type result_set
val top_down_run : Ata.t -> Tree.t -> Tree.node -> result_set
val bottom_up_run : Ata.t -> Tree.t -> Compile.text_query * string -> result_set
- val grammar_run : Ata.t -> Grammar.t -> unit -> result_set
+ val grammar_run : Ata.t -> Grammar2.t -> unit -> result_set
end
type opcode = (t -> t -> t -> Tree.t -> Tree.node -> StateSet.t * t)
- type t = opcode Cache.t Cache.t Cache.t
+ type t = opcode Cache.Lvl3.t
let dummy _ _ _ _ _ = failwith "Uninitialized L3JIT"
let create () = Cache.Lvl3.create 1024 dummy
-
- let stats fmt d =
- let d = Cache.Lvl3.to_array d in
- let len = Array.fold_left
- (fun acc a ->
- Array.fold_left (fun acc2 a2 -> Array.length a2 + acc2) acc a) 0 d
- in
-
- let lvl1 =
- Array.fold_left
- (fun acc a -> if Array.length a == 0 then acc else acc+1) 0 d in
- let lvl2 = Array.fold_left
- (fun acc a ->
- Array.fold_left (fun acc2 a2 -> if Array.length a2 == 0 then acc2 else acc2+1)
- acc a) 0 d
- in
- let lvl3 = Array.fold_left
- (fun acc a ->
- Array.fold_left (fun acc2 a2 ->
- Array.fold_left
- (fun acc3 a3 -> if a3 != dummy then acc3+1 else acc3)
- acc2 a2)
- acc a) 0 d
- in
- fprintf fmt "L3JIT Statistics:
-\t%i entries
-\t%i used L1 lines
-\t%i used L2 lines
-\t%i used L3 lines
-\ttable size: %ikb\n"
- len lvl1 lvl2 lvl3 (Ocaml.size_kb d)
-
let find t tlist s1 s2 =
Cache.Lvl3.find t
- (Uid.to_int tlist.Translist.Node.id)
- (Uid.to_int s1.StateSet.Node.id)
(Uid.to_int s2.StateSet.Node.id)
+ (Uid.to_int s1.StateSet.Node.id)
+ (Uid.to_int tlist.Translist.Node.id)
+
+
let add t tlist s1 s2 v =
Cache.Lvl3.add t
- (Uid.to_int tlist.Translist.Node.id)
- (Uid.to_int s1.StateSet.Node.id)
(Uid.to_int s2.StateSet.Node.id)
+ (Uid.to_int s1.StateSet.Node.id)
+ (Uid.to_int tlist.Translist.Node.id)
v
let compile auto trl s1 s2 =
in
let ns1 = StateSet.inter s1 orig_s1
and ns2 = StateSet.inter s2 orig_s2 in
- let res, ops, todo = eval_trans auto s1 s2 trl in
+ let res, ops, todo = eval_trans auto ns1 ns2 trl in
let code, not_marking = ResJIT.compile ops in
let todo_code, todo_notmarking =
List.fold_left (fun (l, b) (p, q, o) -> let c, b' = ResJIT.compile o in
l2jit_dispatch
_t (tag) (states) (ctx) (L2JIT.find cache2 (tag) (states)))
+
let top_down_run auto tree root states ctx =
- let res_len = (StateSet.max_elt auto.states) + 1 in
+ let res_len = StateSet.max_elt auto.states + 1 in
let empty_slot = Array.create res_len U.NS.empty in
let nil_res = auto.bottom_states, empty_slot in
let cache3 = L3JIT.create () in
in
let cache2 = L2JIT.create () in
- let () = D_TRACE_(at_exit (fun () -> L2JIT.stats Format.err_formatter cache2)) in
let rec l2jit_dispatch t tag states ctx opcode =
match opcode with
- | L2JIT.RETURN () -> nil_res
- | L2JIT.CACHE () ->
+ | L2JIT.RETURN -> nil_res
+ | L2JIT.CACHE ->
let opcode = L2JIT.compile cache2 auto tree tag states in
l2jit_dispatch t tag states ctx opcode
| L2JIT.LEFT (tr_list, instr) ->
let res1, slot1 =
- l2jit_dispatch_instr t tag states (Tree.closing tree t) instr true
+ l2jit_dispatch_instr t tag states (Tree.closing tree t) instr
in
l3jit_dispatch tr_list res1 auto.bottom_states t slot1 empty_slot
| L2JIT.RIGHT (tr_list, instr) ->
- let res2, slot2 = l2jit_dispatch_instr t tag states ctx instr false in
+ let res2, slot2 = l2jit_dispatch_instr t tag states ctx instr in
l3jit_dispatch tr_list auto.bottom_states res2 t empty_slot slot2
| L2JIT.BOTH (tr_list, instr1, instr2) ->
let res1, slot1 =
- l2jit_dispatch_instr t tag states (Tree.closing tree t) instr1 true
+ l2jit_dispatch_instr t tag states (Tree.closing tree t) instr1
in
- let res2, slot2 = l2jit_dispatch_instr t tag states ctx instr2 false in
+ let res2, slot2 = l2jit_dispatch_instr t tag states ctx instr2 in
l3jit_dispatch tr_list res1 res2 t slot1 slot2
- and l2jit_dispatch_instr t tag states ctx instr _left =
+ and l2jit_dispatch_instr t tag states ctx instr =
match instr with
| L2JIT.NOP () -> nil_res
| L2JIT.FIRST_CHILD s -> LOOP ((Tree.first_child tree t), s, ctx)
| L2JIT.NEXT_SIBLING s -> LOOP ((Tree.next_sibling tree t), s, ctx)
-(* | L2JIT.NEXT_SIBLING s -> LOOP ((Tree.next_node_before tree t ctx), s, ctx) *)
| L2JIT.FIRST_ELEMENT s -> LOOP ((Tree.first_element tree t), s, ctx)
| L2JIT.NEXT_ELEMENT s -> LOOP ((Tree.next_element tree t), s, ctx)
-(* | L2JIT.NEXT_ELEMENT s -> LOOP ((Tree.next_node_before tree t ctx), s, ctx) *)
| L2JIT.TAGGED_DESCENDANT (s, tag) ->
LOOP_TAG ((Tree.tagged_descendant tree t tag), s, tag, ctx)
| L2JIT.TAGGED_SUBTREE(s, tag) ->
let count = U.NS.subtree_tags tree t tag in
- if count != U.NS.empty then
- let r = Array.copy empty_slot in
- r.(auto.last) <- count;
- s,r
- else
- s,empty_slot
+ if count != U.NS.empty then
+ let r = Array.copy empty_slot in
+ r.(auto.last) <- count;
+ s,r
+ else
+ s,empty_slot
| L2JIT.ELEMENT_SUBTREE(s) ->
- let count = U.NS.subtree_elements tree t in
- if count != U.NS.empty then
- let r = Array.copy empty_slot in
- r.(auto.last) <- count;
- s,r
- else
- s,empty_slot
+ let count = U.NS.subtree_elements tree t in
+ if count != U.NS.empty then
+ let r = Array.copy empty_slot in
+ r.(auto.last) <- count;
+ s,r
+ else
+ s,empty_slot
- in
+ in
let r = LOOP (root, states, ctx) in
(*L3JIT.stats err_formatter cache3; *)
r
let _, slot = loop_leaves leaves (nil_res) in
slot.(StateSet.min_elt auto.topdown_marking_states)
+let get_trans g auto tag states =
+ StateSet.fold (fun q tr_acc ->
+ List.fold_left
+ (fun ((lstates, rstates, tacc) as acc) (ts, trs) ->
+ if TagSet.mem (Tag.translate tag) ts then
+ if not (TagSet.mem Tag.attribute ts) && Grammar2.is_attribute g tag
+ then acc
+ else
+ let _, _, _, phi = Transition.node trs in
+ let (_,_,l), (_,_,r) = Formula.st phi in
+ (StateSet.union l lstates,
+ StateSet.union r rstates,
+ Translist.cons trs tacc)
+ else acc)
+ tr_acc (Hashtbl.find auto.trans q)
+ ) states (StateSet.empty, StateSet.empty, Translist.nil)
(* Grammar run *)
- module ArrayPool =
- struct
- let pool = Queue.create ()
- let create dummy =
- if Queue.is_empty pool then
- Array.create 16 dummy
- else
- Queue.take pool
- let create dummy = Array.create 16 dummy
- let free p = Queue.add p pool
+let dispatch_param0 conf id2 y0 y1 =
+ match conf with
+ | Grammar2.C0 | Grammar2.C2 -> Grammar2.Node0 id2
+ | Grammar2.C1 | Grammar2.C5 -> Grammar2.Node1(id2,y0)
+ | Grammar2.C3 | Grammar2.C6 -> y0
+ | Grammar2.C4 -> Grammar2.Node2(id2, y0, y1)
+
+let dispatch_param1 conf id2 y0 y1 =
+ match conf with
+ | Grammar2.C2 -> y0
+ | Grammar2.C3 -> Grammar2.Node0 id2
+ | Grammar2.C5 -> y1
+ | Grammar2.C6 -> Grammar2.Node1(id2, y1)
+ | _ -> Grammar2.dummy_param
+
+ module K_down = struct
+ type t = Grammar2.n_symbol * StateSet.t
+ let hash (x,y) = HASHINT2(Node.to_int x, Uid.to_int y.StateSet.Node.id)
+ let equal (x1,y1) (x2,y2) = x1 == x2 && y1 == y2
end
+ module K_up = struct
+ type t = Grammar2.n_symbol * StateSet.t * StateSet.t * StateSet.t
+ let hash (a,b,c,d) =
+ HASHINT4 (Node.to_int a,
+ Uid.to_int b.StateSet.Node.id,
+ Uid.to_int c.StateSet.Node.id,
+ Uid.to_int d.StateSet.Node.id)
+ let equal (a1, b1, c1, d1) (a2, b2, c2, d2) =
+ a1 == a2 && b1 == b2 && c1 == c2 && d1 == d2
+ end
- let grammar_run auto g () =
-
- let start_symbol = (Node.of_int 0) in
- let dummy_leaf = Grammar.Leaf (Node.nil) in
+ module DCache =
+ struct
+ include Hashtbl.Make(K_down)
+ let dummy = StateSet.singleton State.dummy
+ let notfound l = l.(0) == dummy && l.(1) == dummy
+ let find h k =
+ try
+ find h k
+ with
+ Not_found ->
+ let a = [| dummy; dummy |] in
+ add h k a;
+ a
+ end
+ module UCache = Hashtbl.Make(K_up)
+ type result = {
+ in0 : StateSet.t;
+ in1 : StateSet.t;
+ out0 : StateSet.t * U.t;
+ out1 : StateSet.t * U.t;
+ main : StateSet.t * U.t
+ }
+ let mk_empty e =
+ { in0 = StateSet.empty;
+ in1 = StateSet.empty;
+ out0 = e;
+ out1 = e;
+ main = e
+ }
+ let mk_nil s v =
+ {
+ mk_empty (s,v) with
+ out0 = StateSet.empty,v;
+ out1 = StateSet.empty,v;
+ }
+ let grammar_run auto g () =
+ let dummy_leaf = Grammar2.dummy_param in
+ let dummy_set = StateSet.singleton State.dummy in
let res_len = (StateSet.max_elt auto.states) + 1 in
let empty_slot = Array.create res_len U.NS.empty in
- let nil_res = auto.bottom_states, empty_slot in
- let empty_res = StateSet.empty, empty_slot in
+ let nil_res = mk_nil auto.bottom_states empty_slot in
+ let empty_res = mk_empty (StateSet.empty, empty_slot) in
let cache3 = L3JIT.create () in
let dummy2 = (StateSet.empty, StateSet.empty, Translist.nil) in
let cache2 = Cache.Lvl2.create 512 dummy2 in
- let tmp1 = Array.create 16 dummy_leaf in
- let tmp2 = Array.create 16 dummy_leaf in
+ let rule_counter = ref 0 in
+ let preorder_counter = ref 0 in
+ let dcache = DCache.create 1023 in
+ let ucache = UCache.create 1023 in
+ let term_array = [| StateSet.empty; StateSet.empty |] in
let get_trans tag states =
let c = Cache.Lvl2.find cache2 tag (Uid.to_int states.StateSet.Node.id) in
if c == dummy2 then
- let c =
- StateSet.fold (fun q tr_acc ->
- List.fold_left
- (fun ((lstates, rstates, tacc) as acc) (ts, trs) ->
- if TagSet.mem (Tag.translate tag) ts then
- let _, _, _, phi = Transition.node trs in
- let (_,_,l),(_,_,r) = Formula.st phi in
- (StateSet.union l lstates,
- StateSet.union r rstates,
- Translist.cons trs tacc)
- else acc)
- tr_acc (Hashtbl.find auto.trans q)
- ) states (StateSet.empty, StateSet.empty, Translist.nil)
- in
+ let c = get_trans g auto tag states in
begin
Cache.Lvl2.add cache2 tag (Uid.to_int states.StateSet.Node.id) c;
c
end
else c
in
+ let lambda = ref 0 in
let rec start_loop idx states =
TRACE("grammar", 2, __ "Node %i\n%!" (Node.to_int idx));
+ if states == dummy_set then nil_res else
if idx < Node.null then nil_res
- else if StateSet.is_empty states then empty_res
else begin
- let symbol = Grammar.get_symbol_at g start_symbol idx in
- if Grammar.is_terminal symbol then
- let symbol = Grammar.terminal symbol in
- let tag = Grammar.tag symbol in
- let lst, rst, trans = get_trans tag states in
- let fs = Grammar.first_child g start_symbol idx in
- let s1, slot1 = start_loop fs lst in
- let s2, slot2 = start_loop (Grammar.next_sibling g start_symbol fs) rst in
- let opcode = L3JIT.find cache3 trans s1 s2 in
- if opcode == L3JIT.dummy then (L3JIT.cache_apply cache3 auto trans s1 s2) empty_slot slot1 slot2 (Obj.magic ()) (Obj.magic ())
- else opcode empty_slot slot1 slot2 (Obj.magic ()) (Obj.magic())
+ let symbol = Grammar2.start_tag g idx in
+ let fc = Grammar2.start_first_child g idx in
+ let ns = Grammar2.start_next_sibling g fc in
+ if Grammar2.is_terminal g symbol then
+ let t = Grammar2.terminal symbol in
+ terminal_loop t states (Grammar2.Leaf (~-1,0,term_array, fc)) (Grammar2.Leaf (~-1,1,term_array, ns))
else
- let tn = Grammar.non_terminal symbol in
- let nparam = Grammar.num_params tn in
- let a_param = tmp1 (*ArrayPool.create dummy_leaf*) in
- let child = ref (Grammar.first_child g start_symbol idx) in
- for i = 0 to nparam - 1 do
- let c = !child in
- a_param.(i) <- Grammar.Leaf c;
- child := Grammar.next_sibling g start_symbol c;
- done;
- (*let a_param = Array.init nparam
- (fun _ -> let c = !child in
- child := Grammar.next_sibling g start_symbol c;
- Grammar.Leaf c)
- in *)
- rule_loop tn a_param states
-
+ let nt = Grammar2.non_terminal symbol in
+ incr lambda;
+ let lmbd = !lambda in
+ let y0 = (Grammar2.Leaf (lmbd,0, term_array, fc))
+ and y1 = (Grammar2.Leaf (lmbd,1, term_array, ns)) in
+ rule_loop nt states y0 y1
end
- and counter = ref 0
- and rule_loop (t : Grammar.n_symbol) a_param states =
-
- incr counter;
- if !counter land 8191 == 0 then Gc.minor();
-
- let id1 = Grammar.get_id1 g t in
- let id2 = Grammar.get_id2 g t in
- let param_pos = Grammar.get_param_pos g t in
- let nparam1 = Grammar.num_children id1 in
- let nparam2 =
- if Grammar.is_terminal id2 && Grammar.is_nil g (Grammar.terminal id2) then 0
- else Grammar.num_children id2
- in
- let a_param1 = (*ArrayPool.create dummy_leaf*) tmp2 (* Array.create nparam1 dummy_leaf *) in
- let a_param2 = Array.create nparam2 dummy_leaf (* Array.create nparam2 dummy_leaf *) in
- let i = param_pos - 2 in
- (*Array.blit a_param 0 a_param1 0 (i+1); (* Pass parameters before id2 *) *)
- (* Array.blit is too slow *)
- for k = 0 to i do
- a_param1.(k) <- a_param.(k);
- done;
- a_param1.(i+1) <- Grammar.Node(id2, a_param2); (* id2( ... ) *)
- (*Array.blit a_param (i + nparam2 + 1) a_param1 (i+2) (nparam1 - i - 2); (* Pass parameters after id2 *) *)
- for k = 0 to nparam1 - i -3 do
- a_param1.(i+2+k) <- a_param.(i + nparam2 + 1 + k);
- done;
- (*Array.blit a_param (i + 1) a_param2 0 nparam2; (* parameters below id2 *) *)
- for k = 0 to nparam2 - 1 do
- a_param2.(k) <- a_param.(i+1+k)
- done;
- for i = 0 to nparam1 do
- a_param.(i) <- a_param1.(i)
- done;
- if Grammar.is_non_terminal id1 then
- let id1 = Grammar.non_terminal id1 in
- rule_loop id1 a_param states
- else
- let id1 = Grammar.terminal id1 in
- terminal_loop id1 a_param states
+ and rule_loop (t : Grammar2.n_symbol) states y0 y1 =
+ if t = Node.nil || states == dummy_set then nil_res else
+ let () = incr rule_counter in
+ if !rule_counter land 65535 == 0 then begin Gc.minor() end;
+ let k = (t, states) in
+ let pstates = DCache.find dcache k in
+ let notfound = DCache.notfound pstates in
+ let rhs = Grammar2.get_rule g t in
+ let id1 = Grammar2.get_id1 rhs in
+ let id2 = Grammar2.get_id2 rhs in
+ let conf = Grammar2.get_conf rhs in
+ if notfound then
+ let ny0 = dispatch_param0 conf id2 y0 y1 in
+ let ny1 = dispatch_param1 conf id2 y0 y1 in
+ let res = dispatch_loop id1 states ny0 ny1 in
+ pstates.(0) <- res.in0;
+ pstates.(1) <- res.in1;
+ res (*
+ UCache.add ucache (t, states, fst res.out0, fst res.out1)
+ res.main;
+ let h = Hashtbl.create 7 in
+ for i = 0 to res_len - 1 do
+ Hashtbl.add h (0, i) (snd res.out0).(i);
+ Hashtbl.add h (1, i) (snd res.out1).(i);
+ done;
+ { res with
+ main = ((fst res.main), (U.close h (snd res.main)));
+ } *)
- and terminal_loop (symbol : Grammar.t_symbol) a_param states =
- if Grammar.is_nil g symbol then nil_res else begin
- (* todo factor in from start_loop *)
- let tag = Grammar.tag symbol in
- let lst, rst, trans = get_trans tag states in
- let s1, slot1 = partial_loop a_param.(0) lst in
- let s2, slot2 = partial_loop a_param.(1) rst in
- let opcode = L3JIT.find cache3 trans s1 s2 in
- if opcode == L3JIT.dummy then (L3JIT.cache_apply cache3 auto trans s1 s2) empty_slot slot1 slot2 (Obj.magic ()) (Obj.magic ())
+ else
+ let res0 = partial_loop y0 pstates.(0) in
+ let res1 = partial_loop y1 pstates.(1) in
+ let k2 = (t, states, fst res0.main, fst res1.main) in
+ let s, r =
+ try
+ UCache.find ucache k2
+ with
+ Not_found ->
+ let ores0 = { res0 with main = fst res0.main, U.var 0 (snd res0.main) }
+ and ores1 = { res1 with main = fst res1.main, U.var 1 (snd res1.main) }
+ in
+ let res = dispatch_loop id1 states (Grammar2.Cache (0,ores0)) (Grammar2.Cache (1, ores1)) in
+ UCache.add ucache k2 res.main;
+ res.main
+ in
+ let h = Hashtbl.create 7 in
+ for i = 0 to res_len - 1 do
+ Hashtbl.add h (0, i) (snd res0.main).(i);
+ Hashtbl.add h (1, i) (snd res1.main).(i);
+ done;
+ { in0 = pstates.(0);
+ in1 = pstates.(1);
+ out0 = res0.main;
+ out1 = res1.main;
+ main = s, U.close h r;
+ }
+
+ and dispatch_loop id1 states ny0 ny1 =
+ if Grammar2.is_non_terminal g id1 then
+ rule_loop (Grammar2.non_terminal id1) states ny0 ny1
else
- opcode empty_slot slot1 slot2 (Obj.magic()) (Obj.magic())
+ terminal_loop (Grammar2.terminal id1) states ny0 ny1
- (* End: TODO refactor *)
+ and terminal_loop (symbol : Grammar2.t_symbol) states y0 y1 =
+ if symbol == Grammar2.nil_symbol || symbol = Node.nil || states == dummy_set then nil_res else begin
+ let tag = Grammar2.tag symbol in
+ let lst, rst, trans = get_trans tag states in
+ let res0 = partial_loop y0 lst in
+ let res1 = partial_loop y1 rst in
+ let s1, slot1 = res0.main
+ and s2, slot2 = res1.main in
+ let opcode = L3JIT.find cache3 trans s1 s2 in
+ let node = Node.of_int !preorder_counter in
+ incr preorder_counter;
+ let res =
+ if opcode == L3JIT.dummy then
+ (L3JIT.cache_apply cache3 auto trans s1 s2) empty_slot slot1 slot2 (Obj.magic ()) node
+ else
+ opcode empty_slot slot1 slot2 (Obj.magic()) (node)
+ in
+ { in0 = lst;
+ in1 = rst;
+ out0 = res0.main;
+ out1 = res1.main;
+ main = res }
end
and partial_loop l states =
- match l with
- | Grammar.Leaf id -> start_loop id states
- | Grammar.Node (id, a_param) ->
- if Grammar.is_terminal id then terminal_loop (Grammar.terminal id) a_param states
- else rule_loop (Grammar.non_terminal id) a_param states
+ if l == dummy_leaf then nil_res else
+ match l with
+ | Grammar2.Cache (_, r) -> r
+ | Grammar2.Leaf (_,_, _, id) -> start_loop id states
+ | Grammar2.Node0 id ->
+ if (Grammar2.terminal id) == Grammar2.nil_symbol then nil_res
+ else
+ rule_loop (Grammar2.non_terminal id) states dummy_leaf dummy_leaf
+
+ | Grammar2.Node1 (id, y0) ->
+ rule_loop (Grammar2.non_terminal id) states y0 dummy_leaf
+ | Grammar2.Node2 (id, y0, y1) ->
+ if Grammar2.is_terminal g id then
+ terminal_loop (Grammar2.terminal id) states y0 y1
+ else
+ rule_loop (Grammar2.non_terminal id) states y0 y1
in
- (*L3JIT.stats err_formatter cache3; *)
- let _, slot = start_loop (Node.null) auto.init in
+
+ let (_, slot) = (start_loop (Node.null) auto.init).main in
slot.(StateSet.min_elt auto.topdown_marking_states)
;;
projects / SXSI / xpathcomp.git / blobdiff