Add a transaction benchmark using a naïve unscalable leftist heap

bench/bench_leftist_heap.ml

@@ -0,0 +1,139 @@
+open Kcas
+
+(** This is a naïve transactional leftist heap implementation.
+
+    ⚠️ This simply cannot scale, because essentially every delete min operation
+    updates the root.  This can still be a useful benchmark of the transaction
+    mechanism, but please do not believe for a second that this is a scalable
+    concurrent priority queue. *)
+module Leftist_heap : sig
+  type 'a t
+
+  val create : lt:('a -> 'a -> bool) -> 'a t
+
+  module Xt : sig
+    val insert : xt:'x Xt.t -> 'a t -> 'a -> unit
+    val delete_min_opt : xt:'x Xt.t -> 'a t -> 'a option
+  end
+
+  val insert : 'a t -> 'a -> unit
+  val delete_min_opt : 'a t -> 'a option
+end = struct
+  type 'a t = { lt : 'a -> 'a -> bool; root : 'a link Loc.t }
+
+  and ('a, _) tdt =
+    | Null : ('a, [> `Null ]) tdt
+    | Node :
+        'a link Loc.t * int Loc.t * 'a * 'a link Loc.t
+        -> ('a, [> `Node ]) tdt
+
+  and 'a link = Link : ('a, [< `Null | `Node ]) tdt -> 'a link [@@unboxed]
+
+  let create ~lt =
+    let root = Loc.make ~padded:true (Link Null) in
+    Multicore_magic.copy_as_padded { lt; root }
+
+  module Xt = struct
+    let npl_of ~xt = function
+      | Link Null -> 0
+      | Link (Node (_, npl, _, _)) -> Xt.get ~xt npl
+
+    let rec merge ~xt ~lt h1 h2 =
+      match (h1, h2) with
+      | Link Null, h2 -> h2
+      | h1, Link Null -> h1
+      | Link (Node (_, _, v1, _) as h1), Link (Node (_, _, v2, _) as h2) ->
+          let (Node (h1l, npl, _, h1r) as h1 : (_, [ `Node ]) tdt), h2 =
+            if lt v1 v2 then (h1, h2) else (h2, h1)
+          in
+          let l = Xt.get ~xt h1l in
+          if l == Link Null then Xt.set ~xt h1l (Link h2)
+          else begin
+            let r = merge ~xt ~lt (Xt.get ~xt h1r) (Link h2) in
+            match (npl_of ~xt l, npl_of ~xt r) with
+            | l_npl, r_npl when l_npl < r_npl ->
+                Xt.set ~xt h1l r;
+                Xt.set ~xt h1r l;
+                Xt.set ~xt npl (l_npl + 1)
+            | _, r_npl ->
+                Xt.set ~xt h1r r;
+                Xt.set ~xt npl (r_npl + 1)
+          end;
+          Link h1
+
+    let insert ~xt h x =
+      let h1 =
+        Node (Loc.make (Link Null), Loc.make 1, x, Loc.make (Link Null))
+      in
+      Xt.set ~xt h.root (merge ~xt ~lt:h.lt (Link h1) (Xt.get ~xt h.root))
+
+    let delete_min_opt ~xt h =
+      match Xt.get ~xt h.root with
+      | Link Null -> None
+      | Link (Node (h1, _, x, h2)) ->
+          Xt.set ~xt h.root (merge ~xt ~lt:h.lt (Xt.get ~xt h1) (Xt.get ~xt h2));
+          Some x
+  end
+
+  let insert h x = Kcas.Xt.commit { tx = Xt.insert h x }
+  let delete_min_opt h = Kcas.Xt.commit { tx = Xt.delete_min_opt h }
+end
+
+open Multicore_bench
+
+let run_one ~budgetf ~n_domains ~preload () =
+  let n_ops =
+    Float.to_int
+      (Float.of_int (50 * Util.iter_factor) /. Float.log2 (Float.of_int preload))
+  in
+
+  let t = Leftist_heap.create ~lt:(( < ) : int -> int -> bool) in
+
+  let n_ops_todo = Atomic.make 0 |> Multicore_magic.copy_as_padded in
+
+  let before () =
+    Atomic.set n_ops_todo n_ops;
+    while Option.is_some @@ Leftist_heap.delete_min_opt t do
+      ()
+    done;
+    let state = Random.State.make_self_init () in
+    for _ = 1 to preload do
+      Leftist_heap.insert t (Random.State.bits state)
+    done
+  in
+  let init _ = Random.State.make_self_init () in
+  let work _ state =
+    let rec work () =
+      let n = Util.alloc ~batch:100 n_ops_todo in
+      if n <> 0 then
+        let rec loop n =
+          if 0 < n then
+            let value = Random.State.bits state in
+            if value land 1 = 0 then begin
+              Leftist_heap.insert t value;
+              loop (n - 1)
+            end
+            else begin
+              Leftist_heap.delete_min_opt t |> ignore;
+              loop (n - 1)
+            end
+          else work ()
+        in
+        loop n
+    in
+    work ()
+  in
+
+  let config =
+    Printf.sprintf "%d worker%s, %d preload" n_domains
+      (if n_domains = 1 then "" else "s")
+      preload
+  in
+
+  Times.record ~budgetf ~n_domains ~before ~init ~work ()
+  |> Times.to_thruput_metrics ~n:n_ops ~singular:"operation" ~config
+
+let run_suite ~budgetf =
+  Util.cross [ 10; 100; 1000 ] [ 1; 2; 4 ]
+  |> List.concat_map @@ fun (preload, n_domains) ->
+     run_one ~budgetf ~n_domains ~preload ()

bench/main.ml

@@ -3,6 +3,7 @@ let benchmarks =
     ("Kcas Loc", Bench_loc.run_suite);
     ("Kcas Xt", Bench_xt.run_suite);
     ("Kcas Xt read-only", Bench_xt_ro.run_suite);
+    ("Kcas Xt Leftist_heap (unscalable)", Bench_leftist_heap.run_suite);
     ("Kcas parallel CMP", Bench_parallel_cmp.run_suite);
     ("Kcas_data Accumulator", Bench_accumulator.run_suite);
     ("Kcas_data Dllist", Bench_dllist.run_suite);