config
* gettext.m4 (AM_GNU_GETTEXT): Skip checks for the internal
symbols _nl_msg_cat_cntr, _nl_domain_bindings, and
_nl_expand_alias, if __GNU_GETTEXT_SUPPORTED_REVISION is defined.
Backport of gettext serial 68 patch.
gcc/
* config/h8300/logical.md (<code>qi3_1<cczn>): New pattern.
(andqi3_1<cczn>): Removed.
(<ors>qi3_1): Do not split for IOR/XOR a single bit.
(H8/SX bit logicals): Split out from other patterns.
* config/h8300/multiply.md (mulqihi3_const<cczn>): Renamed from
mulqihi3_const_clobber_flags.
(mulqihi3<cczn>, mulhisi3_const<cczn>, mulhisi3<cczn>): Similarly
Add red_zone_used to machine_function to track if red zone is used.
When expanding function prologue, set red_zone_used to true if red
zone is used.
gcc/
PR target/101023
* config/i386/i386.c (ix86_expand_prologue): Set red_zone_used
to true if red zone is used.
(ix86_output_indirect_jmp): Replace ix86_red_zone_size with
ix86_red_zone_used.
* config/i386/i386.h (machine_function): Add red_zone_used.
(ix86_red_zone_size): Removed.
(ix86_red_zone_used): New.
* config/i386/i386.md (peephole2 patterns): Replace
ix86_red_zone_size with ix86_red_zone_used.
gcc/testsuite/
PR target/101023
* g++.target/i386/pr101023a.C: New test.
* g++.target/i386/pr101023b.C: Likewise.
This patch checks that a '[PRnnnn]' and '(PRnnnn)' also appears as PR in the
changelog part of the commit message. And it does likewise for 'PR comp/nnnn'
except that then also the component name is checked. (Note that the reverse
is permitted, i.e. PR(s) only appearing in the changelog.)
To avoid false positives, PR numbers in the subject line are ignored,
if 'revert' appears.
Additionally, reject commits with a nonempty second line.
contrib/ChangeLog:
* gcc-changelog/git_commit.py (pr_regex): Add ?P<pr> for group('pr').
(subject_pr_regex, subject_pr2_regex): New.
(GitInfo.__init__, GitCommit.parse_changelog): Check subject PRs.
* gcc-changelog/git_email.py (SUBJECT_PREFIX, subject_patch_regex): New.
(GitEmail.__init__): Parse 'Subject:' and pass it to GitInfo.
* gcc-changelog/test_email.py (test_pr_only_in_subject,
test_wrong_pr_comp_in_subject, test_copyright_years): New.
* gcc-changelog/test_patches.txt (0030-PR-c-92746, pr-check1.patch):
Update to avoid triggering the new check.
(0001-rs6000-Support-doubleword, pr-wrong-comp.patch,
copyright-years.patch): New.
The C++17 standard (and C2x) says that [[maybe_unused]] may be applied to a
non-static data member, so we shouldn't warn about it. And I don't see a
reason not to handle a FIELD_DECL the same as any other decl, by setting
TREE_USED on it. It doesn't look like anything yet cares about that flag on
a FIELD_DECL, but setting it shouldn't hurt.
gcc/c-family/ChangeLog:
* c-attribs.c (handle_unused_attribute): Handle FIELD_DECL.
gcc/testsuite/ChangeLog:
* g++.dg/ext/attrib62.C: No longer warn.
* g++.dg/diagnostic/maybe_unused1.C: New test.
gcc/ChangeLog:
* doc/extend.texi (unused variable attribute): Applies to
structure fields as well.
build_vec_init_elt exits early if we're initializing a zero-element array,
so build_vec_init needs to do the same to avoid trying to instantiate things
after we've already started throwing important bits away.
PR c++/101029
gcc/cp/ChangeLog:
* init.c (build_vec_init): Shortcut [0] case.
gcc/testsuite/ChangeLog:
* g++.dg/ext/array4.C: New test.
While looking at template instantiation tracing, I noticed that we were
frequently looking up a particular class template instance while
instantiating it. This patch shortcuts that lookup, and speeds up compiling
stdc++.h with my (checking/unoptimized) compiler by about 3%.
gcc/cp/ChangeLog:
* pt.c (lookup_template_class_1): Shortcut current_class_type.
The effective target for a Power 9 runnable test should be
p9vector_hw.
2021-06-11 Carl Love <cel@us.ibm.com>
gcc/testsuite/ChangeLog
* gcc.target/powerpc/check-builtin-vec_rlnm-runnable.c
(dg-require-effective-target): Change target to p9vector_hw.
... instead of just the first clause.
gcc/
* tree-pretty-print.h (dump_omp_clauses): Add 'bool = true'
default argument.
* tree-pretty-print.c (dump_omp_clauses): Update.
(dump_generic_node) <OMP_CLAUSE>: Use it.
gcc/testsuite/
* gcc.dg/gomp/simd-clones-2.c: Enhance.
This makes tsubst_arg_types substitute into a function's parameter types
in left-to-right instead of right-to-left order, in accordance with DR 1227.
DR 1227
PR c++/96560
gcc/cp/ChangeLog:
* pt.c (tsubst_arg_types): Rearrange so that we substitute into
TYPE_ARG_TYPES in forward order while short circuiting
appropriately. Adjust formatting.
gcc/testsuite/ChangeLog:
* g++.dg/template/sfinae-dr1227.C: New test.
In C++23 there is a basic_string_view(Range&&) constructor, which is
constrained to take a range (specifically, a contiguous_range). When the
filesystem::path comparison operators call lhs.compare(rhs) the overload
taking a string_view is considered, which means checking whether path
satisfies the range concept. That satisfaction result changes depending
whether path::iterator is complete, which is ill-formed; no diagnostic
required. To avoid the problem, this change ensures that the overload
resolution is performed in a context where path::iterator is complete
and the range concept is satisfied. (The result of overload resolution
is always that the compare(const path&) overload is the best match, but
we still have to consider the compare(basic_string_view<value_type>) one
to decide if it even participates in overload resolution).
For std::filesystem::path we can't define the comparison operators later
in the file, because they are hidden friends, so a new helper is
introduced that gets defined when everything else is complete.
For std::experimental::filesystem::path we can just move the definitions
of the comparison operators later in the file.
Signed-off-by: Jonathan Wakely <jwakely@redhat.com>
libstdc++-v3/ChangeLog:
* include/bits/fs_path.h (operator==, operator<=>): Use new
_S_compare function.
(path::_S_compare): New function to call path::compare in a
context where path::iterator is complete.
* include/experimental/bits/fs_path.h (operator<, operator==):
Define after path::iterator is complete.
* testsuite/27_io/filesystem/path/native/conv_c++23.cc: New
test.
* testsuite/experimental/filesystem/path/native/conv_c++23.cc:
New test.
Improves semantic passes in the front-end around the `foreach' and
`static foreach' statements to be more resilient to compiling in a
minimal D runtime environment. Checking of the index type has been
improved as well so now there won't be needless compiler errors when
using 8 or 16-bit integers as index types when the size fits the
expected loop range.
gcc/d/ChangeLog:
PR d/100999
* dmd/MERGE: Merge upstream dmd 7a3808254.
libphobos/ChangeLog:
PR d/100999
* src/MERGE: Merge upstream phobos 55bb17543.
The patch implements a missed optimization enhancement. Under usual
IEEE rules, x+0.0 can't be simplified to x when x might potentially
be an IEEE minus zero (-0.0). The current logic in the middle-end
checks whether the type of x should honor signed zeros, but with this
patch we introduce tree_expr_maybe_real_minus_zero_p that allows us
to confirm that the value can't possibly be -0.0, for example, the result
of a conversion from an integer type, or the result of fabs (or has a
type that doesn't honor signed zero).
Whilst modifying match.pd, I also converted some additional folding
transformations from "testing the type" to "testing the value".
2020-06-10 Roger Sayle <roger@nextmovesoftware.com>
gcc/ChangeLog
PR tree-optimization/96392
* fold-const.c (fold_real_zero_addition_p): Take both arguments
of the addition or subtraction, not just the zero. Use this
other argument in tests for signaling NaNs and signed zeros.
(tree_expr_maybe_real_minus_zero_p): New predicate.
* fold-const.h (fold_real_zero_addition_p): Update prototype.
(tree_expr_maybe_real_minus_zero_p): New function prototype.
* match.pd: Update calls to fold_real_zero_addition_p.
Replace HONOR_NANS with tree_expr_maybe_nan_p.
Replace HONOR_SIGNED_ZEROS with tree_expr_maybe_real_minus_zero_p.
Replace HONOR_SNANS with tree_expr_maybe_signaling_nan_p.
* tree-ssa-reassoc.c (eliminate_using_constants): Update
call to fold_real_zero_addition_p.
gcc/testsuite/ChangeLog
PR tree-optimization/96392
* gcc.dg/pr96392.c: New test.
This plugs a hole in store-motion where it fails to perform dependence
checking on conditionally executed but not store-motioned refs.
2021-06-11 Richard Biener <rguenther@suse.de>
PR tree-optimization/101025
* tree-ssa-loop-im.c (sm_seq_valid_bb): Make sure to process
all refs that require dependence checking.
* gcc.dg/torture/pr101025.c: New testcase.
The current CMSE support in the multilib build for
"-march=armv8.1-m.main+mve -mfloat-abi=hard -mfpu=auto" is broken
as specified in PR99939 and this patch fixes the issue.
gcc/testsuite/ChangeLog:
2021-06-11 Srinath Parvathaneni <srinath.parvathaneni@arm.com>
PR target/99939
* gcc.target/arm/cmse/cmse-18.c: Add separate scan-assembler
directives check for target is v8.1-m.main+mve or not before
comparing the assembly output.
* gcc.target/arm/cmse/cmse-20.c: New test.
libgcc/ChangeLog:
2021-06-11 Srinath Parvathaneni <srinath.parvathaneni@arm.com>
PR target/99939
* config/arm/cmse_nonsecure_call.S: Add __ARM_FEATURE_MVE
macro.
* config/arm/t-arm: To link cmse.o and cmse_nonsecure_call.o
on passing -mcmse option.
The following patch implements consteval if support.
There is a new IF_STMT_CONSTEVAL_P flag on IF_STMT and IF_COND is
boolean_false_node to match the non-manifestly constant evaluation
behavior, while constexpr evaluation special-cases it. Perhaps cleaner
would be to set the condition to __builtin_is_constant_evaluated () call
but we need the IF_STMT_CONSTEVAL_P flag anyway and the IL would be larger.
And I'm not changing the libstdc++ side, where perhaps we could change
std::is_constant_evaluated definition for
#ifdef __cpp_if_consteval
case to if consteval { return true; } else { return false; }
but we need to keep it defined to __builtin_is_constant_evaluated ()
for C++20 or older.
2021-06-11 Jakub Jelinek <jakub@redhat.com>
PR c++/100974
gcc/c-family/
* c-cppbuiltin.c (c_cpp_builtins): Predefine __cpp_if_consteval for
-std=c++2b for P1938R3 consteval if support.
gcc/cp/
* cp-tree.h (struct saved_scope): Add consteval_if_p
member. Formatting fix for the discarded_stmt comment.
(in_consteval_if_p, IF_STMT_CONSTEVAL_P): Define.
* parser.c (cp_parser_lambda_expression): Temporarily disable
in_consteval_if_p when parsing lambda body.
(cp_parser_selection_statement): Parse consteval if.
* decl.c (struct named_label_entry): Add in_consteval_if member.
(level_for_consteval_if): New function.
(poplevel_named_label_1, check_previous_goto_1, check_goto): Handle
consteval if.
* constexpr.c (cxx_eval_builtin_function_call): Clarify in comment
why CP_BUILT_IN_IS_CONSTANT_EVALUATED needs to *non_constant_p
for !ctx->manifestly_const_eval.
(cxx_eval_conditional_expression): For IF_STMT_CONSTEVAL_P evaluate
condition as if it was __builtin_is_constant_evaluated call.
(potential_constant_expression_1): For IF_STMT_CONSTEVAL_P always
recurse on both branches.
* cp-gimplify.c (genericize_if_stmt): Genericize IF_STMT_CONSTEVAL_P
as the else branch.
* pt.c (tsubst_expr) <case IF_STMT>: Copy IF_STMT_CONSTEVAL_P.
Temporarily set in_consteval_if_p when recursing on
IF_STMT_CONSTEVAL_P then branch.
(tsubst_lambda_expr): Temporarily disable
in_consteval_if_p when instantiating lambda body.
* call.c (immediate_invocation_p): Return false when
in_consteval_if_p.
gcc/testsuite/
* g++.dg/cpp23/consteval-if1.C: New test.
* g++.dg/cpp23/consteval-if2.C: New test.
* g++.dg/cpp23/consteval-if3.C: New test.
* g++.dg/cpp23/consteval-if4.C: New test.
* g++.dg/cpp23/consteval-if5.C: New test.
* g++.dg/cpp23/consteval-if6.C: New test.
* g++.dg/cpp23/consteval-if7.C: New test.
* g++.dg/cpp23/consteval-if8.C: New test.
* g++.dg/cpp23/consteval-if9.C: New test.
* g++.dg/cpp23/consteval-if10.C: New test.
* g++.dg/cpp23/feat-cxx2b.C: Add __cpp_if_consteval tests.
While debugging another issue I noticed that the analyzer could fail to
merge nodes for control flow in which one path had called a function
and another path hadn't:
BB
/ \
/ \
fn call no fn call
\ /
\ /
join BB
The root cause was that the worklist sort function wasn't prioritizing
call strings, and thus it was fully exploring the "no function called"
path to the exit BB, and only then exploring the "within the function call"
parts of the "funcion called" path.
This patch prioritizes call strings when sorting the worklist so that
the nodes with deeper call strings are processed before those with shallower
call strings, thus allowing such nodes to be merged at the joinpoint.
gcc/analyzer/ChangeLog:
* engine.cc (worklist::key_t::cmp): Move sort by call_string to
before SCC.
gcc/testsuite/ChangeLog:
* gcc.dg/analyzer/loop-0-up-to-n-by-1-with-iter-obj.c: Update
expected number of enodes after the loop.
* gcc.dg/analyzer/paths-8.c: New test.
Signed-off-by: David Malcolm <dmalcolm@redhat.com>
This fixes a missing clearing of mismatched lanes from the
fatal fail path in SLP reassoc discovery in the most conservative
way.
2021-06-11 Richard Biener <rguenther@suse.de>
PR tree-optimization/101028
* tree-vect-slp.c (vect_build_slp_tree_2): When SLP
reassoc discovery fails fatally, mark appropriate lanes
in matches[] so.
* gcc.dg/pr101028.c: New testcase.
Since we cannot yet encode the operation in the SLP node itself
but need a representative stmt require an existing one for now
to avoid the need to build a fake GIMPLE stmt.
2021-06-11 Richard Biener <rguenther@suse.de>
PR tree-optimization/101026
* tree-vect-slp.c (vect_build_slp_tree_2): Make sure we
have a representative for the associated chain nodes.
* gfortran.dg/pr101026.f: New testcase.
simplify_relational_operation callees typically return just const0_rtx
or const_true_rtx and then simplify_relational_operation attempts to fix
that up if the comparison result has vector mode, or floating mode,
or punt if it has scalar mode and vector mode operands (it doesn't know how
exactly to deal with the scalar masks).
But, simplify_logical_relational_operation has a special case, where
it attempts to fold (x < y) | (x >= y) etc. and if it determines it is
always true, it just returns const_true_rtx, without doing the dances that
simplify_relational_operation does.
That results in an ICE on the following testcase, where such folding happens
during expansion (of debug stmts into DEBUG_INSNs) and we ICE because
all of sudden a VOIDmode rtx appears where it expects a vector (V4SImode)
rtx.
The following patch fixes that by moving the adjustement into a separate
helper routine and using it from both simplify_relational_operation and
simplify_logical_relational_operation.
2021-06-11 Jakub Jelinek <jakub@redhat.com>
PR rtl-optimization/101008
* simplify-rtx.c (relational_result): New function.
(simplify_logical_relational_operation,
simplify_relational_operation): Use it.
* gcc.dg/pr101008.c: New test.
Some permutations can be implemented without costly PSHUFB instruction, e.g.:
{ 8,9,10,11,12,13,14,15, 0,1,2,3,4,5,6,7 } with PALIGNR,
{ 0,1,2,3, 4,5,6,7, 4,5,6,7, 12,13,14,15 } with PSHUFD,
{ 0,1, 2,3, 2,3, 6,7, 8,9,10,11,12,13,14,15 } with PSHUFLW and
{ 0,1,2,3,4,5,6,7, 8,9, 10,11, 10,11, 14,15 } with PSHUFHW.
All these instructions have constant shuffle control mask and do not
need to load shuffle mask from a memory to a temporary XMM register.
2021-06-11 Uroš Bizjak <ubizjak@gmail.com>
gcc/
PR target/101021
* config/i386/i386-expand.c (expand_vec_perm_pshufb): Return
false if the permutation can be implemented with constant
permutation instruction in wider mode.
(canonicalize_vector_int_perm): Move above expand_vec_perm_pshufb.
Handle V8QImode and V4HImode.
gcc/testsuite/
PR target/101021
* gcc.target/i386/pr101021-1.c: New test.
* gcc.target/i386/pr101021-2.c: Ditto.
This fixes recording of distance vectors in case the DDR has just
constant equal indexes. In that case we expect distance vectors
with zero distances to be recorded which is what was done when
any distance was computed for affine indexes.
2021-06-11 Richard Biener <rguenther@suse.de>
PR middle-end/101009
* tree-data-ref.c (build_classic_dist_vector_1): Make sure
to set *init_b to true when we encounter a constant equal
index pair.
(compute_affine_dependence): Also dump the actual DR_REF.
* gcc.dg/torture/pr101009.c: New testcase.
For some cases that when we load unsigned char/short values from
the appropriate unsigned char/short memories and convert them to
double/single precision floating point value, there would be
implicit conversions to int first. It makes GCC not leverage the
P9 instructions lxsibzx/lxsihzx. This patch is to add the related
define_insn_and_split to support this kind of scenario.
Bootstrapped/regtested on powerpc64le-linux-gnu P9 and
powerpc64-linux-gnu P8.
gcc/ChangeLog:
* config/rs6000/rs6000.md
(floatsi<SFDF:mode>2_lfiwax_<QHI:mode>_mem_zext): New
define_insn_and_split.
gcc/testsuite/ChangeLog:
* gcc.target/powerpc/p9-fpcvt-3.c: New test.
This should preserve the original association order as much as possible
for the initial SLP discovery attempt and also improve consistency.
2021-06-11 Richard Biener <rguenther@suse.de>
* tree-vect-slp.c (vect_build_slp_tree_2): Use stablesort
to sort operands of the associative chain.
This makes it possible to apply GCCs stable sort algorithm to vec<>
and also use it with the qsort_r compatible interface.
2021-06-10 Richard Biener <rguenther@suse.de>
* system.h (gcc_stablesort_r): Declare.
* sort.cc (gcc_sort_r): Support stable sort.
(gcc_stablesort_r): Define.
* vec.h (vec<>::stablesort): Add.
Jakub pointed me at
<http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2021/p1938r3.html#compiler-warnings>
which shows that our existing warning could be extended to handle more
cases. This patch implements that.
A minor annoyance was handling macros, in libstdc++ we have
reference operator[](size_type __pos) {
__glibcxx_assert(__pos <= size());
...
}
wherein __glibcxx_assert expands to
if (__builtin_is_constant_evaluated() && !bool(__pos <= size())
...
but I'm of a mind to not warn on that.
Once consteval if makes it in, we should tweak this warning one more
time.
PR c++/100995
gcc/cp/ChangeLog:
* constexpr.c (maybe_constexpr_fn): New.
* cp-tree.h (maybe_constexpr_fn): Declare.
* semantics.c (find_std_constant_evaluated_r): New.
(maybe_warn_for_constant_evaluated): New.
(finish_if_stmt_cond): Call it.
gcc/testsuite/ChangeLog:
* g++.dg/cpp2a/is-constant-evaluated9.C: Add dg-warning.
* g++.dg/cpp2a/is-constant-evaluated12.C: New test.
During deduction, when the template of the argument for a bound ttp
is a template template parameter, we need to consider the
TEMPLATE_TEMPLATE_PARAMETER for matching rather than the TEMPLATE_DECL
thereof, because the canonical form of a template template parameter as
a template argument is the former tree, not the latter.
PR c++/67829
gcc/cp/ChangeLog:
* pt.c (unify) <case BOUND_TEMPLATE_TEMPLATE_PARM>: When
the TEMPLATE_DECL of a BOUND_TEMPLATE_TEMPLATE_PARM argument is
a template template parameter, adjust to the
TEMPLATE_TEMPLATE_PARAMETER before falling through.
gcc/testsuite/ChangeLog:
* g++.dg/template/ttp34.C: New test.
* g++.dg/template/ttp34a.C: New test.
* g++.dg/template/ttp34b.C: New test.
Here the satisfaction cache is conflating the satisfaction value of the
two return-type-requirements because the corresponding constrained
'auto's have level 2, but they capture an empty current_template_parms.
This ultimately causes the satisfaction cache to think the type
constraint doesn't depend on the deduced type of the expression.
When normalizing the constraints on an 'auto', the assumption made by
normalize_placeholder_type_constraints is that the level of the 'auto'
is one greater than the depth of the captured current_template_parms, an
assumption which is not holding here. So this patch just makes n_p_t_c
adjust the normalization context appropriately in this situation.
PR c++/100946
gcc/cp/ChangeLog:
* constraint.cc (normalize_placeholder_type_constraints): When
normalizing a non-templated return-type-requirement, add a dummy
level to initial_parms.
gcc/testsuite/ChangeLog:
* g++.dg/cpp2a/concepts-return-req3.C: New test.
The __builtin_vsx_assemble_pair and __builtin_mma_assemble_acc built-ins
currently assign their first source operand to the first VSX register
in a pair/quad, their second operand to the second register in a pair/quad, etc.
This is not endian friendly and forces the user to generate different calls
depending on endianness. In agreement with the POWER LLVM team, we've
decided to lightly deprecate the assemble built-ins and replace them with
"build" built-ins that automatically handle endianness so the same built-in
call and be used for both little-endian and big-endian compiles. We are not
removing the assemble built-ins, since there is code in the wild that use
them, but we are removing their documentation to encourage the use of the
new "build" variants.
gcc/
* config/rs6000/rs6000-builtin.def (build_pair): New built-in.
(build_acc): Likewise.
* config/rs6000/rs6000-call.c (mma_expand_builtin): Swap assemble
source operands in little-endian mode.
(rs6000_gimple_fold_mma_builtin): Handle VSX_BUILTIN_BUILD_PAIR.
(mma_init_builtins): Likewise.
* config/rs6000/rs6000.c (rs6000_split_multireg_move): Handle endianness
ordering for the MMA assemble and build source operands.
* doc/extend.texi (__builtin_vsx_build_acc, __builtin_mma_build_pair):
Document.
(__builtin_mma_assemble_acc, __builtin_mma_assemble_pair): Remove
documentation.
gcc/testsuite/
* gcc.target/powerpc/mma-builtin-4.c (__builtin_vsx_build_pair): Add
tests. Update expected counts.
* gcc.target/powerpc/mma-builtin-5.c (__builtin_mma_build_acc): Add
tests. Update expected counts.
Generate a stub TypeInfo class even if the root Object class is missing.
The front-end will take care of issuing an error and abort the
compilation when running semantic on constructed TypeInfo objects.
The errors issued by the code generation pass relating to missing or
disabled RTTI has been consolidated into a single function, so that a
meaningful error will be emitted before the front-end terminates.
gcc/d/ChangeLog:
PR d/100967
* d-frontend.cc (getTypeInfoType): Move TypeInfo checks to
check_typeinfo_type and call new function.
* d-tree.h (check_typeinfo_type): Declare.
* typeinfo.cc: Include dmd/scope.h.
(create_frontend_tinfo_types): Generate front-end types even if Object
is missing.
(build_typeinfo): Move TypeInfo checks to check_typeinfo_type and call
new function.
(check_typeinfo_type): New function.
gcc/testsuite/ChangeLog:
PR d/100967
* gdc.dg/pr100967.d: New test.
The *TItype_ppc definitions are guarded by _ARCH_PPC64, so all
declarations using it should do so as well.
2021-06-10 Segher Boessenkool <segher@kernel.crashing.org>
libgcc/
* config/rs6000/quad-float128.h: Guard all uses of [U]TItype_ppc by
_ARCH_PPC64 .
There is a mismatch between the new and the delete for the
ssa_equiv_stack class. The correct idiom should have been delete[].
It has been pointed out that perhaps a better alternative is to use
an auto_vec which does everything automatically. Plus, it is more
consistent with m_stack which is already an auto_vec.
This patch fixes the issue in PR100984.
Tested on x86-64 Linux.
gcc/ChangeLog:
PR tree-optimization/100984
* gimple-ssa-evrp.c (ssa_equiv_stack): Use auto_vec for
replacements table.
(ssa_equiv_stack::~ssa_equiv_stack): Remove.