builtin-types.def (BT_FN_BOOL_UINT_LONGPTR_LONGPTR_LONGPTR, [...]): New.

gcc/
2015-10-13  Jakub Jelinek  <jakub@redhat.com>
	    Aldy Hernandez  <aldyh@redhat.com>
	    Ilya Verbin  <ilya.verbin@intel.com>

	* builtin-types.def (BT_FN_BOOL_UINT_LONGPTR_LONGPTR_LONGPTR,
	BT_FN_BOOL_UINT_ULLPTR_ULLPTR_ULLPTR,
	BT_FN_BOOL_UINT_LONGPTR_LONG_LONGPTR_LONGPTR,
	BT_FN_BOOL_UINT_ULLPTR_ULL_ULLPTR_ULLPTR,
	BT_FN_VOID_INT_SIZE_PTR_PTR_PTR_UINT_PTR,
	BT_FN_VOID_INT_OMPFN_SIZE_PTR_PTR_PTR_UINT_PTR,
	BT_FN_VOID_OMPFN_PTR_OMPCPYFN_LONG_LONG_BOOL_UINT_PTR_INT,
	BT_FN_VOID_OMPFN_PTR_OMPCPYFN_LONG_LONG_UINT_LONG_INT_LONG_LONG_LONG,
	BT_FN_VOID_OMPFN_PTR_OMPCPYFN_LONG_LONG_UINT_LONG_INT_ULL_ULL_ULL,
	BT_FN_VOID_LONG_VAR, BT_FN_VOID_ULL_VAR): New.
	(BT_FN_VOID_INT_PTR_SIZE_PTR_PTR_PTR,
	BT_FN_VOID_INT_OMPFN_PTR_SIZE_PTR_PTR_PTR,
	BT_FN_VOID_OMPFN_PTR_OMPCPYFN_LONG_LONG_BOOL_UINT_PTR): Remove.
	* cgraph.h (enum cgraph_simd_clone_arg_type): Add
	SIMD_CLONE_ARG_TYPE_LINEAR_REF_CONSTANT_STEP,
	SIMD_CLONE_ARG_TYPE_LINEAR_UVAL_CONSTANT_STEP and
	SIMD_CLONE_ARG_TYPE_LINEAR_VAL_CONSTANT_STEP.
	(struct cgraph_simd_clone_arg): Adjust comment.
	* coretypes.h (struct gomp_ordered): New forward decl.
	* gimple.c (gimple_build_omp_critical): Add CLAUSES argument,
	set critical clauses to it.
	(gimple_build_omp_ordered): Return gomp_ordered * instead of
	gimple *.  Add CLAUSES argument, set ordered clauses to it.
	(gimple_copy): Unshare clauses on GIMPLE_OMP_CRITICAL and
	GIMPLE_OMP_ORDERED.
	* gimple.def (GIMPLE_OMP_ORDERED): Change from GSS_OMP to
	GSS_OMP_SINGLE_LAYOUT, move it after GIMPLE_OMP_TEAMS.
	* gimple.h (enum gf_mask): Add GF_OMP_TASK_TASKLOOP.  Add another bit
	to GF_OMP_FOR_KIND_MASK mask. Add GF_OMP_FOR_KIND_TASKLOOP, renumber
	GF_OMP_FOR_KIND_CILKFOR and GF_OMP_FOR_KIND_OACC_LOOP.  Adjust
	GF_OMP_FOR_SIMD, GF_OMP_FOR_COMBINED and GF_OMP_FOR_COMBINED_INTO.
	Add another bit to GF_OMP_TARGET_KIND_MASK mask.  Add
	GF_OMP_TARGET_KIND_ENTER_DATA and GF_OMP_TARGET_KIND_EXIT_DATA,
	renumber
	GF_OMP_TARGET_KIND_OACC_{PARALLEL,KERNELS,DATA,UPDATE,ENTER_EXIT_DATA}.
	(gomp_critical): Add clauses field.
	(gomp_ordered): New struct.
	(is_a_helper <gomp_ordered *>::test): New inline.
	(gimple_build_omp_critical): Add CLAUSES argument.
	(gimple_build_omp_ordered): Likewise.  Return gomp_ordered *
	instead of gimple *.
	(gimple_omp_critical_clauses, gimple_omp_critical_clauses_ptr,
	gimple_omp_critical_set_clauses, gimple_omp_ordered_clauses,
	gimple_omp_ordered_clauses_ptr, gimple_omp_ordered_set_clauses,
	gimple_omp_task_taskloop_p, gimple_omp_task_set_taskloop_p): New
	inline functions.
	* gimple-pretty-print.c (dump_gimple_omp_for): Handle taskloop.
	(dump_gimple_omp_target): Handle enter data and exit data.
	(dump_gimple_omp_block): Don't handle GIMPLE_OMP_ORDERED here.
	(dump_gimple_omp_critical): Print clauses.
	(dump_gimple_omp_ordered): New function.
	(dump_gimple_omp_task): Handle taskloop.
	(pp_gimple_stmt_1): Use dump_gimple_omp_ordered for
	GIMPLE_OMP_ORDERED.
	* gimple-walk.c (walk_gimple_op): Walk clauses on
	GIMPLE_OMP_CRITICAL and GIMPLE_OMP_ORDERED.
	* gimplify.c (enum gimplify_omp_var_data): Add GOVD_MAP_0LEN_ARRAY.
	(enum omp_region_type): Add ORT_COMBINED_TARGET and ORT_NONE.
	(struct gimplify_omp_ctx): Add loop_iter_var,
	target_map_scalars_firstprivate, target_map_pointers_as_0len_arrays
	and target_firstprivatize_array_bases fields.
	(delete_omp_context): Release loop_iter_var.
	(gimplify_bind_expr): Handle ORT_NONE.
	(maybe_fold_stmt): Adjust check for ORT_TARGET for the addition of
	ORT_COMBINED_TARGET.
	(is_gimple_stmt): Return true for OMP_TASKLOOP, OMP_TEAMS and
	OMP_TARGET{,_DATA,_UPDATE,_ENTER_DATA,_EXIT_DATA}.
	(omp_firstprivatize_variable): Handle ORT_NONE.  Adjust check for
	ORT_TARGET for the addition of ORT_COMBINED_TARGET.  Handle
	ctx->target_map_scalars_firstprivate.
	(omp_add_variable): Handle ORT_NONE.  Allow map clause together with
	data sharing clauses.  For data sharing clause with VLA decl
	on omp target/target data don't add firstprivate for the pointer.
	Call omp_notice_variable on TYPE_SIZE_UNIT only if it is a DECL_P.
	(omp_notice_threadprivate_variable): Adjust check for ORT_TARGET for
	the addition of ORT_COMBINED_TARGET.
	(omp_notice_variable): Handle ORT_NONE.  Adjust check for ORT_TARGET
	for the addition of ORT_COMBINED_TARGET.  Handle implicit mapping of
	pointers as zero length array sections and
	ctx->target_map_scalars_firstprivate mapping of scalars as firstprivate
	data sharing.
	(omp_check_private): Handle omp_member_access_dummy_var vars.
	(find_decl_expr): New function.
	(gimplify_scan_omp_clauses): Add CODE argument.  For OMP_CLAUSE_IF
	complain if OMP_CLAUSE_IF_MODIFIER is present and does not match code.
	Handle OMP_CLAUSE_GANG separately.  Handle
	OMP_CLAUSE_{PRIORITY,GRAINSIZE,NUM_TASKS,NOGROUP,THREADS,SIMD,SIMDLEN}
	clauses.  Diagnose linear clause on combined
	distribute {, parallel for} simd construct, unless it is the loop
	iterator.  Handle struct element GOMP_MAP_FIRSTPRIVATE_POINTER.
	Handle map clauses with COMPONENT_REF.  Initialize
	ctx->target_map_scalars_firstprivate,
	ctx->target_firstprivatize_array_bases and
	ctx->target_map_pointers_as_0len_arrays.  Add firstprivate for
	linear clause even to target region if combined.  Remove
	map clauses with GOMP_MAP_FIRSTPRIVATE_POINTER kind from
	OMP_TARGET_{,ENTER_,EXIT_}DATA.  For GOMP_MAP_FIRSTPRIVATE_POINTER
	map kind with non-INTEGER_CST OMP_CLAUSE_SIZE firstprivatize the bias.
	Handle OMP_CLAUSE_DEPEND_{SINK,SOURCE}.  Handle
	OMP_CLAUSE_{{USE,IS}_DEVICE_PTR,DEFAULTMAP,HINT}.
	For linear clause on worksharing loop combined with parallel add
	shared clause on the parallel.  Handle OMP_CLAUSE_REDUCTION
	with MEM_REF OMP_CLAUSE_DECL.  Set DECL_NAME on
	omp_member_access_dummy_var vars.  Add lastprivate clause to outer
	taskloop if needed.
	(gimplify_adjust_omp_clauses_1): Handle GOVD_MAP_0LEN_ARRAY.
	If gimplify_omp_ctxp->target_firstprivatize_array_bases, use
	GOMP_MAP_FIRSTPRIVATE_POINTER map kind instead of
	GOMP_MAP_POINTER.
	(gimplify_adjust_omp_clauses): Add CODE argument.  Handle removal
	of GOMP_MAP_FIRSTPRIVATE_POINTER struct elements for struct not seen
	in target body.  Handle removal of struct mapping if struct is not
	seen in target body.  Remove GOMP_MAP_STRUCT map clause on
	OMP_TARGET_EXIT_DATA.  Adjust check for ORT_TARGET for the
	addition of ORT_COMBINED_TARGET.  Use GOMP_MAP_FIRSTPRIVATE_POINTER
	instead of GOMP_MAP_POINTER if ctx->target_firstprivatize_array_bases
	for VLAs.  Set OMP_CLAUSE_MAP_PRIVATE if both data sharing and map
	clause appear together.  Handle
	OMP_CLAUSE_{{USE,IS}_DEVICE_PTR,DEFAULTMAP,HINT}.  Don't remove map
	clause if it has map-type-modifier always.  Handle
	OMP_CLAUSE_{PRIORITY,GRAINSIZE,NUM_TASKS,NOGROUP,THREADS,SIMD,SIMDLEN}
	clauses.
	(gimplify_oacc_cache, gimplify_omp_parallel, gimplify_omp_task):
	Adjust gimplify_scan_omp_clauses and gimplify_adjust_omp_clauses
	callers.
	(gimplify_omp_for): Likewise.  Handle OMP_TASKLOOP.  Initialize
	loop_iter_var.  Use OMP_FOR_ORIG_DECLS.  Fix handling of lastprivate
	iterators in doacross loops.
	(gimplify_omp_workshare): Adjust gimplify_scan_omp_clauses and
	gimplify_adjust_omp_clauses callers.  Use ORT_COMBINED_TARGET
	for OMP_TARGET_COMBINED.  Adjust check for ORT_TARGET
	for the addition of ORT_COMBINED_TARGET.
	(gimplify_omp_target_update): Adjust gimplify_scan_omp_clauses and
	gimplify_adjust_omp_clauses callers.  Handle OMP_TARGET_ENTER_DATA
	and OMP_TARGET_EXIT_DATA.
	(gimplify_omp_ordered): New function.
	(gimplify_expr): Handle OMP_TASKLOOP, OMP_TARGET_ENTER_DATA and
	OMP_TARGET_EXIT_DATA.  Use gimplify_omp_ordered for OMP_ORDERED.
	Gimplify clauses on OMP_CRITICAL.
	* internal-fn.c (expand_GOMP_SIMD_ORDERED_START,
	expand_GOMP_SIMD_ORDERED_END): New functions.
	* internal-fn.def (GOMP_SIMD_ORDERED_START,
	GOMP_SIMD_ORDERED_END): New internal functions.
	* omp-builtins.def (BUILT_IN_GOMP_LOOP_DOACROSS_STATIC_START,
	BUILT_IN_GOMP_LOOP_DOACROSS_DYNAMIC_START,
	BUILT_IN_GOMP_LOOP_DOACROSS_GUIDED_START,
	BUILT_IN_GOMP_LOOP_DOACROSS_RUNTIME_START,
	BUILT_IN_GOMP_LOOP_ULL_DOACROSS_STATIC_START,
	BUILT_IN_GOMP_LOOP_ULL_DOACROSS_DYNAMIC_START,
	BUILT_IN_GOMP_LOOP_ULL_DOACROSS_GUIDED_START,
	BUILT_IN_GOMP_LOOP_ULL_DOACROSS_RUNTIME_START,
	BUILT_IN_GOMP_DOACROSS_POST, BUILT_IN_GOMP_DOACROSS_WAIT,
	BUILT_IN_GOMP_DOACROSS_ULL_POST, BUILT_IN_GOMP_DOACROSS_ULL_WAIT,
	BUILT_IN_GOMP_TARGET_ENTER_EXIT_DATA, BUILT_IN_GOMP_TASKLOOP,
	BUILT_IN_GOMP_TASKLOOP_ULL): New built-ins.
	(BUILT_IN_GOMP_TASK): Add INT argument to the end.
	(BUILT_IN_GOMP_TARGET): Rename from GOMP_target to GOMP_target_41,
	adjust type.
	(BUILT_IN_GOMP_TARGET_DATA): Rename from GOMP_target_data to
	GOMP_target_data_41, adjust type.
	(BUILT_IN_GOMP_TARGET_UPDATE): Rename from GOMP_target_update to
	GOMP_target_update_41, adjust type.
	* omp-low.c (struct omp_region): Adjust comments, add ord_stmt
	field.
	(struct omp_for_data): Add ordered and simd_schedule fields.
	(omp_member_access_dummy_var, unshare_and_remap_1,
	unshare_and_remap, is_taskloop_ctx): New functions.
	(is_taskreg_ctx): Use is_parallel_ctx and is_task_ctx.
	(extract_omp_for_data): Handle taskloops and doacross loops
	and simd schedule modifier.
	(omp_adjust_chunk_size): New function.
	(get_ws_args_for): Use it.
	(lookup_sfield): Change first argument to splay_tree_key,
	add overload with first argument tree.
	(maybe_lookup_field): Likewise.
	(use_pointer_for_field): Handle omp_member_access_dummy_var.
	(omp_copy_decl_2): If var is TREE_ADDRESSABLE listed in
	task_shared_vars, clear TREE_ADDRESSABLE on the copy.
	(build_outer_var_ref): Add LASTPRIVATE argument, handle
	taskloops and omp_member_access_dummy_var vars.
	(build_sender_ref): Change first argument to splay_tree_key,
	add overload with first argument tree.
	(install_var_field): For mask & 8 use &DECL_UID as key instead
	of the tree itself.
	(fixup_child_record_type): Const qualify *.omp_data_i.
	(scan_sharing_clauses): Handle OMP_CLAUSE_SHARED_FIRSTPRIVATE,
	C/C++ array reductions, OMP_CLAUSE_{IS,USE}_DEVICE_PTR clauses,
	OMP_CLAUSE_{PRIORITY,GRAINSIZE,NUM_TASKS,SIMDLEN,THREADS,SIMD} and
	OMP_CLAUSE_{NOGROUP,DEFAULTMAP} clauses, OMP_CLAUSE__LOOPTEMP_ clause
	on taskloop, GOMP_MAP_FIRSTPRIVATE_POINTER, OMP_CLAUSE_MAP_PRIVATE.
	(create_omp_child_function): Set TREE_READONLY on .omp_data_i.
	(find_combined_for): Allow searching for different GIMPLE_OMP_FOR
	kinds.
	(add_taskreg_looptemp_clauses): New function.
	(scan_omp_parallel): Use it.
	(scan_omp_task): Likewise.
	(finish_taskreg_scan): Handle OMP_CLAUSE_SHARED_FIRSTPRIVATE.
	For taskloop, move fields for the first two _LOOPTEMP_ clauses first.
	(check_omp_nesting_restrictions): Handle GF_OMP_TARGET_KIND_ENTER_DATA
	and GF_OMP_TARGET_KIND_EXIT_DATA.  Formatting fixes.  Allow the
	sandwiched taskloop constructs.  Type check
	OMP_CLAUSE_DEPEND_{KIND,SOURCE}.  Allow ordered simd inside of simd
	region.  Diagnose depend(source) or depend(sink:...) on
	target constructs or task/taskloop.
	(handle_simd_reference): Use get_name.
	(lower_rec_input_clauses): Likewise.  Ignore all
	OMP_CLAUSE_LASTPRIVATE_FIRSTPRIVATE clauses on taskloop construct.
	Allow _LOOPTEMP_ clause on GOMP_TASK.  Unshare new_var
	before passing it to omp_clause_{default,copy}_ctor.  Handle
	OMP_CLAUSE_REDUCTION with MEM_REF OMP_CLAUSE_DECL.  Set
	lastprivate_firstprivate flag for linear that needs copyin and
	copyout.  Use BUILT_IN_ALLOCA_WITH_ALIGN instead of BUILT_IN_ALLOCA.
	(lower_lastprivate_clauses): For OMP_CLAUSE_LASTPRIVATE_FIRSTPRIVATE
	on taskloop lookup decl in outer context.  Pass true to
	build_outer_var_ref lastprivate argument.  Handle
	OMP_CLAUSE_LASTPRIVATE_TASKLOOP_IV lastprivate if the decl is global
	outside of outer taskloop for.
	(lower_reduction_clauses): Handle OMP_CLAUSE_REDUCTION with MEM_REF
	OMP_CLAUSE_DECL.
	(lower_send_clauses): Ignore first two _LOOPTEMP_ clauses in taskloop
	GOMP_TASK.  Handle OMP_CLAUSE_SHARED_FIRSTPRIVATE.  Handle
	omp_member_access_dummy_var vars.  Handle OMP_CLAUSE_REDUCTION
	with MEM_REF OMP_CLAUSE_DECL.  Use new lookup_sfield overload.
	(lower_send_shared_vars): Ignore fields with NULL or FIELD_DECL
	abstract origin.  Handle omp_member_access_dummy_var vars.
	(expand_parallel_call): Use expand_omp_build_assign.
	(expand_task_call): Handle taskloop construct expansion.  Add
	REGION argument.  Use GOMP_TASK_* defines instead of hardcoded
	integers.  Add priority argument to GOMP_task* calls.  Or in
	GOMP_TASK_FLAG_PRIORITY into flags if priority is present for
	GOMP_task call.
	(expand_omp_build_assign): Add prototype.  Add AFTER
	argument, if true emit statements after *GSI_P and continue linking.
	(expand_omp_taskreg): Adjust expand_task_call caller.
	(expand_omp_for_init_counts): Rename zero_iter_bb argument to
	zero_iter1_bb and first_zero_iter to first_zero_iter1, add
	zero_iter2_bb and first_zero_iter2 arguments, handle computation
	of counts even for ordered loops.
	(expand_omp_for_init_vars): Handle GOMP_TASK inner_stmt.
	(expand_omp_ordered_source, expand_omp_ordered_sink,
	expand_omp_ordered_source_sink, expand_omp_for_ordered_loops): New
	functions.
	(expand_omp_for_generic): Use omp_adjust_chunk_size.  Handle linear
	clauses on worksharing loop.  Handle DOACROSS loop expansion.
	(expand_omp_for_static_nochunk): Handle linear clauses on
	worksharing loop.  Adjust expand_omp_for_init_counts
	callers.
	(expand_omp_for_static_chunk): Likewise.  Use omp_adjust_chunk_size.
	(expand_omp_simd): Handle addressable fd->loop.v.  Adjust
	expand_omp_for_init_counts callers.
	(expand_omp_taskloop_for_outer, expand_omp_taskloop_for_inner): New
	functions.
	(expand_omp_for): Call expand_omp_taskloop_for_* for taskloop.
	Handle doacross loops.
	(expand_omp_target): Handle GF_OMP_TARGET_KIND_ENTER_DATA and
	GF_OMP_TARGET_KIND_EXIT_DATA.  Pass flags and depend arguments to
	GOMP_target_{41,update_41,enter_exit_data} libcalls.
	(expand_omp): Don't expand ordered depend constructs here, record
	ord_stmt instead for later expand_omp_for_generic.
	(build_omp_regions_1): Handle GF_OMP_TARGET_KIND_ENTER_DATA and
	GF_OMP_TARGET_KIND_EXIT_DATA.  Treat GIMPLE_OMP_ORDERED with depend
	clause as stand-alone directive.
	(lower_omp_ordered_clauses): New function.
	(lower_omp_ordered): Handle OMP_CLAUSE_SIMD, for OMP_CLAUSE_DEPEND
	don't lower anything.
	(lower_omp_for_lastprivate): Use last _looptemp_ clause
	on taskloop for comparison.
	(lower_omp_for): Handle taskloop constructs.  Adjust OMP_CLAUSE_DECL
	and OMP_CLAUSE_LINEAR_STEP so that expand_omp_for_* can use it during
	expansion for linear adjustments.
	(create_task_copyfn): Handle OMP_CLAUSE_SHARED_FIRSTPRIVATE.
	(lower_depend_clauses): Assert not seeing sink/source depend kinds.
	Set TREE_ADDRESSABLE on array.  Change first argument from gimple *
	to tree * pointing to the stmt's clauses.
	(lower_omp_taskreg): Adjust lower_depend_clauses caller.
	(lower_omp_target): Handle GF_OMP_TARGET_KIND_ENTER_DATA
	and GF_OMP_TARGET_KIND_EXIT_DATA, depend clauses,
	GOMP_MAP_{RELEASE,ALWAYS_{TO,FROM,TOFROM},FIRSTPRIVATE_POINTER,STRUCT}
	map kinds, OMP_CLAUSE_{FIRSTPRIVATE,PRIVATE,{IS,USE}_DEVICE_PTR
	clauses.  Always use short kind and 8-bit align shift.
	(lower_omp_regimplify_p): Use IS_TYPE_OR_DECL_P macro.
	(struct lower_omp_regimplify_operands_data): New type.
	(lower_omp_regimplify_operands_p, lower_omp_regimplify_operands):
	New functions.
	(lower_omp_1): Use lower_omp_regimplify_operands instead of
	gimple_regimplify_operands.
	(make_gimple_omp_edges): Handle GF_OMP_TARGET_KIND_ENTER_DATA and
	GF_OMP_TARGET_KIND_EXIT_DATA.  Treat GIMPLE_OMP_ORDERED with depend
	clause as stand-alone directive.
	(simd_clone_clauses_extract): Honor OMP_CLAUSE_LINEAR_KIND.
	(simd_clone_mangle): Mangle the various linear kinds
	per the new ABI.
	(simd_clone_adjust_argument_types): Handle
	SIMD_CLONE_ARG_TYPE_LINEAR_*_CONSTANT_STEP.
	(simd_clone_init_simd_arrays): Don't do anything for uval.
	(simd_clone_adjust): Handle
	SIMD_CLONE_ARG_TYPE_LINEAR_REF_CONSTANT_STEP like
	SIMD_CLONE_ARG_TYPE_LINEAR_CONSTANT_STEP.
	Handle SIMD_CLONE_ARG_TYPE_LINEAR_UVAL_CONSTANT_STEP.
	* omp-low.h (omp_member_access_dummy_var): New prototype.
	* passes.def (pass_simduid_cleanup): Schedule another copy of the
	pass after all optimizations.
	* tree.c (omp_clause_code_name): Add entries for
	OMP_CLAUSE_{TO_DECLARE,LINK,{USE,IS}_DEVICE_PTR,DEFAULTMAP,HINT}
	and OMP_CLAUSE_{PRIORITY,GRAINSIZE,NUM_TASKS,NOGROUP,THREADS,SIMD}.
	(omp_clause_num_ops): Likewise.  Bump number of OMP_CLAUSE_REDUCTION
	arguments to 5 and for OMP_CLAUSE_ORDERED to 1.
	(walk_tree_1): Adjust for OMP_CLAUSE_ORDERED having 1 argument and
	OMP_CLAUSE_REDUCTION 5 arguments.  Handle
	OMP_CLAUSE_{TO_DECLARE,LINK,{USE,IS}_DEVICE_PTR,DEFAULTMAP,HINT}
	and OMP_CLAUSE_{PRIORITY,GRAINSIZE,NUM_TASKS,NOGROUP,THREADS,SIMD}
	clauses.
	* tree-core.h (enum omp_clause_linear_kind): New.
	(struct tree_omp_clause): Change type of map_kind
	from unsigned char to unsigned int.  Add subcode.if_modifier
	and subcode.linear_kind fields.
	(enum omp_clause_code): Add
	OMP_CLAUSE_{TO_DECLARE,LINK,{USE,IS}_DEVICE_PTR,DEFAULTMAP,HINT}
	and OMP_CLAUSE_{PRIORITY,GRAINSIZE,NUM_TASKS,NOGROUP,THREADS,SIMD}.
	(OMP_CLAUSE_REDUCTION): Document
	OMP_CLAUSE_REDUCTION_DECL_PLACEHOLDER.
	(enum omp_clause_depend_kind): Add OMP_CLAUSE_DEPEND_{SOURCE,SINK}.
	* tree.def (OMP_FOR): Add OMP_FOR_ORIG_DECLS operand.
	(OMP_CRITICAL): Move before OMP_SINGLE.  Add OMP_CRITICAL_CLAUSES
	operand.
	(OMP_ORDERED): Move before OMP_SINGLE.  Add OMP_ORDERED_CLAUSES
	operand.
	(OMP_TASKLOOP, OMP_TARGET_ENTER_DATA, OMP_TARGET_EXIT_DATA): New tree
	codes.
	* tree.h (OMP_BODY): Replace OMP_CRITICAL with OMP_TASKGROUP.
	(OMP_CLAUSE_SET_MAP_KIND): Cast to unsigned int rather than unsigned
	char.
	(OMP_CRITICAL_NAME): Adjust to be 3rd operand instead of 2nd.
	(OMP_CLAUSE_NUM_TASKS_EXPR): Formatting fix.
	(OMP_STANDALONE_CLAUSES): Adjust to cover OMP_TARGET_{ENTER,EXIT}_DATA.
	(OMP_CLAUSE_DEPEND_SINK_NEGATIVE, OMP_TARGET_COMBINED,
	OMP_CLAUSE_MAP_PRIVATE, OMP_FOR_ORIG_DECLS, OMP_CLAUSE_IF_MODIFIER,
	OMP_CLAUSE_MAP_MAYBE_ZERO_LENGTH_ARRAY_SECTION, OMP_CRITICAL_CLAUSES,
	OMP_CLAUSE_PRIVATE_TASKLOOP_IV, OMP_CLAUSE_LASTPRIVATE_TASKLOOP_IV,
	OMP_CLAUSE_HINT_EXPR, OMP_CLAUSE_SCHEDULE_SIMD,
	OMP_CLAUSE_LINEAR_KIND, OMP_CLAUSE_REDUCTION_DECL_PLACEHOLDER,
	OMP_CLAUSE_SHARED_FIRSTPRIVATE, OMP_ORDERED_CLAUSES,
	OMP_TARGET_ENTER_DATA_CLAUSES, OMP_TARGET_EXIT_DATA_CLAUSES,
	OMP_CLAUSE_NUM_TASKS_EXPR, OMP_CLAUSE_GRAINSIZE_EXPR,
	OMP_CLAUSE_PRIORITY_EXPR, OMP_CLAUSE_ORDERED_EXPR): Define.
	* tree-inline.c (remap_gimple_stmt): Handle clauses on
	GIMPLE_OMP_ORDERED and GIMPLE_OMP_CRITICAL.  For
	IFN_GOMP_SIMD_ORDERED_{START,END} set has_simduid_loops.
	* tree-nested.c (convert_nonlocal_omp_clauses): Handle
	OMP_CLAUSE_{TO_DECLARE,LINK,{USE,IS}_DEVICE_PTR,SIMDLEN,PRIORITY,SIMD}
	and OMP_CLAUSE_{GRAINSIZE,NUM_TASKS,HINT,NOGROUP,THREADS,DEFAULTMAP}
	clauses.  Handle OMP_CLAUSE_REDUCTION_DECL_PLACEHOLDER.
	(convert_local_omp_clauses): Likewise.
	* tree-pretty-print.c (dump_omp_clause): Handle
	OMP_CLAUSE_{TO_DECLARE,LINK,{USE,IS}_DEVICE_PTR,SIMDLEN,PRIORITY,SIMD}
	and OMP_CLAUSE_{GRAINSIZE,NUM_TASKS,HINT,NOGROUP,THREADS,DEFAULTMAP}
	clauses.  Handle OMP_CLAUSE_IF_MODIFIER, OMP_CLAUSE_ORDERED_EXPR,
	OMP_CLAUSE_SCHEDULE_SIMD, OMP_CLAUSE_LINEAR_KIND,
	OMP_CLAUSE_DEPEND_{SOURCE,SINK}.  Use "delete" for
	GOMP_MAP_FORCE_DEALLOC.  Handle
	GOMP_MAP_{ALWAYS_{TO,FROM,TOFROM},RELEASE,FIRSTPRIVATE_POINTER,STRUCT}.
	(dump_generic_node): Handle OMP_TASKLOOP, OMP_TARGET_{ENTER,EXIT}_DATA
	and clauses on OMP_ORDERED and OMP_CRITICAL.
	* tree-vectorizer.c (adjust_simduid_builtins): Adjust comment.
	Remove IFN_GOMP_SIMD_ORDERED_{START,END}.
	(vectorize_loops): Adjust comments.
	(pass_simduid_cleanup::execute): Likewise.
	* tree-vect-stmts.c (vectorizable_simd_clone_call): Handle
	SIMD_CLONE_ARG_TYPE_LINEAR_{REF,VAL,UVAL}_CONSTANT_STEP.
	* wide-int.h (wi::gcd): New.
gcc/c-family/
2015-10-13  Jakub Jelinek  <jakub@redhat.com>
	    Aldy Hernandez  <aldyh@redhat.com>

	* c-common.c (enum c_builtin_type): Define DEF_FUNCTION_TYPE_9,
	DEF_FUNCTION_TYPE_10 and DEF_FUNCTION_TYPE_11.
	(c_define_builtins): Likewise.
	* c-common.h (enum c_omp_clause_split): Add
	C_OMP_CLAUSE_SPLIT_TASKLOOP.
	(c_finish_omp_critical, c_finish_omp_ordered): Add CLAUSES argument.
	(c_finish_omp_for): Add ORIG_DECLV argument.
	* c-cppbuiltin.c (c_cpp_builtins): Predefine _OPENMP as
	201511 instead of 201307.
	* c-omp.c (c_finish_omp_critical): Add CLAUSES argument, set
	OMP_CRITICAL_CLAUSES to it.
	(c_finish_omp_ordered): Add CLAUSES argument, set
	OMP_ORDERED_CLAUSES to it.
	(c_finish_omp_for): Add ORIG_DECLV argument, set OMP_FOR_ORIG_DECLS
	to it if OMP_FOR.  Clear DECL_INITIAL on the IVs.
	(c_omp_split_clauses): Handle OpenMP 4.5 combined/composite
	constructs and new OpenMP 4.5 clauses.  Clear
	OMP_CLAUSE_SCHEDULE_SIMD if not combined with OMP_SIMD.  Add
	verification code.
	* c-pragma.c (omp_pragmas_simd): Add taskloop.
	* c-pragma.h (enum pragma_kind): Add PRAGMA_OMP_TASKLOOP.
	(enum pragma_omp_clause): Add
	PRAGMA_OMP_CLAUSE_{DEFAULTMAP,GRAINSIZE,HINT,{IS,USE}_DEVICE_PTR}
	and PRAGMA_OMP_CLAUSE_{LINK,NOGROUP,NUM_TASKS,PRIORITY,SIMD,THREADS}.
gcc/c/
2015-10-13  Jakub Jelinek  <jakub@redhat.com>
	    Aldy Hernandez  <aldyh@redhat.com>

	* c-parser.c (c_parser_pragma): Handle PRAGMA_OMP_ORDERED here.
	(c_parser_omp_clause_name): Handle OpenMP 4.5 clauses.
	(c_parser_omp_variable_list): Handle structure elements for
	map, to and from clauses.  Handle array sections in reduction
	clause.  Formatting fixes.
	(c_parser_omp_clause_if): Add IS_OMP argument, handle parsing of
	if clause modifiers.
	(c_parser_omp_clause_num_tasks, c_parser_omp_clause_grainsize,
	c_parser_omp_clause_priority, c_parser_omp_clause_hint,
	c_parser_omp_clause_defaultmap, c_parser_omp_clause_use_device_ptr,
	c_parser_omp_clause_is_device_ptr): New functions.
	(c_parser_omp_clause_ordered): Parse optional parameter.
	(c_parser_omp_clause_reduction): Handle array reductions.
	(c_parser_omp_clause_schedule): Parse optional simd modifier.
	(c_parser_omp_clause_nogroup, c_parser_omp_clause_orderedkind): New
	functions.
	(c_parser_omp_clause_linear): Parse linear clause modifiers.
	(c_parser_omp_clause_depend_sink): New function.
	(c_parser_omp_clause_depend): Parse source/sink depend kinds.
	(c_parser_omp_clause_map): Parse release/delete map kinds and
	optional always modifier.
	(c_parser_oacc_all_clauses): Adjust c_parser_omp_clause_if
	and c_finish_omp_clauses callers.
	(c_parser_omp_all_clauses): Likewise.  Parse OpenMP 4.5 clauses.
	Parse "to" as OMP_CLAUSE_TO_DECLARE if on declare target directive.
	(c_parser_oacc_cache): Adjust c_finish_omp_clauses caller.
	(OMP_CRITICAL_CLAUSE_MASK): Define.
	(c_parser_omp_critical): Parse critical clauses.
	(c_parser_omp_for_loop): Handle doacross loops, adjust
	c_finish_omp_for and c_finish_omp_clauses callers.
	(OMP_SIMD_CLAUSE_MASK): Add simdlen clause.
	(c_parser_omp_simd): Allow ordered clause if it has no parameter.
	(OMP_FOR_CLAUSE_MASK): Add linear clause.
	(c_parser_omp_for): Disallow ordered clause when combined with
	distribute.  Disallow linear clause when combined with distribute
	and not combined with simd.
	(OMP_ORDERED_CLAUSE_MASK, OMP_ORDERED_DEPEND_CLAUSE_MASK): Define.
	(c_parser_omp_ordered): Add CONTEXT argument, remove LOC argument,
	parse clauses and if depend clause is found, don't parse a body.
	(c_parser_omp_parallel): Disallow copyin clause on target parallel.
	Allow target parallel without for after it.
	(OMP_TASK_CLAUSE_MASK): Add priority clause.
	(OMP_TARGET_DATA_CLAUSE_MASK): Add use_device_ptr clause.
	(c_parser_omp_target_data): Diagnose no map clauses or clauses with
	invalid kinds.
	(OMP_TARGET_UPDATE_CLAUSE_MASK): Add depend and nowait clauses.
	(OMP_TARGET_ENTER_DATA_CLAUSE_MASK,
	OMP_TARGET_EXIT_DATA_CLAUSE_MASK): Define.
	(c_parser_omp_target_enter_data, c_parser_omp_target_exit_data): New
	functions.
	(OMP_TARGET_CLAUSE_MASK): Add depend, nowait, private, firstprivate,
	defaultmap and is_device_ptr clauses.
	(c_parser_omp_target): Parse target parallel and target simd.  Set
	OMP_TARGET_COMBINED on combined constructs.  Parse target enter data
	and target exit data.  Diagnose invalid map kinds.
	(OMP_DECLARE_TARGET_CLAUSE_MASK): Define.
	(c_parser_omp_declare_target): Parse OpenMP 4.5 forms of this
	construct.
	(c_parser_omp_declare_reduction): Use STRIP_NOPS when checking for
	&omp_priv.
	(OMP_TASKLOOP_CLAUSE_MASK): Define.
	(c_parser_omp_taskloop): New function.
	(c_parser_omp_construct): Don't handle PRAGMA_OMP_ORDERED here,
	handle PRAGMA_OMP_TASKLOOP.
	(c_parser_cilk_for): Adjust c_finish_omp_clauses callers.
	* c-tree.h (c_finish_omp_clauses): Add two new arguments.
	* c-typeck.c (handle_omp_array_sections_1): Fix comment typo.
	Add IS_OMP argument, handle structure element bases, diagnose
	bitfields, pass IS_OMP recursively, diagnose known zero length
	array sections in depend clauses, handle array sections in reduction
	clause, diagnose negative length even for pointers.
	(handle_omp_array_sections): Add IS_OMP argument, use auto_vec for
	types, pass IS_OMP down to handle_omp_array_sections_1, handle
	array sections in reduction clause, set
	OMP_CLAUSE_MAP_MAYBE_ZERO_LENGTH_ARRAY_SECTION if map could be zero
	length array section, use GOMP_MAP_FIRSTPRIVATE_POINTER for IS_OMP.
	(c_finish_omp_clauses): Add IS_OMP and DECLARE_SIMD arguments.
	Handle new OpenMP 4.5 clauses and new restrictions for the old ones.
gcc/cp/
2015-10-13  Jakub Jelinek  <jakub@redhat.com>
	    Aldy Hernandez  <aldyh@redhat.com>

	* class.c (finish_struct_1): Call finish_omp_declare_simd_methods.
	* cp-gimplify.c (cp_gimplify_expr): Handle OMP_TASKLOOP.
	(cp_genericize_r): Likewise.
	(cxx_omp_finish_clause): Don't diagnose references.
	(cxx_omp_disregard_value_expr): New function.
	* cp-objcp-common.h (LANG_HOOKS_OMP_DISREGARD_VALUE_EXPR): Redefine.
	* cp-tree.h (OMP_FOR_GIMPLIFYING_P): Document for OMP_TASKLOOP.
	(DECL_OMP_PRIVATIZED_MEMBER): Define.
	(finish_omp_declare_simd_methods, push_omp_privatization_clauses,
	pop_omp_privatization_clauses, save_omp_privatization_clauses,
	restore_omp_privatization_clauses, omp_privatize_field,
	cxx_omp_disregard_value_expr): New prototypes.
	(finish_omp_clauses): Add two new arguments.
	(finish_omp_for): Add ORIG_DECLV argument.
	* parser.c (cp_parser_lambda_body): Call
	save_omp_privatization_clauses and restore_omp_privatization_clauses.
	(cp_parser_omp_clause_name): Handle OpenMP 4.5 clauses.
	(cp_parser_omp_var_list_no_open): Handle structure elements for
	map, to and from clauses.  Handle array sections in reduction
	clause.  Parse this keyword.  Formatting fixes.
	(cp_parser_omp_clause_if): Add IS_OMP argument, handle parsing of
	if clause modifiers.
	(cp_parser_omp_clause_num_tasks, cp_parser_omp_clause_grainsize,
	cp_parser_omp_clause_priority, cp_parser_omp_clause_hint,
	cp_parser_omp_clause_defaultmap): New functions.
	(cp_parser_omp_clause_ordered): Parse optional parameter.
	(cp_parser_omp_clause_reduction): Handle array reductions.
	(cp_parser_omp_clause_schedule): Parse optional simd modifier.
	(cp_parser_omp_clause_nogroup, cp_parser_omp_clause_orderedkind):
	New functions.
	(cp_parser_omp_clause_linear): Parse linear clause modifiers.
	(cp_parser_omp_clause_depend_sink): New function.
	(cp_parser_omp_clause_depend): Parse source/sink depend kinds.
	(cp_parser_omp_clause_map): Parse release/delete map kinds and
	optional always modifier.
	(cp_parser_oacc_all_clauses): Adjust cp_parser_omp_clause_if
	and finish_omp_clauses callers.
	(cp_parser_omp_all_clauses): Likewise.  Parse OpenMP 4.5 clauses.
	Parse "to" as OMP_CLAUSE_TO_DECLARE if on declare target directive.
	(OMP_CRITICAL_CLAUSE_MASK): Define.
	(cp_parser_omp_critical): Parse critical clauses.
	(cp_parser_omp_for_incr): Use cp_tree_equal if
	processing_template_decl.
	(cp_parser_omp_for_loop_init): Return tree instead of bool.  Handle
	non-static data member iterators.
	(cp_parser_omp_for_loop): Handle doacross loops, adjust
	finish_omp_for and finish_omp_clauses callers.
	(cp_omp_split_clauses): Adjust finish_omp_clauses caller.
	(OMP_SIMD_CLAUSE_MASK): Add simdlen clause.
	(cp_parser_omp_simd): Allow ordered clause if it has no parameter.
	(OMP_FOR_CLAUSE_MASK): Add linear clause.
	(cp_parser_omp_for): Disallow ordered clause when combined with
	distribute.  Disallow linear clause when combined with distribute
	and not combined with simd.
	(OMP_ORDERED_CLAUSE_MASK, OMP_ORDERED_DEPEND_CLAUSE_MASK): Define.
	(cp_parser_omp_ordered): Add CONTEXT argument, return bool instead
	of tree, parse clauses and if depend clause is found, don't parse
	a body.
	(cp_parser_omp_parallel): Disallow copyin clause on target parallel.
	Allow target parallel without for after it.
	(OMP_TASK_CLAUSE_MASK): Add priority clause.
	(OMP_TARGET_DATA_CLAUSE_MASK): Add use_device_ptr clause.
	(cp_parser_omp_target_data): Diagnose no map clauses or clauses with
	invalid kinds.
	(OMP_TARGET_UPDATE_CLAUSE_MASK): Add depend and nowait clauses.
	(OMP_TARGET_ENTER_DATA_CLAUSE_MASK,
	OMP_TARGET_EXIT_DATA_CLAUSE_MASK): Define.
	(cp_parser_omp_target_enter_data, cp_parser_omp_target_exit_data): New
	functions.
	(OMP_TARGET_CLAUSE_MASK): Add depend, nowait, private, firstprivate,
	defaultmap and is_device_ptr clauses.
	(cp_parser_omp_target): Parse target parallel and target simd.  Set
	OMP_TARGET_COMBINED on combined constructs.  Parse target enter data
	and target exit data.  Diagnose invalid map kinds.
	(cp_parser_oacc_cache): Adjust finish_omp_clauses caller.
	(OMP_DECLARE_TARGET_CLAUSE_MASK): Define.
	(cp_parser_omp_declare_target): Parse OpenMP 4.5 forms of this
	construct.
	(OMP_TASKLOOP_CLAUSE_MASK): Define.
	(cp_parser_omp_taskloop): New function.
	(cp_parser_omp_construct): Don't handle PRAGMA_OMP_ORDERED here,
	handle PRAGMA_OMP_TASKLOOP.
	(cp_parser_pragma): Handle PRAGMA_OMP_ORDERED here directly,
	handle PRAGMA_OMP_TASKLOOP, call push_omp_privatization_clauses
	and pop_omp_privatization_clauses around parsing calls.
	(cp_parser_cilk_for): Adjust finish_omp_clauses caller.
	* pt.c (apply_late_template_attributes): Adjust tsubst_omp_clauses
	and finish_omp_clauses callers.
	(tsubst_omp_clause_decl): Return NULL if decl is NULL.
	For TREE_LIST, copy over OMP_CLAUSE_DEPEND_SINK_NEGATIVE bit.
	Use tsubst_expr instead of tsubst_copy, undo convert_from_reference
	effects.
	(tsubst_omp_clauses): Add ALLOW_FIELDS argument.  Handle new
	OpenMP 4.5 clauses.  Use tsubst_omp_clause_decl for more clauses.
	If ALLOW_FIELDS, handle non-static data members in the clauses.
	Clear OMP_CLAUSE_LINEAR_STEP if it has been cleared before.
	(omp_parallel_combined_clauses): New variable.
	(tsubst_omp_for_iterator): Add ORIG_DECLV argument, recur on
	OMP_FOR_ORIG_DECLS, handle non-static data member iterators.
	Improve handling of clauses on combined constructs.
	(tsubst_expr): Call push_omp_privatization_clauses and
	pop_omp_privatization_clauses around instantiation of certain
	OpenMP constructs, improve handling of clauses on combined
	constructs, handle OMP_TASKLOOP, adjust tsubst_omp_for_iterator,
	tsubst_omp_clauses and finish_omp_for callers, handle clauses on
	critical and ordered, handle OMP_TARGET_{ENTER,EXIT}_DATA.
	(instantiate_decl): Call save_omp_privatization_clauses and
	restore_omp_privatization_clauses around instantiation.
	(dependent_omp_for_p): Fix up comment typo.  Handle SCOPE_REF.
	* semantics.c (omp_private_member_map, omp_private_member_vec,
	omp_private_member_ignore_next): New variables.
	(finish_non_static_data_member): Return dummy decl for privatized
	non-static data members.
	(omp_clause_decl_field, omp_clause_printable_decl,
	omp_note_field_privatization, omp_privatize_field): New functions.
	(handle_omp_array_sections_1): Fix comment typo.
	Add IS_OMP argument, handle structure element bases, diagnose
	bitfields, pass IS_OMP recursively, diagnose known zero length
	array sections in depend clauses, handle array sections in reduction
	clause, diagnose negative length even for pointers.
	(handle_omp_array_sections): Add IS_OMP argument, use auto_vec for
	types, pass IS_OMP down to handle_omp_array_sections_1, handle
	array sections in reduction clause, set
	OMP_CLAUSE_MAP_MAYBE_ZERO_LENGTH_ARRAY_SECTION if map could be zero
	length array section, use GOMP_MAP_FIRSTPRIVATE_POINTER for IS_OMP.
	(finish_omp_reduction_clause): Handle array sections and arrays.
	Use omp_clause_printable_decl.
	(finish_omp_declare_simd_methods, cp_finish_omp_clause_depend_sink):
	New functions.
	(finish_omp_clauses): Add ALLOW_FIELDS and DECLARE_SIMD arguments.
	Handle new OpenMP 4.5 clauses and new restrictions for the old
	ones, handle non-static data members, reject this keyword when not
	allowed.
	(push_omp_privatization_clauses, pop_omp_privatization_clauses,
	save_omp_privatization_clauses, restore_omp_privatization_clauses):
	New functions.
	(handle_omp_for_class_iterator): Handle OMP_TASKLOOP class iterators.
	Add collapse and ordered arguments.  Fix handling of lastprivate
	iterators in doacross loops.
	(finish_omp_for): Add ORIG_DECLV argument, handle doacross loops,
	adjust c_finish_omp_for, handle_omp_for_class_iterator and
	finish_omp_clauses callers.  Fill in OMP_CLAUSE_LINEAR_STEP on simd
	loops with non-static data member iterators.
gcc/fortran/
2015-10-13  Jakub Jelinek  <jakub@redhat.com>
	    Ilya Verbin  <ilya.verbin@intel.com>

	* f95-lang.c (DEF_FUNCTION_TYPE_9, DEF_FUNCTION_TYPE_10,
	DEF_FUNCTION_TYPE_11, DEF_FUNCTION_TYPE_VAR_1): Define.
	* trans-openmp.c (gfc_trans_omp_clauses): Set
	OMP_CLAUSE_IF_MODIFIER to ERROR_MARK, OMP_CLAUSE_ORDERED_EXPR
	to NULL.
	(gfc_trans_omp_critical): Adjust for addition of clauses.
	(gfc_trans_omp_ordered): Likewise.
	* types.def (BT_FN_BOOL_UINT_LONGPTR_LONGPTR_LONGPTR,
	BT_FN_BOOL_UINT_ULLPTR_ULLPTR_ULLPTR,
	BT_FN_BOOL_UINT_LONGPTR_LONG_LONGPTR_LONGPTR,
	BT_FN_BOOL_UINT_ULLPTR_ULL_ULLPTR_ULLPTR,
	BT_FN_VOID_INT_SIZE_PTR_PTR_PTR_UINT_PTR,
	BT_FN_VOID_INT_OMPFN_SIZE_PTR_PTR_PTR_UINT_PTR,
	BT_FN_VOID_OMPFN_PTR_OMPCPYFN_LONG_LONG_BOOL_UINT_PTR_INT,
	BT_FN_VOID_OMPFN_PTR_OMPCPYFN_LONG_LONG_UINT_LONG_INT_LONG_LONG_LONG,
	BT_FN_VOID_OMPFN_PTR_OMPCPYFN_LONG_LONG_UINT_LONG_INT_ULL_ULL_ULL,
	BT_FN_VOID_LONG_VAR, BT_FN_VOID_ULL_VAR): New.
	(BT_FN_VOID_INT_PTR_SIZE_PTR_PTR_PTR,
	BT_FN_VOID_INT_OMPFN_PTR_SIZE_PTR_PTR_PTR,
	BT_FN_VOID_OMPFN_PTR_OMPCPYFN_LONG_LONG_BOOL_UINT_PTR): Remove.
gcc/lto/
2015-10-13  Jakub Jelinek  <jakub@redhat.com>

	* lto-lang.c (DEF_FUNCTION_TYPE_9, DEF_FUNCTION_TYPE_10,
	DEF_FUNCTION_TYPE_11): Define.
gcc/jit/
2015-10-13  Jakub Jelinek  <jakub@redhat.com>

	* jit-builtins.c (DEF_FUNCTION_TYPE_9, DEF_FUNCTION_TYPE_10,
	DEF_FUNCTION_TYPE_11): Define.
	* jit-builtins.h (DEF_FUNCTION_TYPE_9, DEF_FUNCTION_TYPE_10,
	DEF_FUNCTION_TYPE_11): Define.
gcc/ada/
2015-10-13  Jakub Jelinek  <jakub@redhat.com>

	* gcc-interface/utils.c (DEF_FUNCTION_TYPE_9, DEF_FUNCTION_TYPE_10,
	DEF_FUNCTION_TYPE_11): Define.
gcc/testsuite/
2015-10-13  Jakub Jelinek  <jakub@redhat.com>
	    Aldy Hernandez  <aldyh@redhat.com>

	* c-c++-common/gomp/cancel-1.c (f2): Add map clause to target data.
	* c-c++-common/gomp/clauses-1.c: New test.
	* c-c++-common/gomp/clauses-2.c: New test.
	* c-c++-common/gomp/clauses-3.c: New test.
	* c-c++-common/gomp/clauses-4.c: New test.
	* c-c++-common/gomp/declare-target-1.c: New test.
	* c-c++-common/gomp/declare-target-2.c: New test.
	* c-c++-common/gomp/depend-3.c: New test.
	* c-c++-common/gomp/depend-4.c: New test.
	* c-c++-common/gomp/doacross-1.c: New test.
	* c-c++-common/gomp/if-1.c: New test.
	* c-c++-common/gomp/if-2.c: New test.
	* c-c++-common/gomp/linear-1.c: New test.
	* c-c++-common/gomp/map-2.c: New test.
	* c-c++-common/gomp/map-3.c: New test.
	* c-c++-common/gomp/nesting-1.c (f_omp_parallel,
	f_omp_target_data): Add map clause to target data.
	* c-c++-common/gomp/nesting-warn-1.c (f_omp_target): Likewise.
	* c-c++-common/gomp/ordered-1.c: New test.
	* c-c++-common/gomp/ordered-2.c: New test.
	* c-c++-common/gomp/ordered-3.c: New test.
	* c-c++-common/gomp/pr61486-1.c (foo): Remove linear clause
	on non-iterator.
	* c-c++-common/gomp/pr61486-2.c (test, test2): Remove ordered
	clause and ordered construct where no longer allowed.
	* c-c++-common/gomp/priority-1.c: New test.
	* c-c++-common/gomp/reduction-1.c: New test.
	* c-c++-common/gomp/schedule-simd-1.c: New test.
	* c-c++-common/gomp/sink-1.c: New test.
	* c-c++-common/gomp/sink-2.c: New test.
	* c-c++-common/gomp/sink-3.c: New test.
	* c-c++-common/gomp/sink-4.c: New test.
	* c-c++-common/gomp/udr-1.c: New test.
	* c-c++-common/taskloop-1.c: New test.
	* c-c++-common/cpp/openmp-define-3.c: Adjust for the new
	value of _OPENMP macro.
	* c-c++-common/cilk-plus/PS/body.c (foo): Adjust expected diagnostics.
	* c-c++-common/goacc-gomp/nesting-fail-1.c (f_acc_parallel,
	f_acc_kernels, f_acc_data, f_acc_loop): Add map clause to target data.
	* gcc.dg/gomp/clause-1.c:
	* gcc.dg/gomp/reduction-1.c: New test.
	* gcc.dg/gomp/sink-fold-1.c: New test.
	* gcc.dg/gomp/sink-fold-2.c: New test.
	* gcc.dg/gomp/sink-fold-3.c: New test.
	* gcc.dg/vect/vect-simd-clone-15.c: New test.
	* g++.dg/gomp/clause-1.C (T::test): Remove dg-error on privatization
	of non-static data members.
	* g++.dg/gomp/clause-3.C (foo): Remove one dg-error directive.
	Add some linear clause tests.
	* g++.dg/gomp/declare-simd-3.C: New test.
	* g++.dg/gomp/linear-1.C: New test.
	* g++.dg/gomp/member-1.C: New test.
	* g++.dg/gomp/member-2.C: New test.
	* g++.dg/gomp/pr66571-2.C: New test.
	* g++.dg/gomp/pr67504.C (foo): Add test for ordered clause with
	dependent argument.
	* g++.dg/gomp/pr67522.C (foo): Add test for invalid array section
	in reduction clause.
	* g++.dg/gomp/reference-1.C: New test.
	* g++.dg/gomp/sink-1.C: New test.
	* g++.dg/gomp/sink-2.C: New test.
	* g++.dg/gomp/sink-3.C: New test.
	* g++.dg/gomp/task-1.C: Remove both dg-error directives.
	* g++.dg/gomp/this-1.C: New test.
	* g++.dg/gomp/this-2.C: New test.
	* g++.dg/vect/simd-clone-2.cc: New test.
	* g++.dg/vect/simd-clone-2.h: New test.
	* g++.dg/vect/simd-clone-3.cc: New test.
	* g++.dg/vect/simd-clone-4.cc: New test.
	* g++.dg/vect/simd-clone-4.h: New test.
	* g++.dg/vect/simd-clone-5.cc: New test.
include/
2015-10-13  Jakub Jelinek  <jakub@redhat.com>
	    Ilya Verbin  <ilya.verbin@intel.com>

	* gomp-constants.h (GOMP_MAP_FLAG_ALWAYS): Define.
	(enum gomp_map_kind): Add GOMP_MAP_FIRSTPRIVATE,
	GOMP_MAP_FIRSTPRIVATE_INT, GOMP_MAP_USE_DEVICE_PTR,
	GOMP_MAP_ZERO_LEN_ARRAY_SECTION, GOMP_MAP_ALWAYS_TO,
	GOMP_MAP_ALWAYS_FROM, GOMP_MAP_ALWAYS_TOFROM, GOMP_MAP_STRUCT,
	GOMP_MAP_DELETE_ZERO_LEN_ARRAY_SECTION, GOMP_MAP_DELETE,
	GOMP_MAP_RELEASE, GOMP_MAP_FIRSTPRIVATE_POINTER.
	(GOMP_MAP_ALWAYS_TO_P, GOMP_MAP_ALWAYS_FROM_P): Define.
	(GOMP_TASK_FLAG_UNTIED, GOMP_TASK_FLAG_FINAL, GOMP_TASK_FLAG_MERGEABLE,
	GOMP_TASK_FLAG_DEPEND, GOMP_TASK_FLAG_PRIORITY, GOMP_TASK_FLAG_UP,
	GOMP_TASK_FLAG_GRAINSIZE, GOMP_TASK_FLAG_IF, GOMP_TASK_FLAG_NOGROUP,
	GOMP_TARGET_FLAG_NOWAIT, GOMP_TARGET_FLAG_EXIT_DATA,
	GOMP_TARGET_FLAG_UPDATE): Define.
libgomp/
2015-10-13  Jakub Jelinek  <jakub@redhat.com>
	    Aldy Hernandez  <aldyh@redhat.com>
	    Ilya Verbin  <ilya.verbin@intel.com>

	* config/linux/affinity.c (omp_get_place_num_procs,
	omp_get_place_proc_ids, gomp_get_place_proc_ids_8): New functions.
	* config/linux/doacross.h: New file.
	* config/posix/affinity.c (omp_get_place_num_procs,
	omp_get_place_proc_ids, gomp_get_place_proc_ids_8): New functions.
	* config/posix/doacross.h: New file.
	* env.c: Include gomp-constants.h.
	(struct gomp_task_icv): Rename run_sched_modifier to
	run_sched_chunk_size.
	(gomp_max_task_priority_var): New variable.
	(parse_schedule): Rename run_sched_modifier to run_sched_chunk_size.
	(handle_omp_display_env): Change _OPENMP value from 201307 to
	201511.  Print OMP_MAX_TASK_PRIORITY.
	(initialize_env): Parse OMP_MAX_TASK_PRIORITY.
	(omp_set_schedule, omp_get_schedule): Rename modifier argument to
	chunk_size and run_sched_modifier to run_sched_chunk_size.
	(omp_get_max_task_priority, omp_get_initial_device,
	omp_get_num_places, omp_get_place_num, omp_get_partition_num_places,
	omp_get_partition_place_nums): New functions.
	* fortran.c (omp_set_schedule_, omp_set_schedule_8_,
	omp_get_schedule_, omp_get_schedule_8_): Rename modifier argument
	to chunk_size.
	(omp_get_num_places_, omp_get_place_num_procs_,
	omp_get_place_num_procs_8_, omp_get_place_proc_ids_,
	omp_get_place_proc_ids_8_, omp_get_place_num_,
	omp_get_partition_num_places_, omp_get_partition_place_nums_,
	omp_get_partition_place_nums_8_, omp_get_initial_device_,
	omp_get_max_task_priority_): New functions.
	* libgomp_g.h (GOMP_loop_doacross_static_start,
	GOMP_loop_doacross_dynamic_start, GOMP_loop_doacross_guided_start,
	GOMP_loop_doacross_runtime_start, GOMP_loop_ull_doacross_static_start,
	GOMP_loop_ull_doacross_dynamic_start,
	GOMP_loop_ull_doacross_guided_start,
	GOMP_loop_ull_doacross_runtime_start, GOMP_doacross_post,
	GOMP_doacross_wait, GOMP_doacross_ull_post, GOMP_doacross_wait,
	GOMP_taskloop, GOMP_taskloop_ull, GOMP_target_41,
	GOMP_target_data_41, GOMP_target_update_41,
	GOMP_target_enter_exit_data): New prototypes.
	(GOMP_task): Add prototype argument.
	* libgomp.h (_LIBGOMP_CHECKING_): Define to 0 if not yet defined.
	(struct gomp_doacross_work_share): New type.
	(struct gomp_work_share): Add doacross field.
	(struct gomp_task_icv): Rename run_sched_modifier to
	run_sched_chunk_size.
	(enum gomp_task_kind): Rename GOMP_TASK_IFFALSE to
	GOMP_TASK_UNDEFERRED.  Add comments.
	(struct gomp_task_depend_entry): Add comments.
	(struct gomp_task): Likewise.
	(struct gomp_taskgroup): Likewise.
	(struct gomp_target_task): New type.
	(struct gomp_team): Add comment.
	(gomp_get_place_proc_ids_8, gomp_doacross_init,
	gomp_doacross_ull_init, gomp_task_maybe_wait_for_dependencies,
	gomp_create_target_task, gomp_target_task_fn): New prototypes.
	(struct target_var_desc): New type.
	(struct target_mem_desc): Adjust comment.  Use struct
	target_var_desc instead of splay_tree_key for list.
	(REFCOUNT_INFINITY): Define.
	(struct splay_tree_key_s): Remove copy_from field.
	(struct gomp_device_descr): Add dev2dev_func field.
	(enum gomp_map_vars_kind): New enum.
	(gomp_map_vars): Add one argument.
	* libgomp.map (OMP_4.5): Export omp_get_max_task_priority,
	omp_get_max_task_priority_, omp_get_num_places, omp_get_num_places_,
	omp_get_place_num_procs, omp_get_place_num_procs_,
	omp_get_place_num_procs_8_, omp_get_place_proc_ids,
	omp_get_place_proc_ids_, omp_get_place_proc_ids_8_, omp_get_place_num,
	omp_get_place_num_, omp_get_partition_num_places,
	omp_get_partition_num_places_, omp_get_partition_place_nums,
	omp_get_partition_place_nums_, omp_get_partition_place_nums_8_,
	omp_get_initial_device, omp_get_initial_device_, omp_target_alloc,
	omp_target_free, omp_target_is_present, omp_target_memcpy,
	omp_target_memcpy_rect, omp_target_associate_ptr and
	omp_target_disassociate_ptr.
	(GOMP_4.0.2): Renamed to ...
	(GOMP_4.5): ... this.  Export GOMP_target_41, GOMP_target_data_41,
	GOMP_target_update_41, GOMP_target_enter_exit_data, GOMP_taskloop,
	GOMP_taskloop_ull, GOMP_loop_doacross_dynamic_start,
	GOMP_loop_doacross_guided_start, GOMP_loop_doacross_runtime_start,
	GOMP_loop_doacross_static_start, GOMP_doacross_post,
	GOMP_doacross_wait, GOMP_loop_ull_doacross_dynamic_start,
	GOMP_loop_ull_doacross_guided_start,
	GOMP_loop_ull_doacross_runtime_start,
	GOMP_loop_ull_doacross_static_start, GOMP_doacross_ull_post and
	GOMP_doacross_ull_wait.
	* libgomp.texi: Document omp_get_max_task_priority.
	Rename modifier argument to chunk_size for omp_set_schedule and
	omp_get_schedule.  Document OMP_MAX_TASK_PRIORITY env var.
	* loop.c (GOMP_loop_runtime_start): Adjust for run_sched_modifier
	to run_sched_chunk_size renaming.
	(GOMP_loop_ordered_runtime_start): Likewise.
	(gomp_loop_doacross_static_start, gomp_loop_doacross_dynamic_start,
	gomp_loop_doacross_guided_start, GOMP_loop_doacross_runtime_start,
	GOMP_parallel_loop_runtime_start): New functions.
	(GOMP_parallel_loop_runtime): Adjust for run_sched_modifier
	to run_sched_chunk_size renaming.
	(GOMP_loop_doacross_static_start, GOMP_loop_doacross_dynamic_start,
	GOMP_loop_doacross_guided_start): New functions or aliases.
	* loop_ull.c (GOMP_loop_ull_runtime_start): Adjust for
	run_sched_modifier to run_sched_chunk_size renaming.
	(GOMP_loop_ull_ordered_runtime_start): Likewise.
	(gomp_loop_ull_doacross_static_start,
	gomp_loop_ull_doacross_dynamic_start,
	gomp_loop_ull_doacross_guided_start,
	GOMP_loop_ull_doacross_runtime_start): New functions.
	(GOMP_loop_ull_doacross_static_start,
	GOMP_loop_ull_doacross_dynamic_start,
	GOMP_loop_ull_doacross_guided_start): New functions or aliases.
	* oacc-mem.c (acc_map_data, present_create_copy,
	gomp_acc_insert_pointer): Pass GOMP_MAP_VARS_OPENACC instead of false
	to gomp_map_vars.
	(gomp_acc_remove_pointer): Use copy_from from target_var_desc.
	* oacc-parallel.c (GOACC_data_start): Pass GOMP_MAP_VARS_OPENACC
	instead of false to gomp_map_vars.
	(GOACC_parallel_keyed): Likewise.  Use copy_from from target_var_desc.
	* omp.h.in (omp_lock_hint_t): New type.
	(omp_init_lock_with_hint, omp_init_nest_lock_with_hint,
	omp_get_num_places, omp_get_place_num_procs, omp_get_place_proc_ids,
	omp_get_place_num, omp_get_partition_num_places,
	omp_get_partition_place_nums, omp_get_initial_device,
	omp_get_max_task_priority, omp_target_alloc, omp_target_free,
	omp_target_is_present, omp_target_memcpy, omp_target_memcpy_rect,
	omp_target_associate_ptr, omp_target_disassociate_ptr): New
	prototypes.
	* omp_lib.f90.in (omp_lock_hint_kind): New parameter.
	(omp_lock_hint_none, omp_lock_hint_uncontended,
	omp_lock_hint_contended, omp_lock_hint_nonspeculative,
	omp_lock_hint_speculative): New parameters.
	(omp_init_lock_with_hint, omp_init_nest_lock_with_hint,
	omp_get_num_places, omp_get_place_num_procs, omp_get_place_proc_ids,
	omp_get_place_num, omp_get_partition_num_places,
	omp_get_partition_place_nums, omp_get_initial_device,
	omp_get_max_task_priority): New interfaces.
	(omp_set_schedule, omp_get_schedule): Rename modifier argument
	to chunk_size.
	* omp_lib.h.in (omp_lock_hint_kind): New parameter.
	(omp_lock_hint_none, omp_lock_hint_uncontended,
	omp_lock_hint_contended, omp_lock_hint_nonspeculative,
	omp_lock_hint_speculative): New parameters.
	(omp_init_lock_with_hint, omp_init_nest_lock_with_hint,
	omp_get_num_places, omp_get_place_num_procs, omp_get_place_proc_ids,
	omp_get_place_num, omp_get_partition_num_places,
	omp_get_partition_place_nums, omp_get_initial_device,
	omp_get_max_task_priority): New functions and subroutines.
	* ordered.c: Include stdarg.h and string.h.
	(MAX_COLLAPSED_BITS): Define.
	(gomp_doacross_init, GOMP_doacross_post, GOMP_doacross_wait,
	gomp_doacross_ull_init, GOMP_doacross_ull_post,
	GOMP_doacross_ull_wait): New functions.
	* target.c: Include errno.h.
	(resolve_device): If device is not initialized, call
	gomp_init_device on it.
	(gomp_map_lookup): New function.
	(gomp_map_vars_existing): Add tgt_var argument, fill it in.
	Don't bump refcount if REFCOUNT_INFINITY.  Handle
	GOMP_MAP_ALWAYS_TO_P.
	(get_kind): Rename is_openacc argument to short_mapkind.
	(gomp_map_pointer): Use gomp_map_lookup.
	(gomp_map_fields_existing): New function.
	(gomp_map_vars): Rename is_openacc argument to short_mapkind
	and is_target to pragma_kind.  Handle GOMP_MAP_VARS_ENTER_DATA,
	handle GOMP_MAP_FIRSTPRIVATE_INT, GOMP_MAP_STRUCT,
	GOMP_MAP_USE_DEVICE_PTR, GOMP_MAP_ZERO_LEN_ARRAY_SECTION.
	Adjust for tgt->list changed type and copy_from living in there.
	(gomp_copy_from_async): Adjust for tgt->list changed type and
	copy_from living in there.
	(gomp_unmap_vars): Likewise.
	(gomp_update): Likewise.  Rename is_openacc argument to
	short_mapkind.  Don't fail if object is not mapped.
	(gomp_load_image_to_device): Initialize refcount to
	REFCOUNT_INFINITY.
	(gomp_target_fallback): New function.
	(gomp_get_target_fn_addr): Likewise.
	(GOMP_target): Adjust gomp_map_vars caller, use
	gomp_get_target_fn_addr and gomp_target_fallback.
	(GOMP_target_41): New function.
	(gomp_target_data_fallback): New function.
	(GOMP_target_data): Use it, adjust gomp_map_vars caller.
	(GOMP_target_data_41): New function.
	(GOMP_target_update): Adjust gomp_update caller.
	(GOMP_target_update_41): New function.
	(gomp_exit_data, GOMP_target_enter_exit_data,
	gomp_target_task_fn, omp_target_alloc, omp_target_free,
	omp_target_is_present, omp_target_memcpy,
	omp_target_memcpy_rect_worker, omp_target_memcpy_rect,
	omp_target_associate_ptr, omp_target_disassociate_ptr,
	gomp_load_plugin_for_device): New functions.
	* task.c: Include gomp-constants.h.  Include taskloop.c
	twice to get GOMP_taskloop and GOMP_taskloop_ull definitions.
	(gomp_task_handle_depend): New function.
	(GOMP_task): Use it.  Add priority argument.  Use
	gomp-constant.h constants instead of hardcoded numbers.
	Rename GOMP_TASK_IFFALSE to GOMP_TASK_UNDEFERRED.
	(gomp_create_target_task): New function.
	(verify_children_queue, verify_taskgroup_queue,
	verify_task_queue): New functions.
	(gomp_task_run_pre): Call verify_*_queue functions.
	If an upcoming tied task is about to leave the sibling or
	taskgroup queues in an invalid state, adjust appropriately.
	Remove taskgroup argument.  Add comments.
	(gomp_task_run_post_handle_dependers): Add comments.
	(gomp_task_run_post_remove_parent): Likewise.
	(gomp_barrier_handle_tasks): Adjust gomp_task_run_pre caller.
	(GOMP_taskwait): Likewise.  Add comments.
	(gomp_task_maybe_wait_for_dependencies): Fix scheduling
	problem such that the first non parent_depends_on task does not
	end up at the end of the children queue.
	(GOMP_taskgroup_start): Rename GOMP_TASK_IFFALSE to
	GOMP_TASK_UNDEFERRED.
	(GOMP_taskgroup_end): Adjust gomp_task_run_pre caller.
	* taskloop.c: New file.
	* testsuite/lib/libgomp.exp
	(check_effective_target_offload_device_nonshared_as): New proc.
	* testsuite/libgomp.c/affinity-2.c: New test.
	* testsuite/libgomp.c/doacross-1.c: New test.
	* testsuite/libgomp.c/doacross-2.c: New test.
	* testsuite/libgomp.c/examples-4/declare_target-1.c (fib_wrapper):
	Add map clause to target.
	* testsuite/libgomp.c/examples-4/declare_target-4.c (accum): Likewise.
	* testsuite/libgomp.c/examples-4/declare_target-5.c (accum): Likewise.
	* testsuite/libgomp.c/examples-4/device-1.c (main): Likewise.
	* testsuite/libgomp.c/examples-4/device-3.c (main): Likewise.
	* testsuite/libgomp.c/examples-4/target_data-3.c (gramSchmidt):
	Likewise.
	* testsuite/libgomp.c/examples-4/teams-2.c (dotprod): Likewise.
	* testsuite/libgomp.c/examples-4/teams-3.c (dotprod): Likewise.
	* testsuite/libgomp.c/examples-4/teams-4.c (dotprod): Likewise.
	* testsuite/libgomp.c/for-2.h (OMPTGT, OMPTO, OMPFROM): Define if
	not defined.  Use those where needed.
	* testsuite/libgomp.c/for-4.c: New test.
	* testsuite/libgomp.c/for-5.c: New test.
	* testsuite/libgomp.c/for-6.c: New test.
	* testsuite/libgomp.c/linear-1.c: New test.
	* testsuite/libgomp.c/ordered-4.c: New test.
	* testsuite/libgomp.c/pr66199-2.c (f2): Adjust for linear clause
	only allowed on the loop iterator.
	* testsuite/libgomp.c/pr66199-3.c: New test.
	* testsuite/libgomp.c/pr66199-4.c: New test.
	* testsuite/libgomp.c/reduction-7.c: New test.
	* testsuite/libgomp.c/reduction-8.c: New test.
	* testsuite/libgomp.c/reduction-9.c: New test.
	* testsuite/libgomp.c/reduction-10.c: New test.
	* testsuite/libgomp.c/target-1.c (fn2, fn3, fn4): Add
	map(tofrom:s).
	* testsuite/libgomp.c/target-2.c (fn2, fn3, fn4): Likewise.
	* testsuite/libgomp.c/target-7.c (foo): Add map(h) where needed.
	* testsuite/libgomp.c/target-11.c: New test.
	* testsuite/libgomp.c/target-12.c: New test.
	* testsuite/libgomp.c/target-13.c: New test.
	* testsuite/libgomp.c/target-14.c: New test.
	* testsuite/libgomp.c/target-15.c: New test.
	* testsuite/libgomp.c/target-16.c: New test.
	* testsuite/libgomp.c/target-17.c: New test.
	* testsuite/libgomp.c/target-18.c: New test.
	* testsuite/libgomp.c/target-19.c: New test.
	* testsuite/libgomp.c/target-20.c: New test.
	* testsuite/libgomp.c/target-21.c: New test.
	* testsuite/libgomp.c/target-22.c: New test.
	* testsuite/libgomp.c/target-23.c: New test.
	* testsuite/libgomp.c/target-24.c: New test.
	* testsuite/libgomp.c/target-25.c: New test.
	* testsuite/libgomp.c/target-26.c: New test.
	* testsuite/libgomp.c/target-27.c: New test.
	* testsuite/libgomp.c/taskloop-1.c: New test.
	* testsuite/libgomp.c/taskloop-2.c: New test.
	* testsuite/libgomp.c/taskloop-3.c: New test.
	* testsuite/libgomp.c/taskloop-4.c: New test.
	* testsuite/libgomp.c++/ctor-13.C: New test.
	* testsuite/libgomp.c++/doacross-1.C: New test.
	* testsuite/libgomp.c++/examples-4/declare_target-2.C:
	Replace offload_device with offload_device_nonshared_as.
	* testsuite/libgomp.c++/for-12.C: New test.
	* testsuite/libgomp.c++/for-13.C: New test.
	* testsuite/libgomp.c++/for-14.C: New test.
	* testsuite/libgomp.c++/linear-1.C: New test.
	* testsuite/libgomp.c++/member-1.C: New test.
	* testsuite/libgomp.c++/member-2.C: New test.
	* testsuite/libgomp.c++/member-3.C: New test.
	* testsuite/libgomp.c++/member-4.C: New test.
	* testsuite/libgomp.c++/member-5.C: New test.
	* testsuite/libgomp.c++/ordered-1.C: New test.
	* testsuite/libgomp.c++/reduction-5.C: New test.
	* testsuite/libgomp.c++/reduction-6.C: New test.
	* testsuite/libgomp.c++/reduction-7.C: New test.
	* testsuite/libgomp.c++/reduction-8.C: New test.
	* testsuite/libgomp.c++/reduction-9.C: New test.
	* testsuite/libgomp.c++/reduction-10.C: New test.
	* testsuite/libgomp.c++/reference-1.C: New test.
	* testsuite/libgomp.c++/simd14.C: New test.
	* testsuite/libgomp.c++/target-2.C (fn2): Add map(tofrom: s) clause.
	* testsuite/libgomp.c++/target-5.C: New test.
	* testsuite/libgomp.c++/target-6.C: New test.
	* testsuite/libgomp.c++/target-7.C: New test.
	* testsuite/libgomp.c++/target-8.C: New test.
	* testsuite/libgomp.c++/target-9.C: New test.
	* testsuite/libgomp.c++/target-10.C: New test.
	* testsuite/libgomp.c++/target-11.C: New test.
	* testsuite/libgomp.c++/target-12.C: New test.
	* testsuite/libgomp.c++/taskloop-1.C: New test.
	* testsuite/libgomp.c++/taskloop-2.C: New test.
	* testsuite/libgomp.c++/taskloop-3.C: New test.
	* testsuite/libgomp.c++/taskloop-4.C: New test.
	* testsuite/libgomp.c++/taskloop-5.C: New test.
	* testsuite/libgomp.c++/taskloop-6.C: New test.
	* testsuite/libgomp.c++/taskloop-7.C: New test.
	* testsuite/libgomp.c++/taskloop-8.C: New test.
	* testsuite/libgomp.c++/taskloop-9.C: New test.
	* testsuite/libgomp.fortran/affinity1.f90: New test.
	* testsuite/libgomp.fortran/affinity2.f90: New test.
liboffloadmic/
2015-10-13  Ilya Verbin  <ilya.verbin@intel.com>

	* plugin/libgomp-plugin-intelmic.cpp (GOMP_OFFLOAD_dev2dev): New
	function.
	* plugin/offload_target_main.cpp (__offload_target_tgt2tgt): New
	static function, register it in liboffloadmic.

From-SVN: r228777

1 files changed, 533 insertions, 148 deletions

diff --git a/libgomp/task.c b/libgomp/task.c
index 74920d5..1246c6a 100644
--- a/libgomp/task.c
+++ b/libgomp/task.c
@@ -29,6 +29,7 @@
 #include "libgomp.h"
 #include <stdlib.h>
 #include <string.h>
+#include "gomp-constants.h"
 
 typedef struct gomp_task_depend_entry *hash_entry_type;
 
@@ -91,6 +92,8 @@ gomp_end_task (void)
   thr->task = task->parent;
 }
 
+/* Orphan the task in CHILDREN and all its siblings.  */
+
 static inline void
 gomp_clear_parent (struct gomp_task *children)
 {
@@ -105,16 +108,136 @@ gomp_clear_parent (struct gomp_task *children)
     while (task != children);
 }
 
-static void gomp_task_maybe_wait_for_dependencies (void **depend);
+/* Helper function for GOMP_task and gomp_create_target_task.  Depend clause
+   handling for undeferred task creation.  */
+
+static void
+gomp_task_handle_depend (struct gomp_task *task, struct gomp_task *parent,
+			 void **depend)
+{
+  size_t ndepend = (uintptr_t) depend[0];
+  size_t nout = (uintptr_t) depend[1];
+  size_t i;
+  hash_entry_type ent;
+
+  task->depend_count = ndepend;
+  task->num_dependees = 0;
+  if (parent->depend_hash == NULL)
+    parent->depend_hash = htab_create (2 * ndepend > 12 ? 2 * ndepend : 12);
+  for (i = 0; i < ndepend; i++)
+    {
+      task->depend[i].addr = depend[2 + i];
+      task->depend[i].next = NULL;
+      task->depend[i].prev = NULL;
+      task->depend[i].task = task;
+      task->depend[i].is_in = i >= nout;
+      task->depend[i].redundant = false;
+      task->depend[i].redundant_out = false;
+
+      hash_entry_type *slot = htab_find_slot (&parent->depend_hash,
+					      &task->depend[i], INSERT);
+      hash_entry_type out = NULL, last = NULL;
+      if (*slot)
+	{
+	  /* If multiple depends on the same task are the same, all but the
+	     first one are redundant.  As inout/out come first, if any of them
+	     is inout/out, it will win, which is the right semantics.  */
+	  if ((*slot)->task == task)
+	    {
+	      task->depend[i].redundant = true;
+	      continue;
+	    }
+	  for (ent = *slot; ent; ent = ent->next)
+	    {
+	      if (ent->redundant_out)
+		break;
+
+	      last = ent;
+
+	      /* depend(in:...) doesn't depend on earlier depend(in:...).  */
+	      if (i >= nout && ent->is_in)
+		continue;
+
+	      if (!ent->is_in)
+		out = ent;
+
+	      struct gomp_task *tsk = ent->task;
+	      if (tsk->dependers == NULL)
+		{
+		  tsk->dependers
+		    = gomp_malloc (sizeof (struct gomp_dependers_vec)
+				   + 6 * sizeof (struct gomp_task *));
+		  tsk->dependers->n_elem = 1;
+		  tsk->dependers->allocated = 6;
+		  tsk->dependers->elem[0] = task;
+		  task->num_dependees++;
+		  continue;
+		}
+	      /* We already have some other dependency on tsk from earlier
+		 depend clause.  */
+	      else if (tsk->dependers->n_elem
+		       && (tsk->dependers->elem[tsk->dependers->n_elem - 1]
+			   == task))
+		continue;
+	      else if (tsk->dependers->n_elem == tsk->dependers->allocated)
+		{
+		  tsk->dependers->allocated
+		    = tsk->dependers->allocated * 2 + 2;
+		  tsk->dependers
+		    = gomp_realloc (tsk->dependers,
+				    sizeof (struct gomp_dependers_vec)
+				    + (tsk->dependers->allocated
+				       * sizeof (struct gomp_task *)));
+		}
+	      tsk->dependers->elem[tsk->dependers->n_elem++] = task;
+	      task->num_dependees++;
+	    }
+	  task->depend[i].next = *slot;
+	  (*slot)->prev = &task->depend[i];
+	}
+      *slot = &task->depend[i];
+
+      /* There is no need to store more than one depend({,in}out:) task per
+	 address in the hash table chain for the purpose of creation of
+	 deferred tasks, because each out depends on all earlier outs, thus it
+	 is enough to record just the last depend({,in}out:).  For depend(in:),
+	 we need to keep all of the previous ones not terminated yet, because
+	 a later depend({,in}out:) might need to depend on all of them.  So, if
+	 the new task's clause is depend({,in}out:), we know there is at most
+	 one other depend({,in}out:) clause in the list (out).  For
+	 non-deferred tasks we want to see all outs, so they are moved to the
+	 end of the chain, after first redundant_out entry all following
+	 entries should be redundant_out.  */
+      if (!task->depend[i].is_in && out)
+	{
+	  if (out != last)
+	    {
+	      out->next->prev = out->prev;
+	      out->prev->next = out->next;
+	      out->next = last->next;
+	      out->prev = last;
+	      last->next = out;
+	      if (out->next)
+		out->next->prev = out;
+	    }
+	  out->redundant_out = true;
+	}
+    }
+}
 
 /* Called when encountering an explicit task directive.  If IF_CLAUSE is
    false, then we must not delay in executing the task.  If UNTIED is true,
-   then the task may be executed by any member of the team.  */
+   then the task may be executed by any member of the team.
+
+   DEPEND is an array containing:
+	depend[0]: number of depend elements.
+	depend[1]: number of depend elements of type "out".
+	depend[2..N+1]: address of [1..N]th depend element.  */
 
 void
 GOMP_task (void (*fn) (void *), void *data, void (*cpyfn) (void *, void *),
 	   long arg_size, long arg_align, bool if_clause, unsigned flags,
-	   void **depend)
+	   void **depend, int priority)
 {
   struct gomp_thread *thr = gomp_thread ();
   struct gomp_team *team = thr->ts.team;
@@ -126,8 +249,7 @@ GOMP_task (void (*fn) (void *), void *data, void (*cpyfn) (void *, void *),
      might be running on different thread than FN.  */
   if (cpyfn)
     if_clause = false;
-  if (flags & 1)
-    flags &= ~1;
+  flags &= ~GOMP_TASK_FLAG_UNTIED;
 #endif
 
   /* If parallel or taskgroup has been cancelled, don't start new tasks.  */
@@ -136,6 +258,11 @@ GOMP_task (void (*fn) (void *), void *data, void (*cpyfn) (void *, void *),
 	  || (thr->task->taskgroup && thr->task->taskgroup->cancelled)))
     return;
 
+  if ((flags & GOMP_TASK_FLAG_PRIORITY) == 0)
+    priority = 0;
+  /* FIXME, use priority.  */
+  (void) priority;
+
   if (!if_clause || team == NULL
       || (thr->task && thr->task->final_task)
       || team->task_count > 64 * team->nthreads)
@@ -148,12 +275,14 @@ GOMP_task (void (*fn) (void *), void *data, void (*cpyfn) (void *, void *),
 	 depend clauses for non-deferred tasks other than this, because
 	 the parent task is suspended until the child task finishes and thus
 	 it can't start further child tasks.  */
-      if ((flags & 8) && thr->task && thr->task->depend_hash)
+      if ((flags & GOMP_TASK_FLAG_DEPEND)
+	  && thr->task && thr->task->depend_hash)
 	gomp_task_maybe_wait_for_dependencies (depend);
 
       gomp_init_task (&task, thr->task, gomp_icv (false));
-      task.kind = GOMP_TASK_IFFALSE;
-      task.final_task = (thr->task && thr->task->final_task) || (flags & 2);
+      task.kind = GOMP_TASK_UNDEFERRED;
+      task.final_task = (thr->task && thr->task->final_task)
+			|| (flags & GOMP_TASK_FLAG_FINAL);
       if (thr->task)
 	{
 	  task.in_tied_task = thr->task->in_tied_task;
@@ -196,7 +325,7 @@ GOMP_task (void (*fn) (void *), void *data, void (*cpyfn) (void *, void *),
       bool do_wake;
       size_t depend_size = 0;
 
-      if (flags & 8)
+      if (flags & GOMP_TASK_FLAG_DEPEND)
 	depend_size = ((uintptr_t) depend[0]
 		       * sizeof (struct gomp_task_depend_entry));
       task = gomp_malloc (sizeof (*task) + depend_size
@@ -204,7 +333,7 @@ GOMP_task (void (*fn) (void *), void *data, void (*cpyfn) (void *, void *),
       arg = (char *) (((uintptr_t) (task + 1) + depend_size + arg_align - 1)
 		      & ~(uintptr_t) (arg_align - 1));
       gomp_init_task (task, parent, gomp_icv (false));
-      task->kind = GOMP_TASK_IFFALSE;
+      task->kind = GOMP_TASK_UNDEFERRED;
       task->in_tied_task = parent->in_tied_task;
       task->taskgroup = taskgroup;
       thr->task = task;
@@ -219,7 +348,7 @@ GOMP_task (void (*fn) (void *), void *data, void (*cpyfn) (void *, void *),
       task->kind = GOMP_TASK_WAITING;
       task->fn = fn;
       task->fn_data = arg;
-      task->final_task = (flags & 2) >> 1;
+      task->final_task = (flags & GOMP_TASK_FLAG_FINAL) >> 1;
       gomp_mutex_lock (&team->task_lock);
       /* If parallel or taskgroup has been cancelled, don't start new
 	 tasks.  */
@@ -236,123 +365,7 @@ GOMP_task (void (*fn) (void *), void *data, void (*cpyfn) (void *, void *),
 	taskgroup->num_children++;
       if (depend_size)
 	{
-	  size_t ndepend = (uintptr_t) depend[0];
-	  size_t nout = (uintptr_t) depend[1];
-	  size_t i;
-	  hash_entry_type ent;
-
-	  task->depend_count = ndepend;
-	  task->num_dependees = 0;
-	  if (parent->depend_hash == NULL)
-	    parent->depend_hash
-	      = htab_create (2 * ndepend > 12 ? 2 * ndepend : 12);
-	  for (i = 0; i < ndepend; i++)
-	    {
-	      task->depend[i].addr = depend[2 + i];
-	      task->depend[i].next = NULL;
-	      task->depend[i].prev = NULL;
-	      task->depend[i].task = task;
-	      task->depend[i].is_in = i >= nout;
-	      task->depend[i].redundant = false;
-	      task->depend[i].redundant_out = false;
-
-	      hash_entry_type *slot
-		= htab_find_slot (&parent->depend_hash, &task->depend[i],
-				  INSERT);
-	      hash_entry_type out = NULL, last = NULL;
-	      if (*slot)
-		{
-		  /* If multiple depends on the same task are the
-		     same, all but the first one are redundant.
-		     As inout/out come first, if any of them is
-		     inout/out, it will win, which is the right
-		     semantics.  */
-		  if ((*slot)->task == task)
-		    {
-		      task->depend[i].redundant = true;
-		      continue;
-		    }
-		  for (ent = *slot; ent; ent = ent->next)
-		    {
-		      if (ent->redundant_out)
-			break;
-
-		      last = ent;
-
-		      /* depend(in:...) doesn't depend on earlier
-			 depend(in:...).  */
-		      if (i >= nout && ent->is_in)
-			continue;
-
-		      if (!ent->is_in)
-			out = ent;
-
-		      struct gomp_task *tsk = ent->task;
-		      if (tsk->dependers == NULL)
-			{
-			  tsk->dependers
-			    = gomp_malloc (sizeof (struct gomp_dependers_vec)
-					   + 6 * sizeof (struct gomp_task *));
-			  tsk->dependers->n_elem = 1;
-			  tsk->dependers->allocated = 6;
-			  tsk->dependers->elem[0] = task;
-			  task->num_dependees++;
-			  continue;
-			}
-		      /* We already have some other dependency on tsk
-			 from earlier depend clause.  */
-		      else if (tsk->dependers->n_elem
-			       && (tsk->dependers->elem[tsk->dependers->n_elem
-							- 1]
-				   == task))
-			continue;
-		      else if (tsk->dependers->n_elem
-			       == tsk->dependers->allocated)
-			{
-			  tsk->dependers->allocated
-			    = tsk->dependers->allocated * 2 + 2;
-			  tsk->dependers
-			    = gomp_realloc (tsk->dependers,
-					    sizeof (struct gomp_dependers_vec)
-					    + (tsk->dependers->allocated
-					       * sizeof (struct gomp_task *)));
-			}
-		      tsk->dependers->elem[tsk->dependers->n_elem++] = task;
-		      task->num_dependees++;
-		    }
-		  task->depend[i].next = *slot;
-		  (*slot)->prev = &task->depend[i];
-		}
-	      *slot = &task->depend[i];
-
-	      /* There is no need to store more than one depend({,in}out:)
-		 task per address in the hash table chain for the purpose
-		 of creation of deferred tasks, because each out
-		 depends on all earlier outs, thus it is enough to record
-		 just the last depend({,in}out:).  For depend(in:), we need
-		 to keep all of the previous ones not terminated yet, because
-		 a later depend({,in}out:) might need to depend on all of
-		 them.  So, if the new task's clause is depend({,in}out:),
-		 we know there is at most one other depend({,in}out:) clause
-		 in the list (out).  For non-deferred tasks we want to see
-		 all outs, so they are moved to the end of the chain,
-		 after first redundant_out entry all following entries
-		 should be redundant_out.  */
-	      if (!task->depend[i].is_in && out)
-		{
-		  if (out != last)
-		    {
-		      out->next->prev = out->prev;
-		      out->prev->next = out->next;
-		      out->next = last->next;
-		      out->prev = last;
-		      last->next = out;
-		      if (out->next)
-			out->next->prev = out;
-		    }
-		  out->redundant_out = true;
-		}
-	    }
+	  gomp_task_handle_depend (task, parent, depend);
 	  if (task->num_dependees)
 	    {
 	      gomp_mutex_unlock (&team->task_lock);
@@ -374,6 +387,7 @@ GOMP_task (void (*fn) (void *), void *data, void (*cpyfn) (void *, void *),
       parent->children = task;
       if (taskgroup)
 	{
+	  /* If applicable, place task into its taskgroup.  */
 	  if (taskgroup->children)
 	    {
 	      task->next_taskgroup = taskgroup->children;
@@ -412,26 +426,340 @@ GOMP_task (void (*fn) (void *), void *data, void (*cpyfn) (void *, void *),
     }
 }
 
+ialias (GOMP_taskgroup_start)
+ialias (GOMP_taskgroup_end)
+
+#define TYPE long
+#define UTYPE unsigned long
+#define TYPE_is_long 1
+#include "taskloop.c"
+#undef TYPE
+#undef UTYPE
+#undef TYPE_is_long
+
+#define TYPE unsigned long long
+#define UTYPE TYPE
+#define GOMP_taskloop GOMP_taskloop_ull
+#include "taskloop.c"
+#undef TYPE
+#undef UTYPE
+#undef GOMP_taskloop
+
+/* Called for nowait target tasks.  */
+
+void
+gomp_create_target_task (struct gomp_device_descr *devicep,
+			 void (*fn) (void *), size_t mapnum, void **hostaddrs,
+			 size_t *sizes, unsigned short *kinds,
+			 unsigned int flags, void **depend)
+{
+  struct gomp_thread *thr = gomp_thread ();
+  struct gomp_team *team = thr->ts.team;
+
+  /* If parallel or taskgroup has been cancelled, don't start new tasks.  */
+  if (team
+      && (gomp_team_barrier_cancelled (&team->barrier)
+	  || (thr->task->taskgroup && thr->task->taskgroup->cancelled)))
+    return;
+
+  struct gomp_target_task *ttask;
+  struct gomp_task *task;
+  struct gomp_task *parent = thr->task;
+  struct gomp_taskgroup *taskgroup = parent->taskgroup;
+  bool do_wake;
+  size_t depend_size = 0;
+
+  if (depend != NULL)
+    depend_size = ((uintptr_t) depend[0]
+		   * sizeof (struct gomp_task_depend_entry));
+  task = gomp_malloc (sizeof (*task) + depend_size
+		      + sizeof (*ttask)
+		      + mapnum * (sizeof (void *) + sizeof (size_t)
+				  + sizeof (unsigned short)));
+  gomp_init_task (task, parent, gomp_icv (false));
+  task->kind = GOMP_TASK_WAITING;
+  task->in_tied_task = parent->in_tied_task;
+  task->taskgroup = taskgroup;
+  ttask = (struct gomp_target_task *) &task->depend[(uintptr_t) depend[0]];
+  ttask->devicep = devicep;
+  ttask->fn = fn;
+  ttask->mapnum = mapnum;
+  memcpy (ttask->hostaddrs, hostaddrs, mapnum * sizeof (void *));
+  ttask->sizes = (size_t *) &ttask->hostaddrs[mapnum];
+  memcpy (ttask->sizes, sizes, mapnum * sizeof (size_t));
+  ttask->kinds = (unsigned short *) &ttask->sizes[mapnum];
+  memcpy (ttask->kinds, kinds, mapnum * sizeof (unsigned short));
+  ttask->flags = flags;
+  task->fn = gomp_target_task_fn;
+  task->fn_data = ttask;
+  task->final_task = 0;
+  gomp_mutex_lock (&team->task_lock);
+  /* If parallel or taskgroup has been cancelled, don't start new tasks.  */
+  if (__builtin_expect (gomp_team_barrier_cancelled (&team->barrier)
+			|| (taskgroup && taskgroup->cancelled), 0))
+    {
+      gomp_mutex_unlock (&team->task_lock);
+      gomp_finish_task (task);
+      free (task);
+      return;
+    }
+  if (taskgroup)
+    taskgroup->num_children++;
+  if (depend_size)
+    {
+      gomp_task_handle_depend (task, parent, depend);
+      if (task->num_dependees)
+	{
+	  gomp_mutex_unlock (&team->task_lock);
+	  return;
+	}
+    }
+  if (parent->children)
+    {
+      task->next_child = parent->children;
+      task->prev_child = parent->children->prev_child;
+      task->next_child->prev_child = task;
+      task->prev_child->next_child = task;
+    }
+  else
+    {
+      task->next_child = task;
+      task->prev_child = task;
+    }
+  parent->children = task;
+  if (taskgroup)
+    {
+      /* If applicable, place task into its taskgroup.  */
+      if (taskgroup->children)
+	{
+	  task->next_taskgroup = taskgroup->children;
+	  task->prev_taskgroup = taskgroup->children->prev_taskgroup;
+	  task->next_taskgroup->prev_taskgroup = task;
+	  task->prev_taskgroup->next_taskgroup = task;
+	}
+      else
+	{
+	  task->next_taskgroup = task;
+	  task->prev_taskgroup = task;
+	}
+      taskgroup->children = task;
+    }
+  if (team->task_queue)
+    {
+      task->next_queue = team->task_queue;
+      task->prev_queue = team->task_queue->prev_queue;
+      task->next_queue->prev_queue = task;
+      task->prev_queue->next_queue = task;
+    }
+  else
+    {
+      task->next_queue = task;
+      task->prev_queue = task;
+      team->task_queue = task;
+    }
+  ++team->task_count;
+  ++team->task_queued_count;
+  gomp_team_barrier_set_task_pending (&team->barrier);
+  do_wake = team->task_running_count + !parent->in_tied_task
+	    < team->nthreads;
+  gomp_mutex_unlock (&team->task_lock);
+  if (do_wake)
+    gomp_team_barrier_wake (&team->barrier, 1);
+}
+
+#if _LIBGOMP_CHECKING
+/* Sanity check TASK to make sure it is in its parent's children
+   queue, and that the tasks therein are in the right order.
+
+   The expected order is:
+	parent_depends_on WAITING tasks
+	!parent_depends_on WAITING tasks
+	TIED tasks
+
+   PARENT is the alleged parent of TASK.  */
+
+static void
+verify_children_queue (struct gomp_task *task, struct gomp_task *parent)
+{
+  if (task->parent != parent)
+    gomp_fatal ("verify_children_queue: incompatible parents");
+  /* It's OK, Annie was an orphan and she turned out all right.  */
+  if (!parent)
+    return;
+
+  bool seen_tied = false;
+  bool seen_plain_waiting = false;
+  bool found = false;
+  struct gomp_task *t = parent->children;
+  while (1)
+    {
+      if (t == task)
+	found = true;
+      if (seen_tied && t->kind == GOMP_TASK_WAITING)
+	gomp_fatal ("verify_children_queue: WAITING task after TIED");
+      if (t->kind == GOMP_TASK_TIED)
+	seen_tied = true;
+      else if (t->kind == GOMP_TASK_WAITING)
+	{
+	  if (t->parent_depends_on)
+	    {
+	      if (seen_plain_waiting)
+		gomp_fatal ("verify_children_queue: parent_depends_on after "
+			    "!parent_depends_on");
+	    }
+	  else
+	    seen_plain_waiting = true;
+	}
+      t = t->next_child;
+      if (t == parent->children)
+	break;
+    }
+  if (!found)
+    gomp_fatal ("verify_children_queue: child not found in parent queue");
+}
+
+/* Sanity check TASK to make sure it is in its taskgroup queue (if
+   applicable), and that the tasks therein are in the right order.
+
+   The expected order is that GOMP_TASK_WAITING tasks must come before
+   GOMP_TASK_TIED tasks.
+
+   TASK is the task.  */
+
+static void
+verify_taskgroup_queue (struct gomp_task *task)
+{
+  struct gomp_taskgroup *taskgroup = task->taskgroup;
+  if (!taskgroup)
+    return;
+
+  bool seen_tied = false;
+  bool found = false;
+  struct gomp_task *t = taskgroup->children;
+  while (1)
+    {
+      if (t == task)
+	found = true;
+      if (t->kind == GOMP_TASK_WAITING && seen_tied)
+	gomp_fatal ("verify_taskgroup_queue: WAITING task after TIED");
+      if (t->kind == GOMP_TASK_TIED)
+	seen_tied = true;
+      t = t->next_taskgroup;
+      if (t == taskgroup->children)
+	break;
+    }
+  if (!found)
+    gomp_fatal ("verify_taskgroup_queue: child not found in parent queue");
+}
+
+/* Verify that TASK is in the team's task queue.  */
+
+static void
+verify_task_queue (struct gomp_task *task, struct gomp_team *team)
+{
+  struct gomp_task *t = team->task_queue;
+  if (team)
+    while (1)
+      {
+	if (t == task)
+	  return;
+	t = t->next_queue;
+	if (t == team->task_queue)
+	  break;
+      }
+  gomp_fatal ("verify_team_queue: child not in team");
+}
+#endif
+
 static inline bool
 gomp_task_run_pre (struct gomp_task *child_task, struct gomp_task *parent,
-		   struct gomp_taskgroup *taskgroup, struct gomp_team *team)
+		   struct gomp_team *team)
 {
+#if _LIBGOMP_CHECKING
+  verify_children_queue (child_task, parent);
+  verify_taskgroup_queue (child_task);
+  verify_task_queue (child_task, team);
+#endif
+
   if (parent)
     {
+      /* Adjust children such that it will point to a next child,
+	 while the current one is scheduled to be executed.  This way,
+	 GOMP_taskwait (and others) can schedule a next task while
+	 waiting.
+
+	 Do not remove it entirely from the circular list, as it is
+	 still a child, though not one we should consider first (say
+	 by GOMP_taskwait).  */
       if (parent->children == child_task)
 	parent->children = child_task->next_child;
+      /* TIED tasks cannot come before WAITING tasks.  If we're about
+	 to make this task TIED, rewire things appropriately.
+	 However, a TIED task at the end is perfectly fine.  */
+      else if (child_task->next_child->kind == GOMP_TASK_WAITING
+	       && child_task->next_child != parent->children)
+	{
+	  /* Remove from the list.  */
+	  child_task->prev_child->next_child = child_task->next_child;
+	  child_task->next_child->prev_child = child_task->prev_child;
+	  /* Rewire at the end of its siblings.  */
+	  child_task->next_child = parent->children;
+	  child_task->prev_child = parent->children->prev_child;
+	  parent->children->prev_child->next_child = child_task;
+	  parent->children->prev_child = child_task;
+	}
+
+      /* If the current task (child_task) is at the top of the
+	 parent's last_parent_depends_on, it's about to be removed
+	 from it.  Adjust last_parent_depends_on appropriately.  */
       if (__builtin_expect (child_task->parent_depends_on, 0)
 	  && parent->taskwait->last_parent_depends_on == child_task)
 	{
+	  /* The last_parent_depends_on list was built with all
+	     parent_depends_on entries linked to the prev_child.  Grab
+	     the next last_parent_depends_on head from this prev_child if
+	     available...  */
 	  if (child_task->prev_child->kind == GOMP_TASK_WAITING
 	      && child_task->prev_child->parent_depends_on)
 	    parent->taskwait->last_parent_depends_on = child_task->prev_child;
 	  else
-	    parent->taskwait->last_parent_depends_on = NULL;
+	    {
+	      /* ...otherwise, there are no more parent_depends_on
+		 entries waiting to run.  In which case, clear the
+		 list.  */
+	      parent->taskwait->last_parent_depends_on = NULL;
+	    }
 	}
     }
-  if (taskgroup && taskgroup->children == child_task)
-    taskgroup->children = child_task->next_taskgroup;
+
+  /* Adjust taskgroup to point to the next taskgroup.  See note above
+     regarding adjustment of children as to why the child_task is not
+     removed entirely from the circular list.  */
+  struct gomp_taskgroup *taskgroup = child_task->taskgroup;
+  if (taskgroup)
+    {
+      if (taskgroup->children == child_task)
+	taskgroup->children = child_task->next_taskgroup;
+      /* TIED tasks cannot come before WAITING tasks.  If we're about
+	 to make this task TIED, rewire things appropriately.
+	 However, a TIED task at the end is perfectly fine.  */
+      else if (child_task->next_taskgroup->kind == GOMP_TASK_WAITING
+	       && child_task->next_taskgroup != taskgroup->children)
+	{
+	  /* Remove from the list.  */
+	  child_task->prev_taskgroup->next_taskgroup
+	    = child_task->next_taskgroup;
+	  child_task->next_taskgroup->prev_taskgroup
+	    = child_task->prev_taskgroup;
+	  /* Rewire at the end of its taskgroup.  */
+	  child_task->next_taskgroup = taskgroup->children;
+	  child_task->prev_taskgroup = taskgroup->children->prev_taskgroup;
+	  taskgroup->children->prev_taskgroup->next_taskgroup = child_task;
+	  taskgroup->children->prev_taskgroup = child_task;
+	}
+    }
+
+  /* Remove child_task from the task_queue.  */
   child_task->prev_queue->next_queue = child_task->next_queue;
   child_task->next_queue->prev_queue = child_task->prev_queue;
   if (team->task_queue == child_task)
@@ -442,6 +770,7 @@ gomp_task_run_pre (struct gomp_task *child_task, struct gomp_task *parent,
 	team->task_queue = NULL;
     }
   child_task->kind = GOMP_TASK_TIED;
+
   if (--team->task_queued_count == 0)
     gomp_team_barrier_clear_task_pending (&team->barrier);
   if ((gomp_team_barrier_cancelled (&team->barrier)
@@ -479,6 +808,11 @@ gomp_task_run_post_handle_depend_hash (struct gomp_task *child_task)
       }
 }
 
+/* After CHILD_TASK has been run, adjust the various task queues to
+   give higher priority to the tasks that depend on CHILD_TASK.
+
+   TEAM is the team to which CHILD_TASK belongs to.  */
+
 static size_t
 gomp_task_run_post_handle_dependers (struct gomp_task *child_task,
 				     struct gomp_team *team)
@@ -502,6 +836,7 @@ gomp_task_run_post_handle_dependers (struct gomp_task *child_task,
 	      if (parent->taskwait && parent->taskwait->last_parent_depends_on
 		  && !task->parent_depends_on)
 		{
+		  /* Put depender in last_parent_depends_on.  */
 		  struct gomp_task *last_parent_depends_on
 		    = parent->taskwait->last_parent_depends_on;
 		  task->next_child = last_parent_depends_on->next_child;
@@ -509,6 +844,8 @@ gomp_task_run_post_handle_dependers (struct gomp_task *child_task,
 		}
 	      else
 		{
+		  /* Make depender a sibling of child_task, and place
+		     it at the top of said sibling list.  */
 		  task->next_child = parent->children;
 		  task->prev_child = parent->children->prev_child;
 		  parent->children = task;
@@ -518,6 +855,7 @@ gomp_task_run_post_handle_dependers (struct gomp_task *child_task,
 	    }
 	  else
 	    {
+	      /* Make depender a sibling of child_task.  */
 	      task->next_child = task;
 	      task->prev_child = task;
 	      parent->children = task;
@@ -539,6 +877,8 @@ gomp_task_run_post_handle_dependers (struct gomp_task *child_task,
 		parent->taskwait->last_parent_depends_on = task;
 	    }
 	}
+      /* If depender is in a taskgroup, put it at the TOP of its
+	 taskgroup.  */
       if (taskgroup)
 	{
 	  if (taskgroup->children)
@@ -560,6 +900,8 @@ gomp_task_run_post_handle_dependers (struct gomp_task *child_task,
 	      gomp_sem_post (&taskgroup->taskgroup_sem);
 	    }
 	}
+      /* Put depender of child_task at the END of the team's
+	 task_queue.  */
       if (team->task_queue)
 	{
 	  task->next_queue = team->task_queue;
@@ -602,12 +944,18 @@ gomp_task_run_post_handle_depend (struct gomp_task *child_task,
   return gomp_task_run_post_handle_dependers (child_task, team);
 }
 
+/* Remove CHILD_TASK from its parent.  */
+
 static inline void
 gomp_task_run_post_remove_parent (struct gomp_task *child_task)
 {
   struct gomp_task *parent = child_task->parent;
   if (parent == NULL)
     return;
+
+  /* If this was the last task the parent was depending on,
+     synchronize with gomp_task_maybe_wait_for_dependencies so it can
+     clean up and return.  */
   if (__builtin_expect (child_task->parent_depends_on, 0)
       && --parent->taskwait->n_depend == 0
       && parent->taskwait->in_depend_wait)
@@ -615,6 +963,8 @@ gomp_task_run_post_remove_parent (struct gomp_task *child_task)
       parent->taskwait->in_depend_wait = false;
       gomp_sem_post (&parent->taskwait->taskwait_sem);
     }
+
+  /* Remove CHILD_TASK from its sibling list.  */
   child_task->prev_child->next_child = child_task->next_child;
   child_task->next_child->prev_child = child_task->prev_child;
   if (parent->children != child_task)
@@ -637,6 +987,8 @@ gomp_task_run_post_remove_parent (struct gomp_task *child_task)
     }
 }
 
+/* Remove CHILD_TASK from its taskgroup.  */
+
 static inline void
 gomp_task_run_post_remove_taskgroup (struct gomp_task *child_task)
 {
@@ -701,7 +1053,7 @@ gomp_barrier_handle_tasks (gomp_barrier_state_t state)
 	{
 	  child_task = team->task_queue;
 	  cancelled = gomp_task_run_pre (child_task, child_task->parent,
-					 child_task->taskgroup, team);
+					 team);
 	  if (__builtin_expect (cancelled, 0))
 	    {
 	      if (to_free)
@@ -766,7 +1118,9 @@ gomp_barrier_handle_tasks (gomp_barrier_state_t state)
     }
 }
 
-/* Called when encountering a taskwait directive.  */
+/* Called when encountering a taskwait directive.
+
+   Wait for all children of the current task.  */
 
 void
 GOMP_taskwait (void)
@@ -812,8 +1166,7 @@ GOMP_taskwait (void)
 	{
 	  child_task = task->children;
 	  cancelled
-	    = gomp_task_run_pre (child_task, task, child_task->taskgroup,
-				 team);
+	    = gomp_task_run_pre (child_task, task, team);
 	  if (__builtin_expect (cancelled, 0))
 	    {
 	      if (to_free)
@@ -863,6 +1216,9 @@ GOMP_taskwait (void)
 	 finish_cancelled:;
 	  size_t new_tasks
 	    = gomp_task_run_post_handle_depend (child_task, team);
+
+	  /* Remove child_task from children list, and set up the next
+	     sibling to be run.  */
 	  child_task->prev_child->next_child = child_task->next_child;
 	  child_task->next_child->prev_child = child_task->prev_child;
 	  if (task->children == child_task)
@@ -872,8 +1228,12 @@ GOMP_taskwait (void)
 	      else
 		task->children = NULL;
 	    }
+	  /* Orphan all the children of CHILD_TASK.  */
 	  gomp_clear_parent (child_task->children);
+
+	  /* Remove CHILD_TASK from its taskgroup.  */
 	  gomp_task_run_post_remove_taskgroup (child_task);
+
 	  to_free = child_task;
 	  child_task = NULL;
 	  team->task_count--;
@@ -889,9 +1249,11 @@ GOMP_taskwait (void)
 }
 
 /* This is like GOMP_taskwait, but we only wait for tasks that the
-   upcoming task depends on.  */
+   upcoming task depends on.
 
-static void
+   DEPEND is as in GOMP_task.  */
+
+void
 gomp_task_maybe_wait_for_dependencies (void **depend)
 {
   struct gomp_thread *thr = gomp_thread ();
@@ -923,11 +1285,33 @@ gomp_task_maybe_wait_for_dependencies (void **depend)
 	      {
 		tsk->parent_depends_on = true;
 		++num_awaited;
+		/* If a task we need to wait for is not already
+		   running and is ready to be scheduled, move it to
+		   front, so that we run it as soon as possible.
+
+		   We rearrange the children queue such that all
+		   parent_depends_on tasks are first, and
+		   last_parent_depends_on points to the last such task
+		   we rearranged.  For example, given the following
+		   children where PD[123] are the parent_depends_on
+		   tasks:
+
+			task->children
+			|
+			V
+			C1 -> C2 -> C3 -> PD1 -> PD2 -> PD3 -> C4
+
+		   We rearrange such that:
+
+			task->children
+			|	       +--- last_parent_depends_on
+			|	       |
+			V	       V
+			PD1 -> PD2 -> PD3 -> C1 -> C2 -> C3 -> C4
+		*/
+
 		if (tsk->num_dependees == 0 && tsk->kind == GOMP_TASK_WAITING)
 		  {
-		    /* If a task we need to wait for is not already
-		       running and is ready to be scheduled, move it
-		       to front, so that we run it as soon as possible.  */
 		    if (last_parent_depends_on)
 		      {
 			tsk->prev_child->next_child = tsk->next_child;
@@ -941,8 +1325,8 @@ gomp_task_maybe_wait_for_dependencies (void **depend)
 		      {
 			tsk->prev_child->next_child = tsk->next_child;
 			tsk->next_child->prev_child = tsk->prev_child;
-			tsk->prev_child = task->children;
-			tsk->next_child = task->children->next_child;
+			tsk->prev_child = task->children->prev_child;
+			tsk->next_child = task->children;
 			task->children = tsk;
 			tsk->prev_child->next_child = tsk;
 			tsk->next_child->prev_child = tsk;
@@ -983,8 +1367,7 @@ gomp_task_maybe_wait_for_dependencies (void **depend)
 	{
 	  child_task = task->children;
 	  cancelled
-	    = gomp_task_run_pre (child_task, task, child_task->taskgroup,
-				 team);
+	    = gomp_task_run_pre (child_task, task, team);
 	  if (__builtin_expect (cancelled, 0))
 	    {
 	      if (to_free)
@@ -1028,6 +1411,8 @@ gomp_task_maybe_wait_for_dependencies (void **depend)
 	    = gomp_task_run_post_handle_depend (child_task, team);
 	  if (child_task->parent_depends_on)
 	    --taskwait.n_depend;
+
+	  /* Remove child_task from sibling list.  */
 	  child_task->prev_child->next_child = child_task->next_child;
 	  child_task->next_child->prev_child = child_task->prev_child;
 	  if (task->children == child_task)
@@ -1037,6 +1422,7 @@ gomp_task_maybe_wait_for_dependencies (void **depend)
 	      else
 		task->children = NULL;
 	    }
+
 	  gomp_clear_parent (child_task->children);
 	  gomp_task_run_post_remove_taskgroup (child_task);
 	  to_free = child_task;
@@ -1070,7 +1456,7 @@ GOMP_taskgroup_start (void)
   struct gomp_taskgroup *taskgroup;
 
   /* If team is NULL, all tasks are executed as
-     GOMP_TASK_IFFALSE tasks and thus all children tasks of
+     GOMP_TASK_UNDEFERRED tasks and thus all children tasks of
      taskgroup and their descendant tasks will be finished
      by the time GOMP_taskgroup_end is called.  */
   if (team == NULL)
@@ -1137,8 +1523,7 @@ GOMP_taskgroup_end (void)
       if (child_task->kind == GOMP_TASK_WAITING)
 	{
 	  cancelled
-	    = gomp_task_run_pre (child_task, child_task->parent, taskgroup,
-				 team);
+	    = gomp_task_run_pre (child_task, child_task->parent, team);
 	  if (__builtin_expect (cancelled, 0))
 	    {
 	      if (to_free)