Don't vectorise single-iteration epilogues

With a later patch I saw a case in which we peeled a single iteration
for gaps but didn't need to peel further iterations to make up a full
vector.  We then tried to vectorise the single-iteration epilogue.

2019-11-06  Richard Sandiford  <richard.sandiford@arm.com>

gcc/
* tree-vect-loop.c (vect_analyze_loop): Only try to vectorize
the epilogue if there are peeled iterations for it to handle.

From-SVN: r277886

[ARC] Don't split ior/mov predicated insns.

Do not split long immediate constants for predicated instructions.

gcc/
xxxx-xx-xx  Claudiu Zissulescu  <claziss@synopsys.com>

* config/arc/arc.c (arc_split_ior): Add asserts.
(arc_split_mov_const): Likewise.
(arc_check_ior_const): Do not match known short immediate values.
* config/arc/arc.md (movsi): Don't split predicated instructions.
(iorsi): Likewise.

testsuite/
xxxx-xx-xx  Claudiu Zissulescu  <claziss@synopsys.com>
Sahahb Vahedi  <shahab@synopsys.com>
Cupertino Miranda  <cmiranda@synopsys.com>

* gcc.target/arc/or-cnst-size2.c: New test.

Co-Authored-By: Sahahb Vahedi <shahab@synopsys.com>
From-SVN: r277885

[ARC] Update mea option documentation

Update -mea option documentation.

gcc/
xxxx-xx-xx  Claudiu Zissulescu  <claziss@synopsys.com>

* config/arc/arc.opt (mea): Update help string.
* doc/invoke.texi(ARC): Update mea option info.

From-SVN: r277884

[ARC] Cleanup sign/zero extend patterns

Clean up sign/zero extend patterns.

gcc/
xxxx-xx-xx  Claudiu Zissulescu  <claziss@synopsys.com>

* config/arc/arc.md (zero_extendqihi2_i): Cleanup pattern.
(zero_extendqisi2_ac): Likewise.
(zero_extendhisi2_i): Likewise.
(extendqihi2_i): Likewise.
(extendqisi2_ac): Likewise.
(extendhisi2_i): Likewise.

From-SVN: r277883

tree-vect-loop.c (vectorizable_reduction): Remember reduction PHI.

2019-11-06  Richard Biener  <rguenther@suse.de>

* tree-vect-loop.c (vectorizable_reduction): Remember reduction
PHI.  Use STMT_VINFO_REDUC_IDX to skip the reduction operand.
Simplify single_defuse_cycle condition.

From-SVN: r277882

Use scan-tree-dump instead of scan-tree-dump-times for some vect tests

With later patches, we're able to vectorise the epilogues of these tests
on AArch64 and so get two instances of "vectorizing stmts using SLP".
Although it would be possible with a bit of effort to predict when
this happens, it doesn't seem important whether we get 1 vs. 2
occurrences.  All that matters is zero vs. nonzero.

2019-11-06  Richard Sandiford  <richard.sandiford@arm.com>

gcc/testsuite/
* gcc.dg/vect/slp-9.c: Use scan-tree-dump rather than
scan-tree-dump-times.
* gcc.dg/vect/slp-widen-mult-s16.c: Likewise.
* gcc.dg/vect/slp-widen-mult-u8.c: Likewise.

From-SVN: r277881

Check the VF is small enough for an epilogue loop

The number of iterations of an epilogue loop is always smaller than the
VF of the main loop.  vect_analyze_loop_costing was taking this into
account when deciding whether the loop is cheap enough to vectorise,
but that has no effect with the unlimited cost model.  We need to use
a separate check for correctness as well.

This can happen if the sizes returned by autovectorize_vector_sizes
happen to be out of order, e.g. because the target prefers smaller
vectors.  It can also happen with later patches if two vectorisation
attempts happen to end up with the same VF.

2019-11-06  Richard Sandiford  <richard.sandiford@arm.com>

gcc/
* tree-vect-loop.c (vect_analyze_loop_2): When vectorizing an
epilogue loop, make sure that the VF is small enough or that
the epilogue loop can be fully-masked.

From-SVN: r277880

Restructure vect_analyze_loop

Once vect_analyze_loop has found a valid loop_vec_info X, we carry
on searching for alternatives if (1) X doesn't satisfy simdlen or
(2) we want to vectorize the epilogue of X.  I have a patch that
optionally adds a third reason: we want to see if there are cheaper
alternatives to X.

This patch restructures vect_analyze_loop so that it's easier
to add more reasons for continuing.  There's supposed to be no
behavioural change.

If we wanted to, we could allow vectorisation of epilogues once
loop->simdlen has been reached by changing "loop->simdlen" to
"simdlen" in the new vect_epilogues condition.  That should be
a separate change though.

2019-11-06  Richard Sandiford  <richard.sandiford@arm.com>

gcc/
* tree-vect-loop.c (vect_analyze_loop): Break out of the main
loop when we've finished, rather than returning directly from
the loop.  Use a local variable to track whether we're still
searching for the preferred simdlen.  Make vect_epilogues
record whether the next iteration should try to treat the
loop as an epilogue.

From-SVN: r277879

[PATCH] [ARC] Add builtins for identifying floating point support

Currently for hard float we need to check for
 __ARC_FPU_SP__ || __ARC_FPU_DP__ and for soft float inverse of that.
So define single convenience macros for either cases.

gcc/
xxxx-xx-xx  Vineet Gupta  <vgupta@synopsyscom>

* config/arc/arc-c.c (arc_cpu_cpp_builtins): Add
          __arc_hard_float__, __ARC_HARD_FLOAT__,
          __arc_soft_float__, __ARC_SOFT_FLOAT__

From-SVN: r277878

[vect] PR92317: fix skip_epilogue creation for epilogues

gcc/ChangeLog:
2019-11-06  Andre Vieira  <andre.simoesdiasvieira@arm.com>

PR tree-optimization/92317
* tree-vect-loop-manip.c (slpeel_update_phi_nodes_for_guard2): Also
update phi's with constant phi arguments.

gcc/testsuite/ChangeLog:
2019-11-06  Andre Vieira  <andre.simoesdiasvieira@arm.com>

PR tree-optimization/92317
* gcc/testsuite/g++.dg/opt/pr92317.C: New test.

From-SVN: r277877

introduce -fcallgraph-info option

This was first submitted many years ago
https://gcc.gnu.org/ml/gcc-patches/2010-10/msg02468.html

The command line option -fcallgraph-info is added and makes the
compiler generate another output file (xxx.ci) for each compilation
unit (or LTO partitoin), which is a valid VCG file (you can launch
your favorite VCG viewer on it unmodified) and contains the "final"
callgraph of the unit.  "final" is a bit of a misnomer as this is
actually the callgraph at RTL expansion time, but since most
high-level optimizations are done at the Tree level and RTL doesn't
usually fiddle with calls, it's final in almost all cases.  Moreover,
the nodes can be decorated with additional info: -fcallgraph-info=su
adds stack usage info and -fcallgraph-info=da dynamic allocation info.

for  gcc/ChangeLog
From  Eric Botcazou  <ebotcazou@adacore.com>, Alexandre Oliva  <oliva@adacore.com>

* common.opt (-fcallgraph-info[=]): New option.
* doc/invoke.texi (Developer options): Document it.
* opts.c (common_handle_option): Handle it.
* builtins.c (expand_builtin_alloca): Record allocation if
-fcallgraph-info=da.
* calls.c (expand_call): If -fcallgraph-info, record the call.
(emit_library_call_value_1): Likewise.
* flag-types.h (enum callgraph_info_type): New type.
* explow.c: Include stringpool.h.
(set_stack_check_libfunc): Set SET_SYMBOL_REF_DECL on the symbol.
* function.c (allocate_stack_usage_info): New.
(allocate_struct_function): Call it for -fcallgraph-info.
(prepare_function_start): Call it otherwise.
(record_final_call, record_dynamic_alloc): New.
* function.h (struct callinfo_callee): New.
(CALLEE_FROM_CGRAPH_P): New.
(struct callinfo_dalloc): New.
(struct stack_usage): Add callees and dallocs.
(record_final_call, record_dynamic_alloc): Declare.
* gimplify.c (gimplify_decl_expr): Record dynamically-allocated
object if -fcallgraph-info=da.
* optabs-libfuncs.c (build_libfunc_function): Keep SYMBOL_REF_DECL.
* print-tree.h (print_decl_identifier): Declare.
(PRINT_DECL_ORIGIN, PRINT_DECL_NAME, PRINT_DECL_UNIQUE_NAME): New.
* print-tree.c: Include print-tree.h.
(print_decl_identifier): New function.
* toplev.c: Include print-tree.h.
(callgraph_info_file): New global variable.
(callgraph_info_external_printed): Likewise.
(output_stack_usage): Rename to...
(output_stack_usage_1): ... this.  Make it static, add cf
parameter.  If -fcallgraph-info=su, print stack usage to cf.
If -fstack-usage, use print_decl_identifier for
pretty-printing.
(INDIRECT_CALL_NAME): New.
(dump_final_node_vcg_start): New.
(dump_final_callee_vcg, dump_final_node_vcg): New.
(output_stack_usage): New.
(lang_dependent_init): Open and start file if
-fcallgraph-info.  Allocated callgraph_info_external_printed.
(finalize): If callgraph_info_file is not null, finish it,
close it, and release callgraph_info_external_printed.

for  gcc/ada/ChangeLog

* gcc-interface/misc.c (callgraph_info_file): Delete.

Co-Authored-By: Alexandre Oliva <oliva@adacore.com>
From-SVN: r277876

Warn about inconsistent OpenACC nested reduction clauses

OpenACC (cf. OpenACC 2.7, section 2.9.11. "reduction clause";
this was first clarified by OpenACC 2.6) requires that, if a
variable is used in reduction clauses on two nested loops, then
there must be reduction clauses for that variable on all loops
that are nested in between the two loops and all these reduction
clauses must use the same operator.
This commit introduces a check for that property which reports
warnings if it is violated.

2019-11-06  Gergö Barany  <gergo@codesourcery.com>
            Frederik Harwath  <frederik@codesourcery.com>
            Thomas Schwinge  <thomas@codesourcery.com>

gcc/
* omp-low.c (struct omp_context): New fields
local_reduction_clauses, outer_reduction_clauses.
(new_omp_context): Initialize these.
(scan_sharing_clauses): Record reduction clauses on OpenACC constructs.
(scan_omp_for): Check reduction clauses for incorrect nesting.
gcc/testsuite/
* c-c++-common/goacc/nested-reductions-warn.c: New test.
* c-c++-common/goacc/nested-reductions.c: New test.
* gfortran.dg/goacc/nested-reductions-warn.f90: New test.
* gfortran.dg/goacc/nested-reductions.f90: New test.
libgomp/
* testsuite/libgomp.oacc-c-c++-common/par-loop-comb-reduction-1.c:
Add expected warnings about missing reduction clauses.
* testsuite/libgomp.oacc-c-c++-common/par-loop-comb-reduction-2.c:
Likewise.
* testsuite/libgomp.oacc-c-c++-common/par-loop-comb-reduction-3.c:
Likewise.
* testsuite/libgomp.oacc-c-c++-common/par-loop-comb-reduction-4.c:
Likewise.

Reviewed-by: Thomas Schwinge <thomas@codesourcery.com>
From-SVN: r277875

re PR inline-asm/92352 (ICE in force_constant_size)

PR inline-asm/92352
* gimplify.c (gimplify_asm_expr): Reject VLA in output or input
operands with non-memory constraints.

* c-c++-common/pr92352.c: New test.

From-SVN: r277873

PR92090: Fix part of testcase failures by r276469

-finline-functions is enabled by default for O2 since r276469, update the
test cases with -fno-inline-functions.
c11-atomic-exec-5.c stills hit ICE of LRA on BE systems in PR92090.
This commit is NOT a fix for the bug and so it must NOT be closed.

gcc/testsuite/ChangeLog:

2019-11-06  Xiong Hu Luo  <luoxhu@linux.ibm.com>

PR92090
* gcc.target/powerpc/pr72804.c: Add -fno-inline-functions --param
max-inline-insns-single-O2=200.
* gcc.target/powerpc/pr79439-1.c: Add -fno-inline-functions.
* gcc.target/powerpc/vsx-builtin-7.c: Likewise.

From-SVN: r277872

PR tree-optimization/92373 - ICE in -Warray-bounds on access to member array in an initialized char buffer

gcc/testsuite/ChangeLog:

PR tree-optimization/92373
* gcc.dg/Warray-bounds-55.c: New test.
* gcc.dg/Wzero-length-array-bounds-2.c: New test.

gcc/ChangeLog:

PR tree-optimization/92373
* tree.c (component_ref_size): Only consider initializers of objects
of matching struct types.
Return null for instances of interior zero-length arrays.

From-SVN: r277871

Daily bump.

From-SVN: r277870

doc: Insn splitting by combine

The combine pass is perfectly happy if a splitter splits to just one
instruction (instead of two).

* doc/md.texi (Insn Splitting): Fix combiner documentation.

From-SVN: r277866

Implement C++20 operator<=>.

There are three major pieces to this support: scalar operator<=>,
synthesis of comparison operators, and rewritten/reversed overload
resolution (e.g. a < b becomes 0 > b <=> a).

Unlike other defaulted functions, where we use synthesized_method_walk to
semi-simulate what the definition of the function will be like, this patch
determines the characteristics of a comparison operator by trying to define
it.

My handling of non-dependent rewritten operators in templates can still use
some work: build_min_non_dep_op_overload can't understand the rewrites and
crashes, so I'm avoiding it for now by clearing *overload.  This means we'll
do name lookup again at instantiation time, which can incorrectly mean a
different result.  I'll poke at this more in stage 3.

I'm leaving out a fourth section ("strong structural equality") even though
I've implemented it, because it seems likely to change radically tomorrow.

Thanks to Tim van Deurzen and Jakub for implementing lexing of the <=>
operator, and Jonathan for the initial <compare> header.

gcc/cp/
* cp-tree.h (struct lang_decl_fn): Add maybe_deleted bitfield.
(DECL_MAYBE_DELETED): New.
(enum special_function_kind): Add sfk_comparison.
(LOOKUP_REWRITTEN, LOOKUP_REVERSED): New.
* call.c (struct z_candidate): Add rewritten and reversed methods.
(add_builtin_candidate): Handle SPACESHIP_EXPR.
(add_builtin_candidates): Likewise.
(add_candidates): Don't add a reversed candidate if the parms are
the same.
(add_operator_candidates): Split out from build_new_op_1.  Handle
rewritten and reversed candidates.
(add_candidate): Swap conversions of reversed candidate.
(build_new_op_1): Swap them back.  Build a second operation for
rewritten candidates.
(extract_call_expr): Handle rewritten calls.
(same_fn_or_template): New.
(joust): Handle rewritten and reversed candidates.
* class.c (add_implicitly_declared_members): Add implicit op==.
(classtype_has_op, classtype_has_defaulted_op): New.
* constexpr.c (cxx_eval_binary_expression): Handle SPACESHIP_EXPR.
(cxx_eval_constant_expression, potential_constant_expression_1):
Likewise.
* cp-gimplify.c (genericize_spaceship): New.
(cp_genericize_r): Use it.
* cp-objcp-common.c (cp_common_init_ts): Handle SPACESHIP_EXPR.
* decl.c (finish_function): Handle deleted function.
* decl2.c (grokfield): SET_DECL_FRIEND_CONTEXT on defaulted friend.
(mark_used): Check DECL_MAYBE_DELETED.  Remove assumption that
defaulted functions are non-static members.
* error.c (dump_expr): Handle SPACESHIP_EXPR.
* method.c (type_has_trivial_fn): False for sfk_comparison.
(enum comp_cat_tag, struct comp_cat_info_t): New types.
(comp_cat_cache): New array variable.
(lookup_comparison_result, lookup_comparison_category)
(is_cat, cat_tag_for, spaceship_comp_cat)
(spaceship_type, genericize_spaceship)
(common_comparison_type, early_check_defaulted_comparison)
(comp_info, build_comparison_op): New.
(synthesize_method): Handle sfk_comparison.  Handle deleted.
(get_defaulted_eh_spec, maybe_explain_implicit_delete)
(explain_implicit_non_constexpr, implicitly_declare_fn)
(defaulted_late_check, defaultable_fn_check): Handle sfk_comparison.
* name-lookup.c (get_std_name_hint): Add comparison categories.
* tree.c (special_function_p): Add sfk_comparison.
* typeck.c (cp_build_binary_op): Handle SPACESHIP_EXPR.

2019-11-05  Tim van Deurzen  <tim@kompiler.org>

Add new tree code for the spaceship operator.
gcc/cp/
* cp-tree.def: Add new tree code.
* operators.def: New binary operator.
* parser.c: Add new token and tree code.
libcpp/
* cpplib.h: Add spaceship operator for C++.
* lex.c: Implement conditional lexing of spaceship operator for C++20.

2019-11-05  Jonathan Wakely  <jwakely@redhat.com>

libstdc++-v3/
* libsupc++/compare: New header.
* libsupc++/Makefile.am (std_HEADERS): Add compare.
* include/std/version: Define __cpp_lib_three_way_comparison.
* include/std/functional: #include <compare>.

From-SVN: r277865

Fix conversions for built-in operator overloading candidates.

While working on C++20 operator<=>, I noticed that build_new_op_1 was doing
too much conversion when a built-in candidate was selected; the standard
says it should only perform user-defined conversions, and then leave the
normal operator semantics to handle any standard conversions.  This is
important for operator<=> because a comparison of two different unscoped
enums is ill-formed; if we promote the enums to int here, cp_build_binary_op
never gets to see the original operand types, so we can't give the error.

I'm also disabling -Wmaybe-uninitialized for expmed.c to avoid the bootstrap
failure from the last time I applied this patch.

* call.c (build_new_op_1): Don't apply any standard conversions to
the operands of a built-in operator.  Don't suppress conversions in
cp_build_unary_op.
* typeck.c (cp_build_unary_op): Do integral promotions for enums.

PR tree-optimization/91825
* expmed.c: Reduce -Wmaybe-uninitialized to warning.

From-SVN: r277864

Use vec instead of raw array for built-in candidates.

My operator<=> patch wants to split up build_new_op_1, which makes using a
tree array as well as the vec inconvenient.  build_new_op_1 already has a
vec, and build_conditional_expr_1 can release its vec right away, so this
doesn't increase garbage at all.

* call.c (build_builtin_candidate): Take args in a vec.
(add_builtin_candidate, add_builtin_candidates): Likewise.
(build_conditional_expr_1, build_new_op_1): Adjust.

From-SVN: r277863

Various small C++ changes.

Wrappers for lookup_qualified_name and build_x_binary_op to make calling
them more convenient in places, and a function named contextual_conv_bool
for places that want contextual conversion to bool.

I noticed that we weren't showing the declaration location when we complain
about a call to a non-constexpr function where a constant expression is
required.

If maybe_instantiate_noexcept doesn't actually instantiate, there's no
reason for it to mess with clones.

* constexpr.c (explain_invalid_constexpr_fn): Show location of fn.

* pt.c (maybe_instantiate_noexcept): Only update clones if we
instantiated.

* typeck.c (contextual_conv_bool): New.

* name-lookup.c (lookup_qualified_name): Add wrapper overload taking
C string rather than identifier.
* parser.c (cp_parser_userdef_numeric_literal): Use it.
* rtti.c (emit_support_tinfos): Use it.
* cp-tree.h (ovl_op_identifier): Change to inline functions.
(build_x_binary_op): Add wrapper with fewer parms.

From-SVN: r277862

Allow libcalls for complex memcpy when optimizing for size.

The RISC-V backend wants to use a libcall when optimizing for size if
more than 6 instructions are needed.  Emit_move_complex asks for no
libcalls.  This case requires 8 insns for rv64 and 16 insns for rv32,
so we get fallback code that emits a loop.  Commit_one_edge_insertion
doesn't allow code inserted for a phi node on an edge to end with a
branch, and so this triggers an assertion.  This problem goes away if
we allow libcalls when optimizing for size, which gives the code the
RISC-V backend wants, and avoids triggering the assert.

gcc/
PR middle-end/92263
* expr.c (emit_move_complex): Only use BLOCK_OP_NO_LIBCALL when
optimize_insn_for_speed_p is true.

gcc/testsuite/
PR middle-end/92263
* gcc.dg/pr92263.c: New.

From-SVN: r277861

Catch missed uses of function with unsatisfied constraints.

While looking at CA378 I noticed that we weren't properly diagnosing two of
the three erroneous lines in the example.

* decl2.c (mark_used): Diagnose use of a function with unsatisfied
constraints here.
* typeck.c (cp_build_function_call_vec): Not here.

From-SVN: r277860

Make -fconcepts-ts imply -fconcepts.

* c-opts.c (c_common_post_options): -fconcepts-ts implies
-fconcepts.

From-SVN: r277859

PR middle-end/92333 - missing variable name referencing VLA in warnings

PR middle-end/92333 - missing variable name referencing VLA in warnings
PR middle-end/82608 - missing -Warray-bounds on an out-of-bounds VLA index

gcc/testsuite/ChangeLog:

PR middle-end/92333
PR middle-end/82608
* gcc.dg/Warray-bounds-51.c: New test.

gcc/ChangeLog:

PR middle-end/92333
PR middle-end/82608
* tree-vrp.c (vrp_prop::check_array_ref): Handle VLAs with constant
size.
* tree-ssa-ccp.c (fold_builtin_alloca_with_align): Use a meaninful
name and location for a temporary variable.

From-SVN: r277854

[PR c++/92370] ICE with VC marker

https://gcc.gnu.org/ml/gcc-patches/2019-11/msg00323.html
cp/
PR c++/92370
* parser.c (cp_parser_error_1): Check EOF and UNKNOWN_LOCATION
when skipping over version control marker.

testsuite/
PR c++/92370
* g++.dg/pr92370.C: New.

From-SVN: r277853

Fix indentation inconsistencies introduced by previous patch.

From-SVN: r277852

PR middle-end/92341 - missing -Warray-bounds indexing past the end of a compound literal

PR middle-end/92341 - missing -Warray-bounds indexing past the end of a compound literal
PR middle-end/82612 - missing -Warray-bounds on a non-zero offset from the address of a non-array object

gcc/testsuite/ChangeLog:

PR middle-end/92341
PR middle-end/82612
* g++.dg/warn/Warray-bounds-4.C: Adjust text of expected warning.
* gcc.dg/Warray-bounds-53.c: New test.
* gcc.dg/Warray-bounds-54.c: New test.

gcc/ChangeLog:

PR middle-end/92341
PR middle-end/82612
* tree-sra.c (get_access_for_expr): Fail for out-of-bounds offsets.
* tree-vrp.c (vrp_prop::check_array_ref): Correct index and text
of message printed in a warning for empty arrays.
(vrp_prop::check_mem_ref): Also handle function parameters and
empty arrays.

From-SVN: r277851

re PR tree-optimization/92371 (ICE in info_for_reduction, at tree-vect-loop.c:4106)

2019-11-05  Richard Biener  <rguenther@suse.de>

PR tree-optimization/92371
* tree-vect-loop.c (vectorizable_reduction): Set STMT_VINFO_REDUC_DEF
on the original stmt of live stmts in the chain.
(vectorizable_live_operation): Look at the original stmt when
checking STMT_VINFO_REDUC_DEF.

* gcc.dg/torture/pr92371.c: New testcase.

From-SVN: r277850

Fix <version> header for freestanding

* include/std/version [!_GLIBCXX_HOSTED]: Do not define feature test
macros for features that are only present in hosted builds.

From-SVN: r277849

Remove incorrect comment

The negative concept is required for subsumption to work, it's not a
bug.

* include/bits/iterator_concepts.h (__iter_without_nested_types):
Remove incorrect comment.

From-SVN: r277848

The base class for ranges is currently value_range_base, which is rather long and cumbersome.

The base class for ranges is currently value_range_base, which is
rather long and cumbersome.  It also occurs more often than the derived
class of value_range.  To avoid confusion, and save typing, this
patch does a global rename from value_range to value_range_equiv,
and from value_range_base to value_range.

This way, the base class is simply value_range, and the derived
class is value_range_equiv which explicitly states what it does.

From-SVN: r277847

[mid-end] Fix declared type of personality functions

`build_personality_function` generates a declaration for a personality
function.  The type it declares for these functions doesn't match the
type of the actual personality functions that are defined by the C++
unwinding ABI.

This doesn't cause any crashes since the compiler never generates a call
to these decl's, and hence the type of the function is never used.
Nonetheless, for the sake of consistency and readability we update the
type of this declaration.

gcc/ChangeLog:

2019-11-05  Matthew Malcomson  <matthew.malcomson@arm.com>

* expr.c (build_personality_function): Fix generated type to
match actual personality functions.

From-SVN: r277846

[aarch64] Allocate space for err_str in aarch64_handle_attr_branch_protection

-fsanitize=hwaddress found a one-byte overwrite when running the
testsuite here.  aarch64_handle_attr_branch_protection allocates
`strlen(str)` bytes for an error string, which is populated by
`strcpy(..., str)` in the case where the branch protection string is
completely invalid.

Not tested -- I don't want to re-build and it seems obvious.

gcc/ChangeLog:

2019-11-05  Matthew Malcomson  <matthew.malcomson@arm.com>

* config/aarch64/aarch64.c (aarch64_handle_attr_cpu): Allocate
enough bytes for the NULL character.

From-SVN: r277845

Update LOCAL_PATCHES.

From-SVN: r277839

Update scanned patterns in a test-case.

2019-11-05  Martin Liska  <mliska@suse.cz>

* c-c++-common/ubsan/ptr-overflow-2.c: Update based on changed
run-time reporting format.

From-SVN: r277838

Set print_summary for UBSAN.

2019-11-05  Martin Liska  <mliska@suse.cz>

* ubsan/ubsan_flags.cpp (InitializeFlags): Trunk decided to print
summary for all sanitizers, but we want to have UBSAN without it.

From-SVN: r277837

Reapply all revisions mentioned in LOCAL_PATCHES.

2019-11-05  Martin Liska  <mliska@suse.cz>

* asan/asan_globals.cpp (CheckODRViolationViaIndicator): Reapply from
LOCAL_PATCHES.
(CheckODRViolationViaPoisoning): Likewise.
(RegisterGlobal): Likewise.
* asan/asan_interceptors.h (ASAN_INTERCEPT___CXA_RETHROW_PRIMARY_EXCEPTION): Likewise.
(defined): Likewise.
* asan/asan_mapping.h: Likewise.
* sanitizer_common/sanitizer_linux_libcdep.cpp (defined): Likewise.
* sanitizer_common/sanitizer_mac.cpp (defined): Likewise.
* sanitizer_common/sanitizer_platform_limits_linux.cpp (defined): Likewise.
* sanitizer_common/sanitizer_platform_limits_posix.h: Likewise.
* sanitizer_common/sanitizer_stacktrace.cpp (GetCanonicFrame): Likewise.
* tsan/tsan_rtl_ppc64.S: Likewise.
* ubsan/ubsan_handlers.cpp (__ubsan::__ubsan_handle_cfi_bad_icall): Likewise.
(__ubsan::__ubsan_handle_cfi_bad_icall_abort): Likewise.
* ubsan/ubsan_handlers.h (struct CFIBadIcallData): Likewise.
(struct CFICheckFailData): Likewise.
(RECOVERABLE): Likewise.
* ubsan/ubsan_platform.h: Likewise.

From-SVN: r277836

Update Makefile.am.

2019-11-05  Martin Liska  <mliska@suse.cz>

* tsan/Makefile.am: Rename tsan_interceptors.cpp to
tsan_interceptors_posix.
* tsan/Makefile.in: Regenerate.

From-SVN: r277835

Libsanitizer: merge from trunk with merge.sh.

2019-11-05  Martin Liska  <mliska@suse.cz>

* all source files: Merge from upstream r375507.

From-SVN: r277834

IBM Z: gen-vect-26/28: Vectorizing without peeling is ok for Z

These tests check if loop peeling has been applied to avoid
having to vectorize unaligned loops.  On Z we do not have any
alignment requirements for vectorization so we also don't need want
the loop peeling here.

2019-11-05  Andreas Krebbel  <krebbel@linux.ibm.com>

* gcc.dg/tree-ssa/gen-vect-26.c: Disable loop peeling check for
IBM Z.
* gcc.dg/tree-ssa/gen-vect-28.c: Likewise.

From-SVN: r277833

re PR target/92280 (gcc.target/i386/pr83008.c FAILs)

2019-11-05  Richard Biener  <rguenther@suse.de>

PR tree-optimization/92280
* match.pd (BIT_FIELD_REF of CTOR): Unless the original CTOR
had a single use do not create a new CTOR.
* tree-ssa-forwprop.c (simplify_bitfield_ref): Do not re-fold
BIT_FIELD_REF of a CTOR via GENERIC.

From-SVN: r277832

IBM Z: gen-vect-11/32: Set min-vect-loop-bound param back to default

In the Z backend we still set min-vect-loop-bound to 2 to work around
corner cases where awkward epilogue code gets generated in the
vectorizer.  This has a particular bad impact when vectorizing loops
with a low iteration count.  Due to this we do not vectorize the loop
in gen-vect-11/32 - what actually is a pity.

The patch sets min-vect-loop-bound back to the default value of 0 in
order to enable vectorization.

2019-11-05  Andreas Krebbel  <krebbel@linux.ibm.com>

* gcc.dg/tree-ssa/gen-vect-11.c: Add --param min-vect-loop-bound=0
for IBM Z.
* gcc.dg/tree-ssa/gen-vect-23.c: Likewise.

From-SVN: r277831

IBM Z: Fix testsuite useable_hw check

This fixes various issues with the useable_hw check in s390.exp.  The
check is supposed to verify whether a testcase can be run on the
current hardware.

- the test never returned true for -m31 because vzero is not available
  in ESA mode and -m31 defaults to -mesa
- the missing v0 clobber on the vzero instruction made the check fail
  if the stack pointer got saved in f0
- the lcbb instruction used for checking whether we are on a z13
  also requires vx.  Replace it with an instruction from the generic
  instruction set extensions.
- no support for z14 and z15 so far

gcc/testsuite/ChangeLog:

2019-11-05  Andreas Krebbel  <krebbel@linux.ibm.com>

* gcc.target/s390/s390.exp
(check_effective_target_s390_useable_hw): Add inline asm for z14
and z15. Replace instruction for z13 with lochiz. Add register
clobbers. Check also for __zarch__ when doing the __VX__ test.

From-SVN: r277830

IBM Z: Use tree_fits_uhwi_p in vector_alignment hook

This fixes an ICE in gcc.dg/attr-vector_size.c testcase.

gcc/ChangeLog:

2019-11-05  Andreas Krebbel  <krebbel@linux.ibm.com>

* config/s390/s390.c (s390_vector_alignment): Check if the value
fits into uhwi before using it.

From-SVN: r277829

[ARM,testsuite] Skip tests incompatible with -mpure-code.

All these tests fail when using -mpure-code:
* some force A or R profile
* some use Neon
* some use -fpic/-fPIC
all of which are not supported by this option.

2019-11-05  Christophe Lyon  <christophe.lyon@linaro.org>

* gcc.target/arm/attr-crypto.c: Skip if -mpure-code is used.
* gcc.target/arm/attr-neon-fp16.c: Likewise.
* gcc.target/arm/combine-cmp-shift.c: Likewise.
* gcc.target/arm/data-rel-1.c: Likewise.
* gcc.target/arm/data-rel-2.c: Likewise.
* gcc.target/arm/data-rel-3.c: Likewise.
* gcc.target/arm/ftest-armv7a-arm.c: Likewise.
* gcc.target/arm/ftest-armv7a-thumb.c: Likewise.
* gcc.target/arm/ftest-armv7r-arm.c: Likewise.
* gcc.target/arm/ftest-armv7r-thumb.c: Likewise.
* gcc.target/arm/ftest-armv7ve-arm.c: Likewise.
* gcc.target/arm/ftest-armv7ve-thumb.c: Likewise.
* gcc.target/arm/ftest-armv8a-arm.c: Likewise.
* gcc.target/arm/ftest-armv8a-thumb.c: Likewise.
* gcc.target/arm/lceil-vcvt_1.c: Likewise.
* gcc.target/arm/lfloor-vcvt_1.c: Likewise.
* gcc.target/arm/lround-vcvt_1.c: Likewise.
* gcc.target/arm/memset-inline-10.c: Likewise.
* gcc.target/arm/mod_2.c: Likewise.
* gcc.target/arm/mod_256.c: Likewise.
* gcc.target/arm/pr40657-1.c: Likewise.
* gcc.target/arm/pr44788.c: Likewise.
* gcc.target/arm/pr50305.c: Likewise.
* gcc.target/arm/pr51835.c: Likewise.
* gcc.target/arm/pr51915.c: Likewise.
* gcc.target/arm/pr52006.c: Likewise.
* gcc.target/arm/pr53187.c: Likewise.
* gcc.target/arm/pr58784.c: Likewise.
* gcc.target/arm/pr59575.c: Likewise.
* gcc.target/arm/pr59923.c: Likewise.
* gcc.target/arm/pr60650-2.c: Likewise.
* gcc.target/arm/pr60657.c: Likewise.
* gcc.target/arm/pr60663.c: Likewise.
* gcc.target/arm/pr67439_1.c: Likewise.
* gcc.target/arm/pr68620.c: Likewise.
* gcc.target/arm/pr7676.c: Likewise.
* gcc.target/arm/pr79239.c: Likewise.
* gcc.target/arm/pr81863.c: Likewise.
* gcc.target/arm/pragma_attribute.c: Likewise.
* gcc.target/arm/pragma_cpp_fma.c: Likewise.
* gcc.target/arm/require-pic-register-loc.c: Likewise.
* gcc.target/arm/thumb-ltu.c: Likewise.
* gcc.target/arm/thumb1-pic-high-reg.c: Likewise.
* gcc.target/arm/thumb1-pic-single-base.c: Likewise.
* gcc.target/arm/tlscall.c: Likewise.
* gcc.target/arm/unsigned-float.c: Likewise.
* gcc.target/arm/vrinta-ce.c: Likewise.

From-SVN: r277828

Use more ggc_delete.

2019-11-05  Martin Liska  <mliska@suse.cz>

* symbol-summary.h: Use ggc_delete.

From-SVN: r277827

[Aarch64] Fix vec_perm cost for thunderx2t99

Here is the one-liner that fixes the incorrect
vec_perm cost for thunderx2t99 chip.
With the patch applied 526.blender of CPU2017
gets ~5% improvement with no measurable changes
for other benchmarks.

2019-11-05  Anton Youdkevitch  <anton.youdkevitch@bell-sw.com>

* config/aarch64/aarch64.c (thunderx2t99_vector_cost):
Change vec_perm field to 10.

From-SVN: r277826

* doc/install.texi: Further fix syntax for html generation

From-SVN: r277824

Update coding style in symbol-summary.h.

2019-11-05  Martin Liska  <mliska@suse.cz>

* symbol-summary.h: Rename allocator to m_allocator and
add comment.

From-SVN: r277823

re PR tree-optimization/92324 (ICE in expand_direct_optab_fn, at internal-fn.c:2890)

2019-11-05  Richard Biener  <rguenther@suse.de>

PR tree-optimization/92324
* tree-vect-loop.c (check_reduction_path): For MIN/MAX require
all signed or unsigned operations.

* gcc.dg/vect/pr92324-3.c: New testcase.

From-SVN: r277822

hsa-brig.c: Include alloc-pool.h

* hsa-brig.c: Include alloc-pool.h
* hsa-dump.c: Likewise.
* hsa-gen.c: Likewise.
* hse-regalloc.c: Likewise.
* ipa-hsa.c: Likewise.
* ipa-predicate.c: Likewise.
* ipa-reference.c: Likewise.
* ipa-sra.c: Likewise.
* omp-expand.c: Likewise.
* omp-general.c: Likewise.
* omp-low.c: Likewise.
* sumbol-summary.h (function_summary_base): Add allocator.
(function_summary<T *>::function_summary): Update construction.
(fast_function_summary<T *, V>::fast_function_summary): Likewise.
(call_summary_base): Add allcator.
(call_summary<T *>::call_summary): Update construction.
(fast_call_summary<T *, V>::fast_call_summary): Likewise.

From-SVN: r277821

re PR tree-optimization/91945 (ICE: tree check: expected integer_cst, have var_decl in get_len, at tree.h:5837 since r274997)

PR tree-optimization/91945
* builtins.c (compute_objsize): For ARRAY_REF, only multiply off
by tpsize if it is both non-NULL and INTEGER_CST, otherwise punt.
Formatting fix.

* gfortran.dg/pr91945.f90: New test.

From-SVN: r277820

re PR c++/92343 ([[likely]]/[[unlikely]] prevent method from being a constant expression)

PR c++/92343
* constexpr.c (potential_constant_expression_1): Return true rather
than false for PREDICT_EXPR.

* g++.dg/cpp2a/attr-likely6.C: New test.

From-SVN: r277819

Move vrp_set_zero_nonzero_bits from tree-vrp.c into vr-values.c, and make it use wi_set_zero_nonzero_bits.

Move vrp_set_zero_nonzero_bits from tree-vrp.c into vr-values.c, and
make it use wi_set_zero_nonzero_bits.  Remove the now redundant
wide_int_range_set_zero_nonzero_bits.

From-SVN: r277818

Really add gcc/config/gcn/t-omp-device

        gcc/
        * config/gcn/t-omp-device: New.
        Should have been part of Rev. 277797.

From-SVN: r277817

Remove FIELD_DECL leftover.

2019-11-05  Martin Liska  <mliska@suse.cz>

PR c++/92339
* fold-const.c (operand_compare::hash_operand): Remove
FIELD_DECL handling.
2019-11-05  Martin Liska  <mliska@suse.cz>

PR c++/92339
* g++.dg/pr92339.C: New test.

From-SVN: r277816

crontab: Disable snapshots from gcc-7-branch.

2019-10-05  Richard Biener  <rguenther@suse.de>

        * crontab: Disable snapshots from gcc-7-branch.

From-SVN: r277815

decl.c (omp_declare_variant_finalize_one): Call declare_simd_adjust_this not just on the context...

* decl.c (omp_declare_variant_finalize_one): Call
declare_simd_adjust_this not just on the context, but also on the
variant-id expression for methods.  Don't call
cp_get_callee_fndecl_nofold, call cp_get_callee and only if it is
safe cp_get_fndecl_from_callee.  Don't try to print as %qD
NULL in diagnostics.
* pt.c (tsubst_attribute): Handle "omp declare variant base"
attribute.
(tsubst_function_decl): Call omp_declare_variant_finalize
if there are any "omp declare variant base" attributes left.

* g++.dg/gomp/declare-variant-7.C: New test.
* g++.dg/gomp/declare-variant-8.C: New test.

From-SVN: r277814

Move vrp_bitmap_equal_p above its only use and make it static.

From-SVN: r277813

Use value_range_base::equal_p in value_range_base::operator== so we can handle
symbolics without dying.

From-SVN: r277812

Daily bump.

From-SVN: r277810

PR c++/91979 - mangling nullptr expression

2019-11-04  Kamlesh Kumar  <kamleshbhalui@gmail.com>

gcc/cp
        * cp/mangle.c (write_template_arg_literal): Handle nullptr
        mangling.
gcc
* common.opt (-fabi-version): Document =14.
* doc/invoke.texi (C++ Dialect Options): Likewise.
gcc/c-family
        * c-opts.c (c_common_post_options): Update
        latest_abi_version.
libiberty
        * cp-demangle.c (d_expr_primary): Handle
        nullptr demangling.
        * testsuite/demangle-expected: Added test.

From-SVN: r277801

Do not special case pointers in value_range_base::set.

From-SVN: r277799

* typeck.c (check_return_expr): Avoid redundant error.

From-SVN: r277798

Add gcc/config/gcn/t-omp-device for OpenMP declare variant kind/arch/isa

        gcc/
        * config/gcn/gcn.c (gcn_omp_device_kind_arch_isa): New function.
        (TARGET_OMP_DEVICE_KIND_ARCH_ISA): Redefine to
        gcn_omp_device_kind_arch_isa.
        * config/gcn/t-omp-device: New file.
        * configure.ac: Support gcn for omp_device_property.
        * configure: Regenerate.

From-SVN: r277797

Remove handle_pointers argument from all the vrp_val*{min,max} functions.

Remove handle_pointers argument from all the vrp_val*{min,max} functions.  Always
assume pointers should be handled.

From-SVN: r277796

ipa-reference.c (init_function_info): Initialize info->global.statics_read.

* ipa-reference.c (init_function_info): Initialize
info->global.statics_read.

From-SVN: r277795

Use the value_range_base constructors in value_range_base::invert to
make sure we build canonically correct ranges.

From-SVN: r277794

[D] Remove unchecked to_constant in VECTOR_TYPE handling

The SVE port now tries to register variable-length VECTOR_TYPEs
at start-up, so it's no longer possible to use the asserting
to_constant on the number of vector elements.  This patch punts
on variable element counts instead, just like we do for other
things that the frontend doesn't recognise.

The brace indentation matches the surrounding style.

2019-11-04  Richard Sandiford  <richard.sandiford@arm.com>

gcc/d/
* d-builtins.cc (build_frontend_type): Cope with variable
TYPE_VECTOR_SUBPARTS.

From-SVN: r277793

Remove unused range_int_cst_singleton_p.

From-SVN: r277792

Handle VR_UNDEFINED in value_range_base::normalize_addresses().

From-SVN: r277791

Implement debugging functions for assert_info's.

From-SVN: r277790

ipa-inline-transform.c: Include ipa-utils.h

* ipa-inline-transform.c: Include ipa-utils.h
(inline_call): Set thunk_expansion flag.
* ipa-utils.h (thunk_expansion): Declare.
* ipa-devirt.c (thunk_expansion): New global var.
(devirt_node_removal_hook): Do not invalidate cache while
doing thunk expansion.

From-SVN: r277789

SLP: Initialize variable to fix bootstrap after r277784.

This initializes the rstmt variable with NULL and adds an assert to
check that a value has been given by one of the if cases before use.

This fixes the bootstrap failure due to -Werror.

Committed under the gcc obvious rule.

gcc/ChangeLog:

* tree-vect-slp.c (vectorize_slp_instance_root_stmt): Initialize rstmt.

From-SVN: r277788

re PR testsuite/92302 (gcc.target/sparc/sparc-ret-3.c FAILs)

PR testsuite/92302
* gcc.target/sparc/sparc-ret-3.c: Accept more registers in address.

From-SVN: r277787

PR tree-optimization/92349 - ICE in -Warray-bounds of a VLA member

gcc/testsuite/ChangeLog:

PR tree-optimization/92349
* gcc.dg/Warray-bounds-50.c: New test.

gcc/ChangeLog:

PR tree-optimization/92349
* tree-vrp.c (vrp_prop::check_array_ref): Avoid assuming struct
memebers have constant sizes.

From-SVN: r277786

[vect] Clean up orig_loop_vinfo from vect_analyze_loop

gcc/ChangeLog:
2019-11-04  Andre Vieira  <andre.simoesdiasvieira@arm.com>

* tree-vect-loop.c (vect_analyze_loop): Remove orig_loop_vinfo
parameter.
* tree-vectorizer.h (vect_analyze_loop): Update declaration.
* tree-vectorizer.c (try_vectorize_loop_1): Update calls to
vect_analyze_loop.

From-SVN: r277785

[SLP] SLP vectorization: vectorize vector constructors

gcc/ChangeLog:
2019-11-04  Joel Hutton  <Joel.Hutton@arm.com>

* expr.c (store_constructor): Modify to handle single element vectors.
* tree-vect-slp.c (vect_analyze_slp_instance): Add case for vector
constructors.
(vect_slp_check_for_constructors): New function.
(vect_slp_analyze_bb_1): Call new function to check for vector
constructors.
(vectorize_slp_instance_root_stmt): New function.
(vect_schedule_slp): Call new function to vectorize root stmt of vector
constructors.
* tree-vectorizer.h (SLP_INSTANCE_ROOT_STMT): New field.

gcc/testsuite/ChangeLog:
2019-11-04  Joel Hutton  <Joel.Hutton@arm.com>

* gcc.dg/vect/bb-slp-40.c: New test.
* gcc.dg/vect/bb-slp-41.c: New test.

From-SVN: r277784

re PR tree-optimization/92345 (ICE in vec<_stmt_vec_info*, va_heap, vl_embed>::space (vect_get_and_check_slp_defs))

2019-11-04  Richard Biener  <rguenther@suse.de>

PR tree-optimization/92345
* tree-vect-loop.c (vect_is_simple_reduction): Return whether
we produced a reduction chain.
(vect_analyze_scalar_cycles_1): Do not add reduction chains to
LOOP_VINFO_REDUCTIONS.

* gcc.dg/torture/pr92345.c: New testcase.

From-SVN: r277782

cgraphclones.c (cgraph_node::create_version_clone): Do not duplicate summaries.

* cgraphclones.c (cgraph_node::create_version_clone): Do not
duplicate summaries.
* ipa-fnsummary.c (ipa_fn_summary_alloc): Allocate size summary
first.
(ipa_fn_summary_t::duplicate): Use get instead of get_create to
access call summaries.
(dump_ipa_call_summary): Be ready for missing edge summaries.
(analyze_function_body): Use get instead of get_create to access
edge summary.
(estimate_calls_size_and_time): Do not access summaries of
inlined edges; sanity check they are missing.
(ipa_call_context::estimate_size_and_time): Use get instead
of get_create to access node summary.
(inline_update_callee_summaries): Do not update depth of
inlined edge.
(ipa_merge_fn_summary_after_inlining): Remove inline edge from
growth caches.
(ipa_merge_fn_summary_after_inlining): Use get instead
of get_create.
* ipa-fnsummary.h (ipa_remove_from_growth_caches): Declare.
* ipa-inline-analyssi.c (edge_growth_cache): Turn to
fast summary.
(initialize_growth_caches): Update.
(do_estimate_edge_time): Remove redundant copy of context.
(ipa_remove_from_growth_caches): New function.
* ipa-inline.c (flatten_function): Update overall summary
only when optimizing.
(inline_to_all_callers): Update overall summary of function
inlined to.
* ipa-inline.h (edge_growth_cache): Turn to fast summary.
* symbol-summary.h (call_summary_base): Set m_initialize_when_cloning
to false.

From-SVN: r277780

system.h: Include malloc.h if INCLUDE_MALLOC_H and HAVE_MALLINFO.

2019-11-04  Richard Biener  <rguenther@suse.de>

* system.h: Include malloc.h if INCLUDE_MALLOC_H and HAVE_MALLINFO.
* ggc-common.c: Remove inclusion of malloc.h, define INCLUDE_MALLOC_H.

From-SVN: r277779

* ggc-common.c: Include system.h before malloc.h.

From-SVN: r277778

pass --enable-obsolete down to gcc/configure for auto-build.h

Configuring GCC for obsolete targets works as long as build = host.
When it isn't, --enable-obsolete is not passed down to the additional
build configure started by gcc/configure, used to generate
auto-build.h.  The build configure fails and we end up without a
auto-build.h, but the host configure proceeds, so we only get a fatal
failure much later, when make realizes auto-build.h is not there and
there's no rule to create it.

This patch gets the host configure to fail when the build configure
does, leaving the temporary build configure dir behind for
investigation.  It also arranges for --eanble-obsolete to be passed
down to the build configure.

Alas, the latter triggered a warning in the build configure because
--enable-obsolete is not a recognized configure option.  That's not
reported in the host configure because of the
--disable-option-checking passed by the top-level configure, so I
arranged for that to be passed down to the build configure as well.

Finally, since my initial suspicion when investigating this failure
was that auto-build.h had been removed after configuration and there
was no rule to rebuild it, I'm adding rules to gcc/Makefile to get it
created or updated as needed.  Since it is configure that creates it,
as run by e.g. config.status --recheck, and config.status is created
after auto-build.h, I've made config.status depend on auto-build.h,
and added a dummy rule to create auto-build.h.  This would normally
not be enough to create a header when needed, but since Makefile
depends on config.status, and make first updates Makefile, it ends up
working, as long as nothing else that Makefile depends on requires
auto-build.h but not config.status.  The config.status dependency and
the auto-build.h rule are only enabled in the cases in which
auto-build.h is actually used, namely when build != host.

for  gcc/ChangeLog

* configure.ac: Pass --enable-obsolete=* and
--enable-option-checking=* down to build configure, and fail
if it fails.  AC_SUBST HAVE_AUTO_BUILD.
* configure: Rebuild.
* Makefile.in [HAVE_AUTO_BUILD] (auto-build.h): New rule.
[HAVE_AUTO_BUILD] (config.status): Depend on auto-build.h.

From-SVN: r277777

libgomp - remove dg-run if dg-options O2 is used

* testsuite/libgomp.fortran/pr66199-1.f90: Remove
'dg-do run' (implies torture test) as 'dg-options "O2"' is used.
* testsuite/libgomp.fortran/pr66199-2.f90: Ditto.
* testsuite/libgomp.fortran/taskloop2.f90: Ditto.
* testsuite/libgomp.fortran/taskloop3.f90: Ditto.
* testsuite/libgomp.fortran/taskloop4.f90: Ditto.

From-SVN: r277776

libgcc: Fix incorrect use of USE_TM_CLONE_REGISTRY

2019-11-04  Jozef Lawrynowicz  <jozef.l@mittosystems.com>

* crtstuff.c: Define USE_TM_CLONE_REGISTRY to 0 if it's undefined and
the target output object format is not ELF.
s/defined(USE_TM_CLONE_REGISTRY)/USE_TM_CLONE_REGISTRY.

From-SVN: r277775

MSP430: Add -mtiny-printf option

2019-11-04  Jozef Lawrynowicz  <jozef.l@mittosystems.com>

gcc/
* config.in: Regenerate.
* config/msp430/msp430.c (msp430_option_override): Emit an error if
-mtiny-printf is used without GCC being configured with
--enable-newlib-nano-formatted-io.
* config/msp430/msp430.h (LINK_SPEC): Pass
"--wrap puts --wrap printf" when -mtiny-printf is used.
* config/msp430/msp430.opt: Document -mtiny-printf.
* configure: Regenerate.
* configure.ac: Enable --enable-newlib-nano-formatted-io flag.
Define HAVE_NEWLIB_NANO_FORMATTED_IO if
--enable-newlib-nano-formatted-io is passed.
* doc/invoke.texi: Document -mtiny-printf.

gcc/testsuite/
* gcc.target/msp430/tiny-printf.c: New test.

From-SVN: r277774

Regenerate gcc/configure

From-SVN: r277773

MSP430: Add new methods of finding external MCU data file

MCU data file can now be specified with an environment variable or installed
into a toolchain subdirectory.

2019-11-04  Jozef Lawrynowicz  <jozef.l@mittosystems.com>

gcc/
* config/msp430/driver-msp430.c
(msp430_get_linker_devices_include_path): New spec function.
* config/msp430/msp430-devices.c (msp430_dirname): New function.
(extract_devices_dir_from_exec_prefix): New function.
(extract_devices_dir_from_collect_gcc): New function.
(msp430_check_env_var_for_devices): New function.
(msp430_check_path_for_devices): Use xstrdup instead of ASTRDUP.
(parse_devices_csv): Call msp430_check_env_var_for_devices if
devices.csv was not found using other methods.
* config/msp430/msp430-devices.h (msp430_check_env_var_for_devices):
New prototype.
(msp430_dirname): Likewise.
* config/msp430/msp430.c (msp430_register_pre_includes): New function.
* config/msp430/msp430.h (EXTRA_SPEC_FUNCTIONS): Add
msp430_get_linker_devices_include_path.
(TARGET_EXTRA_PRE_INCLUDES): Define.
* doc/invoke.texi: Document new ways of searching for support files.

gcc/testsuite/
* gcc.target/msp430/devices/csv-using-env-var.c: New test.
* gcc.target/msp430/devices/csv-using-installed.c: New test.
* gcc.target/msp430/devices/csv-using-option.c: New test.
* gcc.target/msp430/devices/devices-main.c: New test source file.
* gcc.target/msp430/devices/msp430-devices.h: New test.
* gcc.target/msp430/msp430.exp (msp430_device_permutations_runtest):
Add special cases for csv-using* tests.
Define TESTING_HARD_DATA when running tests that use hard-coded device
data.
(get_installed_device_data_path): New.
(msp430_hide_installed_devices_data): New.
(msp430_restore_installed_devices_data): New.
(msp430_test_installed_device_data): New.
(msp430_install_device_data): New.

From-SVN: r277772

re PR bootstrap/92301 (Wrong vectorizer code since  r277566)

2019-11-04  Richard Biener  <rguenther@suse.de>

PR tree-optimization/92301
* tree-vect-stmts.c (process_use): Force reduction PHI defs live
as required by epilogue generation

* gcc.dg/pr92301.c: New testcase.

From-SVN: r277771

libgomp/testsuite - use unique numbers with Fortran's 'stop'

        PR fortran/92305
        * testsuite/libgomp.fortran/allocatable2.f90: Use
        unique numbers with 'stop'.
        * testsuite/libgomp.fortran/use_device_addr-1.f90: Ditto.
        * testsuite/libgomp.fortran/use_device_addr-2.f90: Ditto.
        * testsuite/libgomp.fortran/use_device_ptr-1.f90: Ditto.
        * testsuite/libgomp.oacc-fortran/lib-15.f90: Ditto.
        * testsuite/libgomp.oacc-fortran/pset-1.f90: Ditto.

From-SVN: r277769

Fix hash_operand for fields of a CONSTRUCTOR.

2019-11-04  Martin Liska  <mliska@suse.cz>

PR ipa/92304
* fold-const.c (operand_compare::hash_operand): Fix field
hashing of CONSTRUCTOR.

From-SVN: r277768

Come up with ggc_delete.

2019-11-04  Martin Liska  <mliska@suse.cz>

* ggc.h (ggc_delete): New function.
* ipa-fnsummary.c (ipa_free_fn_summary): Use it.
* ipa-prop.c (ipa_free_all_edge_args): Likewise.
(ipa_free_all_node_params): Likewise.
* ipa-sra.c (ipa_sra_analysis): Likewise.

From-SVN: r277767

re PR fortran/92113 (r276673 causes segfault in gfortran.dg/pr51434.f90)

2019-11-04  Thomas Koenig  <tkoenig@gcc.gnu.org>

    PR fortran/92113
    * ChangeLog: Fix PR number

From-SVN: r277766

Daily bump.

From-SVN: r277764

re PR fortran/92113 (r276673 causes segfault in gfortran.dg/pr51434.f90)

2019-11-03  Thomas Koenig  <tkoenig@gcc.gnu.org>

    PR fortran/92133
    * trans-decl.c (gfc_get_symbol_decl): If __def_init actually
    contains a value, put it into  the read-only section.

From-SVN: r277760

Improve effectivity of ipa_polymorphi_context cache.

        * ipa-fnsummary.c (set_cond_stmt_execution_predicate,
set_switch_stmt_execution_predicate, compute_bb_predicates,
will_be_nonconstant_expr_predicate,
phi_result_unknown_predicate,
analyze_function_body): Pass arround params summary.
(ipa_call_context::duplicate_from): New comment;
only duplicate useful values.
(ipa_call_context::equal_to): Only compare useful values.
(remap_edge_summaries): Pass params_summary.
(remap_hint_predicate): Likewise.
(ipa_merge_fn_summary_after_inlining): Likewise.
(inline_read_section): Initialize params summary used flags.
* ipa-predicate.c (predicate::remap_after_inlining): Pass
around param_summary.
(add_condition): Initialized used params summary flags.
* ipa-predicate.h (inline_param_summary::equals_to): Make const.
(inline_param_summary::useless_p): New predicate.
(remap_after_inlining, add_condition): Update prototype
* ipa-prop.c (ipa_populate_param_decls): Watch overflow in
move_cost.
(ipa_note_param_call): Add parameter POLYMORPHIC; update params
summaries.
(ipa_analyze_indirect_call_uses): Update use of ipa_note_param_call.
(ipa_analyze_virtual_call_uses): Likewise.
(update_indirect_edges_after_inlining): Update param summaries.
(ipa_print_node_params): Print used flags.
(ipa_read_indirect_edge_info): Update param summareis.
* ipa-prop.h (ipa_param_descriptor): Add
used_by_ipa_predicates, used_by_indirect_call
and used_by_polymorphic_call.
(ipa_set_param_used_by_ipa_predicates,
ipa_set_param_used_by_indirect_call,
ipa_set_param_used_by_polymorphic_call,
ipa_is_param_used_by_ipa_predicates,
ipa_is_param_used_by_indirect_call,
ipa_is_param_used_by_polymorphic_call): New inline functions.

From-SVN: r277759

PR c++/88565 - enhance -Warray-bounds for C++ trailing class member arrays

gcc/testsuite/ChangeLog:

* g++.dg/warn/Warray-bounds-9.C: New test.

From-SVN: r277758

ipa-fnsummary.c (ipa_call_context::duplicate_from): New member function.

* ipa-fnsummary.c (ipa_call_context::duplicate_from): New
member function.
(ipa_call_context::release): Add ALL parameter.
(ipa_call_context::equal_to): New member function.
* ipa-fnsummary.h (ipa_call_context): Add empty constructor;
duplicate_form, release, equal_to and exists_p member functoins.
* ipa-inline-analysis.c (node_context_cache_entry): New
class.
(node_context_summary): Likewise.
(node_context_cache, node_context_cache_hit, node_context_cache_miss,
node_context_clear): New static vars.
(initialize_growth_caches): New function.
(free_growth_caches): Also delete node_context_cache; output stats.
(do_estimate_edge_time): Cache contexts.
(reset_node_cache): New function.
* ipa-inline.c (reset_edge_caches): Reset also node cache.
(inline_small_functions): Initialize growth caches.
* ipa-inline.h (reset_node_cache, initialize_growth_caches):
Declare.
* ipa-predicate.h (inline_param_summary::equal_to): New.
* ipa-prop.c (ipa_agg_jf_item::equal_to): New.
* ipa-prop.h (ipa_agg_jf_item): Declare equal_to member function.
(ipa_agg_jump_function): Implement equal_to member function.

From-SVN: r277757

ipa-fnsummary.c (inline_read_section): Set vector size ahead of time.

* ipa-fnsummary.c (inline_read_section): Set vector size
ahead of time.

From-SVN: r277756