The checks for PR 93244 don't actually pass on Windows (which is the
target where the bug is present) because of a different bug, PR 94063.
This adjusts the tests to not be affected by 94063 so that they verify
that 93244 was fixed.
PR libstdc++/93244
* testsuite/27_io/filesystem/path/generic/generic_string.cc: Adjust
test to not fail due to PR 94063.
* testsuite/27_io/filesystem/path/generic/utf.cc: Likewise.
* testsuite/27_io/filesystem/path/generic/wchar_t.cc: Likewise.
zTPF uses the same numeric value for ENOSYS and ENOTSUP.
libstdc++-v3/ChangeLog:
2020-03-06 Andreas Krebbel <krebbel@linux.ibm.com>
* src/c++11/system_error.cc: Omit the ENOTSUP case statement if it
would match ENOSYS.
2020-03-06 Delia Burduv <delia.burduv@arm.com>
* config/arm/arm_neon.h (bfloat16x4x2_t): New typedef.
(bfloat16x8x2_t): New typedef.
(bfloat16x4x3_t): New typedef.
(bfloat16x8x3_t): New typedef.
(bfloat16x4x4_t): New typedef.
(bfloat16x8x4_t): New typedef.
(vst2_bf16): New.
(vst2q_bf16): New.
(vst3_bf16): New.
(vst3q_bf16): New.
(vst4_bf16): New.
(vst4q_bf16): New.
* config/arm/arm-builtins.c (v2bf_UP): Define.
(VAR13): New.
(arm_init_simd_builtin_types): Init Bfloat16x2_t eltype.
* config/arm/arm-modes.def (V2BF): New mode.
* config/arm/arm-simd-builtin-types.def
(Bfloat16x2_t): New entry.
* config/arm/arm_neon_builtins.def
(vst2): Changed to VAR13 and added v4bf, v8bf
(vst3): Changed to VAR13 and added v4bf, v8bf
(vst4): Changed to VAR13 and added v4bf, v8bf
* config/arm/iterators.md (VDXBF): New iterator.
(VQ2BF): New iterator.
*config/arm/neon.md (neon_vst2<mode>): Used new iterators.
(neon_vst2<mode>): Used new iterators.
(neon_vst3<mode>): Used new iterators.
(neon_vst3<mode>): Used new iterators.
(neon_vst3qa<mode>): Used new iterators.
(neon_vst3qb<mode>): Used new iterators.
(neon_vst4<mode>): Used new iterators.
(neon_vst4<mode>): Used new iterators.
(neon_vst4qa<mode>): Used new iterators.
(neon_vst4qb<mode>): Used new iterators.
* gcc.target/arm/simd/bf16_vstn_1.c: New test.
This patch adds the Armv8.6-a ACLE intrinsics for bfcvtn, bfcvtn2 and
bfcvt as part of the BFloat16 extension.
(https://developer.arm.com/architectures/instruction-sets/simd-isas/neon/intrinsics)
The intrinsics are declared in arm_bf16.h and arm_neon.h and the RTL
patterns are defined in aarch64-simd.md.
2020-03-06 Delia Burduv <delia.burduv@arm.com>
gcc/
* config/aarch64/aarch64-simd-builtins.def
(bfcvtn): New built-in function.
(bfcvtn_q): New built-in function.
(bfcvtn2): New built-in function.
(bfcvt): New built-in function.
* config/aarch64/aarch64-simd.md
(aarch64_bfcvtn<q><mode>): New pattern.
(aarch64_bfcvtn2v8bf): New pattern.
(aarch64_bfcvtbf): New pattern.
* config/aarch64/arm_bf16.h (float32_t): New typedef.
(vcvth_bf16_f32): New intrinsic.
* config/aarch64/arm_bf16.h (vcvt_bf16_f32): New intrinsic.
(vcvtq_low_bf16_f32): New intrinsic.
(vcvtq_high_bf16_f32): New intrinsic.
* config/aarch64/iterators.md (V4SF_TO_BF): New mode iterator.
(UNSPEC_BFCVTN): New UNSPEC.
(UNSPEC_BFCVTN2): New UNSPEC.
(UNSPEC_BFCVT): New UNSPEC.
* config/arm/types.md (bf_cvt): New type.
gcc/testsuite/
* gcc.target/aarch64/advsimd-intrinsics/bfcvt-compile.c: New test.
* gcc.target/aarch64/advsimd-intrinsics/bfcvt-nobf16.c: New test.
* gcc.target/aarch64/advsimd-intrinsics/bfcvt-nosimd.c: New test.
* gcc.target/aarch64/advsimd-intrinsics/bfcvtnq2-untied.c: New test.
gcc/ChangeLog:
2020-03-06 Andreas Krebbel <krebbel@linux.ibm.com>
* config/s390/s390.md ("tabort"): Get rid of two consecutive
blanks in format string.
After add --param max-inline-insns-size=1 all target will remove the
redundant store at dse1, except some targets like AArch64 and MIPS will
expand the struct initialization into loop due to CLEAR_RATIO.
Tested on cross compiler of riscv32, riscv64, x86, x86_64, mips, mips64,
aarch64, nds32 and arm.
gcc/testsuite/ChangeLog
PR tree-optimization/90883
* g++.dg/tree-ssa/pr90883.c: Add --param max-inline-insns-size=1.
Add aarch64-*-* mips*-*-* to XFAIL.
On x86, when AVX and AVX512 are enabled, vector move instructions can
be encoded with either 2-byte/3-byte VEX (AVX) or 4-byte EVEX (AVX512):
0: c5 f9 6f d1 vmovdqa %xmm1,%xmm2
4: 62 f1 fd 08 6f d1 vmovdqa64 %xmm1,%xmm2
We prefer VEX encoding over EVEX since VEX is shorter. Also AVX512F
only supports 512-bit vector moves. AVX512F + AVX512VL supports 128-bit
and 256-bit vector moves. xmm16-xmm31 and ymm16-ymm31 are disallowed in
128-bit and 256-bit modes when AVX512VL is disabled. Mode attributes on
x86 vector move patterns indicate target preferences of vector move
encoding. For scalar register to register move, we can use 512-bit
vector move instructions to move 32-bit/64-bit scalar if AVX512VL isn't
available. With AVX512F and AVX512VL, we should use VEX encoding for
128-bit/256-bit vector moves if upper 16 vector registers aren't used.
This patch adds a function, ix86_output_ssemov, to generate vector moves:
1. If zmm registers are used, use EVEX encoding.
2. If xmm16-xmm31/ymm16-ymm31 registers aren't used, SSE or VEX encoding
will be generated.
3. If xmm16-xmm31/ymm16-ymm31 registers are used:
a. With AVX512VL, AVX512VL vector moves will be generated.
b. Without AVX512VL, xmm16-xmm31/ymm16-ymm31 register to register
move will be done with zmm register move.
There is no need to set mode attribute to XImode explicitly since
ix86_output_ssemov can properly encode xmm16-xmm31/ymm16-ymm31 registers
with and without AVX512VL.
Tested on AVX2 and AVX512 with and without --with-arch=native.
gcc/
PR target/89229
PR target/89346
* config/i386/i386-protos.h (ix86_output_ssemov): New prototype.
* config/i386/i386.c (ix86_get_ssemov): New function.
(ix86_output_ssemov): Likewise.
* config/i386/sse.md (VMOVE:mov<mode>_internal): Call
ix86_output_ssemov for TYPE_SSEMOV. Remove TARGET_AVX512VL
check.
(*movxi_internal_avx512f): Call ix86_output_ssemov for TYPE_SSEMOV.
(*movoi_internal_avx): Call ix86_output_ssemov for TYPE_SSEMOV.
Remove ext_sse_reg_operand and TARGET_AVX512VL check.
(*movti_internal): Likewise.
(*movtf_internal): Call ix86_output_ssemov for TYPE_SSEMOV.
gcc/testsuite/
PR target/89229
PR target/89346
* gcc.target/i386/avx512vl-vmovdqa64-1.c: Updated.
* gcc.target/i386/pr89229-2a.c: New test.
* gcc.target/i386/pr89229-2b.c: Likewise.
* gcc.target/i386/pr89229-2c.c: Likewise.
* gcc.target/i386/pr89229-3a.c: Likewise.
* gcc.target/i386/pr89229-3b.c: Likewise.
* gcc.target/i386/pr89229-3c.c: Likewise.
* gcc.target/i386/pr89346.c: Likewise.
Bug 93577, apparently a regression (although it isn't very clear to me
exactly when it was introduced; tests I made with various past
compilers produced inconclusive results, including e.g. ICEs appearing
with 64-bit-host compilers for some versions but not 32-bit-host
compilers for the same versions) is an C front-end tree-checking ICE
processing initializers for structs using the VLA-in-struct extension.
There is an error for such initializers, but other processing that
still takes place for them results in the ICE.
This patch ensures that processing of initializers for variable-size
types stops earlier to avoid the code that results in the ICE (and
ensures it stops earlier for error_mark_node to avoid ICEs in the
check for variable-size types), adjusts the conditions for the "empty
scalar initializer" diagnostic to avoid consequent excess errors in
the case of a bad type name, and adds tests for a few variations on
what such initializers might look like, as well as tests for cases
identified from ICEs seen with an earlier version of this patch.
Bootstrapped with no regressions for x86_64-pc-linux-gnu.
PR c/93577
gcc/c:
* c-typeck.c (pop_init_level): Do not diagnose initializers as
empty when initialized type is error_mark_node.
(set_designator, process_init_element): Ignore initializers for
elements of a variable-size type or of error_mark_node.
gcc/testsuite:
* gcc.dg/pr93577-1.c, gcc.dg/pr93577-2.c, gcc.dg/pr93577-3.c,
gcc.dg/pr93577-4.c, gcc.dg/pr93577-5.c, gcc.dg/pr93577-6.c: New
tests.
* gcc.dg/vla-init-1.c: Expect fewer errors about VLA initializer.
PR tree-optimization/91890
* gimple-ssa-warn-restrict.c (maybe_diag_overlap): Remove LOC argument.
Use gimple_or_expr_nonartificial_location.
(check_bounds_overlap): Drop LOC argument to maybe_diag_access_bounds.
Use gimple_or_expr_nonartificial_location.
* gimple.c (gimple_or_expr_nonartificial_location): New function.
* gimple.h (gimple_or_expr_nonartificial_location): Declare it.
* tree-ssa-strlen.c (maybe_warn_overflow): Use
gimple_or_expr_nonartificial_location.
(maybe_diag_stxncpy_trunc, handle_builtin_stxncpy_strncat): Likewise.
(maybe_warn_pointless_strcmp): Likewise.
* gcc.dg/pragma-diag-8.c: New test.
As the testcases show, the macros we have for -O0 for intrinsics that require
constant argument(s) should first cast the argument to the type the -O1+
inline uses and afterwards to whatever type e.g. a builtin needs.
The PR reported one which violated this, and I've grepped for all double-casts
and grepped out from that meaningful casts where the __m{128,256,512}{,d,i}
first cast is cast to same sized __v* type and has the same kind of element
type (float, double, integral). These 7 macros were using different casts,
and I've double checked them against the inline function types.
2020-03-05 Jakub Jelinek <jakub@redhat.com>
PR target/94046
* config/i386/avx2intrin.h (_mm_mask_i32gather_ps): Fix first cast of
SRC and MASK arguments to __m128 from __m128d.
(_mm256_mask_i32gather_ps): Fix first cast of MASK argument to __m256
from __m256d.
(_mm_mask_i64gather_ps): Fix first cast of MASK argument to __m128
from __m128d.
* config/i386/xopintrin.h (_mm_permute2_pd): Fix first cast of C
argument to __m128i from __m128d.
(_mm256_permute2_pd): Fix first cast of C argument to __m256i from
__m256d.
(_mm_permute2_ps): Fix first cast of C argument to __m128i from __m128.
(_mm256_permute2_ps): Fix first cast of C argument to __m256i from
__m256.
* g++.target/i386/pr94046-1.C: New test.
* g++.target/i386/pr94046-2.C: New test.
There's a -Wunused-but-set-variable warning in operations/all.cc which
can be fixed with [[maybe_unused]].
The statements in operations/copy.cc give -Wunused-value warnings. I
think I meant to use |= rather than !=.
And operations/file_size.cc gets -Wsign-compare warnings.
* testsuite/27_io/filesystem/operations/all.cc: Mark unused variable.
* testsuite/27_io/filesystem/operations/copy.cc: Fix typo.
* testsuite/experimental/filesystem/operations/copy.cc: Likewise.
* testsuite/27_io/filesystem/operations/file_size.cc: Use correct type
for return value, and in comparison.
* testsuite/experimental/filesystem/operations/file_size.cc: Likewise.
asan_test.cc tries to allocate 0xf0000000 bytes for 32bit targets in
a disabled DISABLED_DemoOOM test. Since the testcase is compiled with
-Werror, the compilation fails with:
error: argument 1 value '4026531840' exceeds maximum object size 2147483647
Compile with -Wno-alloc-size-larger-than to avoid compilation failure.
* g++.dg/asan/asan_test.C (dg-options): Add
-Wno-alloc-size-larger-than.
I don't think this is actually required to compile, because using
operator<< without a definition of the ostream doesn't seem valid to me.
But it's easy to make it work.
PR libstdc++/94051
* include/std/string_view: Include <bits/ostream_insert.h>.
* testsuite/21_strings/basic_string_view/inserters/94051.cc: New test.
A BOZ constant can not appear as a component inialiser for a derived
type.
gcc/fortran/ChangeLog:
PR93792
* decl.c (variable_decl): If param and initializer check
for BOZ, if found, output an error, set m to MATCH_ERROR
and goto cleanup.
gcc/testsuite/ChangeLog:
PR93792
* gfortran.dg/pr93792.f90: New test.
This patch adds the ARMv8.6 ACLE intrinsics for vmmla, vfmab and vfmat
as part of the BFloat16 extension.
(https://developer.arm.com/docs/101028/latest.)
The intrinsics are declared in arm_neon.h and the RTL patterns are
defined in neon.md.
Two new tests are added to check assembler output and lane indices.
2020-03-05 Delia Burduv <delia.burduv@arm.com>
* config/arm/arm_neon.h (vbfmmlaq_f32): New.
(vbfmlalbq_f32): New.
(vbfmlaltq_f32): New.
(vbfmlalbq_lane_f32): New.
(vbfmlaltq_lane_f32): New.
(vbfmlalbq_laneq_f32): New.
(vbfmlaltq_laneq_f32): New.
* config/arm/arm_neon_builtins.def (vmmla): New.
(vfmab): New.
(vfmat): New.
(vfmab_lane): New.
(vfmat_lane): New.
(vfmab_laneq): New.
(vfmat_laneq): New.
* config/arm/iterators.md (BF_MA): New int iterator.
(bt): New int attribute.
(VQXBF): Copy of VQX with V8BF.
* config/arm/neon.md (neon_vmmlav8bf): New insn.
(neon_vfma<bt>v8bf): New insn.
(neon_vfma<bt>_lanev8bf): New insn.
(neon_vfma<bt>_laneqv8bf): New expand.
(neon_vget_high<mode>): Changed iterator to VQXBF.
* config/arm/unspecs.md (UNSPEC_BFMMLA): New UNSPEC.
(UNSPEC_BFMAB): New UNSPEC.
(UNSPEC_BFMAT): New UNSPEC.
2020-03-05 Delia Burduv <delia.burduv@arm.com>
* gcc.target/arm/simd/bf16_ma_1.c: New test.
* gcc.target/arm/simd/bf16_ma_2.c: New test.
* gcc.target/arm/simd/bf16_mmla_1.c: New test.
The following testcase fails to assemble, as CONST_STRING in the DEBUG_INSNs
is printed as is, so if it contains \n and/or \r, we are in trouble:
.loc 1 14 3
# DEBUG haystack => [si]
# DEBUG needle => "
"
In the gimple dumps we print those (STRING_CSTs) as
# DEBUG haystack => D#1
# DEBUG needle => "\n"
so this patch uses what we use in tree printing for the CONST_STRINGs too.
2020-03-05 Jakub Jelinek <jakub@redhat.com>
PR middle-end/93399
* tree-pretty-print.h (pretty_print_string): Declare.
* tree-pretty-print.c (pretty_print_string): Remove forward
declaration, no longer static. Change nbytes parameter type
from unsigned to size_t.
* print-rtl.c (print_value) <case CONST_STRING>: Use
pretty_print_string and for shrink way too long strings.
* gcc.dg/pr93399.c: New test.
> > where POINTER_PLUS_EXPR last operand has sizetype type, thus unsigned,
> > and in the testcase gimple_assign_rhs2 (def) is thus 0xf000000000000001ULL
> > which multiplied by 8 doesn't fit into signed HWI. If it would be treated
> > as signed offset instead, it would fit (-0xfffffffffffffffLL, multiplied
> > by 8 is -0x7ffffffffffffff8LL). Unfortunately with the poly_int obfuscation
> > I'm not sure how to convert it from unsigned to signed poly_int.
>
> mem_ref_offset provides a boiler-plate for this:
>
> poly_offset_int::from (wi::to_poly_wide (TREE_OPERAND (t, 1)), SIGNED);
Thanks, that seems to work.
The test now works on both big-endian and little-endian.
2020-03-05 Richard Biener <rguenther@suse.de>
Jakub Jelinek <jakub@redhat.com>
PR tree-optimization/93582
* tree-ssa-sccvn.c (vn_reference_lookup_3): Treat POINTER_PLUS_EXPR
last operand as signed when looking for memset offset. Formatting
fix.
* gcc.dg/tree-ssa/pr93582-11.c: New test.
Co-authored-by: Richard Biener <rguenther@suse.de>
output_constructor doesn't like two consecutive entries with fields at the
same position; let's avoid adding the one for the empty field.
gcc/cp/ChangeLog
2020-03-04 Jason Merrill <jason@redhat.com>
PR c++/90432
* init.c (perform_member_init): Don't do aggregate initialization of
empty field.
* constexpr.c (cx_check_missing_mem_inits): Don't enforce
initialization of empty field.
gcc/testsuite/ChangeLog:
PR tree-optimization/93986
* gcc.dg/pr93986.c: New test.
gcc/ChangeLog:
PR tree-optimization/93986
* tree-ssa-strlen.c (maybe_warn_overflow): Convert all wide_int
operands to the same precision widest_int to avoid ICEs.
This patch converts region::validate to a vfunc, implementing
additional checking per subclass: verifying that various
region_id fields within map_region, array_region, stack_region and
root_region are valid, rather than just those within the base class.
Doing so caught bugs earlier in follow-up work I have on
canonicalization and purging of region_model.
gcc/analyzer/ChangeLog:
* region-model.cc (region::validate): Convert model param from ptr
to reference. Update comment to reflect that it's now a vfunc.
(map_region::validate): New vfunc implementation.
(array_region::validate): New vfunc implementation.
(stack_region::validate): New vfunc implementation.
(root_region::validate): New vfunc implementation.
(region_model::validate): Pass a reference rather than a pointer
to the region::validate vfunc.
* region-model.h (region::validate): Make virtual. Convert model
param from ptr to reference.
(map_region::validate): New vfunc decl.
(array_region::validate): New vfunc decl.
(stack_region::validate): New vfunc decl.
(root_region::validate): New vfunc decl.
The C++ reproducer for PR analyzer/94028 generates a similar ICE
to that of the Fortran reproducer for PR analyzer/93993 and, like
it, was fixed by r10-7023-g3d66e153b40ed000af30a9e569a05f34d5d576aa.
This patch adds the C++ reproducer as a regression test.
gcc/testsuite/ChangeLog:
PR analyzer/94028
* g++.dg/analyzer/pr94028.C: New test.
The XFAILs in this testcase (20050603-3.c) are no longer necessary
since the fix to PR68803 was committed with svn revision r242681.
2020-03-04 Will Schmidt <will_schmidt@vnet.ibm.com>
testsuite
* gcc.target/powerpc/20050603-3.c: Remove XFAILS.
The false warning:
pr93993.f90:19:0:
19 | allocate (tm) ! { dg-warning "dereference of possibly-NULL" }
|
Warning: dereference of possibly-NULL ‘_6’ [CWE-690] [-Wanalyzer-possible-null-dereference]
in the reproducer for PR analyzer/93993 is due to a BUILTIN_EXPECT in
the chain of SSA expressions between the malloc and the condition
guarding the edge: the analyzer didn't "know" about the relationship
between initial argument to BUILTIN_EXPECT and the return value.
This patch implements support for BUILTIN_EXPECT so that the return
value is known to be equal to the initial argument. This adds
constraints when exploring the CFG edges, eliminating the above
false positive.
Doing so also eliminated the leak warning from the reproducer. The
issue was that leaked_pvs was empty within
impl_region_model_context::on_state_leak, due to the leaking region
being a view, of type struct Pdtet_8 *, of a region of type
struct pdtet_8 *, which led region_model::get_representative_path_var to
return a NULL_TREE value.
Hence the patch also implements view support for
region_model::get_representative_path_var, restoring the leak
diagnostic, albeit changing the wording to:
Warning: leak of ‘(struct Pdtet_8) qb’ [CWE-401] [-Wanalyzer-malloc-leak]
It's not clear to me if we should emit leaks at a fortran "end program"
(currently we suppress them for leaks at the end of main).
gcc/analyzer/ChangeLog:
PR analyzer/93993
* region-model.cc (region_model::on_call_pre): Handle
BUILT_IN_EXPECT and its variants.
(region_model::add_any_constraints_from_ssa_def_stmt): Split out
gassign handling into add_any_constraints_from_gassign; add gcall
handling.
(region_model::add_any_constraints_from_gassign): New function,
based on the above. Add handling for NOP_EXPR.
(region_model::add_any_constraints_from_gcall): New function.
(region_model::get_representative_path_var): Handle views.
* region-model.h
(region_model::add_any_constraints_from_ssa_def_stmt): New decl.
(region_model::add_any_constraints_from_gassign): New decl.
gcc/testsuite/ChangeLog:
PR analyzer/93993
* gcc.dg/analyzer/expect-1.c: New test.
* gcc.dg/analyzer/malloc-4.c: New test.
* gfortran.dg/analyzer/pr93993.f90: Remove xfail from dg-bogus.
Move location of leak warning and update message.
PR analyzer/93993 reports another ICE within
diagnostic_manager::prune_for_sm_diagnostic in which the expression
of interest becomes a non-lvalue (similar to PR 93544, PR 93647, and
PR 93950), due to attempting to get an lvalue for a non-lvalue with a
NULL context, leading to an ICE when the failure is reported to
make_region_for_unexpected_tree_code. The tree in question is
an ADDR_EXPR of a VAR_DECL, due to:
event 11: switching var of interest from ‘tm’ in callee to ‘&qb’ in caller
This patch adds more bulletproofing to the routine by introducing
a tentative_region_model_context class that can be passed in such
circumstances which records that an error occurred, and then
checking to see if an error was recorded, thus avoiding the ICE.
This is papering over the problem, but a better solution seems more
like stage 1 material.
The patch also refactors the error-checking for CONSTANT_CLASS_P.
The testcase pr93993.f90 has a false positive:
pr93993.f90:19:0:
19 | allocate (tm) ! { dg-warning "dereference of possibly-NULL" }
|
Warning: dereference of possibly-NULL ‘_6’ [CWE-690] [-Wanalyzer-possible-null-dereference]
which appears to be a pre-existing bug affecting any allocate call in
Fortran, which I will fix in a followup.
gcc/analyzer/ChangeLog:
PR analyzer/93993
* checker-path.h (state_change_event::get_lvalue): Add ctxt param
and pass it to region_model::get_value call.
* diagnostic-manager.cc (get_any_origin): Pass a
tentative_region_model_context to the calls to get_lvalue and reject
the comparison if errors occur.
(can_be_expr_of_interest_p): New function.
(diagnostic_manager::prune_for_sm_diagnostic): Replace checks for
CONSTANT_CLASS_P with calls to update_for_unsuitable_sm_exprs.
Pass a tentative_region_model_context to the calls to
state_change_event::get_lvalue and reject the comparison if errors
occur.
(diagnostic_manager::update_for_unsuitable_sm_exprs): New.
* diagnostic-manager.h
(diagnostic_manager::update_for_unsuitable_sm_exprs): New decl.
* region-model.h (class tentative_region_model_context): New class.
gcc/testsuite/ChangeLog:
PR analyzer/93993
* gfortran.dg/analyzer/pr93993.f90: New test.
The discussion of iterator_traits<volatile T*>::value_type and the
example with three tempalte arguments related to an earlier version of
the patch, not the one committed.
Also improve the comment on __memcmpable.
* include/bits/cpp_type_traits.h (__memcpyable): Fix comment.
PR87560 reports an ICE when a test case is compiled with -mpower9-vector
and -mno-altivec. This patch terminates compilation with an error when
this combination (and other unreasonable ones) are requested.
Bootstrapped and tested on powerpc64le-unknown-linux-gnu with no
regressions. Reported error is now:
f951: Error: '-mno-altivec' turns off '-mpower9-vector'
2020-03-02 Bill Schmidt <wschmidt@linux.ibm.com>
PR target/87560
* rs6000-cpus.def (OTHER_ALTIVEC_MASKS): New #define.
* rs6000.c (rs6000_disable_incompatible_switches): Add table entry
for OPTION_MASK_ALTIVEC.
Building a zTPF cross currently fails when building libstdc++
complaining about the __UINTPTR_TYPE__ to be missing.
Fixed by including the glibc-stdint.h header.
2020-03-04 Andreas Krebbel <krebbel@linux.ibm.com>
* config.gcc: Include the glibc-stdint.h header for zTPF.
For the zTPF we must not use floating point registers.
gcc/ChangeLog:
2020-03-04 Andreas Krebbel <krebbel@linux.ibm.com>
* config/s390/s390.c (s390_secondary_memory_needed): Disallow
direct FPR-GPR copies.
(s390_register_info_gprtofpr): Disallow GPR content to be saved in
FPRs.
libgcc is supposed to be built with the trace skip flags and branch
targets. Add a zTPF header file fragment and add the -mtpf-trace-skip
option.
libgcc/ChangeLog:
2020-03-04 Andreas Krebbel <krebbel@linux.ibm.com>
* config.host: Include the new makefile fragment.
* config/s390/t-tpf: New file.
The zTPF OS implements a tracing facility for function entry and exit
which uses global flags and trace function addresses. The addresses of
the flags as well as the trace functions are currently hard-coded in
the zTPF specific GCC parts of the IBM Z back-end.
With this patch these addresses can be changed at compile-time using
the new command line options. For convenience one additional command
line option (-mtpf-trace-skip) implements a new set of hard-coded
addresses.
gcc/ChangeLog:
2020-03-04 Andreas Krebbel <krebbel@linux.ibm.com>
* config/s390/s390.c (s390_emit_prologue): Specify the 2 new
operands to the prologue_tpf expander.
(s390_emit_epilogue): Likewise.
(s390_option_override_internal): Do error checking and setup for
the new options.
* config/s390/tpf.h (TPF_TRACE_PROLOGUE_CHECK)
(TPF_TRACE_EPILOGUE_CHECK, TPF_TRACE_PROLOGUE_TARGET)
(TPF_TRACE_EPILOGUE_TARGET, TPF_TRACE_PROLOGUE_SKIP_TARGET)
(TPF_TRACE_EPILOGUE_SKIP_TARGET): New macro definitions.
* config/s390/tpf.md ("prologue_tpf", "epilogue_tpf"): Add two new
operands for the check flag and the branch target.
* config/s390/tpf.opt ("mtpf-trace-hook-prologue-check")
("mtpf-trace-hook-prologue-target")
("mtpf-trace-hook-epilogue-check")
("mtpf-trace-hook-epilogue-target", "mtpf-trace-skip"): New
options.
* doc/invoke.texi: Document -mtpf-trace-skip option. The other
options are for debugging purposes and will not be documented
here.
* gcc.target/i386/pr91623.c: Add -fcommon in order
to re-trigger the needed code for the test-case which
was added in r10-2910-g9151048d854e352a9b83b771c6711b8221c73f7e.
In the following testcase we emit wrong debug info for the karg
parameter in the DW_TAG_inlined_subroutine into main.
The problem is that the karg PARM_DECL is DECL_BY_REFERENCE and thus
in the IL has const K & type, but in the source just const K.
When the function is inlined, we create a VAR_DECL for it, but don't
set DECL_BY_REFERENCE, so when emitting DW_AT_location, we treat it like
a const K & typed variable, but it has DW_AT_abstract_origin which has
just the const K type and thus the debugger thinks the variable has
const K type.
Fixed by copying the DECL_BY_REFERENCE flag. Not doing it in
copy_decl_for_dup_finish, because copy_decl_no_change already copies
that flag through copy_node and in copy_result_decl_to_var it is
undesirable, as we handle DECL_BY_REFERENCE in that case instead
by changing the type.
2020-03-04 Jakub Jelinek <jakub@redhat.com>
PR debug/93888
* tree-inline.c (copy_decl_to_var): Copy DECL_BY_REFERENCE flag.
* g++.dg/guality/pr93888.C: New test.
The following patch attempts to avoid dangerous overflows in the various
push_partial_def HOST_WIDE_INT computations.
This is achieved by performing the subtraction offset2i - offseti in
the push_partial_def function and before doing that doing some tweaks.
If a constant store (non-CONSTRUCTOR) is too large (perhaps just
hypothetical case), native_encode_expr would fail for it, but we don't
necessarily need to fail right away, instead we can treat it like
non-constant store and if it is already shadowed, we can ignore it.
Otherwise, if it at most 64-byte and the caller ensured that there is
a range overlap and push_partial_def ensures the load is at most 64-byte,
I think we should be fine, offset (relative to the load)
can be from -64*8+1 to 64*8-1 only and size at most 64*8, so no risks of
overflowing HOST_WIDE_INT computations.
For CONSTRUCTOR (or non-constant) stores, those can be indeed arbitrarily
large, the caller just checks that both the absolute offset and size fit
into signed HWI. But, we store the same bytes in that case over and over
(both in the {} case where it is all 0, and in the hypothetical future case
where we handle in push_partial_def also memset (, 123, )), so we can tweak
the write range for our purposes. For {} store we could just cap it at the
start offset and/or offset+size because all the bits are 0, but I wrote it
in anticipation of the memset case and so the relative offset can now be
down to -7 and similarly size can grow up to 64 bytes + 14 bits, all this
trying to preserve the offset difference % BITS_PER_UNIT or end as well.
2020-03-04 Jakub Jelinek <jakub@redhat.com>
* tree-ssa-sccvn.c (vn_walk_cb_data::push_partial_def): Add offseti
argument. Change pd argument so that it can be modified. Turn
constant non-CONSTRUCTOR store into non-constant if it is too large.
Adjust offset and size of CONSTRUCTOR or non-constant store to avoid
overflows.
(vn_walk_cb_data::vn_walk_cb_data, vn_reference_lookup_3): Adjust
callers.