Do more aggressive lambda lifting

With @vouillon we realized that the `Lambda_lifting_simple` pass that is
performed by double translation makes some programs significantly
faster. We measured roughly a 1.45 speedup on a large (proprietary)
Bonsai benchmark. Presumably V8 is much faster with more toplevel
functions and less nested closures.

Author: OlivierNicole

compiler/lib/driver.ml

@@ -165,6 +165,17 @@ let effects_and_exact_calls
       Deadcode.f pure_fun p
     else Deadcode.f pure_fun p
   in
+  let p =
+    match Config.(target (), effects ()) with
+    | `JavaScript, `Disabled ->
+        (* If effects are disabled, we lambda-lift aggressively. While not
+           necessary, it results in a substantial gain in performance in some
+           programs in Javascript. *)
+        let to_lift = all_functions p in
+        let p, _ = Lambda_lifting_simple.f ~to_lift p in
+        p
+    | _ -> p
+  in
   match Config.effects () with
   | `Cps | `Double_translation ->
       if debug () then Format.eprintf "Effects...@.";
@@ -696,6 +707,16 @@ let link_and_pack ?(standalone = true) ?(wrap_with_fun = `Iife) ?(link = `No) p
   |> pack ~wrap_with_fun ~standalone
   |> check_js
 
+let all_functions p =
+  let open Code in
+  fold_closures
+    p
+    (fun name _ _ _ acc ->
+      match name with
+      | Some name -> Var.Set.add name acc
+      | None -> acc)
+    Var.Set.empty
+
 let optimize ~shapes ~profile ~keep_flow_data p =
   let deadcode_sentinal =
     (* If deadcode is disabled, this field is just fresh variable *)