for building the vector mask.
The transposition nolto -> notlo is confusing and it makes the long
name even harder to read than it already is - I kept reading it as
"not lo" until I realized it was a simple typo.
Fixes: 5269b24605 (Silence warning in LTO mode on VxWorks)
lto-plugin/
* lto-plugin.c: Fix -linker-output-auto-notlo-rel ->
-linker-output-auto-nolto-rel typo.
(process_option): Adjust accordingly, accepting both old and
new spelling.
gcc/
* config/vxworks.h (LTO_PLUGIN_SPEC): Adapt to corrected
spelling of -linker-output-auto-nolto-rel.
Fix issue with the Fortran front-end when mapping arrays: when creating the
data MEM_REF for the map clause, there was a convention of casting the
referencing pointer to 'c_char *' by
fold_convert (build_pointer_type (char_type_node), ptr).
This causes the alignment passed to the libgomp runtime for array data
hardwared to '1', and causes alignment errors on the offload target.
This patch fixes this by removing the char_type_node pointer converts, and
adding gcc_asserts to ensure POINTER_TYPE_P (TREE_TYPE (ptr)).
PR fortran/90030
gcc/fortran/ChangeLog:
* trans-openmp.c (gfc_omp_finish_clause): Remove fold_convert to pointer
to char_type_node, add gcc_assert of POINTER_TYPE_P.
(gfc_trans_omp_array_section): Likewise.
(gfc_trans_omp_clauses): Likewise.
gcc/testsuite/ChangeLog:
* gfortran.dg/goacc/finalize-1.f: Adjust scan test.
* gfortran.dg/gomp/affinity-clause-1.f90: Likewise.
* gfortran.dg/gomp/affinity-clause-5.f90: Likewise.
* gfortran.dg/gomp/defaultmap-4.f90: Likewise.
* gfortran.dg/gomp/defaultmap-5.f90: Likewise.
* gfortran.dg/gomp/defaultmap-6.f90: Likewise.
* gfortran.dg/gomp/map-3.f90: Likewise.
* gfortran.dg/gomp/pr78260-2.f90: Likewise.
* gfortran.dg/gomp/pr78260-3.f90: Likewise.
libgomp/ChangeLog:
* testsuite/libgomp.oacc-fortran/pr90030.f90: New test.
* testsuite/libgomp.fortran/pr90030.f90: New test.
Fixes a typo that occurred during the splitting of the std.math module
into a package.
libphobos/ChangeLog:
* src/std/math/hardware.d (FloatingPointControl.getControlState): Add
missing ControlState variable for AArch64.
Not all targets that support building libdruntime have a stable garbage
collector, so to avoid running into problems where live memory allocated
by the D GC is freed, disable all in-flight collections until a time
when scanning is more reliably implemented everywhere.
PR d/103520
gcc/d/ChangeLog:
* d-frontend.h (gc_disable): Declare.
* d-lang.cc (d_init_options): Disable the D runtime garbage collector
after initializing.
None of the front-end module names in either the dmd or root package
collide just yet, but that does not mean they won't in the future.
gcc/d/ChangeLog:
* Make-lang.in (D_FRONTEND_OBJS): Prefix object files from the root
package with root-.
(d/root-%.o): New recipe.
Juliet 1.3's CWE415_Double_Free__malloc_free_*_67a.c
were showing leak false positives in non-LTO builds; fixed thusly.
gcc/analyzer/ChangeLog:
PR analyzer/102471
* region-model-reachability.cc (reachable_regions::handle_parm):
Treat all svalues within a compound parm has reachable, and those
wrapped in a cast.
gcc/testsuite/ChangeLog:
PR analyzer/102471
* gcc.dg/analyzer/leak-3.c: New test.
Signed-off-by: David Malcolm <dmalcolm@redhat.com>
Introduce vec_set_0 pattern for V8HI and V8HF modes to implement scalar
element 0 inserts to from a GP register, SSE register or memory. Also
add V8HI and V8HF AVX2 (x,x,x) alternative to PINSR insn pattern, which is
split after reload to a sequence of PBROADCASTW and PBLENDW.
The V8HF inserts from memory improve from:
- vpbroadcastw 4(%esp), %xmm1
- vpblendw $16, %xmm1, %xmm0, %xmm0
+ vpinsrw $4, 4(%esp), %xmm0, %xmm0
and V8HF inserts from SSE register to element 0 improve from:
vpxor %xmm2, %xmm2, %xmm2
- vpbroadcastw %xmm0, %xmm0
vpblendw $1, %xmm0, %xmm2, %xmm0
Based on the above improvements, the register allocator is able to determine
the optimal instruction (or instruction sequence) based on the register set
of the input value, so there is no need to manually expand V8HI and V8HF
inserts to the sequence of VEC_DUPLICATE and VEC_MERGE RTXes.
2021-12-01 Uroš Bizjak <ubizjak@gmail.com>
gcc/ChangeLog:
PR target/102811
* config/i386/sse.md (VI2F): Remove mode iterator.
(VI2F_256_512): New mode iterator.
(vec_set<V8_128:mode>_0): New insn pattern.
(vec_set<VI2F_256_512:mode>_0>): Rename from vec_set<VI2F:mode>mode.
Use VI2F_256_512 mode iterator instead of VI2F.
(*axv512fp16_movsh): Remove.
(<sse2p4_1>_pinsr<ssemodesuffix>): Add (x,x,x) AVX2 alternative.
Do not disable V8HF mode insn on AVX2 targets.
(pinsrw -> pbroadcast + pblendw peephole2): New peephole.
(pinsrw -> pbroadcast + pblendw splitter): New post-reload splitter.
* config/i386/i386.md (extendhfsf): Call gen_vec_setv8hf_0.
* config/i386/i386-expand.c (ix86_expand_vector_set)
<case E_V8HFmode>: Use vec_merge path for TARGET_AVX2.
gcc/testsuite/ChangeLog:
PR target/102881
* gcc.target/i386/pr102811-1.c: New test.
* gcc.target/i386/avx512fp16-1c.c (dg-final): Update
scan-assembler-times scan strings for ia32 targets.
* gcc.target/i386/pr102327-1.c (dg-final): Ditto.
* gcc.target/i386/pr102811.c: Rename from ...
* gcc.target/i386/avx512vl-vcvtps2ph-pr102811.c: ... this.
This middle-end patch is inspired by the Richard Beiner's until-wrap
loop example in PR tree-optimization/101145.
unsigned foo(unsigned val, unsigned start)
{
unsigned cnt = 0;
for (unsigned i = start; i > val; ++i)
cnt++;
return cnt;
}
For this loop, the tree optimizers currently generate:
unsigned int foo (unsigned int val, unsigned int start)
{
unsigned int cnt;
unsigned int _1;
unsigned int _5;
<bb 2> [local count: 118111600]:
if (start_3(D) > val_4(D))
goto <bb 3>; [89.00%]
else
goto <bb 4>; [11.00%]
<bb 3> [local count: 105119324]:
_1 = start_3(D) + 1;
_5 = -start_3(D);
cnt_2 = _1 > val_4(D) ? _5 : 1;
<bb 4> [local count: 118111600]:
# cnt_11 = PHI <cnt_2(3), 0(2)>
return cnt_11;
}
or perhaps slightly easier to read:
if (start > val) {
cnt = (start+1) > val ? -start : 1;
} else cnt = 0;
In this snippet, if we know start > val, then (start+1) > val
unless start+1 overflows, i.e. (start+1) == 0 and start == ~0.
We can use this (loop header) context to simplify the ternary
expression to "(start != -1) ? -start : 1", which with a little
help from match.pd can be folded to -start. Hence the optimal
final value replacement should be:
cnt = (start > val) ? -start : 0;
Or as now generated by this patch:
unsigned int foo (unsigned int val, unsigned int start)
{
unsigned int cnt;
<bb 2> [local count: 118111600]:
if (start_3(D) > val_4(D))
goto <bb 3>; [89.00%]
else
goto <bb 4>; [11.00%]
<bb 3> [local count: 105119324]:
cnt_2 = -start_3(D);
<bb 4> [local count: 118111600]:
# cnt_11 = PHI <cnt_2(3), 0(2)>
return cnt_11;
}
We can also improve until-wrap loops that don't have a (suitable) loop
header, as determined by simplify_using_initial_conditions.
unsigned bar(unsigned val, unsigned start)
{
unsigned cnt = 0;
unsigned i = start;
do {
cnt++;
i++;
} while (i > val);
return cnt;
}
which is currently optimized to:
unsigned int foo (unsigned int val, unsigned int start)
{
unsigned int cnt;
unsigned int _9;
unsigned int _10;
<bb 2> [local count: 118111600]:
_9 = start_4(D) + 1;
_10 = -start_4(D);
cnt_3 = val_7(D) < _9 ? _10 : 1;
return cnt_3;
}
Here we have "val < (start+1) ? -start : 1", which again with the
help of match.pd can be slightly simplified to "val <= start ? -start : 1"
when dealing with unsigned types, because at the complicating value where
start == ~0, we fortunately have -start == 1, hence it doesn't matter
whether the second or third operand of the ternary operator is returned.
To summarize, this patch (in addition to tweaking may_be_zero in
number_of_iterations_until_wrap) adds three new constant folding
transforms to match.pd.
X != C1 ? -X : C2 simplifies to -X when -C1 == C2.
which is the generalized form of the simplification above.
X != C1 ? ~X : C2 simplifies to ~X when ~C1 == C2.
which is the BIT_NOT_EXPR analog of the NEGATE_EXPR case.
and the "until-wrap final value replacement without context":
(X + 1) > Y ? -X : 1 simplifies to X >= Y ? -X : 1 when
X is unsigned, as when X + 1 overflows, X is -1, so -X == 1.
2021-12-01 Roger Sayle <roger@nextmovesoftware.com>
Richard Biener <rguenther@suse.de>
gcc/ChangeLog
* tree-ssa-loop-niter.c (number_of_iterations_until_wrap):
Check if simplify_using_initial_conditions allows us to
simplify the expression for may_be_zero.
* match.pd (X != C ? -X : -C -> -X): New transform.
(X != C ? ~X : ~C -> ~X): Likewise.
((X+1) > Y ? -X : 1 -> X >= Y ? -X : 1): Likewise.
gcc/testsuite/ChangeLog
* gcc.dg/fold-condneg-1.c: New test case.
* gcc.dg/fold-condneg-2.c: New test case.
* gcc.dg/fold-condnot-1.c: New test case.
* gcc.dg/pr101145-1.c: New test case.
* gcc.dg/pr101145-2.c: New test case.
For -ftrivial-auto-var-init=*, skip initializing the variable if it is an
opaque type, because CONST0_RTX(mode) is not defined for opaque modes.
2021-12-01 Peter Bergner <bergner@linux.ibm.com>
gcc/
PR middle-end/103127
* gimplify.c (is_var_need_auto_init): Handle opaque types.
gcc/testsuite/
PR middle-end/103127
* gcc.target/powerpc/pr103127.c: New test.
In the gdc driver, this takes the previous fix for the Darwin D
bootstrap, and extends it to the -static-libphobos option as well.
Rather than pushing the -static-libphobos option back onto the command
line, the setting of SKIPOPT is instead conditionally removed. The same
change has been repeated for -static-libstdc++ so there is now no need
to call generate_option to re-add it.
In the gcc driver, -static-libphobos has been added as a common option,
validated, and a new outfile substition added to config/darwin.h to
correctly replace -lgphobos with libgphobos.a.
gcc/ChangeLog:
* common.opt (static-libphobos): Add option.
* config/darwin.h (LINK_SPEC): Substitute -lgphobos with libgphobos.a
when linking statically.
* gcc.c (driver_handle_option): Set -static-libphobos as always valid.
gcc/d/ChangeLog:
* d-spec.cc (lang_specific_driver): Set SKIPOPT on -static-libstdc++
and -static-libphobos only when target supports LD_STATIC_DYNAMIC.
Remove generate_option to re-add -static-libstdc++.
libphobos/ChangeLog:
* testsuite/testsuite_flags.in: Add libphobos library directory as
search path to --gdcldflags.
For PR61825, honza changed tree_single_nonzero_warnv_p to prevent a later
declaration from marking a function as weak after we've determined that it
wasn't weak before. But we shouldn't do that for speculative folding; we
should only do it when we actually need a constant value. In C++, such a
context is called "manifestly constant-evaluated". In fold, this seems to
correspond to the folding_initializer flag, since in C this situation only
occurs in static initializers.
This change makes nonzero-1.c well-formed; I've added a nonzero-1a.c to
verify that we delete the null check eventually if there is no weak
redeclaration.
The varasm.c change is so that if we do get the weak redeclaration error, we
get it at the position of the weak declaration rather than the previous
declaration.
Using the FOLD_INIT paths also affects floating point arithmetic: notably,
this makes floating point division by zero in a manifestly
constant-evaluated context constant, as in a C static initializer. I've had
some success convincing CWG that this is the right direction; C++ should
follow C's floating point semantics more than we have been doing, and Joseph
says that the C policy is that Annex F overrides other parts of the standard
that say that some operations are undefined. But since we're in stage 3,
I'm only making this change with the new flag -fconstexpr-fp-except. It may
turn on by default in a future release.
I think this distinction is only relevant for binary operations; arithmetic
for the floating point case, comparison for possibly non-zero addresses.
PR c++/103310
gcc/ChangeLog:
* fold-const.c (maybe_nonzero_address): Use get_create or get
depending on folding_initializer.
(fold_binary_initializer_loc): New.
* fold-const.h (fold_binary_initializer_loc): Declare.
* varasm.c (mark_weak): Don't use the decl location.
* doc/invoke.texi: Document -fconstexpr-fp-except.
gcc/c-family/ChangeLog:
* c.opt: Add -fconstexpr-fp-except.
gcc/cp/ChangeLog:
* constexpr.c (cxx_eval_binary_expression): Use
fold_binary_initializer_loc if manifestly cxeval.
gcc/testsuite/ChangeLog:
* g++.dg/cpp0x/constexpr-fp-except1.C: New test.
* g++.dg/cpp1z/constexpr-if36.C: New test.
* gcc.dg/tree-ssa/nonzero-1.c: Now well-formed.
* gcc.dg/tree-ssa/nonzero-1a.c: New test.
Recently Kewen fixed a problem in the old builtins support where
rs6000_builtin_decl prematurely indicated that a target builtin is
unavailable. This also needs to be done for the new builtins support, but in
this case we have to ensure the error message is still produced from the
overload support in rs6000-c.c. Unfortunately, this is less straightforward
than it could be, because header file includes need to be adjusted to make this
happen. Someday we'll consolidate all the builtin code in one file and this
won't have to be so ugly.
2021-12-01 Bill Schmidt <wschmidt@linux.ibm.com>
gcc/
PR target/102347
* config/rs6000/rs6000-c.c (rs6000-builtins.h): Stop including.
(rs6000-internal.h): Include.
(altivec_resolve_new_overloaded_builtin): Move call to
rs6000_invalid_new_builtin here from rs6000_new_builtin_decl.
* config/rs6000/rs6000-call.c (rs6000-builtins.h): Stop including.
(rs6000_invalid_new_builtin): Remove static qualifier.
(rs6000_new_builtin_decl): Remove test for supported builtin.
* config/rs6000/rs6000-internal.h (rs6000-builtins.h): Include.
(rs6000_invalid_new_builtin): Declare.
* config/rs6000/rs6000.c (rs6000-builtins.h): Don't include.
We're using a temporary range cache while computing ranges for PHIs to
make sure the real cache doesn't get set until all PHIs are computed.
With the ltrans beast in LTO mode this causes undue overhead.
Since we already have a bitmap to indicate whether there's a cache
entry, we can avoid the extra cache object by clearing it while PHIs
are being calculated.
gcc/ChangeLog:
PR tree-optimization/103409
* gimple-range-path.cc (path_range_query::compute_ranges_in_phis):
Do all the work with just one ssa_global_cache.
* gimple-range-path.h: Remove m_tmp_phi_cache.
If the allocator-extended move constructor move-constructs each element
into the new container, the contents of the old container are left in
moved-from states. We cannot know if those states preserve the
container's ordering and uniqueness guarantees, so just erase all
moved-from elements.
libstdc++-v3/ChangeLog:
PR libstdc++/103501
* include/bits/stl_tree.h (_Rb_tree(_Rb_tree&&, false_type)):
Clear container if elements have been moved-from.
* testsuite/23_containers/map/allocator/move_cons.cc: Expect
moved-from container to be empty.
* testsuite/23_containers/multimap/allocator/move_cons.cc:
Likewise.
* testsuite/23_containers/multiset/allocator/103501.cc: New test.
* testsuite/23_containers/set/allocator/103501.cc: New test.
This adds std::__is_constant_evaluated() as a C++11 wrapper for
__builtin_is_constant_evaluated, but just returning false if the
built-in isn't supported by the compiler. This allows us to use it
throughout the library without checking __has_builtin every time.
Some uses in std::vector and std::string can only be constexpr when the
std::is_constant_evaluated() function actually works, so we might as
well guard them with a relevant macro and call that function directly,
rather than the built-in or std::__is_constant_evaluated().
The remaining checks of the __cpp_lib_is_constant_evaluated macro could
now be replaced by checking __cplusplus >= 202002 instead, but there's
no practical difference. We still need some kind of preprocessor check
there anyway.
libstdc++-v3/ChangeLog:
* doc/doxygen/user.cfg.in (PREDEFINED): Change macro name.
* include/bits/allocator.h (allocate, deallocate): Use
std::__is_constant_evaluated() unconditionally, instead of
checking whether std::is_constant_evaluated() (or the built-in)
can be used.
* include/bits/basic_string.h: Check new macro. call
std::is_constant_evaluated() directly in C++20-only code that is
guarded by a suitable macro.
* include/bits/basic_string.tcc: Likewise.
* include/bits/c++config (__is_constant_evaluated): Define.
(_GLIBCXX_HAVE_BUILTIN_IS_CONSTANT_EVALUATED): Replace with ...
(_GLIBCXX_HAVE_IS_CONSTANT_EVALUATED): New macro.
* include/bits/char_traits.h (char_traits): Replace conditional
calls to std::is_constant_evaluated with unconditional calls to
std::__is_constant_evaluated.
* include/bits/cow_string.h: Use new macro.
* include/bits/ranges_algobase.h (__copy_or_move): Replace
conditional calls to std::is_constant_evaluated with unconditional
calls to std::__is_constant_evaluated.
(__copy_or_move_backward, __fill_n_fn): Likewise.
* include/bits/ranges_cmp.h (ranges::less): Likewise.
* include/bits/stl_algobase.h (lexicographical_compare_three_way):
Likewise.
* include/bits/stl_bvector.h: Call std::is_constant_evaluated
directly in C++20-only code that is guarded by a suitable macro.
* include/bits/stl_construct.h (_Construct, _Destroy, _Destroy_n):
Replace is_constant_evaluated with __is_constant_evaluated.
* include/bits/stl_function.h (greater, less, greater_equal)
(less_equal): Replace __builtin_is_constant_evaluated and
__builtin_constant_p with __is_constant_evaluated.
* include/bits/stl_vector.h: Call std::is_constant_evaluated()
in C++20-only code.
* include/debug/helper_functions.h (__check_singular): Use
__is_constant_evaluated instead of built-in, or remove check
entirely.
* include/std/array (operator<=>): Use __is_constant_evaluated
unconditionally.
* include/std/bit (__bit_ceil): Likewise.
* include/std/type_traits (is_constant_evaluated): Define using
'if consteval' if possible.
* include/std/version: Use new macro.
* libsupc++/compare: Use __is_constant_evaluated instead of
__builtin_is_constant_evaluated.
* testsuite/23_containers/array/tuple_interface/get_neg.cc:
Adjust dg-error lines.
Most ref-count updates in the COW string are done via the functions in
<ext/atomicity.h>, which will use non-atomic ops when the program is
known to be single-threaded. The _M_is_leaked() and _M_is_shared()
functions use __atomic_load_n directly, because <ext/atomicity.h>
doesn't provide a load operation. Those functions can check the
__is_single_threaded() predicate to avoid using __atomic_load_n when not
needed.
The move constructor for the fully-dynamic-string increments the
ref-count by either 2 or 1, for leaked or non-leaked strings
respectively. That can be changed to use a non-atomic store of 1 for all
non-shared strings. It can be non-atomic because even if the program is
multi-threaded, conflicting access to the rvalue object while it's being
moved from would be data race anyway. It can store 1 directly for all
non-shared strings because it doesn't matter whether the initial
refcount was -1 or 0, it should be 1 after the move constructor creates
a second owner.
libstdc++-v3/ChangeLog:
* include/bits/cow_string.h (basic_string::_M_is_leaked): Use
non-atomic load when __is_single_threaded() is true.
(basic_string::_M_is_shared): Likewise.
(basic_string::(basic_string&&)) [_GLIBCXX_FULLY_DYNAMIC_STRING]:
Use non-atomic store when rvalue is not shared.
When using COW strings, accessing _M_pathname[0] and similar non-const
accessors can cause the string to "leak", meaning it reallocates itself
if it shares ownership with another string object.
This causes test failures for --enable-fully-dynamic-string builds:
/home/jwakely/src/gcc/libstdc++-v3/testsuite/experimental/filesystem/path/construct/90634.cc:62: void test01(): Assertion 'bytes_allocated == 0' failed.
FAIL: experimental/filesystem/path/construct/90634.cc execution test
This FAIL happens because the fully-dynamic move constructor results in
shared ownership, so for path(std::move(std::string("foo"))) the
_M_pathname member shares ownership with the temporary, and the
non-const accesses in _M_split_cmpts() cause a new copy of the string to
be allocated. This un-sharing is wasteful, and entirely unnecessary when
sharing ownership with an rvalue that is about to release its ownership
anyway. Even for lvalues, sharing ownership is not a problem and
reallocating a unique copy of the string is wasteful.
This removes non-const accesses of _M_pathname in the
path::_M_split_cmpts() members.
libstdc++-v3/ChangeLog:
* src/c++17/fs_path.cc (path::_M_split_cmpts()): Remove
micro-optimization for "/" path.
* src/filesystem/path.cc (path::_M_split_cmpts()): Only access
the contents of _M_pathname using const member functions.
When checking for compatible stmts, vect_build_slp_tree_1 did:
&& !(STMT_VINFO_GROUPED_ACCESS (stmt_info)
&& (first_stmt_code == ARRAY_REF
|| first_stmt_code == BIT_FIELD_REF
|| first_stmt_code == INDIRECT_REF
|| first_stmt_code == COMPONENT_REF
|| first_stmt_code == MEM_REF)))
That is, it allowed any rhs_code as long as the first_stmt_code
looked valid. This had the effect of allowing IFN_MASK_LOAD
to be paired with an earlier non-call code (but didn't allow
the reverse).
This patch makes the check symmetrical.
gcc/
PR tree-optimization/103517
* tree-vect-slp.c (vect_build_slp_tree_1): When allowing two
different component references, check the codes of both them,
rather than just the first.
gcc/testsuite/
PR tree-optimization/103517
* gcc.dg/vect/pr103517.c: New test.
gcc/ChangeLog:
* doc/install.texi (Prerequisites): Add note that D front end now
requires GDC installed in order to bootstrap.
(Building): Add D compiler section, referencing prerequisites.
Adds documentation for the following:
- New switch that controls what code is generated on a contract
failure (throw or abort).
- New switch that controls mangling of D types in `extern(C++)`
code, as well as setting the compile-time value of
`__traits(getTargetInfo "cppStd")`
- New switches that generate C++ headers from D source files.
- New switch to save expanded mixins to a file.
- New switches that now distinguish between D language changes that
are either (a) an experimental feature or an upcoming breaking
change, (b) a warning or help on an upcoming change, or (c) revert
of a change for users who don't want to deal with the breaking
change for now.
gcc/d/ChangeLog:
* gdc.texi (Runtime Options): Document -fcheckaction=, -fextern-std=,
-fpreview=, -frevert=.
(Code Generation): Document -fdump-c++-spec=, -fdump-c++-spec-verbose,
-fsave-mixins=.
(Warnings): Update list of supported -ftransitions=.
gcc/ada/
* einfo.ads (E_Decimal_Fixed_Point_Subtype): Fix pasto.
* freeze.adb (Freeze_Fixed_Point_Type): Retrieve the underlying type
of the first subtype and do not use a stale value of Small_Value.
* sem_res.adb (Resolve_Real_Literal): In the case of a fixed-point
type, make sure that the base type is frozen and use its Small_Value
to compute the corresponding integer value of the literal.
gcc/ada/
* doc/gnat_rm/implementation_defined_pragmas.rst: Add
documentation of the new form of formal subprogram default in
the section on language extensions (pragma Extensions_Allowed).
* gnat_rm.texi: Regenerate.
* gen_il-gen-gen_nodes.adb: Add Expression as a syntactic field
of N_Formal_(Abstract|Concrete)_Subprogram_Declaration nodes.
* par-ch12.adb (P_Formal_Subprogram_Declaration): Add parsing
support for the new default of a parenthesized expression for
formal functions. Issue an error when extensions are not
allowed, suggesting use of -gnatX. Update comment with extended
syntax for SUBPROGRAM_DEFAULT.
* sem_ch12.adb (Analyze_Formal_Subprogram_Declaration): Issue an
error when an expression default is given for an abstract formal
function. When a default expression is present for a formal
function, install the function's formals and preanalyze the
expression.
(Instantiate_Formal_Subprogram): Fix typo in RM paragraph in a
comment. When a formal function has a default expression,
create a body for the function that will evaluate the expression
and will be called when the default applies in an instantiation.
The implicit function is marked as inlined and as having
convention Intrinsic.
