Uros Bizjak [Mon, 28 Dec 2020 20:30:08 +0000 (21:30 +0100)]
i386: Fix __builtin_rint with FE_DOWNWARD rounding direction [PR96793]
x86_expand_rint expander uses x86_sse_copysign_to_positive, which
is unable to change the sign from - to +. When FE_DOWNWARD rounding
direction is in effect, the expanded sequence that involves subtraction
can trigger x - x = -0.0 special rule. x86_sse_copysign_to_positive
fails to change the sign of the intermediate value, assumed to always
be positive, back to positive.
The patch adds one extra fabs that strips the sign from the intermediate
value when flag_rounding_math is in effect.
2020-12-28 Uroš Bizjak <ubizjak@gmail.com>
gcc/
PR target/96793
* config/i386/i386-expand.c (ix86_expand_rint):
Remove the sign of the intermediate value for flag_rounding_math.
gcc/testsuite/
PR target/96793
* gcc.target/i386/pr96793-2.c: New test.
Uros Bizjak [Mon, 28 Dec 2020 15:52:45 +0000 (16:52 +0100)]
i386: Use existing temporary register in rounding functions
It is possible to avoid the call to force_reg and use existing
temporary register in ix86_expand_trunc, ix86_expand_round and
ix86_expand_rounddf_32 expanders.
2020-12-28 Uroš Bizjak <ubizjak@gmail.com>
gcc/
* config/i386/i386-expand.c (ix86_expand_trunc): Use
existing temporary register to avoid a call to force_reg.
Paul Thomas [Sun, 27 Dec 2020 14:59:38 +0000 (14:59 +0000)]
Fortran: Fix some select rank issues [PR97694 and 97723].
2020-12-27 Paul Thomas <pault@gcc.gnu.org>
gcc/fortran
PR fortran/97694
PR fortran/97723
* check.c (allocatable_check): Select rank temporaries are
permitted even though they are treated as associate variables.
* resolve.c (gfc_resolve_code): Break on select rank as well as
select type so that the block os resolved.
* trans-stmt.c (trans_associate_var): Class associate variables
that are optional dummies must use the backend_decl.
gcc/testsuite/
PR fortran/97694
PR fortran/97723
* gfortran.dg/select_rank_5.f90: New test.
Paul Thomas [Sat, 26 Dec 2020 16:44:24 +0000 (16:44 +0000)]
Fortran: Correction to recent patch in light of comments [PR98022].
2020-12-26 Paul Thomas <pault@gcc.gnu.org>
gcc/fortran
PR fortran/98022
* data.c (gfc_assign_data_value): Throw an error for inquiry
references. Follow with corrected code that would provide the
expected result and provides clean error recovery.
gcc/testsuite/
PR fortran/98022
* gfortran.dg/data_inquiry_ref.f90: Change to dg-compile and
add errors for inquiry references.
Paul Thomas [Sat, 26 Dec 2020 15:08:11 +0000 (15:08 +0000)]
Fix failures with -m32 and some memory leaks.
2020-12-23 Paul Thomas <pault@gcc.gnu.org>
gcc/fortran
PR fortran/83118
* trans-array.c (gfc_alloc_allocatable_for_assignment): Make
sure that class expressions are captured for dummy arguments by
use of gfc_get_class_from_gfc_expr otherwise the wrong vptr is
used.
* trans-expr.c (gfc_get_class_from_gfc_expr): New function.
(gfc_get_class_from_expr): If a constant expression is
encountered, return NULL_TREE;
(gfc_trans_assignment_1): Deallocate rhs allocatable components
after passing derived type function results to class lhs.
* trans.h : Add prototype for gfc_get_class_from_gfc_expr.
Harald Anlauf [Fri, 25 Dec 2020 14:40:39 +0000 (15:40 +0100)]
PR93685 - ICE in gfc_constructor_append_expr, at fortran/constructor.c:135
Fix handling of F2018 enhancements to DATA statements that allows
initialization of pointer components to derived types, and adjust error
handling for the CHARACTER case.
gcc/fortran/ChangeLog:
* data.c (gfc_assign_data_value): Restrict use of
create_character_initializer to constant initializers.
* trans-expr.c (gfc_conv_initializer): Ensure that character
initializer is constant, otherwise fall through to get the same
error handling as for non-character cases.
gcc/testsuite/ChangeLog:
* gfortran.dg/pr93685_1.f90: New test.
* gfortran.dg/pr93685_2.f90: New test.
Iain Sandoe [Wed, 27 Nov 2019 13:58:10 +0000 (13:58 +0000)]
C++ : Add the -stdlib= option.
This option allows the user to specifiy alternate C++ runtime libraries,
for example when a platform uses libc++ as the installed C++ runtime.
We introduce the command line option: -stdlib= which is the user-facing
mechanism to select the C++ runtime to be used when compiling and linking
code. This is the same option spelling as that used by clang to allow the
use of libstdc++.
The availability (and thus function) of the option are a configure-time
choice using the configuration control:
--with-gxx-libcxx-include-dir=
Specification of the path for the libc++ headers, enables the -stdlib=
option (using the path as given), default values are set when the path
is unconfigured.
If --with-gxx-libcxx-include-dir is given together with --with-sysroot=,
then we test to see if the include path starts with the sysroot and, if so,
record the sysroot-relative component as the local path. At runtime, we
prepend the sysroot that is actually active.
At link time, we use the C++ runtime in force and (if that is libc++) also
append the libc++abi ABI library. As for other cases, if a target sets the
name pointer for the ABI library to NULL the G++ driver will omit it from
the link line.
gcc/ChangeLog:
* configure.ac: Add gxx-libcxx-include-dir handled
in the same way as the regular cxx header directory.
* Makefile.in: Regenerated.
* config.in: Likewise.
* configure: Likewise.
* cppdefault.c: Pick up libc++ headers if the option
is enabled.
* cppdefault.h (struct default_include): Amend comments
to reflect the extended use of the cplusplus field.
* incpath.c (add_standard_paths): Allow for multiple
c++ header include path variants.
* doc/invoke.texi: Document the -stdlib= option.
gcc/c-family/ChangeLog:
* c.opt: Add -stdlib= option and enumerations for
libstdc++ and libc++.
Nathan Sidwell [Wed, 23 Dec 2020 22:17:17 +0000 (14:17 -0800)]
c++: EOF location [PR 96045]
Setting the EOF token location to be the start of a line just after
the ending newline is not most helpful. While that location is
probably the right place to report preprocessing and lexing issues,
when parsing, the location just after the last token is better. That
way we get to point at some actual text. Setting the location from
the previous token has the advantage over just setting the location to
be the end of the final line, in that any ending comments do not get
considered, which I think is better.
Jason Merrill [Tue, 22 Dec 2020 20:41:56 +0000 (15:41 -0500)]
c++: Fix initializing empty base from prvalue [PR97597]
unsafe_return_slot_p wasn't recognizing an empty base as
potentially-overlapping, which it definitely is.
The change to build_base_path is to make the virtual conversion also
recognized by is_empty_base_ref; unsafe_return_slot_p doesn't to handle
virtual conversions, but hypothetical future callers might.
gcc/cp/ChangeLog:
PR c++/97597
* class.c (is_empty_base_ref): New.
(build_base_path): Add NOP_EXPR after offset.
* cp-tree.h (is_empty_base_ref): Declare it.
* call.c (unsafe_return_slot_p): Call it.
Ian Lance Taylor [Wed, 23 Dec 2020 21:54:35 +0000 (13:54 -0800)]
compiler: nil-checked pointers and index expressions can trap
The compiler was treating indirections through pointers that had been
explicitly checked against nil and slice and string index expressions
as non-trapping memory references. That is true for ordinary Go
programs, but it isn't true if the programs construct their own memory
addresses. In particular it isn't true for the kinds of programs that
want to use runtime.SetPanicOnFault.
The effect of this will be slightly larger binaries, due to additional
exception information, and perhaps slightly less optimization.
Jakub Jelinek [Wed, 23 Dec 2020 21:45:56 +0000 (22:45 +0100)]
c++: Fix up floating point complex handling in build_zero_init_1 [PR98353]
While the gimplifier patch I've just committed fixed an ICE, in some cases
like on the committed testcase cp folding doesn't happen after
build_zero_init_1 because it is called already during gimplification.
For the scalar types, if we want to use convert, the problem with complex floats
is that it returns a COMPLEX_EXPR with FLOAT_EXPR arguments which have
INTEGER_CST 0 as argument. As fold isn't recursive, it doesn't do anything
in that case, we need to first fold those FLOAT_EXPRs to REAL_CST 0.0 and
only afterwards the COMPLEX_EXPR can be folded into COMPLEX_CST with 0.0
arguments.
This patch instead just uses build_zero_cst that creates the zero constant for
any scalar types (and more) directly, instead of going through multiple hops.
2020-12-23 Jakub Jelinek <jakub@redhat.com>
PR c++/98353
* init.c (build_zero_init_1): Use build_zero_cst for SCALAR_TYPE_P
zero initializers.
Jason Merrill [Tue, 22 Dec 2020 21:40:37 +0000 (16:40 -0500)]
c++: Fix constexpr array ICE [PR98332]
The element initializer was non-constant, so its CONSTRUCTOR element ended
up NULL, so unshare_constructor crashed trying to look at it. This patch
fixes this in two places: First, by returning when we see a non-constant
initializer; second, by not crashing on NULL.
gcc/cp/ChangeLog:
PR c++/98332
* constexpr.c (unshare_constructor): Check for NULL.
(cxx_eval_vec_init_1): Always exit early if non-constant.
gcc/testsuite/ChangeLog:
PR c++/98332
* g++.dg/cpp0x/constexpr-overflow3.C: New test.
Iain Sandoe [Wed, 23 Dec 2020 17:16:08 +0000 (17:16 +0000)]
Darwin : Adjust handling of MACOSX_DEPLOYMENT_TARGET for macOS 11.
The shift to macOS version 11 also means that '11' without any
following '.x' is accepted as a valid version number. This adjusts
the validation code to accept this and map it to 11.0.0 which
matches what the clang toolchain appears to do.
gcc/ChangeLog:
* config/darwin-driver.c (validate_macosx_version_min): Allow
MACOSX_DEPLOYMENT_TARGET=11.
(darwin_default_min_version): Adjust warning spelling to avoid
an apostrophe.
Iain Buclaw [Tue, 22 Dec 2020 08:47:22 +0000 (09:47 +0100)]
d: Force TYPE_MODE of classes and non-POD structs as BLKmode
Without this being forced, the optimizer could still make decisions that
require objects of the non-POD types to need a temporary, which would
result in an ICE during the expand to RTL passes.
gcc/d/ChangeLog:
PR d/98427
* types.cc (TypeVisitor::visit (TypeStruct *)): Set TYPE_MODE of all
non-trivial types as BLKmode.
(TypeVisitor::visit (TypeClass *)): Likewise.
Uros Bizjak [Wed, 23 Dec 2020 15:47:52 +0000 (16:47 +0100)]
i386: Fix __builtin_trunc with FE_DOWNWARD rounding direction [PR96793]
x86_expand_truncdf_32 expander uses x86_sse_copysign_to_positive, which
is unable to change the sign from - to +. When FE_DOWNWARD rounding
direction is in effect, the expanded sequence that involves subtraction
can trigger x - x = -0.0 special rule. x86_sse_copysign_to_positive
fails to change the sign of the intermediate value, assumed to always
be positive, back to positive.
The patch adds one extra fabs that strips the sign from the intermediate
value when flag_rounding_math is in effect.
2020-12-23 Uroš Bizjak <ubizjak@gmail.com>
gcc/
PR target/96793
* config/i386/i386-expand.c (ix86_expand_truncdf_32):
Remove the sign of the intermediate value for flag_rounding_math.
gcc/testsuite/
PR target/96793
* gcc.target/i386/pr96793-1.c: New test.
Ian Lance Taylor [Tue, 22 Dec 2020 19:04:35 +0000 (11:04 -0800)]
runtime: eliminate scase.kind field
This is the gofrontend version of https://golang.org/cl/245125.
Original CL description:
Currently, we include a "kind" field on scase to distinguish the three
kinds of cases in a select statement: sends, receives, and defaults.
This commit removes by kind field by instead arranging for the
compiler to always place sends before receives, and to provide their
counts separately. It also passes an explicit "block bool" parameter
to avoid needing to include a default case in the array.
It's safe to shuffle cases like this because the runtime will
randomize the order they're polled in anyway.
For golang/go#40410.
This is being brought over to gofrontend as a step toward upgrading to
Go1.16beta1.
Ian Lance Taylor [Tue, 22 Dec 2020 05:20:59 +0000 (21:20 -0800)]
runtime: add "success" field to sudog
This is the gofrontend version of https://golang.org/cl/245019.
Original CL description:
The current wakeup protocol for channel communications is that the
second goroutine sets gp.param to the sudog when a value is
successfully communicated over the channel, and to nil when the wakeup
is due to closing the channel.
Setting nil to indicate channel closure works okay for chansend and
chanrecv, because they're only communicating with one channel, so they
know it must be the channel that was closed. However, it means
selectgo has to re-poll all of the channels to figure out which one
was closed.
This commit adds a "success" field to sudog, and changes the wakeup
protocol to always set gp.param to sg, and to use sg.success to
indicate successful communication vs channel closure.
While here, this also reorganizes the chansend code slightly so that
the sudog is still released to the pool if the send blocks and then is
awoken because the channel closed.
For golang/go#40410
This is being brought over to gofrontend as a step toward upgrading to
Go1.16beta1, setting up for more compiler changes related to select handling.
Ian Lance Taylor [Tue, 22 Dec 2020 04:51:18 +0000 (20:51 -0800)]
runtime: omit nil-channel cases from selectgo's orders
This is the gofrontend version of https://golang.org/cl/245123.
Original CL description:
Currently, selectgo does an initial pass over the cases array to look
for entries with nil channels, so they can be easily recognized and
skipped later on. But this still involves actually visiting the cases.
This commit changes selectgo to omit cases with nil channels when
constructing pollorder, so that they'll be skipped over entirely later
on. It also checks for caseDefault up front, which will facilitate
changing it to use a "block bool" parameter instead.
Updates golang/go#40410
This is being brought over to gofrontend as a step toward upgrading to
Go1.16beta1, setting up for more compiler changes related to select handling.
Ian Lance Taylor [Mon, 21 Dec 2020 22:58:14 +0000 (14:58 -0800)]
runtime: remove scase.releasetime field
This is the gofrontend version of https://golang.org/cl/245122.
Original CL description:
selectgo will report at most one block event, so there's no need to
keep a releasetime for every select case. It suffices to simply track
the releasetime of the case responsible for the wakeup.
Updates golang/go#40410.
This is being brought over to gofrontend as a step toward upgrading to
Go1.16beta1.
Jakub Jelinek [Tue, 22 Dec 2020 19:18:10 +0000 (20:18 +0100)]
c++: Handle array members in build_comparison_op [PR93480]
http://eel.is/c++draft/class.compare.default#6 says for the
expanded list of subobjects:
"In that list, any subobject of array type is recursively expanded
to the sequence of its elements, in the order of increasing subscript."
but build_comparison_op just tried to compare the whole arrays, which
failed and therefore the defaulted comparison was deleted.
The following patch instead compares the array elements, and
if info.defining, adds runtime loops around it so that it iterates
over increasing subscripts.
For flexible array members it punts, we don't know how large those will be,
for zero sized arrays it doesn't even try to compare the elements,
because if there are no elements, there is nothing to compare, and
for [1] arrays it will not emit a loop because it is enough to use
[0] array ref to cover everything.
2020-12-21 Jakub Jelinek <jakub@redhat.com>
PR c++/93480
* method.c (common_comparison_type): If comps[i] is a TREE_LIST,
use its TREE_VALUE instead.
(build_comparison_op): Handle array members.
* g++.dg/cpp2a/spaceship-synth10.C: New test.
* g++.dg/cpp2a/spaceship-synth-neg5.C: New test.
Qian Jianhua [Tue, 22 Dec 2020 18:54:34 +0000 (18:54 +0000)]
arm&aarch64: subdivide the type attribute "alu_shfit_imm"
The type attribute "alu_shfit_imm" is subdivided into
"alu_shift_imm_lsl_1to4" and "alu_shift_imm_other", to accommodate
optimazations of some microarchitectures.
Here is the detailed discussion.
https://gcc.gnu.org/pipermail/gcc/2020-September/233594.html
Uros Bizjak [Tue, 22 Dec 2020 17:13:24 +0000 (18:13 +0100)]
i386: Fix __builtin_floor with FE_DOWNWARD rounding direction [PR96793]
x86_expand_floorceil expander uses x86_sse_copysign_to_positive, which
is unable to change the sign from - to +. When FE_DOWNWARD rounding
direction is in effect, the expanded sequence that involves subtraction
can trigger x - x = -0.0 special rule. x86_sse_copysign_to_positive
fails to change the sign of the intermediate value, assumed to always
be positive, back to positive.
The patch adds one extra fabs that strips the sign from the intermediate
value when flag_rounding_math is in effect.
2020-12-22 Uroš Bizjak <ubizjak@gmail.com>
gcc/
PR target/96793
* config/i386/i386-expand.c (ix86_expand_floorceil):
Remove the sign of the intermediate value for flag_rounding_math.
(ix86_expand_floorceildf_32): Ditto.
gcc/testsuite/
PR target/96793
* gcc.target/i386/pr96793.c: New test.
Gerald Pfeifer [Tue, 22 Dec 2020 12:37:56 +0000 (13:37 +0100)]
c++: Fix build with clang
After the addition of libcody building with clang 10.0.1 would fail
as follows:
In file included from .../libcody/cody.hh:24:
In file included from /usr/include/c++/v1/memory:653:
/usr/include/c++/v1/typeinfo:346:5: error: no member named 'fancy_abort'
in namespace 'std::__1'; did you mean simply 'fancy_abort'?
_VSTD::abort();
^~~~~~~
mapper-client.cc and mapper-resolver.cc have addressed this already
by getting <string> and <vector> included; do the same for module.cc.
Jason Merrill [Mon, 21 Dec 2020 22:31:16 +0000 (17:31 -0500)]
c++: Add some asserts in mangle.c
The added asserts are places I noticed that we aren't adding the "on" to
distinguish an operator name from an expression involving the operator.
Hopefully that's because operator names never get there; if not, crash bug
reports will let us fix the issue.
Jason Merrill [Mon, 21 Dec 2020 22:36:25 +0000 (17:36 -0500)]
c++: Fix demangling of <unresolved-name>
The ABI for unresolved scoped names on the RHS of . and -> used to be
sr <type> <unqualified-id>
That changed years ago to something more complex, but G++ was never updated.
This change was particularly incompatible for simple qualified-ids like
A::x, which were previously mangled as sr1A1x, and now sr1AE1x.
This obviously makes life hard for demanglers, which can't know whether to
consume that E or not. To work around this, we now try demangling with the
newer ABI, and if that fails and we saw an "sr", try again with the older
ABI.
libiberty/ChangeLog:
PR c++/67343
* cp-demangle.h (struct d_info): Add unresolved_name_state.
* cp-demangle.c (d_prefix): Add subst parm.
(d_nested_name): Pass it.
(d_unresolved_name): Split out from...
(d_expression_1): ...here.
(d_demangle_callback): Maybe retry with old sr mangling.
* testsuite/demangle-expected: Add test.
Jakub Jelinek [Mon, 21 Dec 2020 23:01:34 +0000 (00:01 +0100)]
gimplify: Gimplify value in gimplify_init_ctor_eval_range [PR98353]
gimplify_init_ctor_eval_range wasn't gimplifying value, so if it wasn't
a gimple val, verification at the end of gimplification would ICE (or with
release checking some random pass later on would ICE or misbehave).
2020-12-21 Jakub Jelinek <jakub@redhat.com>
PR c++/98353
* gimplify.c (gimplify_init_ctor_eval_range): Gimplify value before
storing it into cref.
Jason Merrill [Sat, 19 Dec 2020 18:10:01 +0000 (13:10 -0500)]
c++: Fix demangling of x.operator type
d_operator_name decides whether "cv" indicates a cast or a conversion
operator based on is_expression. "on" specifies that we want the conversion
operator.
Iain Sandoe [Sat, 19 Dec 2020 13:05:34 +0000 (13:05 +0000)]
Darwin : Update the kernel version to macOS version mapping.
With the change to macOS 11 and Darwin20, the algorithm for mapping
kernel version to macOS version has changed.
We now have darwin 20.X.Y => macOS 11.(X > 0 ? X - 1 : 0).??.
It currently unclear if the Y will be mapped to macOS patch version
and, if so, whether it will be one-based or 0-based.
Likewise, it's unknown if Darwin 21 will map to macOS 12, so these
entries are unchanged for the present.
gcc/ChangeLog:
* config/darwin-driver.c (darwin_find_version_from_kernel):
Compute the minor OS version from the minor kernel version.
Patrick Palka [Mon, 21 Dec 2020 20:15:36 +0000 (15:15 -0500)]
libstdc++: Disable floating-point std::to_chars on unsupported targets
This patch conditionally disables the floating-point std::to_chars
implementation on targets whose float and double aren't IEEE binary32
and binary64, until a proper fallback can be added for such targets.
This fixes a bootstrap failure on non-IEEE-754 FP targets such as
vax-netbsdelf.
The new preprocessor tests in c++config that detect the binary32 and
binary64 formats were copied from gcc/testsuite/gcc.dg/float-exact-1.c.
libstdc++-v3/ChangeLog:
* include/bits/c++config (_GLIBCXX_FLOAT_IS_IEEE_BINARY_32):
Define this macro.
(_GLIBCXX_DOUBLE_IS_IEEE_BINARY_64): Likewise.
* include/std/charconv (to_chars): Use these macros to
conditionally hide the overloads for floating-point types.
* src/c++17/floating_to_chars.cc: Use the macros to
conditionally disable this file.
(floating_type_traits<float>): Remove redundant static assert.
(floating_type_traits<double>): Likewise.
* testsuite/20_util/to_chars/double.cc: Run this test only on
ieee-floats effective targets.
* testsuite/20_util/to_chars/float.cc: Likewise.
* testsuite/20_util/to_chars/long_double.cc: Likewise.
* testsuite/lib/libstdc++.exp
(check_effective_target_ieee-floats): Define new proc for
detecting whether float and double have the IEEE binary32 and
binary64 formats.
Nathan Sidwell [Mon, 21 Dec 2020 13:32:28 +0000 (05:32 -0800)]
c++tools: Fix exe suffix [PR 98409]
I had a thinko about variable case, and, coupled with Make's behaviour
of just consing up variables out of nothing, and linux not having an
executable extension, didn't notice.
Jakub Jelinek [Mon, 21 Dec 2020 09:14:46 +0000 (10:14 +0100)]
fold-const: Fix up a buffer overflow in native_encode_initializer [PR98407]
For flexible array members we need to incrementally clear just from
ptr + total_bytes up to new ptr + total_bytes, but memset has been called
with the length from ptr, so was missing - total_bytes. Additionally,
in this code off is guaranteed to be -1 and thus o 0, so don't bother pretending
we could handle anything else, it would be more complicated than that.
2020-12-21 Jakub Jelinek <jakub@redhat.com>
PR tree-optimization/98407
* fold-const.c (native_encode_initializer): When handling flexible
array members, fix up computation of length for memset. Also remove
" - o" as o is always guaranteed to be 0 in this code path.
Jakub Jelinek [Mon, 21 Dec 2020 07:59:05 +0000 (08:59 +0100)]
openmp: Fix up handling of addressable temporaries in simd lb, b and incr expressions [PR98383]
For simd, we have code to artificially add locally defined variables into
private clauses if they are addressable, so that omplower turns them into
"omp simd array" variables. As the testcase shows, this is undesirable if
those temporaries only show in the lb, b or incr expressions and nowhere else,
if it is just used there, we really want normal scalar temporaries.
This patch implements that by making sure we don't set for those GOVD_LOCAL-ish
temporaries turned into GOVD_PRIVATE the GOVD_SEEN flag during gimplification
of the lb, b and incr expressions, which means that the private clause isn't
added for those.
2020-12-21 Jakub Jelinek <jakub@redhat.com>
PR c++/98383
* gimplify.c (struct gimplify_omp_ctx): Add in_for_exprs flag.
(gimple_add_tmp_var): For addressable temporaries appearing in
simd lb, b or incr expressions, don't add a private clause unless
it is seen also outside of those expressions in the simd body.
(omp_notice_variable): Likewise.
(gimplify_omp_for): Set and reset in_for_exprs around gimplification
of lb, b or incr expressions.
Nikhil Benesch [Wed, 16 Dec 2020 13:41:20 +0000 (08:41 -0500)]
libgo: adjust sysinfo scripts for changed -fdump-go-spec
The -fdump-go-spec flag to GCC recently changed to be more fastidious
about handling incomplete types. This caused some breakage in
mk[r]sysinfo.sh on Solaris. This commit adjusts for the new behavior.
Specifically:
* Types that refer to _in6_addr may be hidden behind a typedef and can
no longer be filtered out with `grep -v in6_addr`. Instead just
rewrite the definition of _in6_addr to [16]byte wherever it appears.
* timestruc_t is now (correctly) emitted as an alias for timespec, so this
case is handled specially.
* stdio.h is included in sysinfo.c to avoid emitting an incomplete
definition of the FILE type.
* Dummy definitions for _u?pad128_t are now emitted automatically,
which conflict with the definitions installed by mk[r]sysinfo.sh.
These definitions were actually dead code, so just remove them.
Thomas Koenig [Sun, 20 Dec 2020 15:27:01 +0000 (16:27 +0100)]
Set DECL_ARTIFICIAL on gfortran internal variables.
It seems we sometimes use DECL_ARTIFICIAL as choosing between
different code paths. In order not to make -fdebug-aux-vars
do different things, set DECL_ARTIFICIAL on the variables to
avoid these different code paths (and the corresponding
regressions).
gcc/fortran/ChangeLog:
* trans.c (create_var_debug_raw): Set DECL_ARTIFICIAL
on variables.
Jakub Jelinek [Sat, 19 Dec 2020 21:24:10 +0000 (22:24 +0100)]
expr: Fix up constant_byte_string bitfield handling [PR98366]
constant_byte_string now uses a convert_to_bytes function, which doesn't
handle bitfields at all (don't punt on them, just puts them into wrong bits
or bytes). Furthermore, I don't see a reason why that function should exist
at all, it duplicates native_encode_initializer functionality.
Except that native_encode_initializer punted on flexible array members and 2
tests in the testsuite relied on constant_byte_string handling those.
So, this patch throws away convert_to_bytes, uses native_encode_initializer
instead, but teaches it to handle flexible array members (only in the
non-mask mode with off == -1 for now), furthermore, it adds various corner
case checks that the old implementation was missing (like that STRING_CSTs
use int as length and therefore we shouldn't try to build larger than that
strings, or that native_encode*/native_interpret* APIs require sane
host and target bytes (8-bit on both).
2020-12-19 Jakub Jelinek <jakub@redhat.com>
PR middle-end/98366
* fold-const.c (native_encode_initializer): Don't try to
memset more than total_bytes with off == -1 even if len is large.
Handle flexible array member initializers if off == -1 and mask is
NULL.
* expr.c (convert_to_bytes): Remove.
(constant_byte_string): Use native_encode_initializer instead of
convert_to_bytes. Remove extraneous semicolon. Punt on various
corner-cases the APIs don't handle, like sizes > INT_MAX,
BITS_PER_UNIT != 8, CHAR_BIT != 8.
Iain Buclaw [Tue, 15 Dec 2020 23:17:40 +0000 (00:17 +0100)]
d: Fix ICE in in force_decl_die, at dwarf2out.c with -gdwarf-2 -gstrict-dwarf [PR98067]
Manifest constants in D are represented as CONST_DECLs, which can be
imported from one module to another. However, when compiling on strict
dwarf2 targets such as *-*-darwin10, importing CONST_DECLs cannot be
represented in debug as D did not exist as an AT_language until dwarf3,
and the only available fallback being DW_LANG_C. As CONST_DECLs are
treated as enumerators in C, and not outputted individually in
gen_decl_die, this causes an internal error in force_decl_die to occur.
To handle this, similar to other places in dwarf2out, if a CONST_DECL is
seen in dwarf2out_imported_module_or_decl_1, then we simply return early
if the language is not one of Ada, D, or Fortran.
gcc/ChangeLog:
PR d/98067
* dwarf2out.c (dwarf2out_imported_module_or_decl_1): Handle
CONST_DECL only if is_fortran, is_ada, or is_dlang.
gcc/testsuite/ChangeLog:
PR d/98067
* gdc.dg/debug/debug.exp: New test.
* gdc.dg/debug/dwarf2/dwarf2.exp: New test.
* gdc.dg/debug/dwarf2/imports/pr98067.d: New test.
* gdc.dg/debug/dwarf2/langdw2.d: New test.
* gdc.dg/debug/dwarf2/langdw3.d: New test.
* gdc.dg/debug/dwarf2/pr98067.d: New test.
* gdc.dg/debug/trivial.d: New test.
Jakub Jelinek [Sat, 19 Dec 2020 11:46:40 +0000 (12:46 +0100)]
bswap: Fix up a thinko with empty CONSTRUCTORs [PR98378]
The code I've added recently in find_bswap_or_nop for VECTOR CONSTRUCTORs
is missing punt on an important case - namely empty CONSTRUCTORs, because in that
case the loop will not initialize *n and the code after the loop will then
use the uninitialized structure.
2020-12-19 Jakub Jelinek <jakub@redhat.com>
PR tree-optimization/98378
* gimple-ssa-store-merging.c (find_bswap_or_nop): Punt if CONSTRUCTOR
has no elements.
Jakub Jelinek [Fri, 18 Dec 2020 20:43:20 +0000 (21:43 +0100)]
openmp: Don't optimize shared to firstprivate on task with depend clause
The attached testcase is miscompiled, because we optimize shared clauses
to firstprivate when task body can't modify the variable even when the
task has depend clause. That is wrong, because firstprivate means the
variable will be copied immediately when the task is created, while with
depend clause some other task might change it later before the dependencies
are satisfied and the task should observe the value only after the change.
2020-12-18 Jakub Jelinek <jakub@redhat.com>
* gimplify.c (struct gimplify_omp_ctx): Add has_depend member.
(gimplify_scan_omp_clauses): Set it to true if OMP_CLAUSE_DEPEND
appears on OMP_TASK.
(gimplify_adjust_omp_clauses_1, gimplify_adjust_omp_clauses): Force
GOVD_WRITTEN on shared variables if task construct has depend clause.
aarch64: SVE: ICE in expand_direct_optab_fn [PR98177]
Problem comes from using the wrong interface to get the index type for a
COND_REDUCTION. For fixed-length SVE we get a V2SI (a 64-bit Advanced
SIMD vector) instead of a VNx2SI (an SVE vector that stores SI elements
in DI containers).
Credits to Richard Sandiford for pointing out the issue's root cause.
Original PR snippet proposed to reproduce issue was only causing ICE for C++
compiler (see pr98177-1 test cases). I've slightly modified original
snippet in order to reproduce issue on both C and C++ compilers. These
are pr98177-2 test cases.
gcc/ChangeLog:
PR target/98177
* tree-vect-loop.c (vect_create_epilog_for_reduction): Use
get_same_sized_vectype to obtain index type.
(vectorizable_reduction): Likewise.
gcc/testsuite/ChangeLog:
PR target/98177
* g++.target/aarch64/sve/pr98177-1.C: New test.
* g++.target/aarch64/sve/pr98177-2.C: New test.
* gcc.target/aarch64/sve/pr98177-1.c: New test.
* gcc.target/aarch64/sve/pr98177-2.c: New test.
Patrick Palka [Fri, 18 Dec 2020 18:01:49 +0000 (13:01 -0500)]
libstdc++: Fix mistake in PR98374 change [PR98377]
The #ifdef RADIXCHAR directive should be moved one line up so that it
also guards the outer if statement, or else when RADIXCHAR is not
defined the outer if statement will end up nonsensically guarding the
declaration of output_length_upper_bound a few lines below it.
Patrick Palka [Fri, 18 Dec 2020 16:52:24 +0000 (11:52 -0500)]
libstdc++: Fix build failure on AArch64 ILP32 [PR98370]
This should fix a build failure on AArch64 ILP32 due to int32_t mapping
to long int instead of int on this platform, which causes type deduction
to fail in the below call to std::max as reported in the PR.
libstdc++-v3/ChangeLog:
PR libstdc++/98370
* src/c++17/floating_to_chars.cc (__floating_to_chars_shortest):
Provide explicit template arguments to the call to std::max.
Patrick Palka [Fri, 18 Dec 2020 16:52:17 +0000 (11:52 -0500)]
libstdc++: Fix build failure due to missing <langinfo.h> [PR98374]
This should fix a build failure on Windows which lacks <langinfo.h>,
from which we use nl_langinfo() to obtain the radix character of the
current locale. (We can't use the more portable localeconv() from
<clocale> to obtain the radix character of the current locale here
because it's not thread-safe, unfortunately.)
This change means that on Windows and other such platforms, we'll just
always assume the radix character used by printf is '.' when formatting
a long double through it.
libstdc++-v3/ChangeLog:
PR libstdc++/98374
* src/c++17/floating_to_chars.cc: Guard include of <langinfo.h>
with __has_include.
(__floating_to_chars_precision) [!defined(RADIXCHAR)]: Don't
attempt to obtain the radix character of the current locale,
just assume it's '.'.
Patrick Palka [Fri, 18 Dec 2020 16:52:14 +0000 (11:52 -0500)]
libstdc++: Check FE_TONEAREST is defined before using it
We need to test that FE_TONEAREST is defined before we may use it along
with fegetround/fesetround to adjust the floating-point rounding mode.
This fixes a build failure with older versions of newlib.
aarch64: Extend aarch64-autovec-preference==2 to 128-bit SVE
When compiling with -msve-vector-bits=128, aarch64_preferred_simd_mode
would pass the same vector width to aarch64_simd_container_mode for
both SVE and Advanced SIMD, and so Advanced SIMD would always “win”.
This patch instead makes it choose directly between SVE and Advanced
SIMD modes, so that aarch64-autovec-preference==2 and
aarch64-autovec-preference==4 work for this configuration.
(aarch64-autovec-preference shouldn't affect aarch64_simd_container_mode
because that would have an ABI impact for things like GNU vectors.)
gcc/
* config/aarch64/aarch64.c (aarch64_preferred_simd_mode): Use
aarch64_full_sve_mode and aarch64_vq_mode directly, instead of
going via aarch64_simd_container_mode.
projects / thirdparty / gcc.git / log
