Remove verification of merge buffer nodes inside device_to_device

CHANGES.md

@@ -10,6 +10,7 @@
 - Verifying that code is linked with the right contexts, by tracking `embedded_nodes` with assignments.
 - Renaming: (virtual) `device` -> `stream`, `physical_device` -> `device`.
 - New files: split out `backend_types.ml` from `backends.ml`; moved `Tnode.task` to `task.ml`; renamed `backend_utils.ml` to `c_syntax.ml`.
+- Removed half-static verification of merge buffer nodes inside `device_to_device`.
 - TODO: Moved the multicore backend from a `device = stream` model to a single device model.
 - TODO: Fixed #286: cross-stream-sharing incorporated into `Tnode.memory_mode`.
 - TODO: Built per-tensor-node stream-to-stream synchronization into device-to-device copying functions, removed obsolete blocking synchronizations.

arrayjit/lib/backend_types.ml

@@ -156,11 +156,8 @@ module type Backend = sig
         given node. If [into_merge_buffer=Streaming], remembers the buffer pointer of the source
         node to use for streaming, without blocking. If [into_merge_buffer=Copy], schedules copying
         from [src] to the merge buffer of [dst]'s stream.
-      - If the [dst] context resulted from a compilation with [Streaming] or [Copy] specific merge
-        buffer code, the [device_to_device] call should fail immediately if there's a mismatch with
-        [into_merge_buffer].
 
-      NOTE: If [into_merge_buffer:Streaming], after scheduling the work on [dst] using the merge
+      NOTE: If [into_merge_buffer=Streaming], after scheduling the work on [dst] using the merge
       buffer but before scheduling work on [src] that modifies [tn], execute
       [will_wait_for src (all_work (get_ctx_stream dst))]. *)
 
@@ -287,7 +284,7 @@ module type Lowered_backend = sig
 
   val device_to_device :
     Tnode.t -> into_merge_buffer:merge_buffer_use -> dst:context -> src:context -> bool
-  (** If the tensor node is in both contexts, copies from [dst] to [src]. *)
+  (** See {!Backend.device_to_device}. *)
 
   type buffer_ptr [@@deriving sexp_of]
 

arrayjit/lib/backends.ml

@@ -280,15 +280,6 @@ struct
 
   let device_to_device tn ~into_merge_buffer ~dst ~src =
     let dev = dst.stream in
-    if
-      (not (equal_merge_buffer_use into_merge_buffer No))
-      && not (Option.equal Tnode.equal (Some tn) dst.expected_merge_node)
-    then
-      raise
-      @@ Utils.User_error
-           ("Multicore_backend.device_to_device: merge node mismatch, expected "
-           ^ Option.(value ~default:"none" @@ map ~f:Tnode.debug_name dst.expected_merge_node)
-           ^ ", actual " ^ Tnode.debug_name tn);
     let schedule dst =
       let work =
         (* TODO: log the operation if [Utils.settings.with_log_level > 0]. *)
@@ -479,15 +470,6 @@ module Sync_backend (Backend : Backend_types.No_device_backend) : Backend_types.
 
   let device_to_device tn ~into_merge_buffer ~dst ~src =
     let dev = dst.stream in
-    if
-      (not (equal_merge_buffer_use into_merge_buffer No))
-      && not (Option.equal Tnode.equal (Some tn) dst.expected_merge_node)
-    then
-      raise
-      @@ Utils.User_error
-           ("Multicore_backend.device_to_device: merge node mismatch, expected "
-           ^ Option.(value ~default:"none" @@ map ~f:Tnode.debug_name dst.expected_merge_node)
-           ^ ", actual " ^ Tnode.debug_name tn);
     (* TODO: log the operation if [Utils.settings.with_log_level > 0]. *)
     match (Backend.get_buffer tn dst.ctx, Backend.get_buffer tn src.ctx) with
     | None, _ | _, None -> false
@@ -855,15 +837,6 @@ module Lowered_backend (Device : Backend_types.Lowered_backend) : Backend_types.
   let to_host context tn = to_host context.ctx tn
 
   let device_to_device tn ~into_merge_buffer ~dst ~src =
-    if
-      (not (equal_merge_buffer_use into_merge_buffer No))
-      && not (Option.equal Tnode.equal (Some tn) dst.expected_merge_node)
-    then
-      raise
-      @@ Utils.User_error
-           ("Multicore_backend.device_to_device: merge node mismatch, expected "
-           ^ Option.(value ~default:"none" @@ map ~f:Tnode.debug_name dst.expected_merge_node)
-           ^ ", actual " ^ Tnode.debug_name tn);
     device_to_device tn ~into_merge_buffer ~dst:dst.ctx ~src:src.ctx
 end
 

bin/moons_demo_parallel.ml

@@ -13,16 +13,16 @@ let experiment ~seed ~backend_name ~config () =
   (* Utils.set_log_level 3; *)
   (* Utils.settings.output_debug_files_in_build_directory <- true; *)
   (* Utils.settings.debug_log_from_routines <- true; *)
-  (* let hid_dim = 16 in *)
-  let hid_dim = 4 in
+  let hid_dim = 16 in
+  (* let hid_dim = 4 in *)
   (* let batch_size = 120 in *)
-  (* let batch_size = 60 in *)
-  let batch_size = 20 in
+  let batch_size = 60 in
+  (* let batch_size = 20 in *)
   let len = batch_size * 20 in
   let init_lr = 0.1 in
   (* let epochs = 10 in *)
-  (* let epochs = 40 in *)
-  let epochs = 1 in
+  let epochs = 40 in
+  (* let epochs = 1 in *)
   let noise () = Rand.float_range (-0.1) 0.1 in
   let moons_flat =
     Array.concat_map (Array.create ~len ())

lib/train.ml

@@ -364,14 +364,16 @@ let%track3_sexp parallel_update (type context)
         let grad_merge =
           Option.value_exn ~here:[%here] ~message:(Tn.debug_name p.value) grad_merges_to.(to_).(i)
         in
+        (* NOTE: we no longer have to to pass [grad_merge.context] as [dst]. *)
         assert (
           Backend.device_to_device (Option.value_exn ~here:[%here] p.diff).grad ~into_merge_buffer
-            ~dst:grad_merge.context ~src:ctxs.(from));
+            ~dst:ctxs.(to_) ~src:ctxs.(from));
         (Task.run grad_merge.schedule : unit))
   in
   let merge_loss ~src =
+    (* NOTE: we no longer have to to pass [loss_merge.context] as [dst]. *)
     assert (
-      Backend.device_to_device updaten.loss.value ~into_merge_buffer ~dst:loss_merge.context ~src);
+      Backend.device_to_device updaten.loss.value ~into_merge_buffer ~dst:sgd_update.context ~src);
     Task.run loss_merge.schedule
   in
   (* FIXME: missing backcopy. *)