Fixed extract_types_for_resx function

The function was recursive in nature and there is a chance of runnign out of stack, so now ann iterative approach was used to get the types for resx

Formatting for all functions done according to GNU standards and fixed testsuite fail bugs

Added function bool extract_types_for_resx and fixed more functions for testcases

1. bool extract_types_for_resx (gimple *,vec<tree> *) - This is for intiating the function static bool extract_types_for_resx (basic_block, vec<tree> *)
which gets the resx stmt from a basic block and aclears the aux value for all blocks
2. update_stmt_eh_region - This is changed to unlink and reinsert the regions incase of a resx stmt properly

Updated code for unlinking regions and update stmt_eh_regions as well

Updated code for unlinking regions

Updated code for inserting regions and changed the return parameters of extract_types_for call

Added code for inserting regions

Updated tree-eh.h

resolved more conflicts

removed conflicts

Reolved some conflicts

Added functions extract_fun_resx_types

Merge remote-tracking branch 'origin/devel/nothrow-detection' into devel/nothrow-detection

Updated parameters of functions and typecasted resx stmts

Integrated the resx functions with stmt_throw_types

update_stmt_eh_region updated for resx stmt support and extract_fun_resx_types fixed

Added the previous functions to the tree-eh.h file

Functions added:
1. void extract_types_for_resx (gimple *, vec<tree> *);
2. void extract_fun_resx_types (function *);

RESX statement processing functions added:

1. extract_types_for_resx - Used as a helper for getting the types from a resx statement to use later in the extract_fun_resx_types
2. extract_fun_resx_types - Used for getting the types thrown and propagated outside a function

Fixed some indentations and function names

1. Changed function name extract_exception_types_for_call to extract_types_for_call due to length of the name
2. Fixed space indentation in some places

Added some functions and fixed some testcase failures

1. odr_equivalent_or_derived_p in ipa-devirt.cc
2. same_or_derived_type in in tree-eh.cc
3. Fixed catch-all handling in match_lp function

Edge redirection for exceptions.

 This commit is contains change in code for the tree-eh.cc, tree-eh.h, MAINTAINERS and tree-cfg.cc files.
 Specifically it contains four functions -
1. void extract_exception_types_for_call which extracts the exception types in a call stmt and adds them into a vector tree.
2. bool stmt_throw_types does the same as stmt_could_throw the difference being that it also gives the list of exception types as given by the
 extract_exception_types_for_call function.
3. bool match_lp checks if a landing pad can handle any of the exception types given as input parameters by looking into the catch handlers.
4. update_stmt_eh_region is the function that walks up the EH tree and changes the landing pad for the last statement in a basic block in the control
 flow graph so that when the edge by make_eh_edge is created it points to the correct handlers.

Further work to be done regarding RESX stmts.

Daily bump.

libstdc++-v3: Fix signed-overflow warning for newlib/ctype_base.h, PR116895

There are 100+ regressions when running the g++ testsuite for newlib
targets (probably excepting ARM-based ones) e.g cris-elf after commit
r15-3859-g63a598deb0c9fc "libstdc++: #ifdef out #pragma GCC
system_header", which effectively no longer silences warnings for
gcc-installed system headers.  Some of these regressions are fixed by
r15-3928.  For the remaining ones, there's in g++.log:

FAIL: g++.old-deja/g++.robertl/eb79.C  -std=c++26 (test for excess errors)
Excess errors:
/gccobj/cris-elf/libstdc++-v3/include/cris-elf/bits/ctype_base.h:50:53: \
 warning: overflow in conversion from 'int' to 'std::ctype_base::mask' \
 {aka 'char'} changes value from '151' to '-105' [-Woverflow]

This is because the _B macro in newlib's ctype.h (from where the
"_<letter>" macros come) is bit 7, the sign-bit of 8-bit types:

#define _B 0200

Using it in an int-expression that is then truncated to 8 bits will
"change" the value to negative for a default-signed char.  If this
code was created from scratch, it should have been an unsigned type,
however it's not advisable to change the type of mask as this affects
the API.  The least ugly option seems to be to silence the warning by
explict casts in the initializer, and for consistency, doing it for
all members.

PR libstdc++/116895
* config/os/newlib/ctype_base.h: Avoid signed-overflow warnings by
explicitly casting initializer expressions to mask.

[testcase] Fix-absfloat16.c-testcase

gcc/testsuite/ChangeLog:
* gcc.dg/tree-ssa/absfloat16.c: Fix testcase.

Signed-off-by: Kugan Vivekanandarajah <kvivekananda@nvidia.com>

c++: concept in default argument [PR109859]

1) We're hitting the assert in cp_parser_placeholder_type_specifier.
It says that if it turns out to be false, we should do error() instead.
Do so, then.

2) lambda-targ8.C should compile fine, though.  The problem was that
local_variables_forbidden_p wasn't cleared when we're about to parse
the optional template-parameter-list for a lambda in a default argument.

PR c++/109859

gcc/cp/ChangeLog:

* parser.cc (cp_parser_lambda_declarator_opt): Temporarily clear
local_variables_forbidden_p.
(cp_parser_placeholder_type_specifier): Turn an assert into an
error.

gcc/testsuite/ChangeLog:

* g++.dg/cpp2a/concepts-defarg3.C: New test.
* g++.dg/cpp2a/lambda-targ8.C: New test.

Reviewed-by: Jason Merrill <jason@redhat.com>

Fix internal error during inlining after ICF pass

The problem is that the ICF pass identifies two functions, one of which has
a static chain while the other does not.  The fix is simply to prevent this
identification from occurring.

gcc/
PR ipa/113996
* ipa-icf.cc (sem_function::get_hash): Hash DECL_STATIC_CHAIN.
(sem_function::equals_wpa): Compare it.
(sem_function::equals_private): Likewise.

gcc/testsuite/
* gnat.dg/lto27.adb: New test.

diagnostics: return text buffer from test_show_locus [PR116613]

As work towards supporting multiple diagnostic outputs (where each
output has its own pretty_printer), avoid referencing dc.m_printer
throughout the selftests of diagnostic-show-locus.cc.  Instead
have test_diagnostic_context::test_show_locus return the result
buffer, hiding the specifics of which printer is in use in such
test cases.

No functional change intended.

gcc/ChangeLog:
PR other/116613
* diagnostic-show-locus.cc
(selftest::test_diagnostic_show_locus_unknown_location): Move call
to dc.test_show_locus into ASSERT_STREQ, and compare against its
result, rather than explicitly using dc.m_printer.
(selftest::test_one_liner_simple_caret): Likewise.
(selftest::test_one_liner_no_column): Likewise.
(selftest::test_one_liner_caret_and_range): Likewise.
(selftest::test_one_liner_multiple_carets_and_ranges): Likewise.
(selftest::test_one_liner_fixit_insert_before): Likewise.
(selftest::test_one_liner_fixit_insert_after): Likewise.
(selftest::test_one_liner_fixit_remove): Likewise.
(selftest::test_one_liner_fixit_replace): Likewise.
(selftest::test_one_liner_fixit_replace_non_equal_range):
Likewise.
(selftest::test_one_liner_fixit_replace_equal_secondary_range):
Likewise.
(selftest::test_one_liner_fixit_validation_adhoc_locations):
Likewise.
(selftest::test_one_liner_many_fixits_1): Likewise.
(selftest::test_one_liner_many_fixits_2): Likewise.
(selftest::test_one_liner_labels): Likewise.
(selftest::test_one_liner_simple_caret_utf8): Likewise.
(selftest::test_one_liner_multiple_carets_and_ranges_utf8):
Likewise.
(selftest::test_one_liner_fixit_insert_before_utf8): Likewise.
(selftest::test_one_liner_fixit_insert_after_utf8): Likewise.
(selftest::test_one_liner_fixit_remove_utf8): Likewise.
(selftest::test_one_liner_fixit_replace_utf8): Likewise.
(selftest::test_one_liner_fixit_replace_non_equal_range_utf8):
Likewise.
(selftest::test_one_liner_fixit_replace_equal_secondary_range_utf8):
Likewise.
(selftest::test_one_liner_fixit_validation_adhoc_locations_utf8):
Likewise.
(selftest::test_one_liner_many_fixits_1_utf8): Likewise.
(selftest::test_one_liner_many_fixits_2_utf8): Likewise.
(selftest::test_one_liner_labels_utf8): Likewise.
(selftest::test_one_liner_colorized_utf8): Likewise.
(selftest::test_add_location_if_nearby): Likewise.
(selftest::test_diagnostic_show_locus_fixit_lines): Likewise.
(selftest::test_overlapped_fixit_printing): Likewise.
(selftest::test_overlapped_fixit_printing_utf8): Likewise.
(selftest::test_overlapped_fixit_printing_utf8): Likewise.
(selftest::test_overlapped_fixit_printing_2): Likewise.
(selftest::test_fixit_insert_containing_newline): Likewise.
(selftest::test_fixit_insert_containing_newline_2): Likewise.
(selftest::test_fixit_replace_containing_newline): Likewise.
(selftest::test_fixit_deletion_affecting_newline): Likewise.
(selftest::test_tab_expansion): Likewise.
(selftest::test_escaping_bytes_1): Likewise.
(selftest::test_escaping_bytes_2): Likewise.
(selftest::test_line_numbers_multiline_range): Likewise.
* selftest-diagnostic.cc
(selftest::test_diagnostic_context::test_show_locus): Return the
formatted text of m_printer.
* selftest-diagnostic.h
(selftest::test_diagnostic_context::test_show_locus): Convert
return type from void to const char *.

Signed-off-by: David Malcolm <dmalcolm@redhat.com>

diagnostics: require callers of diagnostic_show_locus to be explicit about the printer [PR116613]

As work towards supporting multiple diagnostic outputs (where each
output has its own pretty_printer), update diagnostic_show_locus
so that the pretty_printer must always be explicitly passed in.

No functional change intended.

gcc/c-family/ChangeLog:
PR other/116613
* c-format.cc (selftest::test_type_mismatch_range_labels):
Explicitly pass in dc.m_printer to diagnostic_show_locus.

gcc/ChangeLog:
PR other/116613
* diagnostic-show-locus.cc (diagnostic_context::maybe_show_locus):
Convert param "pp" from * to &.  Drop logic for using the
context's m_printer when the param is null.
* diagnostic.h (diagnostic_context::maybe_show_locus): Convert
param "pp" from * to &.
(diagnostic_show_locus): Drop default "nullptr" value for pp
param.  Assert that it and context are nonnull.  Pass pp by
reference to maybe_show_locus.

gcc/testsuite/ChangeLog:
PR other/116613
* gcc.dg/plugin/expensive_selftests_plugin.c (test_richloc):
Explicitly pass in dc.m_printer to diagnostic_show_locus.

Signed-off-by: David Malcolm <dmalcolm@redhat.com>

diagnostics: isolate diagnostic_context with interface classes [PR116613]

As work towards supporting multiple diagnostic outputs (where each
output has its own pretty_printer), avoid passing around
diagnostic_context to the various printing routines, so that we
can be more explicit about which pretty_printer is in use.

Introduce a set of "policy" classes that capture the parts of
diagnostic_context that are needed, and use these rather than
diagnostic_context *.  Pass around pretty_printer & rather than
taking value from context.  Split out the pretty_printer-using code
from class layout into a new class layout_printer, separating the
responsibilities of determining layout when quoting source versus
actually printing the source.

No functional change intended.

gcc/analyzer/ChangeLog:
PR other/116613
* program-point.cc (function_point::print_source_line): Replace
call to diagnostic_show_locus with a call to
diagnostic_source_print_policy::print.

gcc/ChangeLog:
PR other/116613
* diagnostic-format-json.cc (json_from_expanded_location): Replace
call to diagnostic_context::converted_column with call to
diagnostic_column_policy::converted_column.
* diagnostic-format-sarif.cc
(sarif_builder::make_location_object): Replace call to
diagnostic_show_locus with call to
diagnostic_source_print_policy::print.
* diagnostic-format-text.cc (get_location_text): Replace call to
diagnostic_context::get_location_text with call to
diagnostic_column_policy::get_location_text.
(diagnostic_text_output_format::report_current_module): Replace call
to diagnostic_context::converted_column with call to
diagnostic_column_policy::converted_column.
* diagnostic-format-text.h
(diagnostic_text_output_format::diagnostic_output_format):
Initialize m_column_policy.
(diagnostic_text_output_format::get_column_policy): New.
(diagnostic_text_output_format::m_column_policy): New.
* diagnostic-path.cc (class path_print_policy): New.
(event_range::maybe_add_event): Replace diagnostic_context param
with path_print_policy.
(event_range::print): Convert "pp" from * to &.  Convert first
param of start_span callback from diagnostic_context to
diagnostic_location_print_policy.
(path_summary::path_summary): Convert first param from
diagnostic_text_output_format to path_print_policy.  Add
colorize param.  Update for changes to
event_range::maybe_add_event.
(thread_event_printer::print_swimlane_for_event_range): Assert
that pp is non-null.  Update for change to event_range::print.
(diagnostic_text_output_format::print_path): Pass
path_print_policy to path_summary's ctor.
(selftest::test_empty_path): Likewise.
(selftest::test_intraprocedural_path): Likewise.
(selftest::test_interprocedural_path_1): Likewise.
(selftest::test_interprocedural_path_2): Likewise.
(selftest::test_recursion): Likewise.
(selftest::test_control_flow_1): Likewise.
(selftest::test_control_flow_2): Likewise.
(selftest::test_control_flow_3): Likewise.
(selftest::assert_cfg_edge_path_streq): Likewise.
(selftest::test_control_flow_5): Likewise.
(selftest::test_control_flow_6): Likewise.
* diagnostic-show-locus.cc (colorizer::set_range): Update for
change to m_pp.
(colorizer::m_pp): Convert from * to &.
(class layout): Add friend class layout_printer and move various
decls to it.
(layout::m_pp): Drop field.
(layout::m_policy): Rename to...
(layout::m_char_policy): ...this.
(layout::m_colorizer): Move field to class layout_printer.
(layout::m_diagnostic_path_p): Drop field.
(class layout_printer): New class, by refactoring class layout.
(colorizer::colorizer): Convert "pp" param from * to &.
(colorizer::set_named_color): Update for above change.
(colorizer::begin_state): Likewise.
(colorizer::finish_state): Likewise.
(make_policy): Rename to...
(make_char_policy): ...this, and update param from
diagnostic_context to diagnostic_source_print_policy.
(layout::layout): Update param from diagnostic_context to
diagnostic_source_print_policy.  Drop params "diagnostic_kind" and
"pp", moving these and other material to class layout_printer.
(layout::maybe_add_location_range): Update for renamed field.
(layout::print_gap_in_line_numbering): Convert to...
(layout_printer::print_gap_in_line_numbering): ...this.
(layout::calculate_x_offset_display): Update for renamed field.
(layout::print_source_line): Convert to...
(layout_printer::print_source_line): ...this.
(layout::print_leftmost_column): Convert to...
(layout_printer::print_leftmost_column): ...this.
(layout::start_annotation_line): Convert to...
(layout_printer::start_annotation_line): ...this.
(layout::print_annotation_line): Convert to...
(layout_printer::print_annotation_line): ...this.
(layout::print_any_labels): Convert to...
(layout_printer::print_any_labels): ...this.
(layout::print_leading_fixits): Convert to...
(layout_printer::print_leading_fixits): ...this.
(layout::print_trailing_fixits): Convert to...
(layout_printer::print_trailing_fixits): ...this.
(layout::print_newline): Convert to...
(layout_printer::print_newline): ...this.
(layout::get_state_at_point): Make const.
(layout::get_x_bound_for_row): Make const.
(layout::move_to_column): Convert to...
(layout_printer::move_to_column): ...this.
(layout::show_ruler): Convert to...
(layout_printer::show_ruler): ...this.
(layout::print_line): Convert to...
(layout_printer::print_line): ...this.
(layout::print_any_right_to_left_edge_lines): Convert to...
(layout_printer::print_any_right_to_left_edge_lines): ...this.
(layout::print_any_right_to_left_edge_lines): Likewise.
(layout_printer::layout_printer): New.
(layout::update_any_effects): Delete, moving logic to
layout_printer::print.
(gcc_rich_location::add_location_if_nearby): Update param from
diagnostic_context to diagnostic_source_print_policy.  Add
overload taking a diagnostic_context.
(diagnostic_context::maybe_show_locus): Move handling of null
pretty_printer here, from layout ctor.  Convert call to
diagnostic_context::show_locus to
diagnostic_source_print_policy::print.
(diagnostic_source_print_policy::diagnostic_source_print_policy):
New.
(diagnostic_context::show_locus): Convert to...
(diagnostic_source_print_policy::print): ...this.  Convert pp
from * to &.
(layout_printer::print): New, based on material in
diagnostic_context::show_locus.
(selftest::make_char_policy): New.
(selftest::test_display_widths): Update for above changes.
(selftest::test_offset_impl): Likewise.
(selftest::test_layout_x_offset_display_utf8): Likewise.
(selftest::test_layout_x_offset_display_tab): Likewise.
(selftest::test_diagnostic_show_locus_unknown_location): Use
test_diagnostic_context::test_show_locus rather than
diagnostic_show_locus.
(selftest::test_one_liner_no_column): Likewise.
(selftest::test_one_liner_caret_and_range): Likewise.
(selftest::test_one_liner_multiple_carets_and_ranges): Likewise.
(selftest::test_one_liner_fixit_insert_before): Likewise.
(selftest::test_one_liner_fixit_insert_after): Likewise.
(selftest::test_one_liner_fixit_remove): Likewise.
(selftest::test_one_liner_fixit_replace): Likewise.
(selftest::test_one_liner_fixit_replace_non_equal_range):
Likewise.
(selftest::test_one_liner_fixit_replace_equal_secondary_range):
Likewise.
(selftest::test_one_liner_fixit_validation_adhoc_locations):
Likewise.
(selftest::test_one_liner_many_fixits_1): Likewise.
(selftest::test_one_liner_many_fixits_2): Likewise.
(selftest::test_one_liner_labels): Likewise.
(selftest::test_one_liner_simple_caret_utf8): Likewise.
(selftest::test_one_liner_caret_and_range_utf8): Likewise.
(selftest::test_one_liner_multiple_carets_and_ranges_utf8):
Likewise.
(selftest::test_one_liner_fixit_insert_before_utf8): Likewise.
(selftest::test_one_liner_fixit_insert_after_utf8): Likewise.
(selftest::test_one_liner_fixit_remove_utf8): Likewise.
(selftest::test_one_liner_fixit_replace_utf8): Likewise.
(selftest::test_one_liner_fixit_replace_non_equal_range_utf8):
Likewise.
(selftest::test_one_liner_fixit_replace_equal_secondary_range_utf8):
Likewise.
(selftest::test_one_liner_fixit_validation_adhoc_locations_utf8):
Likewise.
(selftest::test_one_liner_many_fixits_1_utf8): Likewise.
(selftest::test_one_liner_many_fixits_2_utf8): Likewise.
(selftest::test_one_liner_labels_utf8): Likewise.
(selftest::test_one_liner_colorized_utf8): Likewise.
(selftest::test_add_location_if_nearby): Likewise.
(selftest::test_diagnostic_show_locus_fixit_lines): Likewise.
(selftest::test_overlapped_fixit_printing): Likewise.
(selftest::test_overlapped_fixit_printing_utf8): Likewise.
(selftest::test_overlapped_fixit_printing_2): Likewise.
(selftest::test_fixit_insert_containing_newline): Likewise.
(selftest::test_fixit_insert_containing_newline_2): Likewise.
(selftest::test_fixit_replace_containing_newline): Likewise.
(selftest::test_fixit_deletion_affecting_newline): Likewise.
(selftest::test_tab_expansion): Likewise.
(selftest::test_escaping_bytes_1): Likewise.
(selftest::test_escaping_bytes_2): Likewise.
(selftest::test_line_numbers_multiline_range): Likewise.
* diagnostic.cc
(diagnostic_column_policy::diagnostic_column_policy): New.
(diagnostic_context::converted_column): Convert to...
(diagnostic_column_policy::converted_column): ...this.
(diagnostic_context::get_location_text): Convert to...
(diagnostic_column_policy::get_location_text): ...this, adding
"show_column" param.
(diagnostic_location_print_policy::diagnostic_location_print_policy):
New ctors.
(default_diagnostic_start_span_fn): Convert param from
diagnostic_context * to const diagnostic_location_print_policy &.
Add "pp" param.
(selftest::assert_location_text): Update for above changes.
(selftest::test_diagnostic_get_location_text): Rename to...
(selftest::test_get_location_text): ...this.
(selftest::c_diagnostic_cc_tests): Update for renaming.
* diagnostic.h (class diagnostic_location_print_policy): New
forward decl.
(class diagnostic_source_print_policy): New forward decl.
(diagnostic_start_span_fn): Convert first param from
diagnostic_context * to const diagnostic_location_print_policy &
and add pretty_printer * param.
(class diagnostic_column_policy): New.
(class diagnostic_location_print_policy): New.
(class diagnostic_source_print_policy): New.
(class diagnostic_context): Add friend class
diagnostic_source_print_policy.
(diagnostic_context::converted_column): Drop decl in favor of
diagnostic_column_policy::converted_column.
(diagnostic_context::get_location_text): Drop decl in favor of
diagnostic_column_policy::get_location_text.
(diagnostic_context::show_locus): Drop decl in favor of
diagnostic_source_print_policy::print.
(default_diagnostic_start_span_fn): Update for change to
diagnostic_start_span_fn.
* gcc-rich-location.h (class diagnostic_source_print_policy): New
forward decl.
(gcc_rich_location::add_location_if_nearby): Convert first param
from diagnostic_context to diagnostic_source_print_policy.  Add
overload taking diagnostic_context.
* selftest-diagnostic.cc
(selftest::test_diagnostic_context::test_diagnostic_context): Turn
off colorization.
(selftest::test_diagnostic_context::start_span_cb): Update for
change to callback type.
(test_diagnostic_context::test_show_locus): New.
* selftest-diagnostic.h
(selftest::test_diagnostic_context::start_span_cb): Update for
change to callback type.
(test_diagnostic_context::test_show_locus): New decl.

gcc/fortran/ChangeLog:
PR other/116613
* error.cc (gfc_diagnostic_build_locus_prefix): Convert first
param from diagnostic_context * to
const diagnostic_location_print_policy &.  Add colorize param.
Likewise for the "two expanded_locations" overload.
(gfc_diagnostic_text_starter): Update for above changes.
(gfc_diagnostic_start_span): Update for change to callback type.

gcc/testsuite/ChangeLog:
PR other/116613
* gcc.dg/plugin/diagnostic_group_plugin.c
(test_diagnostic_start_span_fn): Update for change to callback
type.

Signed-off-by: David Malcolm <dmalcolm@redhat.com>

diagnostics: avoid using diagnostic_context's m_printer [PR116613]

As work towards supporting multiple diagnostic outputs (where each
output has its own pretty_printer), avoid using diagnostic_context's
m_printer field.  Instead, use the output format's printer.  Currently
this *is* the dc's printer, but eventually it might not be.

No functional change intended.

gcc/ChangeLog:
PR other/116613
* diagnostic-format-json.cc (diagnostic_output_format_init_json):
Pass in the format.  Use the format's printer when disabling
colorization.  Move the call to set_output_format into here.
(diagnostic_output_format_init_json_stderr): Update for above
change.
(diagnostic_output_format_init_json_file): Likewise.
* diagnostic-format-sarif.cc
(diagnostic_output_format_init_sarif): Use the format's printer
when disabling colorization.
* diagnostic-path.cc (selftest::test_empty_path): Use the
text_output's printer.
(selftest::test_intraprocedural_path): Likewise.
(selftest::test_interprocedural_path_1): Likewise.
(selftest::test_interprocedural_path_2): Likewise.
(selftest::test_recursion): Likewise.
(selftest::test_control_flow_1): Likewise.
(selftest::test_control_flow_2): Likewise.
(selftest::test_control_flow_3): Likewise.
(selftest::assert_cfg_edge_path_streq): Likewise.
(selftest::test_control_flow_5): Likewise.
(selftest::test_control_flow_6): Likewise.

gcc/testsuite/ChangeLog:
PR other/116613
* gcc.dg/plugin/diagnostic_group_plugin.c
(test_output_format::on_begin_group): Use get_printer () rather
than accessing m_context.m_printer.
(test_output_format::on_end_group): Likewise.
* gcc.dg/plugin/diagnostic_plugin_xhtml_format.c
(xhtml_builder::m_printer): New field.
(xhtml_builder::xhtml_builder): Add "pp" param and use it to
initialize m_printer.
(xhtml_builder::on_report_diagnostic): Drop "context" param.
(xhtml_builder::make_element_for_diagnostic): Likewise.  Use
this->m_printer rather than the context's m_printer.  Pass
m_printer to call to diagnostic_show_locus.
(xhtml_builder::emit_diagram): Drop "context" param.
(xhtml_output_format::on_report_diagnostic): Drop context param
from call to m_builder.
(xhtml_output_format::on_diagram): Likewise.
(xhtml_output_format::xhtml_output_format): Pass result of
get_printer as printer for builder.
(diagnostic_output_format_init_xhtml): Use the fmt's printer
rather than the context's.

Signed-off-by: David Malcolm <dmalcolm@redhat.com>

diagnostics: use "%e" to avoid intermediate strings [PR116613]

Various diagnostics build an intermediate string, potentially with
colorization, and then use this in a diagnostic message.

This won't work if we have multiple diagnostic sinks, where some might
be colorized and some not.

This patch reworks such places using "%e" and pp_element subclasses, so
that any colorization happens within report_diagnostic's call to
pp_format.

gcc/analyzer/ChangeLog:
PR other/116613
* kf-analyzer.cc: Include "pretty-print-markup.h".
(kf_analyzer_dump_escaped::impl_call_pre): Defer colorization
choices by eliminating the construction of a intermediate string,
replacing it with a new pp_element subclass via "%e".

gcc/ChangeLog:
PR other/116613
* attribs.cc: Include "pretty-print-markup.h".
(decls_mismatched_attributes): Defer colorization choices by
replacing printing to a pretty_printer * param with appending
to a vec of strings.
(maybe_diag_alias_attributes): As above, replacing pretty_printer
with usage of pp_markup::comma_separated_quoted_strings and "%e"
in two places.
* attribs.h (decls_mismatched_attributes): Update decl.
* gimple-ssa-warn-access.cc: Include "pretty-print-markup.h".
(pass_waccess::maybe_warn_memmodel): Defer colorization choices by
replacing printing to a pretty_printer * param with use of
pp_markup::comma_separated_quoted_strings and "%e".
(pass_waccess::maybe_warn_memmodel): Likewise, replacing printing
to a temporary buffer.
* pretty-print-markup.h
(class pp_markup::comma_separated_quoted_strings): New.
* pretty-print.cc
(pp_markup::comma_separated_quoted_strings::add_to_phase_2): New.
(selftest::test_pp_printf_within_pp_element): New.
(selftest::test_comma_separated_quoted_strings): New.
(selftest::pretty_print_cc_tests): Call the new tests.

gcc/cp/ChangeLog:
PR other/116613
* pt.cc: Include "pretty-print-markup.h".
(warn_spec_missing_attributes): Defer colorization choices by
replacing printing to a pretty_printer * param with appending
to a vec of strings.  Replace pretty_printer with usage of
pp_markup::comma_separated_quoted_strings and "%e".

gcc/testsuite/ChangeLog:
PR other/116613
* c-c++-common/analyzer/escaping-1.c: Update expected results to
remove type information from C++ results.  Previously we were
using %qD with default_tree_printer, which used
lang_hooks.decl_printable_name, whereas now we're using %qD with
a clone of the cxx_pretty_printer.

Signed-off-by: David Malcolm <dmalcolm@redhat.com>

diagnostics: add "dump" to pretty_printer and output_buffer

No functional change intended.

gcc/ChangeLog:
* pretty-print.cc (output_buffer::dump): New.
(pretty_printer::dump): New.
* pretty-print.h (output_buffer::dump): New decls.
(pretty_printer::dump): New decls.

Signed-off-by: David Malcolm <dmalcolm@redhat.com>

diagnostics: fix typo in XHTML output [PR116792]

gcc/testsuite/ChangeLog:
PR other/116792
* gcc.dg/plugin/diagnostic_plugin_xhtml_format.c: Fix stray
reference to JSON.

Signed-off-by: David Malcolm <dmalcolm@redhat.com>

diagnostics: fix memory leak in SARIF selftests

"make selftest-valgrind" was complaining about leaks of artifact objects
in SARIF's selftest::test_make_location_object:

-fself-test: 7638695 pass(es) in 89.999249 seconds
==3306525==
==3306525== HEAP SUMMARY:
==3306525==     in use at exit: 1,215,639 bytes in 2,808 blocks
==3306525==   total heap usage: 2,860,898 allocs, 2,858,090 frees, 1,336,446,579 bytes allocated
==3306525==
==3306525== 11,728 (1,536 direct, 10,192 indirect) bytes in 16 blocks are definitely lost in loss record 353 of 375
==3306525==    at 0x514FE7D: operator new(unsigned long) (vg_replace_malloc.c:342)
==3306525==    by 0x36E5FD2: sarif_builder::get_or_create_artifact(char const*, diagnostic_artifact_role, bool) (diagnostic-format-sarif.cc:2884)
==3306525==    by 0x36E3D57: sarif_builder::maybe_make_physical_location_object(unsigned int, diagnostic_artifact_role, int, content_renderer const*) (diagnostic-format-sarif.cc:2097)
==3306525==    by 0x36E34CE: sarif_builder::make_location_object(sarif_location_manager&, rich_location const&, logical_location const*, diagnostic_artifact_role) (diagnostic-format-sarif.cc:1922)
==3306525==    by 0x36E72C6: selftest::test_make_location_object(selftest::line_table_case const&) (diagnostic-format-sarif.cc:3500)
==3306525==    by 0x375609B: selftest::for_each_line_table_case(void (*)(selftest::line_table_case const&)) (input.cc:3898)
==3306525==    by 0x36E9668: selftest::diagnostic_format_sarif_cc_tests() (diagnostic-format-sarif.cc:3910)
==3306525==    by 0x3592A11: selftest::run_tests() (selftest-run-tests.cc:100)
==3306525==    by 0x17DBEF3: toplev::run_self_tests() (toplev.cc:2268)
==3306525==    by 0x17DC2BF: toplev::main(int, char**) (toplev.cc:2376)
==3306525==    by 0x36A1919: main (main.cc:39)
==3306525==
==3306525== 12,400 (1,536 direct, 10,864 indirect) bytes in 16 blocks are definitely lost in loss record 355 of 375
==3306525==    at 0x514FE7D: operator new(unsigned long) (vg_replace_malloc.c:342)
==3306525==    by 0x36E5FD2: sarif_builder::get_or_create_artifact(char const*, diagnostic_artifact_role, bool) (diagnostic-format-sarif.cc:2884)
==3306525==    by 0x36E2323: sarif_builder::sarif_builder(diagnostic_context&, line_maps const*, char const*, bool) (diagnostic-format-sarif.cc:1500)
==3306525==    by 0x36E70AA: selftest::test_make_location_object(selftest::line_table_case const&) (diagnostic-format-sarif.cc:3469)
==3306525==    by 0x375609B: selftest::for_each_line_table_case(void (*)(selftest::line_table_case const&)) (input.cc:3898)
==3306525==    by 0x36E9668: selftest::diagnostic_format_sarif_cc_tests() (diagnostic-format-sarif.cc:3910)
==3306525==    by 0x3592A11: selftest::run_tests() (selftest-run-tests.cc:100)
==3306525==    by 0x17DBEF3: toplev::run_self_tests() (toplev.cc:2268)
==3306525==    by 0x17DC2BF: toplev::main(int, char**) (toplev.cc:2376)
==3306525==    by 0x36A1919: main (main.cc:39)
==3306525==
==3306525== LEAK SUMMARY:
==3306525==    definitely lost: 3,072 bytes in 32 blocks
==3306525==    indirectly lost: 21,056 bytes in 368 blocks
==3306525==      possibly lost: 0 bytes in 0 blocks
==3306525==    still reachable: 1,191,511 bytes in 2,408 blocks
==3306525==         suppressed: 0 bytes in 0 blocks
==3306525== Reachable blocks (those to which a pointer was found) are not shown.
==3306525== To see them, rerun with: --leak-check=full --show-leak-kinds=all
==3306525==
==3306525== For lists of detected and suppressed errors, rerun with: -s
==3306525== ERROR SUMMARY: 2 errors from 2 contexts (suppressed: 0 from 0)

Fixed thusly.

gcc/ChangeLog:
* diagnostic-format-sarif.cc (sarif_builder::~sarif_builder): New,
deleting any remaining artifact objects.
(sarif_builder::make_run_object): Empty the artifact map.
* ordered-hash-map.h (ordered_hash_map::empty): New.

Signed-off-by: David Malcolm <dmalcolm@redhat.com>

autovectorizer: Test autovectorization of different dot-prod modes.

Given the novel treatment of the dot product optab as a conversion, we
are now able to target different relationships between output modes and
input modes.

This is made clearer by way of example. Previously, on AArch64, the
following loop was vectorizable:

uint32_t udot4(int n, uint8_t* data) {
  uint32_t sum = 0;
  for (int i=0; i<n; i+=1)
    sum += data[i] * data[i];
  return sum;
}

while the following was not:

uint32_t udot2(int n, uint16_t* data) {
  uint32_t sum = 0;
  for (int i=0; i<n; i+=1)
    sum += data[i] * data[i];
  return sum;
}

Under the new treatment of the dot product optab, they are both now
vectorizable.

This adds the relevant target-agnostic check to ensure this behavior
in the autovectorizer, gated behind the new check_effective_target
`vect_dotprod_hisi' as well a runtime check targeting aarch64.

gcc/testsuite/ChangeLog:

* lib/target-supports.exp (check_effective_target_vect_dotprod_hisi):
New.
* gcc.dg/vect/vect-dotprod-conv-optab.c: Likewise.
* gcc.target/aarch64/vect-dotprod-twoway-hisi.c: Likewise.

c6x:  Adjust dot-product backend patterns

Following the migration of the dot_prod optab from a direct to a
conversion-type optab, ensure all back-end patterns incorporate the
second machine mode into pattern names.

gcc/ChangeLog:

* config/c6x/c6x.md (sdot_prodv2hi): Renamed to...
(sdot_prodsiv2hi): ...this.

rs6000: Adjust altivec dot-product backend patterns

Following the migration of the dot_prod optab from a direct to a
conversion-type optab, ensure all back-end patterns incorporate the
second machine mode into pattern names.

gcc/ChangeLog:

* config/rs6000/altivec.md (udot_prod<mode>): Renamed to...
(udot_prodv4si<mode>): ...this.
(sdot_prodv8hi): Renamed to...
(sdot_prodv4siv8hi): ...this.

mips:  Adjust dot-product backend patterns

Following the migration of the dot_prod optab from a direct to a
conversion-type optab, ensure all back-end patterns incorporate the
second machine mode into pattern names.

gcc/ChangeLog:

* config/mips/loongson-mmi.md (sdot_prodv4hi): Renamed to...
(sdot_prodv2siv4hi): ...this.

arc: Adjust dot-product backend patterns

Following the migration of the dot_prod optab from a direct to a
conversion-type optab, ensure all back-end patterns incorporate the
second machine mode into pattern names.

gcc/ChangeLog:

* config/arc/simdext.md (sdot_prodv2hi): Renamed to...
(sdot_prodsiv2hi): ...this.
(udot_prodv2hi): Renamed to...
(udot_prodsiv2hi): ...this.
(sdot_prodv4hi): Renamed to...
(sdot_prodv2siv4hi): ...this.
(udot_prodv4hi): Renamed to...
(udot_prodv2siv4hi): ...this.

i386: Fix dot_prod backend patterns for mmx and sse targets

Following the migration of the dot_prod optab from a direct to a
conversion-type optab, ensure all back-end patterns incorporate the
second machine mode into pattern names.

gcc/ChangeLog:

* config/i386/mmx.md (usdot_prodv8qi): Renamed to...
(usdot_prodv2siv8qi): ...this.
(sdot_prodv8qi): Renamed to...
(sdot_prodv2siv8qi): ...this.
(udot_prodv8qi): Renamed to...
(udot_prodv2siv8qi): ...this.
(usdot_prodv4hi): Renamed to...
(usdot_prodv2siv4hi): ...this.
(udot_prodv4hi): Renamed to...
(udot_prodv2siv4hi): ...this.
(sdot_prodv4hi): Renamed to...
(sdot_prodv2siv4hi): ...this.
* config/i386/sse.md (sdot_prod<mode>): Renamed to...
(sdot_prod<sseunpackmodelower><mode>): ...this.
(sdot_prodv4si): Renamed to...
(sdot_prodv2div4si): ...this.
(usdot_prod<mode>): Renamed to...
(usdot_prod<ssedvecmodelower><mode>): ...this.
(sdot_prod<mode>): Renamed to...
(sdot_prod<ssedvecmodelower><mode>): ...this.
(sdot_prodv64qi): Renamed to...
(sdot_prodv16siv64qi): ...this.
(udot_prod<mode>): Renamed to...
(udot_prod<ssedvecmodelower><mode>): ...this.
(udot_prodv64qi): Renamed to...
(udot_prodv16qiv64qi): ...this.
(usdot_prod<mode>): Renamed to...
(usdot_prod<sseunpackmodelower><mode>): ...this.
(udot_prod<mode>): Renamed to...
(udot_prod<sseunpackmodelower><mode>): ...this.

arm: Fix arm backend-use of (u|s|us)dot_prod patterns

Given recent changes to the dot_prod standard pattern name, this patch
fixes the arm back-end by implementing the following changes:

1. Add 2nd mode to all patterns relating to the dot-product in .md
files.
2. redirect the single-mode CODE_FOR_neon_(u|s|us)dot<mode> values
generated from `arm_neon_builtins.def' to their new 2-mode
equivalents via means of simple aliases, as per the following example:

  constexpr insn_code CODE_FOR_neon_sdotv8qi
    = CODE_FOR_neon_sdotv2siv8qi;

gcc/ChangeLog:

* config/arm/neon.md (<sup>dot_prod<vsi2qi>): Renamed to...
(<sup>dot_prod<mode><vsi2qi>): ...this.
(neon_<sup>dot<vsi2qi>): Renamed to...
(neon_<sup>dot<mode><vsi2qi>): ...this.
(neon_usdot<vsi2qi>): Renamed to...
(neon_usdot<mode><vsi2qi>): ...this.
(usdot_prod<vsi2qi>): Renamed to...
(usdot_prod<mode><vsi2qi>): ...this.
* config/arm/arm-builtins.cc
(CODE_FOR_neon_sdotv8qi): Definie as alias to
new CODE_FOR_neon_sdotv2siv8qi.
(CODE_FOR_neon_udotv8qi): Definie as alias to
new CODE_FOR_neon_udotv2siv8qi.
(CODE_FOR_neon_usdotv8qi): Definie as alias to
new CODE_FOR_neon_usdotv2siv8qi.
(CODE_FOR_neon_sdotv16qi): Definie as alias to
new CODE_FOR_neon_sdotv4siv16qi.
(CODE_FOR_neon_udotv16qi): Definie as alias to
new CODE_FOR_neon_udotv4siv16qi.
(CODE_FOR_neon_usdotv16qi): Definie as alias to
new CODE_FOR_neon_usdotv4siv16qi.

aarch64: Fix aarch64 backend-use of (u|s|us)dot_prod patterns

Given recent changes to the dot_prod standard pattern name, this patch
fixes the aarch64 back-end by implementing the following changes:

1. Add 2nd mode to all (u|s|us)dot_prod patterns in .md files.
2. Rewrite initialization and function expansion mechanism for simd
builtins.
3. Fix all direct calls to back-end `dot_prod' patterns in SVE
builtins.

Finally, given that it is now possible for the compiler to
differentiate between the two- and four-way dot product, we add a test
to ensure that autovectorization picks up on dot-product patterns
where the result is twice the width of the operands.

gcc/ChangeLog:

* config/aarch64/aarch64-simd.md
(<sur>dot_prod<vsi2qi><vczle><vczbe>): Renamed to...
(<sur>dot_prod<mode><vsi2qi><vczle><vczbe>): ...this.
(usdot_prod<vsi2qi><vczle><vczbe>): Renamed to...
(usdot_prod<mode><vsi2qi><vczle><vczbe>): ...this.
(<su>sadv16qi): Adjust call to gen_udot_prod take second mode.
(popcount<mode2>): fix use of `udot_prod_optab'.
* config/aarch64/aarch64-sve.md
(<sur>dot_prod<vsi2qi>): Renamed to...
(<sur>dot_prod<mode><vsi2qi>): ...this.
(@<sur>dot_prod<vsi2qi>): Renamed to...
(@<sur>dot_prod<mode><vsi2qi>): ...this.
(<su>sad<vsi2qi>): Adjust call to gen_udot_prod take second mode.
* config/aarch64/aarch64-sve2.md
(@aarch64_sve_<sur>dotvnx4sivnx8hi): Renamed to...
(<sur>dot_prodvnx4sivnx8hi): ...this.
* config/aarch64/aarch64-simd-builtins.def: Modify macro
expansion-based initialization and expansion
of (u|s|us)dot_prod builtins.
* config/aarch64/aarch64-builtins.cc
(CODE_FOR_aarch64_sdot_prodv8qi): Define as alias to
new CODE_FOR_sdot_prodv2siv8qi.
(CODE_FOR_aarch64_udot_prodv8qi): Define as alias to
new CODE_FOR_udot_prodv2siv8qi.
(CODE_FOR_aarch64_usdot_prodv8qi): Define as alias to
new CODE_FOR_usdot_prodv2siv8qi.
(CODE_FOR_aarch64_sdot_prodv16qi): Define as alias to
new CODE_FOR_sdot_prodv4siv16qi.
(CODE_FOR_aarch64_udot_prodv16qi): Define as alias to
new CODE_FOR_udot_prodv4siv16qi.
(CODE_FOR_aarch64_usdot_prodv16qi): Define as alias to
new CODE_FOR_usdot_prodv4siv16qi.
* config/aarch64/aarch64-sve-builtins-base.cc
(svdot_impl::expand): s/direct/convert/ in
`convert_optab_handler_for_sign' function call.
(svusdot_impl::expand): add second mode argument in call to
`code_for_dot_prod'.
* config/aarch64/aarch64-sve-builtins.cc
(function_expander::convert_optab_handler_for_sign): New class
method.
* config/aarch64/aarch64-sve-builtins.h
(class function_expander): Add prototype for new
`convert_optab_handler_for_sign' method.

gcc/testsuite/ChangeLog:
* gcc.target/aarch64/sme/vect-dotprod-twoway.c (udot2): New.

autovectorizer: Add basic support for convert optabs

Given the shift from modeling dot products as direct optabs to
treating them as conversion optabs, we make necessary changes to the
autovectorizer code to ensure that given the relevant tree code,
together with the input and output data modes, we can retrieve the
relevant optab and subsequently the insn_code for it.

gcc/ChangeLog:

* gimple-match-exports.cc (directly_supported_p): Add overload
for conversion-type optabs.
* gimple-match.h (directly_supported_p): Add new function
prototype.
* optabs.cc (expand_widen_pattern_expr): Make the
DOT_PROD_EXPR tree code use `find_widening_optab_handler' to
retrieve icode.
* tree-vect-loop.cc (vect_is_emulated_mixed_dot_prod): make it
call conversion-type overloaded `directly_supported_p'.
* tree-vect-patterns.cc (vect_supportable_conv_optab_p): New.
(vect_recog_dot_prod_pattern): s/direct/conv/ in call to
`vect_supportable_direct_optab_p'.

optabs: Make all `*dot_prod_optab's modeled as conversions

Given the specification in the GCC internals manual defines the
{u|s}dot_prod<m> standard name as taking "two signed elements of the
same mode, adding them to a third operand of wider mode", there is
currently ambiguity in the relationship between the mode of the first
two arguments and that of the third.

This vagueness means that, in theory, different modes may be
supportable in the third argument.  This flexibility would allow for a
given backend to add to the accumulator a different number of
vectorized products, e.g. A backend may provide instructions for both:

  accum += a[0] * b[0] + a[1] * b[1] + a[2] * b[2] + a[3] * b[3]

and

  accum += a[0] * b[0] + a[1] * b[1],

as is now seen in the SVE2.1 extension to AArch64.  In spite of the
aforementioned flexibility, modeling the dot-product operation as a
direct optab means that we have no way to encode both input and the
accumulator data modes into the backend pattern name, which prevents
us from harnessing this flexibility.

We therefore make all dot_prod optabs conversions, allowing, for
example, for the encoding of both 2-way and 4-way dot product backend
patterns.

gcc/ChangeLog:

* optabs.def (sdot_prod_optab): Convert from OPTAB_D to
OPTAB_CD.
(udot_prod_optab): Likewise.
(usdot_prod_optab): Likewise.
* doc/md.texi (Standard Names): update entries for u,s and us
dot_prod names.

tree-optimization/116879 - failure to recognize non-empty latch

When we relaxed the vectorizers constraint on loop structure verifying
the emptiness of the latch became too lose as can be seen in the case
for PR116879 where the latch effectively contains two basic-blocks
which one being an unmerged forwarder that's not empty.

PR tree-optimization/116879
* tree-vect-loop.cc (vect_analyze_loop_form): Scan all
blocks that form the latch.

* gcc.dg/pr116879.c: New testcase.

Merge branch 'master' of git+ssh://gcc.gnu.org/git/gcc into devel/nothrow-detection

middle-end: check explicitly for external or constants when checking for loop invariant [PR116817]

The previous check if a value was external was checking
!vect_get_internal_def (vinfo, var) but this of course isn't completely right
as they could reductions etc.

This changes the check to just explicitly look at externals and constants.
Note that reductions remain unhandled here, but we don't support codegen of
boolean reductions today anyway.

So at the time we do then this would have the be handled as well in lowering.

gcc/ChangeLog:

PR tree-optimization/116817
* tree-vect-patterns.cc (vect_recog_bool_pattern): Check for const or
externals.

gcc/testsuite/ChangeLog:

PR tree-optimization/116817
* g++.dg/vect/pr116817.cc: New test.

Ensure coarrays in calls use a descriptor [PR81265]

gcc/fortran/ChangeLog:

PR fortran/81265

* trans-expr.cc (gfc_conv_procedure_call): Ensure coarrays use a
descriptor when passed.

gcc/testsuite/ChangeLog:

* gfortran.dg/coarray/pr81265.f90: New test.

tree-optimization/116842 - vectorizer load hosting breaks UID order

The following fixes the case when vectorizing a load hoists an invariant
load and dependent stmts, thereby breaking UID order of said stmts.
While we duplicate the load we just move the dependences.

PR tree-optimization/116842
* tree-vect-stmts.cc (hoist_defs_of_uses): Sort stmts to hoist
after UID to avoid breaking vect_stmt_dominates_stmt_p.

* g++.dg/torture/pr116842.C: New testcase.

tree-optimization/116785 - relax volatile handling in PTA

When there's volatile qualified stores we do not have to treat the
destination as pointing to ANYTHING.  It's only when reading from
it that we want to treat the resulting pointers as pointing to ANYTHING.

PR tree-optimization/116785
* tree-ssa-structalias.cc (get_constraint_for_1): Only
volatile qualified reads produce ANYTHING.

tree-optimization/116850 - corrupt post-dom info

Path isolation computes post-dominators on demand but can end up
splitting blocks after that, wrecking it.  We can delay splitting
of blocks until we no longer need the post-dom info which is what
the following patch does to solve the issue.

PR tree-optimization/116850
* gimple-ssa-isolate-paths.cc (bb_split_points): New global.
(insert_trap): Delay BB splitting if post-doms are computed.
(find_explicit_erroneous_behavior): Process delayed BB
splitting after releasing post dominators.
(gimple_ssa_isolate_erroneous_paths): Do not free post-dom
info here.

* gcc.dg/pr116850.c: New testcase.

Match: Support form 1 for scalar signed integer SAT_SUB

This patch would like to support the form 1 of the scalar signed
integer SAT_SUB.  Aka below example:

Form 1:
  #define DEF_SAT_S_SUB_FMT_1(T, UT, MIN, MAX) \
  T __attribute__((noinline))                  \
  sat_s_sub_##T##_fmt_1 (T x, T y)             \
  {                                            \
    T minus = (UT)x - (UT)y;                   \
    return (x ^ y) >= 0                        \
      ? minus                                  \
      : (minus ^ x) >= 0                       \
        ? minus                                \
        : x < 0 ? MIN : MAX;                   \
  }

DEF_SAT_S_SUB_FMT_1(int8_t, uint8_t, INT8_MIN, INT8_MAX)

Before this patch:
   4   │ __attribute__((noinline))
   5   │ int8_t sat_s_sub_int8_t_fmt_1 (int8_t x, int8_t y)
   6   │ {
   7   │   int8_t minus;
   8   │   unsigned char x.0_1;
   9   │   unsigned char y.1_2;
  10   │   unsigned char _3;
  11   │   signed char _4;
  12   │   signed char _5;
  13   │   int8_t _6;
  14   │   _Bool _11;
  15   │   signed char _12;
  16   │   signed char _13;
  17   │   signed char _14;
  18   │   signed char _15;
  19   │
  20   │ ;;   basic block 2, loop depth 0
  21   │ ;;    pred:       ENTRY
  22   │   x.0_1 = (unsigned char) x_7(D);
  23   │   y.1_2 = (unsigned char) y_8(D);
  24   │   _3 = x.0_1 - y.1_2;
  25   │   minus_9 = (int8_t) _3;
  26   │   _4 = x_7(D) ^ y_8(D);
  27   │   _5 = x_7(D) ^ minus_9;
  28   │   _15 = _4 & _5;
  29   │   if (_15 < 0)
  30   │     goto <bb 3>; [41.00%]
  31   │   else
  32   │     goto <bb 4>; [59.00%]
  33   │ ;;    succ:       3
  34   │ ;;                4
  35   │
  36   │ ;;   basic block 3, loop depth 0
  37   │ ;;    pred:       2
  38   │   _11 = x_7(D) < 0;
  39   │   _12 = (signed char) _11;
  40   │   _13 = -_12;
  41   │   _14 = _13 ^ 127;
  42   │ ;;    succ:       4
  43   │
  44   │ ;;   basic block 4, loop depth 0
  45   │ ;;    pred:       2
  46   │ ;;                3
  47   │   # _6 = PHI <minus_9(2), _14(3)>
  48   │   return _6;
  49   │ ;;    succ:       EXIT
  50   │
  51   │ }

After this patch:
   4   │ __attribute__((noinline))
   5   │ int8_t sat_s_sub_int8_t_fmt_1 (int8_t x, int8_t y)
   6   │ {
   7   │   int8_t _6;
   8   │
   9   │ ;;   basic block 2, loop depth 0
  10   │ ;;    pred:       ENTRY
  11   │   _6 = .SAT_SUB (x_7(D), y_8(D)); [tail call]
  12   │   return _6;
  13   │ ;;    succ:       EXIT
  14   │
  15   │ }

The below test suites are passed for this patch.
* The rv64gcv fully regression test.
* The x86 bootstrap test.
* The x86 fully regression test.

gcc/ChangeLog:

* match.pd: Add case 1 matching pattern for signed SAT_SUB.
* tree-ssa-math-opts.cc (gimple_signed_integer_sat_sub): Add new
decl for generated SAT_SUB matching func.
(match_unsigned_saturation_sub): Rename from...
(match_saturation_sub): ...Rename to and add signed SAT_SUB matching.
(math_opts_dom_walker::after_dom_children): Leverage the named
match func for both the unsigned and signed SAT_SUB.

Signed-off-by: Pan Li <pan2.li@intel.com>

Daily bump.

RISC-V: Add testcases for form 1 of scalar signed SAT_SUB

Form 1:
  #define DEF_SAT_S_SUB_FMT_1(T, UT, MIN, MAX) \
  T __attribute__((noinline))                  \
  sat_s_sub_##T##_fmt_1 (T x, T y)             \
  {                                            \
    T minus = (UT)x - (UT)y;                   \
    return (x ^ y) >= 0                        \
      ? minus                                  \
      : (minus ^ x) >= 0                       \
        ? minus                                \
        : x < 0 ? MIN : MAX;                   \
  }

DEF_SAT_S_SUB_FMT_1(int8_t, uint8_t, INT8_MIN, INT8_MAX)

The below test are passed for this patch.
* The rv64gcv fully regression test.

It is test only patch and obvious up to a point, will commit it
directly if no comments in next 48H.

gcc/testsuite/ChangeLog:

* gcc.target/riscv/sat_arith.h: Add test helper macros.
* gcc.target/riscv/sat_arith_data.h: Add test data for SAT_SUB.
* gcc.target/riscv/sat_s_sub-1-i16.c: New test.
* gcc.target/riscv/sat_s_sub-1-i32.c: New test.
* gcc.target/riscv/sat_s_sub-1-i64.c: New test.
* gcc.target/riscv/sat_s_sub-1-i8.c: New test.
* gcc.target/riscv/sat_s_sub-run-1-i16.c: New test.
* gcc.target/riscv/sat_s_sub-run-1-i32.c: New test.
* gcc.target/riscv/sat_s_sub-run-1-i64.c: New test.
* gcc.target/riscv/sat_s_sub-run-1-i8.c: New test.

Signed-off-by: Pan Li <pan2.li@intel.com>

RISC-V: Implement scalar SAT_SUB for signed integer

This patch would like to implement the sssub form 1.  Aka:

Form 1:
  #define DEF_SAT_S_SUB_FMT_1(T, UT, MIN, MAX) \
  T __attribute__((noinline))                  \
  sat_s_sub_##T##_fmt_1 (T x, T y)             \
  {                                            \
    T minus = (UT)x - (UT)y;                   \
    return (x ^ y) >= 0                        \
      ? minus                                  \
      : (minus ^ x) >= 0                       \
        ? minus                                \
        : x < 0 ? MIN : MAX;                   \
  }

DEF_SAT_S_SUB_FMT_1(int8_t, uint8_t, INT8_MIN, INT8_MAX)

Before this patch:
  10   │ sat_s_sub_int8_t_fmt_1:
  11   │     subw    a5,a0,a1
  12   │     slliw   a5,a5,24
  13   │     sraiw   a5,a5,24
  14   │     xor a1,a0,a1
  15   │     xor a4,a0,a5
  16   │     and a1,a1,a4
  17   │     blt a1,zero,.L4
  18   │     mv  a0,a5
  19   │     ret
  20   │ .L4:
  21   │     srai    a0,a0,63
  22   │     xori    a5,a0,127
  23   │     mv  a0,a5
  24   │     ret

After this patch:
  10   │ sat_s_sub_int8_t_fmt_1:
  11   │     sub a4,a0,a1
  12   │     xor a5,a0,a4
  13   │     xor a1,a0,a1
  14   │     and a5,a5,a1
  15   │     srli    a5,a5,7
  16   │     andi    a5,a5,1
  17   │     srai    a0,a0,63
  18   │     xori    a3,a0,127
  19   │     neg a0,a5
  20   │     addi    a5,a5,-1
  21   │     and a3,a3,a0
  22   │     and a0,a4,a5
  23   │     or  a0,a0,a3
  24   │     slliw   a0,a0,24
  25   │     sraiw   a0,a0,24
  26   │     ret

The below test suites are passed for this patch.
* The rv64gcv fully regression test.

gcc/ChangeLog:

* config/riscv/riscv-protos.h (riscv_expand_sssub): Add new func
decl for expanding signed SAT_SUB.
* config/riscv/riscv.cc (riscv_expand_sssub): Add new func impl
for expanding signed SAT_SUB.
* config/riscv/riscv.md (sssub<mode>3): Add new pattern sssub
for scalar signed integer.

Signed-off-by: Pan Li <pan2.li@intel.com>

cselib: Discard useless locs of preserved VALUEs [PR116627]

remove_useless_values iteratively discards useless locs (locs of
cselib_val which refer to non-preserved VALUEs with no locations),
which in turn can make further values useless until no further VALUEs
are made useless and then discards the useless VALUEs.

Preserved VALUEs (something done during var-tracking only I think)
live in a different hash table, cselib_preserved_hash_table rather
than cselib_hash_table.  cselib_find_slot first looks up slot in
cselib_preserved_hash_table and only if not found looks it up in
cselib_hash_table (and INSERTs only into the latter), whereas preservation
of a VALUE results in move of a cselib_val from the latter to the former
hash table.

The testcase in the PR (apparently too fragile, it only reproduces on 14
branch with various flags on a single arch, not on trunk) ICEs, because
we have a preserved VALUE (QImode with (const_int 0) as one of the locs).
In a different BB SImode r2 is looked up, a non-preserved VALUE is created
for it, and the r13-2916 added code attempts to lookup also SUBREGs of that
in narrower modes, among those QImode, so adds to that SImode r2
non-preserve VALUE a new loc of (subreg:QI (value:SI) 0).  That SImode
value is considered useless, so remove_useless_value discards it, but
nothing discarded it from the preserved VALUE's loc_list, so when looking
something up in the hash table we ICE trying to derevence CSELIB_VAL
of the discarded VALUE.

I think we need to discuard useless locs even from the preserved VALUEs.
That IMHO shouldn't create any further useless VALUEs, the preserved
VALUEs are never useless, so we don't need to iterate with it, can do it
just once, but IMHO it needs to be done because actually
discard_useless_values.

The following patch does that.

2024-09-29  Jakub Jelinek  <jakub@redhat.com>

PR target/116627
* cselib.cc (remove_useless_values): Discard useless locs
even from preserved cselib_vals in cselib_preserved_hash_table
hash table.

testsuite: XFAIL gfortran.dg/initialization_25.f90 properly (again)

dg-error needs an argument for "why" / a comment.

gcc/testsuite/ChangeLog:
PR fortran/116858

* gfortran.dg/initialization_25.f90: Fix dg-error arguments.

[PATCH] SH: Document extended asm operand modifers

From: Pietro Monteiro <pietro@sociotechnical.xyz>

SH: Document extended asm operand modifers

Tested by running "make info pdf html" and looking at the pdf and html output. I used the comment on "gcc/config/sh.cc:sh_print_operand()", SH's TARGET_PRINT_OPERAND function, as a guide.

gcc/ChangeLog:
* doc/extend.texi (SH Operand Modifiers): New.

Signed-off-by: Pietro Monteiro <pietro@sociotechnical.xyz>

[PATCH] [PATCH] Avoid integer overflow in gcc.dg/cpp/charconst-3.c (PR testsuite/116806)

The intermediate expression (unsigned char) '\234' * scale overflows
int on int16 targets, causing the test case to fail there.  Fixed by
performing the arithmetic in unsigned type, as suggested by Andrew Pinski.

Regression tested on x86_64-pc-linux-gnu, and on an out-of-tree 16-bit
target with simulator.  Manually checked the generated code for pdp11
and xstormy16.

Ok for trunk? (I don't have commit rights so I'd need help committing it.)

gcc/testsuite/

PR testsuite/116806
* gcc.dg/cpp/charconst-3.c: Perform arithmetic in unsigned
type to avoid integer overflow.

[PATCH v2] RISC-V: Improve code generation for select of consecutive constants

Based on the valuable feedback I received, I decided to implement the patch
in the RTL pipeline. Since a similar optimization already exists in
simplify_binary_operation_1, I chose to generalize my original approach
and place it directly below that code.

The expression (X xor C1) + C2 is simplified to X xor (C1 xor C2) under
the conditions described in the patch. This is a more general optimization,
but it still applies to the RISC-V case, which was my initial goal:

long f1(long x, long y) {
    return (x > y) ? 2 : 3;
}

Before the patch, the generated assembly is:

f1(long, long):
        sgt     a0,a0,a1
        xori    a0,a0,1
        addi    a0,a0,2
        ret

After the patch, the generated assembly is:

f1(long, long):
        sgt     a0,a0,a1
        xori    a0,a0,3
        ret

The patch optimizes cases like x LT/GT y ? 2 : 3 (and x GE/LE y ? 3 : 2),
as initially intended. Since this optimization is more general, I noticed
it also optimizes cases like x < CONST ? 3 : 2 when CONST < 0. I’ve added
tests for these cases as well.

A bit of logic behind the patch: The equality A + B == A ^ B + 2 * (A & B)
always holds true. This can be simplified to A ^ B if 2 * (A & B) == 0.
In our case, we have A == X ^ C1, B == C2 and X is either 0 or 1.

PR target/108038

gcc/ChangeLog:

* simplify-rtx.cc (simplify_context::simplify_binary_operation_1): New
simplification.

gcc/testsuite/ChangeLog:

* gcc.target/riscv/slt-1.c: New test.

doc: Document struct-layout-1.exp for ABI checks

This test helped discover PR116621, so it is worth being documented.

gcc/ChangeLog:

* doc/sourcebuild.texi: Document struct-layout-1.exp.

Signed-off-by: Dimitar Dimitrov <dimitar@dinux.eu>

Daily bump.

Implement FINDLOC for UNSIGNED.

gcc/fortran/ChangeLog:

* check.cc (intrinsic_type_check): Handle unsigned.
(gfc_check_findloc): Likewise.
* gfortran.texi: Include FINDLOC in unsigned documentation.
* iresolve.cc (gfc_resolve_findloc): Use INTEGER version
for UNSIGNED.

gcc/testsuite/ChangeLog:

* gfortran.dg/unsigned_33.f90: New test.

Implement CSHIFT and EOSHIFT for unsigned.

gcc/fortran/ChangeLog:

* check.cc (gfc_check_eoshift): Handle BT_UNSIGNED.
* simplify.cc (gfc_simplify_eoshift): Likewise.
* gfortran.texi: Document CSHIFT and EOSHIFT for UNSIGNED.

gcc/testsuite/ChangeLog:

* gfortran.dg/unsigned_31.f90: New test.
* gfortran.dg/unsigned_32.f90: New test.

doc: Remove i?86-*-linux* installation note from 2003

gcc:
PR target/69374
* doc/install.texi (Specific) <i?86-*-linux*>: Remove note
from 2003.

Daily bump.

c++: Implement resolution for DR 36 [PR116160]

This implements part of P1787 to no longer complain about redeclaring an
entity via using-decl other than in a class scope.

PR c++/116160

gcc/cp/ChangeLog:

* name-lookup.cc (supplement_binding): Allow redeclaration via
USING_DECL if not in class scope.
(do_nonmember_using_decl): Remove function-scope exemption.
(push_using_decl_bindings): Remove outdated comment.

gcc/testsuite/ChangeLog:

* g++.dg/cpp0x/using-enum-3.C: No longer expect an error.
* g++.dg/lookup/using53.C: Remove XFAIL.
* g++.dg/cpp2a/using-enum-11.C: New test.

Signed-off-by: Nathaniel Shead <nathanieloshead@gmail.com>
Reviewed-by: Jason Merrill <jason@redhat.com>

c++: Don't strip USING_DECLs when updating local bindings [PR116748]

Currently update_binding strips USING_DECLs too eagerly, leading to ICEs
in pop_local_decl as it can't find the decl it's popping in the binding
list.  Let's rather try to keep the original USING_DECL around.

This also means that using59.C can point to the location of the
using-decl rather than the underlying object directly; this is in the
direction required to fix PR c++/106851 (though more work is needed to
emit properly helpful diagnostics here).

PR c++/116748

gcc/cp/ChangeLog:

* name-lookup.cc (update_binding): Maintain USING_DECLs in the
binding slots.

gcc/testsuite/ChangeLog:

* g++.dg/lookup/using59.C: Update location.
* g++.dg/lookup/using69.C: New test.

Signed-off-by: Nathaniel Shead <nathanieloshead@gmail.com>
Reviewed-by: Jason Merrill <jason@redhat.com>

c++/modules: Propagate purview/import for templates in duplicate_decls [PR116803]

We need to ensure that for a declaration in the module purview, that the
resulting declaration has PURVIEW_P set and IMPORT_P cleared so that we
understand it might be something requiring exporting.  This is normally
handled for a declaration by set_instantiating_module, but when this
declaration is a redeclaration duplicate_decls needs to propagate this
to olddecl.

This patch only changes the logic for template declarations, because in
the non-template case the whole contents of olddecl's DECL_LANG_SPECIFIC
is replaced with newdecl's (which includes these flags), so there's
nothing to do.

PR c++/116803

gcc/cp/ChangeLog:

* decl.cc (duplicate_decls): Propagate DECL_MODULE_PURVIEW_P and
DECL_MODULE_IMPORT_P for template redeclarations.

gcc/testsuite/ChangeLog:

* g++.dg/modules/merge-18_a.H: New test.
* g++.dg/modules/merge-18_b.H: New test.
* g++.dg/modules/merge-18_c.C: New test.

Signed-off-by: Nathaniel Shead <nathanieloshead@gmail.com>
Reviewed-by: Jason Merrill <jason@redhat.com>

libstdc++: Fix more pedwarns in headers for C++98

Some tests e.g.  17_intro/headers/c++1998/all_pedantic_errors.cc FAIL
with GLIBCXX_TESTSUITE_STDS=98 due to numerous C++11 extensions still in
use in the library headers. The recent changes to not make them system
headers means we get warnings now.

This change adds more diagnostic pragmas to suppress those warnings.

libstdc++-v3/ChangeLog:

* include/bits/istream.tcc: Add diagnostic pragmas around uses
of long long and extern template.
* include/bits/locale_facets.h: Likewise.
* include/bits/locale_facets.tcc: Likewise.
* include/bits/locale_facets_nonio.tcc: Likewise.
* include/bits/ostream.tcc: Likewise.
* include/bits/stl_algobase.h: Likewise.
* include/c_global/cstdlib: Likewise.
* include/ext/pb_ds/detail/resize_policy/hash_prime_size_policy_imp.hpp:
Likewise.
* include/ext/pointer.h: Likewise.
* include/ext/stdio_sync_filebuf.h: Likewise.
* include/std/istream: Likewise.
* include/std/ostream: Likewise.
* include/tr1/cmath: Likewise.
* include/tr1/type_traits: Likewise.
* include/tr1/functional_hash.h: Likewise. Remove semi-colons
at namespace scope that aren't needed after macro expansion.
* include/tr1/tuple: Remove semi-colon at namespace scope.
* include/bits/vector.tcc: Change LL suffix to just L.

libstdc++: Refactor experimental::filesystem::path string conversions

I noticed a -Wc++17-extensions warning due to use of if-constexpr in
std::experimental::filesystem::path, which was not protected by
diagnostic pragmas to disable the warning.

While adding the pragmas I noticed that other places in the same file
use tag dispatching and multiple overloads instead of if-constexpr.
Since we're already using it in that file, we might as well just use it
everywhere.

libstdc++-v3/ChangeLog:

* include/experimental/bits/fs_path.h (path::_Cvt): Refactor to
use if-constexpr.
(path::string(const Allocator&)): Likewise.

libstdc++: Fix -Wsign-compare warning in std::string::resize_for_overwrite

libstdc++-v3/ChangeLog:

* include/bits/basic_string.tcc (resize_for_overwrite): Fix
-Wsign-compare warning.
* include/bits/cow_string.h (resize_for_overwrite): Likewise.

c++: ICE with structured bindings and m-d array [PR102594]

We ICE in decay_conversion with this test:

  struct S {
    S() {}
  };
  S arr[1][1];
  auto [m](arr3);

But not when the last line is:

  auto [n] = arr3;

Therefore the difference is between copy- and direct-init.  In
particular, in build_vec_init we have:

  if (direct_init)
    from = build_tree_list (NULL_TREE, from);

and then we call build_vec_init again with init==from.  Then
decay_conversion gets the TREE_LIST and it crashes.

build_aggr_init has:

              /* Wrap the initializer in a CONSTRUCTOR so that build_vec_init
                 recognizes it as direct-initialization.  */
              init = build_constructor_single (init_list_type_node,
                                               NULL_TREE, init);
              CONSTRUCTOR_IS_DIRECT_INIT (init) = true;

so I propose to do the same in build_vec_init.

PR c++/102594

gcc/cp/ChangeLog:

* init.cc (build_vec_init): Build up a CONSTRUCTOR to signal
direct-initialization rather than a TREE_LIST.

gcc/testsuite/ChangeLog:

* g++.dg/cpp1z/decomp61.C: New test.

Reviewed-by: Jason Merrill <jason@redhat.com>

libstdc++: Fix test FAILs due to -Wreturn-local-addr

This fixes two FAILs due to -Wpointer-arith warnings when testing with
c++11 or c++14 dialects.

libstdc++-v3/ChangeLog:

* testsuite/20_util/bind/dangling_ref.cc: Add an additional
dg-warning for -Wreturn-local-addr warning.
* testsuite/30_threads/packaged_task/cons/dangling_ref.cc:
Likewise.

libstdc++: Fix test FAIL due to -Wpointer-arith

This fixes a FAIL due to a -Wpointer-arith warning when testing with
c++11 or c++14 dialects. As an extension our std::atomic<void*> supports
pointer arithmetic in C++11 and C++14, but due to the system header
changes there is now a warning about it. The warning seems reasonable,
so rather than suppress it we should make the test expect it.

While looking into this I decided to simplify some of the code related
to atomic<T*> arithmetic.

libstdc++-v3/ChangeLog:

* include/bits/atomic_base.h (__atomic_base<T*>::_M_type_size):
Replace overloaded functions with static _S_type_size.
* include/std/atomic (atomic<T*>): Use is_object_v instead of
is_object.
* testsuite/29_atomics/atomic/operators/pointer_partial_void.cc:
Add dg-warning for -Wpointer-arith warning.

aarch64: fix build failure on aarch64-none-elf

A previous patch ([1]) introduced a build regression on aarch64-none-elf
target. The changes were primarilly tested on aarch64-unknown-linux-gnu,
so the issue was missed during development.
The includes are slighly different between the two targets, and due to some
include rules ([2]), "aarch64-unwind-def.h" was not found.

[1]: https://gcc.gnu.org/git/?p=gcc.git;a=commit;h=bdf41d627c13bc5f0dc676991f4513daa9d9ae36

[2]: https://gcc.gnu.org/onlinedocs/cpp/Include-Syntax.html
> include "file"
> ...  It searches for a file named file first in the directory
> containing the current file, ...

libgcc/ChangeLog:

* config/aarch64/aarch64-unwind.h: Fix header path.

diagnostic: Save/restore diagnostic context history and push/pop state for PCH [PR116847]

The following patch on top of the just posted cleanup patch
saves/restores the m_classification_history and m_push_list
vectors for PCH.  Without that as the testcase shows during parsing
of the templates we don't report ignored diagnostics, but after loading
PCH header when instantiating those templates those warnings can be
emitted.  This doesn't show up on x86_64-linux build because configure
injects there -fcf-protection -mshstk flags during library build (and so
also during PCH header creation), but make check doesn't use those flags
and so the PCH header is ignored.

2024-09-26  Jakub Jelinek  <jakub@redhat.com>

PR libstdc++/116847
gcc/
* diagnostic.h (diagnostic_option_classifier): Add pch_save and
pch_restore method declarations.
(diagnostic_context): Add pch_save and pch_restore inline method
definitions.
* diagnostic.cc (diagnostic_option_classifier::pch_save): New method.
(diagnostic_option_classifier::pch_restore): Likewise.
gcc/c-family/
* c-pch.cc: Include diagnostic.h.
(c_common_write_pch): Call global_dc->pch_save.
(c_common_read_pch): Call global_dc->pch_restore.
gcc/testsuite/
* g++.dg/pch/pr116847.C: New test.
* g++.dg/pch/pr116847.Hs: New test.

diagnostic: Use vec instead of custom array reallocations for m_classification_history/m_push_list [PR116847]

diagnostic.h already relies on vec.h, it uses auto_vec in one spot.

The following patch converts m_classification_history and m_push_list
hand-managed arrays to vec templates.
The main advantage is exponential rather than linear reallocation,
e.g. with current libstdc++ headers if one includes all the standard
headers there could be ~ 300 reallocations of the m_classification_history
array (sure, not all of them will result in actually copying the data, but
still).
In addition to that it fixes some formatting issues in the code.

2024-09-26  Jakub Jelinek  <jakub@redhat.com>

PR libstdc++/116847
* diagnostic.h (diagnostic_option_classifier): Change type
of m_classification_history from diagnostic_classification_change_t *
to vec<diagnostic_classification_change_t>.  Change type of
m_push_list from int * to vec<int>.  Remove m_n_classification_history
and m_n_push members.
* diagnostic.cc (diagnostic_option_classifier::init): Set m_push_list
to vNULL rather than nullptr.  Don't initialize m_n_push.  Initialize
m_classification_history to vNULL.
(diagnostic_option_classifier::fini): Call release () method on
m_push_list instead of free on it.  Call release () on
m_classification_history.  Don't clear m_n_push.
(diagnostic_option_classifier::push): Adjust for m_push_list and
m_classification_history being vectors rather than custom allocated
arrays with counter.
(diagnostic_option_classifier::pop): Likewise.
(classify_diagnostic): Adjust for m_classification_history being
vector rather than custom allocated array with counter.
(update_effective_level_from_pragmas): Likewise.

i386: Modernize AMD processor types

Use iterative PTA definitions for members of the same AMD processor family.

Also, fix a couple of related M_CPU_TYPE/M_CPU_SUBTYPE inconsistencies.

No functional changes intended.

gcc/ChangeLog:

* config/i386/i386.h: Add PTA_BDVER1, PTA_BDVER2, PTA_BDVER3,
PTA_BDVER4, PTA_BTVER1 and PTA_BTVER2.
* common/config/i386/i386-common.cc (processor_alias_table)
<"bdver1">: Use PTA_BDVER1.
<"bdver2">: Use PTA_BDVER2.
<"bdver3">: Use PTA_BDVER3.
<"bdver4">: Use PTA_BDVER4.
<"btver1">: Use PTA_BTVER1.  Use M_CPU_TYPE (AMD_BTVER1).
<"btver2">: Use PTA_BTVER2.
<"shanghai>: Use M_CPU_SUBTYPE (AMDFAM10H_SHANGHAI).
<"istanbul>: Use M_CPU_SUBTYPE (AMDFAM10H_ISTANBUL).

Widening-Mul: Fix one ICE when iterate on phi node

We iterate all phi node of bb to try to match the SAT_* pattern
for scalar integer.  We also remove the phi mode when the relevant
pattern matched.

Unfortunately the iterator may have no idea the phi node is removed
and continue leverage the free data and then ICE similar as below.

[0] psi ptr 0x75216340c000
[0] psi ptr 0x75216340c400
[1] psi ptr 0xa5a5a5a5a5a5a5a5 <=== GC freed pointer.

during GIMPLE pass: widening_mul
tmp.c: In function ‘f’:
tmp.c:45:6: internal compiler error: Segmentation fault
   45 | void f(int rows, int cols) {
      |      ^
0x36e2788 internal_error(char const*, ...)
        ../../gcc/diagnostic-global-context.cc:517
0x18005f0 crash_signal
        ../../gcc/toplev.cc:321
0x752163c4531f ???
        ./signal/../sysdeps/unix/sysv/linux/x86_64/libc_sigaction.c:0
0x103ae0e bool is_a_helper<gphi*>::test<gimple>(gimple*)
../../gcc/gimple.h:1256
0x103f9a5 bool is_a<gphi*, gimple>(gimple*)
        ../../gcc/is-a.h:232
0x103dc78 gphi* as_a<gphi*, gimple>(gimple*)
../../gcc/is-a.h:255
0x104f12e gphi_iterator::phi() const
        ../../gcc/gimple-iterator.h:47
0x1a57bef after_dom_children
        ../../gcc/tree-ssa-math-opts.cc:6140
0x3344482 dom_walker::walk(basic_block_def*)
        ../../gcc/domwalk.cc:354
0x1a58601 execute
        ../../gcc/tree-ssa-math-opts.cc:6312

This patch would like to fix the iterate on modified collection problem
by backup the next phi in advance.

The below test suites are passed for this patch.
* The rv64gcv fully regression test.
* The x86 bootstrap test.
* The x86 fully regression test.

PR middle-end/116861

gcc/ChangeLog:

* tree-ssa-math-opts.cc (math_opts_dom_walker::after_dom_children): Backup
the next psi iterator before remove the phi node.

gcc/testsuite/ChangeLog:

* gcc.dg/torture/pr116861-1.c: New test.

Signed-off-by: Pan Li <pan2.li@intel.com>

Fix sorting in Contributors.html

The following moves my entry to where it belongs alphabetically
(it wasn't moved when s/Guenther/Biener/).

* doc/contrib.texi (Richard Biener): Move entry.

libgcc, Darwin: Don't build legacy libgcc_s.1 on macOS 14 [PR116809]

d9cafa0c4f0a stopped building libgcc_s.1 on macOS >= 15, in part because
that is required to bootstrap the compiler using the macOS 15 SDK. The
macOS 15 SDK ships in Xcode 16, which also runs on macOS 14. libgcc_s.1
can no longer be built on macOS 14 using Xcode 16 by the same logic that
the previous change disabled it for macOS 15.

PR target/116809

libgcc/ChangeLog:

* config.host: Don't build legacy libgcc_s.1 on macOS 14.

Signed-off-by: Mark Mentovai <mark@mentovai.com>

c++/coro: ignore cleanup_point_exprs while expanding awaits [PR116793]

If we reach a CLEANUP_POINT_EXPR while trying to walk statements, we
actually care about the statement or statement list contained within it.

Indeed, such a construction started happening with
r15-3513-g964577c31df206, after temporary promotion.  In the test case
presented in PR116793, the compiler generated:

  <<cleanup_point {
    struct _cleanup_task Aw0 [value-expr: frame_ptr->Aw0_2_3];
    int T002 [value-expr: frame_ptr->T002_2_3];

      int T002 [value-expr: frame_ptr->T002_2_3];
    <<cleanup_point <<< Unknown tree: expr_stmt
      (void) (T002 = TARGET_EXPR <D.20994, 3>) >>>>>;
      struct _cleanup_task Aw0 [value-expr: frame_ptr->Aw0_2_3];
    <<cleanup_point <<< Unknown tree: expr_stmt
      (void) (Aw0 = TARGET_EXPR <D.20995, func ((int &) &T002)>) >>>>>;
    <<cleanup_point <<< Unknown tree: expr_stmt
      (void) (D.22450 = <<< Unknown tree: co_await
        TARGET_EXPR <D.20995, func ((int &) &T002)>
        Aw0

        {_cleanup_task::await_ready (&Aw0), _cleanup_task::await_suspend<_task1::promise_type> (&Aw0, TARGET_EXPR <D.21078, _Coro_self_handle>), <<< Unknown tree: aggr_init_expr
          4
          await_resume
          D.22443
          &Aw0 >>>}
        0 >>>) >>>>>;
    <<cleanup_point <<< Unknown tree: expr_stmt
      (void) (D.20991 = (struct tuple &) &D.22450) >>>>>;
  }
  D.22467 = 1;
  int & i [value-expr: frame_ptr->i_1_2];
  <<cleanup_point <<< Unknown tree: expr_stmt
    (void) (i = std::get<0, int&> (NON_LVALUE_EXPR <D.20991>)) >>>>>;>>;

... i.e. a statement list within a cleanup point.  In such a case, we
don't actually care about the cleanup point, but we do care about the
statement inside, so, we can just walk down into the CLEANUP_POINT_EXPR.

PR c++/116793

gcc/cp/ChangeLog:

* coroutines.cc (await_statement_expander): Just process
subtrees if encountering a CLEANUP_POINT_EXPR.

gcc/testsuite/ChangeLog:

* g++.dg/coroutines/pr116793-1.C: New test.

c++: simplify handling implicit INDIRECT_REF and co_await in convert_to_void

convert_to_void has, so far, when converting a co_await expression to
void altered the await_resume expression of a co_await so that it is
also converted to void.  This meant that the type of the await_resume
expression, which is also supposed to be the type of the whole co_await
expression, was not the same as the type of the CO_AWAIT_EXPR tree.

While this has not caused problems so far, it is unexpected, I think.

Also, convert_to_void had a special case when an INDIRECT_REF wrapped a
CALL_EXPR.  In this case, we also diagnosed maybe_warn_nodiscard.  This
was a duplication of logic related to converting call expressions to
void.

Instead, we can generalize a bit, and rather discard the expression that
was implicitly dereferenced instead.

This patch changes the diagnostic of:

  void f(struct S* x) { static_cast<volatile S&>(*x); }

... from:

  warning: indirection will not access object of incomplete type
           'volatile S' in statement

... to:

  warning: implicit dereference will not access object of type
           ‘volatile S’ in statement

... but should have no impact in other cases.

gcc/cp/ChangeLog:

* coroutines.cc (co_await_get_resume_call): Return a tree
directly, rather than a tree pointer.
* cp-tree.h (co_await_get_resume_call): Adjust signature
accordingly.
* cvt.cc (convert_to_void): Do not alter CO_AWAIT_EXPRs when
discarding them.  Simplify handling implicit INDIRECT_REFs.

gcc/testsuite/ChangeLog:

* g++.dg/coroutines/nodiscard-1.C: New test.

c++/coro: prevent ICV_STATEMENT diagnostics in temp promotion [PR116502]

If such a diagnostic is necessary, it has already been emitted,
otherwise, it is not correct and emitting it here is inactionable by the
user, and bogus.

PR c++/116502

gcc/cp/ChangeLog:

* coroutines.cc (maybe_promote_temps): Convert temporary
initializers to void without complaining.

gcc/testsuite/ChangeLog:

* g++.dg/coroutines/maybe-unused-1.C: New test.
* g++.dg/coroutines/pr116502.C: New test.

[MAINTAINERS]: Add myself as MVE Reviewer for the AArch32 (arm) port

ChangeLog:
* MAINTAINERS: Add myself as MVE Reviewer for the AArch32 (arm)
port.

libgomp.texi: Remove now duplicate TR13 item

Remove an item under "Other new TR 13 features" that since the last commit
(r15-3917-g6b7eaec20b046e) to this file is is covered by the added
  "New @code{storage} map-type modifier; context-dependent @code{alloc} and
   @code{release} are aliases"
  "Update of the map-type decay for mapping and @code{declare_mapper}"

libgomp/
* libgomp.texi (TR13 status): Update semi-duplicated, semi-obsoleted
item; remove left-over half-sentence.

RISC-V/libgcc: Save/Restore routines for E goes with ABI.

That Save/Restore routines for E can be used for RVI with ILP32E ABI.

libgcc/ChangeLog:

* config/riscv/save-restore.S: Check with __riscv_abi_rve rather than
__riscv_32e.

libgomp.texi: fix formatting; add post-TR13 OpenMP impl. status items

libgomp/
* libgomp.texi (OpenMP Technical Report 13): Change @emph to @code;
add two post-TR13 OpenMP 6.0 items.

tree-optimization/116818 - try VMAT_GATHER_SCATTER also for SLP

When not doing SLP and we end up with VMAT_ELEMENTWISE we consider
using strided loads, aka VMAT_GATHER_SCATTER.  The following moves
this logic down to also apply to SLP where we now can end up
using VMAT_ELEMENTWISE as well.

PR tree-optimization/116818
* tree-vect-stmts.cc (get_group_load_store_type): Consider
VMAT_GATHER_SCATTER instead of VMAT_ELEMENTWISE also for SLP.
(vectorizable_load): For single-lane VMAT_GATHER_SCATTER also
ignore permutations.

Fix bogus SLP nvector compute in check_load_store_for_partial_vectors

We have a new overload for vect_get_num_copies that handles both
SLP and non-SLP.  Use it and avoid the division by group_size
for SLP when not using load-store lanes.

* tree-vect-stmts.cc (check_load_store_for_partial_vectors):
Use the new vect_get_num_copies overload.  Only divide by
group_size for SLP for load-store lanes.

unswitch: Replace manual ondemand maybe_undef with ssa_name_maybe_undef_p/mark_ssa_maybe_undefs [PR116848]

The ondemand maybe_undef that follows phis was added in r7-6427-g8b670f93ab1136
but then later ssa_name_maybe_undef_p/mark_ssa_maybe_undefs was added in
r13-972-gbe2861fe8c527a. This moves the ondemand one to use
mark_ssa_maybe_undefs/ssa_name_maybe_undef_p instead. Which itself will be
faster since the mark_ssa_maybe_undefs is a walk based on the uses of
undefined names (and only once) rather than a walk based on the def of
ones which are more likely defined (and on demand).

Even though the ondemand maybe_undef had some extra special cases, those won't make
a big difference in most code.

Bootstrapped and tested on x86_64-linux-gnu.

PR tree-optimization/116848

gcc/ChangeLog:

* tree-ssa-loop-unswitch.cc (tree_ssa_unswitch_loops): Call mark_ssa_maybe_undefs.
(is_maybe_undefined): Call ssa_name_maybe_undef_p instead of ondemand undef.

Signed-off-by: Andrew Pinski <quic_apinski@quicinc.com>

c++/modules: Allow imported references in constant expressions

Currently the streaming code uses TREE_CONSTANT to determine whether an
entity will have a definition that is interesting to stream out.  This
is not sufficient, however; we also need to write the definition of
references, since although not TREE_CONSTANT they can still be usable in
constant expressions.

As such this patch uses the existing decl_maybe_constant_var function
which correctly handles this case.

gcc/cp/ChangeLog:

* module.cc (has_definition): Use decl_maybe_constant_var
instead of TREE_CONSTANT.

gcc/testsuite/ChangeLog:

* g++.dg/modules/cexpr-5_a.C: New test.
* g++.dg/modules/cexpr-5_b.C: New test.

Signed-off-by: Nathaniel Shead <nathanieloshead@gmail.com>
Reviewed-by: Jason Merrill <jason@redhat.com>

c++/modules: Fix linkage checks for exported using-decls

This fixes some inconsistencies with what kinds of linkage various
entities are assumed to have.  This also fixes handling of exported
using-decls binding to GM entities and type aliases to better align with
the standard's requirements.

gcc/cp/ChangeLog:

* name-lookup.cc (check_can_export_using_decl): Handle internal
linkage GM entities (but ignore in header units); use linkage
of entity ultimately referred to by aliases.

gcc/testsuite/ChangeLog:

* g++.dg/modules/using-10.C: Add tests for no-linkage, fix
expected linkage of aliases.
* g++.dg/modules/using-12.C: Likewise.
* g++.dg/modules/using-27.C: New test.
* g++.dg/modules/using-28_a.C: New test.
* g++.dg/modules/using-28_b.C: New test.
* g++.dg/modules/using-29.H: New test.

Signed-off-by: Nathaniel Shead <nathanieloshead@gmail.com>
Reviewed-by: Jason Merrill <jason@redhat.com>

c++/modules: Use decl_linkage in maybe_record_mergeable_decl

This avoids any possible inconsistencies (current or future) about
whether a declaration is internal or not.

gcc/cp/ChangeLog:

* name-lookup.cc (maybe_record_mergeable_decl): Use decl_linkage
instead of ad-hoc checks.

Signed-off-by: Nathaniel Shead <nathanieloshead@gmail.com>
Reviewed-by: Jason Merrill <jason@redhat.com>

c++: Update decl_linkage for C++11

Currently modules code uses a variety of ad-hoc methods to attempt to
determine whether an entity has internal linkage, which leads to
inconsistencies and some correctness issues as different edge cases are
neglected.  While investigating this I discovered 'decl_linkage', but it
doesn't seem to have been updated to account for the C++11 clarification
that all entities declared in an anonymous namespace are internal.

I'm not convinced that even in C++98 it was intended that e.g. types in
anonymous namespaces should be external, but some tests in the testsuite
rely on this, so for compatibility I restricted those modifications to
C++11 and later.

This should have relatively minimal impact as not much seems to actually
rely on decl_linkage, but does change the mangling of symbols in
anonymous namespaces slightly.  Previously, we had

  namespace {
    int x;  // mangled as '_ZN12_GLOBAL__N_11xE'
    static int y;  // mangled as '_ZN12_GLOBAL__N_1L1yE'
  }

but with this patch the x is now mangled like y (with the extra 'L').
For contrast, Clang currently mangles neither x nor y with the 'L'.
Since this only affects internal-linkage entities I don't believe this
should break ABI in any observable fashion.

gcc/cp/ChangeLog:

* name-lookup.cc (do_namespace_alias): Propagate TREE_PUBLIC for
namespace aliases.
* tree.cc (decl_linkage): Update rules for C++11.

gcc/testsuite/ChangeLog:

* g++.dg/modules/mod-sym-4.C: Update test to account for
non-static internal-linkage variables new mangling.

Signed-off-by: Nathaniel Shead <nathanieloshead@gmail.com>
Reviewed-by: Jason Merrill <jason@redhat.com>

testsuite/gfortran.dg/open_errors_2.f90: Remove now-redundant file deletion

Now that fort.N files are removed by the testsuite
framework, remove this single "manual" file deletion.
(Also, it should have been "remote_file target delete",
since it's the target that creates the file, not the build
framework, which might matter to some setups.)

* gfortran.dg/open_errors_2.f90: Remove now-redundant file deletion.

Daily bump.

libstdc++: Add missing 'inline' to always_inline function

This fixes a -Wattributes warning for the COW std::string which was
previously suppressed due to being in a system header.

libstdc++-v3/ChangeLog:

* include/bits/cow_string.h (__resize_for_overwrite): Add
inline keyword to function with always_inline attribute.

libgcc, libstdc++: Make declarations no longer TU-local [PR115126]

In C++20, modules streaming check for exposures of TU-local entities.
In general exposing internal linkage functions in a header is liable to
cause ODR violations in C++, and this is now detected in a module
context.

This patch goes through and removes 'static' from many declarations
exposed through libstdc++ to prevent code like the following from
failing:

  export module M;
  extern "C++" {
    #include <bits/stdc++.h>
  }

Since gthreads is used from C as well, we need to choose whether to use
'inline' or 'static inline' depending on whether we're compiling for C
or C++ (since the semantics of 'inline' are different between the
languages).  Additionally we need to remove static global variables, so
we migrate these to function-local statics to avoid the ODR issues.

There doesn't seem to be a good workaround for weakrefs, so I've left
them as-is and will work around it in the modules streaming code to
consider them as not TU-local.

The same issue occurs in the objective-C specific parts of gthreads, but
I'm not familiar with the surrounding context and we don't currently
test modules with Objective C++ anyway so I've left it as-is.

PR libstdc++/115126

libgcc/ChangeLog:

* gthr-posix.h (__GTHREAD_ALWAYS_INLINE): New macro.
(__GTHREAD_INLINE): New macro.
(__gthread_active): Convert from variable to (hidden) function.
(__gthread_active_p): Mark as __GTHREAD_INLINE instead of
static; make visibility("hidden") when it has a static local
variable.
(__gthread_trigger): Mark as __GTHREAD_INLINE instead of static.
(__gthread_create): Likewise.
(__gthread_join): Likewise.
(__gthread_detach): Likewise.
(__gthread_equal): Likewise.
(__gthread_self): Likewise.
(__gthread_yield): Likewise.
(__gthread_once): Likewise.
(__gthread_key_create): Likewise.
(__gthread_key_delete): Likewise.
(__gthread_getspecific): Likewise.
(__gthread_setspecific): Likewise.
(__gthread_mutex_init_function): Likewise.
(__gthread_mutex_destroy): Likewise.
(__gthread_mutex_lock): Likewise.
(__gthread_mutex_trylock): Likewise.
(__gthread_mutex_timedlock): Likewise.
(__gthread_mutex_unlock): Likewise.
(__gthread_recursive_mutex_init_function): Likewise.
(__gthread_recursive_mutex_lock): Likewise.
(__gthread_recursive_mutex_trylock): Likewise.
(__gthread_recursive_mutex_timedlock): Likewise.
(__gthread_recursive_mutex_unlock): Likewise.
(__gthread_recursive_mutex_destroy): Likewise.
(__gthread_cond_init_function): Likewise.
(__gthread_cond_broadcast): Likewise.
(__gthread_cond_signal): Likewise.
(__gthread_cond_wait): Likewise.
(__gthread_cond_timedwait): Likewise.
(__gthread_cond_wait_recursive): Likewise.
(__gthread_cond_destroy): Likewise.
(__gthread_rwlock_rdlock): Likewise.
(__gthread_rwlock_tryrdlock): Likewise.
(__gthread_rwlock_wrlock): Likewise.
(__gthread_rwlock_trywrlock): Likewise.
(__gthread_rwlock_unlock): Likewise.
* gthr-single.h: (__GTHREAD_ALWAYS_INLINE): New macro.
(__GTHREAD_INLINE): New macro.
(__gthread_active_p): Mark as __GTHREAD_INLINE instead of static.
(__gthread_once): Likewise.
(__gthread_key_create): Likewise.
(__gthread_key_delete): Likewise.
(__gthread_getspecific): Likewise.
(__gthread_setspecific): Likewise.
(__gthread_mutex_destroy): Likewise.
(__gthread_mutex_lock): Likewise.
(__gthread_mutex_trylock): Likewise.
(__gthread_mutex_unlock): Likewise.
(__gthread_recursive_mutex_lock): Likewise.
(__gthread_recursive_mutex_trylock): Likewise.
(__gthread_recursive_mutex_unlock): Likewise.
(__gthread_recursive_mutex_destroy): Likewise.

libstdc++-v3/ChangeLog:

* include/bits/shared_ptr.h (std::__is_shared_ptr): Remove
unnecessary 'static'.
* include/bits/unique_ptr.h (std::__is_unique_ptr): Likewise.
* include/std/future (std::__create_task_state): Likewise.
* include/std/shared_mutex (_GLIBCXX_GTRHW): Likewise.
(__glibcxx_rwlock_init): Likewise.
(__glibcxx_rwlock_timedrdlock): Likewise.
(__glibcxx_rwlock_timedwrlock): Likewise.
(__glibcxx_rwlock_rdlock): Likewise.
(__glibcxx_rwlock_tryrdlock): Likewise.
(__glibcxx_rwlock_wrlock): Likewise.
(__glibcxx_rwlock_trywrlock): Likewise.
(__glibcxx_rwlock_unlock): Likewise.
(__glibcxx_rwlock_destroy): Likewise.
(__glibcxx_rwlock_init): Likewise.
* include/pstl/algorithm_impl.h
(__pstl::__internal::__set_algo_cut_off): Mark inline.
* include/pstl/unseq_backend_simd.h
(__pstl::__unseq_backend::__lane_size): Mark inline.

Signed-off-by: Nathaniel Shead <nathanieloshead@gmail.com>
Reviewed-by: Jonathan Wakely <jwakely@redhat.com>
Reviewed-by: Jakub Jelinek <jakub@redhat.com>

c++: tweak for -Wrange-loop-construct [PR116731]

This PR reports that the warning would be better off using a check
for trivially constructible rather than trivially copyable.

LLVM accepted a similar fix:
https://github.com/llvm/llvm-project/issues/47355

PR c++/116731

gcc/cp/ChangeLog:

* parser.cc (warn_for_range_copy): Check if TYPE is trivially
constructible, not copyable.

gcc/testsuite/ChangeLog:

* g++.dg/warn/Wrange-loop-construct3.C: New test.

Reviewed-by: Jason Merrill <jason@redhat.com>