gcc.git
5 years ago[libbacktrace] Unify function name preference handling
Tom de Vries [Wed, 16 Jan 2019 07:46:56 +0000 (07:46 +0000)]
[libbacktrace] Unify function name preference handling

Both read_function_entry and read_referenced_name implement a priority scheme
for names.  The priorities are:
- 1st: DW_AT_linkage_name
- 2nd: Name from DW_AT_abstract_origin or DW_AT_specification
- 3rd: DW_AT_name.

Ensure both functions fully adhere to it.

2019-01-16  Tom de Vries  <tdevries@suse.de>

* dwarf.c (read_referenced_name): Don't allow DW_AT_name to override any
name.
(read_function_entry): Same.  Don't allow name found via
DW_AT_abstract_origin or case DW_AT_specification to override linkage
name.

From-SVN: r267963

5 years agoAdd myself
Xiong Hu Luo [Wed, 16 Jan 2019 07:41:23 +0000 (07:41 +0000)]
Add myself

From-SVN: r267962

5 years agoDaily bump.
GCC Administrator [Wed, 16 Jan 2019 00:16:17 +0000 (00:16 +0000)]
Daily bump.

From-SVN: r267961

5 years agogodump.c (go_output_typedef): When outputting a typedef...
Nikhil Benesch [Tue, 15 Jan 2019 23:54:37 +0000 (23:54 +0000)]
godump.c (go_output_typedef): When outputting a typedef...

gcc/:
        * godump.c (go_output_typedef): When outputting a typedef, refer
to the underlying type by its name and not its structure.
gcc/testsuite:
        * gcc.misc-tests/godump-1.c: Add test case for typedef before
struct.

From-SVN: r267958

5 years agoFix ICE on class-template argument deduction (PR c++/88795)
David Malcolm [Tue, 15 Jan 2019 23:29:15 +0000 (23:29 +0000)]
Fix ICE on class-template argument deduction (PR c++/88795)

PR c++/88795 reports an ICE building a function_type for a deduction guide
when the substitution into the function signature fails, due to an
error_mark_node being returned from tsubst_arg_types but not being checked
for.  This error_mark_node gets used as the TYPE_ARG_TYPES, leading to
ICEs in various places that assume this is a TREE_LIST.

This patch checks the result of tsubst_arg_types and propagates the failure
if it returns error_mark_node.  It also adds an assertion to
build_function_type, to fail faster if passed in error_mark_node.

gcc/cp/ChangeLog:
PR c++/88795
* pt.c (build_deduction_guide): Bail out if tsubst_arg_types
fails.

gcc/testsuite/ChangeLog:
PR c++/88795
* g++.dg/template/pr88795.C: New test.

gcc/ChangeLog:
PR c++/88795
* tree.c (build_function_type): Assert that arg_types is not
error_mark_node.

From-SVN: r267957

5 years agoruntime: add padding to FFI type of struct ending with zero-sized field
Ian Lance Taylor [Tue, 15 Jan 2019 23:21:24 +0000 (23:21 +0000)]
runtime: add padding to FFI type of struct ending with zero-sized field

    CL 157557 changes the compiler to add one byte padding to
    non-empty struct ending with a zero-sized field. Add the same
    padding to the FFI type, so reflect.Call works.

    This fixes test/fixedbugs/issue26335.go in the main repo.

    Reviewed-on: https://go-review.googlesource.com/c/158018

From-SVN: r267956

5 years ago[D] Move building of typeof(null) values to a common function.
Iain Buclaw [Tue, 15 Jan 2019 23:02:43 +0000 (23:02 +0000)]
[D] Move building of typeof(null) values to a common function.

gcc/d/ChangeLog:

* d-codegen.cc (build_typeof_null_value): New function.
* d-tree.h (build_typeof_null_value): Declare.
* d-convert.cc (convert_expr): Use build_typeof_null_value.
* expr.cc (ExprVisitor::visit(NullExp)): Likewise.

From-SVN: r267955

5 years agore PR fortran/43136 (Excess copy-in/copy-out with character argument)
Thomas Koenig [Tue, 15 Jan 2019 22:20:26 +0000 (22:20 +0000)]
re PR fortran/43136 (Excess copy-in/copy-out with character argument)

2019-01-15  Thomas Koenig  <tkoenig@gcc.gnu.org>

PR fortran/43136
* resolve.c (resolve_array_ref): Add equal_length argument; set it
if the length of the substring equals that of the orignal
variable.
(resolve_ref): Remove the substring if it is equal in length to
the original variable, unless it is an EXPR_SUBSTRING).

2019-01-15  Thomas Koenig  <tkoenig@gcc.gnu.org>

PR fortran/43136
* gfortran.dg/actual_array_substr_3.f90: New test.

From-SVN: r267954

5 years agore PR fortran/43072 (unneeded temporary (s=s+f(a)))
Thomas Koenig [Tue, 15 Jan 2019 22:18:55 +0000 (22:18 +0000)]
re PR fortran/43072 (unneeded temporary (s=s+f(a)))

2019-01-15  Thomas Koenig  <tkoenig@gcc.gnu.org>

PR fortran/43072
* resolve.c (resolve_array_ref): Add equal_length argument; set it
if the length of the substring equals that of the orignal
variable.
(resolve_ref): Remove the substring if it is equal in length to
the original variable, unless it is an EXPR_SUBSTRING).

2019-01-15  Thomas Koenig  <tkoenig@gcc.gnu.org>

PR fortran/43072
* gfortran.dg/actual_array_substr_3.f90: New test.

From-SVN: r267953

5 years agocompiler, runtime: panic on uncomparable map key, even if map is empty
Ian Lance Taylor [Tue, 15 Jan 2019 20:32:39 +0000 (20:32 +0000)]
compiler, runtime: panic on uncomparable map key, even if map is empty

    This ports https://golang.org/cl/155918 from the master repo.

        runtime: panic on uncomparable map key, even if map is empty

        Reorg map flags a bit so we don't need any extra space for the extra flag.

    This is a pre-req for updating libgo to the Go 1.12beta2 release.

    Reviewed-on: https://go-review.googlesource.com/c/157858

From-SVN: r267950

5 years agore PR fortran/81849 (Size of automatic array argument specified by host-associated...
Steven G. Kargl [Tue, 15 Jan 2019 20:17:35 +0000 (20:17 +0000)]
re PR fortran/81849 (Size of automatic array argument specified by host-associated  variable.)

2019-01-15  Steven G. Kargl  <kargl@gcc.gnu.org>

PR fortran/81849
* resolve.c (resolve_symbol): Host associated varaibles can appear
in the specification statement of a RESULT array.

2019-01-15  Steven G. Kargl  <kargl@gcc.gnu.org>

PR fortran/81849
* gfortran.dg/pr81849.f90: New test.

From-SVN: r267948

5 years agotrans-expr.c (gfc_conv_gfc_desc_to_cfi_desc): Deal with exprs that are indirect refer...
Paul Thomas [Tue, 15 Jan 2019 19:52:08 +0000 (19:52 +0000)]
trans-expr.c (gfc_conv_gfc_desc_to_cfi_desc): Deal with exprs that are indirect references; ie.

2019-01-15  Paul Thomas  <pault@gcc.gnu.org>

* trans-expr.c (gfc_conv_gfc_desc_to_cfi_desc): Deal with exprs
that are indirect references; ie. dummy arguments.

2019-01-15  Paul Thomas  <pault@gcc.gnu.org>

* gfortran.dg/ISO_Fortran_binding_2.c : Change reference to
ISO_Fortran_binding_2.h.

From-SVN: r267946

5 years agore PR c++/88866 (g++.dg/cpp0x/variadic126.C fails with -std=c++2a)
Marek Polacek [Tue, 15 Jan 2019 18:35:01 +0000 (18:35 +0000)]
re PR c++/88866 (g++.dg/cpp0x/variadic126.C fails with -std=c++2a)

PR c++/88866
* g++.dg/cpp0x/variadic126.C: Tweak dg-error.

From-SVN: r267944

5 years ago* g++.dg/ext/utf-cvt-char8_t.C: Pass -fsigned-char.
Jason Merrill [Tue, 15 Jan 2019 18:26:00 +0000 (13:26 -0500)]
* g++.dg/ext/utf-cvt-char8_t.C: Pass -fsigned-char.

From-SVN: r267942

5 years agoPR inline-asm/52813 revisited
Richard Sandiford [Tue, 15 Jan 2019 16:46:54 +0000 (16:46 +0000)]
PR inline-asm/52813 revisited

The original patch for this PR made it an error to list the stack
pointer in the clobber list of an inline asm.  However, the general
feeling seemed to be that going straight to a hard error was too harsh,
since there's quite a bit of existing code that has the clobber.

This patch implements the compromise discussed on IRC of making it
a -Wdeprecated warning instead.

2019-01-15  Richard Sandiford  <richard.sandiford@arm.com>

gcc/
PR inline-asm/52813
* doc/extend.texi: Document that listing the stack pointer in the
clobber list of an asm is a deprecated feature.
* common.opt (Wdeprecated): Moved from c-family/c.opt.
* cfgexpand.c (asm_clobber_reg_is_valid): Issue a -Wdeprecated
warning instead of an error for clobbers of the stack pointer.
Add a note explaining why.

gcc/c-family/
PR inline-asm/52813
* c.opt (Wdeprecated): Move documentation and variable to common.opt.

gcc/d/
PR inline-asm/52813
* lang.opt (Wdeprecated): Reference common.opt instead of c.opt.

gcc/testsuite/
PR inline-asm/52813
* gcc.target/i386/pr52813.c (test1): Turn the diagnostic into a
-Wdeprecated warning and expect a following note:.

From-SVN: r267941

5 years agore PR debug/88046 (ICE in add_data_member_location_attribute at gcc/dwarf2out.c:19237...
Richard Biener [Tue, 15 Jan 2019 16:06:42 +0000 (16:06 +0000)]
re PR debug/88046 (ICE in add_data_member_location_attribute at gcc/dwarf2out.c:19237 since r261885)

2019-01-15  Richard Biener  <rguenther@suse.de>

PR debug/88046
* dwarf2out.c (gen_member_die): Do not generate inheritance
DIEs late.

* g++.dg/lto/pr88046_0.C: New testcase.

From-SVN: r267940

5 years agore PR tree-optimization/88855 (ICE: verify_ssa failed (error: SSA_NAME_OCCURS_IN_ABNO...
Richard Biener [Tue, 15 Jan 2019 15:37:29 +0000 (15:37 +0000)]
re PR tree-optimization/88855 (ICE: verify_ssa failed (error: SSA_NAME_OCCURS_IN_ABNORMAL_PHI should be set))

2019-01-15  Richard Biener  <rguenther@suse.de>

PR tree-optimization/88855
* tree-if-conv.c (combine_blocks): Collect
SSA_NAME_OCCURS_IN_ABNORMAL_PHI from propagated out virtuals.

* gcc.dg/pr88855.c: New testcase.

From-SVN: r267939

5 years agoUpdate value of __cpp_lib_shared_ptr_arrays macro
Jonathan Wakely [Tue, 15 Jan 2019 13:13:11 +0000 (13:13 +0000)]
Update value of __cpp_lib_shared_ptr_arrays macro

* include/bits/shared_ptr_base.h (__cpp_lib_shared_ptr_arrays): Define
as 201611L, because P0497R0 changes are supported.
* include/std/version (__cpp_lib_shared_ptr_arrays): Likewise.

From-SVN: r267938

5 years agoFix more missing or incorrect feature test macros
Jonathan Wakely [Tue, 15 Jan 2019 12:58:19 +0000 (12:58 +0000)]
Fix more missing or incorrect feature test macros

* include/bits/erase_if.h [__cplusplus > 201703L]
(__cpp_lib_erase_if): Only define for C++2a.
* include/std/iterator [__cplusplus >= 201402L && !_GLIBCXX_DEBUG]
(__cpp_lib_null_iterators): Define.
* include/std/version [__cplusplus >= 201402L && !_GLIBCXX_DEBUG]
(__cpp_lib_null_iterators): Define.
[__cpp_impl_destroying_delete] (__cpp_lib_destroying_delete): Define.

From-SVN: r267937

5 years agoFix missing or incorrect feature test macros
Jonathan Wakely [Tue, 15 Jan 2019 12:01:12 +0000 (12:01 +0000)]
Fix missing or incorrect feature test macros

* doc/xml/manual/status_cxx2017.xml: Document P0032R3 and P0307R2
status.
* include/bits/stl_uninitialized.h (__cpp_lib_raw_memory_algorithms):
Define.
* include/std/any (__cpp_lib_any): Define as 201606L, because P0032R3
changes are supported.
* include/std/optional (__cpp_lib_optional): Likewise.
* include/std/variant (__cpp_lib_variant): Likewise.
* include/std/version [!__STRICT_ANSI__]
(__cpp_lib_uncaught_exceptions): Define as long integer.
[__cplusplus >= 201703L] (__cpp_lib_any)
(__cpp_lib_raw_memory_algorithms, __cpp_lib_uncaught_exceptions)
(__cpp_lib_variant): Define for C++17.
[__cplusplus >= 201703L] (__cpp_lib_optional): Update value and define
as long integer.
* libsupc++/exception (__cpp_lib_uncaught_exceptions): Define as long
integer.

From-SVN: r267936

5 years ago* doc/xml/manual/status_cxx2020.xml: Update P1123R0 status.
Jonathan Wakely [Tue, 15 Jan 2019 10:34:18 +0000 (10:34 +0000)]
* doc/xml/manual/status_cxx2020.xml: Update P1123R0 status.

From-SVN: r267935

5 years ago[nvptx] Handle assignment to gang-level reduction variable
Tom de Vries [Tue, 15 Jan 2019 10:11:16 +0000 (10:11 +0000)]
[nvptx] Handle assignment to gang-level reduction variable

2019-01-15  Tom de Vries  <tdevries@suse.de>

PR target/80547
* config/nvptx/nvptx.c (nvptx_goacc_reduction_init): Handle
lhs == NULL_TREE for gang-level reduction.

* testsuite/libgomp.oacc-c-c++-common/gang-reduction-var-assignment.c:
New test.

From-SVN: r267934

5 years agore PR ipa/88788 (Infinite loop in malloc_candidate_p_1 since r264838)
Richard Biener [Tue, 15 Jan 2019 09:37:22 +0000 (09:37 +0000)]
re PR ipa/88788 (Infinite loop in malloc_candidate_p_1 since r264838)

2019-01-15  Richard Biener  <rguenther@suse.de>
    Prathamesh Kulkarni  <prathamesh.kulkarni@linaro.org>

PR ipa/88788
* ipa-pure-const.c (malloc_candidate_p_1): Add parameter visited and
return true if SSA_NAME is already marked in visited bitmap.
(malloc_candidate_p): Pass visited to malloc_candidate_p_1.

Co-Authored-By: Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
From-SVN: r267933

5 years agodecl.c (start_decl): Improve error location.
Paolo Carlini [Tue, 15 Jan 2019 09:36:43 +0000 (09:36 +0000)]
decl.c (start_decl): Improve error location.

/cp
2019-01-15  Paolo Carlini  <paolo.carlini@oracle.com>

* decl.c (start_decl): Improve error location.
* decl2.c (grokfield): Likewise.

/testsuite
2019-01-15  Paolo Carlini  <paolo.carlini@oracle.com>

* g++.dg/diagnostic/typedef-initialized.C: New.

/cp
2019-01-15  Paolo Carlini  <paolo.carlini@oracle.com>

* decl.c (grokdeclarator): Move further up the location_t loc
declaration and use the location when building a TYPE_DECL for
a typedef name.
* decl2.c (grokbitfield): Use DECL_SOURCE_LOCATION in the error
about an ill-formed bit-field as typedef.

/testsuite
2019-01-15  Paolo Carlini  <paolo.carlini@oracle.com>

* g++.dg/diagnostic/bitfld3.C: New.

From-SVN: r267932

5 years agore PR tree-optimization/88775 (Optimize std::string assignment)
Jakub Jelinek [Tue, 15 Jan 2019 08:11:00 +0000 (09:11 +0100)]
re PR tree-optimization/88775 (Optimize std::string assignment)

PR tree-optimization/88775
* match.pd (cmp (convert1?@2 addr@0) (convert2? addr@1)): Optimize
equal == 0 equality pointer comparisons some more if compared in
integral types and either one points to an automatic var and the
other to a global, or we can prove at least one points to the middle
or both point to start or both point to end.

* gcc.dg/tree-ssa/pr88775-1.c: New test.
* gcc.dg/tree-ssa/pr88775-2.c: New test.

From-SVN: r267931

5 years agoLower sampling rate for autofdo bootstrap
Andi Kleen [Tue, 15 Jan 2019 03:49:45 +0000 (03:49 +0000)]
Lower sampling rate for autofdo bootstrap

autofdo create_gcov uses a lot of memory for large sample files.
Since gcc runs quite long the sample files generated during
the bootstrap are fairly ig.

Currently I can't even build make autoprofiledbootstrap on my system at
home because create_gcov needs more than 12GB and runs out of memory.

This should probably be fixed in create_gcov, but for now
lowering the sampling rate works well enough for me. The bootstrap
run is long enough that it gets good enough data in any case.

gcc/:
2019-01-14  Andi Kleen  <ak@linux.intel.com>

* Makefile.in: Lower autofdo sampling rate by 10x.
* Makefile.tpl: Dito.

From-SVN: r267930

5 years agoDaily bump.
GCC Administrator [Tue, 15 Jan 2019 00:16:22 +0000 (00:16 +0000)]
Daily bump.

From-SVN: r267929

5 years agoPR c++/88825 - ICE with bogus function return type deduction.
Marek Polacek [Mon, 14 Jan 2019 22:01:24 +0000 (22:01 +0000)]
PR c++/88825 - ICE with bogus function return type deduction.

* typeck.c (can_do_nrvo_p): Check error_mark_node.

* g++.dg/cpp1y/auto-fn55.C: New test.

From-SVN: r267926

5 years agoc-cppbuiltin.c (c_cpp_builtin): Define __cpp_guaranteed_copy_elision and __cpp_nontyp...
Jakub Jelinek [Mon, 14 Jan 2019 20:12:11 +0000 (21:12 +0100)]
c-cppbuiltin.c (c_cpp_builtin): Define __cpp_guaranteed_copy_elision and __cpp_nontype_template_parameter_auto.

* c-cppbuiltin.c (c_cpp_builtin): Define __cpp_guaranteed_copy_elision
and __cpp_nontype_template_parameter_auto.  Add a comment that
__cpp_template_auto is deprecated.

* g++.dg/cpp1z/feat-cxx1z.C: Add tests for
__cpp_guaranteed_copy_elision and __cpp_nontype_template_parameter_auto
feature test macros.
* g++.dg/cpp2a/feat-cxx2a.C: Likewise.

From-SVN: r267925

5 years agoPR c++/88830 - ICE with abstract class.
Marek Polacek [Mon, 14 Jan 2019 20:09:10 +0000 (20:09 +0000)]
PR c++/88830 - ICE with abstract class.

* decl2.c (maybe_emit_vtables): Check CLASSTYPE_LAZY_DESTRUCTOR.
Fix formatting.

* g++.dg/other/abstract7.C: New test.

From-SVN: r267924

5 years agoImplement P0482R5, char8_t: A type for UTF-8 characters and strings
Tom Honermann [Mon, 14 Jan 2019 19:55:51 +0000 (19:55 +0000)]
Implement P0482R5, char8_t: A type for UTF-8 characters and strings

gcc/cp/
* cvt.c (type_promotes_to): Handle char8_t promotion.
* decl.c (grokdeclarator): Handle invalid type specifier
combinations involving char8_t.
* lex.c (init_reswords): Add char8_t as a reserved word.
* mangle.c (write_builtin_type): Add name mangling for char8_t (Du).
* parser.c (cp_keyword_starts_decl_specifier_p)
(cp_parser_simple_type_specifier): Recognize char8_t as a simple
type specifier.
(cp_parser_string_literal): Use char8_array_type_node for the type
of CPP_UTF8STRING.
(cp_parser_set_decl_spec_type): Tolerate char8_t typedefs in system
headers.
* rtti.c (emit_support_tinfos): type_info support for char8_t.
* tree.c (char_type_p): Recognize char8_t as a character type.
* typeck.c (string_conv_p): Handle conversions of u8 string
literals of char8_t type.
(check_literal_operator_args): Handle UDLs with u8 string literals
of char8_t type.
* typeck2.c (ordinary_char_type_p): New.
(digest_init_r): Disallow initializing a char array with a u8 string
literal.
gcc/c-family/
* c-common.c (c_common_reswords): Add char8_t.
(fix_string_type): Use char8_t for the type of u8 string literals.
(c_common_get_alias_set): char8_t doesn't alias.
(c_common_nodes_and_builtins): Define char8_t as a builtin type in
C++.
(c_stddef_cpp_builtins): Add __CHAR8_TYPE__.
(keyword_begins_type_specifier): Add RID_CHAR8.
* c-common.h (rid): Add RID_CHAR8.
(c_tree_index): Add CTI_CHAR8_TYPE and CTI_CHAR8_ARRAY_TYPE.
Define D_CXX_CHAR8_T and D_CXX_CHAR8_T_FLAGS.
Define char8_type_node and char8_array_type_node.
* c-cppbuiltin.c (cpp_atomic_builtins): Predefine
__GCC_ATOMIC_CHAR8_T_LOCK_FREE.
(c_cpp_builtins): Predefine __cpp_char8_t.
* c-lex.c (lex_string): Use char8_array_type_node as the type of
CPP_UTF8STRING.
(lex_charconst): Use char8_type_node as the type of CPP_UTF8CHAR.
* c-opts.c: If not otherwise specified, enable -fchar8_t when
targeting C++2a.
* c.opt: Add the -fchar8_t command line option.
libiberty/
* cp-demangle.c (cplus_demangle_builtin_types)
(cplus_demangle_type): Add name demangling for char8_t (Du).
* cp-demangle.h: Increase D_BUILTIN_TYPE_COUNT to accommodate the
new char8_t type.

From-SVN: r267923

5 years agoPR target/88638 - FAIL: fsf-nsstring-format-1.s on darwin
Martin Sebor [Mon, 14 Jan 2019 18:44:00 +0000 (18:44 +0000)]
PR target/88638 - FAIL: fsf-nsstring-format-1.s on darwin

gcc/c-family/ChangeLog:

PR target/88638
* c-attribs.c (positional_argument): Call valid_format_string_type_p
and issue errors if it fails.
* c-common.h (valid_format_string_type_p): Declare.
* c-format.c (valid_stringptr_type_p): Rename...
(valid_format_string_type_p): ...to this and make extern.
(handle_format_arg_attribute): Adjust to new name.
(check_format_string): Same.

gcc/testsuite/ChangeLog:

PR target/88638
* gcc.dg/format/attr-8.c: New test.
* gcc.dg/darwin-cfstring-format-1.c: Adjust diagnostics.
* gcc.dg/format/attr-3.c: Same.
* obj-c++.dg/fsf-nsstring-format-1.mm: Same.
* objc.dg/fsf-nsstring-format-1.m: Same.

gcc/ChangeLog:

PR target/88638
* doc/extend.texi (Darwin Format Checks): Clarify.

From-SVN: r267922

5 years agoFix location of tls_wrapper_fn (PR gcov-profile/88263).
Martin Liska [Mon, 14 Jan 2019 18:40:34 +0000 (19:40 +0100)]
Fix location of tls_wrapper_fn (PR gcov-profile/88263).

2019-01-14  Martin Liska  <mliska@suse.cz>

PR gcov-profile/88263
* decl2.c (get_tls_wrapper_fn): Use DECL_SOURCE_LOCATION
as location of the TLS wrapper.
2019-01-14  Martin Liska  <mliska@suse.cz>

PR gcov-profile/88263
* g++.dg/gcov/pr88263-2.C: New test.

From-SVN: r267921

5 years agoinvoke.texi (-Wmemset-transposed-args): Fix typos, adjust wording.
Martin Sebor [Mon, 14 Jan 2019 18:26:32 +0000 (11:26 -0700)]
invoke.texi (-Wmemset-transposed-args): Fix typos, adjust wording.

gcc/ChangeLog:

* invoke.texi (-Wmemset-transposed-args): Fix typos, adjust wording.

From-SVN: r267920

5 years agogenmatch.c (dt_simplify::gen_1): Change dumping dependent on whether we are in (simpl...
Richard Biener [Mon, 14 Jan 2019 13:11:43 +0000 (13:11 +0000)]
genmatch.c (dt_simplify::gen_1): Change dumping dependent on whether we are in (simplify ...) or (match ...) context.

2019-01-14  Richard Biener  <rguenther@suse.de>

* genmatch.c (dt_simplify::gen_1): Change dumping dependent on
whether we are in (simplify ...) or (match ...) context.

From-SVN: r267917

5 years agore PR rtl-optimization/88796 (-fstack-protector* kills RTL DSE opportunities)
Jakub Jelinek [Mon, 14 Jan 2019 12:01:01 +0000 (13:01 +0100)]
re PR rtl-optimization/88796 (-fstack-protector* kills RTL DSE opportunities)

PR rtl-optimization/88796
* emit-rtl.h (struct rtl_data): Add stack_protect_guard_decl field.
* cfgexpand.c (stack_protect_prologue): Initialize
crtl->stack_protect_guard_decl.
* function.c (stack_protect_epilogue): Use it instead of calling
targetm.stack_protect_guard again.
* dse.c (check_mem_read_rtx): Ignore MEM_VOLATILE_P reads from
MEMs with MEM_EXPR equal to crtl->stack_protect_guard or
crtl->stack_protect_guard_decl.
* config/i386/i386.c (ix86_stack_protect_guard): Set TREE_THIS_VOLATILE
on the returned MEM_EXPR.

* gcc.target/i386/pr88796.c: New test.

From-SVN: r267916

5 years agoUndo undesirable part of r267850 2019-01-11 commit.
Jakub Jelinek [Mon, 14 Jan 2019 10:47:09 +0000 (11:47 +0100)]
Undo undesirable part of r267850 2019-01-11 commit.

From-SVN: r267914

5 years ago[D] Merge upstream dmd cd2034cd7
Iain Buclaw [Mon, 14 Jan 2019 10:38:00 +0000 (10:38 +0000)]
[D] Merge upstream dmd cd2034cd7

One fix in the asm statement parser to stop parsing if the end of the
statement has been reached, and moves all inline asm tests to gdc.dg.
These being adjusted where necessary to test the GCC style instead.

gcc/testsuite/ChangeLog:

2019-01-14  Iain Buclaw  <ibuclaw@gdcproject.org>

* gdc.dg/asm1.d: New test.
* gdc.dg/asm2.d: New test.
* gdc.dg/asm3.d: New test.
* gdc.dg/asm4.d: New test.
* lib/gdc.exp (gdc_init): Set gcc_error_prefix and gcc_warning_prefix.

From-SVN: r267913

5 years agore PR libfortran/88776 (Namelist read from stdin: loss of data)
Jerry DeLisle [Mon, 14 Jan 2019 00:22:00 +0000 (00:22 +0000)]
re PR libfortran/88776 (Namelist read from stdin: loss of data)

2019-01-13  Jerry DeLisle  <jvdelisle@gcc.gnu.org>

PR libfortran/88776
* io/open.c (newunit): Free format buffer if the unit specified is for
stdin, stdout, or stderr.

* gfortran.dg/namelist_96.f90: New test.

From-SVN: r267910

5 years agoDaily bump.
GCC Administrator [Mon, 14 Jan 2019 00:16:12 +0000 (00:16 +0000)]
Daily bump.

From-SVN: r267909

5 years agore PR fortran/88803 (gfortran documentation warning: '.' or ',' must follow @xref)
Dominique d'Humieres [Sun, 13 Jan 2019 19:18:46 +0000 (20:18 +0100)]
re PR fortran/88803 (gfortran documentation warning: '.' or ',' must follow @xref)

2019-01-13  Dominique d'Humieres  <dominiq@gcc.gnu.org>

PR fortran/88803
* gfortran.texi: Replace @xref with @ref and adjust the sentence.

From-SVN: r267906

5 years agore PR fortran/57992 (Pointless packing of contiguous arrays for simply contiguous...
Thomas Koenig [Sun, 13 Jan 2019 14:57:39 +0000 (14:57 +0000)]
re PR fortran/57992 (Pointless packing of contiguous arrays for simply contiguous functions results as actual arguments)

2019-01-13  Thomas Koenig  <tkoenig@gcc.gnu.org>

PR fortran/57992
* trans-array.c (gfc_conv_array_parameter):  Do not pack/unpack
functions with contiguous results.

2019-01-13  Thomas Koenig  <tkoenig@gcc.gnu.org>

PR fortran/59345
* gfortran.dg/internal_pack_18.f90: New test.

From-SVN: r267905

5 years agoC-family: Replace "may may" with "may" in warning message
H.J. Lu [Sun, 13 Jan 2019 12:53:47 +0000 (12:53 +0000)]
C-family: Replace "may may" with "may" in warning message

gcc/c-family/

* c-warn.c (warn_for_address_or_pointer_of_packed_member):
Replace "may may" with "may" in warning message.

gcc/testsuite/

* gcc.dg/pr51628-20.c: Updated.
* gcc.dg/pr51628-21.c: Likewise.
* gcc.dg/pr51628-25.c: Likewise.

From-SVN: r267904

5 years agore PR fortran/59345 (_gfortran_internal_pack on compiler generated temps)
Thomas Koenig [Sun, 13 Jan 2019 11:06:03 +0000 (11:06 +0000)]
re PR fortran/59345 (_gfortran_internal_pack on compiler generated temps)

2019-01-13  Thomas Koenig  <tkoenig@gcc.gnu.org>

PR fortran/59345
* trans-array.c (gfc_conv_array_parameter): Remove TODO.  Do not
pack/unpack results of functions which return an explicit-shaped
or allocatable array.

2019-01-13  Thomas Koenig  <tkoenig@gcc.gnu.org>

PR fortran/59345
* gfortran.dg/internal_pack_17.f90: New test.
* gfortran.dg/alloc_comp_auto_array_3.f90: Adjust number of calls
to builtin_free.

From-SVN: r267903

5 years agore PR fortran/61765 ([F03] Rejects valid BIND(C) ENTRY)
Steven G. Kargl [Sun, 13 Jan 2019 04:02:46 +0000 (04:02 +0000)]
re PR fortran/61765 ([F03] Rejects valid BIND(C) ENTRY)

2019-01-12  Steven G. Kargl  <kargl@gcc.gnu.org>

PR fortran/61765
* resolve.c (gfc_verify_binding_labels): Break if-elseif-elseif
structure into independent if's with a return to simplify logic.
Avoid a check for ENTRY name with bind(c).

2019-01-12  Steven G. Kargl  <kargl@gcc.gnu.org>

PR fortran/61765
* gfortran.dg/pr61765.f90: New test.

From-SVN: r267902

5 years agoDaily bump.
GCC Administrator [Sun, 13 Jan 2019 00:16:28 +0000 (00:16 +0000)]
Daily bump.

From-SVN: r267901

5 years agore PR libfortran/88776 (Namelist read from stdin: loss of data)
Jerry DeLisle [Sat, 12 Jan 2019 23:06:47 +0000 (23:06 +0000)]
re PR libfortran/88776 (Namelist read from stdin: loss of data)

2019-01-12  Jerry DeLisle  <jvdelisle@gcc.gnu.org>

PR libfortran/88776
* io/list_read.c (namelist_read): Use nml_err_ret path on read error
not based on stdin_unit.

From-SVN: r267898

5 years ago[nvptx] Enable setting vector length using -fopenacc-dim -- testcases
Tom de Vries [Sat, 12 Jan 2019 22:19:31 +0000 (22:19 +0000)]
[nvptx] Enable setting vector length using -fopenacc-dim -- testcases

Add some test-cases that set vector length using -fopenacc-dim.

2019-01-12  Tom de Vries  <tdevries@suse.de>

* testsuite/libgomp.oacc-c-c++-common/pr85486-2.c: New test.
* testsuite/libgomp.oacc-c-c++-common/vector-length-128-2.c: New test.
* testsuite/libgomp.oacc-c-c++-common/vector-length-128-5.c: New test.
* testsuite/libgomp.oacc-fortran/gemm-2.f90: New test.

From-SVN: r267897

5 years ago[nvptx] Enable setting vector length using -fopenacc-dim
Tom de Vries [Sat, 12 Jan 2019 22:19:15 +0000 (22:19 +0000)]
[nvptx] Enable setting vector length using -fopenacc-dim

Enable setting vector length using -fopenacc-dim, f.i. -fopenacc-dim=::128.

2019-01-12  Tom de Vries  <tdevries@suse.de>

* config/nvptx/nvptx.c (nvptx_goacc_validate_dims_1): Alow setting
vector length using -fopenacc-dim.

* plugin/plugin-nvptx.c (nvptx_exec): Update error message.

From-SVN: r267896

5 years ago[nvptx] Add vector_length 64 test-cases
Tom de Vries [Sat, 12 Jan 2019 22:19:02 +0000 (22:19 +0000)]
[nvptx] Add vector_length 64 test-cases

Add some test-cases using vector_length 64.

2019-01-12  Tom de Vries  <tdevries@suse.de>

* testsuite/libgomp.oacc-c-c++-common/vector-length-64-1.c: New test.
* testsuite/libgomp.oacc-c-c++-common/vector-length-64-2.c: New test.
* testsuite/libgomp.oacc-c-c++-common/vector-length-64-3.c: New test.

From-SVN: r267895

5 years ago[nvptx] Force vl32 if calling vector-partitionable routines -- test-cases
Tom de Vries [Sat, 12 Jan 2019 22:18:50 +0000 (22:18 +0000)]
[nvptx] Force vl32 if calling vector-partitionable routines -- test-cases

Add test-cases for "[nvptx] Force vl32 if calling vector-partitionable
routines".

2019-01-12  Tom de Vries  <tdevries@suse.de>

PR target/85486
* testsuite/libgomp.oacc-c-c++-common/pr85486-3.c: New test.
* testsuite/libgomp.oacc-c-c++-common/pr85486.c: New test.

From-SVN: r267894

5 years ago[nvptx] Don't emit barriers for empty loops -- test-cases
Tom de Vries [Sat, 12 Jan 2019 22:18:39 +0000 (22:18 +0000)]
[nvptx] Don't emit barriers for empty loops -- test-cases

Add test-cases for PR85381.

2019-01-12  Tom de Vries  <tdevries@suse.de>

PR target/85381
* testsuite/libgomp.oacc-c-c++-common/pr85381-5.c: New test.
* testsuite/libgomp.oacc-c-c++-common/pr85381.c: New test.

From-SVN: r267893

5 years ago[nvptx] Enable large vectors -- reduction testcases
Tom de Vries [Sat, 12 Jan 2019 22:18:27 +0000 (22:18 +0000)]
[nvptx] Enable large vectors -- reduction testcases

Add various reduction test-cases with vector length 128.

2019-01-12  Tom de Vries  <tdevries@suse.de>

* testsuite/libgomp.oacc-c-c++-common/vred2d-128.c: New test.
* testsuite/libgomp.oacc-fortran/gemm.f90: New test.
* testsuite/libgomp.oacc-c-c++-common/vector-length-128-10.c: New test.

From-SVN: r267892

5 years ago[nvptx] Enable large vectors -- test-cases
Tom de Vries [Sat, 12 Jan 2019 22:18:11 +0000 (22:18 +0000)]
[nvptx] Enable large vectors -- test-cases

Add various test-cases with vector length 128.

2019-01-12  Tom de Vries  <tdevries@suse.de>

* testsuite/libgomp.oacc-c-c++-common/vector-length-128-7.c: New test.
* testsuite/libgomp.oacc-c-c++-common/vector-length-128-4.c: New test.
* testsuite/libgomp.oacc-c-c++-common/vector-length-128-6.c: New test.

From-SVN: r267891

5 years ago[nvptx] Update insufficient launch message for variable vector_length
Tom de Vries [Sat, 12 Jan 2019 22:18:00 +0000 (22:18 +0000)]
[nvptx] Update insufficient launch message for variable vector_length

Update message in nvptx libgomp plugin about insufficient resources to launch
kernel, to accommodate for the fact the vector_length can now be variable.

2019-01-12  Tom de Vries  <tdevries@suse.de>

* plugin/plugin-nvptx.c (nvptx_exec): Update insufficient hardware
resources diagnostic.

From-SVN: r267890

5 years ago[nvptx] Enable large vectors
Tom de Vries [Sat, 12 Jan 2019 22:17:42 +0000 (22:17 +0000)]
[nvptx] Enable large vectors

Allow vector_length clauses to accept values larger than warp size.  Note that
this does not enable setting vector_length to values larger than warp size using
-fopenacc-dim.

2019-01-12  Tom de Vries  <tdevries@suse.de>

* config/nvptx/nvptx.c (nvptx_goacc_validate_dims): Take larger vector
lengths into account.

* testsuite/libgomp.oacc-c-c++-common/vector-length-128-1.c: Expect
vector length to be 128.
* testsuite/libgomp.oacc-c-c++-common/parallel-dims.c: Expect vector
length 2097152 to be reduced to 1024 instead of 32.

From-SVN: r267889

5 years agodecl.c (cp_finish_decl): Improve error location.
Paolo Carlini [Sat, 12 Jan 2019 21:59:50 +0000 (21:59 +0000)]
decl.c (cp_finish_decl): Improve error location.

/cp
2019-01-12  Paolo Carlini  <paolo.carlini@oracle.com>

* decl.c (cp_finish_decl): Improve error location.
* decl2.c (grokfield): Likewise, improve two locations.

/testsuite
2019-01-12  Paolo Carlini  <paolo.carlini@oracle.com>

* g++.dg/cpp0x/pr62101.C: Test locations too.
* g++.dg/inherit/pure1.C: Likewise.

From-SVN: r267888

5 years agodecl.c (cp_finish_decl): Improve error location.
Paolo Carlini [Sat, 12 Jan 2019 21:59:03 +0000 (21:59 +0000)]
decl.c (cp_finish_decl): Improve error location.

/cp
2019-01-12  Paolo Carlini  <paolo.carlini@oracle.com>

* decl.c (cp_finish_decl): Improve error location.
* decl2.c (grokfield): Likewise, improve two locations.

/testsuite
2019-01-12  Paolo Carlini  <paolo.carlini@oracle.com>

* g++.dg/cpp0x/pr62101.C: Test locations too.
* g++.dg/inherit/pure1.C: Likewise.

From-SVN: r267887

5 years agognu.h (TARGET_THREAD_SSP_OFFSET): Define.
Svante Signell [Sat, 12 Jan 2019 21:17:56 +0000 (21:17 +0000)]
gnu.h (TARGET_THREAD_SSP_OFFSET): Define.

* config/i386/gnu.h (TARGET_THREAD_SSP_OFFSET): Define.
(TARGET_CAN_SPLIT_STACK): Define.
(TARGET_THREAD_SPLIT_STACK_OFFSET): Define.

From-SVN: r267886

5 years agoDocument merge process for dmd, druntime and phobos.
Iain Buclaw [Sat, 12 Jan 2019 19:50:41 +0000 (19:50 +0000)]
Document merge process for dmd, druntime and phobos.

gcc/d/ChangeLog:

* README.gcc: New file.

libphobos/ChangeLog:

* README.gcc: New file.

From-SVN: r267885

5 years agoISO_Fortran_binding_2.f90: Remove because of reports of ICEs.
Paul Thomas [Sat, 12 Jan 2019 18:34:30 +0000 (18:34 +0000)]
ISO_Fortran_binding_2.f90: Remove because of reports of ICEs.

2019-01-12  Paul Thomas  <pault@gcc.gnu.org>

* gfortran.dg/ISO_Fortran_binding_2.f90 : Remove because of
reports of ICEs.
* gfortran.dg/ISO_Fortran_binding_2.c : Ditto.

From-SVN: r267884

5 years ago* params.def (inline-unit-growth): Set to 40.
Jan Hubicka [Sat, 12 Jan 2019 18:18:23 +0000 (19:18 +0100)]
* params.def (inline-unit-growth): Set to 40.

From-SVN: r267883

5 years agotree-ssa-loop-ivopts.c (find_inv_vars): Fix a comment typo.
Jakub Jelinek [Sat, 12 Jan 2019 16:55:45 +0000 (17:55 +0100)]
tree-ssa-loop-ivopts.c (find_inv_vars): Fix a comment typo.

* tree-ssa-loop-ivopts.c (find_inv_vars): Fix a comment typo.

* c-typeck.c (convert_for_assignment): Fix a comment typo.

From-SVN: r267882

5 years agogfortran.texi: Add description in sections on TS 29113 and further interoperability...
Paul Thomas [Sat, 12 Jan 2019 15:25:52 +0000 (15:25 +0000)]
gfortran.texi: Add description in sections on TS 29113 and further interoperability with C.

2019-01-12  Paul Thomas  <pault@gcc.gnu.org>

* gfortran.texi : Add description in sections on TS 29113 and
further interoperability with C.
* trans-array.c (gfc_conv_descriptor_attribute): New function.
(gfc_get_dataptr_offset): Remove static function attribute.
* trans-array.h : Add prototypes for above functions.
* trans-decl.c : Add declarations for the library functions
cfi_desc_to_gfc_desc and gfc_desc_to_cfi_desc.
* trans-expr.c (gfc_conv_gfc_desc_to_cfi_desc): New function.
(gfc_conv_procedure_call): Call it for scalar and array actual
arguments, when the formal arguments are bind_c with assumed
shape or assumed rank.
* trans.h : External declarations for gfor_fndecl_cfi_to_gfc
and gfor_fndecl_gfc_to_cfi.

2019-01-12  Paul Thomas  <pault@gcc.gnu.org>

* gfortran.dg/ISO_Fortran_binding_1.f90 : New test.
* gfortran.dg/ISO_Fortran_binding_1.c : Auxilliary file for test.
* gfortran.dg/ISO_Fortran_binding_2.f90 : New test.
* gfortran.dg/ISO_Fortran_binding_2.c : Auxilliary file for test.
* gfortran.dg/bind_c_array_params_2.f90 : Change search string
for dump tree scan.

2019-01-12  Paul Thomas  <pault@gcc.gnu.org>

* ISO_Fortran_binding.h : New file.
* Makefile.am : Include ISO_Fortran_binding.c in the list of
files to compile.
* Makefile.in : Regenerated.
* gfortran.map : Add _gfortran_cfi_desc_to_gfc_desc,
_gfortran_gfc_desc_to_cfi_desc and the CFI API functions.
* runtime/ISO_Fortran_binding.c : New file containing the new
functions added to the map.

From-SVN: r267881

5 years ago[nvptx] Apply vector-partitionable routines workaround to default vl
Tom de Vries [Sat, 12 Jan 2019 10:27:34 +0000 (10:27 +0000)]
[nvptx] Apply vector-partitionable routines workaround to default vl

Make "[nvptx] Force vl32 if calling vector-partitionable routines" work as well
if vector length is set by modifying PTX_DEFAULT_VECTOR_LENGTH.

2019-01-12  Tom de Vries  <tdevries@suse.de>

* config/nvptx/nvptx.c (nvptx_goacc_validate_dims_1): In offloading
region calling vector-partitionable routine, set default_vector_length
to WARP_SIZE.

From-SVN: r267879

5 years ago[nvptx] Allow default vl to be overridden in nvptx_goacc_validate_dims_1
Tom de Vries [Sat, 12 Jan 2019 10:27:23 +0000 (10:27 +0000)]
[nvptx] Allow default vl to be overridden in nvptx_goacc_validate_dims_1

In nvptx_goacc_validate_dims_1, allow oacc_default_dims[DIM_VECTOR] to be
overridden, by assigning it to a new variable default_vector_length at the
start, and using it at the end.

2019-01-12  Tom de Vries  <tdevries@suse.de>

* config/nvptx/nvptx.c (nvptx_goacc_validate_dims_1): Add an use new
variable default_vector_length.

From-SVN: r267878

5 years ago[nvptx] Verify dimension limits after applying defaults
Tom de Vries [Sat, 12 Jan 2019 10:27:12 +0000 (10:27 +0000)]
[nvptx] Verify dimension limits after applying defaults

There's a problem in oacc_validate_dims that when f.i. the worker dimension
is set using -fopenacc-dim=:32, and the vector_length is set using a
"vector_length (128)" clause, the compiler combines, accepts and emits the
values, while the combination of the two is invalid.

The reason for this is that while oacc_validate_dims validates the dimensions
using targetm.goacc.validate_dims before applying default or minimum values,
it does not do so afterwards.

Work around this in the nvptx port by applying the defaults from
oacc_default_dims at the end of nvptx_goacc_validate_dims_1, as
oacc_validate_dims would do it, and then apply the dimensions limits.

2019-01-12  Tom de Vries  <tdevries@suse.de>

PR middle-end/88703
* config/nvptx/nvptx.c (nvptx_goacc_validate_dims_1): Apply defaults
from oacc_default_dims, as oacc_validate_dims would do it, and apply
dimensions limits.

From-SVN: r267877

5 years ago[openacc] Add used parameter to TARGET_GOACC_VALIDATE_DIMS
Tom de Vries [Sat, 12 Jan 2019 10:27:00 +0000 (10:27 +0000)]
[openacc] Add used parameter to TARGET_GOACC_VALIDATE_DIMS

Add a used parameter to TARGET_GOACC_VALIDATE_DIMS, allowing a target to make
decisions in the hook implementation based on whether a dimension is used or
not.

2019-01-12  Tom de Vries  <tdevries@suse.de>

* config/nvptx/nvptx.c (nvptx_goacc_validate_dims_1)
(nvptx_goacc_validate_dims): Add used parameter.
* doc/tm.texi: Regenerate.
* omp-offload.c (oacc_parse_default_dims, oacc_validate_dims): Add
argument to call to targetm.goacc.validate_dims.
(default_goacc_validate_dims): Add used
parameter.
* target.def (validate_dims): Add used parameter in DEFHOOK.
* targhooks.h (default_goacc_validate_dims): Add used parameter.

From-SVN: r267876

5 years agore PR libfortran/88807 (misleading indentation warnings building libgfortran)
Jakub Jelinek [Sat, 12 Jan 2019 09:15:54 +0000 (10:15 +0100)]
re PR libfortran/88807 (misleading indentation warnings building libgfortran)

PR libfortran/88807
* m4/minloc0.m4: Reindent to avoid -Wmisleading-indentation warnings.
* generated/minloc0_4_i1.c: Regenerated.
* generated/minloc0_4_i2.c: Regenerated.
* generated/minloc0_4_i4.c: Regenerated.
* generated/minloc0_4_i8.c: Regenerated.
* generated/minloc0_4_i16.c: Regenerated.
* generated/minloc0_4_r4.c: Regenerated.
* generated/minloc0_4_r8.c: Regenerated.
* generated/minloc0_4_r10.c: Regenerated.
* generated/minloc0_4_r16.c: Regenerated.
* generated/minloc0_8_i1.c: Regenerated.
* generated/minloc0_8_i2.c: Regenerated.
* generated/minloc0_8_i4.c: Regenerated.
* generated/minloc0_8_i8.c: Regenerated.
* generated/minloc0_8_i16.c: Regenerated.
* generated/minloc0_8_r4.c: Regenerated.
* generated/minloc0_8_r8.c: Regenerated.
* generated/minloc0_8_r10.c: Regenerated.
* generated/minloc0_8_r16.c: Regenerated.
* generated/minloc0_16_i1.c: Regenerated.
* generated/minloc0_16_i2.c: Regenerated.
* generated/minloc0_16_i4.c: Regenerated.
* generated/minloc0_16_i8.c: Regenerated.
* generated/minloc0_16_i16.c: Regenerated.
* generated/minloc0_16_r4.c: Regenerated.
* generated/minloc0_16_r8.c: Regenerated.
* generated/minloc0_16_r10.c: Regenerated.
* generated/minloc0_16_r16.c: Regenerated.

From-SVN: r267875

5 years agoRemove svn:executable property from a couple of text files
Jakub Jelinek [Sat, 12 Jan 2019 00:20:04 +0000 (01:20 +0100)]
Remove svn:executable property from a couple of text files
which shouldn't be executable.

From-SVN: r267873

5 years agoDaily bump.
GCC Administrator [Sat, 12 Jan 2019 00:16:18 +0000 (00:16 +0000)]
Daily bump.

From-SVN: r267872

5 years agoPR libstdc++/88811 fix typo introduced in r266569
Jonathan Wakely [Sat, 12 Jan 2019 00:12:20 +0000 (00:12 +0000)]
PR libstdc++/88811 fix typo introduced in r266569

PR libstdc++/88811
PR libstdc++/83306
* src/filesystem/path.cc: Fix typo. If first path is empty, show []
before second path.
* testsuite/experimental/filesystem/filesystem_error/cons.cc: New
test.

From-SVN: r267868

5 years agoDocument C++20 library status
Jonathan Wakely [Fri, 11 Jan 2019 23:41:15 +0000 (23:41 +0000)]
Document C++20 library status

* doc/xml/manual/intro.xml: Include new section.
* doc/xml/manual/status_cxx2017.xml: Document more
implementation-defined properties of the library.
* doc/xml/manual/status_cxx2020.xml: Document C++2a status.
* doc/html/*: Regenerate.

From-SVN: r267867

5 years agoP0357R3 reference_wrapper for incomplete types
Jonathan Wakely [Fri, 11 Jan 2019 23:41:11 +0000 (23:41 +0000)]
P0357R3 reference_wrapper for incomplete types

This patch implements the C++2a proposal to allow incomplete types in
std::reference_wrapper, which was previously undefined.

The change cannot be implemented for earlier standards, because prior to
C++2a std::reference_wrapper has a weak result type, so must inspect the
template argument to see if it defines a nested result_type member. That
is deprecated (but still required) in C++17, and removed from C++2a.

The removal of the base class from reference_wrapper is a potential ABI
change, as it could alter the layout of a type which derives from
reference_wrapper<T> and from an empty type with _Weak_result_type<T> as
a base class.  Previously the repeated _Weak_result_type<T> base class
would have prevented the empty base-class optimization, but if
reference_wrapper<T> no longer derives from it, the empty class could be
placed at the same address as the reference_wrapper<T> base.  In
practice, the only types which derive from _Weak_result_type or from
_Reference_wrapper_base_memfun or any of its base classes are non-empty
types defined in libstdc++: std::reference_wrapper, std::function, and
std::_Bind. As they are non-empty types, they are not eligible for EBO
anyway.

* include/bits/refwrap.h [__cplusplus > 201703L]
(_Refwrap_base_arg1, _Refwrap_base_arg2, _Reference_wrapper_base)
(_Reference_wrapper_base_memfun): Do not define for C++2a.
(reference_wrapper): Do not derive from _Reference_wrapper_base_memfun
for C++2a.
(reference_wrapper::operator()): Add static assertion.
* testsuite/20_util/reference_wrapper/incomplete.cc: New test.

From-SVN: r267866

5 years agoP0972R0 <chrono> zero(), min(), and max() should be noexcept
Jonathan Wakely [Fri, 11 Jan 2019 23:41:05 +0000 (23:41 +0000)]
P0972R0 <chrono> zero(), min(), and max() should be noexcept

This paper has been included in the C++20 draft, but the changes to add
noexcept can be made unconditionally, to apply for C++11 too.

* include/std/chrono (duration_values::zero(), duration_values::min())
(duration_values::max()): Add noexcept.
(duration::zero(), duration::min(), duration::max()): Likewise.
(time_point::zero(), time_point::min(), time_point::max()): Likewise.
* testsuite/20_util/duration/requirements/noexcept.cc: New test.
* testsuite/20_util/time_point/requirements/noexcept.cc: New test.

From-SVN: r267865

5 years agore PR fortran/35031 (ELEMENTAL procedure with BIND(C))
Steven G. Kargl [Fri, 11 Jan 2019 23:41:04 +0000 (23:41 +0000)]
re PR fortran/35031 (ELEMENTAL procedure with BIND(C))

2019-01-11  Steven G. Kargl  <kargl@gcc.gnu.org>

PR fortran/35031
* decl.c (gfc_match_entry): Check for F2018:C1546.  Fix nearby
mis-indentation.

2019-01-11  Steven G. Kargl  <kargl@gcc.gnu.org>

PR fortran/35031
* gfortran.dg/pr35031.f90: new test.

From-SVN: r267864

5 years agoFix location of __cpp_lib_erase_if macro
Jonathan Wakely [Fri, 11 Jan 2019 23:40:58 +0000 (23:40 +0000)]
Fix location of __cpp_lib_erase_if macro

This macro should only be defined for C++2a, not C++17.

* include/std/version (__cpp_lib_erase_if): Move to C++20 group.

From-SVN: r267863

5 years agoPR c++/88692, c++/87882 - -Wredundant-move false positive with *this.
Marek Polacek [Fri, 11 Jan 2019 23:21:40 +0000 (23:21 +0000)]
PR c++/88692, c++/87882 - -Wredundant-move false positive with *this.

* typeck.c (maybe_warn_pessimizing_move): Return if ARG isn't
ADDR_EXPR.

* g++.dg/cpp0x/Wredundant-move5.C: New test.
* g++.dg/cpp0x/Wredundant-move6.C: New test.

From-SVN: r267862

5 years agocompiler: pad structs ending with zero-sized field
Ian Lance Taylor [Fri, 11 Jan 2019 23:16:38 +0000 (23:16 +0000)]
compiler: pad structs ending with zero-sized field

    For a struct with zero-sized last field, the address of the
    field falls out of the object boundary, which confuses the
    garbage collector. Pad an extra byte in this case.

    Reviewed-on: https://go-review.googlesource.com/c/157557

From-SVN: r267861

5 years agoPR c++/88312 - pack expansion of decltype.
Jason Merrill [Fri, 11 Jan 2019 22:37:01 +0000 (17:37 -0500)]
PR c++/88312 - pack expansion of decltype.

The standard doesn't really talk about an expression depending on the number
of elements of a pack, but that's definitely an important form of template
argument dependence.

* pt.c (instantiation_dependent_r): A template non-type parameter
pack is instantiation-dependent.

From-SVN: r267860

5 years agoPR c++/88613 - ICE with use of const var in lambda.
Jason Merrill [Fri, 11 Jan 2019 22:36:20 +0000 (17:36 -0500)]
PR c++/88613 - ICE with use of const var in lambda.

The issue here was that we were cp_folding a location wrapper around a
lambda capture proxy before it had been mark_rvalue_used.  I considered
adding mark_rvalue_use calls to build_new_op_1, but it seems appropriate to
have them in cp_fold_maybe_rvalue when we know we're trying to produce an
rvalue.

The change to mark_use is for a related issue: when we change the operand of
the location wrapper from VAR_DECL to INTEGER_CST, we need the TREE_CODE of
the location wrapper to change as well, from VIEW_CONVERT_EXPR to
NON_LVALUE_EXPR.

* expr.c (mark_use): Fix location wrapper handling.
* cp-gimplify.c (cp_fold_maybe_rvalue): Call mark_rvalue_use.

From-SVN: r267859

5 years agore PR middle-end/85956 (ICE in wide_int_to_tree_1, at tree.c:1549)
Jakub Jelinek [Fri, 11 Jan 2019 21:03:53 +0000 (22:03 +0100)]
re PR middle-end/85956 (ICE in wide_int_to_tree_1, at tree.c:1549)

PR middle-end/85956
PR lto/88733
* tree-inline.h (struct copy_body_data): Add adjust_array_error_bounds
field.
* tree-inline.c (remap_type_1): Formatting fix.  If TYPE_MAX_VALUE of
ARRAY_TYPE's TYPE_DOMAIN is newly error_mark_node, replace it with
a dummy "omp dummy var" variable if id->adjust_array_error_bounds.
* omp-low.c (new_omp_context): Set cb.adjust_array_error_bounds.
fortran/
* trans-openmp.c: Include attribs.h.
(gfc_walk_alloc_comps, gfc_omp_clause_linear_ctor): Handle
VAR_DECL max bound with "omp dummy var" attribute like NULL or
error_mark_node - recompute number of elts independently.
testsuite/
* c-c++-common/gomp/pr85956.c: New test.
* g++.dg/gomp/pr88733.C: New test.

From-SVN: r267858

5 years agoPR C++/88114 Gen destructor of an abstract class
Tobias Burnus [Fri, 11 Jan 2019 19:40:13 +0000 (20:40 +0100)]
PR C++/88114 Gen destructor of an abstract class

        PR C++/8811
        * decl2.c (maybe_emit_vtables): If needed, generate code for
        the destructor of an abstract class.
        (mark_used): Update comment for older function-name change.

        PR C++/88114
        * g++.dg/cpp0x/defaulted61.C: New.
        * g++.dg/cpp0x/defaulted62.C: New.

From-SVN: r267855

5 years agore PR rtl-optimization/87305 (Segfault in end_hard_regno in setup_live_pseudos_and_sp...
Vladimir Makarov [Fri, 11 Jan 2019 19:25:31 +0000 (19:25 +0000)]
re PR rtl-optimization/87305 (Segfault in end_hard_regno in setup_live_pseudos_and_spill_after_risky_transforms on aarch64 big-endian)

2019-01-11  Vladimir Makarov  <vmakarov@redhat.com>

PR rtl-optimization/87305
* lra-assigns.c
(setup_live_pseudos_and_spill_after_risky_transforms): Add code
for little endian pseudos used as paradoxical subreg.

From-SVN: r267854

5 years agore PR tree-optimization/88693 (Wrong code since r263018)
Jakub Jelinek [Fri, 11 Jan 2019 19:04:32 +0000 (20:04 +0100)]
re PR tree-optimization/88693 (Wrong code since r263018)

PR tree-optimization/88693
* tree-ssa-strlen.c (get_min_string_length): Don't set *full_string_p
for STRING_CSTs that don't contain any NUL characters in the first
TREE_STRING_LENGTH bytes.

* gcc.c-torture/execute/pr88693.c: New test.

From-SVN: r267852

5 years agore PR target/88777 (Out-of-range offsets building glibc test-tgmath2.c for hppa-linux...
Alan Modra [Fri, 11 Jan 2019 18:44:00 +0000 (05:14 +1030)]
re PR target/88777 (Out-of-range offsets building glibc test-tgmath2.c for hppa-linux-gnu)

PR 88777
PR 88614
* genattrtab.c (min_fn): Don't translate values.
(min_attr_value): Return INT_MAX when the value can't be calculated.
Return minimum among any values that can be calculated.
(max_attr_value): Adjust.

From-SVN: r267851

5 years ago* Makefile.in (PLUGIN_HEADERS): Add $(INSN_ATTR_H).
Jakub Jelinek [Fri, 11 Jan 2019 17:01:25 +0000 (18:01 +0100)]
* Makefile.in (PLUGIN_HEADERS): Add $(INSN_ATTR_H).

From-SVN: r267850

5 years agoRemove trailing whitespace from latest commit.
Jakub Jelinek [Fri, 11 Jan 2019 16:55:13 +0000 (17:55 +0100)]
Remove trailing whitespace from latest commit.

From-SVN: r267849

5 years agoaarch64.c (aarch64_simd_call_p): New function.
Steve Ellcey [Fri, 11 Jan 2019 16:50:17 +0000 (16:50 +0000)]
aarch64.c (aarch64_simd_call_p): New function.

2019-01-11  Steve Ellcey  <sellcey@marvell.com>

* config/aarch64/aarch64.c (aarch64_simd_call_p): New function.
(aarch64_hard_regno_call_part_clobbered): Add insn argument.
(aarch64_return_call_with_max_clobbers): New function.
(TARGET_RETURN_CALL_WITH_MAX_CLOBBERS): New macro.
* config/avr/avr.c (avr_hard_regno_call_part_clobbered): Add insn
argument.
* config/i386/i386.c (ix86_hard_regno_call_part_clobbered): Ditto.
* config/mips/mips.c (mips_hard_regno_call_part_clobbered): Ditto.
* config/rs6000/rs6000.c (rs6000_hard_regno_call_part_clobbered): Ditto.
* config/s390/s390.c (s390_hard_regno_call_part_clobbered): Ditto.
* cselib.c (cselib_process_insn): Add argument to
targetm.hard_regno_call_part_clobbered call.
* ira-conflicts.c (ira_build_conflicts): Ditto.
* ira-costs.c (ira_tune_allocno_costs): Ditto.
* lra-constraints.c (inherit_reload_reg): Ditto.
* lra-int.h (struct lra_reg): Add call_insn field, remove call_p field.
* lra-lives.c (check_pseudos_live_through_calls): Add call_insn
argument.  Call targetm.return_call_with_max_clobbers.
Add argument to targetm.hard_regno_call_part_clobbered call.
(calls_have_same_clobbers_p): New function.
(process_bb_lives): Add call_insn and last_call_insn variables.
Pass call_insn to check_pseudos_live_through_calls.
Modify if stmt to check targetm.return_call_with_max_clobbers.
Update setting of flush variable.
(lra_create_live_ranges_1): Set call_insn to NULL instead of call_p
to false.
* lra.c (initialize_lra_reg_info_element): Set call_insn to NULL.
* regcprop.c (copyprop_hardreg_forward_1): Add argument to
        targetm.hard_regno_call_part_clobbered call.
* reginfo.c (choose_hard_reg_mode): Ditto.
* regrename.c (check_new_reg_p): Ditto.
* reload.c (find_equiv_reg): Ditto.
* reload1.c (emit_reload_insns): Ditto.
* sched-deps.c (deps_analyze_insn): Ditto.
* sel-sched.c (init_regs_for_mode): Ditto.
(mark_unavailable_hard_regs): Ditto.
* targhooks.c (default_dwarf_frame_reg_mode): Ditto.
* target.def (hard_regno_call_part_clobbered): Add insn argument.
(return_call_with_max_clobbers): New target function.
* doc/tm.texi: Regenerate.
* doc/tm.texi.in (TARGET_RETURN_CALL_WITH_MAX_CLOBBERS): New hook.
* hooks.c (hook_bool_uint_mode_false): Change to
hook_bool_insn_uint_mode_false.
* hooks.h (hook_bool_uint_mode_false): Ditto.

From-SVN: r267848

5 years agotype_traits (__cpp_lib_is_constant_evaluated): Define.
Jakub Jelinek [Fri, 11 Jan 2019 16:01:23 +0000 (17:01 +0100)]
type_traits (__cpp_lib_is_constant_evaluated): Define.

* include/std/type_traits (__cpp_lib_is_constant_evaluated): Define.
* include/std/version (__cpp_lib_is_constant_evaluated): Define.

From-SVN: r267847

5 years agoaarch64.c (aarch64_simd_call_p): New function.
Steve Ellcey [Fri, 11 Jan 2019 15:53:02 +0000 (15:53 +0000)]
aarch64.c (aarch64_simd_call_p): New function.

2019-01-11  Steve Ellcey  <sellcey@marvell.com>

* config/aarch64/aarch64.c (aarch64_simd_call_p): New function.
(aarch64_remove_extra_call_preserved_regs): New function.
(TARGET_REMOVE_EXTRA_CALL_PRESERVED_REGS): New macro.
* doc/tm.texi.in (TARGET_REMOVE_EXTRA_CALL_PRESERVED_REGS): New hook.
* doc/tm.texi: Regenerate.
* final.c (get_call_reg_set_usage): Call new hook.
* target.def (remove_extra_call_preserved_regs): New hook.
* targhooks.c (default_remove_extra_call_preserved_regs): New function.
* targhooks.h (default_remove_extra_call_preserved_regs): New function.

From-SVN: r267846

5 years agoPR libstdc++/88802 define std::hash<nullptr_t> for C++17
Jonathan Wakely [Fri, 11 Jan 2019 14:54:49 +0000 (14:54 +0000)]
PR libstdc++/88802 define std::hash<nullptr_t> for C++17

PR libstdc++/88802
* include/bits/functional_hash.h (hash<nullptr_t>): Define
specialization for C++17 (P0513R0, LWG 2817).
* testsuite/20_util/hash/nullptr.cc: New test.

From-SVN: r267845

5 years agoFix arm testism regression.
Tamar Christina [Fri, 11 Jan 2019 14:20:58 +0000 (14:20 +0000)]
Fix arm testism regression.

gcc/testsuite/ChangeLog:

2019-01-11  Tamar Christina  <tamar.christina@arm.com>

* gcc.target/aarch64/advsimd-intrinsics/vector-complex_f16.c: Require neon
and add options.

From-SVN: r267843

5 years agoBetter distinguish OpenACC and OpenMP sections in libgomp.texi
Thomas Schwinge [Fri, 11 Jan 2019 13:55:01 +0000 (14:55 +0100)]
Better distinguish OpenACC and OpenMP sections in  libgomp.texi

2019-01-11  Thomas Schwinge  <thomas@codesourcery.com>
            James Norris  <jnorris@codesourcery.com>

* libgomp.texi: Better distinguish OpenACC and OpenMP "Runtime
Library Routines", and "Environment Variables".

Co-Authored-By: James Norris <jnorris@codesourcery.com>
From-SVN: r267841

5 years agoAdd a testcase (PR middle-end/88758).
Martin Liska [Fri, 11 Jan 2019 13:20:01 +0000 (14:20 +0100)]
Add a testcase (PR middle-end/88758).

2019-01-11  Martin Liska  <mliska@suse.cz>

PR middle-end/88758
* g++.dg/lto/pr88758_0.C: New test.
* g++.dg/lto/pr88758_1.C: New test.

From-SVN: r267840

5 years agore PR bootstrap/88714 (bootstrap comparison failure on armv7l since r265398)
Jakub Jelinek [Fri, 11 Jan 2019 12:05:54 +0000 (13:05 +0100)]
re PR bootstrap/88714 (bootstrap comparison failure on armv7l since r265398)

PR bootstrap/88714
* passes.c (finish_optimization_passes): Call print_combine_total_stats
inside of pass_combine_1 dump rather than pass_profile_1.

From-SVN: r267839

5 years ago[nvptx] Don't allow vector_length 64 with num_workers 16
Tom de Vries [Fri, 11 Jan 2019 11:46:43 +0000 (11:46 +0000)]
[nvptx] Don't allow vector_length 64 with num_workers 16

When using a compiler build with:
...
+#define PTX_DEFAULT_VECTOR_LENGTH PTX_CTA_SIZE
...

consider a test-case:
...
int
main (void)
{
  #pragma acc parallel vector_length (64)
    #pragma acc loop worker
    for (unsigned int i = 0; i < 32; i++)
      #pragma acc loop vector
      for (unsigned int j = 0; j < 64; j++)
        ;

  return 0;
}
...

If num_workers is 16, either because:
- we add a "num_workers (16)" clause on the parallel directive, or
- we set "GOMP_OPENACC_DIM=:16:", or
- the libgomp plugin chooses 16 num_workers
we run into an illegal instruction at runtime, because a bar.sync instruction
tries to use a barrier 16.  The instruction is illegal, because ptx supports
only 16 barriers per CTA, and the valid range is 0..15.

The problem is that with a warp-multiple vector length, we use a code generation
scheme with a per-worker barrier.  And because barrier zero is reserved for
per-cta barrier, only the remaining 15 barriers can be used as per-worker
barrier, and consequently we can't use num_workers larger than 15.

This problem occurs only for vector_length 64.  For vector_length 32, we use a
different code generation scheme, and for vector_length >= 96, the maximum
num_workers is not big enough not to trigger this problem.

Also, this problem only occurs for num_workers 16.  As explained above,
num_workers 15 is safe to use, and 16 is already the maximum num_workers for
vector_length 64.

This patch fixes the problem in both the compiler (handling "num_workers (16)")
and in the libgomp nvptx plugin (with and without "GOMP_OPENACC_DIM=:16:").

2019-01-11  Tom de Vries  <tdevries@suse.de>

* config/nvptx/nvptx.c (PTX_CTA_NUM_BARRIERS, PTX_PER_CTA_BARRIER)
(PTX_NUM_PER_CTA_BARRIER, PTX_FIRST_PER_WORKER_BARRIER)
(PTX_NUM_PER_WORKER_BARRIERS): Define.
(nvptx_apply_dim_limits): Prevent vector_length 64 and
num_workers 16.

* plugin/plugin-nvptx.c (nvptx_exec): Prevent vector_length 64 and
num_workers 16.

From-SVN: r267838

5 years ago[nvptx] Move PTX_CTA_SIZE up
Tom de Vries [Fri, 11 Jan 2019 11:46:31 +0000 (11:46 +0000)]
[nvptx] Move PTX_CTA_SIZE up

Move the defition of PTX_CTA_SIZE up in nvptx.c.

2019-01-11  Tom de Vries  <tdevries@suse.de>

* config/nvptx/nvptx.c (PTX_CTA_SIZE): Move up.

From-SVN: r267837

5 years ago[libgomp, testsuite, openacc] Remove -foffload=-w in reduction-[1-5].c
Tom de Vries [Fri, 11 Jan 2019 11:46:06 +0000 (11:46 +0000)]
[libgomp, testsuite, openacc] Remove -foffload=-w in reduction-[1-5].c

Before the commit "[libgomp, testsuite, openacc] Don't use const int for
dimensions", the "const int" construct was used to set launch dimensions in
reductions-[1-5].c.  In the case of -xc -O0, the const int is implemented as a
variable by the C front-end.  Consequently, the nvptx back-end generated
warnings that vector_length was overridden to be hard-coded, rather than left to
be set at runtime.  The test-cases silenced these warnings by switching off all
warnings in the accelerator compiler using "-foffload=-w".

Given that no warnings occur anymore, remove the "-foffload=-w" setting.

2019-01-11  Tom de Vries  <tdevries@suse.de>

* testsuite/libgomp.oacc-c-c++-common/reduction-1.c: Remove
-foffload=-w.
* testsuite/libgomp.oacc-c-c++-common/reduction-2.c: Same.
* testsuite/libgomp.oacc-c-c++-common/reduction-3.c: Same.
* testsuite/libgomp.oacc-c-c++-common/reduction-4.c: Same.
* testsuite/libgomp.oacc-c-c++-common/reduction-5.c: Same.

From-SVN: r267836

5 years ago[nvptx, testsuite, openacc, libgomp] Add insufficient-resources.c
Tom de Vries [Fri, 11 Jan 2019 11:45:55 +0000 (11:45 +0000)]
[nvptx, testsuite, openacc, libgomp] Add insufficient-resources.c

Add a test-case that tests the "insufficient resources" fatal in the nvptx
libgomp plugin.

2019-01-11  Tom de Vries  <tdevries@suse.de>

* testsuite/libgomp.oacc-c-c++-common/insufficient-resources.c: New
test.

From-SVN: r267835