Support torch.distributed.scatter collective

test/pjrt/test_collective_ops_tpu.py

@@ -103,6 +103,29 @@ def test_reduce_scatter(self, pin_layout):
     for ordinal, value in results.items():
       np.testing.assert_array_equal(value, [-ordinal])
 
+  @staticmethod
+  def _scatter():
+    dist.init_process_group("xla", init_method='xla://')
+    device = torch_xla.device()
+    world_size = xr.world_size()
+    tensors = None
+    if xr.global_ordinal() == 0:
+      tensors = [
+          torch.tensor([i], device=device, dtype=torch.float)
+          for i in range(world_size)
+      ]
+
+    output_tensor = torch.tensor([-1], dtype=torch.float, device=device)
+    dist.scatter(output_tensor, tensors, src=0)
+    return output_tensor.cpu()
+
+  def test_scatter(self):
+    """self._scatter instantiates a list of tensors [[0], [1], ..., [n-1]]
+    on device 0, then scatters it. Device i should therefore receive [i]."""
+    results = pjrt.run_multiprocess(self._scatter)
+    for ordinal, value in results.items():
+      np.testing.assert_array_equal(value, [ordinal])
+
   @staticmethod
   def _all_to_all(pin_layout):
     device = torch_xla.device()

test/test_torch_distributed_xla_backend.py

@@ -360,7 +360,6 @@ def test_barrier(self):
       'allreduce_coalesced',
       'alltoall',
       'gather',
-      'scatter',
       'recv_anysource',
       'monitored_barrier',
   )

torch_xla/distributed/xla_backend.py

@@ -5,7 +5,7 @@
 from torch_xla._internal import rendezvous
 import logging
 import os
-from torch._C._distributed_c10d import ProcessGroup, AllgatherOptions
+from torch._C._distributed_c10d import ProcessGroup, ScatterOptions, ReduceScatterOptions, AllgatherOptions
 
 
 def _create_xla_process_group(prefix_store, rank, size, timeout):
@@ -253,8 +253,22 @@ def alltoall_base(self, output, input, output_split_sizes, input_split_sizes,
   def gather(self, *args):
     raise NotImplementedError
 
-  def scatter(self, *args):
-    raise NotImplementedError
+  # Called by torch.distributed.scatter. Call site example:
+  # https://github.com/pytorch/pytorch/blob/v2.7.1/torch/distributed/distributed_c10d.py#L4146
+  # Input tensors are defined on the source device and scattered
+  # to the output tensors.
+  def scatter(self, output_tensor_list: list[torch.Tensor],
+              input_tensors_list: list[list[torch.Tensor]],
+              opts: ScatterOptions):
+    if xr.global_ordinal() == opts.rootRank:
+      inputs = input_tensors_list
+    else:
+      inputs = [[torch.zeros_like(output_tensor)] * xr.world_size()
+                for output_tensor in output_tensor_list]
+
+    rs_opts = ReduceScatterOptions()
+    rs_opts.reduceOp = dist.ReduceOp.SUM
+    return self.reduce_scatter(output_tensor_list, inputs, rs_opts)
 
   # Dummy channel id maker. Different backend (TPU, GPU, etc) should replace
   # the maker with their specific one. See unit test in