9 val top_down_run : Ata.t -> Tree.t -> Tree.node -> result_set
10 val bottom_up_run : Ata.t -> Tree.t -> Compile.text_query * string -> result_set
11 val grammar_run : Ata.t -> Grammar2.t -> unit -> result_set
15 module Make (U : ResJIT.S) : S with type result_set = U.NS.t =
18 type result_set = U.NS.t;;
20 let eval_form auto s1 s2 f =
22 match Formula.expr f with
23 | Formula.False | Formula.True | Formula.Pred _ -> f, []
24 | Formula.Atom(`Left, b, q) ->
25 Formula.of_bool (b == (StateSet.mem q s1)),
26 if b && StateSet.mem q auto.topdown_marking_states then [ResJIT.LEFT q] else []
27 | Formula.Atom (`Right, b, q) ->
28 Formula.of_bool(b == (StateSet.mem q s2)),
29 if b && StateSet.mem q auto.topdown_marking_states then [ResJIT.RIGHT q] else []
30 | Formula.Atom (`Epsilon, _, _) -> assert false
32 | Formula.Or(f1, f2) ->
33 let b1, i1 = loop f1 in
34 let b2, i2 = loop f2 in
35 Formula.or_pred b1 b2, i1 @ i2
36 | Formula.And(f1, f2) ->
37 let b1, i1 = loop f1 in
38 let b2, i2 = loop f2 in
39 Formula.and_pred b1 b2, i1 @ i2
44 let eval_trans auto s1 s2 trans =
46 (fun t ((a_st, a_op, a_todo) as acc)->
47 let q, _, m, f = Transition.node t in
48 let form, ops = eval_form auto s1 s2 f in
49 match Formula.expr form with
52 (q, (if m then (ResJIT.SELF() :: ops) else ops)):: a_op,
54 | Formula.False -> acc
55 | Formula.Pred p -> a_st, a_op,
56 (p.Tree.Predicate.node, q, [(q,(if m then (ResJIT.SELF() :: ops) else ops))]) :: a_todo
58 ) trans (StateSet.empty, [], [])
65 type opcode = (t -> t -> t -> Tree.t -> Tree.node -> StateSet.t * t)
67 type t = opcode Cache.Lvl3.t
69 let dummy _ _ _ _ _ = failwith "Uninitialized L3JIT"
71 let create () = Cache.Lvl3.create 1024 dummy
72 let find t tlist s1 s2 =
74 (Uid.to_int s2.StateSet.Node.id)
75 (Uid.to_int s1.StateSet.Node.id)
76 (Uid.to_int tlist.Translist.Node.id)
80 let add t tlist s1 s2 v =
82 (Uid.to_int s2.StateSet.Node.id)
83 (Uid.to_int s1.StateSet.Node.id)
84 (Uid.to_int tlist.Translist.Node.id)
87 let compile auto trl s1 s2 =
88 let orig_s1, orig_s2 =
89 Translist.fold (fun t (a1, a2) ->
90 let _, _, _, f = Transition.node t in
91 let (_, _, fs1), (_, _, fs2) = Formula.st f in
92 (StateSet.union a1 fs1, StateSet.union a2 fs2)
93 ) trl (StateSet.empty, StateSet.empty)
95 let ns1 = StateSet.inter s1 orig_s1
96 and ns2 = StateSet.inter s2 orig_s2 in
97 let res, ops, todo = eval_trans auto ns1 ns2 trl in
98 let code, not_marking = ResJIT.compile ops in
99 let todo_code, todo_notmarking =
100 List.fold_left (fun (l, b) (p, q, o) -> let c, b' = ResJIT.compile o in
101 (p, q, c)::l, b && b')
102 ([], not_marking) todo
104 let opcode = res, code, todo_notmarking, todo_code in
107 let gen_code auto tlist s1 s2 =
108 let res, code, not_marking, todo_code = compile auto tlist s1 s2 in
110 if todo_code == [] then
111 if not_marking then begin fun empty_slot sl1 sl2 _ node ->
112 let slot1_empty = sl1 == empty_slot
113 and slot2_empty = sl2 == empty_slot in
114 if slot1_empty && slot2_empty then res,sl2
119 Array.copy empty_slot
123 U.exec sl sl1 sl2 node code;
126 else (* marking *) begin fun empty_slot sl1 sl2 _ node ->
128 if sl2 == empty_slot then
129 if sl1 == empty_slot then
130 Array.copy empty_slot
134 U.exec sl sl1 sl2 node code;
137 else (* todo != [] *)
138 begin fun empty_slot sl1 sl2 tree node ->
140 if sl2 == empty_slot then
141 if sl1 == empty_slot then
142 Array.copy empty_slot
146 U.exec sl sl1 sl2 node code;
148 (fun ares (p, q, code) ->
149 if !p tree node then begin
150 if code != ResJIT.Nil then U.exec sl sl1 sl2 node code;
153 else ares) res todo_code, sl
159 let cache_apply cache auto tlist s1 s2 =
160 let f = gen_code auto tlist s1 s2 in
161 TRACE("grammar", 2, __ "Inserting: %i, %a, %a\n%!"
162 (Uid.to_int tlist.Translist.Node.id) StateSet.print s1 StateSet.print s2);
163 add cache tlist s1 s2 f; f
166 DEFINE LOOP (t, states, ctx) = (
168 TRACE("top-down-run", 3,
169 __ "Entering node %i with loop (tag %s, context %i) with states %a\n%!"
171 (Tag.to_string (Tree.tag tree _t))
173 (StateSet.print) (states));
174 if _t == Tree.nil then nil_res
176 let tag = Tree.tag tree _t in
178 _t tag (states) (ctx) (L2JIT.find cache2 tag (states))
181 DEFINE LOOP_TAG (t, states, tag, ctx) = (
182 let _t = (t) in (* to avoid duplicating expression t *)
183 TRACE("top-down-run", 3,
184 __ "Entering node %i with loop_tag (tag %s, context %i) with states %a\n%!"
186 (Tag.to_string (tag))
188 (StateSet.print) (states));
189 if _t == Tree.nil then nil_res
192 _t (tag) (states) (ctx) (L2JIT.find cache2 (tag) (states)))
195 let top_down_run auto tree root states ctx =
196 let res_len = StateSet.max_elt auto.states + 1 in
197 let empty_slot = Array.create res_len U.NS.empty in
198 let nil_res = auto.bottom_states, empty_slot in
199 let cache3 = L3JIT.create () in
201 let l3jit_dispatch trl s1 s2 t sl1 sl2 =
202 let f = L3JIT.find cache3 trl s1 s2 in
203 if f == L3JIT.dummy then (L3JIT.cache_apply cache3 auto trl s1 s2) empty_slot sl1 sl2 tree t
204 else f empty_slot sl1 sl2 tree t
207 let cache2 = L2JIT.create () in
210 let rec l2jit_dispatch t tag states ctx opcode =
212 | L2JIT.RETURN -> nil_res
214 let opcode = L2JIT.compile cache2 auto tree tag states in
215 l2jit_dispatch t tag states ctx opcode
217 | L2JIT.LEFT (tr_list, instr) ->
219 l2jit_dispatch_instr t tag states (Tree.closing tree t) instr
221 l3jit_dispatch tr_list res1 auto.bottom_states t slot1 empty_slot
223 | L2JIT.RIGHT (tr_list, instr) ->
224 let res2, slot2 = l2jit_dispatch_instr t tag states ctx instr in
225 l3jit_dispatch tr_list auto.bottom_states res2 t empty_slot slot2
227 | L2JIT.BOTH (tr_list, instr1, instr2) ->
229 l2jit_dispatch_instr t tag states (Tree.closing tree t) instr1
231 let res2, slot2 = l2jit_dispatch_instr t tag states ctx instr2 in
232 l3jit_dispatch tr_list res1 res2 t slot1 slot2
234 and l2jit_dispatch_instr t tag states ctx instr =
236 | L2JIT.FIRST_CHILD s -> LOOP ((Tree.first_child tree t), s, ctx)
237 | L2JIT.NEXT_SIBLING s -> LOOP ((Tree.next_sibling tree t), s, ctx)
239 | L2JIT.FIRST_ELEMENT s -> LOOP ((Tree.first_element tree t), s, ctx)
240 | L2JIT.NEXT_ELEMENT s -> LOOP ((Tree.next_element tree t), s, ctx)
242 | L2JIT.TAGGED_DESCENDANT (s, tag) ->
243 LOOP_TAG ((Tree.tagged_descendant tree t tag), s, tag, ctx)
245 | L2JIT.TAGGED_FOLLOWING (s, tag) ->
246 LOOP_TAG((Tree.tagged_following_before tree t tag ctx), s, tag, ctx)
248 | L2JIT.SELECT_DESCENDANT (s, _, us) ->
249 LOOP((Tree.select_descendant tree t us), s, ctx)
251 | L2JIT.SELECT_FOLLOWING (s, pt, us) ->
252 LOOP ((Tree.select_following_before tree t us ctx), s, ctx)
254 | L2JIT.TAGGED_CHILD (s, tag) ->
255 LOOP_TAG((Tree.tagged_child tree t tag), s, tag, ctx)
257 | L2JIT.TAGGED_FOLLOWING_SIBLING (s, tag) ->
258 LOOP_TAG((Tree.tagged_following_sibling tree t tag), s, tag, ctx)
260 | L2JIT.SELECT_CHILD (s, _, us) ->
261 LOOP ((Tree.select_child tree t us), s, ctx)
263 | L2JIT.SELECT_FOLLOWING_SIBLING (s, _, us) ->
264 LOOP ((Tree.select_following_sibling tree t us), s, ctx)
266 | L2JIT.TAGGED_SUBTREE(s, tag) ->
268 let count = U.NS.subtree_tags tree t tag in
269 if count != U.NS.empty then
270 let r = Array.copy empty_slot in
271 r.(auto.last) <- count;
276 | L2JIT.ELEMENT_SUBTREE(s) ->
278 let count = U.NS.subtree_elements tree t in
279 if count != U.NS.empty then
280 let r = Array.copy empty_slot in
281 r.(auto.last) <- count;
287 let r = LOOP (root, states, ctx) in
288 (*L3JIT.stats err_formatter cache3; *)
291 let full_top_down_run auto states tree root =
293 top_down_run auto tree root states (Tree.closing tree root)
295 let top_down_run auto tree root =
297 let res, slot = full_top_down_run auto auto.init tree root in
299 slot.(StateSet.min_elt auto.topdown_marking_states)
302 (*** Bottom-up evaluation function **)
305 Format.fprintf fmt "{ ";
306 U.NS.iter begin fun node ->
307 Format.fprintf fmt "%a " Node.print node;
309 Format.fprintf fmt "}"
311 let slot_print fmt t =
312 Array.iteri begin fun state ns ->
313 Format.eprintf "%a -> %a\n" State.print state ns_print ns;
317 let eval_trans auto tree parent res1 res2 = assert false
320 let bottom_up_run auto tree (query, pat) =
321 let leaves = Array.to_list (Tree.full_text_query query tree pat) in
322 let states = auto.states in
323 let res_len = (StateSet.max_elt states) + 1 in
324 let empty_slot = Array.create res_len U.NS.empty in
325 let nil_res = auto.bottom_states, empty_slot in
326 let cache = Cache.Lvl3.create 1024 L3JIT.dummy in
327 let rec loop_leaves l acc =
331 let res, lll = bottom_up_next node ll Tree.nil in
332 if (lll <> []) then Printf.eprintf "Leftover elements\n%!";
335 and bottom_up_next node rest stop =
336 let fs = Tree.first_child tree node in
338 if fs == Tree.nil then nil_res
339 else full_top_down_run auto states tree fs
341 move_up node res1 true rest stop
343 and move_up node res is_left rest stop =
344 if node == stop then res, rest
346 let prev_sibling = Tree.prev_sibling tree node in
347 let is_left' = prev_sibling == Tree.nil in
348 let real_parent = Tree.parent tree node in
350 if is_left' then real_parent else max (Tree.first_child tree real_parent) stop
352 (* let parent = if is_left' then Tree.parent tree node else prev_sibling in *)
353 let (s1, sl1), (s2, sl2), rest' =
354 if is_left then match rest with
355 [] -> res, nil_res, rest
357 if Tree.is_right_descendant tree node next
359 let res2, rest' = bottom_up_next next rest' node in
361 else res, nil_res, rest
365 let tag = Tree.tag tree node in
366 let id1 = Uid.to_int s1.StateSet.Node.id in
367 let id2 = Uid.to_int s2.StateSet.Node.id in
369 let code = Cache.Lvl3.find cache tag id1 id2 in
370 if code == L3JIT.dummy then
374 List.fold_left (fun acc' (labels, tr) ->
375 if labels == TagSet.any || TagSet.mem tag labels
376 then Translist.cons tr acc' else acc')
378 (Hashtbl.find auto.trans q)
383 let code = L3JIT.gen_code auto trl s1 s2 in
384 Cache.Lvl3.add cache tag id1 id2 code; code
387 let res' = code empty_slot sl1 sl2 tree node in
388 move_up parent res' is_left' rest' stop
390 let _, slot = loop_leaves leaves (nil_res) in
391 slot.(StateSet.min_elt auto.topdown_marking_states)
393 let get_trans g auto tag states =
394 StateSet.fold (fun q tr_acc ->
396 (fun ((lstates, rstates, tacc) as acc) (ts, trs) ->
397 if TagSet.mem (Tag.translate tag) ts then
398 if not (TagSet.mem Tag.attribute ts) && Grammar2.is_attribute g tag
401 let _, _, _, phi = Transition.node trs in
402 let (_,_,l), (_,_,r) = Formula.st phi in
403 (StateSet.union l lstates,
404 StateSet.union r rstates,
405 Translist.cons trs tacc)
407 tr_acc (Hashtbl.find auto.trans q)
408 ) states (StateSet.empty, StateSet.empty, Translist.nil)
411 let dispatch_param0 conf id2 y0 y1 =
413 | Grammar2.C0 | Grammar2.C2 -> Grammar2.Node0 id2
414 | Grammar2.C1 | Grammar2.C5 -> Grammar2.Node1(id2,y0)
415 | Grammar2.C3 | Grammar2.C6 -> y0
416 | Grammar2.C4 -> Grammar2.Node2(id2, y0, y1)
418 let dispatch_param1 conf id2 y0 y1 =
421 | Grammar2.C3 -> Grammar2.Node0 id2
423 | Grammar2.C6 -> Grammar2.Node1(id2, y1)
424 | _ -> Grammar2.dummy_param
426 module K_down = struct
427 type t = Grammar2.n_symbol * StateSet.t
428 let hash (x,y) = HASHINT2(Node.to_int x, Uid.to_int y.StateSet.Node.id)
429 let equal (x1,y1) (x2,y2) = x1 == x2 && y1 == y2
433 type t = Grammar2.n_symbol * StateSet.t * StateSet.t * StateSet.t
435 HASHINT4 (Node.to_int a,
436 Uid.to_int b.StateSet.Node.id,
437 Uid.to_int c.StateSet.Node.id,
438 Uid.to_int d.StateSet.Node.id)
439 let equal (a1, b1, c1, d1) (a2, b2, c2, d2) =
440 a1 == a2 && b1 == b2 && c1 == c2 && d1 == d2
445 include Hashtbl.Make(K_down)
446 let dummy = StateSet.singleton State.dummy
447 let notfound l = l.(0) == dummy && l.(1) == dummy
453 let a = [| dummy; dummy |] in
457 module UCache = Hashtbl.Make(K_up)
461 out0 : StateSet.t * U.t;
462 out1 : StateSet.t * U.t;
463 main : StateSet.t * U.t
466 { in0 = StateSet.empty;
467 in1 = StateSet.empty;
475 out0 = StateSet.empty,v;
476 out1 = StateSet.empty,v;
479 let grammar_run auto g () =
480 let dummy_leaf = Grammar2.dummy_param in
481 let dummy_set = StateSet.singleton State.dummy in
482 let res_len = (StateSet.max_elt auto.states) + 1 in
483 let empty_slot = Array.create res_len U.NS.empty in
484 let nil_res = mk_nil auto.bottom_states empty_slot in
485 let empty_res = mk_empty (StateSet.empty, empty_slot) in
486 let cache3 = L3JIT.create () in
487 let dummy2 = (StateSet.empty, StateSet.empty, Translist.nil) in
488 let cache2 = Cache.Lvl2.create 512 dummy2 in
489 let rule_counter = ref 0 in
490 let preorder_counter = ref 0 in
491 let dcache = DCache.create 1023 in
492 let ucache = UCache.create 1023 in
493 let term_array = [| StateSet.empty; StateSet.empty |] in
494 let get_trans tag states =
495 let c = Cache.Lvl2.find cache2 tag (Uid.to_int states.StateSet.Node.id) in
497 let c = get_trans g auto tag states in
499 Cache.Lvl2.add cache2 tag (Uid.to_int states.StateSet.Node.id) c;
504 let lambda = ref 0 in
505 let rec start_loop idx states =
506 TRACE("grammar", 2, __ "Node %i\n%!" (Node.to_int idx));
507 if states == dummy_set then nil_res else
508 if idx < Node.null then nil_res
510 let symbol = Grammar2.start_tag g idx in
511 let fc = Grammar2.start_first_child g idx in
512 let ns = Grammar2.start_next_sibling g fc in
513 if Grammar2.is_terminal g symbol then
514 let t = Grammar2.terminal symbol in
515 terminal_loop t states (Grammar2.Leaf (~-1,0,term_array, fc)) (Grammar2.Leaf (~-1,1,term_array, ns))
517 let nt = Grammar2.non_terminal symbol in
519 let lmbd = !lambda in
520 let y0 = (Grammar2.Leaf (lmbd,0, term_array, fc))
521 and y1 = (Grammar2.Leaf (lmbd,1, term_array, ns)) in
522 rule_loop nt states y0 y1
524 and rule_loop (t : Grammar2.n_symbol) states y0 y1 =
525 if t = Node.nil || states == dummy_set then nil_res else
526 let () = incr rule_counter in
527 if !rule_counter land 65535 == 0 then begin Gc.minor() end;
528 let k = (t, states) in
529 let pstates = DCache.find dcache k in
530 let notfound = DCache.notfound pstates in
531 let rhs = Grammar2.get_rule g t in
532 let id1 = Grammar2.get_id1 rhs in
533 let id2 = Grammar2.get_id2 rhs in
534 let conf = Grammar2.get_conf rhs in
536 let ny0 = dispatch_param0 conf id2 y0 y1 in
537 let ny1 = dispatch_param1 conf id2 y0 y1 in
538 let res = dispatch_loop id1 states ny0 ny1 in
539 pstates.(0) <- res.in0;
540 pstates.(1) <- res.in1;
542 UCache.add ucache (t, states, fst res.out0, fst res.out1)
544 let h = Hashtbl.create 7 in
545 for i = 0 to res_len - 1 do
546 Hashtbl.add h (0, i) (snd res.out0).(i);
547 Hashtbl.add h (1, i) (snd res.out1).(i);
550 main = ((fst res.main), (U.close h (snd res.main)));
554 let res0 = partial_loop y0 pstates.(0) in
555 let res1 = partial_loop y1 pstates.(1) in
556 let k2 = (t, states, fst res0.main, fst res1.main) in
559 UCache.find ucache k2
562 let ores0 = { res0 with main = fst res0.main, U.var 0 (snd res0.main) }
563 and ores1 = { res1 with main = fst res1.main, U.var 1 (snd res1.main) }
565 let res = dispatch_loop id1 states (Grammar2.Cache (0,ores0)) (Grammar2.Cache (1, ores1)) in
566 UCache.add ucache k2 res.main;
569 let h = Hashtbl.create 7 in
570 for i = 0 to res_len - 1 do
571 Hashtbl.add h (0, i) (snd res0.main).(i);
572 Hashtbl.add h (1, i) (snd res1.main).(i);
578 main = s, U.close h r;
581 and dispatch_loop id1 states ny0 ny1 =
582 if Grammar2.is_non_terminal g id1 then
583 rule_loop (Grammar2.non_terminal id1) states ny0 ny1
585 terminal_loop (Grammar2.terminal id1) states ny0 ny1
587 and terminal_loop (symbol : Grammar2.t_symbol) states y0 y1 =
589 if symbol == Grammar2.nil_symbol || symbol = Node.nil || states == dummy_set then nil_res else begin
590 let tag = Grammar2.tag symbol in
591 let lst, rst, trans = get_trans tag states in
592 let res0 = partial_loop y0 lst in
593 let res1 = partial_loop y1 rst in
594 let s1, slot1 = res0.main
595 and s2, slot2 = res1.main in
596 let opcode = L3JIT.find cache3 trans s1 s2 in
597 let node = Node.of_int !preorder_counter in
598 incr preorder_counter;
600 if opcode == L3JIT.dummy then
601 (L3JIT.cache_apply cache3 auto trans s1 s2) empty_slot slot1 slot2 (Obj.magic ()) node
603 opcode empty_slot slot1 slot2 (Obj.magic()) (node)
612 and partial_loop l states =
613 if l == dummy_leaf then nil_res else
615 | Grammar2.Cache (_, r) -> r
616 | Grammar2.Leaf (_,_, _, id) -> start_loop id states
617 | Grammar2.Node0 id ->
618 if (Grammar2.terminal id) == Grammar2.nil_symbol then nil_res
620 rule_loop (Grammar2.non_terminal id) states dummy_leaf dummy_leaf
622 | Grammar2.Node1 (id, y0) ->
623 rule_loop (Grammar2.non_terminal id) states y0 dummy_leaf
624 | Grammar2.Node2 (id, y0, y1) ->
625 if Grammar2.is_terminal g id then
626 terminal_loop (Grammar2.terminal id) states y0 y1
628 rule_loop (Grammar2.non_terminal id) states y0 y1
631 let (_, slot) = (start_loop (Node.null) auto.init).main in
632 slot.(StateSet.min_elt auto.topdown_marking_states)