[Ada] Add Package_Body helper routine to be used in GNATprove

gcc/ada/

* sem_aux.adb, sem_aux.ads (Package_Body): Moved from GNATprove.
* sem_elab.adb (Spec_And_Body_From_Entity): Refine type of parameter.

[Ada] Minor doc improvement

gcc/ada/

* doc/gnat_ugn/platform_specific_information.rst: Improve doc
on permission and containers.
* gnat_ugn.texi: Regenerate.

[Ada] Fix infinite loop in compilation of illegal code

gcc/ada/

* atree.adb (Rewrite): Fix parent node of shared aspects.
* atree.ads (Rewrite): Add ??? comment on incorrect
documentation.
* einfo-utils.adb (Known_Esize): Fix logic.
* sem_ch13.adb (Alignment_Check_For_Size_Change,
Analyze_Attribute_Definition_Clause): Protect against unset
Size.

[Ada] More precise analysis of function renamings in GNATprove

gcc/ada/

* freeze.adb (Build_Renamed_Body): Special case for GNATprove.
* sem_ch6.adb (Analyze_Expression_Function): Remove useless test
for a node to come from source, which becomes harmful otherwise.

[Ada] Removal of technical debt

gcc/ada/

* ali.adb, ali.ads (Scan_ALI): Remove use of deprecated
parameter Ignore_ED, and all specification for Lower in call to
Get_File_Name.
* ali-util.adb (Read_Withed_ALIs): Modify call to Scan_ALI.
* clean.adb (Clean_Executables): Likewise.
* gnatbind.adb (Add_Artificial_ALI_File, Executable section):
Likewise.
* gnatlink.adb (Executable section): Likewise.
* gnatls.adb (Executable section): Likewise.
* make.adb (Check, Wait_For_Available_Slot): Likewise.
* aspects.ads: Add Aspect_No_Controlled_Parts to
Nonoverridable_Aspect_Id
* opt.ads: Remove function pointers used as a workaround for
ASIS.
* osint-c.adb (Executable section): Remove setting of function
pointer workarounds needed for ASIS.
* osint.adb (Read_Default_Search_Dirs): Correct behavior to
detect EOL characters.
* par_sco.adb (Output_Header): Remove comment regarding use of
First_Sloc.
(Traverse_Sync_Definition): Renamed to
Traverse_Protected_Or_Task_Definition.
* pprint.adb (Interal_List_Name): Add description about purpose,
and refactor conditional statement.
(Prepend): Removed.
* repinfo.adb (List_Rep_Info, Write_Info_Line): Remove use of
subprogram pointer.
* scng.adb (Scan): Remove CODEFIX question, and minor comment
change.
* sem_attr.adb (Analyze_Image_Attribute): Remove special
processing for 'Img.
* sem_ch6.adb (Check_Untagged_Equality): Add RM reference.
(FCE): Add comment describing behavior.
(Is_Non_Overriding_Operation): Minor comment formatting change.
* sem_type.adb (Is_Actual_Subprogram): Add comment about
Comes_From_Source test.
(Matching_Types): Describe non-matching cases.
* sem_util.adb (Is_Confirming): Add stub case for
No_Controlled_Parts.

[Ada] Fix access to predicated parent in Itype

gcc/ada/

* sem_ch13.adb (Build_Predicate_Functions): Access
Predicated_Parent only on subtypes.

[Ada] Allow more cases of import with Relaxed_RM_Semantics

gcc/ada/

* sem_prag.adb (Process_Import_Or_Interface): Relax error when
Relaxed_RM_Semantics.

[Ada] Improve performance for case-insensitive regular expressions

gcc/ada/

* libgnat/s-regpat.adb (Match): Handle the case where Self.First
is not NUL (so we know the first character we are looking for),
but case-insensitive matching has
been specified.
(Optimize): In the case of an EXACTF Op, set Self.First as is
done in the EXACT case, except with the addition of a call to
Lower_Case.

[Ada] Remove System.Img_Enum_New unit

gcc/ada/

* libgnat/s-imenne.ads, libgnat/s-imenne.adb: Delete.
* gcc-interface/Make-lang.in (GNAT_ADA_OBJS): Remove s-imenne.o.
(GNATBIND_OBJS): Likewise.

[Ada] Fix obsolete comments/name referring to girder discriminants

gcc/ada/

* einfo.ads: Fix comments.
* exp_aggr.adb: Fix variable name.
* exp_util.adb: Fix comments.
* sem_ch13.adb: Fix comments.
* sem_ch3.adb: Fix comments and variable name.

[Ada] VxWorks inconsistent use of return type (Int_Unlock)

gcc/ada/

* libgnarl/s-osinte__vxworks.ads: Make procedure vice function.
* libgnarl/s-vxwext.ads: Likewise.
* libgnarl/s-vxwext__kernel-smp.adb: Likewise.
* libgnarl/s-vxwext__kernel.adb: Likewise.
* libgnarl/s-vxwext__kernel.ads: Likewise.
* libgnarl/s-vxwext__rtp-smp.adb: Likewise.
* libgnarl/s-vxwext__rtp.adb: Likewise.
* libgnarl/s-vxwext__rtp.ads: Likewise.
* libgnarl/s-taprop__vxworks.adb (Stop_All_Tasks): Call
Int_Unlock as a procedure.

[Ada] VxWorks inconsistent use of return type (vx_freq_t)

gcc/ada/

* libgnarl/s-osinte__vxworks.ads (SVE): New package renaming
(vx_freq_t): New subtype.
(sysClkRateGet): Return vx_freq_t.
* libgnarl/s-vxwext.ads (vx_freq_t): New type.
* libgnarl/s-vxwext__kernel.ads: Likewise.
* libgnarl/s-vxwext__rtp.ads: Likewise.

[Ada] Replace use of 'Image with use of Error_Msg_Uint

gcc/ada/

* sem_case.adb (Composite_Case_Ops): Replace 'Image with
Error_Msg_Uint.

[Ada] Generate temporary for if-expression with -fpreserve-control-flow

gcc/ada/

* exp_ch4.adb (Expand_N_If_Expression): Generate an intermediate
temporary when the expression is a condition in an outer decision
and control-flow optimizations are suppressed.

[Ada] Add -gnatX support for casing on array values

gcc/ada/

* exp_ch5.adb (Expand_General_Case_Statement.Pattern_Match): Add
new function Indexed_Element to handle array element
comparisons. Handle case choices that are array aggregates,
string literals, or names denoting constants.
* sem_case.adb (Composite_Case_Ops.Array_Case_Ops): New package
providing utilities needed for casing on arrays.
(Composite_Case_Ops.Choice_Analysis): If necessary, include
array length as a "component" (like a discriminant) when
traversing components. We do not (yet) partition choice analysis
to deal with unequal length choices separately. Instead, we
embed everything in the minimum-dimensionality Cartesian product
space needed to handle all choices properly; this is determined
by the length of the longest choice pattern.
(Composite_Case_Ops.Choice_Analysis.Traverse_Discrete_Parts):
Include length as a "component" in the traversal if necessary.
(Composite_Case_Ops.Choice_Analysis.Parse_Choice.Traverse_Choice):
Add support for case choices that are string literals or names
denoting constants.
(Composite_Case_Ops.Choice_Analysis): Include length as a
"component" in the analysis if necessary.
(Check_Choices.Check_Case_Pattern_Choices.Ops.Value_Sets.Value_Index_Count):
Improve error message when capacity exceeded.
* doc/gnat_rm/implementation_defined_pragmas.rst: Update
documentation to reflect current implementation status.
* gnat_rm.texi: Regenerate.

[Ada] Fix imprecise wording for error on scalar storage order

gcc/ada/

* freeze.adb (Check_Component_Storage_Order): Give a specific error
message for non-byte-aligned component in the packed case.  Replace
"composite" with "record" in both cases.

[Ada] Make Ada.Task_Initialization compatible with No_Elaboration_Code_All

gcc/ada/

* libgnarl/a-tasini.ads, libgnarl/a-tasini.adb: Make compatible
with No_Elaboration_Code_All.
* libgnarl/s-taskin.ads, libgnarl/s-tassta.adb: Adjust
accordingly.

[Ada] Change message format on missing return

gcc/ada/

* sem_ch6.adb (Check_Returns): Change message on missing return.

[Ada] Mark gnatfind and gnatxref obsolete

gcc/ada/

* gnatfind.adb, gnatxref.adb: Mark these tools as obsolete
before removing them completely.

Set bound/cmp/control for until wrap loop.

In patch r12-3136, niter->control, niter->bound and niter->cmp are
derived from number_of_iterations_lt.  While for 'until wrap condition',
the calculation in number_of_iterations_lt is not align the requirements
on the define of them and requirements in determine_exit_conditions.

This patch calculate niter->control, niter->bound and niter->cmp in
number_of_iterations_until_wrap.

gcc/ChangeLog:

2021-09-22  Jiufu Guo  <guojiufu@linux.ibm.com>

PR tree-optimization/102087
* tree-ssa-loop-niter.c (number_of_iterations_until_wrap):
Update bound/cmp/control for niter.

gcc/testsuite/ChangeLog:

2021-09-22  Jiufu Guo  <guojiufu@linux.ibm.com>

* gcc.dg/pr102087.c: New test.
PR tree-optimization/102087

Check for BB before calling register_outgoing_edges.

We may be asked to fold an artificial statement not in the CFG.  Since
there are no outgoing edges from those, avoid calling
register_outgoing_edges.

Tested on x86-64 Linux.

gcc/ChangeLog:

* gimple-range-fold.cc (fold_using_range::range_of_range_op):
Move check for non-empty BB here.
(fur_source::register_outgoing_edges): ...from here.

path solver: Use range_on_path_entry instead of looking at equivalences.

Cycling through equivalences to improve a range is nowhere near as
efficient as asking the ranger what the range on entry is.

Testing on a hybrid VRP threader, shows that this improves our VRP
threading benefit from 14.5% to 18.5% and our overall jump threads from
0.85% to 1.28%.

Tested on x86-64 Linux.

gcc/ChangeLog:

* gimple-range-path.cc (path_range_query::internal_range_of_expr):
Remove call to improve_range_with_equivs.
(path_range_query::improve_range_with_equivs): Remove
* gimple-range-path.h: Remove improve_range_with_equivs.

AVX512FP16: Add permutation and mask blend intrinsics.

gcc/ChangeLog:

* config/i386/avx512fp16intrin.h:
(_mm512_mask_blend_ph): New intrinsic.
(_mm512_permutex2var_ph): Ditto.
(_mm512_permutexvar_ph): Ditto.
* config/i386/avx512fp16vlintrin.h:
(_mm256_mask_blend_ph): New intrinsic.
(_mm256_permutex2var_ph): Ditto.
(_mm256_permutexvar_ph): Ditto.
(_mm_mask_blend_ph): Ditto.
(_mm_permutex2var_ph): Ditto.
(_mm_permutexvar_ph): Ditto.

gcc/testsuite/ChangeLog:

* gcc.target/i386/avx512fp16-14.c: New test.

AVX512FP16: Add complex conjugation intrinsic instructions.

gcc/ChangeLog:

* config/i386/avx512fp16intrin.h: Add new intrinsics.
(_mm512_conj_pch): New intrinsic.
(_mm512_mask_conj_pch): Ditto.
(_mm512_maskz_conj_pch): Ditto.
* config/i386/avx512fp16vlintrin.h: Add new intrinsics.
(_mm256_conj_pch): New intrinsic.
(_mm256_mask_conj_pch): Ditto.
(_mm256_maskz_conj_pch): Ditto.
(_mm_conj_pch): Ditto.
(_mm_mask_conj_pch): Ditto.
(_mm_maskz_conj_pch): Ditto.

gcc/testsuite/ChangeLog:

* gcc.target/i386/avx512fp16-conjugation-1.c: New test.
* gcc.target/i386/avx512fp16vl-conjugation-1.c: New test.

AVX512FP16: Add reduce operators(add/mul/min/max).

gcc/ChangeLog:

* config/i386/avx512fp16intrin.h (_MM512_REDUCE_OP): New macro
(_mm512_reduce_add_ph): New intrinsic.
(_mm512_reduce_mul_ph): Ditto.
(_mm512_reduce_min_ph): Ditto.
(_mm512_reduce_max_ph): Ditto.
* config/i386/avx512fp16vlintrin.h
(_MM256_REDUCE_OP/_MM_REDUCE_OP): New macro.
(_mm256_reduce_add_ph): New intrinsic.
(_mm256_reduce_mul_ph): Ditto.
(_mm256_reduce_min_ph): Ditto.
(_mm256_reduce_max_ph): Ditto.
(_mm_reduce_add_ph): Ditto.
(_mm_reduce_mul_ph): Ditto.
(_mm_reduce_min_ph): Ditto.
(_mm_reduce_max_ph): Ditto.

gcc/testsuite/ChangeLog:

* gcc.target/i386/avx512fp16-reduce-op-1.c: New test.
* gcc.target/i386/avx512fp16vl-reduce-op-1.c: Ditto.

AVX512FP16: Support load/store/abs intrinsics.

gcc/ChangeLog:

* config/i386/avx512fp16intrin.h (__m512h_u, __m256h_u,
__m128h_u): New typedef.
(_mm512_load_ph): New intrinsic.
(_mm256_load_ph): Ditto.
(_mm_load_ph): Ditto.
(_mm512_loadu_ph): Ditto.
(_mm256_loadu_ph): Ditto.
(_mm_loadu_ph): Ditto.
(_mm512_store_ph): Ditto.
(_mm256_store_ph): Ditto.
(_mm_store_ph): Ditto.
(_mm512_storeu_ph): Ditto.
(_mm256_storeu_ph): Ditto.
(_mm_storeu_ph): Ditto.
(_mm512_abs_ph): Ditto.
* config/i386/avx512fp16vlintrin.h
(_mm_abs_ph): Ditto.
(_mm256_abs_ph): Ditto.

gcc/testsuite/ChangeLog:

* gcc.target/i386/avx512fp16-13.c: New test.

Small tweak to comments in range_cannot_be_superflat

gcc/ada/
* gcc-interface/decl.c (range_cannot_be_superflat): Tweak comments.

IBM Z: TPF: Add cc clobber to profiling expanders

The code sequence emitted uses CC internally.

gcc/ChangeLog:

* config/s390/tpf.md (prologue_tpf, epilogue_tpf): Add cc clobber.

Fortran: Add gfc_simple_for_loop aux function

Function to generate a simple loop (to be used internally).
Callers will be added in follow-up commits.

gcc/fortran/
* trans-expr.c (gfc_simple_for_loop): New.
* trans.h (gfc_simple_for_loop): New prototype.

obstack.h __PTR_ALIGN vs. ubsan

Current ubsan complains on every use of __PTR_ALIGN (when ptrdiff_t is
as large as a pointer), due to making calculations relative to a NULL
pointer.  This patch avoids the problem by extracting out and
simplifying __BPTR_ALIGN for the usual case.  I've continued to use
ptrdiff_t here, where it might be better to throw away __BPTR_ALIGN
entirely and just assume uintptr_t exists.

* obstack.h (__PTR_ALIGN): Expand and simplify __BPTR_ALIGN
rather than calculating relative to a NULL pointer.

IBM Z: Fix PR102222

Avoid emitting a strict low part move if the insv target actually
affects the whole target reg.

gcc/ChangeLog:

PR target/102222
* config/s390/s390.c (s390_expand_insv): Emit a normal move if it
is actually a full copy of the source operand into the target.
Don't emit a strict low part move if source and target mode match.

gcc/testsuite/ChangeLog:

* gcc.target/s390/pr102222.c: New test.

openmp: Fix OpenMP expansion of scope with non-fallthrugh body [PR102415]

I've used function for omp single expansion also for omp scope.  That is
mostly ok, but as the testcase shows, there is one important difference.
The omp single expansion always has a fallthru body, because it during
omp lowering expands the body as if wrapped in an if to simulate that
one thread runs the body and others wait (unless nowait) until it completes
and continue.  omp scope is invoked by all threads and so if the body
is non-fallthru, the barrier (unless nowait) at the end will not be reached
by any of the threads.

The following patch fixes that by handling the case where cfg pass optimizes
away the exit bb of it gracefully.

2021-09-22  Jakub Jelinek  <jakub@redhat.com>

PR middle-end/102415
* omp-expand.c (expand_omp_single): If region->exit is NULL,
assert region->entry is GIMPLE_OMP_SCOPE region and return.

* c-c++-common/gomp/scope-3.c: New test.

openmp: Add support for allocator and align modifiers on allocate clauses

As the allocate-2.c testcase shows, this change isn't 100% backwards compatible,
one could have allocate and/or align functions that return an OpenMP allocator
handle and previously it would call those functions and now would use those
names as keywords for the modifiers.  But it allows specify extra alignment
requirements for the allocations.

2021-09-22  Jakub Jelinek  <jakub@redhat.com>

gcc/
* tree.h (OMP_CLAUSE_ALLOCATE_ALIGN): Define.
* tree.c (omp_clause_num_ops): Change number of OMP_CLAUSE_ALLOCATE
arguments from 2 to 3.
* tree-pretty-print.c (dump_omp_clause): Print allocator() around
allocate clause allocator and print align if present.
* omp-low.c (scan_sharing_clauses): Force allocate_map entry even
for omp_default_mem_alloc if align modifier is present.  If align
modifier is present, use TREE_LIST to encode both allocator and
align.
(lower_private_allocate, lower_rec_input_clauses, create_task_copyfn):
Handle align modifier on allocator clause if present.
gcc/c-family/
* c-omp.c (c_omp_split_clauses): Copy over OMP_CLAUSE_ALLOCATE_ALIGN.
gcc/c/
* c-parser.c (c_parser_omp_clause_allocate): Parse allocate clause
modifiers.
gcc/cp/
* parser.c (cp_parser_omp_clause_allocate): Parse allocate clause
modifiers.
* semantics.c (finish_omp_clauses) <OMP_CLAUSE_ALLOCATE>: Perform
semantic analysis of OMP_CLAUSE_ALLOCATE_ALIGN.
* pt.c (tsubst_omp_clauses) <case OMP_CLAUSE_ALLOCATE>: Handle
also OMP_CLAUSE_ALLOCATE_ALIGN.
gcc/testsuite/
* c-c++-common/gomp/allocate-6.c: New test.
* c-c++-common/gomp/allocate-7.c: New test.
* g++.dg/gomp/allocate-4.C: New test.
libgomp/
* testsuite/libgomp.c-c++-common/allocate-2.c: New test.
* testsuite/libgomp.c-c++-common/allocate-3.c: New test.

libgcc, X86: Adjust guard for Mach-O code.

Existing code in the sfp-machine header has been using __MACH__
as a guard for Mach-O, where currently symbols aliases are not
supported.

__MACH__ is not a sufficient guard for this, since the define
is also emitted for HURD, at least.

Fixed by amending the guard to use __APPLE__ instead.

Signed-off-by: Iain Sandoe <iain@sandoe.co.uk>
libgcc/ChangeLog:

* config/i386/sfp-machine.h: Guard Mach-O-specific code
using __APPLE__.

Support 64bit fma/fms/fnma/fnms under avx512vl.

gcc/ChangeLog:

* config/i386/i386.md (define_attr "isa"): Add
fma_or_avx512vl.
(define_attr "enabled"): Correspond fma_or_avx512vl to
TARGET_FMA || TARGET_AVX512VL.
* config/i386/mmx.md (fmav2sf4): Extend to AVX512 fma.
(fmsv2sf4): Ditto.
(fnmav2sf4): Ditto.
(fnmsv2sf4): Ditto.

gcc/testsuite/ChangeLog:

* gcc.target/i386/avx512vl-pr95046.c: New test.

Adjust testcase.

gcc/testsuite/ChangeLog:

* gcc.target/i386/pr92658-avx512f.c: Refine testcase.
* gcc.target/i386/pr92658-avx512vl.c: Adjust scan-assembler,
only v2di->v2qi truncate is not supported, v4di->v4qi should
be supported.

AVX512FP16: Add expander for cstorehf4.

gcc/ChangeLog:

* config/i386/i386.md (cstorehf3): New define_expand.

gcc/testsuite/ChangeLog:

* gcc.target/i386/avx512fp16-builtin-fpcompare-1.c: New test.
* gcc.target/i386/avx512fp16-builtin-fpcompare-2.c: New test.

AVX512FP16: Add expander for ceil/floor/trunc/roundeven.

gcc/ChangeLog:

* config/i386/i386.md (<rounding_insn>hf2): New expander.
(sse4_1_round<mode>2): Extend from MODEF to MODEFH.
* config/i386/sse.md (*sse4_1_round<ssescalarmodesuffix>):
Extend from VF_128 to VFH_128.

gcc/testsuite/ChangeLog:

* gcc.target/i386/avx512fp16-builtin-round-1.c: New test.

AVX512FP16: Add expander for sqrthf2.

gcc/ChangeLog:

* config/i386/i386-features.c (i386-features.c): Handle
E_HFmode.
* config/i386/i386.md (sqrthf2): New expander.
(*sqrthf2): New define_insn.
* config/i386/sse.md
(*<sse>_vmsqrt<mode>2<mask_scalar_name><round_scalar_name>):
Extend to VFH_128.

gcc/testsuite/ChangeLog:

* gcc.target/i386/avx512fp16-builtin-sqrt-1.c: New test.
* gcc.target/i386/avx512fp16vl-builtin-sqrt-1.c: New test.

AVX512FP16: Add testcases for vfcmaddcsh/vfmaddcsh/vfcmulcsh/vfmulcsh.

gcc/testsuite/ChangeLog:

* gcc.target/i386/avx512fp16-vfcmaddcsh-1a.c: New test.
* gcc.target/i386/avx512fp16-vfcmaddcsh-1b.c: Ditto.
* gcc.target/i386/avx512fp16-vfcmulcsh-1a.c: Ditto.
* gcc.target/i386/avx512fp16-vfcmulcsh-1b.c: Ditto.
* gcc.target/i386/avx512fp16-vfmaddcsh-1a.c: Ditto.
* gcc.target/i386/avx512fp16-vfmaddcsh-1b.c: Ditto.
* gcc.target/i386/avx512fp16-vfmulcsh-1a.c: Ditto.
* gcc.target/i386/avx512fp16-vfmulcsh-1b.c: Ditto.
* gcc.target/i386/avx512fp16-complex-constraints.c: Ditto.

AVX512FP16: Add vfcmaddcsh/vfmaddcsh/vfcmulcsh/vfmulcsh.

gcc/ChangeLog:

* config/i386/avx512fp16intrin.h (_mm_mask_fcmadd_sch):
New intrinsic.
(_mm_mask3_fcmadd_sch): Likewise.
(_mm_maskz_fcmadd_sch): Likewise.
(_mm_fcmadd_sch): Likewise.
(_mm_mask_fmadd_sch): Likewise.
(_mm_mask3_fmadd_sch): Likewise.
(_mm_maskz_fmadd_sch): Likewise.
(_mm_fmadd_sch): Likewise.
(_mm_mask_fcmadd_round_sch): Likewise.
(_mm_mask3_fcmadd_round_sch): Likewise.
(_mm_maskz_fcmadd_round_sch): Likewise.
(_mm_fcmadd_round_sch): Likewise.
(_mm_mask_fmadd_round_sch): Likewise.
(_mm_mask3_fmadd_round_sch): Likewise.
(_mm_maskz_fmadd_round_sch): Likewise.
(_mm_fmadd_round_sch): Likewise.
(_mm_fcmul_sch): Likewise.
(_mm_mask_fcmul_sch): Likewise.
(_mm_maskz_fcmul_sch): Likewise.
(_mm_fmul_sch): Likewise.
(_mm_mask_fmul_sch): Likewise.
(_mm_maskz_fmul_sch): Likewise.
(_mm_fcmul_round_sch): Likewise.
(_mm_mask_fcmul_round_sch): Likewise.
(_mm_maskz_fcmul_round_sch): Likewise.
(_mm_fmul_round_sch): Likewise.
(_mm_mask_fmul_round_sch): Likewise.
(_mm_maskz_fmul_round_sch): Likewise.
* config/i386/i386-builtin.def: Add corresponding new builtins.
* config/i386/sse.md
(avx512fp16_fmaddcsh_v8hf_maskz<round_expand_name>): New expander.
(avx512fp16_fcmaddcsh_v8hf_maskz<round_expand_name>): Ditto.
(avx512fp16_fma_<complexopname>sh_v8hf<mask_scalarcz_name><round_scalarcz_name>):
New define insn.
(avx512fp16_<complexopname>sh_v8hf_mask<round_name>): Ditto.
(avx512fp16_<complexopname>sh_v8hf<mask_scalarc_name><round_scalarcz_name>):
Ditto.
* config/i386/subst.md (mask_scalarcz_name): New.
(mask_scalarc_name): Ditto.
(mask_scalarc_operand3): Ditto.
(mask_scalarcz_operand4): Ditto.
(round_scalarcz_name): Ditto.
(round_scalarc_mask_operand3): Ditto.
(round_scalarcz_mask_operand4): Ditto.
(round_scalarc_mask_op3): Ditto.
(round_scalarcz_mask_op4): Ditto.
(round_scalarcz_constraint): Ditto.
(round_scalarcz_nimm_predicate): Ditto.
(mask_scalarcz): Ditto.
(mask_scalarc): Ditto.
(round_scalarcz): Ditto.

gcc/testsuite/ChangeLog:

* gcc.target/i386/avx-1.c: Add test for new builtins.
* gcc.target/i386/sse-13.c: Ditto.
* gcc.target/i386/sse-23.c: Ditto.
* gcc.target/i386/sse-14.c: Add test for new intrinsics.
* gcc.target/i386/sse-22.c: Ditto.

AVX512FP16: Add testcases for vfcmaddcph/vfmaddcph/vfcmulcph/vfmulcph.

gcc/testsuite/ChangeLog:

* gcc.target/i386/avx512fp16-helper.h
(init_src): Adjust init value.
(NET_CMASK): New net mask for complex input.
* gcc.target/i386/avx512fp16-vfcmaddcph-1a.c: New test.
* gcc.target/i386/avx512fp16-vfcmaddcph-1b.c: Ditto.
* gcc.target/i386/avx512fp16-vfcmulcph-1a.c: Ditto.
* gcc.target/i386/avx512fp16-vfcmulcph-1b.c: Ditto.
* gcc.target/i386/avx512fp16-vfmaddcph-1a.c: Ditto.
* gcc.target/i386/avx512fp16-vfmaddcph-1b.c: Ditto.
* gcc.target/i386/avx512fp16-vfmulcph-1a.c: Ditto.
* gcc.target/i386/avx512fp16-vfmulcph-1b.c: Ditto.
* gcc.target/i386/avx512fp16vl-vfcmaddcph-1a.c: Ditto.
* gcc.target/i386/avx512fp16vl-vfcmaddcph-1b.c: Ditto.
* gcc.target/i386/avx512fp16vl-vfcmulcph-1a.c: Ditto.
* gcc.target/i386/avx512fp16vl-vfcmulcph-1b.c: Ditto.
* gcc.target/i386/avx512fp16vl-vfmaddcph-1a.c: Ditto.
* gcc.target/i386/avx512fp16vl-vfmaddcph-1b.c: Ditto.
* gcc.target/i386/avx512fp16vl-vfmulcph-1a.c: Ditto.
* gcc.target/i386/avx512fp16vl-vfmulcph-1b.c: Ditto.

AVX512FP16: Add vfcmaddcph/vfmaddcph/vfcmulcph/vfmulcph

gcc/ChangeLog:

* config/i386/avx512fp16intrin.h (_mm512_fcmadd_pch):
New intrinsic.
(_mm512_mask_fcmadd_pch): Likewise.
(_mm512_mask3_fcmadd_pch): Likewise.
(_mm512_maskz_fcmadd_pch): Likewise.
(_mm512_fmadd_pch): Likewise.
(_mm512_mask_fmadd_pch): Likewise.
(_mm512_mask3_fmadd_pch): Likewise.
(_mm512_maskz_fmadd_pch): Likewise.
(_mm512_fcmadd_round_pch): Likewise.
(_mm512_mask_fcmadd_round_pch): Likewise.
(_mm512_mask3_fcmadd_round_pch): Likewise.
(_mm512_maskz_fcmadd_round_pch): Likewise.
(_mm512_fmadd_round_pch): Likewise.
(_mm512_mask_fmadd_round_pch): Likewise.
(_mm512_mask3_fmadd_round_pch): Likewise.
(_mm512_maskz_fmadd_round_pch): Likewise.
(_mm512_fcmul_pch): Likewise.
(_mm512_mask_fcmul_pch): Likewise.
(_mm512_maskz_fcmul_pch): Likewise.
(_mm512_fmul_pch): Likewise.
(_mm512_mask_fmul_pch): Likewise.
(_mm512_maskz_fmul_pch): Likewise.
(_mm512_fcmul_round_pch): Likewise.
(_mm512_mask_fcmul_round_pch): Likewise.
(_mm512_maskz_fcmul_round_pch): Likewise.
(_mm512_fmul_round_pch): Likewise.
(_mm512_mask_fmul_round_pch): Likewise.
(_mm512_maskz_fmul_round_pch): Likewise.
* config/i386/avx512fp16vlintrin.h (_mm_fmadd_pch):
New intrinsic.
(_mm_mask_fmadd_pch): Likewise.
(_mm_mask3_fmadd_pch): Likewise.
(_mm_maskz_fmadd_pch): Likewise.
(_mm256_fmadd_pch): Likewise.
(_mm256_mask_fmadd_pch): Likewise.
(_mm256_mask3_fmadd_pch): Likewise.
(_mm256_maskz_fmadd_pch): Likewise.
(_mm_fcmadd_pch): Likewise.
(_mm_mask_fcmadd_pch): Likewise.
(_mm_mask3_fcmadd_pch): Likewise.
(_mm_maskz_fcmadd_pch): Likewise.
(_mm256_fcmadd_pch): Likewise.
(_mm256_mask_fcmadd_pch): Likewise.
(_mm256_mask3_fcmadd_pch): Likewise.
(_mm256_maskz_fcmadd_pch): Likewise.
(_mm_fmul_pch): Likewise.
(_mm_mask_fmul_pch): Likewise.
(_mm_maskz_fmul_pch): Likewise.
(_mm256_fmul_pch): Likewise.
(_mm256_mask_fmul_pch): Likewise.
(_mm256_maskz_fmul_pch): Likewise.
(_mm_fcmul_pch): Likewise.
(_mm_mask_fcmul_pch): Likewise.
(_mm_maskz_fcmul_pch): Likewise.
(_mm256_fcmul_pch): Likewise.
(_mm256_mask_fcmul_pch): Likewise.
(_mm256_maskz_fcmul_pch): Likewise.
* config/i386/i386-builtin-types.def (V8HF_FTYPE_V8HF_V8HF_V8HF,
V8HF_FTYPE_V16HF_V16HF_V16HF, V16HF_FTYPE_V16HF_V16HF_V16HF_UQI,
V32HF_FTYPE_V32HF_V32HF_V32HF_INT,
V32HF_FTYPE_V32HF_V32HF_V32HF_UHI_INT): Add new builtin types.
* config/i386/i386-builtin.def: Add new builtins.
* config/i386/i386-expand.c: Handle new builtin types.
* config/i386/subst.md (SUBST_CV): New.
(maskc_name): Ditto.
(maskc_operand3): Ditto.
(maskc): Ditto.
(sdc_maskz_name): Ditto.
(sdc_mask_op4): Ditto.
(sdc_mask_op5): Ditto.
(sdc_mask_mode512bit_condition): Ditto.
(sdc): Ditto.
(round_maskc_operand3): Ditto.
(round_sdc_mask_operand4): Ditto.
(round_maskc_op3): Ditto.
(round_sdc_mask_op4): Ditto.
(round_saeonly_sdc_mask_operand5): Ditto.
* config/i386/sse.md (unspec): Add complex fma unspecs.
(avx512fmaskcmode): New.
(UNSPEC_COMPLEX_F_C_MA): Ditto.
(UNSPEC_COMPLEX_F_C_MUL): Ditto.
(complexopname): Ditto.
(<avx512>_fmaddc_<mode>_maskz<round_expand_name>): New expander.
(<avx512>_fcmaddc_<mode>_maskz<round_expand_name>): Ditto.
(fma_<complexopname>_<mode><sdc_maskz_name><round_name>): New
define insn.
(<avx512>_<complexopname>_<mode>_mask<round_name>): Ditto.
(<avx512>_<complexopname>_<mode><maskc_name><round_name>): Ditto.

gcc/testsuite/ChangeLog:

* gcc.target/i386/avx-1.c: Add test for new builtins.
* gcc.target/i386/sse-13.c: Ditto.
* gcc.target/i386/sse-23.c: Ditto.
* gcc.target/i386/sse-14.c: Add test for new intrinsics.
* gcc.target/i386/sse-22.c: Ditto.

rs6000: Parameterize some const values for density test

This patch follows the discussion here[1], where Segher suggested
parameterizing those exact magic constants for density heuristics,
to make it easier to tweak if need.

The change here should be "No Functional Change".  But I verified
it with SPEC2017 at option sets O2-vect and Ofast-unroll on Power8,
the result is neutral as expected.

[1]https://gcc.gnu.org/pipermail/gcc-patches/2021-September/579121.html

gcc/ChangeLog:

* config/rs6000/rs6000.opt (rs6000-density-pct-threshold,
rs6000-density-size-threshold, rs6000-density-penalty,
rs6000-density-load-pct-threshold,
rs6000-density-load-num-threshold): New parameter.
* config/rs6000/rs6000.c (rs6000_density_test): Adjust with
corresponding parameters.

c++: fix template instantiation comparison in redeclarations

This change fixes a primordial c++11 frontend defect where function template
redeclarations with trailing return types that used dependent
sizeof/alignof/noexcept expressions in template value arguments failed to
compare as equivalent to the identical primary template declaration. By
forcing structural AST comparison of the template arguments, we no longer
require TYPE_CANONICAL to match in this case. The new canon-type-{15..18}.C
tests failed with all prior GCC versions, where the redeclarations were
incorrectly reported as ambiguous overloads. The new dependent-name{15,16}.C
tests are regression tests for sneaky problems encountered during
development of this fix. Note that this fix does not address the use of parm
objects' constexpr members as template arguments within a declaration (a
superficially similar longstanding defect).

gcc/cp/ChangeLog:

* pt.c (find_parm_usage_r): New walk_tree callback to find func
parms.
(any_template_arguments_need_structural_equality_p): New special
case.

gcc/testsuite/ChangeLog:

* g++.dg/cpp0x/constexpr-52830.C: Remove unwanted dg-ice.
* g++.dg/template/canon-type-15.C: New test.
* g++.dg/template/canon-type-16.C: New test.
* g++.dg/template/canon-type-17.C: New test.
* g++.dg/template/canon-type-18.C: New test.
* g++.dg/template/dependent-name15.C: New regression test.
* g++.dg/template/dependent-name16.C: New regression test.

Daily bump.

runtime: set runtime.GOROOT value at build time

In Go 1.17 the gc toolchain changed to set runtime.GOROOT in cmd/link
(previously it was runtime/internal/sys.GOROOT).  Do the same in libgo.

Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/351313

gotools/:
* Makefile.am (check-runtime): Add goroot.go to --extrafiles.
* Makefile.in: Regenerate.

path solver: Use ranger to solve unknowns.

The default behavior for the path solver is to resort to VARYING when
the range for an unknown SSA is outside the given path.  This is both
cheap and fast, but fails to get a significant amount of ranges that
traditionally the DOM and VRP threaders could get.

This patch uses the ranger to resolve any unknown names upon entry to
the path.  It also uses equivalences to improve ranges.

gcc/ChangeLog:

* gimple-range-path.cc (path_range_query::defined_outside_path):
New.
(path_range_query::range_on_path_entry): New.
(path_range_query::internal_range_of_expr): Resolve unknowns
with ranger.
(path_range_query::improve_range_with_equivs): New.
(path_range_query::ssa_range_in_phi): Resolve unknowns with
ranger.
* gimple-range-path.h (class path_range_query): Add
defined_outside_path, range_on_path_entry, and
improve_range_with_equivs.

path solver: Add related SSAs to solvable set.

The path solver takes an initial set of SSA names which are deemed
interesting.  These are then solved along the path.  Adding any copies
of said SSA names to the list of interesting names yields significantly
better results.  This patch adds said copies to the already provided
list.

Currently this code is guarded by "m_resolve", which is the more
expensive mode, but it would be reasonable to make it available always,
especially since adding more imports usually has minimal impact on the
processing time.  I will investigate and make it universally available
if this is indeed the case.

gcc/ChangeLog:

* gimple-range-path.cc (path_range_query::add_to_imports): New.
(path_range_query::add_copies_to_imports): New.
(path_range_query::precompute_ranges): Call
add_copies_to_imports.
* gimple-range-path.h (class path_range_query): Add prototypes
for add_copies_to_imports and add_to_imports.

path solver: Remove useless code.

gcc/ChangeLog:

* gimple-range-path.cc (path_range_query::range_defined_in_block):
Remove useless code.

path solver: Add relation support.

This patch adds relational support to the path solver.  It uses a
path_oracle that keeps track of relations within a path which are
augmented by relations on entry to the path.  With it, range_of_stmt,
range_of_expr, and friends can give relation aware answers.

gcc/ChangeLog:

* gimple-range-fold.h (class fur_source): Make oracle protected.
* gimple-range-path.cc (path_range_query::path_range_query): Add
resolve argument.  Initialize oracle.
(path_range_query::~path_range_query): Delete oracle.
(path_range_query::range_of_stmt): Adapt to use relations.
(path_range_query::precompute_ranges): Pre-compute relations.
(class jt_fur_source): New
(jt_fur_source::jt_fur_source): New.
(jt_fur_source::register_relation): New.
(jt_fur_source::query_relation): New.
(path_range_query::precompute_relations): New.
(path_range_query::precompute_phi_relations): New.
* gimple-range-path.h (path_range_query): Add resolve argument.
Add oracle, precompute_relations, precompute_phi_relations.
* tree-ssa-threadbackward.c (back_threader::back_threader): Pass
resolve argument to solver.

Move postfold_gcond_edges into fur_source.

The code registering outgoing edges from a cond is living in
fold_using_range, which makes it difficult to be called from other
places.  Also, it refuses to register relations on the outgoing
destinations that have more than one predecessor.  This latter issue is
a problem because we would like to register outgoing edges along a path
in the path solver (regardless of single_pred_p).

gcc/ChangeLog:

* gimple-range-fold.cc (fold_using_range::range_of_range_op):
Rename postfold_gcond_edges to register_outgoing_edges and
adapt.
(fold_using_range::postfold_gcond_edges): Rename...
(fur_source::register_outgoing_edges): ...to this.
* gimple-range-fold.h (postfold_gcond_edges): Rename to
register_outgoing_edges and move to fur_source.

Do not query SCEV in range_of_phi unless dominators are available.

SCEV won't work without dominators and we can get called without
dominators from debug_ranger.

Another option would be to rename scev_initialized_p to something like
scev_available_p and move the check there.  For now, this will do.

gcc/ChangeLog:

* gimple-range-fold.cc (fold_using_range::range_of_phi): Check
dom_info_available_p.

Allocate non_null_ref tables at creation.

Preallocating the space is slightly cheaper than calling
safe_grow_cleared.

gcc/ChangeLog:

* gimple-range-cache.cc (non_null_ref::non_null_ref): Use create
and quick_grow_cleared instead of safe_grow_cleared.

[Ada] Add some comments in init.c about the lynx178 signal handler

gcc/ada/

* init.c (__gnat_error_handler) [LynxOS]: Add a comment about
missing optional args.

[Ada] Rename "optional" node subtypes that allow Empty

gcc/ada/

* gen_il-gen.adb (Put_Opt_Subtype): Add suffix.

[Ada] Spurious dynamic accessibility check on allocator

gcc/ada/

* sem_util.adb (Accessibility_Level): Remove spurious special
case for protected type components.
* exp_ch4.adb (Generate_Accessibility_Check): Use general
Accessibility_Level instead of the low-level function
Type_Access_Level.

[Ada] Regenerate gnat_ugn.texi

gcc/ada/

* gnat_ugn.texi: Regenerate.

[Ada] SCOs: generate 'P' decisions for [Type_]Invariant pragmas

gcc/ada/

* par_sco.adb (Traverse_One): Add support for pragma Invariant /
Type_Invariant.

[Ada] Add "optional" node subtypes that allow Empty

gcc/ada/

* gen_il-gen.adb (Put_Opt_Subtype): Print out subtypes of the
form:
subtype Opt_N_Declaration is
Node_Id with Predicate =>
Opt_N_Declaration = Empty or else
Opt_N_Declaration in N_Declaration_Id;
One for each node or entity type, with the predicate allowing
Empty.
* atree.adb (Parent, Set_Parent): Remove unnecessary "Atree.".

[Ada] bindgen: support additional features on targets suppressing the standard lib

gcc/ada/

* bindgen.adb (Gen_Adainit): For targets that suppress the
standard library: set the default stack size global variable if
a value is provided via the -d switch, and generate a call to
__gnat_initialize_stack_limit if stack checking using stack
limits is enabled.

[Ada] Fix regression in ACATS bdd2006 and bdd2007

gcc/ada/

* sem_ch13.adb (Stream_Size): Print message about allowed stream
sizes even if other error were already found. This avoids
falling into the 'else', which prints "Stream_Size cannot be
given for...", which is misleading -- the Size COULD be given if
it were correct.

[Ada] Set related expression for external DISCR symbols in Build_Temporary

gcc/ada/

* exp_util.adb (Build_Temporary): In case of an external DISCR
symbol, set the related expression for CodePeer so that a more
comprehensible message can be emitted to the user.

[Ada] Minor tweaks to System.Dwarf_Line

gcc/ada/

* libgnat/s-dwalin.adb (Parse_Header): Tweak comments.
(Read_Entry_Format_Array): Tweak exception message.
(Symbolic_Address.Set_Result): Likewise.

[Ada] Crash on build of Initialization procedure for derived container

gcc/ada/

* exp_ch7.adb (Make_Init_Call): Add guard to protect against a
missing initialization procedure for a type.

[Ada] Cleanup old VxWorks in Makefile.rtl

gcc/ada/

* Makefile.rtl: Remove unused VxWorks sections.
* libgnarl/s-vxwext__noints.adb: Remove.
* libgnarl/s-vxwext__vthreads.ads: Remove.
* libgnat/a-elchha__vxworks-ppc-full.adb: Remove.
* libgnat/s-osprim__vxworks.adb: Remove.
* libgnat/s-osvers__vxworks-653.ads: Remove.
* libgnat/system-vxworks-e500-vthread.ads: Remove.
* libgnat/system-vxworks-ppc-vthread.ads: Remove.
* libgnat/system-vxworks-x86-vthread.ads: Remove.

[Ada] Add assertions to Uintp (UI_Is_In_Int_Range)

gcc/ada/

* uintp.ads, uintp.adb (UI_Is_In_Int_Range): Change the type of
the formal parameter to Valid_Uint. Remove code that preserved
the previous behavior, and replace it with an assertion. The
previous behavior is no longer needed given the recent change to
gigi.
(No, Present): Add comment.

[Ada] Remove if_expression

gcc/ada/

* sem_eval.adb (Fold_Shift): Replace an if_expression with an
if_statement.

[Ada] Add assertions to Uintp

gcc/ada/

* uintp.ads, uintp.adb: Add assertions.
(Ubool, Opt_Ubool): New "boolean" subtypes.
(UI_Is_In_Int_Range): The parameter should probably be
Valid_Uint, but we don't change that for now, because it causes
failures in gigi.
* sem_util.ads, sem_util.adb (Is_True, Is_False,
Static_Boolean): Use Opt_Ubool subtype.  Document the fact that
Is_True (No_Uint) = True.  Implement Is_False in terms of
Is_True.  We considered changing Static_Boolean to return Uint_1
in case of error, but that doesn't fit in well with
Static_Integer.
(Has_Compatible_Alignment_Internal): Deal with cases where Offs
is No_Uint. Change one "and" to "and then" to ensure we don't
pass No_Uint to ">", which would violate the new assertions.
* exp_util.adb, freeze.adb, sem_ch13.adb: Avoid violating new
assertions in Uintp.

[Ada] Small optimization to DWARF 5 mode in System.Dwarf_Line

gcc/ada/

* libgnat/s-dwalin.adb (To_File_Name): Fetch only the last string
from the .debug_line_str section.
(Symbolic_Address.Set_Result): Likewise.

[Ada] Follow-up tweaks to System.Dwarf_Line

gcc/ada/

* libgnat/s-dwalin.adb (Skip_Form): Fix cases of DW_FORM_addrx
and DW_FORM_implicit_const.  Replace Constraint_Error with
Dwarf_Error.

[Ada] exp_pakd.adb: work around spurious Codepeer warnings

gcc/ada/

* exp_pakd.adb (Expand_Packed_Not): Replace expression with
statement.

[Ada] Fix ignored dynamic predicates specified through "predicate" aspect

gcc/ada/

* sem_eval.adb (Is_Static_Subtype): Take predicates created
through "predicate" pragma into account.

[Ada] rtems: add 128bit support for aarch64

gcc/ada/

* Makefile.rtl (aarch64-rtems*): Add GNATRTL_128BIT_PAIRS to
the LIBGNAT_TARGET_PAIRS list and also GNATRTL_128BIT_OBJS to
the EXTRA_GNATRTL_NONTASKING_OBJS list.

[Ada] Presence of abstract operator function causes resolution problems

gcc/ada/

* sem_ch4.adb (Remove_Abstract_Operations): Add condition to
test for an E_Operator as part of criteria for setting
Abstract_Op on interpretations involving predefined operators.

[Ada] Interface behaves differently from abstract tagged null

gcc/ada/

* exp_ch6.adb (Expand_Simple_Function_Return): For explicit
dereference of type conversion, enable code that ensures that
the tag of the result is that of the result type.

[Ada] Clean up uses of Esize and RM_Size

gcc/ada/

* einfo-utils.adb: Add support (currently disabled) for using
"initial zero" instead of "Uint_0" to represent "unknown".  Call
Known_ functions, instead of evilly duplicating their code
inline.
* fe.h (No_Uint_To_0): New function to convert No_Uint to
Uint_0, in order to preserve existing behavior.
(Copy_Esize, Copy_RM_Size): New imports from Einfo.Utils.
* cstand.adb: Set size fields of Standard_Debug_Renaming_Type
and Standard_Exception_Type.
* checks.adb, exp_attr.adb, exp_ch3.adb, exp_ch5.adb,
exp_ch6.adb, exp_pakd.adb, exp_util.adb, freeze.adb, itypes.adb,
layout.adb, repinfo.adb, sem_attr.adb, sem_ch12.adb,
sem_ch13.adb, sem_ch13.ads, sem_ch3.adb, sem_ch7.adb,
sem_util.adb: Protect calls with Known_..., use Copy_...  Remove
assumption that Uint_0 represents "unknown".
* types.ads (Nonzero_Int): New subtype.
* gcc-interface/decl.c, gcc-interface/trans.c: Protect calls
with Known_... and use Copy_...  as appropriate, to avoid
blowing up in unknown cases. Similarly, call No_Uint_To_0 to
preserve existing behavior.

[Ada] Enforce legality rule for Predicate_Failure aspect specifications

gcc/ada/

* sem_ch13.adb (Analyze_Aspect_Specifications): Add a new nested
function, Directly_Specified, and then use it in the
implementation of the required check.

[Ada] Refactor sort procedures of doubly linked list containers

gcc/ada/

* libgnat/a-costso.ads, libgnat/a-costso.adb: A new library
unit, Ada.Containers.Stable_Sorting, which exports a pair of
generics (one within the other) which are instantiated by each
of the 5 doubly-linked list container generics to implement
their respective Sort procedures. We use a pair of generics,
rather than a single generic, in order to further reduce code
duplication. The outer generic takes a formal private Node_Ref
type representing a reference to a linked list element. For some
instances, the corresponding actual parameter will be an access
type; for others, it will be the index type for an array.
* Makefile.rtl: Include new Ada.Containers.Stable_Sorting unit.
* libgnat/a-cbdlli.adb, libgnat/a-cdlili.adb,
libgnat/a-cfdlli.adb, libgnat/a-cidlli.adb, libgnat/a-crdlli.adb
(Sort): Replace existing Sort implementation with a call to an
instance of
Ada.Containers.Stable_Sorting.Doubly_Linked_List_Sort. Declare
the (trivial) actual parameters needed to declare that instance.
* libgnat/a-cfdlli.ads: Fix a bug encountered during testing in
the postcondition for M_Elements_Sorted. With a partial
ordering, it is possible for all three of (X < Y), (Y < X),
and (X = Y) to be simultaneously false, so that case needs to
handled correctly.

[Ada] Update comment for Error_Msg_Internal

gcc/ada/

* errout.adb (Error_Msg_Internal): Fix references to Sptr and
Optr in comment; fix grammar of "low-level" where it is used as
an adjective.

[Ada] Simplify iteration when printing error message spans

gcc/ada/

* errout.adb (Write_Source_Code_Lines): Use Cur_Loc before
incrementing it, so that we don't need to decrement it.

[Ada] Exception raised on empty file in GNATprove mode

gcc/ada/

* errout.adb (Get_Line_End): Do not allow the result to go past
the end of the buffer.

[Ada] Refine patch for spurious link error involving discriminated types

gcc/ada/

* sem_ch3.adb (Process_Discriminant_Expressions): If the
constraint is for a Component_Definition that appears in a
Component_Declaration, the entity to be used to create the
potentially global symbol is the Defining_Identifier of the
Component_Declaration.

[Ada] Remove "with GNAT.OS_Lib;" from libgnat/a-stbufi.ads

gcc/ada/

* libgnat/a-stbufi.ads, libgnat/a-stbufi.adb: Change all
occurrences of GNAT.OS_Lib to System.OS_Lib.

Evaluate 'random ()' to '0' in 'pass_omp_oacc_neuter_broadcast'

Julian Brown,
<http://mid.mail-archive.com/20210920134603.16459021@squid.athome>:

| [...] the randomness shouldn't be necessary for the
| correctness of the patch (i.e. it could just be "base = bounds_lo", or
| indeed folded into the line after).
|
| The "ar.invalid ()" case happens when we fail to allocate a block of
| memory in LDS space for broadcasting a particular set of variables,
| and trigger a fall-back path in the broadcasting code that adds extra
| barriers around the broadcast in question. I imagine I was thinking
| that adding randomness could mean we can "get lucky" sometimes and
| avoid needing those barriers in some cases, but in fact I don't think
| that was implemented, so the randomness is useless. (Or it could just
| have been leftover debug code... oops).

gcc/
PR other/102408
* omp-oacc-neuter-broadcast.cc (oacc_do_neutering): Evaluate
'random ()' to '0'.

arm: pass architecture extensions to assembler if supported

When I originally added the new extended architecture features support
to GCC, the assembler was unable to parse the new feature lists on the
command-line and would throw an error.  This has now been fixed in GAS
and the behaviour is the same as GCC.

So this patch adds a configure-time test for the assembler in use to
detect that it is recent enough to support this behaviour and then
enables passing the architecture extensions to the assembler when this
is the case.

Although the assembly output generated by GCC does not rely on this,
this should ease some issues when using GCC as a driver for compiling
source written directly in assembly language when the programmer was
expecting the appropriate options to be passed on the command line.

gcc/ChangeLog:

* configure.ac: Detect when the assembler supports new-style
architecture extensions.
* common/config/arm/arm-common.c (arm_rewrite_mcpu): Return
the full CPU string if the assembler can grok it.
(arm_rewrite_march): Likewise but for the architecture.
* config.in: Regenerate.
* configure: Regenerate.

tree-optimization/102421 - copy alignment info when splitting groups

This makes sure to copy and adjust alignment info when we are splitting
DR groups after alignment analysis.

2021-09-21  Richard Biener  <rguenther@suse.de>

PR tree-optimization/102421
* tree-vect-loop.c (vect_dissolve_slp_only_groups): Copy and
adjust alignment info.

* g++.dg/vect/pr102421.cc: New testcase.

Fix no_fsanitize_address effective target

The implementation of the no_fsanitize_address effective target was copied
from asan-dg.exp without realizing that it does not work outside of this
context (there is a comment explaining why).  As a consequence, it always
returns 0, so for example the directive in gnat.dg/asan1.adb:

{ dg-skip-if "no address sanitizer" { no_fsanitize_address } }

does not work.  This led some people to add the nonsensical:

{ dg-require-effective-target no_fsanitize_address }

to sanitizer tests, e.g. g++.dg/warn/uninit-pr93100.C, thus disabling them
everywhere instead of just for the problematic targets.

gcc/testsuite/
* lib/target-supports.exp (no_fsanitize_address): Add missing bits.
* gcc.dg/uninit-pr93100.c: Skip if no_fsanitize_address.
* gcc.dg/pr91441.c: Likewise.
* gcc.dg/pr96260.c: Likewise.
* gcc.dg/pr96307.c: Likewise.
* g++.dg/warn/uninit-pr93100.C: Likewise.
* gnat.dg/asan1.adb: Likewise.

* gcc.dg/Wstringop-overflow-70.c: Adjust for SPARC.
* g++.dg/abi/anon4.C: Likewise.

Fortran: Fix -Wno-missing-include-dirs handling [PR55534]

gcc/fortran/ChangeLog:

PR fortran/55534
* cpp.c: Define GCC_C_COMMON_C for #include "options.h" to make
cpp_reason_option_codes available.
(gfc_cpp_register_include_paths): Make static, set pfile's
warn_missing_include_dirs and move before caller.
(gfc_cpp_init_cb): New, cb code moved from ...
(gfc_cpp_init_0): ... here.
(gfc_cpp_post_options): Call gfc_cpp_init_cb.
(cb_cpp_diagnostic_cpp_option): New. As implemented in c-family
to match CppReason flags to -W... names.
(cb_cpp_diagnostic): Use it to replace single special case.
* cpp.h (gfc_cpp_register_include_paths): Remove as now static.
* gfortran.h (gfc_check_include_dirs): New prototype.
(gfc_add_include_path): Add new bool arg.
* options.c (gfc_init_options): Don't set -Wmissing-include-dirs.
(gfc_post_options): Set it here after commandline processing. Call
gfc_add_include_path with defer_warn=false.
(gfc_handle_option): Call it with defer_warn=true.
* scanner.c (gfc_do_check_include_dir, gfc_do_check_include_dirs,
gfc_check_include_dirs): New. Diagnostic moved from ...
(add_path_to_list): ... here, which came before cmdline processing.
Take additional bool defer_warn argument.
(gfc_add_include_path): Take additional defer_warn arg.
* scanner.h (struct gfc_directorylist): Reorder for alignment issues,
add new 'bool warn'.

libgfortran/ChangeLog:
PR fortran/55534
* configure.ac (AM_FCFLAGS): Add -Wno-missing-include-dirs.
* configure: Regenerate.

libgomp/ChangeLog:
PR fortran/55534
* testsuite/libgomp.fortran/fortran.exp: Add -Wno-missing-include-dirs
to ALWAYS_CFLAGS.
* testsuite/libgomp.oacc-fortran/fortran.exp: Likewise.

gcc/testsuite/ChangeLog:
* gfortran.dg/include_6.f90: Change dg-error to
dg-warning and update pattern.
* gfortran.dg/include_14.f90: New test.
* gfortran.dg/include_15.f90: New test.
* gfortran.dg/include_16.f90: New test.
* gfortran.dg/include_17.f90: New test.
* gfortran.dg/include_18.f90: New test.
* gfortran.dg/include_19.f90: New test.
* gfortran.dg/include_20.f90: New test.
* gfortran.dg/include_21.f90: New test.

ipa-fnsummary: Remove inconsistent bp_pack_value

There is one inconsistent bit-field streaming out and in.
On the side of streaming in:

    bp_pack_value (&bp, info->inlinable, 1);
    bp_pack_value (&bp, false, 1);
    bp_pack_value (&bp, info->fp_expressions, 1);

while on the side of the streaming out:

    info->inlinable = bp_unpack_value (&bp, 1);
    info->fp_expressions = bp_unpack_value (&bp, 1)

The removal of Cilk Plus support r8-4956 missed to remove
the streaming out of the bit, instead just change the value
for streaming out to be always false.

By hacking fp_expression_p to always return true, I can see
it reads the wrong fp_expressions value (false) out in wpa.

gcc/ChangeLog:

* ipa-fnsummary.c (ipa_fn_summary_write): Remove inconsistent
bitfield stream out.

C++: add type checking for static local vector variable in template

This patch moves verify_type_context from start_decl_1 to cp_finish_decl to
do more type checking such as static local vector variable in C++ template.

2021-08-06  wangpc  <pc.wang@linux.alibaba.com>

gcc/cp/ChangeLog

* decl.c (start_decl_1): Move verify_type_context to ...
(cp_finish_decl): ... to here.

gcc/testsuite/ChangeLog

* g++.target/aarch64/sve/static-var-in-template.C: New test.

Daily bump.

Use EDGE_EXECUTABLE in ranger and return UNDEFINED for those edges.

If an incoming edge is UNDEFINED, don't process it.  Track if other edges
equate to a single value, and add an equivalence if appropriate.

gcc/
* gimple-range-fold.cc (fold_using_range::range_of_phi): Ignore
undefined edges, apply an equivalence if appropriate.
* gimple-range-gori.cc (gori_compute::outgoing_edge_range_p): Return
UNDEFINED if EDGE_EXECUTABLE is not set.
* gimple-range.cc (gimple_ranger::gimple_ranger): Set all edges
as EXECUTABLE upon startup.
(gimple_ranger::range_on_edge): Return UNDEFINED for edges without
EDGE_EXECUTABLE set.
* vr-values.c (set_and_propagate_unexecutable): New.
(simplify_using_ranges::fold_cond): Call set_and_propagate.
(simplify_using_ranges::simplify_switch_using_ranges): Ditto.
* vr-values.h: Add prototype.

gcc/testsuite/
* gcc.dg/tree-ssa/evrp-ignore.c: New.

Make each def a new equivalency record.

Create a new equivalency set at each def point killing any equivalencies
coming into the block from back edges.  Do not add equivalences for PHI
arguments defined in this block.

* value-relation.cc (equiv_oracle::register_initial_def): New.
(equiv_oracle::register_relation): Call register_initial_def.
(equiv_oracle::add_equiv_to_block): New.  Split register_relation.
(relation_oracle::register_stmt): Check def block of PHI arguments.
* value-relation.h (equiv_oracle): Add new prototypes.

MAINTAINERS: add myself as Fortran reviewer

ChangeLog:

* MAINTAINERS (Reviewers): Add myself as Fortran reviewer.

c-family: Add more predefined macros for math flags

Library code, especially in headers, sometimes needs to know how the
compiler interprets / optimizes floating-point types and operations.
This information can be used for additional optimizations or for
ensuring correctness. This change makes -freciprocal-math,
-fno-signed-zeros, -fno-trapping-math, -fassociative-math, and
-frounding-math report their state via corresponding pre-defined macros.

Signed-off-by: Matthias Kretz <m.kretz@gsi.de>
gcc/testsuite/ChangeLog:

* gcc.dg/associative-math-1.c: New test.
* gcc.dg/associative-math-2.c: New test.
* gcc.dg/no-signed-zeros-1.c: New test.
* gcc.dg/no-signed-zeros-2.c: New test.
* gcc.dg/no-trapping-math-1.c: New test.
* gcc.dg/no-trapping-math-2.c: New test.
* gcc.dg/reciprocal-math-1.c: New test.
* gcc.dg/reciprocal-math-2.c: New test.
* gcc.dg/rounding-math-1.c: New test.
* gcc.dg/rounding-math-2.c: New test.

gcc/c-family/ChangeLog:

* c-cppbuiltin.c (c_cpp_builtins_optimize_pragma): Define or
undefine __RECIPROCAL_MATH__, __NO_SIGNED_ZEROS__,
__NO_TRAPPING_MATH__, __ASSOCIATIVE_MATH__, and
__ROUNDING_MATH__ according to the new optimization flags.

gcc/ChangeLog:

* cppbuiltin.c (define_builtin_macros_for_compilation_flags):
Define __RECIPROCAL_MATH__, __NO_SIGNED_ZEROS__,
__NO_TRAPPING_MATH__, __ASSOCIATIVE_MATH__, and
__ROUNDING_MATH__ according to their corresponding flags.
* doc/cpp.texi: Document __RECIPROCAL_MATH__,
__NO_SIGNED_ZEROS__, __NO_TRAPPING_MATH__, __ASSOCIATIVE_MATH__,
and __ROUNDING_MATH__.

[Ada] Remove redundant checks for non-empty list of aspects

gcc/ada/

* inline.adb (Has_Excluded_Declaration): Remove redundant guard;
the guarded code will call First on a No_List, which is
well-defined and gives Empty.

[Ada] Fix shadowing in conditions for inlining

gcc/ada/

* inline.adb (Has_Excluded_Declaration): Rename and reduce scope
of a local variable.

[Ada] Present and No functions for type Uint

gcc/ada/

* uintp.ads, uintp.adb (Present, No): New functions for
comparing with No_Uint.
* checks.adb, einfo-utils.adb, exp_aggr.adb, exp_attr.adb,
exp_ch3.adb, exp_ch4.adb, exp_dbug.adb, exp_disp.adb,
exp_util.adb, repinfo.adb, repinfo-input.adb, scn.adb,
sem_attr.adb, sem_ch13.adb, sem_eval.adb, sem_util.adb,
sinfo-utils.adb, treepr.adb: Use Present (...) instead of "...
/= No_Uint", and No (...) instead of "... = No_Uint".

[Ada] Add SPARK_Mode => Off to System.File_Control_Block

gcc/ada/

* libgnat/s-ficobl.ads: The entire package has a SPARK_Mode =>
Off aspect.