Two simd tests FAIL on Solaris, both SPARC and x86:
FAIL: experimental/simd/standard_abi_usable.cc -msse2 -O2 -Wno-psabi (test for excess errors)
FAIL: experimental/simd/standard_abi_usable_2.cc -msse2 -O2 -Wno-psabi (test for excess errors)
This happens because the simd headers use identifiers documented in the
libstdc++ manual as reserved by system headers.
Fixed as follows, tested on i386-pc-solaris2.11, sparc-sun-solaris2.11,
and x86_64-pc-linux-gnu.
2021-02-01 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
libstdc++-v3:
* include/experimental/bits/simd.h: Replace reserved _X, _B by
_Xp, _Bp.
* include/experimental/bits/simd_builtin.h: Likewise.
* include/experimental/bits/simd_x86.h: Likewise.
gcc/
PR target/99234
* config/i386/i386.c (ix86_compute_frame_layout): For a SEH target,
point back the hard frame pointer to its default location when the
frame is larger than SEH_MAX_FRAME_SIZE.
As the testcase shows, the
(define_peephole2
[(set (match_operand 0 "sse_reg_operand")
(match_operand 1 "sse_reg_operand"))
(set (match_dup 0)
(match_operator 3 "commutative_operator"
[(match_dup 0)
(match_operand 2 "memory_operand")]))]
peephole2 can for AVX512VL without AVX512BW (I guess it is a hyphothetical
CPU, but unfortunately they are separate CPUID bits and we have separate
options for them) turn something that is valid without that peephole2
into something that is invalid (and in this case ICEs).
The problem is that the vpadd[bw], vpmullw, vpmin[su][bw] and vpmax[su][bw]
instructions require both AVX512BW and AVX512VL when they have
16-byte or 32-byte operands. If operands[0] is %[xy]mm0 .. %[xy]mm15
but operands[1] is %[xy]mm16 .. %[xy]mm31, then before we have
a vector move which uses vmovdqa{32,64} and doesn't need AVX512BW,
AVX512VL is I think implied from HARD_REGNO_MODE_OK only supporting
V{16Q,32Q,8H,16H}imode in EXT_REX_SSE_REGNO_P regs with AVX512VL, and then
we have a commutative operation with that %[xy]mm0 .. %[xy]mm15 destination
and one source and a memory operand, so VEX encoded operation.
And, the peephole2 wants to replace it with a load into the destination
register from memory (ok) and then the commutative arith instruction.
But that needs EVEX encoding because of the high register and so requires
AVX512BW which might not be enabled.
The exception is and/ior/xor, because the hw doesn't have
vp{and,or,xor}{b,w} instructions at all, it uses vp{and,or,xor}d instead
and that of course doesn't need AVX512BW.
BTW, there are other bugs I need to look at, while the vp{min,max}ub with
16-byte operands instruction properly requires avx512bw for v constraints
and otherwise uses x, e.g. the vpadd[bw] etc. instructions don't.
I'll try to handle that incrementally later this week.
2021-03-03 Jakub Jelinek <jakub@redhat.com>
PR target/99321
* config/i386/predicates.md (logic_operator): New define_predicate.
* config/i386/i386.md (mov + mem using comm arith peephole2):
Punt if operands[1] is EXT_REX_SSE_REGNO_P, AVX512BW is not enabled
and the inner mode is [QH]Imode.
* gcc.target/i386/pr99321.c: New test.
build_va_arg calls the middle-end mark_addressable, which e.g. requires that
cfun is non-NULL. The following patch calls instead c_common_mark_addressable_vec
which is the c-family variant similarly to the FE c_mark_addressable and
cxx_mark_addressable, except that it doesn't error on addresses of register
variables. As the taking of the address is artificial for the .VA_ARG
ifn and when that is lowered goes away, it is similar case to the vector
subscripting for which c_common_mark_addressable_vec has been added.
2021-03-03 Jakub Jelinek <jakub@redhat.com>
PR c/99324
* c-common.c (build_va_arg): Call c_common_mark_addressable_vec
instead of mark_addressable. Fix a comment typo -
neutrallly -> neutrally.
* gcc.c-torture/compile/pr99324.c: New test.
As mentioned in the PR, riscv* only supports .uleb128 with constant
arguments, doesn't support difference of two labels because of aggressive
linker relaxations. But I bet various other targets, especially those not
using GNU assembler, might suffer from the same problem.
As the FIXME comment in output_loc_list indicates, we ICE on
-gsplit-dwarf on those targets whenever we need .debug_loclists, because
we only emit DW_LLE_startx_length which requires working .uleb128 delta
of 2 code section labels. We can't use DW_LLE_base_addressx
once followed by DW_LLE_offset_pair either because the latter suffers
from the same issue - need .uleb128 difference of code section labels
(and in that case not just for the second operand but also for the first).
So, this patch implements what the comment said and emits DW_LLE_startx_endx
instead, which wastes more space in .debug_addr, but will work.
Bootstrapped/regtested on x86_64-linux and i686-linux and as written in the
PR, Jim has tested it on riscv*linux. Ok for trunk?
BTW, for HAVE_AS_LEB128 -gdwarf-5 -gsplit-dwarf, maybe we should consider
instead of always emitting DW_LLE_startx_length do all the optimizations
that we do for HAVE_AS_LEB128 -gdwarf-5, or at least a subset of them.
For !have_multiple_function_sections, we in that case emit just
DW_LLE_offset_pair (that can certainly be a win for small TUs, we wouldn't
need any .debug_addr entry in that case; on the other side, just using
DW_LLE_offset_pair can be harmful for very large TUs especially if the
loclist has many entries, emitting in that case a single DW_LLE_base_address
or for -gsplit-dwarf DW_LLE_base_addressx followed by DW_LLE_offset_pair
might be much smaller), and for have_multiple_function_sections figuring
out if DW_LLE_base_address followed by DW_LLE_offset_pair entries
or DW_LLE_start_length is bettter. So perhaps a middle-ground for
-gsplit-dwarf would be to always do the have_multiple_function_sections
behavior, i.e. DW_LLE_base_addressx followed by DW_LLE_offset_pair vs.
DW_LLE_startx_length decisions based on the ranges and their counts.
And perhaps dwz could optimize afterwards, on linked binaries or shared
libraries it knows all the offsets and could figure out optimal DW_LLE_*.
2021-03-03 Jakub Jelinek <jakub@redhat.com>
PR debug/99090
* dwarf2out.c (dw_loc_list_struct): Add end_entry member.
(new_loc_list): Clear end_entry.
(output_loc_list): Only use DW_LLE_startx_length for -gsplit-dwarf
if HAVE_AS_LEB128, otherwise use DW_LLE_startx_endx. Fix comment
typo.
(index_location_lists): For dwarf_version >= 5 without HAVE_AS_LEB128,
initialize also end_entry.
fixup_partitions sometimes changes some basic blocks from hot partition to
cold partition, in particular if after unreachable block removal or other
optimizations a hot partition block is dominated by cold partition block(s).
It fixes up the edges and jumps on those edges, but when after reorder
blocks and in rtl (non-cfglayout) mode that is clearly not enough, because
it keeps the block order the same and so we can end up with more than
1 hot/cold section transition in the same function.
So, this patch fixes that up too.
2021-03-03 Jakub Jelinek <jakub@redhat.com>
PR target/99085
* cfgrtl.c (fixup_partitions): When changing some bbs from hot to cold
partitions, if in non-layout mode after reorder_blocks also move
affected blocks to ensure a single partition transition.
* gcc.dg/graphite/pr99085.c: New test.
The patch for PR92372 made us start warning about a flatten attribute on an
alias. But in the case of C++ 'tor base/complete variants, the user didn't
create the alias. If the alias target also has the attribute, the alias
points to a flattened function, so we shouldn't warn.
gcc/ChangeLog:
PR c++/96078
* cgraphunit.c (process_function_and_variable_attributes): Don't
warn about flatten on an alias if the target also has it.
* cgraph.h (symtab_node::get_alias_target_tree): New.
gcc/testsuite/ChangeLog:
PR c++/96078
* g++.dg/ext/attr-flatten1.C: New test.
It's no longer necessary as file 0 is now set up in all cases.
* dwarf.c (read_line_program): Don't special case file 0.
(read_function_entry): Likewise.
This patch adds missing periods to the symbol name for TLS helper functions.
gcc/ChangeLog:
* config/rs6000/rs6000.md (tls_get_tpointer_internal): Prepend
period to symbol name.
(tls_get_addr_internal<mode>): Same.
The outputs.exp tests check what temporary files are created
and left behind with e.g. -save-temps.
Additional files are created in presence of @file option.
Adding an -I or -L option causes *another* temporary file to
appear. I take it that's deliberate, as there are tests for
that behavior.
For native testing, the default test-framework baseboard
file unix.exp doesn't add any -I or -L options and all tests
pass. For a newlib target however, you'll have a couple of
-L options (see the nopts handling in outputs.exp), leading
to:
Running /x/gcc/gcc/testsuite/gcc.misc-tests/outputs.exp ...
FAIL: outputs exe savetmp namedb: extra
outputs.args.1
FAIL: outputs exe savetmp named2: extra
outputs.args.1
FAIL: outputs exe savetmp named2: extra
outputs.args.3
FAIL: outputs lto sing unnamed: extra
a.args.1
The failing tests are among the actual tests that check the
behavior of @file, and are confused by the additional -L.
Identify presence of -I or -L from the test framework and
skip those tests.
Tested cris-elf and x86_64-pc-linux-gnu.
gcc/testsuite:
* gcc.misc-tests/outputs.exp: Skip @file -save-temps
tests if target test-framework has -L or -I options.
The embed spec allows for //go:embed to be used in files that
underscore-import package "embed". This is useful for embeds to
[]byte and string vars because the embed.FS type may not be referenced
if those are the only types of embeds in a file. Because the compiler
previously checked whether there were any aliases to the embed
package to decide if //go:embed could be used, it would reject
files with only underscore imports of embed. Instead, record
whether the embed import is encountered at all, similar to what
is done with unsafe, to decide whether //go:embed is allowed.
Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/297553
PR c/99323 describes an ICE due to a failed assertion deep inside the
fix-it printing machinery, where the fix-it hints on one line have not
been properly sorted in layout's constructor.
The underlying issue occurs when multiple fix-it hints affect a line
wider that LINE_MAP_MAX_COLUMN_NUMBER, where the location_t values for
characters after that threshold fall back to having column zero.
It's not meaningful to try to handle fix-it hints without column
information, so this patch rejects them as they are added to the
rich_location, falling back to the "no fix-it hints on this diagnostic"
case, fixing the crash.
gcc/ChangeLog:
PR c/99323
* diagnostic-show-locus.c
(selftest::test_one_liner_many_fixits_2): Fix accidental usage of
column 0.
gcc/testsuite/ChangeLog:
PR c/99323
* gcc.dg/pr99323-1.c: New test.
* gcc.dg/pr99323-2.c: New test.
libcpp/ChangeLog:
PR c/99323
* line-map.c (rich_location::maybe_add_fixit): Reject fix-it hints
at column 0.
The -gsplit-dwarf changes came a few months after .debug_macro
and the r0-120109 changes just changed the 2nd operand of
DW_MACRO_GNU_{define,undef}_indirect from the usual .debug_str
section offset argument to leb128 index into .debug_str_offsets
without changing the opcodes.
DWARF5 standardized different opcodes for those, but GCC hasn't been changed
yet for that.
This patch starts using DW_MACRO_define_strx and DW_MACRO_undef_strx
instead of DW_MACRO_define_strp and DW_MACRO_undef_strp when -gsplit-dwarf
-gdwarf-5 -g3. I'm not sure what to do if anything with the -gdwarf-4
-gsplit-dwarf -g3 -gno-strict-dwarf case, we've been emitting it that way
for 8 years and it is an extension, so presumably the consumers that cared
have already hacks to handle DW_MACRO_GNU_{define,undef}_indirect
differently in .debug_macro 4 sections depending on if it is
.debug_macro.dwo or .debug_macro.
Another change the patch does is that it will use
DW_MACRO_{define,undef}_str{p,x} even with -gdwarf-5 -gstrict-dwarf -g3,
for DWARF 4 we were doing that only for -gno-strict-dwarf as we've emitted
.debug_macro section only in that case.
2021-03-02 Jakub Jelinek <jakub@redhat.com>
PR debug/99319
* dwarf2out.c (output_macinfo_op): Use DW_MACRO_*_str* even with
-gdwarf-5 -gstrict-dwarf. For -gsplit-dwarf -gdwarf-5 use
DW_MACRO_*_strx instead of DW_MACRO_*_strp. Handle
DW_MACRO_define_strx and DW_MACRO_undef_strx.
(save_macinfo_strings): Use DW_MACRO_*_str* even with
-gdwarf-5 -gstrict-dwarf. Handle DW_MACRO_define_strx and
DW_MACRO_undef_strx.
This is a regression present on the mainline and 10 branch, where we fail
to make the bounds explicit for the return value of a function returning
an unconstrained array of a limited record type.
gcc/ada/
PR ada/99095
* sem_ch8.adb (Check_Constrained_Object): Restrict again the special
optimization for limited types to non-array types except in the case
of an extended return statement.
gcc/testsuite/
* gnat.dg/limited5.adb: New test.
The issue is reported by Clang:
warning: private field 'm_engine' is not used [-Wunused-private-field]
gcc/analyzer/ChangeLog:
* diagnostic-manager.cc (diagnostic_manager::emit_saved_diagnostics):
Do not pass engine.
This fixes the way we check satisfaction of constraints on placeholder
types in various deduction contexts, and in particular when the
constraint is dependent.
Firstly, when evaluating the return type requirement of a compound
requirement, we currently substitute the outer template arguments into
the constraint before checking satisfaction. But we should instead be
passing in the complete set of template arguments to satisfaction and
not do a prior separate substitution. Our current approach leads to us
incorrectly rejecting the testcase concepts-return-req2.C below.
Secondly, when checking the constraints on a placeholder variable or
return type, we don't consider the template arguments of the enclosing
context at all. This leads to bogus errors during satisfaction when the
constraint is dependent as in the testcase concepts-placeholder3.C
below.
In order to fix these two issues, we need to be able to normalize the
constraints on a placeholder 'auto' on demand, which in turn requires us
to know the template parameters that were in scope where the 'auto' was
introduced. This information currently doesn't seem to be easily available
when we need it, so this patch turns PLACEHOLDER_TYPE_CONSTRAINTS into a
TREE_LIST whose TREE_PURPOSE additionally holds the value of
current_template_parms whence a constrained 'auto' was formed.
This patch also removes some seemingly wrong handling of placeholder
type arguments from tsubst_parameter_mapping. The code doesn't trigger
with the example used in the comments, because type_uses_auto doesn't
look inside non-deduced contexts such as the operand of decltype. And
the call to do_auto_deduction seems confused because if 'arg' is a type,
then so is 'parm', and therefore 'init' too is a type, but
do_auto_deduction expects it to be an expression. Before this patch,
this code was dead (as far as our testsuite can tell), but now it breaks
other parts of this patch, so let's remove it.
gcc/cp/ChangeLog:
PR c++/96443
PR c++/96960
* constraint.cc (type_deducible_p): Don't substitute into the
constraints, and instead just pass 'args' to do_auto_deduction
as the outer template arguments.
(tsubst_parameter_mapping): Remove confused code for handling
placeholder type arguments.
(normalize_placeholder_type_constraint): Define.
(satisfy_constraint_expression): Use it to handle placeholder
'auto' types.
* cp-tree.h (PLACEHOLDER_TYPE_CONSTRAINTS_INFO): Define.
(PLACEHOLDER_TYPE_CONSTRAINTS): Redefine in terms of the above.
* pt.c (tsubst) <case TEMPLATE_TYPE_PARM>: Use
PLACEHOLDER_TYPE_CONSTRAINTS_INFO instead.
(make_constrained_placeholder_type): Set
PLACEHOLDER_TYPE_CONSTRAINTS_INFO instead.
(do_auto_deduction): Clarify comments about the outer_targs
parameter. Rework satisfaction of a placeholder type constraint
to pass in the complete set of template arguments directly to
constraints_satisfied_p.
(splice_late_return_type): Use PLACEHOLDER_TYPE_CONSTRAINTS_INFO
instead. Also rebuild the the constraint info on the new auto.
gcc/testsuite/ChangeLog:
PR c++/96443
PR c++/96960
* g++.dg/concepts/abbrev9.C: New test.
* g++.dg/cpp2a/concepts-lambda15.C: New test.
* g++.dg/cpp2a/concepts-placeholder3.C: New test.
* g++.dg/cpp2a/concepts-return-req2.C: New test.
* g++.dg/cpp2a/concepts-ts1.C: Add dg-bogus directive to the
call to f15 that we expect to accept.
This sets up the functionality for controlling the initial set of
template parameters to pass to normalization when dealing with a
constraint-expression that is not associated with some constrained
declaration, for instance when normalizing a nested requirement of a
requires expression, or the constraints on a placeholder type.
The main new ingredient here is the data member norm_info::initial_parms
which can be set by callers of the normalization routines to communicate
the in-scope template parameters for the supplied constraint-expression,
rather than always falling back to using current_template_parms.
This patch then uses this functionality in our handling of nested
requirements so that we can delay normalizing them until needed for
satisfaction. We currently immediately normalize nested requirements at
parse time, where we have the necessary template context, and cache the
normal form in their TREE_TYPE node. With this patch, we now delay
normalization until needed (as with other constraint expressions), and
instead store the current value of current_template_parms in their
TREE_TYPE node (which we use to restore the template context at
normalization time).
In the subsequent patch, this functionality will also be used to
normalize placeholder type constraints during auto deduction.
gcc/cp/ChangeLog:
* constraint.cc (build_parameter_mapping): Rely on the caller to
determine the in-scope template parameters.
(norm_info::norm_info): Delegate the tsubst_flags_t constructor
to the two-parameter constructor. In the two-parameter
constructor, fold in the definition of make_context, set
initial_parms appropriately, and don't set the now-removed
orig_decl member.
(norm_info::make_context): Remove, now that its only use is
inlined into the caller.
(norm_info::update_context): Adjust call to
build_parameter_mapping to pass in the relevant set of in-scope
template parameters.
(norm_info::ctx_parms): Define this member function.
(norm_info::context): Initialize to NULL_TREE.
(norm_info::orig_decl): Remove this data member.
(norm_info::initial_parms): Define this data member.
(normalize_atom): Adjust call to build_parameter_mapping to pass
in the relevant set of in-scope template parameters. Use
info.initial_parms instead of info.orig_decl.
(normalize_constraint_expression): Take a norm_info object
instead of a bool. Cache the result of normalization.
(tsubst_nested_requirement): Call satisfy_constraint_expression
instead of satisfy_constraint, so that we normalize on demand.
(satisfy_constraint_expression): Handle a NESTED_REQ argument.
Adjust call to normalize_constraint_expression.
(finish_nested_requirement): Set the TREE_TYPE of the NESTED_REQ
to current_template_parms.
(diagnose_nested_requirements): Go through
satisfy_constraint_expression, as with tsubst_nested_requirement.
The patches that follow revealed some latent type canonicalization
issues during normalization/satisfaction, due to normalization of
some constraint-expressions now being performed in different contexts
and more frequently (since the normalization caches get frequently
emptied in a checking compiler). The issues are:
1. In tsubst_parameter_mapping, we canonicalize the arguments of a
substituted TYPE_ARGUMENT_PACK only if the argument we started with
wasn't a TYPE_ARGUMENT_PACK. We should canonicalize a substituted
TYPE_ARGUMENT_PACK regardless of what we started with.
2. We currently set DECL_CONTEXT and CONSTRAINT_VAR_P on each of the
parameters introduced in a requires expression _after_ we're done
processing the requirements. But meanwhile we may have already
built and computed the canonical form of a type that uses one of
these PARM_DECLs (as say an operand to decltype). But the canonical
form depends on the result of cp_tree_equal, which in turn depends on
the value of CONSTRAINT_VAR_P and DECL_CONTEXT. So we must set these
fields earlier, before processing requirements.
3. In do_auto_deduction, we use the result of finish_decltype_type later
as a template argument, so we should canonicalize the result too.
(While we're here, we should pass 'complain' to finish_decltype_type,
which fixes the testcase auto1.C below.)
gcc/cp/ChangeLog:
* constraint.cc (tsubst_parameter_mapping): Canonicalize the
arguments of a substituted TYPE_ARGUMENT_PACK even if we've
started with a TYPE_ARGUMENT_PACK.
(finish_requires_expr): Don't set DECL_CONTEXT and
CONSTRAINT_VAR_P on each of the introduced parameters here.
* parser.c (cp_parser_requirement_parameter_list): Instead set
these fields earlier, here.
* pt.c (do_auto_deduction): Canonicalize the result of
do_auto_deduction. Pass 'complain' to finish_decltype_type.
gcc/testsuite/ChangeLog:
* g++.dg/cpp1z/auto1.C: New test.
Since we no longer partially instantiate REQUIRES_EXPRs, we don't need
to rebuild its requirements during tsubst_requires_expr.
gcc/cp/ChangeLog:
* constraint.cc (tsubst_simple_requirement): Just return
boolean_true_node on success.
(tsubst_type_requirement): Likewise.
(tsubst_compound_requirement): Likewise.
(tsubst_nested_requirement): Likewise.
(tsubst_requirement_body): Remove.
(check_constaint_variables): Rename to ...
(check_constraint_variables): ... this.
(tsubst_constraint_variables): Adjust.
(tsubst_requires_expr): Fold tsubst_requirement_body into here.
The Wstringop-overflow-25.c testcase doesn't emit one of the expected
warnings on targets that don't do short curcuiting due to target costs
(or e.g. with --param=logical-op-non-short-circuit=0 on all targets).
The problem is that only reassoc2 optimizes:
_49 ={v} unsigned_value_source;
if (_49 == 0)
goto <bb 7>; [50.00%]
else
goto <bb 5>; [50.00%]
<bb 5> [local count: 536870913]:
if (_49 > 2)
goto <bb 7>; [50.00%]
else
goto <bb 6>; [50.00%]
<bb 6> [local count: 268435457]:
_53 = _49 + 1;
into:
_49 ={v} unsigned_value_source;
_48 = _49 + 18446744073709551615;
_1 = _48 > 1;
if (_1 != 0)
goto <bb 6>; [50.00%]
else
goto <bb 5>; [50.00%]
<bb 5> [local count: 268435457]:
_53 = _49 + 1;
(but, note the _1 = _48 > 1; if (_1 != 0)),
then dom3 is run and because of that if (_1 != 0) vs. if (_48 > 1) doesn't
register edge asserts for _48 and _49) and so we don't get
SSA_NAME_RANGE_INFO for _53 (and ditto for vrp2) and only afterwards comes
forwprop4 that canonicalizes it to if (_48 > 1). While with
--param=logical-op-non-short-circuit=1 it is already reassoc1 that optimizes
it and forwprop3 that propagates it, so we have on the SSA_NAME
corresponding to _53 above SSA_NAME_RANGE_INFO and during expansion we warn.
The following patch fixes it by handling those not yet propagated
comparisons into GIMPLE_COND in register_edge_assert_for. We already
have all the infrastructure there to handle the --param=logical-op-non-short-circuit=1
| and &s.
2021-03-02 Jakub Jelinek <jakub@redhat.com>
PR middle-end/95757
* tree-vrp.c (register_edge_assert_for): Remove superfluous ()s around
condition. Call register_edge_assert_for_1 for == 0, != 0, == 1 and
!= 1 comparisons if name is lhs of a comparison.
The prototype of __sprintfkf in _sprintfkf.h did not match the function in
_sprintfkf.c. This patch fixes the prototype. I also included the
_sprintfkf.h file in _sprintfkf.c to make sure the prototype is correct and to
eliminate a warning about declaring the function without a previous
declaration.
libgcc/
2021-03-01 Michael Meissner <meissner@linux.ibm.com>
* config/rs6000/_sprintfkf.h (__sprintfkf): Fix prototype to match
the function.
* config/rs6000/_sprintfkf.c: Include _sprintfkf.h.
Build_Discriminated_Subtype may be invoked on a E_Record_Type_With_Private,
in which case it builds a E_Record_Subtype_With_Private which does not have
the Cloned_Subtype field.
gcc/ada/
PR ada/99020
* sem_ch3.adb (Build_Discriminated_Subtype): Set the Cloned_Subtype
only if the type is not private.
This address one of the more long-standing and serious regressions
for Darwin. GCC emits unwind code by default on the assumption that
the unwinder will be (of have the same capability) as the one in the
current libgcc_s. For Darwin platforms, this is not the case - some
of them are based on the libgcc_s from GCC-4.2.1 and some are using
the unwinder provided by libunwind (part of the LLVM project). The
latter implementation has gradually adopted a section that deals with
GNU unwind.
The most serious problem for some of the platform versions is in
handling DW_CFA_remember/restore_state pairs. The DWARF description
talks about these in terms of saving/restoring register rows; this is
what GCC originally did (and is what the unwinders do for the Darwin
versions based on libgcc_s).
However, in r118068, this was changed so that not only the registers
but also the current frame address expression were saved. The unwind
code assumes that the unwinder will do this; some of Darwin's unwinders
do not, leading to lockups etc. To date, the only solution has been
to replace the system libgcc_s with a newer one which is not a viable
solution for many end-users (since that means overwritting the one
provided with the system installation).
The fix here provides a target hook that allows the target to specify
that the CFA should be reinstated after a DW_CFA_restore. This fixes
the issue (and also the closed WONTFIX of 44107).
(As a matter of record, it also fixes reported Java issues if
backported to GCC-5).
gcc/ChangeLog:
PR target/44107
PR target/48097
* config/darwin-protos.h (darwin_should_restore_cfa_state): New.
* config/darwin.c (darwin_should_restore_cfa_state): New.
* config/darwin.h (TARGET_ASM_SHOULD_RESTORE_CFA_STATE): New.
* doc/tm.texi: Regenerated.
* doc/tm.texi.in: Document TARGET_ASM_SHOULD_RESTORE_CFA_STATE.
* dwarf2cfi.c (connect_traces): If the target requests, restore
the CFA expression after a DW_CFA_restore.
* target.def (TARGET_ASM_SHOULD_RESTORE_CFA_STATE): New hook.
When we read in a class definition, we use fixup_type_variants to
propagate the now-completed fields of the class's TYPE to other
variants. Unfortunately that doesn't propagate all of them, and in
this case we had a typedef to an (incomplete) instantiation. That
typedef ended up with a VOIDmode, which blew up gimple expansion as
the type itself isn't VOID. Without modules, that information is
propagated in finalize_type_size when laying out the class. But that
doesn't happen with stream-in -- we already know the layout. There is
already some overlap between the two functions, now there's a bit
more. In fixup_type_variants, I pay attention to the TYPE_NAME to
decide whether to override a user's TYPE_ALIGN -- variants of the
main-variant typedef just copy the main-variant. Other variants
recalculate. Overaligning is still permitted.
I also added a TYPE_ALIGN_RAW accessor, and fixed a bug in the
alignment streaming I noticed. I did not refactor TYPE_ALIGN beyond
using the new accessor. (It could be written as ((1 << align_raw) >>
1), rather than use the conditional.)
PR c++/99294
gcc/
* tree.h (TYPE_ALIGN_RAW): New accessor.
(TYPE_ALIGN): Use it.
gcc/cp/
* class.c (fixup_type_variants): Propagate mode, precision,
alignment & emptiness.
* module.cc (trees_out::type_node): Use TYPE_ALIGN_RAW.
(trees_in::tree_node): Rematerialize alignment here.
gcc/testsuite/
* g++.dg/modules/pr99294.h: New.
* g++.dg/modules/pr99294_a.C: New.
* g++.dg/modules/pr99294_b.C: New.
This wrong-code PR for the C++ compiler on x86-64/Windows is a regression
in GCC 9 and later, but the underlying issue has probably been there since
SEH was implemented and is exposed by this comment in config/i386/winnt.c:
/* SEH records saves relative to the "current" stack pointer, whether
or not there's a frame pointer in place. This tracks the current
stack pointer offset from the CFA. */
HOST_WIDE_INT sp_offset;
That's not what the (current) Microsoft documentation says; instead it says:
/* SEH records offsets relative to the lowest address of the fixed stack
allocation. If there is no frame pointer, these offsets are from the
stack pointer; if there is a frame pointer, these offsets are from the
value of the stack pointer when the frame pointer was established, i.e.
the frame pointer minus the offset in the .seh_setframe directive. */
That's why the implementation is correct only under the condition that the
frame pointer be established *after* the fixed stack allocation; as a matter
of fact, that's clearly the model underpinning SEH, but is the opposite of
what is done e.g. on Linux.
However the issue is mostly papered over in practice because:
1. SEH forces use_fast_prologue_epilogue to false, which in turns forces
save_regs_using_mov to false, so the general regs are always pushed when
they need to be saved, which eliminates the offset computation for them.
2. As soon as a frame is larger than 240 bytes, the frame pointer is fixed
arbitrarily to 128 bytes above the stack pointer, which of course requires
that it be established after the fixed stack allocation.
So you need a small frame clobbering one of the call-saved XMM registers in
order to generate wrong SEH unwind info.
The attached fix makes sure that the frame pointer is always established
after the fixed stack allocation by pointing it at or below the lowest used
register save area, i.e. the SSE save area, and removing the special early
saves in the prologue; the end result is a uniform prologue sequence for
SEH whatever the frame size. And it avoids a discrepancy between cases
where the number of saved general regs is even and cases where it is odd.
gcc/
PR target/99234
* config/i386/i386.c (ix86_compute_frame_layout): For a SEH target,
point the hard frame pointer to the SSE register save area instead
of the general register save area. Perform only minimal adjustment
for small frames if it is initially not correctly aligned.
(ix86_expand_prologue): Remove early saves for a SEH target.
* config/i386/winnt.c (struct seh_frame_state): Document constraint.
gcc/testsuite/
* g++.dg/eh/seh-xmm-unwind.C: New test.
This is part of the requirement of [dcl.fct.def.coroutine]/15.
In addition to promise final_suspend() calls, the following cases must
also be noexcept as per discussion in PR95616.
- finalSuspendObj.operator co_await()
- finalSuspendAwaiter.await_ready()
- finalSuspendAwaiter.await_suspend()
- finalSuspendAwaiter.await_resume()
- finalSuspedObj destructor
- finalSuspendAwaiter destructor
Fixed for missing cases in the testsuite as a prerequisite to fixing
PR95616.
gcc/testsuite/ChangeLog:
* g++.dg/coroutines/pr94879-folly-1.C: Make final suspend
expression components noexcept.
* g++.dg/coroutines/pr94883-folly-2.C: Likewise.
* g++.dg/coroutines/pr95345.C: Likewise.
Some testcase for a change I was testing had an unnecessary ; at the end of
a line that caused a test failure on the 10 branch, so I mistakenly removed
it from this one. Oops.
gcc/testsuite/ChangeLog:
* g++.dg/ext/attr-lambda3.C: Add missing ;
As reported in the PR, some diagnostic messages need wording improvements.
2021-02-28 Jakub Jelinek <jakub@redhat.com>
PR fortran/99303
* openmp.c (gfc_omp_requires_add_clause): Fix up diagnostic message
wordings.
(resolve_omp_clauses): Likewise.
The diagnostic had " " in it instead of just " ".
2021-02-28 Jakub Jelinek <jakub@redhat.com>
PR fortran/99300
* frontend-passes.c (doloop_code): Replace double space in diagnostics
with a single space.
The PR is about a typo in handle_malloc_attribute diagnostic message,
but grepping around I found many other cases and while fixing those I've
noticed a couple of other typos.
2021-02-28 Jakub Jelinek <jakub@redhat.com>
PR c/99304
* ipa.c (symbol_table::remove_unreachable_nodes): Fix a comment
typo - referneced -> referenced.
* tree.c (component_ref_size): Fix comment typo -
refernce -> reference.
* tree-ssa-alias.c (access_path_may_continue_p): Fix comment typo -
traling -> trailing.
(aliasing_component_refs_p): Fix comment typos -
refernce -> reference and refernece -> reference and
traling -> trailing.
(nonoverlapping_refs_since_match_p): Fix comment typo -
referneces -> references.
* doc/invoke.texi (--param modref-max-bases): Fix a typo -
referneces -> references.
gcc/c-family/
* c-attribs.c (handle_malloc_attribute): Fix a typo in inform
message - refernced -> referenced. Remove superfluous space before
closing paren of function calls.
gcc/lto/
* lto-symtab.c (lto_symtab_prevailing_virtual_decl): Fix comment
typos - refernced -> referenced and
devirtualizaiton -> devirtualization.
In my 9.3/10 patch for 90333 I allowed attributes between [] and (), and
after the trailing return type, but not in the place that GCC 8 expected
them, and we've gotten several bug reports about that. So let's allow them
there, as well.
gcc/cp/ChangeLog:
PR c++/90333
* parser.c (cp_parser_lambda_declarator_opt): Accept GNU attributes
between () and ->.
gcc/testsuite/ChangeLog:
PR c++/90333
* g++.dg/ext/attr-lambda3.C: New test.
The avoids a contraction and a format diagnostic warning.
gcc/ChangeLog:
* config/host-darwin.c (darwin_gt_pch_use_address): Modify
diagnostic message to avoid use of a contraction and format
warning.
This fails everywhere on Darwin, which does not have support for
symbol aliases. Add a dg-require-alias to UNSUPPORT it.
libgomp/ChangeLog:
* testsuite/libgomp.c-c++-common/pr96390.c: Require alias
support from the target.
The conversions to integer types are explicit, so need to use the
correct type. Converting to uint32_t only works if that is the same type
as unsigned.
libstdc++-v3/ChangeLog:
PR libstdc++/99301
* include/std/chrono (year_month_day::_M_days_since_epoch()):
Convert chrono::month and chrono::day to unsigned before
converting to uint32_t.
