From 738218592e41da4ceb46f4dba41f292a60ba1f7b Mon Sep 17 00:00:00 2001 From: =?utf8?q?Kim=20Nguy=E1=BB=85n?= Date: Wed, 13 Mar 2013 18:41:06 +0100 Subject: [PATCH] Implement runtime optimisation via Hashing of transitions. --- src/auto/ata.ml | 16 ++++- src/auto/eval.ml | 128 ++++++++++++++++++++++++---------- src/utils.mlpack | 2 + src/utils/cache.ml | 31 +++++++- src/utils/cache.mli | 12 +++- src/utils/hlist.ml | 82 ++++++++++++++++++++++ src/utils/hlist.mli | 30 ++++++++ src/utils/hlist_sig.ml | 35 ++++++++++ tests/xmark_small.xml.queries | 4 +- 9 files changed, 296 insertions(+), 44 deletions(-) create mode 100644 src/utils/hlist.ml create mode 100644 src/utils/hlist.mli create mode 100644 src/utils/hlist_sig.ml diff --git a/src/auto/ata.ml b/src/auto/ata.ml index 4418996..9e5e57f 100644 --- a/src/auto/ata.ml +++ b/src/auto/ata.ml @@ -14,7 +14,7 @@ (***********************************************************************) (* - Time-stamp: + Time-stamp: *) INCLUDE "utils.ml" @@ -153,14 +153,24 @@ let create () = { id = next (); } +module Transition = Hcons.Make (struct + type t = State.t * QNameSet.t * SFormula.t + let equal (a, b, c) (d, e, f) = + a == d && b == e && c == f + let hash (a, b, c) = + HASHINT4 (PRIME1, a, ((QNameSet.uid b) :> int), ((SFormula.uid c) :> int)) +end) + +module TransList : Hlist.S with type elt = Transition.t = Hlist.Make(Transition) + let get_trans a states tag = StateSet.fold (fun q acc0 -> try let trs = Hashtbl.find a.transitions q in List.fold_left (fun acc1 (labs, phi) -> - if QNameSet.mem tag labs then (q,phi)::acc1 else acc1) acc0 trs + if QNameSet.mem tag labs then TransList.cons (Transition.make (q, labs, phi)) acc1 else acc1) acc0 trs with Not_found -> acc0 - ) states [] + ) states TransList.nil (* [add_trans a q labels f] adds a transition [(q,labels) -> f] to the diff --git a/src/auto/eval.ml b/src/auto/eval.ml index bf9788f..5c346d0 100644 --- a/src/auto/eval.ml +++ b/src/auto/eval.ml @@ -14,89 +14,145 @@ (***********************************************************************) (* - Time-stamp: + Time-stamp: *) INCLUDE "utils.ml" open Format open Utils -module Make (T : Tree.Sig.S) = struct +module Make (T : Tree.Sig.S) : + sig + val eval : Ata.t -> T.t -> T.node -> T.node list + end + = struct type cache = (int, StateSet.t) Hashtbl.t - let get c t n = try Hashtbl.find c (T.preorder t n) with Not_found -> StateSet.empty let set c t n v = Hashtbl.replace c (T.preorder t n) v - let eval_form phi tree node fcs nss ps ss = + + module Info = struct + type t = { is_left : bool; + is_right : bool; + has_left : bool; + has_right : bool; + kind : Tree.Common.NodeKind.t; + } + let equal a b = a = b + let hash a = Hashtbl.hash a + end + + module NodeInfo = Hcons.Make(Info) + + let eval_form phi node_info fcs nss ps ss = + let open NodeInfo in + let open Info in let rec loop phi = - match Ata.SFormula.expr phi with + begin match Ata.SFormula.expr phi with Formula.True -> true | Formula.False -> false | Formula.Atom a -> let p, b, q = Ata.Atom.node a in let pos = let open Ata in - match p with - | First_child -> StateSet.mem q fcs - | Next_sibling -> StateSet.mem q nss - | Parent | Previous_sibling -> StateSet.mem q ps - | Stay -> StateSet.mem q ss - | Is_first_child -> - node == (T.first_child tree (T.parent tree node)) - | Is_next_sibling -> - node == (T.next_sibling tree (T.parent tree node)) - | Is k -> k == (T.kind tree node) - | Has_first_child -> - T.nil != T.first_child tree node - | Has_next_sibling -> - T.nil != T.next_sibling tree node - in - if Ata.is_move p && (not b) then - eprintf "Warning: Invalid negative atom %a" Ata.Atom.print a; - b == pos + match p with + | First_child -> StateSet.mem q fcs + | Next_sibling -> StateSet.mem q nss + | Parent | Previous_sibling -> StateSet.mem q ps + | Stay -> StateSet.mem q ss + | Is_first_child -> node_info.node.is_left + | Is_next_sibling -> node_info.node.is_right + | Is k -> k == node_info.node.kind + | Has_first_child -> node_info.node.has_left + | Has_next_sibling -> node_info.node.has_right + in + if Ata.is_move p && (not b) then + eprintf "Warning: Invalid negative atom %a" Ata.Atom.print a; + b == pos | Formula.And(phi1, phi2) -> loop phi1 && loop phi2 | Formula.Or (phi1, phi2) -> loop phi1 || loop phi2 + end in loop phi - let eval_trans l tree node fcs nss ps ss acc = - List.fold_left (fun (acct, accs) ((q, phi) as trs) -> - if StateSet.mem q accs then (acct, accs) else - if eval_form phi tree node fcs nss ps ss then - (acct, StateSet.add q accs) - else - (trs::acct, accs) - ) ([], acc) l + let eval_trans cache ltrs node_info fcs nss ps ss = + let i = (ltrs.Ata.TransList.id :> int) + and j = (node_info.NodeInfo.id :> int) + and k = (fcs.StateSet.id :> int) + and l = (nss.StateSet.id :> int) + and m = (ps.StateSet.id :> int) + and n = (ss.StateSet.id :> int) in + let res = Cache.N6.find cache i j k l m n in + if res == Cache.N6.dummy cache then + let res = + Ata.TransList.fold (fun trs (acct, accs) -> + let q, _, phi = Ata.Transition.node trs in + if StateSet.mem q accs then (acct, accs) else + if eval_form phi node_info fcs nss ps accs then + (acct, StateSet.add q accs) + else + (Ata.TransList.cons trs acct, accs) + ) ltrs (Ata.TransList.nil, ss) + in + Cache.N6.add cache i j k l m n res; res + else + res let top_down_run auto tree node cache _i = let redo = ref false in + let dummy2 = Ata.TransList.cons + (Ata.Transition.make (State.dummy,QNameSet.empty, Ata.SFormula.false_)) + Ata.TransList.nil + in + let dummy6 = (dummy2, StateSet.empty) in + let trans_cache6 = Cache.N6.create 17 dummy6 in + let trans_cache2 = Cache.N2.create 17 dummy2 in let rec loop node = if node != T.nil then begin let parent = T.parent tree node in let fc = T.first_child tree node in let ns = T.next_sibling tree node in - let states0 = get cache tree node in let tag = T.tag tree node in - let trans0 = Ata.get_trans auto auto.Ata.states tag in + let states0 = get cache tree node in + let trans0 = + let trs = + Cache.N2.find trans_cache2 + (tag.QName.id :> int) (auto.Ata.states.StateSet.id :> int) + in + if trs == dummy2 then + let trs = Ata.get_trans auto auto.Ata.states tag in + (Cache.N2.add + trans_cache2 + (tag.QName.id :> int) + (auto.Ata.states.StateSet.id :> int) trs; trs) + else trs + in let ps = get cache tree parent in let fcs = get cache tree fc in let nss = get cache tree ns in + let node_info = NodeInfo.make + (Info.({ is_left = node == T.first_child tree parent; + is_right = node == T.next_sibling tree parent; + has_left = fc != T.nil; + has_right = ns != T.nil; + kind = T.kind tree node })) + in let trans1, states1 = - eval_trans trans0 tree node fcs nss ps states0 states0 + eval_trans trans_cache6 trans0 node_info fcs nss ps states0 in if states1 != states0 then set cache tree node states1; let () = loop fc in let fcs1 = get cache tree fc in let trans2, states2 = - eval_trans trans1 tree node fcs1 nss ps states1 states1 + eval_trans trans_cache6 trans1 node_info fcs1 nss ps states1 in if states2 != states1 then set cache tree node states2; let () = loop ns in let _, states3 = - eval_trans trans2 tree node fcs1 (get cache tree ns) ps states2 states2 + eval_trans trans_cache6 trans2 node_info fcs1 (get cache tree ns) ps states2 in if states3 != states2 then set cache tree node states3; if states0 != states3 && (not !redo) then redo := true diff --git a/src/utils.mlpack b/src/utils.mlpack index 711e03f..755f521 100644 --- a/src/utils.mlpack +++ b/src/utils.mlpack @@ -4,6 +4,8 @@ utils/FiniteCofinite utils/FiniteCofinite_sig utils/Hcons utils/Hcons_sig +utils/Hlist +utils/Hlist_sig utils/Misc utils/Pretty utils/Ptset diff --git a/src/utils/cache.ml b/src/utils/cache.ml index a38fe0c..7ba67d5 100644 --- a/src/utils/cache.ml +++ b/src/utils/cache.ml @@ -14,7 +14,7 @@ (***********************************************************************) (* - Time-stamp: + Time-stamp: *) let realloc l old_size new_size dummy = @@ -30,7 +30,7 @@ struct dummy : 'a; mutable offset : int; } - let create n a = { + let create _ a = { line = Array.create 0 a; dummy = a; offset = ~-1; @@ -241,3 +241,30 @@ struct done end + +module N6 = +struct + type 'a t = 'a N3.t N3.t + + let create _n a = + let dummy1 = N3.create 512 a in + N3.create 512 dummy1 + + let add a i j k l m n v = + let line = N3.find a i j k in + if line == N3.dummy a then + let nline = N3.create 0 (N3.dummy line) in + N3.add a i j k nline; + N3.add nline l m n v + else + N3.add line l m n v + + let find a i j k l m n = + let v = N3.find a i j k in + if v == N3.dummy a then N3.dummy (N3.dummy a) + else N3.find v l m n + + + let dummy a = N3.dummy (N3.dummy a) + let iteri _f _a = assert false +end diff --git a/src/utils/cache.mli b/src/utils/cache.mli index dba9917..d6c0ebb 100644 --- a/src/utils/cache.mli +++ b/src/utils/cache.mli @@ -14,7 +14,7 @@ (***********************************************************************) (* - Time-stamp: + Time-stamp: *) (** N-dimentional caches *) @@ -69,3 +69,13 @@ module N5 : val dummy : 'a t -> 'a val iteri : (int -> int -> int -> int -> int -> 'a -> bool -> unit) -> 'a t -> unit end + +module N6 : + sig + type 'a t + val create : int -> 'a -> 'a t + val find : 'a t -> int -> int -> int -> int -> int -> int -> 'a + val add : 'a t -> int -> int -> int -> int -> int -> int -> 'a -> unit + val dummy : 'a t -> 'a + val iteri : (int -> int -> int -> int -> int -> int -> 'a -> bool -> unit) -> 'a t -> unit + end diff --git a/src/utils/hlist.ml b/src/utils/hlist.ml new file mode 100644 index 0000000..3ffb8fc --- /dev/null +++ b/src/utils/hlist.ml @@ -0,0 +1,82 @@ +INCLUDE "utils.ml" + +include Hlist_sig + +module type HConsBuilder = + functor (H : Common_sig.HashedType) -> Hcons.S with type data = H.t + +module Builder (HCB : HConsBuilder) (H : Hcons.Abstract) : + S with type elt = H.t = +struct + type elt = H.t + + module rec Node : Hcons.S with type data = Data.t = HCB(Data) + and Data : Common_sig.HashedType with type t = (elt, Node.t) node = + struct + type t = (elt, Node.t) node + let equal x y = + match x,y with + | Nil, Nil -> true + | Cons(e1, l1), Cons(e2, l2) -> e1 == e2 && l1 == l2 + | _ -> false + + let hash = function + | Nil -> 0 + | Cons(e, l) -> HASHINT3 (PRIME1, Uid.to_int (H.uid e), Uid.to_int (Node.uid l)) + end + + include Node + + let nil = make Nil + + let rec sorted_cons e l = + match l.Node.node with + | Nil -> Node.make (Cons(e, l)) + | Cons (x, ll) -> + if H.uid e < H.uid x + then Node.make (Cons(e, l)) + else Node.make (Cons(x, sorted_cons e ll)) + + let cons e l = + Node.make(Cons(e, l)) + + let cons ?(sorted=true) e l = + if sorted then sorted_cons e l else cons e l + + let hd = function { Node.node = Cons(e, _); _ } -> e | _ -> failwith "hd" + let tl = function { Node.node = Cons(_, l); _ } -> l | _ -> failwith "tl" + + let fold f l acc = + let rec loop acc l = match l.Node.node with + | Nil -> acc + | Cons (a, aa) -> loop (f a acc) aa + in + loop acc l + + let map f l = + let rec loop l = match l.Node.node with + | Nil -> nil + | Cons(a, aa) -> cons (f a) (loop aa) + in + loop l + + let iter f l = + let rec loop l = match l.Node.node with + | Nil -> () + | Cons(a,aa) -> (f a);(loop aa) + in + loop l + + let rev l = fold cons l nil + let rev_map f l = fold (fun x acc -> cons (f x) acc) l nil + let length l = fold (fun _ c -> c+1) l 0 + let rec mem e l = + match l.Node.node with + | Nil -> false + | Cons (x, ll) -> x == e || mem e ll + +end + +module Make = Builder(Hcons.Make) +module Weak = Builder(Hcons.Weak) + diff --git a/src/utils/hlist.mli b/src/utils/hlist.mli new file mode 100644 index 0000000..80204d2 --- /dev/null +++ b/src/utils/hlist.mli @@ -0,0 +1,30 @@ +(***********************************************************************) +(* *) +(* TAToo *) +(* *) +(* Kim Nguyen, LRI UMR8623 *) +(* Université Paris-Sud & CNRS *) +(* *) +(* Copyright 2010-2012 Université Paris-Sud and Centre National de la *) +(* Recherche Scientifique. All rights reserved. This file is *) +(* distributed under the terms of the GNU Lesser General Public *) +(* License, with the special exception on linking described in file *) +(* ../LICENSE. *) +(* *) +(***********************************************************************) + +(* + Time-stamp: +*) + +include module type of Hlist_sig + +module Make (H : Hcons.Abstract) : S with type elt = H.t +(** Builds an implementation of hashconsed lists of hashconsed elements. + See {!Hcons.Make}. +*) + +module Weak (H : Hcons.Abstract) : S with type elt = H.t +(** Builds an implementation of hashconsed lists of hashconsed elements + with weak internal storage. See {!Hcons.Weak}. +*) diff --git a/src/utils/hlist_sig.ml b/src/utils/hlist_sig.ml new file mode 100644 index 0000000..227768c --- /dev/null +++ b/src/utils/hlist_sig.ml @@ -0,0 +1,35 @@ +(***********************************************************************) +(* *) +(* TAToo *) +(* *) +(* Kim Nguyen, LRI UMR8623 *) +(* Université Paris-Sud & CNRS *) +(* *) +(* Copyright 2010-2012 Université Paris-Sud and Centre National de la *) +(* Recherche Scientifique. All rights reserved. This file is *) +(* distributed under the terms of the GNU Lesser General Public *) +(* License, with the special exception on linking described in file *) +(* ../LICENSE. *) +(* *) +(***********************************************************************) + +(* + Time-stamp: +*) +type ('a,'b) node = Nil | Cons of ('a * 'b) + +module type S = sig + type elt + include Hcons.S + val nil : t + val cons : ?sorted:bool -> elt -> t -> t + val hd : t -> elt + val tl : t -> t + val fold : (elt -> 'a -> 'a) -> t -> 'a -> 'a + val map : (elt -> elt) -> t -> t + val iter : (elt -> 'a) -> t -> unit + val rev : t -> t + val rev_map : (elt -> elt) -> t -> t + val length : t -> int + val mem : elt -> t -> bool +end diff --git a/tests/xmark_small.xml.queries b/tests/xmark_small.xml.queries index a51fc3c..237bba7 100644 --- a/tests/xmark_small.xml.queries +++ b/tests/xmark_small.xml.queries @@ -14,8 +14,8 @@ B5 /site/regions/*/item[following::item]/name B6 /site/regions/*/item[preceding::item]/name B7 //person[profile/@income]/name B8 /site/open_auctions/open_auction[bidder and not(bidder/preceding-sibling::bidder)]/interval -#B9 /site/open_auctions/open_auction[(not(bidder/following::bidder) or not(bidder/preceding::bidder)) or (bidder/following::bidder and bidder/preceding::bidder)]/interval -#B10 /site/open_auctions/open_auction[(not(bidder/following::bidder) or not(bidder/preceding::bidder)) and (bidder/following::bidder and bidder/preceding::bidder)]/interval +B9 /site/open_auctions/open_auction[(not(bidder/following::bidder) or not(bidder/preceding::bidder)) or (bidder/following::bidder and bidder/preceding::bidder)]/interval +B10 /site/open_auctions/open_auction[(not(bidder/following::bidder) or not(bidder/preceding::bidder)) and (bidder/following::bidder and bidder/preceding::bidder)]/interval B11 //open_auction/bidder/../bidder/../bidder/../interval B12 //item/@id/../@id/../@id/../@id/../name B13 //keyword/ancestor::parlist/descendant::keyword/ancestor::parlist/descendant::keyword/ancestor::parlist/descendant::keyword -- 2.17.1