Jakub Jelinek [Sat, 8 Dec 2018 08:58:24 +0000 (09:58 +0100)]
re PR libgomp/87995 (libgomp.c/../libgomp.c-c++-common/cancel-taskgroup-3.c fails consistently after r265930)
PR libgomp/87995
* testsuite/libgomp.c-c++-common/cancel-taskgroup-3.c: Require
tls_runtime effective target.
(t): New threadprivate variable.
(main): Set t in threads which execute iterations of the worksharing
loop. Propagate that to the task after the loop and don't abort
if the current taskgroup hasn't been cancelled.
Paolo Carlini [Fri, 7 Dec 2018 17:48:39 +0000 (17:48 +0000)]
decl2.c (grokbitfield): Use DECL_SOURCE_LOCATION in error messages about bit-fields with function type...
/cp
2018-12-07 Paolo Carlini <paolo.carlini@oracle.com>
* decl2.c (grokbitfield): Use DECL_SOURCE_LOCATION in error messages
about bit-fields with function type, warn_if_not_aligned type, and
static bit-fields; avoid DECL_NAME for unnamed declarations.
/testsuite
2018-12-07 Paolo Carlini <paolo.carlini@oracle.com>
* g++.dg/other/bitfield7.C: New.
* g++.dg/parse/bitfield8.C: Likewise.
* g++.dg/parse/bitfield9.C: Likewise.
* g++.dg/pr53037-4.C: Test the locations too.
Peter Bergner [Fri, 7 Dec 2018 17:33:55 +0000 (17:33 +0000)]
re PR target/87496 (ICE in aggregate_value_p at gcc/function.c:2046)
gcc/
PR target/87496
* config/rs6000/rs6000.c (rs6000_option_override_internal): Disallow
-mabi=ieeelongdouble and -mabi=ibmlongdouble without -mlong-double-128.
Do not error for -mabi=ibmlongdouble and no ISA 2.06 support.
* doc/invoke.texi: Document -mabi=ibmlongdouble and -mabi=ieeelongdouble
require -mlong-double-128.
gcc/testsuite/
PR target/87496
* gcc.target/powerpc/pr87496.c: Rename from this...
* gcc.target/powerpc/pr87496-1.c: ...to this. Update comment.
* gcc.target/powerpc/pr87496-2.c: New test.
* gcc.target/powerpc/pr87496-3.c: New test.
Kyrylo Tkachov [Fri, 7 Dec 2018 17:05:21 +0000 (17:05 +0000)]
[AArch64][2/2] Add sve_width -moverride tunable
On top of the previous patch that implements TARGET_ESTIMATED_POLY_VALUE
and adds an sve_width tuning field to the CPU structs, this patch implements
an -moverride knob to adjust this sve_width field to allow for experimentation.
Again, reminder that this only has an effect when compiling for VLA-SVE that is,
without msve-vector-bits=<foo>. This just adjusts tuning heuristics in the compiler,,
like profitability thresholds for vectorised versioned loops, and others.
It can be used, for example like -moverride=sve_width=256 to set the sve_width
tuning field to 256. Widths outside of the accepted SVE widths [128 - 2048] are rejected
as you'd expect.
* gimple-ssa-evrp-analyze.h (class evrp_range_analyzer): Add
m_update_global_ranges member. Add corresponding argument to ctor.
* gimple-ssa-evrp-analyze.c
(evrp_range_analyzer::evrp_range_analyzer): Add new argument and
initialize m_update_global_ranges.
(evrp_range_analyzer::set_ssa_range_info): Assert that we are
updating global ranges.
(evrp_range_analyzer::record_ranges_from_incoming_edge): Only
update global ranges if explicitly requested.
(evrp_range_analyzer::record_ranges_from_phis): Similarly.
(evrp_range_analyzer::record_ranges_from_stmt): Similarly.
* gimple-ssa-evrp.c (evrp_dom_walker): Pass new argument to
evrp_range_analyzer ctor.
* gimple-ssa-sprintf.c (sprintf_dom_walker): Similarly.
* tree-ssa-dom.c (dom_opt_dom_walker): Similarly.
* gcc.c-torture/builtins/strnlen.x: New file to filter -Og from
options to test.
The hook TARGET_ESTIMATED_POLY_VALUE allows a target to give an estimate for a poly_int run-time value.
It is used exclusively in tuning decisions, things like estimated loop iterations, probabilities etc.
It is not relied on for correctness.
If we know the SVE width implemented in hardware we can make more more
informed decisions in the implementation of TARGET_ESTIMATED_POLY_VALUE,
even when compiling for VLA vectorisation.
This patch adds an sve_width field to our tuning structs and sets it for
the current CPU tunings.
A new value is introduced to the aarch64_sve_vector_bits_enum enum that indicates
that SVE is not available: SVE_NOT_IMPLEMENTED. I set it to the same value as SVE_SCALABLE
so that parts of the aarch64 backend that follow the pattern:
if (vector_width == SVE_SCALABLE)
do_vla_friendly_action ()
else
assume_specific_width_for_correctness ()
continue to work without change, but the CPU tuning structs can use a more
appropriate moniker for indicating the absence of SVE.
This sets sve_width to SVE_NOT_IMPLEMENTED for all cores.
I aim to add an -moverride switch in the next patch that allows a power user to experiment
with different values of it for investigations.
[AArch64][SVE] Remove unnecessary PTRUEs from integer arithmetic
When using the unpredicated immediate forms of MUL, LSL, LSR and ASR,
the rtl patterns would still have the predicate operand we created for
the other forms. This patch splits the patterns after reload in order
to get rid of the predicate, like we already do for WHILE.
2018-12-07 Richard Sandiford <richard.sandiford@arm.com>
gcc/
* config/aarch64/aarch64-sve.md (*mul<mode>3, *v<optab><mode>3):
Split the patterns after reload if we don't need the predicate
operand.
(*post_ra_mul<mode>3, *post_ra_v<optab><mode>3): New patterns.
gcc/testsuite/
* gcc.target/aarch64/sve/pred_elim_2.c: New test.
[AArch64][SVE] Remove unnecessary PTRUEs from FP arithmetic
When using the unpredicated all-register forms of FADD, FSUB and FMUL,
the rtl patterns would still have the predicate operand we created for
the other forms. This patch splits the patterns after reload in order
to get rid of the predicate, like we already do for WHILE.
2018-12-07 Richard Sandiford <richard.sandiford@arm.com>
gcc/
* config/aarch64/iterators.md (SVE_UNPRED_FP_BINARY): New code
iterator.
(sve_fp_op): Handle minus and mult.
* config/aarch64/aarch64-sve.md (*add<mode>3, *sub<mode>3)
(*mul<mode>3): Split the patterns after reload if we don't
need the predicate operand.
(*post_ra_<sve_fp_op><mode>3): New pattern.
gcc/testsuite/
* gcc.target/aarch64/sve/pred_elim_1.c: New test.
Nick Clifton [Fri, 7 Dec 2018 10:33:30 +0000 (10:33 +0000)]
Add a recursion limit to libiberty's demangling code. The limit is enabled by default, but can be disabled via a new demangling option.
include * demangle.h (DMGL_NO_RECURSE_LIMIT): Define.
(DEMANGLE_RECURSION_LIMIT): Define
PR 87681
PR 87675
PR 87636
PR 87350
PR 87335
libiberty * cp-demangle.h (struct d_info): Add recursion_level field.
* cp-demangle.c (d_function_type): Add recursion counter.
If the recursion limit is reached and the check is not disabled,
then return with a failure result.
(cplus_demangle_init_info): Initialise the recursion_level field.
(d_demangle_callback): If the recursion limit is enabled, check
for a mangled string that is so long that there is not enough
stack space for the local arrays.
* cplus-dem.c (struct work): Add recursion_level field.
(squangle_mop_up): Set the numb and numk fields to zero.
(work_stuff_copy_to_from): Handle the case where a btypevec or
ktypevec field is NULL.
(demangle_nested_args): Add recursion counter. If
the recursion limit is not disabled and reached, return with a
failure result.
Jakub Jelinek [Thu, 6 Dec 2018 23:29:04 +0000 (00:29 +0100)]
re PR fortran/88377 (ICE in gfc_omp_clause_copy_ctor, at fortran/trans-openmp.c:614)
PR fortran/88377
* trans-openmp.c (gfc_omp_clause_default_ctor,
gfc_omp_clause_copy_ctor, gfc_omp_clause_assign_op,
gfc_omp_clause_linear_ctor, gfc_omp_clause_dtor): Only consider
GFC_DECL_GET_SCALAR_ALLOCATABLE vars as scalar allocatables if they
have pointer type.
Jakub Jelinek [Thu, 6 Dec 2018 23:28:04 +0000 (00:28 +0100)]
re PR tree-optimization/88367 (-fno-delete-null-pointer-checks doesn't work properly)
PR c/88367
* tree-vrp.c (extract_range_from_binary_expr): For POINTER_PLUS_EXPR
with -fno-delete-null-pointer-checks, set_nonnull only if the pointer
is non-NULL and offset is known to have most significant bit clear.
* vr-values.c (vr_values::vrp_stmt_computes_nonzero): For ADDR_EXPR
of MEM_EXPR, return true if the MEM_EXPR has non-zero offset with
most significant bit clear. If offset does have most significant bit
set and -fno-delete-null-pointer-checks, don't return true even if
the base pointer is non-NULL.
Paolo Carlini [Thu, 6 Dec 2018 23:20:16 +0000 (23:20 +0000)]
class.c (check_bitfield_decl): In error message about non-integral type print the type itself too.
/cp
2018-12-06 Paolo Carlini <paolo.carlini@oracle.com>
* class.c (check_bitfield_decl): In error message about non-integral
type print the type itself too.
* decl.c (grokdeclarator): Do not ICE on unnamed bit-fields declared
friends; when calling build_decl for a FIELD_DECL possibly pass the
declarator->id_loc.
/testsuite
2018-12-06 Paolo Carlini <paolo.carlini@oracle.com>
Alexandre Oliva [Thu, 6 Dec 2018 23:18:40 +0000 (23:18 +0000)]
[PR86747] tsubst friend tpl ctxt before looking it up for dupes
When a member template is redeclared as a friend, we enter the context
of the member before looking it up, and then we check that the decls
are compatible. However, when the member template references template
types of the enclosing context, say an enclosing template class, the
compare fails because the friend decl is already tsubsted, whereas the
looked up name isn't.
The problem is that the enclosing context is taken from the friend
declaration before tsubsting it, so we look up in the context of the
generic template instead of that of the tsubsted one we're
specializing. The solution is to tsubst the enclosing context when
it's a non-namespace scope.
for gcc/cp/ChangeLog
PR c++/86747
* pt.c (tsubst_friend_class): Enter tsubsted class context.
Alexandre Oliva [Thu, 6 Dec 2018 23:18:30 +0000 (23:18 +0000)]
[PR86397] resolve nondependent noexcept specs early in C++1[14]
build_noexcept_spec refrained from resolving nondependent noexcept
expressions when they were not part of the function types (C++ 11 and
14). This caused problems during mangling: canonical_eh_spec, when
called on the template function type, would find an unresolved but not
explicitly deferred expression, and nothrow_spec_p would reject it.
We could relax the mangling logic to skip canonical_eh_spec, but since
-Wnoexcept-type warns when mangling function names that change as
noexcept specs become part of types and of mangling in C++17, and the
test at mangling time may give incorrect results if the spec is not
resolved, we might as well keep things simple and resolve nondependent
noexcept specs sooner rather than later. This is what this patch does.
Alexandre Oliva [Thu, 6 Dec 2018 23:18:19 +0000 (23:18 +0000)]
avoid null ptr deref in cselib_record_sets
Jeff Law tells me h8300-elf fails gcc.c-torture/compile/pr49029.c
with -O2 -g -mint32 -mh. This patch fixes it.
The problem is that strict low part handling in cselib_record_sets
assumes src_elt is not NULL. That src_elt is taken from a strict low
part set, but it won't always have a src_elt to begin with. In this
case, it's because src is a volatile MEM; we don't record values for
those.
Although we could fix the problem by testing for a NULL src_elt before
creating the zero extends corresponding to strict low part sets of
formerly const0_rtx REGs, there's no point in recording the additional
set that we won't be able to use anyway.
We could still record that the whole register has a zero-extend of
the value stored in the narrower-mode strict low part of the register,
but is that of any use? I guess not, but if we find otherwise, we can
change that later.
for gcc/ChangeLog
* cselib.c (cselib_record_sets): Skip strict low part sets
with NULL src_elt.
The testsuite tests for the compatibility implementations of x86 vector
intrinsics for "powerpc" had been inadvertently made to PASS
without actually running the test code.
This patch removes the code which kept the tests from running the actual
test code.
2018-12-06 Paul A. Clarke <pc@us.ibm.com>
[gcc/testsuite]
PR target/88316
* gcc.target/powerpc/bmi-check.h: Remove test for
__BUILTIN_CPU_SUPPORTS__, thereby enabling test code to run.
* gcc.target/powerpc/bmi2-check.h: Likewise.
* gcc.target/powerpc/mmx-check.h: Likewise.
* gcc.target/powerpc/sse-check.h: Likewise.
* gcc.target/powerpc/sse2-check.h: Likewise.
* gcc.target/powerpc/sse3-check.h: Likewise.
Paul A. Clarke [Thu, 6 Dec 2018 22:11:01 +0000 (22:11 +0000)]
[rs6000] Fix x86-compat vector intrinsics testcases for BE, 32bit
Fix general endian issues found in the test cases for thecompatibility
implementations of the x86 vector intrinsics. (The tests had been
inadvertently made to PASS without actually running the test code.
A later patch fixes this issue.)
Additionally, a new <smmintrin.h> is added, as some of the APIs therein are
now used by the test cases. It is _not_ a complete implementation of the
SSE4 interfaces, only the few "extract" interfaces uses by the tests.
2018-12-06 Paul A. Clarke <pc@us.ibm.com>
[gcc]
PR target/88316
* config/rs6000/smmintrin.h: New file.
* config.gcc: Add smmintrin.h to extra_headers for powerpc*-*-*.
Paul A. Clarke [Thu, 6 Dec 2018 22:03:25 +0000 (22:03 +0000)]
[rs6000] x86-compat vector intrinsics fixes for BE, 32bit
Fix general endian and 32-bit mode issues found in the
compatibility implementations of the x86 vector intrinsics when running the
associated test suite tests. (The tests had been inadvertently made to PASS
without actually running the test code. A later patch fixes this issue.)
2018-12-03 Paul A. Clarke <pc@us.ibm.com>
PR target/88316
* config/rs6000/mmintrin.h (_mm_unpackhi_pi8): Fix for big-endian.
(_mm_unpacklo_pi8): Likewise.
(_mm_mulhi_pi16): Likewise.
(_mm_packs_pi16): Fix for big-endian. Use preferred API.
(_mm_packs_pi32): Likewise.
(_mm_packs_pu16): Likewise.
* config/rs6000/xmmintrin.h (_mm_cvtss_si32): Fix for big-endian.
(_mm_cvtss_si64): Likewise.
(_mm_cvtpi32x2_ps): Likewise.
(_mm_shuffle_ps): Likewise.
(_mm_movemask_pi8): Likewise.
(_mm_mulhi_pu16): Likewise.
(_mm_sad_pu8): Likewise.
(_mm_sad_pu8): Likewise.
(_mm_cvtpu16_ps): Fix for big-endian. Use preferred API.
(_mm_cvtpu8_ps): Likewise.
(_mm_movemask_ps): Better #else case for big-endian (no functional
change).
(_mm_shuffle_pi16): Likewise.
* config/rs6000/emmintrin.h (_mm_movemask_pd): Fix for big-endian.
Better #else case for big-endian (no functional change).
(_mm_movemask_epi8): Likewise.
(_mm_shufflehi_epi16): Likewise.
(_mm_shufflelo_epi16): Likewise.
(_mm_shuffle_epi32): Likewise.
(_mm_mul_epu32): Fix for big-endian.
(_mm_bsrli_si128): Likewise.
(_mm_cvtps_pd): Better #else case for big endian.
(_mm_mulhi_epi16): Likewise.
(_mm_mul_epu32): Likewise.
(_mm_slli_si128): Likewise.
(_mm_sll_epi16): Likewise.
(_mm_sll_epi32): Likewise.
(_mm_sra_epi16): Likewise.
(_mm_sra_epi32): Likewise.
(_mm_srl_epi16): Likewise.
(_mm_srl_epi32): Likewise.
(_mm_mulhi_epu16): Likewise.
(_mm_sad_epu8): Likewise.
* config/rs6000/pmmintrin.h (_mm_hadd_ps): Fix for big-endian.
(_mm_sub_ps): Likewise.
* config/rs6000/mmintrin.h (_mm_cmpeq_pi8): Fix for 32-bit mode.
* gcc/config/rs6000/tmmintrin.h (_mm_alignr_epi8): Use ENDIAN
macros consistently (no functional changes).
(_mm_alignr_pi8): Likewise.
Jason Merrill [Thu, 6 Dec 2018 21:17:08 +0000 (16:17 -0500)]
PR c++/88136 - -Wdeprecated-copy false positives
Deprecating the copy operations because the class has a user-provided
destructor turns out to have too many false positives; this patch adjusts
-Wdeprecated-copy to only deprecate if the other copy operation is
user-provided. To get the earlier behavior, people can explicitly request
it with -Wdeprecated-copy-dtor.
gcc/c-family/
* c.opt (Wdeprecated-copy-dtor): New.
(Wdeprecated-copy): Move to -Wextra.
gcc/cp/
* class.c (classtype_has_depr_implicit_copy): Rename from
classtype_has_user_copy_or_dtor.
* method.c (lazily_declare_fn): Adjust.
* decl2.c (cp_warn_deprecated_use): Refer to -Wdeprecated-copy-dtor
if deprecation is due to a destructor.
Iain Sandoe [Thu, 6 Dec 2018 19:42:21 +0000 (19:42 +0000)]
Darwin fix libstdc++ build warnings.
GCC does not export construction vtable symbols from shared libraries.
The symbols are marked hidden in the objects; for Darwin that makes
them also external (“private_extern” is Darwin’s hidden) which means
that they show up in the list of possible symbols for export from
libstdc++, and there are sufficiently relaxed match conditions that
they reach the exports list. When Darwin’s static linker encounters
them it generates a warning that they cannot be exported. This patch
prunes them from the list of symbols to be considered, thus
eliminating the warnings. No functional Change inended to the library
exports.
2018-12-06 Iain Sandoe <iain@sandoe.co.uk>
* scripts/make_exports.pl (check names): Don’t try to export
construction vtable symbols.
Jonathan Wakely [Thu, 6 Dec 2018 19:21:32 +0000 (19:21 +0000)]
Fix PR libstdc++/64883 Darwin headers use always_inline so don't test that
Because darwin system headers use always_inline rather than
__always_inline__ the libstdc++ test will fail, even if our headers only
use the reserved form of the attribute. Don't test it on Darwin, and
assume that testing on other targets will catch any accidental misuses
in libstdc++ headers.
2018-12-06 Jonathan Wakely <jwakely@redhat.com>
Iain Sandoe <iain@sandoe.co.uk>
re PR target/88282 (ICE in df_install_refs at gcc/df-scan.c:2379)
2018-12-06 Vladimir Makarov <vmakarov@redhat.com>
PR target/88282
* ira.c (ira_init_register_move_cost): Use info from
hard_regno_mode_ok instead of contains_reg_of_mode.
* ira-costs.c (contains_reg_of_mode): Don't use cost from bigger
hard register class for some fixed hard registers.
The Linux kernel people want a feature that makes GCC pretend some
inline assembler code is tiny (while it would think it is huge), so
that such code will be inlined essentially always instead of
essentially never.
This patch lets you say "asm inline" instead of just "asm", with the
result that that inline assembler is always counted as minimum cost
for inlining. It implements this for C and C++, making "inline"
another asm-qualifier (supplementing "volatile" and "goto").
* doc/extend.texi (Using Assembly Language with C): Document asm inline.
(Size of an asm): Fix typo. Document asm inline.
* gimple-pretty-print.c (dump_gimple_asm): Handle asm inline.
* gimple.h (enum gf_mask): Add GF_ASM_INLINE.
(gimple_asm_set_volatile): Fix typo.
(gimple_asm_inline_p): New.
(gimple_asm_set_inline): New.
* gimplify.c (gimplify_asm_expr): Propagate the asm inline flag from
tree to gimple.
* ipa-icf-gimple.c (func_checker::compare_gimple_asm): Compare the
gimple_asm_inline_p flag, too.
* tree-core.h (tree_base): Document that protected_flag is ASM_INLINE_P
in an ASM_EXPR.
* tree-inline.c (estimate_num_insns): If gimple_asm_inline_p return
a minimum size for an asm.
* tree.h (ASM_INLINE_P): New.
gcc/c/
* c-parser.c (c_parser_asm_statement): Detect the inline keyword
after asm. Pass a flag for it to build_asm_expr.
* c-tree.h (build_asm_expr): Update declaration.
* c-typeck.c (build_asm_stmt): Add is_inline parameter. Use it to
set ASM_INLINE_P.
gcc/cp/
* cp-tree.h (finish_asm_stmt): Update declaration.
* parser.c (cp_parser_asm_definition): Detect the inline keyword
after asm. Pass a flag for it to finish_asm_stmt.
* pt.c (tsubst_expr): Pass the ASM_INLINE_P flag to finish_asm_stmt.
* semantics.c (finish_asm_stmt): Add inline_p parameter. Use it to
set ASM_INLINE_P.
gcc/testsuite/
* c-c++-common/torture/asm-inline.c: New testcase.
* gcc.dg/asm-qual-2.c: Test asm inline, too.
PR55681 observes that currently only one qualifier is allowed for
inline asm, so that e.g. "volatile asm" is allowed, "const asm" is also
okay (with a warning), but "const volatile asm" gives an error. Also
"goto" has to be last.
This patch changes things so that only "asm-qualifiers" are allowed,
that is "volatile" and "goto", in any combination, in any order, but
without repetitions.
gcc/c/
PR inline-asm/55681
* c-parser.c (c_parser_asm_statement): Update grammar. Allow any
combination of volatile and goto, in any order, without repetitions.
gcc/cp/
PR inline-asm/55681
* parser.c (cp_parser_asm_definition): Update grammar. Allow any
combination of volatile and goto, in any order, without repetitions.
gcc/testsuite/
PR inline-asm/55681
* gcc.dg/asm-qual-1.c: Test that "const" and "restrict" are refused.
* gcc.dg/asm-qual-2.c: New test, test that asm-qualifiers are allowed
in any order, but that duplicates are not allowed.
David Malcolm [Thu, 6 Dec 2018 15:25:12 +0000 (15:25 +0000)]
C++: improvements to diagnostics using %P (more PR c++/85110)
This patch is based on grepping the C++ frontend for %P
i.e. diagnostics that refer to a parameter number. It fixes up
these diagnostics to highlight the pertinent param where appropriate
(and possible), along with various other tweaks, as described in the
ChangeLog.
gcc/cp/ChangeLog:
PR c++/85110
* call.c (conversion_null_warnings): Try to use the location of
the expression for the warnings. Add notes showing the parameter
of the function decl, where available.
(get_fndecl_argument_location): For implicitly-declared functions,
use the fndecl location rather than that of the param.
(maybe_inform_about_fndecl_for_bogus_argument_init): New function.
(convert_like_real): Use it in various places to avoid repetition.
(complain_about_bad_argument): Likewise.
* cp-tree.h (maybe_inform_about_fndecl_for_bogus_argument_init):
New declaration.
* decl2.c (check_default_args): Put all diagnostics for the fndecl
into a diagnostic group. Use the location of the parameter when
complaining about parameters with missing default arguments in
preference to that of the fndecl. Attempt to record the location
of the first parameter with a default argument and emit a note
for the first parameter that's missing one.
* typeck.c (convert_arguments): When complaining about parameters
with incomplete types, attempt to use the location of the
argument. Where available, add a note showing the pertinent
parameter in the fndecl.
(convert_for_assignment): When complaining about bad conversions
at function calls, use the location of the unstripped argument.
(convert_for_initialization): When checking for bogus references,
add an auto_diagnostic_group, and update the note to use the
location of the pertinent parameter, rather than just the callee.
gcc/testsuite/ChangeLog:
PR c++/85110
* g++.dg/diagnostic/missing-default-args.C: New test.
* g++.dg/diagnostic/param-type-mismatch-3.C: New test.
* g++.dg/diagnostic/param-type-mismatch.C: Add tests for invalid
references and incomplete types.
* g++.dg/warn/Wconversion-null-4.C: New test.
Richard Biener [Thu, 6 Dec 2018 11:32:09 +0000 (11:32 +0000)]
bitmap.c (bitmap_head::crashme): Define.
2018-12-06 Richard Biener <rguenther@suse.de>
* bitmap.c (bitmap_head::crashme): Define.
* bitmap.h (bitmap_head): Add constexpr default constructor
poisoning the obstack member.
(bitmap_head::crashme): Declare.
(bitmap_release): New function clearing a bitmap and poisoning
the obstack member.
* gengtype.c (main): Make it recognize CONSTEXPR.
* lra-constraints.c (lra_inheritance): Use bitmap_release
instead of bitmap_clear.
* ira.c (ira): Work around class-memaccess warning.
* regrename.c (create_new_chain): Likewise.
Jakub Jelinek [Thu, 6 Dec 2018 10:45:47 +0000 (11:45 +0100)]
re PR tree-optimization/85726 (div C1 to div C2 match.pd suboptimization)
PR tree-optimization/85726
* generic-match-head.c (optimize_successive_divisions_p): New function.
* gimple-match-head.c (optimize_successive_divisions_p): Likewise.
* match.pd: Don't combine successive divisions if they aren't exact
and optimize_successive_divisions_p is false.
* gcc.dg/tree-ssa/pr85726-1.c: New test.
* gcc.dg/tree-ssa/pr85726-2.c: New test.
* gcc.dg/tree-ssa/pr85726-3.c: New test.
* gcc.dg/tree-ssa/pr85726-4.c: New test.
This CL adds support of precise stack scan using stack maps to
the runtime. The stack maps are generated by the compiler (if
supported). Each safepoint is associated with a (real or dummy)
landing pad, and its "type info" in the exception table is a
pointer to the stack map. When a stack is scanned, the stack map
is found by the stack unwinding code by inspecting the exception
table (LSDA).
For precise stack scan we need to unwind the stack. There are
three cases:
- If a goroutine is scanning its own stack, it can unwind the
stack and scan the frames.
- If a goroutine is scanning another, stopped, goroutine, it
cannot directly unwind the target stack. We handle this by
switching (runtime.gogo) to the target g, letting it unwind
and scan the stack, and switch back.
- If we are scanning a goroutine that is blocked in a syscall,
we send a signal to the target goroutine's thread, and let the
signal handler unwind and scan the stack. Extra care is needed
as this races with enter/exit syscall.
Use unsigned arithmetic for demoted vector plus/minus/mult (PR 88064)
As Jakub pointed out, if we narrow a plus, minus or mult operation based
on the number of bits that consumers need, we have to convert a signed
operation to an unsigned one in order to avoid new undefined behaviour.
This patch does that and generalises vect_convert_input and
vect_recog_over_widening_pattern to cope with the extra casts.
(The changes to both functions are covered by existing tests.)
2018-12-03 Richard Sandiford <richard.sandiford@arm.com>
gcc/
PR tree-optimization/88064
* tree-vect-patterns.c (vect_convert_input): Convert the result of
an existing cast if it has the right width but the wrong sign.
Do not test the signedness of the required result when
considering whether to split an existing cast; instead split to
a type with the same signedness as the source of the cast, then
convert it to the opposite signedness where necessary.
(vect_recog_over_widening_pattern): Handle sign changes between
the final PLUS_EXPR and the RSHIFT_EXPR.
(vect_recog_average_pattern): Use an unsigned operation when
truncating an addition, subtraction or multiplication. Cast the
result back to the "real" signedness before promoting.
gcc/testsuite/
PR tree-optimization/88064
* gcc.dg/vect/vect-over-widen-23.c: New test.
David Malcolm [Wed, 5 Dec 2018 15:16:05 +0000 (15:16 +0000)]
testsuite: turn down verbosity of "process-message"
When debugging a failing test, I typically invoke DejaGnu at
verbosity level 2 (via RUNTESTFLAGS="-v -v dg.exp=something"),
so that DejaGnu prints the command line used to invoke the
compiler; specifically these two sites:
target.exp "Invoking the compiler as "
remote.exp "Executing on $hostname"
which are both verbosity level 2.
Unfortunately I run into an O(n^2) issue with logging from
process-message:
verbose "process-message:\n${dg-messages}" 2
where, as each message each processed, it emits the state
of dg-messages, containing the new message and all messages so far,
leading to exponentially-increasing output at level 2 as more test
messages are added.
This patch papers over the problem by moving the
problematic message to verbosity level 3.
gcc/testsuite/ChangeLog:
* lib/gcc-dg.exp (process-message): Change verbosity level of
"verbose" from 2 to 3.
(dg-locus): Likewise.
Alexandre Oliva [Wed, 5 Dec 2018 06:47:11 +0000 (06:47 +0000)]
[PR85569] skip constexpr target_expr constructor dummy type conversion
The testcase is the work-around testcase for the PR; even that had
started failing. The problem was that, when unqualifying the type of
a TARGET_EXPR, we'd create a variant of the type, then request the
conversion of the TARGET_EXPR_INITIAL to that variant type. Though
the types are different pointer-wise, they're the same_type_p, so the
resulting modified expr compares cp_tree_equal to the original, which
maybe_constant_value flags as an error. There's no reason to
construct an alternate TARGET_EXPR or CONSTRUCTOR just because of an
equivalent type, except for another spot that expected pointer
equality that would no longer be satisfied. Without relaxing the
assert in constexpr_call_hasher::equal, g++.robertl/eb73.C would
trigger an assertion failure.
for gcc/cp/ChangeLog
PR c++/85569
* constexpr.c (adjust_temp_type): Test for type equality with
same_type_p.
(constexpr_call_hasher::equal): Likewise.
David's fix for the AIX aggregate passing from yesterday unfortunately
also triggers on powerpc64-linux. This fixes it.
* config/rs6000/rs6000.c (rs6000_function_arg): Only do the special
aggregate handling on actual AIX, not on somewhat similar systems.
(rs6000_arg_partial_bytes): Ditto.