2019-01-22 Richard Biener <rguenther@suse.de>
* tree-vect-loop.c (vect_analyze_loop_operations): Use
auto_vec for cost vector to fix memleak.
(vectorize_fold_left_reduction): Properly gather SLP defs.
(vectorizable_comparison): Do not swap operands to properly
gather SLP defs.
From-SVN: r268137
The direct cause of this PR is the fact that tls_gdld_nomark didn't
handle indirect calls. Also, most indirect calls were being optimised
back to direct calls anyway, due to tls_gdld_nomark not checking any
of the parallel elements except the first (plus the extra element that
distinguishes this call from normal calls). There were other unwanted
substitutions too.
So this patch attacks the problem of handling special calls in a
different way. Rather than adding another element to the call insn
parallel to distinguish -mno-tls-markers __tls_get_addr calls from any
other calls, we now inspect the second CALL arg. Each
call_value_nonlocal and call_value_indirect insn now checks for the
tlsgd/ld unspecs when !TARGET_TLS_MARKERS and emits the arg setup
insns. I disallow the local call patterns since we'll only see local
calls to __tls_get_addr in testcases, and it doesn't seem a good idea
to complicate the patterns just for a minor optimisation. Sibling
call insns aren't used for libcalls, so none of these insns need to
change.
The patch also fixes a minor problem with -mno-tls-markers
__tls_get_addr calls causing a "li 3,0" instruction to be emitted
prior to the arg setup instructions, due to using a libcall with one
arg. That isn't correct when the call insn itself sets up its arg.
Also, I've tidied the V4 secure-plt calls, generating them in
rs6000_call_sysv rather than by splitting in rs6000.md. The
CALL_INSN_FUNCTION_USAGE added in edit_tls_call_insn is no longer
needed (since git commit 0a4b5c66df9).
On the subject of unwanted substitutions, I also saw a
_GLOBAL_OFFSET_TABLE_ symbol_ref being substituted for the GOT reg,
resulting in code like "addi 3,_GLOBAL_OFFSET_TABLE_,tls_ld@got@tlsld".
Fixed by the unspec_tls change.
PR target/88614
* config/rs6000/predicates.md (unspec_tls): Ensure GOT reg
stays a reg. Allow a const_int.
* config/rs6000/rs6000-protos.h (rs6000_output_tlsargs): Declare.
* config/rs6000/rs6000.h (IS_V4_FP_ARGS): Define.
(IS_NOMARK_TLSGETADDR): Define.
* config/rs6000/rs6000.c (edit_tls_call_insn): Delete.
(rs6000_output_tlsargs): New function.
(rs6000_legitimize_tls_address): Don't say a !TARGET_TLS_MARKERS
__tls_get_addr call takes an arg.
(rs6000_call_sysv): Generate sysv4 secure plt call pattern here..
* config/rs6000/rs6000.md (call_nonlocal_sysv): ..rather than here,
delete split..
(call_value_nonlocal_sysv): ..or here, delete split.
(tls_gdld_nomark): Delete.
(call_value_indirect_nonlocal_sysv): Use unspec_tls as operand2
predicate. Call rs6000_output_tlsargs. Adjust length to suit.
(call_value_nonlocal_sysv): Likewise.
(call_value_nonlocal_sysv_secure): Likewise.
(call_value_nonlocal_aix): Likewise.
(call_value_indirect_aix): Likewise.
(call_value_indirect_elfv2): Likewise.
(call_value_local32, call_value_local64): Disable for no-mark tls.
(call_value_local_aix): Likewise.
From-SVN: r268135
PR go/88927
runtime, internal/cpu: fix build for ARM GNU/Linux
Was failing with
../../../libgo/go/internal/cpu/cpu.go:138:2: error: reference to undefined name 'doinit'
138 | doinit()
| ^
Fix it by adding in Go 1.12 internal/cpu/cpu_arm.go, and the code in
runtime that initializes the values.
Fixes https://gcc.gnu.org/PR88927.
Reviewed-on: https://go-review.googlesource.com/c/158717
From-SVN: r268131
Restore some of the fixes that were applied to golang_org/x/net/lif
but were lost when 1.12 moved the directory to internal/x/net/lif.
Add support for reading /proc to fetch argc/argv/env for c-archive mode.
Reviewed-on: https://go-review.googlesource.com/c/158640
From-SVN: r268130
If there is an error reading or parsing an archive header, the
Archive_iterator code would return a dummy header but would not mark
itself as done. The effect is that an invalid archive leads to an
endless loop reading and re-reading the same archive header. Avoid
that by setting the offset to the end of the file, which will cause
the iterator to == archive_end.
No test since it doesn't seem worth constructing an invalid archive.
Reviewed-on: https://go-review.googlesource.com/c/158217
From-SVN: r268129
PR c++/88949
* optimize.c (cxx_copy_decl): New function.
(clone_body): Use it instead of copy_decl_no_change.
* g++.dg/gomp/pr88949.C: New test.
From-SVN: r268127
Reduces the memory footprint of the CTFE interpreter by replacing new
with emplacement new in many places.
gcc/d/ChangeLog:
2019-01-21 Iain Buclaw <ibuclaw@gdcproject.org>
* d-frontend.cc (Compiler::paintAsType): Update for new signature.
From-SVN: r268124
gcc/testsuite/ChangeLog:
* g++.dg/vect/simd-clone-7.cc: Fix assembler scan.
--This line, a d those below, will be ignored--
M gcc/testsuite/ChangeLog
M gcc/testsuite/g++.dg/vect/simd-clone-7.cc
From-SVN: r268120
2019-01-21 Bernd Edlinger <bernd.edlinger@hotmail.de>
PR c/88928
* c-warn.c (check_alignment_of_packed_member): Add a boolean parameter
for rvalue context. Handle rvalues correctly. Use min_align_of_type
instead of TYPE_ALIGN.
(check_address_or_pointer_of_packed_member): Handle rvalues coorrectly.
Use min_align_of_type instead of TYPE_ALIGN_UNIT. Check for NULL
pointer from TYPE_STUB_DECL.
testsuite:
2019-01-21 Bernd Edlinger <bernd.edlinger@hotmail.de>
PR c/88928
* c-c++-common/Waddress-of-packed-member-1.c: New test case.
* gcc.dg/pr88928.c: New test case.
From-SVN: r268118
Philippe Waroquiers noticed a memory leak in gdb, which he tracked
down to a bug in splay-tree. splay_tree_remove does not call the
`delete_key' function when it removes the old node; but it should.
I looked at every splay tree in GCC and there is only one that passes
a non-NULL delete function -- the one in lto.c. That file does not
call splay_tree_remove. So, I think this is safe to check in.
I re-ran the LTO tests to double check.
libiberty/
* splay-tree.c (splay_tree_remove): Delete the key if necessary.
From-SVN: r268116
2019-01-20 Martin Jambor <mjambor@suse.cz>
PR ipa/87615
* ipa-prop.h (struct ipa_func_body_info): Replaced field aa_walked
with aa_walk_budget.
* cgraph.h (ipa_polymorphic_call_context::get_dynamic_type): Add
aa_walk_budget_p parameter.
* ipa-fnsummary.c (unmodified_parm_1): New parameter fbi. Limit AA
walk. Updated all callers.
(unmodified_parm): New parameter fbi, pass it to unmodified_parm_1.
(eliminated_by_inlining_prob): New parameter fbi, pass it on to
unmodified_parm.
(will_be_nonconstant_expr_predicate): New parameter fbi, removed
parameter info. Extract info from fbi. Pass fbi to recursive calls
and to unmodified_parm.
(phi_result_unknown_predicate): New parameter fbi, removed parameter
info, updated call to will_be_nonconstant_expr_predicate.
(param_change_prob): New parameter fbi, limit AA walking.
(analyze_function_body): Initialize aa_walk_budget in fbi. Update
calls to various above functions.
* ipa-polymorphic-call.c (get_dynamic_type): Add aa_walk_budget_p
parameter. Use it to limit AA walking.
* ipa-prop.c (detect_type_change_from_memory_writes): New parameter
fbi, limit AA walk.
(detect_type_change): New parameter fbi, pass it on to
detect_type_change_from_memory_writes.
(detect_type_change_ssa): Likewise.
(aa_overwalked): Removed.
(parm_preserved_before_stmt_p): Assume fbi is never NULL, stream line
accordingly, adjust to the neew AA limiting scheme.
(parm_ref_data_preserved_p): Likewise.
(ipa_compute_jump_functions_for_edge): Adjust call to
get_dynamic_type.
(ipa_analyze_call_uses): Likewise.
(ipa_analyze_virtual_call_uses): Pass fbi to detect_type_change_ssa.
(ipa_analyze_node): Initialize aa_walk_budget.
(ipcp_transform_function): Likewise.
* tree-ssa-sccvn.c (eliminate_dom_walker::eliminate_stmt): Update call
to get_dynamic_type.
From-SVN: r268107
Every back-end defines STACK_GROWS_DOWNWARD, the condition should
instead be checking whether it was defined as 1 or 0.
gcc/d/ChangeLog:
2019-01-20 Iain Buclaw <ibuclaw@gdcproject.org>
* d-builtins.cc (d_init_versions): Check value of
STACK_GROWS_DOWNWARD.
From-SVN: r268106
gcc/d/ChangeLog:
2019-01-20 Iain Buclaw <ibuclaw@gdcproject.org>
* d-codegen.cc (identity_compare_p): Return false if seen built-in
type with padding.
(build_float_identity): Moved here from expr.cc.
(lower_struct_comparison): Handle real and complex types.
* d-tree.h (build_float_identity): New.
* expr.cc (build_float_identity): Move to d-codegen.cc.
gcc/testsuite/ChangeLog:
2019-01-20 Iain Buclaw <ibuclaw@gdcproject.org>
* gdc.dg/runnable.d: Add more tests for comparing complex types.
From-SVN: r268104
2019-01-19 Steven G. Kargl <kargl@gcc.gnu.org>
PR fortran/77960
* io.c (match_io_element): input-item cannot be an external function.
2019-01-19 Steven G. Kargl <kargl@gcc.gnu.org>
PR fortran/77960
* gfortran.dg/pr77960.f90: New test.
From-SVN: r268097
2018-01-19 Thomas Koenig <tkoenig@gcc.gnu.org>
Paul Thomas <pault@gcc.gnu.org>
PR fortran/56789
* trans-expr.c (gfc_conv_procedure_call): Call
gfc_conv_subref_array_arg if the formal arg is contiguous
and the actual arg may not be.
2018-01-19 Thomas Koenig <tkoenig@gcc.gnu.org>
Paul Thomas <pault@gcc.gnu.org>
PR fortran/56789
* gfortran.dg/contiguous_3.f90: Make code compilant. Remove
scan-tree tests that fail with patch.
* gfortran.dg/contiguous_8.f90: New test.
Co-Authored-By: Paul Thomas <pault@gcc.gnu.org>
From-SVN: r268096
This patch tries harder to detect cases in which the inner dimension
of an array access is invariant, such as:
x(i, :) = 100
It fixes some of the code size regression in 548.exchange2_r, with
size improving by 5% compared to before the patch. Of the two other
SPEC 2017 tests affected by loop versioning, 554.roms_r improved by a
trivial amount (0.3%) and 549.fotonik3d_r didn't change. All three
results are with -Ofast -flto.
2019-01-19 Richard Sandiford <richard.sandiford@arm.com>
gcc/
* gimple-loop-versioning.cc (loop_versioning::dump_inner_likelihood):
New function, split out from...
(loop_versioning::analyze_stride): ...here.
(loop_versioning::find_per_loop_multiplication): Use gassign.
(loop_versioning::analyze_term_using_scevs): Return a success code.
(loop_versioning::analyze_arbitrary_term): New function.
(loop_versioning::analyze_address_fragment): Use
analyze_arbitrary_term if all else fails.
gcc/testsuite/
* gfortran.dg/loop_versioning_1.f90: Bump the number of identified
inner strides.
* gfortran.dg/loop_versioning_9.f90: New test.
* gfortran.dg/loop_versioning_10.f90: Likewise.
From-SVN: r268093
PR fortran/88902
* trans-decl.c (gfc_get_symbol_decl): Don't add length to function
or parent function if it has been added there already.
* gfortran.dg/pr88902.f90: New test.
From-SVN: r268091
The old version of _Sp_counted_ptr_inplace::_M_get_deleter (up to GCC
8.2.0) expects to be passed a real std::typeinfo object, so mixing that
with the new definition of the __shared_ptr constructor (which always
passes the fake tag) leads to accessing the fake object as a real
std::typeinfo. Instead of trying to make it safe to mix the old and new
definitions, just stop using that function. By passing a reference to
__shared_ptr::_M_ptr to the __shared_count constructor it can be set
directly, without needing to obtain the pointer via the _M_get_deleter
back-channel. This avoids a virtual dispatch (which fixes PR 87514).
This means that code built against new libstdc++ headers doesn't use
_M_get_deleter at all, and so make_shared works the same whether RTTI is
enabled or not.
Also change _M_get_deleter so that it checks for a real type_info object
even when RTTI is disabled, by calling a library function. Unless
libstdc++ itself is built without RTTI that library function will be
able to test if it's the right type_info. This means the new definition
of _M_get_deleter can handle both the fake type_info tag and a real
type_info object, even if built without RTTI.
If linking to objects built against older versions of libstdc++ then if
all objects use -frtti or all use -fno-rtti, then the caller of
_M_get_deleter and the definition of _M_get_deleter will be consistent
and it will work. If mixing -frtti with -fno-rtti it can still fail if
the linker picks an old definition of _M_get_deleter and an old
__shared_ptr constructor that are incompatible. In that some or all
objects might need to be recompiled.
PR libstdc++/87514
PR libstdc++/87520
PR libstdc++/88782
* config/abi/pre/gnu.ver (GLIBCXX_3.4.26): Export new symbol.
* include/bits/shared_ptr.h
(shared_ptr(_Sp_make_shared_tag, const Alloc&, Args&&...))
(allocate_shared): Change to use new tag type.
* include/bits/shared_ptr_base.h (_Sp_make_shared_tag::_S_eq):
Declare new member function.
(_Sp_alloc_shared_tag): Define new type.
(_Sp_counted_ptr_inplace): Declare __shared_count<_Lp> as a friend.
(_Sp_counted_ptr_inplace::_M_get_deleter) [!__cpp_rtti]: Use
_Sp_make_shared_tag::_S_eq to check type_info.
(__shared_count(Ptr, Deleter),__shared_count(Ptr, Deleter, Alloc)):
Constrain to prevent being called with _Sp_alloc_shared_tag.
(__shared_count(_Sp_make_shared_tag, const _Alloc&, Args&&...)):
Replace constructor with ...
(__shared_count(Tp*&, _Sp_alloc_shared_tag<_Alloc>, Args&&...)): Use
reference parameter so address of the new object can be returned to
the caller. Obtain the allocator from the tag type.
(__shared_ptr(_Sp_make_shared_tag, const Alloc&, Args&&...)): Replace
constructor with ...
(__shared_ptr(_Sp_alloc_shared_tag<Alloc>, Args&&...)): Pass _M_ptr
to the __shared_count constructor.
(__allocate_shared): Change to use new tag type.
* src/c++11/shared_ptr.cc (_Sp_make_shared_tag::_S_eq): Define.
From-SVN: r268086
In my patch for CWG issue 2267, I changed reference_binding to clear
CONSTRUCTOR_IS_DIRECT_INIT on the argument init-list. But that breaks if
there's another candidate for which CONSTRUCTOR_IS_DIRECT_INIT is correct.
So instead, let's encode in the conversion that we want to override the
flag.
* call.c (reference_binding): Don't modify EXPR. Set
need_temporary_p on the ck_user conversion for a temporary.
(convert_like_real): Check it.
From-SVN: r268085
Reviewed-on: https://go-review.googlesource.com/c/158019
gotools/:
* Makefile.am (go_cmd_vet_files): Update for Go1.12beta2 release.
(GOTOOLS_TEST_TIMEOUT): Increase to 600.
(check-runtime): Export LD_LIBRARY_PATH before computing GOARCH
and GOOS.
(check-vet): Copy golang.org/x/tools into check-vet-dir.
* Makefile.in: Regenerate.
gcc/testsuite/:
* go.go-torture/execute/names-1.go: Stop using debug/xcoff, which
is no longer externally visible.
From-SVN: r268084
The memory store instructions (stfs[u][x], stxssp[x]) can result in
garbage if the value to be stored isn't already a valid single
precision floating point number. So we cannot use this here.
PR target/88892
* config/rs6000/rs6000.md (*movsi_from_df): Allow only register
operands.
From-SVN: r268083