Stephan Bergmann reported that our -Wbidi-chars breaks the build
of LibreOffice because we warn about UCNs even when their usage
is correct: LibreOffice constructs strings piecewise, as in:
aText = u"\u202D" + aText;
and warning about that is overzealous. Since no editor (AFAIK)
interprets UCNs to show them as Unicode characters, there's less
risk in misinterpreting them, and so perhaps we shouldn't warn
about them by default. However, identifiers containing UCNs or
programs generating other programs could still cause confusion,
so I'm keeping the UCN checking. To turn it on, you just need
to use -Wbidi-chars=unpaired,ucn or -Wbidi-chars=any,ucn.
The implementation is done by using the new EnumSet feature.
PR preprocessor/104030
gcc/c-family/ChangeLog:
* c.opt (Wbidi-chars): Mark as EnumSet. Also accept =ucn.
gcc/ChangeLog:
* doc/invoke.texi: Update documentation for -Wbidi-chars.
libcpp/ChangeLog:
* include/cpplib.h (enum cpp_bidirectional_level): Add
bidirectional_ucn. Set values explicitly.
* internal.h (cpp_reader): Adjust warn_bidi_p.
* lex.cc (maybe_warn_bidi_on_close): Don't warn about UCNs
unless UCN checking is on.
(maybe_warn_bidi_on_char): Likewise.
gcc/testsuite/ChangeLog:
* c-c++-common/Wbidi-chars-10.c: Turn on UCN checking.
* c-c++-common/Wbidi-chars-11.c: Likewise.
* c-c++-common/Wbidi-chars-14.c: Likewise.
* c-c++-common/Wbidi-chars-16.c: Likewise.
* c-c++-common/Wbidi-chars-17.c: Likewise.
* c-c++-common/Wbidi-chars-4.c: Likewise.
* c-c++-common/Wbidi-chars-5.c: Likewise.
* c-c++-common/Wbidi-chars-6.c: Likewise.
* c-c++-common/Wbidi-chars-7.c: Likewise.
* c-c++-common/Wbidi-chars-8.c: Likewise.
* c-c++-common/Wbidi-chars-9.c: Likewise.
* c-c++-common/Wbidi-chars-ranges.c: Likewise.
* c-c++-common/Wbidi-chars-18.c: New test.
* c-c++-common/Wbidi-chars-19.c: New test.
* c-c++-common/Wbidi-chars-20.c: New test.
* c-c++-common/Wbidi-chars-21.c: New test.
* c-c++-common/Wbidi-chars-22.c: New test.
* c-c++-common/Wbidi-chars-23.c: New test.
For targets with IEEE support but without the issignaling macro in libc
(currently, everywhere except glibc), this allows us to provide a fallback
implementation. In order to keep the code in ieee_helper.c relatively
readable, I've put that new implementation in a separate file,
issignaling_fallback.h.
libgfortran/ChangeLog:
* ieee/issignaling_fallback.h: New file.
* ieee/ieee_helper.c: Include issignaling_fallback.h when target
does not define issignaling macro.
gcc/testsuite/ChangeLog:
* gfortran.dg/ieee/signaling_1.f90: Do not require issignaling.
* gfortran.dg/ieee/signaling_2.f90: Add comment.
* gfortran.dg/ieee/signaling_3.f90: New test.
This optimizations were originally in glibc, but was removed
and suggested that they were a good fit as gcc builtins[1].
feclearexcept and feraiseexcept were extended (in comparison to the
glibc version) to accept any combination of the accepted flags, not
limited to just one flag bit at a time anymore.
The builtin expanders needs knowledge of the target libc's FE_*
values, so they are limited to expand only to suitable libcs.
[1] https://sourceware.org/legacy-ml/libc-alpha/2020-03/msg00047.htmlhttps://sourceware.org/legacy-ml/libc-alpha/2020-03/msg00080.html
2020-08-13 Raoni Fassina Firmino <raoni@linux.ibm.com>
gcc/
PR target/94193
* builtins.cc (expand_builtin_fegetround): New function.
(expand_builtin_feclear_feraise_except): New function.
(expand_builtin): Add cases for BUILT_IN_FEGETROUND,
BUILT_IN_FECLEAREXCEPT and BUILT_IN_FERAISEEXCEPT.
* config/rs6000/rs6000.md (fegetroundsi): New pattern.
(feclearexceptsi): New Pattern.
(feraiseexceptsi): New Pattern.
* doc/extend.texi: Add a new introductory paragraph about the
new builtins.
* doc/md.texi: (fegetround@var{m}): Document new optab.
(feclearexcept@var{m}): Document new optab.
(feraiseexcept@var{m}): Document new optab.
* optabs.def (fegetround_optab): New optab.
(feclearexcept_optab): New optab.
(feraiseexcept_optab): New optab.
gcc/testsuite/
PR target/94193
* gcc.target/powerpc/builtin-feclearexcept-feraiseexcept-1.c: New test.
* gcc.target/powerpc/builtin-feclearexcept-feraiseexcept-2.c: New test.
* gcc.target/powerpc/builtin-fegetround.c: New test.
Signed-off-by: Raoni Fassina Firmino <raoni@linux.ibm.com>
Avoid using the macro current_template_depth, which expands to an
expression that uses __FUNCTION__, within the signature of a function.
PR c++/104197
gcc/cp/ChangeLog:
* pt.cc (make_auto_1): Use -1 as a placeholder default argument
for level.
Here the call to (the &&-qualified) toLower() is incorrectly rejected
during overload resolution because the object expression is encoded as
an lvalue when it's really a prvalue. The object expression,
instance()->applicationName(), is encoded as an INDIRECT_REF of a
COMPOUND_EXPR
*(*instance ();, &TARGET_EXPR <D.2383, QCoreApplication::applicationName ()>;);
which lvalue_kind deems an lvalue.
This issue is similar to PR88103 except that here the original compound
object expression is a prvalue rather than an xvalue. The fix there was to
adjust the result of unary_complex_lvalue in build_class_member_access_expr
so that xvalueness of the original expression is preserved. This patch
extends that fix so that rvalueness is preserved more generally.
PR c++/104173
gcc/cp/ChangeLog:
* typeck.cc (build_class_member_access_expr): Extend
unary_complex_lvalue result adjustment to preserve all
rvalues, not just xvalues.
gcc/testsuite/ChangeLog:
* g++.dg/cpp0x/ref-qual21.C: New test.
contrib/ChangeLog:
* gcc-changelog/git_commit.py: New files in toplev must
be explicitly marked as "New file".
* gcc-changelog/test_email.py: Test.
* gcc-changelog/test_patches.txt: Add test.
On Sat, Jan 22, 2022 at 01:47:08AM +0100, Jakub Jelinek via Gcc-patches wrote:
> I think with the 2) patch I achieve what we want for Fortran, for 1)
> the only behavior from gcc 11 is that
> -fsanitize-coverage=trace-cmp,trace-cmp is now rejected.
> This is mainly from the desire to disallow
> -fconvert=big-endian,little-endian or -Wbidi-chars=bidirectional,any
> etc. where it would be confusing to users what exactly it means.
> But it is the only from these options that actually acts as an Enum
> bit set, each enumerator can be specified with all the others.
> So one option would be stop requiring the EnumSet implies Set properties
> must be specified and just require that either they are specified on all
> EnumValues, or on none of them; the latter case would be for
> -fsanitize-coverage= and the non-Set case would mean that all the
> EnumValues need to have disjoint Value bitmasks and that they can
> be all specified and unlike the Set case also repeated.
> Thoughts on this?
Here is an incremental patch to the first two patches of the series
that implements EnumBitSet that fully restores the -fsanitize-coverage
GCC 11 behavior.
2022-01-24 Jakub Jelinek <jakub@redhat.com>
PR sanitizer/104158
* opt-functions.awk (var_set): Handle EnumBitSet property.
* optc-gen.awk: Don't disallow RejectNegative if EnumBitSet is
specified.
* opts.h (enum cl_enum_var_value): New type.
* opts-common.cc (decode_cmdline_option): Use CLEV_* values.
Handle CLEV_BITSET.
(cmdline_handle_error): Handle CLEV_BITSET.
* opts.cc (test_enum_sets): Also test EnumBitSet requirements.
* doc/options.texi (EnumBitSet): Document.
* common.opt (fsanitize-coverage=): Use EnumBitSet instead of
EnumSet.
(trace-pc, trace-cmp): Drop Set properties.
* gcc.dg/sancov/pr104158-7.c: Adjust for repeating of arguments
being allowed.
This patch on top of the previously posted option handling changes patch
allows specifying -fconvert=swap,r16_ieee etc. (but will error on it
when not on powerpc64le because in the library such swapping is only
implemented for HAVE_REAL_17).
2022-01-24 Jakub Jelinek <jakub@redhat.com>
* lang.opt (fconvert=): Add EnumSet property and mention also
r16_ieee and r16_ibm arguments.
(big-endian, little-endian, native, swap): Add Set(1) property.
(r16_ieee, r16_ibm): New EnumValue entries with Set(2) property.
* trans-types.cc (gfc_init_kinds): Emit gfc_fatal_error for
-fconvert=r16_ieee or -fconvert=r16_ibm when R16_IEEE <=> R16_IBM
conversions aren't supported.
This is incremental patch to fix up -fsanitize-coverage= option
handling, allow -fno-sanitize-coverage= again, allow both
options together in one option or make
-fsanitize-coverage=trace-pc -fsanitize-coverage=trace-cmp
actually enable both suboptions rather than the last one.
2022-01-24 Jakub Jelinek <jakub@redhat.com>
PR sanitizer/104158
* common.opt (flag_sanitize_coverage): Remove Variable entry.
(fsanitize-coverage=): Remove RejectNegative property, add
Var(flag_sanitize_coverage) and EnumSet properties.
(trace-pc): Add Set(1) property.
(trace-cmp): Add Set(2) property.
* opts.cc (common_handle_option): Don't handle
OPT_fsanitize_coverage_.
* gcc.dg/spellcheck-options-24.c: New test.
* gcc.dg/sancov/pr104158-1.c: New test.
* gcc.dg/sancov/pr104158-2.c: New test.
* gcc.dg/sancov/pr104158-3.c: New test.
* gcc.dg/sancov/pr104158-4.c: New test.
* gcc.dg/sancov/pr104158-5.c: New test.
* gcc.dg/sancov/pr104158-6.c: New test.
* gcc.dg/sancov/pr104158-7.c: New test.
The following patch is infrastructure support for at least 3 different
options that need changes:
1) PR104158 talks about a regression with the -fsanitizer-coverage=
option; in GCC 11 and older and on trunk prior to r12-1177, this
option behaved similarly to -f{,no-}sanitizer{,-recover}= options,
namely that the option allows negative and argument of the option
is a list of strings, each of them has some enumerator and
-fsanitize-coverage= enabled those bits in the underlying
flag_sanitize_coverage, while -fno-sanitize-coverage= disabled them.
So, -fsanitize-coverage=trace-pc,trace-cmp was equivalent to
-fsanitize-coverage=trace-pc -fsanitize-coverage=trace-cmp and both
set flag_sanitize_coverage to
(SANITIZE_COV_TRACE_PC | SANITIZE_COV_TRACE_CMP)
Also, e.g.
-fsanitize-coverage=trace-pc,trace-cmp -fno-sanitize-coverage=trace-pc
would in the end set flag_sanitize_coverage to
SANITIZE_COV_TRACE_CMP (first set both bits, then subtract one)
The r12-1177 change, I think done to improve argument misspelling
diagnostic, changed the option incompatibly in multiple ways,
-fno-sanitize-coverage= is now rejected, only a single argument
is allowed, not multiple and
-fsanitize-coverage=trace-pc -fsanitize-coverage=trace-cmp
enables just SANITIZE_COV_TRACE_CMP and not both (each option
overrides the previous value)
2) Thomas Koenig wants to extend Fortran -fconvert= option for the
ppc64le real(kind=16) swapping support; currently the option
accepts -fconvert={native,swap,big-endian,little-endian} and the
intent is to add support for -fconvert=r16_ibm and -fconvert=r16_ieee
(that alone is just normal Enum), but also to handle
-fconvert=swap,r16_ieee or -fconvert=r16_ieee,big-endian but not
-fconvert=big-endian,little-endian - the
native/swap/big-endian/little-endian are one mutually exclusive set
and r16_ieee/r16_ibm another one.
See https://gcc.gnu.org/pipermail/gcc-patches/2022-January/587943.html
and thread around that.
3) Similarly Marek Polacek wants to extend the -Wbidi-chars= option,
such that it will handle not just the current
-Wbidi-chars={none,bidirectional,any}, but also -Wbidi-chars=ucn
and bidirectional,ucn and ucn,any etc. Again two separate sets,
one none/bidirectional/any and another one ucn.
See https://gcc.gnu.org/pipermail/gcc-patches/2022-January/588960.html
The following patch adds framework for this and I'll post incremental
patches for 1) and 2).
As I've tried to document, such options are marked by additional
EnumSet property on the option and in that case all the EnumValues
in the Enum referenced from it must use a new Set property with set
number (initially I wanted just mark last enumerator in each mutually
exclusive set, but optionlist is sorted and so it doesn't really work
well). So e.g. for the Fortran -fconvert=, one specifies:
fconvert=
Fortran RejectNegative Joined Enum(gfc_convert) EnumSet Var(flag_convert) Init(GFC_FLAG_CONVERT_NATIVE)
-fconvert=<big-endian|little-endian|native|swap|r16_ieee|r16_ibm> The endianness used for unformatted files.
Enum
Name(gfc_convert) Type(enum gfc_convert) UnknownError(Unrecognized option to endianness value: %qs)
EnumValue
Enum(gfc_convert) String(big-endian) Value(GFC_FLAG_CONVERT_BIG) Set(1)
EnumValue
Enum(gfc_convert) String(little-endian) Value(GFC_FLAG_CONVERT_LITTLE) Set(1)
EnumValue
Enum(gfc_convert) String(native) Value(GFC_FLAG_CONVERT_NATIVE) Set(1)
EnumValue
Enum(gfc_convert) String(swap) Value(GFC_FLAG_CONVERT_SWAP) Set(1)
EnumValue
Enum(gfc_convert) String(r16_ieee) Value(GFC_FLAG_CONVERT_R16_IEEE) Set(2)
EnumValue
Enum(gfc_convert) String(r16_ibm) Value(GFC_FLAG_CONVERT_R16_IBM) Set(2)
and this says to the option handling code that
1) if only one arg is specified to one instance of the option, it can be any
of those 6
2) if two args are specified, one has to be from the first 4 and another
from the last 2, in any order
3) at most 2 args may be specified (there are just 2 sets)
There is a requirement on the Value values checked in self-test, the
values from one set ored together must be disjunct from values from
another set ored together. In the Fortran case, the first 4 are 0-3
so mask is 3, and the last 2 are 4 and 8, so mask is 12.
When say -fconvert=big-endian is specified, it sets the first set
to GFC_FLAG_CONVERT_BIG (2) but doesn't modify whatever value the
other set had, so e.g.
-fconvert=big-endian -fconvert=r16_ieee
-fconvert=r16_ieee -fconvert=big-endian
-fconvert=r16_ieee,big_endian
-fconvert=big_endian,r16_ieee
all behave the same.
Also, with the EnumSet support, it is now possible to allow
not specifying RejectNegative - we can set some set's value and
then clear it and set it again to some other value etc.
I think with the 2) patch I achieve what we want for Fortran, for 1)
the only behavior from gcc 11 is that
-fsanitize-coverage=trace-cmp,trace-cmp is now rejected.
This is mainly from the desire to disallow
-fconvert=big-endian,little-endian or -Wbidi-chars=bidirectional,any
etc. where it would be confusing to users what exactly it means.
But it is the only from these options that actually acts as an Enum
bit set, each enumerator can be specified with all the others.
So one option would be stop requiring the EnumSet implies Set properties
must be specified and just require that either they are specified on all
EnumValues, or on none of them; the latter case would be for
-fsanitize-coverage= and the non-Set case would mean that all the
EnumValues need to have disjoint Value bitmasks and that they can
be all specified and unlike the Set case also repeated.
Thoughts on this?
2022-01-24 Jakub Jelinek <jakub@redhat.com>
PR sanitizer/104158
* opt-functions.awk (var_set): Handle EnumSet property.
* optc-gen.awk: Don't disallow RejectNegative if EnumSet is
specified.
* opt-read.awk: Handle Set property.
* opts.h (CL_ENUM_SET_SHIFT, CL_ERR_ENUM_SET_ARG): Define.
(struct cl_decoded_option): Mention enum in value description.
Add mask member.
(set_option): Add mask argument defaulted to 0.
* opts.cc (test_enum_sets): New function.
(opts_cc_tests): Call it.
* opts-common.cc (enum_arg_to_value): Change return argument
from bool to int, on success return index into the cl_enum_arg
array, on failure -1. Add len argument, if non-0, use strncmp
instead of strcmp.
(opt_enum_arg_to_value): Adjust caller.
(decode_cmdline_option): Handle EnumSet represented as
CLVC_ENUM with non-zero var_value. Initialize decoded->mask.
(decode_cmdline_options_to_array): CLear opt_array[0].mask.
(handle_option): Pass decoded->mask to set_options last argument.
(generate_option): Clear decoded->mask.
(generate_option_input_file): Likewise.
(cmdline_handle_error): Handle CL_ERR_ENUM_SET_ARG.
(set_option): Add mask argument, use it for CLVC_ENUM.
(control_warning_option): Adjust enum_arg_to_value caller.
* doc/options.texi: Document Set and EnumSet properties.
On Sat, Jan 22, 2022 at 10:32:21AM +0100, Martin Liška wrote:
> I've just noticed the patch broke a few cross compilers:
>
> s390x-ibm-tpf:
>
> /home/marxin/buildworker/zen2-cross-compilers/build/gcc/common/config/s390/s390-common.cc: In function ‘bool s390_supports_split_stack(bool, gcc_options*)’:
> /home/marxin/buildworker/zen2-cross-compilers/build/gcc/common/config/s390/s390-common.cc:126:13: error: ‘struct gcc_options’ has no member named ‘x_linux_libc’
> 126 | if (opts->x_linux_libc == LIBC_GLIBC)
> | ^~~~~~~~~~~~
>
> i686-kopensolaris-gnu, i686-symbolics-gnu
>
> /home/marxin/buildworker/zen2-cross-compilers/build/gcc/common/config/i386/i386-common.cc: In function ‘bool ix86_supports_split_stack(bool, gcc_options*)’:
> /home/marxin/buildworker/zen2-cross-compilers/build/gcc/common/config/i386/i386-common.cc:1721:13: error: ‘struct gcc_options’ has no member named ‘x_linux_libc’
> 1721 | if (opts->x_linux_libc != LIBC_GLIBC)
> | ^~~~~~~~~~~~
> make[1]: *** [Makefile:2418: i386-common.o] Error 1
>
> Can you please take a look? Btw. do you have a bugzilla account?
I bet instead of opts->x_linux_libc != LIBC_GLIBC it needs to use
#ifdef OPTION_GLIBC
if (!OPTION_GLIBC)
#endif
or so. I think the first committed patch actually used that
but used it in #if directive, which is wrong because it is something
that needs to be evaluated at runtime.
That doesn't work well either, because the *supports_split_stack
hooks have opts argument and OPTION_GLIBC doesn't take that.
So, here is a patch that introduces OPTION_*_P macros that take opts
as an argument and redefines OPTION_* using those (similarly to how
the option scripts create TARGET_*_P and TARGET_* macros).
2022-01-24 Jakub Jelinek <jakub@redhat.com>
PR bootstrap/104170
* config/linux.h (OPTION_GLIBC_P, OPTION_UCLIBC_P,
OPTION_BIONIC_P, OPTION_MUSL_P): Define.
(OPTION_GLIBC, OPTION_UCLIBC, OPTION_BIONIC, OPTION_MUSL): Redefine
using OPTION_*_P macros.
* config/alpha/linux.h (OPTION_GLIBC_P, OPTION_UCLIBC_P,
OPTION_BIONIC_P, OPTION_MUSL_P): Define.
(OPTION_GLIBC, OPTION_UCLIBC, OPTION_BIONIC, OPTION_MUSL): Redefine
using OPTION_*_P macros.
* config/rs6000/linux.h (OPTION_GLIBC_P, OPTION_UCLIBC_P,
OPTION_BIONIC_P, OPTION_MUSL_P): Define.
(OPTION_GLIBC, OPTION_UCLIBC, OPTION_BIONIC, OPTION_MUSL): Redefine
using OPTION_*_P macros.
* config/rs6000/linux64.h (OPTION_GLIBC_P, OPTION_UCLIBC_P,
OPTION_BIONIC_P, OPTION_MUSL_P): Define.
(OPTION_GLIBC, OPTION_UCLIBC, OPTION_BIONIC, OPTION_MUSL): Redefine
using OPTION_*_P macros.
* config/fuchsia.h (OPTION_MUSL_P): Redefine.
* config/glibc-stdint.h (OPTION_MUSL_P): Define if not defined.
* common/config/s390/s390-common.cc (s390_supports_split_stack): Re-add
ATTRIBUTE_UNUSED to opts parameter. If OPTION_GLIBC_P is defined, use
OPTION_GLIBC_P (opts) as condition, otherwise assume if (false).
* common/config/i386/i386-common.cc (ix86_supports_split_stack): If
OPTION_GLIBC_P is defined use !OPTION_GLIBC_P (opts) as condition,
otherwise assume if (true).
Extension version might be different among different ISA spec version,
add explicitly isa-spec version to prevent that might fail when build
GCC with different default ISA version.
gcc/testsuite/ChangeLog
* gcc.target/riscv/attribute-19.c: Add -misa-spec=2.2
I-ext 2.0 already included zicsr and zifencei, skip that prevent
confusing binutils.
gcc/ChangeLog
* common/config/riscv/riscv-common.cc (riscv_subset_list::to_string):
Skip zicsr and zifencei if I-ext is 2.0.
Bump default ISA spec to newer version 20191213, current default ISA spec
is 2.2, but it's already out of date for a long time, sync with binutils
ISA version, convention in toolchain use.
gcc/ChangeLog:
* config.gcc: Modify default isa_spec version.
This patch correct the type of niter->control.base, when it is updated
as a PLUS expr.
During build PLUS expr, the result type should align with the type of
the operands.
PR tree-optimization/102087
gcc/ChangeLog:
* tree-ssa-loop-niter.cc (number_of_iterations_until_wrap):
Correct PLUS result type.
gcc/testsuite/ChangeLog:
* gcc.dg/pr102087_1.c: New test.
After commit 591b6e00d1 " riscv: fix -Wformat-diag errors ", some strings in implement was changed.
This patch update the check info in testcases to sync with it.
gcc/testsuite/ChangeLog:
* gcc.target/riscv/arch-9.c: Update the check info.
* gcc.target/riscv/arch-10.c: Ditto.
* gcc.target/riscv/arch-12.c: Ditto.
libstdc++-v3/ChangeLog:
PR libstdc++/104032
* include/std/spanstream (basic_spanbuf(basic_spanbuf&&)): Use
mem-initializer for _M_buf.
(basic_spanbuf::Operator=(basic_spanbuf&&)): Fix ill-formed
member access.
* testsuite/27_io/spanstream/2.cc: New test.
We can use the new from_chars implementation when long double and double
have the same representation.
libstdc++-v3/ChangeLog:
* src/c++17/floating_from_chars.cc (USE_STRTOD_FOR_FROM_CHARS):
Define macro for case where std::from_chars is implemented in
terms of strtod, strtof or strtold.
(buffer_resource, valid_fmt, find_end_of_float, pattern)
(from_chars_impl, make_result, reserve_string): Do not define
unless USE_STRTOD_FOR_FROM_CHARS is defined.
(from_chars): Define when at least one of USE_LIB_FAST_FLOAT and
USE_STRTOD_FOR_FROM_CHARS is defined, instead of
_GLIBCXX_HAVE_USELOCALE. Use fast_float for long double when it
is binary64.
I broke this unintentionally in r12-4259.
libstdc++-v3/ChangeLog:
PR libstdc++/104174
* include/bits/hashtable_policy.h (_Map_base): Add partial
specialization for maps with const key types.
* testsuite/23_containers/unordered_map/104174.cc: New test.
The non-atomic store that sets both reference counts to zero uses a
type-punned pointer, which has undefined behaviour. We could use memset
to write 8 bytes, but we don't actually need it to be a single store
anyway. No other thread can observe the values, that's why it's safe to
use non-atomic stores in the first place. So we can just set each count
to zero.
With -fstore-merging (which is enabled by default at -O2) GCC produces
the same code for this as for memset or the type punned store. Clang
does that store merging even at -O1.
libstdc++-v3/ChangeLog:
PR libstdc++/104019
* include/bits/shared_ptr_base.h (_Sp_counted_base<>::_M_release):
Set members to zero without type punning.
There are two underlying bugs in the designated initialization of char array
fields by string literals that cause:
(1) Rejection of valid cases with:
(a) brace-enclosed string literal initializer (of any valid size), or
(b) unbraced string literal shorter than the target char array field.
(2) Acceptance of invalid cases with designators appearing within the braces
of a braced string literal, in which case the bogus 'designator' was
being entirely ignored and the string literal treated as a positional
initializer.
The fixes above allow to address a FIXME in cp_complete_array_type:
/* FIXME: this code is duplicated from reshape_init.
Probably we should just call reshape_init here? */
I believe that this was obstructed by the designator bugs (see comment here
https://patchwork.ozlabs.org/project/gcc/list/?series=199783)
PR c++/55227
gcc/cp/ChangeLog:
* decl.cc (reshape_init_r): Only call has_designator_check when
first_initializer_p or for the inner constructor element.
(cp_complete_array_type): Call reshape_init on braced-init-list.
gcc/testsuite/ChangeLog:
* g++.dg/cpp2a/desig21.C: New test.
This is a simple patch which simplifies the __builtin_aarch64_sqrt* builtins
into the internal function SQRT which allows for constant folding and other
optimizations at the gimple level. It was originally suggested we do to
__builtin_sqrt just for __builtin_aarch64_sqrtdf when -fno-math-errno
but since r6-4969-g686ee9719a4 we have the internal function SQRT which does
the same so it makes we don't need to check -fno-math-errno either now.
Applied as approved after bootstrapped and tested on aarch64-linux-gnu with no regressions.
PR target/64821
gcc/ChangeLog:
* config/aarch64/aarch64-builtins.cc
(aarch64_general_gimple_fold_builtin): Handle
__builtin_aarch64_sqrt* and simplify into SQRT internal
function.
gcc/testsuite/ChangeLog:
* gcc.target/aarch64/vsqrt-1.c: New test.
* gcc.target/aarch64/vsqrt-2.c: New test.
This is the same issue as PR104031, but that patch doesn't fix this
testcase because in this case, current_function_decl isn't set when we get
to cp_genericize_target_expr. But there seems to be no need for
is_local_temp to check for function scope; !TREE_STATIC should be enough.
PR c++/104182
gcc/cp/ChangeLog:
* cp-gimplify.cc (cp_genericize_target_expr): Make sure nothing
has set DECL_INITIAL on a TARGET_EXPR slot.
* tree.cc (is_local_temp): Don't check DECL_CONTEXT.
gcc/testsuite/ChangeLog:
* g++.dg/cpp0x/initlist127.C: New test.
We've been trying for a while to avoid TARGET_EXPRs in template code, but
there were still a few that snuck through, and the one in this case broke
the code that tried to handle it. Fixed by using IMPLICIT_CONV_EXPR, as we
have done elsewhere.
I also noticed that finish_compound_literal was assuming that all T{init}
were for aggregate T, and we got a few more TARGET_EXPRs from that. Fixed
by only messing with TARGET_EXPR if we actually have an aggregate init.
PR c++/101072
gcc/cp/ChangeLog:
* cp-tree.h (build_implicit_conv_flags): Declare.
* call.cc (build_implicit_conv_flags): Split out from...
(perform_implicit_conversion_flags): ...here.
* decl.cc (check_initializer): Use it.
* pt.cc (tsubst_copy_and_build): Remove TARGET_EXPR handling.
* semantics.cc (finish_compound_literal): Don't treat
scalar values like CONSTRUCTORs.
gcc/testsuite/ChangeLog:
* g++.dg/cpp0x/constexpr-empty14a.C: New test.
With --disable-plugin, bootstrap fails on x86_64-linux or probably
all other targets with:
../../gcc/opts-global.cc: In function ‘void handle_common_deferred_options()’:
../../gcc/opts-global.cc:420:62: error: unquoted option name ‘--enable-plugin’ in format [-Werror=format-diag]
420 | error ("plugin support is disabled; configure with --enable-plugin");
| ^~~~~~~~~~~~~~~
../../gcc/opts-global.cc:428:62: error: unquoted option name ‘--enable-plugin’ in format [-Werror=format-diag]
428 | error ("plugin support is disabled; configure with --enable-plugin");
| ^~~~~~~~~~~~~~~
The following patch fixes that.
2022-01-22 Jakub Jelinek <jakub@redhat.com>
PR other/104176
* opts-global.cc (handle_common_deferred_options): Quote
--enable-plugin in diagnostics to avoid -Werror=format-diag.
PR analyzer/104159 describes an ICE attempting to convert a vector_cst,
which occurs when symbolically executing within a recursive call on:
_4 = BIT_FIELD_REF <w_3(D), 32, 0>;
_1 = VIEW_CONVERT_EXPR<T>(_4);
where the BIT_FIELD_REF leads to a get_or_create_cast from
VEC<long, 8> to VEC<unsigned 4>
which get_code_for_cast erroneously picks NOP_EXPR for the cast, leading
to a bogus input to the VIEW_CONVERT_EXPR.
This patch fixes the issue by giving up on attempts to cast symbolic
values of vector types, treating the result of such casts as unknowable.
gcc/analyzer/ChangeLog:
PR analyzer/104159
* region-model-manager.cc
(region_model_manager::get_or_create_cast): Bail out if the types
are the same. Don't attempt to handle casts involving vector
types.
gcc/testsuite/ChangeLog:
PR analyzer/104159
* gcc.dg/analyzer/torture/pr104159.c: New test.
Signed-off-by: David Malcolm <dmalcolm@redhat.com>
gcc/fortran/ChangeLog:
PR fortran/104127
* simplify.cc (gfc_simplify_transfer): Ensure that the result
typespec is set up for TRANSFER with MOLD of type CHARACTER
including character length even if the result is a zero-sized
array.
gcc/testsuite/ChangeLog:
PR fortran/104127
* gfortran.dg/transfer_simplify_11.f90: Fix logic.
* gfortran.dg/transfer_simplify_13.f90: New test.
This has been added in r12-6342-ge7a7dbb5ca5dd69689f1a probably
by accident.
2022-01-22 Jakub Jelinek <jakub@redhat.com>
PR other/104181
* build.log: Remove.
When I added support for generating XXSPLTIDP on December 15th, 2021, I
missed updating the fold-vec-splat-floatdouble.c test to add to the regex
for the instructions generated. This patch fixes that.
2022-01-20 Michael Meissner <meissner@the-meissners.org>
gcc/testsuite/
PR testsuite/103763
* gcc.target/powerpc/fold-vec-splat-floatdouble.c: Fix insn regex
on power10.
If you compile module_advect_em.F90 with -Ofast -mcpu=power10, one module
is large enough that we can't use a single conditional jump to span the
function. Instead, GCC has to reverse the condition, and do a conditional
jump around an unconditional branch. It turns out when xxspltiw and
xxspltdp instructions were generated, they were not marked as being
prefixed (i.e. length of 12 bytes instead of 4 bytes). This meant the
calculations for the branch length were off, which in turn meant the
assembler raised an error because it couldn't do the conditional jump.
The fix is to explicitly set the prefixed attribute when we are loading up
vector constants with the xxspltiw or xxspltidp instructions.
I have removed the code that sets the prefixed attribute for xxspltiw,
xxspltidp, and xxsplti32dx instructions, since it no longer will be invoked.
I have also explicitly set the prefixed attribute for load SF and DF mode
constants with xxsplitw and xxspltidp. Previously, it was not set on these
insns, but when the insn was split to get the XXSPLTIW/XXSPLTIDP forms, those
forms already had the prefixed attribute set.
2022-01-21 Michael Meissner <meissner@the-meissners.org>
gcc/
PR target/104136
* config/rs6000/rs6000-protos.h (prefixed_xxsplti_p): Delete.
* config/rs6000/rs6000.cc (prefixed_xxsplti_p): Delete.
* config/rs6000/rs6000.md (prefixed attribute): Delete section
that sets the prefixed attribute for xxspltiw, xxspltidp, and
xxsplti32dx instructions.
(movsf_hardfloat): Explicitly set the prefixed attribute
when xxspltiw and xxspltidp instructions are generated.
(mov<mode>_hardfloat32): Likewise.
(mov<mode>_hardfloat64): Likewise.
* config/rs6000/vsx.md (vsx_mov<mode>_64bit): Explicitly set the
prefixed attribute for xxspltiw and xxspltidp instructions.
(vsx_mov<mode>_32bit): Likewise.
Revert x86 changes in
commit c163647ffb
Author: Soren Tempel <soeren@soeren-tempel.net>
Date: Fri Jan 21 19:22:46 2022 +0000
Disable -fsplit-stack support on non-glibc targets
and change ix86_supports_split_stack to return true only on glibc.
PR bootstrap/104170
* common/config/i386/i386-common.cc (ix86_supports_split_stack):
Return true only on glibc.
* config/i386/gnu-user-common.h (STACK_CHECK_STATIC_BUILTIN):
Revert commit c163647ffb.
* config/i386/gnu.h (TARGET_LIBC_PROVIDES_SSP): Likewise.
As can be seen on the testcase, GCC 11 no longer warns if the format
string is wrapped inside of ()s.
This regressed with r11-2457-gdf5cf47a978, which added
if (TREE_NO_WARNING (param)) return;
to check_function_arguments_recurse. That function is used with a callback
for two cases, for -Wformat and for -Wnonnull. For the latter it is
desirable to not warn in parameters or their subexpressions where that
warning is suppressed, but for -Wformat the function is used solely
to discover the string literals if any so that the c-format.cc code can
diagnose them. I believe no warning suppression should stand in the
way of that, -Wformat* warnings should be decided from warning suppression
on the CALL_EXPR only.
In the PR Martin argued that now that we have specialized
warning_suppressed_p we should use it, so instead of adding a bool
arg to check_function_arguments_recurse I've added opt_code to the
function, but will defer the warning_suppressed_p change to him.
For OPT_Wformat_ we don't want to call it anyway at all (as I said,
I think there should be no suppression for it during the string discovery,
there isn't just one -Wformat= option, there are many and
warning_suppression_p even with no_warnings actually tests the
TREE_NO_WARNING bit).
Initially, I thought I'd restrict also call to fn with format_arg attribute
handling in check_function_arguments_recurse to OPT_Wformat_ only, but
after looking around, it perhaps is intentional that way, most functions
with format_arg attribute don't have nonnull attribute for that arg too,
various gettext implementations handle NULL argument by passing it through,
but when result of gettext (NULL) etc. is passed to non-NULL argument, it
makes sense to warn.
2022-01-21 Jakub Jelinek <jakub@redhat.com>
PR c++/104148
* c-common.h (check_function_arguments_recurse): Add for_format
arg.
* c-common.cc (check_function_nonnull): Pass false to
check_function_arguments_recurse's last argument.
(check_function_arguments_recurse): Add for_format argument,
if true, don't stop on warning_suppressed_p.
* c-format.cc (check_format_info): Pass true to
check_function_arguments_recurse's last argument.
* c-c++-common/Wformat-pr104148.c: New test.
While looking at another bug I wanted the compiler to tell me what the two
unequal values were.
gcc/cp/ChangeLog:
* semantics.cc (find_failing_clause): Return expr if not
decomposable.
(finish_static_assert): Show constant values in failing
comparison.
gcc/testsuite/ChangeLog:
* g++.dg/template/explicit-args6.C: Add expected message.
My patch for PR20040 made us stop exiting early from build_new_1 in
cases of trivial initialization if there's a class operator delete; as a
result, code later in the function needs to handle this case properly.
PR c++/104084
PR c++/20040
gcc/cp/ChangeLog:
* init.cc (build_new_1): Only pull out TARGET_EXPR_INITIAL if
alloc_expr is a TARGET_EXPR.
gcc/testsuite/ChangeLog:
* g++.dg/init/new50.C: New test.
The -fsplit-stack option requires the pthread_t TCB definition in the
libc to provide certain struct fields at specific hardcoded offsets. As
far as I know, only glibc provides these fields at the required offsets.
Most notably, musl libc does not have these fields. However, since gcc
accesses the fields using a fixed offset, this does not cause a
compile-time error, but instead results in a silent memory corruption at
run-time with musl libc. For example, on s390x libgcc's
__stack_split_initialize CTOR will overwrite the cancel field in the
pthread_t TCB on musl.
The -fsplit-stack option is used within the gcc code base itself by
gcc-go (if available). On musl-based systems with split-stack support
(i.e. s390x or x86) this causes Go programs compiled with gcc-go to
misbehave at run-time.
This patch fixes gcc-go on musl by disabling -fsplit-stack in gcc itself
since it is not supported on non-glibc targets anyhow. This is achieved
by checking if gcc targets a glibc-based system. This check has been
added for x86 and s390x, the rs6000 config already checks for
TARGET_GLIBC_MAJOR. Other architectures do not have split-stack
support. With this patch applied, the gcc-go configure script will
detect that -fsplit-stack support is not available and will not use it.
See https://www.openwall.com/lists/musl/2012/10/16/12
This patch was written under the assumption that glibc is the only libc
implementation which supports the required fields at the required
offsets in the pthread_t TCB. The patch has been tested on Alpine Linux
Edge on the s390x and x86 architectures by bootstrapping Google's Go
implementation with gcc-go.
Signed-off-by: Sören Tempel <soeren@soeren-tempel.net>
gcc/ChangeLog:
* common/config/s390/s390-common.cc (s390_supports_split_stack):
Only support split-stack on glibc targets.
* config/i386/gnu-user-common.h (STACK_CHECK_STATIC_BUILTIN): Ditto.
* config/i386/gnu.h (defined): Ditto.
Fix a compilation issue in stage2 bootstrap. Fixed as obvious (re:
discussion with Bill Schmidt).
2022-01-21 Bill Seurer <seurer@gcc.gnu.org>
gcc/
* config/rs6000/rs6000.cc (rs6000_get_function_versions_dispatcher):
Fix mention of ifunc in string.
This patch resolves the P1 "ice-on-valid-code" regression boostrapping
GCC on risv-unknown-linux-gnu caused by my recent MULT_HIGHPART_EXPR
functionality. RISC-V differs from x86_64 and many targets by
supporting a usmusidi3 instruction, basically a widening multiply
where one operand is signed and the other is unsigned. Alas the
final version of my patch to recognize MULT_HIGHPART_EXPR didn't
sufficiently defend against the operands of WIDEN_MULT_EXPR having
different signedness. This is fixed by the two-line change to
tree-ssa-math-opts.cc's convert_mult_to_highpart in the patch below.
The majority of the rest of the patch is to the documentation
(in tree.def and generic.texi). It turns out that WIDEN_MULT_EXPR
wasn't previously documented in generic.texi, let alone the slightly
unusual semantics of allowing mismatched (signed vs unsigned) operands.
This also clarifies that MULT_HIGHPART_EXPR currently requires the
signedness of operands to match [but this might change in a future
release of GCC to support targets with usmul<mode>3_highpart].
The one final chunk of this patch (that is hopefully sufficiently
close to obvious for stage 4) is a similar (NULL pointer) sanity
check in riscv_cpu_cpp_builtins. Currently running cc1 from the
command line (or from gdb) without specifying -march results in a
segmentation fault (ICE). This is a minor annoyance tracking down
issues (in cross compilers) for riscv, and trivially fixed as below.
2022-01-22 Roger Sayle <roger@nextmovesoftware.com>
gcc/ChangeLog
PR middle-end/104140
* tree-ssa-math-opts.cc (convert_mult_to_highpart): Check that the
operands of the widening multiplication are either both signed or
both unsigned, and abort the conversion if mismatched.
* doc/generic.texi (WIDEN_MULT_EXPR): Describe expression node.
(MULT_HIGHPART_EXPR): Clarify that operands must have the same
signedness.
* tree.def (MULT_HIGHPART_EXPR): Document both operands must have
integer types with the same precision and signedness.
(WIDEN_MULT_EXPR): Document that operands must have integer types
with the same precision, but possibly differing signedness.
* config/riscv/riscv-c.cc (riscv_cpu_cpp_builtins): Defend against
riscv_current_subset_list returning a NULL pointer (empty list).
gcc/testsuite/ChangeLog
PR middle-end/104140
* gcc.target/riscv/pr104140.c: New test case.
LRA and old reload pass uses only one register class for reload pseudos even if
operand constraints contain more one register class. Let us consider
constraint 'lh' for thumb arm which means low and high thumb registers.
Reload pseudo for such constraint will have general reg class (union of
low and high reg classes). Assigning the last low register to the reload
pseudo is wrong if the pseudo is of DImode as it requires two hard regs.
But it is considered OK if we use general reg class. The following patch
solves this problem for LRA.
gcc/ChangeLog:
PR target/103676
* ira.h (struct target_ira): Add member
x_ira_exclude_class_mode_regs.
(ira_exclude_class_mode_regs): New macro.
* lra.h (lra_create_new_reg): Add arg exclude_start_hard_regs and
move from here ...
* lra-int.h: ... to here.
(lra_create_new_reg_with_unique_value): Add arg
exclude_start_hard_regs.
(class lra_reg): Add member exclude_start_hard_regs.
* lra-assigns.cc (find_hard_regno_for_1): Setup
impossible_start_hard_regs from exclude_start_hard_regs.
* lra-constraints.cc (get_reload_reg): Add arg exclude_start_hard_regs and pass
it lra_create_new_reg[_with_unique_value].
(match_reload): Ditto.
(check_and_process_move): Pass NULL
exclude_start_hard_regs to lra_create_new_reg_with_unique_value.
(goal_alt_exclude_start_hard_regs): New static variable.
(process_addr_reg, simplify_operand_subreg): Pass NULL
exclude_start_hard_regs to lra_create_new_reg_with_unique_value
and get_reload_reg.
(process_alt_operands): Setup goal_alt_exclude_start_hard_regs.
Use this_alternative_exclude_start_hard_regs additionally to find
winning operand alternative.
(base_to_reg, base_plus_disp_to_reg, index_part_to_reg): Pass NULL
exclude_start_hard_regs to lra_create_new_reg.
(process_address_1, emit_inc): Ditto.
(curr_insn_transform): Pass exclude_start_hard_regs value to
lra_create_new_reg, get_reload_reg, match_reload.
(inherit_reload_reg, split_reg): Pass NULL exclude_start_hard_regs
to lra_create_new_reg.
(process_invariant_for_inheritance): Ditto.
* lra-remat.cc (update_scratch_ops): Ditto.
* lra.cc (lra_create_new_reg_with_unique_value): Add arg
exclude_start_hard_regs. Setup the corresponding member of
lra reg info.
(lra_create_new_reg): Add arg exclude_start_hard_regs and pass it
to lra_create_new_reg_with_unique_value.
(initialize_lra_reg_info_element): Initialize member
exclude_start_hard_regs.
(get_scratch_reg): Pass NULL to lra_create_new_reg.
* ira.cc (setup_prohibited_class_mode_regs): Rename to
setup_prohibited_and_exclude_class_mode_regs and calculate
ira_exclude_class_mode_regs.
gcc/testsuite/ChangeLog:
PR target/103676
* g++.target/arm/pr103676.C: New.
This is a "canonical types differ for identical types" ICE, which started
with r11-4682. It's a bit tricky to explain. Consider:
template <typename T> struct S {
S<T> bar() noexcept(T::value); // #1
S<T> foo() noexcept(T::value); // #2
};
template <typename T> S<T> S<T>::foo() noexcept(T::value) {} // #3
We ICE because #3 and #2 have the same type, but their canonical types
differ: TYPE_CANONICAL (#3) == #2 but TYPE_CANONICAL (#2) == #1.
The member functions #1 and #2 have the same type. However, since their
noexcept-specifier is deferred, when parsing them, we create a variant for
both of them, because DEFERRED_PARSE cannot be compared. In other words,
build_cp_fntype_variant's
tree v = TYPE_MAIN_VARIANT (type);
for (; v; v = TYPE_NEXT_VARIANT (v))
if (cp_check_qualified_type (v, type, type_quals, rqual, raises, late))
return v;
will *not* find an existing variant when creating a method_type for #2, so we
have to create a new one.
But then we perform delayed parsing and call fixup_deferred_exception_variants
for #1 and #2. f_d_e_v will replace TYPE_RAISES_EXCEPTIONS with the newly
parsed noexcept-specifier. It also sets TYPE_CANONICAL (#2) to #1. Both
noexcepts turned out to be the same, so now we have two equivalent variants in
the list! I.e.,
+-----------------+ +-----------------+ +-----------------+
| main | | #2 | | #1 |
| S S::<T379>(S*) |----->| S S::<T37c>(S*) |----->| S S::<T37a>(S*) |----->NULL
| - | | noex(T::value) | | noex(T::value) |
+-----------------+ +-----------------+ +-----------------+
Then we get to #3. As for #1 and #2, grokdeclarator calls build_memfn_type,
which ends up calling build_cp_fntype_variant, which will use the loop
above to look for an existing variant. The first one that matches
cp_check_qualified_type will be used, so we use #2 rather than #1, and the
TYPE_CANONICAL mismatch follows. Hopefully that makes sense.
As for the fix, I didn't think I could rewrite the method_type #2 with #1
because the type may have escaped via decltype. So my approach is to
elide #2 from the list, so when looking for a matching variant, we always
find #1 (#2 remains live though, which admittedly sounds sort of dodgy).
PR c++/101715
gcc/cp/ChangeLog:
* tree.cc (fixup_deferred_exception_variants): Remove duplicate
variants after parsing the exception specifications.
gcc/testsuite/ChangeLog:
* g++.dg/cpp0x/noexcept72.C: New test.
* g++.dg/cpp0x/noexcept73.C: New test.
Martin Sebor
Marek Polacek
Francois-Xavier Coudert
Raoni Fassina Firmino
Patrick Palka
Martin Liska
Richard Biener
Jakub Jelinek
Kito Cheng
Jia-Wei Chen
Jiufu Guo
LiaoShihua
David Edelsohn
H.J. Lu
GCC Administrator
Jonathan Wakely
Will Wray
Andrew Pinski
Jason Merrill
David Malcolm
Harald Anlauf
Michael Meissner
Sören Tempel
Bill Schmidt
Bill Seurer
Roger Sayle
Vladimir N. Makarov