From 9.7.4 in Parallelism TS 2. For some reason I overlooked these two
functions. Implement them via call to _S_reduce.
libstdc++-v3/ChangeLog:
* include/experimental/bits/simd.h: Add __detail::_Minimum and
__detail::_Maximum to use them as _BinaryOperation to _S_reduce.
Add hmin and hmax overloads for simd and const_where_expression.
* include/experimental/bits/simd_scalar.h
(_SimdImplScalar::_S_reduce): Make unused _BinaryOperation
parameter const-ref to allow calling _S_reduce with an rvalue.
* testsuite/experimental/simd/tests/reductions.cc: Add tests for
hmin and hmax. Since the compiler statically determined that all
tests pass, repeat the test after a call to make_value_unknown.
In many failure cases it is helpful to inspect the instructions leading
up to the test failure. After this change the location is easier to find
and the branch after failure is easier to find.
libstdc++-v3/ChangeLog:
* testsuite/experimental/simd/tests/bits/verify.h (verify): Add
instruction pointer data member. Ensure that the `if (m_failed)`
branch is always inlined into the calling code. The body of the
conditional can still be a function call. Move the get_ip call
into the verify ctor to simplify the ctor calls.
(COMPARE): Don't mention the use of all_of for reduction of a
simd_mask. It only distracts from the real issue.
libstdc++-v3/ChangeLog:
* testsuite/experimental/simd/driver.sh: Abstract reading test
options into read_src_option function. Read skip, only,
expensive, and xfail via read_src_option. Add timeout and
timeout-factor options and adjust timeout variable accordingly.
* testsuite/experimental/simd/tests/loadstore.cc: Set
timeout-factor 2.
Handle overly large output by aborting the log and thus the test. This
is a similar condition to a timeout.
libstdc++-v3/ChangeLog:
* testsuite/experimental/simd/driver.sh: When handling the pipe
to log (and on verbose to stdout) count the lines. If it exceeds
1000 log the issue and exit 125, which is then handled as a
failure.
std::hypot(a, b, c) is imprecise and makes this test fail even though
the failure is unrelated to simd.
libstdc++-v3/ChangeLog:
* testsuite/experimental/simd/tests/hypot3_fma.cc: Add skip:
markup for long double on powerpc64*.
POWER7 does not support __vector long long reductions, making the
generic _S_popcount implementation ill-formed. Specializing _S_popcount
for PPC allows optimization and avoids the issue.
libstdc++-v3/ChangeLog:
* include/experimental/bits/simd.h: Add __have_power10vec
conditional on _ARCH_PWR10.
* include/experimental/bits/simd_builtin.h: Forward declare
_MaskImplPpc and use it as _MaskImpl when __ALTIVEC__ is
defined.
(_MaskImplBuiltin::_S_some_of): Call _S_popcount from the
_SuperImpl for optimizations and correctness.
* include/experimental/bits/simd_ppc.h: Add _MaskImplPpc.
(_MaskImplPpc::_S_popcount): Implement via vec_cntm for POWER10.
Otherwise, for >=int use -vec_sums divided by a sizeof factor.
For <int use -vec_sums(vec_sum4s(...)) to sum all mask entries.
libstdc++-v3/ChangeLog:
* testsuite/experimental/simd/driver.sh: Remove executable on
SIGINT. Process compiler and test executable output: In verbose
mode print messages immediately, limited to 1000 lines and
breaking long lines to below $COLUMNS (or 1024 if not set).
Communicating the exit status of the compiler / test with the
necessary pipe is done via a message through stdout/-in.
libstdc++-v3/ChangeLog:
* testsuite/experimental/simd/generate_makefile.sh: Use
different variables internally than documented for user
overrides. This makes internal append/prepend work as intended.
libstdc++-v3/ChangeLog:
* testsuite/experimental/simd/driver.sh (verify_test): Print
test output on run xfail. Do not repeat lines from the log that
were already printed on stdout.
(test_selector): Make the compiler flags pattern usable as a
substring selector.
(toplevel): Trap on SIGINT and remove the log and sum files.
Call timout with --foreground to quickly terminate on SIGINT.
* testsuite/experimental/simd/generate_makefile.sh: Simplify run
targets via target patterns. Default DRIVEROPTS to -v for run
targets. Remove log and sum files after completion of the run
target (so that it's always recompiled).
Place help text into text file for reasonable 'make help'
performance.
libstdc++-v3/ChangeLog:
* include/experimental/bits/simd.h: Let __intrinsic_type<long
double, N> be valid if sizeof(long double) == sizeof(double) and
use a __vector double as member type.
Intrinsics types for NEON differ from gnu::vector_size types now. This
requires explicit specializations for __intrinsic_type and a new
__is_intrinsic_type trait.
libstdc++-v3/ChangeLog:
* include/experimental/bits/simd.h (__is_intrinsic_type): New
internal type trait. Alias for __is_vector_type on x86.
(_VectorTraitsImpl): Enable for __intrinsic_type in addition for
__vector_type.
(__intrin_bitcast): Allow casting to & from vector & intrinsic
types.
(__intrinsic_type): Explicitly specialize for NEON intrinsic
vector types.
libstdc++-v3/ChangeLog:
* testsuite/experimental/simd/driver.sh: Implement skip, only,
expensive, and xfail markers. They can select on type, ABI tag
subset number, target-triplet, and compiler flags.
* testsuite/experimental/simd/generate_makefile.sh: The summary
now includes lines for unexpected passes and expected failures.
If the skip or only markers are only conditional on the type, do
not generate rules for those types.
* testsuite/experimental/simd/tests/abs.cc: Mark test expensive
for ABI tag subsets 1-9.
* testsuite/experimental/simd/tests/algorithms.cc: Ditto.
* testsuite/experimental/simd/tests/broadcast.cc: Ditto.
* testsuite/experimental/simd/tests/casts.cc: Ditto.
* testsuite/experimental/simd/tests/generator.cc: Ditto.
* testsuite/experimental/simd/tests/integer_operators.cc: Ditto.
* testsuite/experimental/simd/tests/loadstore.cc: Ditto.
* testsuite/experimental/simd/tests/mask_broadcast.cc: Ditto.
* testsuite/experimental/simd/tests/mask_conversions.cc: Ditto.
* testsuite/experimental/simd/tests/mask_implicit_cvt.cc: Ditto.
* testsuite/experimental/simd/tests/mask_loadstore.cc: Ditto.
* testsuite/experimental/simd/tests/mask_operator_cvt.cc: Ditto.
* testsuite/experimental/simd/tests/mask_operators.cc: Ditto.
* testsuite/experimental/simd/tests/mask_reductions.cc: Ditto.
* testsuite/experimental/simd/tests/operator_cvt.cc: Ditto.
* testsuite/experimental/simd/tests/operators.cc: Ditto.
* testsuite/experimental/simd/tests/reductions.cc: Ditto.
* testsuite/experimental/simd/tests/simd.cc: Ditto.
* testsuite/experimental/simd/tests/split_concat.cc: Ditto.
* testsuite/experimental/simd/tests/splits.cc: Ditto.
* testsuite/experimental/simd/tests/where.cc: Ditto.
* testsuite/experimental/simd/tests/fpclassify.cc: Ditto. In
addition replace "test only floattypes" marker by unconditional
"float|double|ldouble" only marker.
* testsuite/experimental/simd/tests/frexp.cc: Ditto.
* testsuite/experimental/simd/tests/hypot3_fma.cc: Ditto.
* testsuite/experimental/simd/tests/ldexp_scalbn_scalbln_modf.cc:
Ditto.
* testsuite/experimental/simd/tests/logarithm.cc: Ditto.
* testsuite/experimental/simd/tests/math_1arg.cc: Ditto.
* testsuite/experimental/simd/tests/math_2arg.cc: Ditto.
* testsuite/experimental/simd/tests/remqo.cc: Ditto.
* testsuite/experimental/simd/tests/trigonometric.cc: Ditto.
* testsuite/experimental/simd/tests/trunc_ceil_floor.cc: Ditto.
* testsuite/experimental/simd/tests/sincos.cc: Ditto. In
addition, xfail on run because the reference data is missing.
Somehow I lost my fixes based on Segher's review of the p10
logical-logical fusion patch. This commit adds the changes that
should have been there before. Bootstrap/regtest passed.
gcc/ChangeLog:
* config/rs6000/genfusion.pl (gen_2logical): Add missing
fixes based on patch review.
* config/rs6000/fusion.md: Regenerate file.
In a previous fusion-combine patch for rs6000, Segher had asked me to
comment out the automatic regeneration of fusion.md. And more recently
Edelsohn pointed out that gcc_update needed to fix the timestamp of
fusion.md so it didn't get unnecessarily regenerated.
contrib/ChangeLog:
* gcc_update (files_and_dependencies): Add dependency for
gcc/config/rs6000/fusion.md on gcc/config/rs6000/genfusion.md.
gcc/ChangeLog:
* config/rs6000/t-rs6000: Comment out auto generation of
fusion.md for now.
Here we crash with a noexcept-specifier in a nested template class,
because my handling of such deferred-parse noexcept-specifiers was
gronked when we need to instantiate a DEFERRED_PARSE before it was
actually parsed at the end of the outermost class.
In
struct S {
template<class> struct B {
B() noexcept(noexcept(x));
int x;
};
struct A : B<int> {
A() : B() {}
};
};
we call complete_type for B<int> which triggers tsubsting S::B<int>::B()
whose noexcept-specifier still contains a DEFERRED_PARSE. The trick is
to stash such noexcept-specifiers into DEFPARSE_INSTANTIATIONS so that
we can replace it later when we've finally parsed all deferred
noexcept-specifiers.
In passing, fix missing usage of UNPARSED_NOEXCEPT_SPEC_P.
gcc/cp/ChangeLog:
PR c++/98899
* parser.c (cp_parser_class_specifier_1): Use any possible
DEFPARSE_INSTANTIATIONS to update DEFERRED_NOEXCEPT_PATTERN.
(cp_parser_save_noexcept): Initialize DEFPARSE_INSTANTIATIONS.
* pt.c (tsubst_exception_specification): Stash new_specs into
DEFPARSE_INSTANTIATIONS.
* tree.c (fixup_deferred_exception_variants): Use
UNPARSED_NOEXCEPT_SPEC_P.
gcc/testsuite/ChangeLog:
PR c++/98899
* g++.dg/cpp0x/noexcept65.C: New test.
Suppose we have:
(insn (set (reg:FPRX2 70) (subreg:FPRX2 (reg/v:TF 63) 0)))
where operand_loc[0] points to r70 and operand_loc[1] points to r63.
If r63 is spilled, remove_pseudos() will change this insn to:
(insn (set (reg:FPRX2 70)
(subreg:FPRX2 (mem/c:TF (plus:DI (reg:DI %fp)
(const_int 144))))))
This is fine so far: rtx pointed to by operand_loc[1] has been changed
from (reg) to (mem), but its slot is still under (subreg). However,
alter_subreg() will simplify this insn to:
(insn (set (reg:FPRX2 70)
(mem/c:FPRX2 (plus:DI (reg:DI %fp) (const_int 144)))))
The (subreg) is gone, and therefore operand_loc[1] is no longer valid.
This will prevent process_insn_for_elimination() from updating the spill
slot offset, causing miscompilation: different instructions will refer
to the same spill slot using different offsets.
Fix by clearing all the cached data, and not just used_insn_alternative.
gcc/ChangeLog:
2021-01-13 Ilya Leoshkevich <iii@linux.ibm.com>
* lra-spills.c (remove_pseudos): Call lra_update_insn_recog_data()
after calling alter_subreg() on a (mem).
It can yield an incorrect layout when there is a partial representation
clause on a discriminated record type with a variant part.
gcc/ada/
* gcc-interface/decl.c (components_to_record): If the first component
with rep clause is the _Parent field with variable size, temporarily
set it aside when computing the internal layout of the REP part again.
* gcc-interface/utils.c (finish_record_type): Revert to taking the
maximum when merging sizes for all record types with rep clause.
(merge_sizes): Put SPECIAL parameter last and adjust recursive calls.
This polishes a few rough edges visible in LTO mode.
gcc/ada/
* gcc-interface/decl.c (gnat_to_gnu_entity) <E_Array_Type>: Make the
two fields of the fat pointer type addressable, and do not make the
template type read-only.
<E_Record_Type>: If the type has discriminants mark it as may_alias.
* gcc-interface/utils.c (make_dummy_type): Likewise.
(build_dummy_unc_pointer_types): Likewise.
gcc/fortran/ChangeLog:
PR fortran/98913
* dependency.c (gfc_dep_resolver): Treat local access
to coarrays like any array access in dependency analysis.
gcc/testsuite/ChangeLog:
PR fortran/98913
* gfortran.dg/coarray/array_temporary.f90: New test.
This fixes more memory leaks as discovered by building 521.wrf_r.
2021-02-03 Richard Biener <rguenther@suse.de>
* lto-streamer.c (lto_get_section_name): Free temporary
buffer.
* tree-loop-distribution.c
(loop_distribution::merge_dep_scc_partitions): Free edge data.
As the testcase shows, RTL ifcvt can throw random RTL (whatever it found in
some insns) at expand_binop or expand_unop and expects it to do something
(and then will check if it created valid insns and punts if not).
These functions in the end if the operands don't match try to
copy_to_mode_reg the operands, which does
if (!general_operand (x, VOIDmode))
x = force_operand (x, temp);
but, force_operand is far from handling all possible RTLs, it will ICE for
all more unusual RTL codes. Basically handles just simple arithmetic and
unary RTL operations if they have an optab and
expand_simple_binop/expand_simple_unop ICE on others.
The following patch fixes it by adding some operand verification (whether
there is a hope that copy_to_mode_reg will succeed on those). It is added
both to noce_emit_move_insn (not needed for this exact testcase,
that function simply tries to recog the insn as is and if it fails,
handles some simple binop/unop cases; the patch performs the verification
of their operands) and noce_try_sign_mask.
2021-02-03 Jakub Jelinek <jakub@redhat.com>
PR middle-end/97487
* ifcvt.c (noce_can_force_operand): New function.
(noce_emit_move_insn): Use it.
(noce_try_sign_mask): Likewise. Formatting fix.
* gcc.dg/pr97487-1.c: New test.
* gcc.dg/pr97487-2.c: New test.
The following testcase has ice-on-invalid, it can't be reloaded, but we
shouldn't ICE the compiler because the user typed non-sense.
In current_insn_transform we have:
if (process_alt_operands (reused_alternative_num))
alt_p = true;
if (check_only_p)
return ! alt_p || best_losers != 0;
/* If insn is commutative (it's safe to exchange a certain pair of
operands) then we need to try each alternative twice, the second
time matching those two operands as if we had exchanged them. To
do this, really exchange them in operands.
If we have just tried the alternatives the second time, return
operands to normal and drop through. */
if (reused_alternative_num < 0 && commutative >= 0)
{
curr_swapped = !curr_swapped;
if (curr_swapped)
{
swap_operands (commutative);
goto try_swapped;
}
else
swap_operands (commutative);
}
if (! alt_p && ! sec_mem_p)
{
/* No alternative works with reloads?? */
if (INSN_CODE (curr_insn) >= 0)
fatal_insn ("unable to generate reloads for:", curr_insn);
error_for_asm (curr_insn,
"inconsistent operand constraints in an %<asm%>");
lra_asm_error_p = true;
...
and so handle inline asms there differently (and delete/nullify them after
this) - fatal_insn is only called for non-inline asm.
But in process_alt_operands we do:
/* Both the earlyclobber operand and conflicting operand
cannot both be user defined hard registers. */
if (HARD_REGISTER_P (operand_reg[i])
&& REG_USERVAR_P (operand_reg[i])
&& operand_reg[j] != NULL_RTX
&& HARD_REGISTER_P (operand_reg[j])
&& REG_USERVAR_P (operand_reg[j]))
fatal_insn ("unable to generate reloads for "
"impossible constraints:", curr_insn);
and thus ICE even for inline-asms.
I think it is inappropriate to delete/nullify the insn in
process_alt_operands, as it could be done e.g. in the check_only_p mode,
so this patch just returns false in that case, which results in the
caller have alt_p false, and as inline asm isn't simple move, sec_mem_p
will be also false (and it isn't commutative either), so for check_only_p
it will suggests to the callers it isn't ok and otherwise will emit
error and delete/nullify the inline asm insn.
2021-02-03 Jakub Jelinek <jakub@redhat.com>
PR middle-end/97971
* lra-constraints.c (process_alt_operands): For inline asm, don't call
fatal_insn, but instead return false.
* gcc.target/i386/pr97971.c: New test.
On Tue, Feb 02, 2021 at 02:23:55PM +0100, Richard Biener wrote:
> All I say is that the x86 target
> should either not advertise V1DF shifts or advertise the basic
> ops that reasonable simplification would expect to exist.
The backend has several V1?Imode shifts, but optab only for those V1DImode
ones:
grep '[la]sh[lr]v1[qhsdtox]' tmp-mddump.md
(define_insn ("mmx_ashlv1di3")
(define_insn ("mmx_lshrv1di3")
(define_insn ("avx512bw_ashlv1ti3")
(define_insn ("avx512bw_lshrv1ti3")
(define_insn ("sse2_ashlv1ti3")
(define_insn ("sse2_lshrv1ti3")
(define_expand ("ashlv1di3")
(define_expand ("lshrv1di3")
emit_insn (gen_sse2_lshrv1ti3 (tmp, gen_lowpart (V1TImode, operands[1]),
I think it has been introduced with
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89021#c13
Before we didn't have any V1DImode expanders (except mov/movmisalign, but
those are needed and are supplied for other V1??mode modes too).
This patch just removes the two V1DImode shift expanders with standard names.
2021-02-03 Jakub Jelinek <jakub@redhat.com>
PR tree-optimization/98287
* config/i386/mmx.md (<insn><mode>3): For shifts don't enable expander
for V1DImode.
* gcc.dg/pr98287.c: New test.
Previously the SLP pattern matcher was using STMT_VINFO_SLP_VECT_ONLY as a way
to dissolve the SLP only patterns during SLP cancellation. However it seems
like the semantics for STMT_VINFO_SLP_VECT_ONLY are slightly different than what
I expected.
Namely that the non-SLP path can still use a statement marked
STMT_VINFO_SLP_VECT_ONLY. One such example is masked loads which are used both
in the SLP and non-SLP path.
To fix this I now introduce a new flag STMT_VINFO_SLP_VECT_ONLY_PATTERN which is
used only by the pattern matcher.
gcc/ChangeLog:
PR tree-optimization/98928
* tree-vect-loop.c (vect_analyze_loop_2): Change
STMT_VINFO_SLP_VECT_ONLY to STMT_VINFO_SLP_VECT_ONLY_PATTERN.
* tree-vect-slp-patterns.c (complex_pattern::build): Likewise.
* tree-vectorizer.h (STMT_VINFO_SLP_VECT_ONLY_PATTERN): New.
(class _stmt_vec_info): Add slp_vect_pattern_only_p.
gcc/testsuite/ChangeLog:
PR tree-optimization/98928
* gcc.target/i386/pr98928.c: New test.
My patch for 96199 had us re-substitute the parameter types of a constructor
in order to rewrite mentions of members into dependent references. We need
to do that for member functions, too.
gcc/cp/ChangeLog:
PR c++/98929
PR c++/96199
* error.c (dump_expr): Ignore dummy object.
* pt.c (tsubst_baselink): Handle dependent scope.
gcc/testsuite/ChangeLog:
PR c++/98929
* g++.dg/cpp1z/class-deduction-decltype1.C: New test.
Another very simple move from inline asm to builtins.
Only two intrinsics this time.
gcc/ChangeLog:
* config/aarch64/aarch64-simd-builtins.def (ursqrte): Define builtin.
* config/aarch64/aarch64-simd.md (aarch64_ursqrte<mode>): New pattern.
* config/aarch64/arm_neon.h (vrsqrte_u32): Reimplement using builtin.
(vrsqrteq_u32): Likewise.
Another transition from inline asm to builtin.
Only 3 intrinsics converted this time but they use the "+w" constraint in their inline asm
so are more likely to generate redundant moves so benefit more from reimplementation.
gcc/ChangeLog:
* config/aarch64/aarch64-simd-builtins.def (sqxtun2): Define builtin.
* config/aarch64/aarch64-simd.md (aarch64_sqxtun2<mode>_le): Define.
(aarch64_sqxtun2<mode>_be): Likewise.
(aarch64_sqxtun2<mode>): Likewise.
* config/aarch64/arm_neon.h (vqmovun_high_s16): Reimplement using builtin.
(vqmovun_high_s32): Likewise.
(vqmovun_high_s64): Likewise.
* config/aarch64/iterators.md (UNSPEC_SQXTUN2): Define.
gcc/testsuite/ChangeLog:
* gcc.target/aarch64/narrow_high-intrinsics.c: Adjust sqxtun2 scan.
This patch updates the flags for the bfloat16 builtins.
The bfdot ones aren't affected by the FPCR/FPSR so can be AUTO_FP
whereas the bfmlal ones follow the normal floating-point instructions and get FP.
gcc/ChangeLog:
* config/aarch64/aarch64-simd-builtins.def (bfdot_lane, bfdot_laneq): Use
AUTO_FP flags.
(bfmlalb_lane, bfmlalt_lane, bfmlalb_lane_q, bfmlalt_lane_q): Use FP flags.
We already have a STORE flag that we use for builtins. This patch introduces a LOAD set
that uses AUTO_FP and FLAG_READ_MEMORY. This allows for more aggressive optimisation of the load
intrinsics.
Turns out we have a great many testcases that do:
float16x4x2_t
f_vld2_lane_f16 (float16_t * p, float16x4x2_t v)
{
float16x4x2_t res;
/* { dg-error "lane 4 out of range 0 - 3" "" { target *-*-* } 0 } */
res = vld2_lane_f16 (p, v, 4);
/* { dg-error "lane -1 out of range 0 - 3" "" { target *-*-* } 0 } */
res = vld2_lane_f16 (p, v, -1);
return res;
}
but since the first res is unused it now gets eliminated early on before we get to give an error
message. Ideally we'd like to warn for both.
This patch takes the conservative approach and doesn't convert the load-lane builtins to LOAD ;
that's something we can improve later.
gcc/ChangeLog:
* config/aarch64/aarch64-builtins.c (FLAG_LOAD): Define.
* config/aarch64/aarch64-simd-builtins.def (ld1x2, ld2, ld3, ld4, ld2r,
ld3r, ld4r, ld1, ld1x3, ld1x4): Use LOAD flags.
This patch relaxes the flags for some builtins to AUTO_FP. These
builtins do permutes and similar, so they shouldn't get the FP flags
when operating on floating-point modes as they don't care about
FPCR/FPSR and exceptions.
gcc/ChangeLog:
* config/aarch64/aarch64-simd-builtins.def (combine, zip1, zip2,
uzp1, uzp2, trn1, trn2, simd_bsl): Use AUTO_FP flags.
This patch relaxes the flags for most integer builtins to NONE as they don't read/write memory
and don't care about the FPCR/FPSR or exceptions so we should be more aggressive with them.
This leads to fallout in a testcase where the result of an intrinsic was unused and it is now
DCE'd. The testcase is adjusted.
gcc/ChangeLog:
* config/aarch64/aarch64-simd-builtins.def (clrsb, clz, ctz, popcount,
vec_smult_lane_, vec_smlal_lane_, vec_smult_laneq_, vec_smlal_laneq_,
vec_umult_lane_, vec_umlal_lane_, vec_umult_laneq_, vec_umlal_laneq_,
ashl, sshl, ushl, srshl, urshl, sdot_lane, udot_lane, sdot_laneq,
udot_laneq, usdot_lane, usdot_laneq, sudot_lane, sudot_laneq, ashr,
ashr_simd, lshr, lshr_simd, srshr_n, urshr_n, ssra_n, usra_n, srsra_n,
ursra_n, sshll_n, ushll_n, sshll2_n, ushll2_n, ssri_n, usri_n, ssli_n,
ssli_n, usli_n, bswap, rbit, simd_bsl, eor3q, rax1q, xarq, bcaxq): Use
NONE builtin flags.
gcc/testsuite/ChangeLog:
* gcc.target/aarch64/arg-type-diagnostics-1.c: Return result from foo.
As discussed in the PR, the reduction code isn't able to cope with type
promotions/demotions in the reduction computation, so if we recognize an
over-widening pattern that has vect_reduction_def type, we most likely make
it non-vectorizable.
2021-02-02 Jakub Jelinek <jakub@redhat.com>
PR tree-optimization/98848
* tree-vect-patterns.c (vect_recog_over_widening_pattern): Punt if
STMT_VINFO_DEF_TYPE (last_stmt_info) is vect_reduction_def.
* gcc.dg/vect/pr98848.c: New test.
* gcc.dg/vect/pr92205.c: Remove xfail.
This testcase has been fixed by
r11-5904-g4cf70c20cb10acd6fb1016611d05540728176b60
so I'm checking it in so that we can close the PR.
2021-02-02 Jakub Jelinek <jakub@redhat.com>
PR tree-optimization/97960
* g++.dg/torture/pr97960.C: New test.
- Check `from` mode is not BLMmode before call store_expr, calling store_expr
with BLKmode will cause ICE.
- Verified with riscv64, x86_64 and aarch64, no introduce new regression.
Note: Those logic was introduced by 3e60ddeb82,
so I cc Jakub for reivew.
Changes for V2:
- Checking mode of `from` rather than mode of `to`.
- Verified on riscv64, x86_64 and aarch64 again.
gcc/ChangeLog:
PR target/98743
* expr.c: Check mode before calling store_expr.
gcc/testsuite/ChangeLog:
PR target/98743
* g++.dg/opt/pr98743.C: New.
