Cherrypick #8524 into r2.6

test/scan/test_scan.py

@@ -9,6 +9,7 @@
 from torch.utils._pytree import tree_map, tree_flatten, tree_iter, tree_leaves, PyTree
 
 import torch_xla
+import torch_xla.debug.metrics as met
 from torch_xla.experimental.scan import scan, value_and_grad_partitioned, tree_flatten_none
 
 parent_folder = os.path.dirname(os.path.dirname(__file__))
@@ -178,6 +179,29 @@ def fn(carry, x):
                       device=self.device)
     self.run_test(fn, init, xs)
 
+  def test_scan_create_tensors_no_transfers_from_device(self):
+    """Test that scanning over a function that internally creates tensors
+    will not transfer those tensor to host, which can be potentially expensive.
+    """
+
+    def fn(carry, x):
+      a = torch.tensor([1.0, 2.0], device=self.device)
+      b = torch.tensor([3.0, 4.0], device=self.device)
+      return carry + a, x + b
+
+    init = torch.tensor([0.0, 0.0], requires_grad=True, device=self.device)
+    xs = torch.tensor([[1.0, 2.0], [3.0, 4.0], [5.0, 6.0]],
+                      requires_grad=True,
+                      device=self.device)
+    met.clear_all()
+    self.assertFalse(met.metric_data("TransferFromDeviceTime"))
+    # Use `scan` to lower `fn` into HLO and run it. Doing so should not
+    # transfer anything from the device to host. `carry` and `ys` still
+    # reference data on the device.
+    carry, ys = scan(fn, init, xs)
+    torch_xla.sync(wait=True)
+    self.assertFalse(met.metric_data("TransferFromDeviceTime"))
+
   def test_scan_internal_in_place_mutation(self):
     """
     Test internal in-place mutations inside the `fn` to be scanned over.

torch_xla/experimental/scan.py

@@ -521,7 +521,8 @@ def make_fake_tensor(v: torch.Tensor) -> torch.Tensor:
     builder.add_param(val)
 
   # Detect hoisted variables.
-  hoisted_vars: Dict[int, torch.Tensor] = fn_ctx.parameter_id_tensor_mapping()
+  hoisted_vars: Dict[
+      int, torch.Tensor] = fn_ctx.device_parameter_id_tensor_mapping()
   for v in itertools.chain(fake_carry, fake_x):
     param_id = fn_ctx.tensor_parameter_id(v)
     if param_id != -1: