[JAX] add support for gather/scatter batching dims following the new …

…attributes in stablehlo.

This change also uses the new batching dims for gather/scatter batching rules, to avoid concatenating the indices with iota.

See openxla/stablehlo#2259

PiperOrigin-RevId: 647647825

tf2jax/_src/xla_utils.py

@@ -68,16 +68,24 @@ def gather_dimension_numbers_from_proto(
     message) -> jax.lax.GatherDimensionNumbers:
   proto = xla_data_pb2.GatherDimensionNumbers().FromString(message)
   return jax.lax.GatherDimensionNumbers(
-      tuple(proto.offset_dims), tuple(proto.collapsed_slice_dims),
-      tuple(proto.start_index_map))
+      tuple(proto.offset_dims),
+      tuple(proto.collapsed_slice_dims),
+      tuple(proto.start_index_map),
+      tuple(proto.operand_batching_dims),
+      tuple(proto.start_indices_batching_dims),
+  )
 
 
 def scatter_dimension_numbers_from_proto(
     message) -> jax.lax.ScatterDimensionNumbers:
   proto = xla_data_pb2.ScatterDimensionNumbers().FromString(message)
   return jax.lax.ScatterDimensionNumbers(
-      tuple(proto.update_window_dims), tuple(proto.inserted_window_dims),
-      tuple(proto.scatter_dims_to_operand_dims))
+      tuple(proto.update_window_dims),
+      tuple(proto.inserted_window_dims),
+      tuple(proto.scatter_dims_to_operand_dims),
+      tuple(proto.input_batching_dims),
+      tuple(proto.scatter_indices_batching_dims),
+  )
 
 
 def precision_config_from_proto(