INCLUDE "debug.ml"
-INCLUDE "trace.ml"
+INCLUDE "log.ml"
INCLUDE "utils.ml"
open Format
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 -> Grammar2.t -> unit -> result_set
-
+ val naive_top_down_run : Ata.t -> Tree.t -> Tree.node -> result_set
+ val twopass_top_down_run : Ata.t -> Tree.t -> Tree.node -> result_set
end
module Make (U : ResJIT.S) : S with type result_set = U.NS.t =
let eval_trans auto s1 s2 trans =
+ LOG(__ "top-down-run" 3 "Evaluating transition list:@\n%a" Translist.print trans);
Translist.fold
(fun t ((a_st, a_op, a_todo) as acc)->
let q, _, m, f = Transition.node t in
let dummy _ _ _ _ _ = failwith "Uninitialized L3JIT"
- let create () = Cache.Lvl3.create 1024 dummy
+
+ let show_stats a =
+ let count = ref 0 in
+ Cache.Lvl3.iteri (fun _ _ _ _ b -> if not b then incr count) a;
+ Logger.print err_formatter "@?L3JIT: %i used entries@\n@?" !count
+ let create () =
+ let v = Cache.Lvl3.create 1024 dummy in
+ if !Options.verbose then at_exit (fun () -> show_stats v);
+ v
+
let find t tlist s1 s2 =
Cache.Lvl3.find t
(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 s2.StateSet.Node.id)
let orig_s1, orig_s2 =
Translist.fold (fun t (a1, a2) ->
let _, _, _, f = Transition.node t in
- let (_, _, fs1), (_, _, fs2) = Formula.st f in
+ let fs1, fs2 = Formula.st f in
(StateSet.union a1 fs1, StateSet.union a2 fs2)
) trl (StateSet.empty, StateSet.empty)
in
let ns1 = StateSet.inter s1 orig_s1
and ns2 = StateSet.inter s2 orig_s2 in
- let res, ops, todo = eval_trans auto ns1 ns2 trl in
+ let res, ops, todo = eval_trans auto orig_s1 orig_s2 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
let cache_apply cache auto tlist s1 s2 =
let f = gen_code auto tlist s1 s2 in
- TRACE("grammar", 2, __ "Inserting: %i, %a, %a\n%!"
+ LOG(__ "grammar" 2 "Inserting: %i, %a, %a\n%!"
(Uid.to_int tlist.Translist.Node.id) StateSet.print s1 StateSet.print s2);
add cache tlist s1 s2 f; f
end
DEFINE LOOP (t, states, ctx) = (
- let _t = (t) in
- TRACE("top-down-run", 3,
- __ "Entering node %i with loop (tag %s, context %i) with states %a\n%!"
- (Node.to_int _t)
- (Tag.to_string (Tree.tag tree _t))
- (Node.to_int (ctx))
- (StateSet.print) (states));
+ let _t = t in
+ LOG(__ "top-down-run" 3
+ "Entering node %i with loop (tag %s, context %i) with states %a"
+ (Node.to_int _t)
+ (Tag.to_string (Tree.tag tree _t))
+ (Node.to_int (ctx))
+ (StateSet.print) (states));
if _t == Tree.nil then nil_res
else
let tag = Tree.tag tree _t in
DEFINE LOOP_TAG (t, states, tag, ctx) = (
let _t = (t) in (* to avoid duplicating expression t *)
- TRACE("top-down-run", 3,
- __ "Entering node %i with loop_tag (tag %s, context %i) with states %a\n%!"
+ LOG(__ "top-down-run" 3
+ "Entering node %i with loop_tag (tag %s, context %i) with states %a"
(Node.to_int _t)
(Tag.to_string (tag))
(Node.to_int (ctx))
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 empty_slot = Array.create res_len U.NS.empty in
let nil_res = auto.bottom_states, empty_slot in
let cache3 = L3JIT.create () in
-
+ let mark_subtree =
+ fun s subtree -> if subtree != U.NS.empty then
+ let r = Array.copy empty_slot in
+ r.(auto.last) <- subtree;
+ s,r
+ else
+ s,empty_slot
+ in
let l3jit_dispatch trl s1 s2 t sl1 sl2 =
let f = L3JIT.find cache3 trl s1 s2 in
if f == L3JIT.dummy then (L3JIT.cache_apply cache3 auto trl s1 s2) empty_slot sl1 sl2 tree t
in
let cache2 = L2JIT.create () in
-
let rec l2jit_dispatch t tag states ctx opcode =
match opcode with
| L2JIT.RETURN -> nil_res
| L2JIT.CACHE ->
- let opcode = L2JIT.compile cache2 auto tree tag states in
- l2jit_dispatch t tag states ctx opcode
+ LOG(__ "top-down-run" 3
+ "Top-down cache miss for configuration %s %a"
+ (Tag.to_string tag) StateSet.print states);
+ 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
+ l2jit_dispatch_instr t (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 in
- l3jit_dispatch tr_list auto.bottom_states res2 t empty_slot slot2
+ let res2, slot2 =
+ l2jit_dispatch_instr t 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
+ l2jit_dispatch_instr t (Tree.closing tree t) instr1
+ in
+ let res2, slot2 =
+ l2jit_dispatch_instr t ctx instr2
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 =
- match instr with
+ and l2jit_dispatch_instr t 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_ELEMENT s -> LOOP ((Tree.next_element tree t), s, ctx)
| L2JIT.TAGGED_DESCENDANT (s, tag) ->
- LOOP_TAG ((Tree.tagged_descendant tree t tag), s, tag, ctx)
+ LOOP_TAG ((Tree.tagged_descendant tree t tag), s, tag, ctx)
| L2JIT.TAGGED_FOLLOWING (s, tag) ->
- LOOP_TAG((Tree.tagged_following_before tree t tag ctx), s, tag, ctx)
+ LOOP_TAG((Tree.tagged_following_before tree t tag ctx), s, tag, ctx)
| L2JIT.SELECT_DESCENDANT (s, _, us) ->
- LOOP((Tree.select_descendant tree t us), s, ctx)
+ LOOP((Tree.select_descendant tree t us), s, ctx)
| L2JIT.SELECT_FOLLOWING (s, pt, us) ->
- LOOP ((Tree.select_following_before tree t us ctx), s, ctx)
+ LOOP ((Tree.select_following_before tree t us ctx), s, ctx)
| L2JIT.TAGGED_CHILD (s, tag) ->
- LOOP_TAG((Tree.tagged_child tree t tag), s, tag, ctx)
+ LOOP_TAG((Tree.tagged_child tree t tag), s, tag, ctx)
| L2JIT.TAGGED_FOLLOWING_SIBLING (s, tag) ->
- LOOP_TAG((Tree.tagged_following_sibling tree t tag), s, tag, ctx)
+ LOOP_TAG((Tree.tagged_following_sibling tree t tag), s, tag, ctx)
| L2JIT.SELECT_CHILD (s, _, us) ->
- LOOP ((Tree.select_child tree t us), s, ctx)
+ LOOP ((Tree.select_child tree t us), s, ctx)
| L2JIT.SELECT_FOLLOWING_SIBLING (s, _, us) ->
- LOOP ((Tree.select_following_sibling tree t us), s, ctx)
+ LOOP ((Tree.select_following_sibling tree t us), s, 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
+ mark_subtree s (U.NS.subtree_tags tree t tag)
| 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
-
+ mark_subtree s (U.NS.subtree_elements tree t)
in
let r = LOOP (root, states, ctx) in
(*L3JIT.stats err_formatter cache3; *)
(*** Bottom-up evaluation function **)
- let ns_print fmt t =
- Format.fprintf fmt "{ ";
- U.NS.iter begin fun node ->
- Format.fprintf fmt "%a " Node.print node;
- end t;
- Format.fprintf fmt "}"
-
- let slot_print fmt t =
- Array.iteri begin fun state ns ->
- Format.eprintf "%a -> %a\n" State.print state ns_print ns;
- end t
-
-
let eval_trans auto tree parent res1 res2 = assert false
+ let rec uniq = function
+ | ([] | [ _ ]) as l -> l
+ | e1 :: ((e2 :: ll) as l) -> if e1 == e2 then uniq l
+ else e1 :: e2 :: (uniq ll);;
let bottom_up_run auto tree (query, pat) =
- let leaves = Array.to_list (Tree.full_text_query query tree pat) in
+ let array = time ~msg:"Timing text query" (Tree.full_text_query query tree) pat in
+ let leaves = Array.to_list array in
let states = auto.states in
let res_len = (StateSet.max_elt states) + 1 in
let empty_slot = Array.create res_len U.NS.empty in
[] -> acc
| node :: ll ->
let res, lll = bottom_up_next node ll Tree.nil in
- if (lll <> []) then Printf.eprintf "Leftover elements\n%!";
+ if (lll <> []) then
+ begin
+ eprintf "Leftover nodes: %i\n" (List.length lll);
+ end;
res
and bottom_up_next node rest stop =
then acc
else
let _, _, _, phi = Transition.node trs in
- let (_,_,l), (_,_,r) = Formula.st phi in
+ let l, r = Formula.st phi in
(StateSet.union l lstates,
StateSet.union r rstates,
Translist.cons trs tacc)
in
let lambda = ref 0 in
let rec start_loop idx states =
- TRACE("grammar", 2, __ "Node %i\n%!" (Node.to_int idx));
+ LOG(__ "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 begin
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 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
+(* 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;
+(* pstates.(0) <- res.in0;
+ pstates.(1) <- res.in1; *)
res (*
UCache.add ucache (t, states, fst res.out0, fst res.out1)
res.main;
{ res with
main = ((fst res.main), (U.close h (snd res.main)));
} *)
-
+(*
else
let res0 = partial_loop y0 pstates.(0) in
let res1 = partial_loop y1 pstates.(1) in
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
;;
+ (* Slow reference top-down implementation *)
+ let naive_top_down auto tree root states ctx =
+ 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
+ let l3jit_dispatch trl s1 s2 t sl1 sl2 =
+ let f = L3JIT.find cache3 trl s1 s2 in
+ if f == L3JIT.dummy then (L3JIT.cache_apply cache3 auto trl s1 s2) empty_slot sl1 sl2 tree t
+ else f empty_slot sl1 sl2 tree t
+ in
+ let dummy = Translist.nil, StateSet.singleton State.dummy, StateSet.singleton State.dummy in
+ let cache2 = Cache.Lvl2.create 512 dummy in
+ let rec loop t states ctx =
+ if states == StateSet.empty then nil_res
+ else if t == Tree.nil then (*StateSet.inter states auto.bottom_states, empty_slot *) nil_res
+ else
+ let tag = Tree.tag tree t in
+
+ let trans, lstates, rstates =
+ let c = Cache.Lvl2.find cache2 (Uid.to_int states.StateSet.Node.id) tag in
+ if c == dummy then
+ let c = Ata.get_trans auto states tag in
+ Cache.Lvl2.add cache2 (Uid.to_int states.StateSet.Node.id) tag c;
+ c
+ else c
+ in
+ let s1, res1 = loop (Tree.first_child tree t) lstates ctx
+ and s2, res2 = loop (Tree.next_sibling tree t) rstates ctx in
+ l3jit_dispatch trans s1 s2 t res1 res2
+ in
+ loop root states ctx
+
+
+
+
+ let naive_top_down_run auto tree root =
+ let res, slot = naive_top_down auto tree root auto.init (Tree.closing tree root) in
+ slot.(StateSet.min_elt auto.topdown_marking_states)
+
+
+
+ let eval_form auto s1 s2 f =
+ let rec loop f =
+ match Formula.expr f with
+ | Formula.False | Formula.True | Formula.Pred _ -> f
+ | Formula.Atom(`Left, b, q) ->
+ Formula.of_bool (b == (StateSet.mem q s1))
+ | Formula.Atom (`Right, b, q) ->
+ Formula.of_bool(b == (StateSet.mem q s2))
+ | Formula.Atom (`Epsilon, _, _) -> assert false
+
+ | Formula.Or(f1, f2) ->
+ let b1 = loop f1 in
+ let b2 = loop f2 in
+ Formula.or_pred b1 b2
+ | Formula.And(f1, f2) ->
+ let b1 = loop f1 in
+ let b2 = loop f2 in
+ Formula.and_pred b1 b2
+ in
+ loop f
+
+ let eval_trans auto s1 s2 trans =
+ Translist.fold
+ (fun t ((a_st, mark) as acc)->
+ let q, _, m, f = Transition.node t in
+ let form = eval_form auto s1 s2 f in
+ match Formula.expr form with
+ | Formula.True -> StateSet.add q a_st, mark || m
+ | Formula.False -> acc
+ | _ -> assert false
+ ) trans (StateSet.empty, false)
+
+
+ let set a i v =
+ LOG(__ "twopass" 2 "Setting node %i to state %a\n%!"
+ i StateSet.print v);
+ a.(i) <- v
+
+ let twopass_top_down states_array auto tree root states ctx =
+ let dummy3 = StateSet.singleton State.dummy in
+ let cache3 = Cache.Lvl3.create 512 dummy3 in
+ let dummy2 = Translist.nil, StateSet.singleton State.dummy, StateSet.singleton State.dummy in
+ let cache2 = Cache.Lvl2.create 512 dummy2 in
+ let attributes = TagSet.inj_positive (Tree.attribute_tags tree) in
+ let rec loop t states ctx =
+ if t == Tree.nil then auto.bottom_states
+ else if states == StateSet.empty then
+ let () = set states_array (Node.to_int t) auto.bottom_states in
+ auto.bottom_states
+ else
+ let tag = Tree.tag tree t in
+ LOG(__ "twopass" 2 "Traversing node %i (tag %s) in states %a\n%!" (Node.to_int t) (Tag.to_string tag)
+ StateSet.print states
+ );
+ let trans, lstates, rstates =
+ let c = Cache.Lvl2.find cache2 (Uid.to_int states.StateSet.Node.id) tag in
+ if c == dummy2 then
+ let c = Ata.get_trans ~attributes:attributes auto states tag in
+ Cache.Lvl2.add cache2 (Uid.to_int states.StateSet.Node.id) tag c;
+ c
+ else c
+ in
+ LOG(__ "twopass" 2 "\nTransitions are:\n%!");
+ LOG(__ "twopass" 2"\nTransitions are:\n%a\n%!"
+ Translist.print trans
+ );
+ let s1 = loop (Tree.first_child tree t) lstates ctx
+ and s2 = loop (Tree.next_sibling tree t) rstates ctx in
+ let st =
+ let c = Cache.Lvl3.find cache3
+ (Uid.to_int s1.StateSet.Node.id)
+ (Uid.to_int s2.StateSet.Node.id)
+ (Uid.to_int trans.Translist.Node.id)
+ in
+ if c == dummy3 then
+ let c, _ = eval_trans auto s1 s2 trans in
+ Cache.Lvl3.add cache3
+ (Uid.to_int s1.StateSet.Node.id)
+ (Uid.to_int s2.StateSet.Node.id)
+ (Uid.to_int trans.Translist.Node.id) c;c
+ else c
+ in
+ set states_array (Node.to_int t) st;
+ st
+ in
+ loop root states ctx, (dummy2, cache2)
+
+
+ type action = Nop | Mark | Dummy
+
+ let twopass_top_down_scan states_array (dummy2, cache2) auto tree root states ctx =
+ let attributes = TagSet.inj_positive (Tree.attribute_tags tree) in
+ let cache3 = Cache.Lvl3.create 512 Dummy in
+ let rec loop t states acc =
+ if states == StateSet.empty || t = Tree.nil then acc
+ else
+ let tag = Tree.tag tree t in
+ let trans, _, _ =
+ let c = Cache.Lvl2.find cache2 (Uid.to_int states.StateSet.Node.id) tag in
+ if c == dummy2 then
+ let c = Ata.get_trans ~attributes:attributes auto states tag in
+ Cache.Lvl2.add cache2 (Uid.to_int states.StateSet.Node.id) tag c;
+ c
+ else c
+ in
+ let fs = Tree.first_child tree t in
+ let ns = Tree.next_sibling tree t in
+ let s1 = if fs != Tree.nil then states_array.(Node.to_int fs) else auto.bottom_states
+ and s2 = if ns != Tree.nil then states_array.(Node.to_int ns) else auto.bottom_states
+ in
+ let mark =
+ let c = Cache.Lvl3.find cache3
+ (Uid.to_int s1.StateSet.Node.id)
+ (Uid.to_int s2.StateSet.Node.id)
+ (Uid.to_int trans.Translist.Node.id)
+ in
+ if c == Dummy then
+ let _, c = eval_trans auto s1 s2 trans in
+ let c = if c then Mark else Nop in
+ Cache.Lvl3.add cache3
+ (Uid.to_int s1.StateSet.Node.id)
+ (Uid.to_int s2.StateSet.Node.id)
+ (Uid.to_int trans.Translist.Node.id) c;c
+ else c
+ in
+ LOG(__ "twopass" 2 "Evaluating node %i (tag %s).\n%!States=%a\n%!"
+ (Node.to_int t)
+ (Tag.to_string tag)
+ StateSet.print states
+ );
+ LOG(__ "twopass" 2 "Translist=%a\nLeft=%a\nRight=%a\nMark=%s\n\n%!"
+ Translist.print trans
+ StateSet.print s1
+ StateSet.print s2
+ (match mark with
+ Dummy -> "Dummy"
+ | Mark -> "Mark"
+ | Nop -> "Nop"));
+ if mark == Mark then
+ loop ns s2 (loop fs s1 (U.NS.snoc acc t))
+ else
+ loop ns s2 (loop fs s1 acc)
+ in
+ loop root states U.NS.empty
+
+ let twopass_top_down_run auto tree root =
+ let len = Node.to_int (Tree.closing tree root) + 1 in
+ LOG(__ "twopass" 2 "Creating array of size: %i\n%!" len);
+ let states_array = Array.make len StateSet.empty in
+ let _, cache =
+ twopass_top_down states_array auto tree root auto.init Tree.nil
+ in
+ twopass_top_down_scan states_array cache auto tree root auto.init Tree.nil
+
+
+
+
+
projects / SXSI / xpathcomp.git / blobdiff