Add a padding_spec argument into tpu.replicate, which provides the functionality of auto bucketing inputs.

PiperOrigin-RevId: 295226850
Change-Id: I486a82739bd7d2629b218371f5b50f1e37cb4b2d

tensorflow/python/distribute/central_storage_strategy.py

@@ -161,7 +161,7 @@ class CentralStorageStrategy(distribute_lib.Strategy):
     """
     return super(CentralStorageStrategy, self).experimental_local_results(value)
 
-  def experimental_run_v2(self, fn, args=(), kwargs=None):  # pylint: disable=useless-super-delegation
+  def experimental_run_v2(self, fn, args=(), kwargs=None, options=None):  # pylint: disable=useless-super-delegation
     """Run `fn` on each replica, with the given arguments.
 
     In `CentralStorageStrategy`, `fn` is  called on each of the compute
@@ -171,12 +171,14 @@ class CentralStorageStrategy(distribute_lib.Strategy):
       fn: The function to run. The output must be a `tf.nest` of `Tensor`s.
       args: (Optional) Positional arguments to `fn`.
       kwargs: (Optional) Keyword arguments to `fn`.
+      options: (Optional) An instance of `tf.distribute.RunOptions` specifying
+        the options to run `fn`.
 
     Returns:
       Return value from running `fn`.
     """
-    return super(CentralStorageStrategy, self).experimental_run_v2(fn, args,
-                                                                   kwargs)
+    return super(CentralStorageStrategy,
+                 self).experimental_run_v2(fn, args, kwargs, options)
 
   def reduce(self, reduce_op, value, axis):  # pylint: disable=useless-super-delegation
     """Reduce `value` across replicas.

tensorflow/python/distribute/custom_training_loop_input_test.py

@@ -23,6 +23,7 @@ from absl.testing import parameterized
 from tensorflow.python import tf2
 from tensorflow.python.data.ops import dataset_ops
 from tensorflow.python.distribute import combinations
+from tensorflow.python.distribute import distribute_lib
 from tensorflow.python.distribute import reduce_util
 from tensorflow.python.distribute import strategy_combinations
 from tensorflow.python.eager import def_function
@@ -333,6 +334,31 @@ class InputIterationTest(test.TestCase, parameterized.TestCase,
     # This assumes that there are exactly 2 replicas
     self.assertAllEqual([5.5, 7.], run(input_iterator))
 
+  @combinations.generate(
+      combinations.combine(
+          distribution=strategy_combinations.multidevice_strategies,
+          mode=["eager"]))
+  def testDynamicShapesWithRunOptions(self, distribution):
+    dataset = get_dataset_from_tensor_slices([5., 6., 7.]).batch(4)
+    input_iterator = iter(distribution.experimental_distribute_dataset(dataset))
+    options = distribute_lib.RunOptions
+    options.experimental_bucketizing_dynamic_shape = True
+
+    @def_function.function
+    def run(iterator):
+
+      def computation(x):
+        return math_ops.reduce_mean(x)
+
+      inputs = next(iterator)
+      outputs = distribution.experimental_local_results(
+          distribution.experimental_run_v2(
+              computation, args=(inputs,), options=options))
+      return outputs
+
+    # This assumes that there are exactly 2 replicas
+    self.assertAllEqual([5.5, 7.], run(input_iterator))
+
   @combinations.generate(
       combinations.combine(
           distribution=strategy_combinations.multidevice_strategies,

tensorflow/python/distribute/distribute_lib.py

@@ -95,6 +95,7 @@ from __future__ import absolute_import
 from __future__ import division
 from __future__ import print_function
 
+import collections
 import copy
 import enum  # pylint: disable=g-bad-import-order
 import threading
@@ -411,6 +412,35 @@ class InputContext(object):
         self.input_pipeline_id, self.num_input_pipelines)
 
 
+@tf_export("distribute.RunOptions")
+class RunOptions(
+    collections.namedtuple("RunOptions", [
+        "experimental_enable_dynamic_batch_size",
+        "experimental_bucketizing_dynamic_shape",
+    ])):
+  """Run options for `strategy.experimental_run_v2`.
+
+  This can be used to hold some strategy specific configs.
+
+  Attributes:
+    experimental_enable_dynamic_batch_size: Boolean. Only applies to
+      TPUStrategy. Default to False. If True, TPUStrategy will enable dynamic
+      padder to support dynamic batch size for the inputs. Otherwise only static
+      shape inputs are allowed.
+    experimental_bucketizing_dynamic_shape: Boolean. Only applies to
+      TPUStrategy. Default to False. If True, TPUStrategy will automatic
+      bucketize inputs passed into `experimental_run_v2` if the input shape is
+      dynamic. This is a performance optimization to reduce XLA recompilation,
+      which should not have impact on correctness.
+  """
+
+  def __new__(cls,
+              experimental_enable_dynamic_batch_size=True,
+              experimental_bucketizing_dynamic_shape=False):
+    return super(RunOptions,
+                 cls).__new__(cls, experimental_enable_dynamic_batch_size,
+                              experimental_bucketizing_dynamic_shape)
+
 # ------------------------------------------------------------------------------
 # Base classes for all distribution strategies.
 
@@ -778,7 +808,7 @@ class StrategyBase(object):
     return self._extended._experimental_distribute_datasets_from_function(  # pylint: disable=protected-access
         dataset_fn)
 
-  def experimental_run_v2(self, fn, args=(), kwargs=None):
+  def experimental_run_v2(self, fn, args=(), kwargs=None, options=None):
     """Run `fn` on each replica, with the given arguments.
 
     Executes ops specified by `fn` on each replica. If `args` or `kwargs` have
@@ -800,6 +830,8 @@ class StrategyBase(object):
       fn: The function to run. The output must be a `tf.nest` of `Tensor`s.
       args: (Optional) Positional arguments to `fn`.
       kwargs: (Optional) Keyword arguments to `fn`.
+      options: (Optional) An instance of `tf.distribute.RunOptions` specifying
+        the options to run `fn`.
 
     Returns:
       Merged return value of `fn` across replicas. The structure of the return
@@ -807,6 +839,8 @@ class StrategyBase(object):
       structure can either be "per-replica" `Tensor` objects or `Tensor`s
       (for example, if running on a single replica).
     """
+    del options
+
     if not isinstance(args, (list, tuple)):
       raise ValueError(
           "positional args must be a list or tuple, got {}".format(type(args)))

tensorflow/python/distribute/one_device_strategy.py

@@ -163,7 +163,7 @@ class OneDeviceStrategy(distribute_lib.Strategy):
     """
     return super(OneDeviceStrategy, self).experimental_local_results(value)
 
-  def experimental_run_v2(self, fn, args=(), kwargs=None):  # pylint: disable=useless-super-delegation
+  def experimental_run_v2(self, fn, args=(), kwargs=None, options=None):  # pylint: disable=useless-super-delegation
     """Run `fn` on each replica, with the given arguments.
 
     In `OneDeviceStrategy`, `fn` is simply called within a device scope for the
@@ -173,11 +173,14 @@ class OneDeviceStrategy(distribute_lib.Strategy):
       fn: The function to run. The output must be a `tf.nest` of `Tensor`s.
       args: (Optional) Positional arguments to `fn`.
       kwargs: (Optional) Keyword arguments to `fn`.
+      options: (Optional) An instance of `tf.distribute.RunOptions` specifying
+        the options to run `fn`.
 
     Returns:
       Return value from running `fn`.
     """
-    return super(OneDeviceStrategy, self).experimental_run_v2(fn, args, kwargs)
+    return super(OneDeviceStrategy,
+                 self).experimental_run_v2(fn, args, kwargs, options)
 
   def reduce(self, reduce_op, value, axis):  # pylint: disable=useless-super-delegation
     """Reduce `value` across replicas.

tensorflow/python/distribute/tpu_strategy.py

@@ -158,14 +158,15 @@ class TPUStrategy(distribute_lib.Strategy):
   # TODO(cjfj): Modify `_call_for_each_replica` in `TPUExtended` such that this
   # can use the default implementation.
   # This implementation runs a single step. It does not use infeed or outfeed.
-  def experimental_run_v2(self, fn, args=(), kwargs=None):
+  def experimental_run_v2(self, fn, args=(), kwargs=None, options=None):
     """See base class."""
     validate_experimental_run_function(fn)
 
     # Note: the target function is converted to graph even when in Eager mode,
     # so autograph is on by default here.
     fn = autograph.tf_convert(fn, autograph_ctx.control_status_ctx())
-    return self.extended.tpu_run(fn, args, kwargs)
+    options = options or distribute_lib.RunOptions()
+    return self.extended.tpu_run(fn, args, kwargs, options)
 
 
 @tf_export(v1=["distribute.experimental.TPUStrategy"])
@@ -206,12 +207,62 @@ class TPUStrategyV1(distribute_lib.StrategyV1):
   # TODO(cjfj): Modify `_call_for_each_replica` in `TPUExtended` such that this
   # can use the default implementation.
   # This implementation runs a single step. It does not use infeed or outfeed.
-  def experimental_run_v2(self, fn, args=(), kwargs=None):
-    """See base class."""
+  def experimental_run_v2(self, fn, args=(), kwargs=None, options=None):
+    """Run `fn` on each replica, with the given arguments.
+
+    Executes ops specified by `fn` on each replica. If `args` or `kwargs` have
+    "per-replica" values, such as those produced by a "distributed `Dataset`",
+    when `fn` is executed on a particular replica, it will be executed with the
+    component of those "per-replica" values that correspond to that replica.
+
+    `fn` may call `tf.distribute.get_replica_context()` to access members such
+    as `all_reduce`.
+
+    All arguments in `args` or `kwargs` should either be nest of tensors or
+    per-replica objects containing tensors or composite tensors.
+
+    Users can pass strategy specific options to `options` argument. An example
+    to enable bucketizing dynamic shapes in `TPUStrategy.experimental_run_v2`
+    is:
+    ```python
+
+    resolver = tf.distribute.cluster_resolver.TPUClusterResolver(tpu='')
+    tf.config.experimental_connect_to_cluster(resolver)
+    tf.tpu.experimental.initialize_tpu_system(resolver)
+    strategy = tf.distribute.experimental.TPUStrategy(tpu='')
+
+    options = tf.distribute.RunOptions()
+    options.experimental_bucketizing_dynamic_shape = True
+
+    iterator = iter(inputs)
+
+    @tf.function()
+    def step_fn(inputs):
+      output = tf.reduce_sum(inputs)
+      return output
+
+      strategy.experimental_run_v2(step_fn, args=(next(iterator),),
+                                   options=options)
+    ```
+
+    Args:
+      fn: The function to run. The output must be a `tf.nest` of `Tensor`s.
+      args: (Optional) Positional arguments to `fn`.
+      kwargs: (Optional) Keyword arguments to `fn`.
+      options: (Optional) An instance of `tf.distribute.RunOptions` specifying
+        the options to run `fn`.
+
+    Returns:
+      Merged return value of `fn` across replicas. The structure of the return
+      value is the same as the return value from `fn`. Each element in the
+      structure can either be "per-replica" `Tensor` objects or `Tensor`s
+      (for example, if running on a single replica).
+    """
     validate_experimental_run_function(fn)
 
     fn = autograph.tf_convert(fn, autograph_ctx.control_status_ctx())
-    return self.extended.tpu_run(fn, args, kwargs)
+    options = options or distribute_lib.RunOptions()
+    return self.extended.tpu_run(fn, args, kwargs, options)
 
 
 # TODO(josh11b): Switch to V2 when we no longer need to support tf.compat.v1.
@@ -288,7 +339,6 @@ class TPUExtended(distribute_lib.StrategyExtendedV1):
     self._retrace_functions_for_each_device = False
 
     self.experimental_enable_get_next_as_optional = True
-    self.experimental_enable_dynamic_batch_size = True
     self._prefetch_on_host = False
 
     self._logical_device_stack = [0]
@@ -801,11 +851,11 @@ class TPUExtended(distribute_lib.StrategyExtendedV1):
     """
     return True
 
-  def tpu_run(self, fn, args, kwargs):
-    func = self._tpu_function_creator(fn)
+  def tpu_run(self, fn, args, kwargs, options=None):
+    func = self._tpu_function_creator(fn, options)
     return func(args, kwargs)
 
-  def _tpu_function_creator(self, fn):
+  def _tpu_function_creator(self, fn, options):
     if fn in self._tpu_function_cache:
       return self._tpu_function_cache[fn]
 
@@ -844,7 +894,7 @@ class TPUExtended(distribute_lib.StrategyExtendedV1):
 
       # Construct and pass `maximum_shapes` so that we could support dynamic
       # shapes using dynamic padder.
-      if self.experimental_enable_dynamic_batch_size and replicate_inputs:
+      if options.experimental_enable_dynamic_batch_size and replicate_inputs:
         maximum_shapes = []
         flattened_list = nest.flatten(replicate_inputs[0])
         for input_tensor in flattened_list:
@@ -859,12 +909,18 @@ class TPUExtended(distribute_lib.StrategyExtendedV1):
       else:
         maximum_shapes = None
 
+      if options.experimental_bucketizing_dynamic_shape:
+        padding_spec = tpu.PaddingSpec.POWER_OF_TWO
+      else:
+        padding_spec = None
+
       with strategy.scope():
         replicate_outputs = tpu.replicate(
             replicated_fn,
             replicate_inputs,
             device_assignment=self._device_assignment,
-            maximum_shapes=maximum_shapes)
+            maximum_shapes=maximum_shapes,
+            padding_spec=padding_spec)
 
       # Remove all no ops that may have been added during 'tpu.replicate()'
       if isinstance(result[0], list):

tensorflow/python/tpu/tpu.py

@@ -20,6 +20,8 @@ from __future__ import division
 from __future__ import print_function
 
 from absl import logging
+import enum
+
 import numpy as np
 from six.moves import xrange  # pylint: disable=redefined-builtin
 
@@ -784,15 +786,58 @@ def outside_compilation(computation, *args, **kwargs):
   return retval
 
 
+@tf_export(v1=["tpu.PaddingSpec"])
+class PaddingSpec(enum.IntEnum):
+  """Represents the type of padding policies for tpu.replicate."""
+  # By default the policy is set to AUTO, the dynamic input shape dimension will
+  # be pad to maximum of all the replicas.
+  AUTO = 0
+  # Bucketize the dynamic input shape dimension into a power of 2.
+  POWER_OF_TWO = 1
+
+
 @tf_export(v1=["tpu.replicate"])
 def replicate(computation,
               inputs=None,
               infeed_queue=None,
               device_assignment=None,
               name=None,
-              maximum_shapes=None):
+              maximum_shapes=None,
+              padding_spec=None):
   """Builds a graph operator that runs a replicated TPU computation.
 
+  Example for the basic usage that `inputs` has static shape:
+
+  ```python
+
+  def computation(x):
+    x = x + 1
+    return tf.math.reduce_mean(x)
+
+  x = tf.convert_to_tensor([1., 2., 3.])
+  y = tf.convert_to_tensor([4., 5., 6.])
+  tf.compat.v1.tpu.replicate(computation, inputs=[[x], [y]])
+  ```
+
+  If the `inputs` has dynamic shapes and you would like to automatically
+  bucketize the inputs to avoid XLA recompilation. See the advanced example
+  below:
+
+  ```python
+
+  def computation(x):
+    x = x + 1
+    return tf.math.reduce_mean(x)
+
+  # Assume input tensors in two replicas `x` and `y` both have dynamic shape
+  # ([None, 2]).
+  tf.compat.v1.tpu.replicate(
+    computation,
+    inputs=[x, y],
+    maximum_shapes=[tf.TensorShape([None, None])],
+    padding_spec=tf.compat.v1.tpu.PaddingSpec.POWER_OF_TWO)
+  ```
+
   Args:
     computation: A Python function that builds the computation to replicate.
     inputs: A list of lists of input tensors or `None` (equivalent to
@@ -818,6 +863,11 @@ def replicate(computation,
       object) will be padded to the maximum size of that dimension over all
       replicas. The structure of `maximum_shapes` needs to be the same as
       `inputs[0]`.
+    padding_spec: An enum specified by `tpu.PaddingSpec`. This describes the
+      padding policy when the `inputs` to `tpu.replicate` is dynamic.
+      One usage is to enable automatic bucketizing on the inputs by setting the
+      value to `tpu.PaddingSpec.POWER_OF_TWO`, which can help to reduce the
+      recompilation in the XLA side.
   Returns:
     A list of outputs, indexed by `[replica_num]` each output can be a nested
     structure same as what computation() returns with a few exceptions.
@@ -845,10 +895,21 @@ def replicate(computation,
       infeed_queue,
       device_assignment,
       name,
-      maximum_shapes=maximum_shapes)[1]
+      maximum_shapes=maximum_shapes,
+      padding_spec=padding_spec)[1]
 
 
-def _pad_all_input(inputs, padded_shapes):
+def _ceil_to_pow_of_n(x, n):
+  """Ceil input `x` to power of `n`."""
+  x = math_ops.cast(x, dtypes.float32)
+  lognx = math_ops.log(x) / math_ops.log(n * 1.0)
+  lognx = math_ops.ceil(lognx)
+  result = math_ops.pow(n * 1.0, lognx)
+  result = math_ops.cast(result, dtypes.int32)
+  return result
+
+
+def _pad_all_input(inputs, padded_shapes, padding_spec):
   """Pad all input tensors given padded_shapes.
 
   The real shape tensors will be concatenated with the padded original inputs.
@@ -856,6 +917,11 @@ def _pad_all_input(inputs, padded_shapes):
   Args:
     inputs: The original inputs.
     padded_shapes: A list of padded shapes for each input.
+    padding_spec: An enum specified by `tpu.PaddingSpec`. This describes the
+      padding policy when the `inputs` to `tf.tpu.replicate` is dynamic.
+      One usage is to enable automatic bucketizing on the inputs by setting the
+      value to `tpu.PaddingSpec.POWER_OF_TWO`, which can help to reduce the
+      recompilation in the XLA side.
 
   Returns:
     The padded inputs and a PaddingMap list which maps the padded input
@@ -933,6 +999,8 @@ def _pad_all_input(inputs, padded_shapes):
               # among all the cores.
               max_dim_size = math_ops.maximum(maximum_shapes[idx][i],
                                               minimum_dynamic_dim_size)
+              if padding_spec == PaddingSpec.POWER_OF_TWO:
+                max_dim_size = _ceil_to_pow_of_n(max_dim_size, 2)
             # Pad to the given maximum value.
             padding = [0, max_dim_size - input_shape_tensor[i]]
           else:
@@ -972,7 +1040,8 @@ def split_compile_and_replicate(computation,
                                 device_assignment=None,
                                 name=None,
                                 use_tpu=True,
-                                maximum_shapes=None):
+                                maximum_shapes=None,
+                                padding_spec=None):
   """Builds graph operators that runs compilation and replicated computation.
 
   This is a lower level interface than replicate that returns a separate compile
@@ -1009,6 +1078,11 @@ def split_compile_and_replicate(computation,
       object) will be padded to the maximum size of that dimension over all
       replicas. The structure of `maximum_shapes` needs to be the same as
       `inputs[0]`.
+    padding_spec: An enum specified by `tf.tpu.PaddingSpec`. This describes the
+      padding policy when the `inputs` to `tf.tpu.replicate` is dynamic.
+      One usage is to enable automatic bucketizing on the inputs by setting the
+      value to `tpu.PaddingSpec.POWER_OF_TWO`, which can help to reduce the
+      recompilation in the XLA side.
 
   Returns:
     A list of lists with the first list corresponding to the compile op and the
@@ -1108,7 +1182,8 @@ def split_compile_and_replicate(computation,
         tensor_shape.TensorShape(s) for s in flat_maximum_shapes
     ]
 
-    flat_inputs, padding_maps = _pad_all_input(flat_inputs, flat_maximum_shapes)
+    flat_inputs, padding_maps = _pad_all_input(flat_inputs, flat_maximum_shapes,
+                                               padding_spec)
 
     serialized_padding_maps = []
     for padding_map in padding_maps:

tensorflow/tools/api/golden/v1/tensorflow.distribute.-mirrored-strategy.pbtxt

@@ -46,7 +46,7 @@ tf_class {
   }
   member_method {
     name: "experimental_run_v2"
-    argspec: "args=[\'self\', \'fn\', \'args\', \'kwargs\'], varargs=None, keywords=None, defaults=[\'()\', \'None\'], "
+    argspec: "args=[\'self\', \'fn\', \'args\', \'kwargs\', \'options\'], varargs=None, keywords=None, defaults=[\'()\', \'None\', \'None\'], "
   }
   member_method {
     name: "group"

tensorflow/tools/api/golden/v1/tensorflow.distribute.-one-device-strategy.pbtxt

@@ -46,7 +46,7 @@ tf_class {
   }
   member_method {
     name: "experimental_run_v2"
-    argspec: "args=[\'self\', \'fn\', \'args\', \'kwargs\'], varargs=None, keywords=None, defaults=[\'()\', \'None\'], "
+    argspec: "args=[\'self\', \'fn\', \'args\', \'kwargs\', \'options\'], varargs=None, keywords=None, defaults=[\'()\', \'None\', \'None\'], "
   }
   member_method {
     name: "group"

tensorflow/tools/api/golden/v1/tensorflow.distribute.-run-options.pbtxt

@@ -0,0 +1,23 @@
+path: "tensorflow.distribute.RunOptions"
+tf_class {
+  is_instance: "<class \'tensorflow.python.distribute.distribute_lib.RunOptions\'>"
+  is_instance: "<class \'tensorflow.python.distribute.distribute_lib.RunOptions\'>"
+  is_instance: "<type \'tuple\'>"
+  member {
+    name: "experimental_bucketizing_dynamic_shape"
+    mtype: "<type \'property\'>"
+  }
+  member {
+    name: "experimental_enable_dynamic_batch_size"
+    mtype: "<type \'property\'>"
+  }
+  member_method {
+    name: "__init__"
+  }
+  member_method {
+    name: "count"
+  }
+  member_method {
+    name: "index"
+  }
+}

tensorflow/tools/api/golden/v1/tensorflow.distribute.-strategy.pbtxt

@@ -45,7 +45,7 @@ tf_class {
   }
   member_method {
     name: "experimental_run_v2"
-    argspec: "args=[\'self\', \'fn\', \'args\', \'kwargs\'], varargs=None, keywords=None, defaults=[\'()\', \'None\'], "
+    argspec: "args=[\'self\', \'fn\', \'args\', \'kwargs\', \'options\'], varargs=None, keywords=None, defaults=[\'()\', \'None\', \'None\'], "
   }
   member_method {
     name: "group"

tensorflow/tools/api/golden/v1/tensorflow.distribute.experimental.-central-storage-strategy.pbtxt

@@ -46,7 +46,7 @@ tf_class {
   }
   member_method {
     name: "experimental_run_v2"
-    argspec: "args=[\'self\', \'fn\', \'args\', \'kwargs\'], varargs=None, keywords=None, defaults=[\'()\', \'None\'], "
+    argspec: "args=[\'self\', \'fn\', \'args\', \'kwargs\', \'options\'], varargs=None, keywords=None, defaults=[\'()\', \'None\', \'None\'], "
   }
   member_method {
     name: "group"

tensorflow/tools/api/golden/v1/tensorflow.distribute.experimental.-multi-worker-mirrored-strategy.pbtxt

@@ -46,7 +46,7 @@ tf_class {
   }
   member_method {
     name: "experimental_run_v2"
-    argspec: "args=[\'self\', \'fn\', \'args\', \'kwargs\'], varargs=None, keywords=None, defaults=[\'()\', \'None\'], "
+    argspec: "args=[\'self\', \'fn\', \'args\', \'kwargs\', \'options\'], varargs=None, keywords=None, defaults=[\'()\', \'None\', \'None\'], "
   }
   member_method {
     name: "group"

tensorflow/tools/api/golden/v1/tensorflow.distribute.experimental.-parameter-server-strategy.pbtxt

@@ -46,7 +46,7 @@ tf_class {
   }
   member_method {
     name: "experimental_run_v2"
-    argspec: "args=[\'self\', \'fn\', \'args\', \'kwargs\'], varargs=None, keywords=None, defaults=[\'()\', \'None\'], "
+    argspec: "args=[\'self\', \'fn\', \'args\', \'kwargs\', \'options\'], varargs=None, keywords=None, defaults=[\'()\', \'None\', \'None\'], "
   }
   member_method {
     name: "group"

tensorflow/tools/api/golden/v1/tensorflow.distribute.experimental.-t-p-u-strategy.pbtxt

@@ -50,7 +50,7 @@ tf_class {
   }
   member_method {
     name: "experimental_run_v2"
-    argspec: "args=[\'self\', \'fn\', \'args\', \'kwargs\'], varargs=None, keywords=None, defaults=[\'()\', \'None\'], "
+    argspec: "args=[\'self\', \'fn\', \'args\', \'kwargs\', \'options\'], varargs=None, keywords=None, defaults=[\'()\', \'None\', \'None\'], "
   }
   member_method {
     name: "group"

tensorflow/tools/api/golden/v1/tensorflow.distribute.pbtxt

@@ -40,6 +40,10 @@ tf_module {
     name: "ReplicaContext"
     mtype: "<type \'type\'>"
   }
+  member {
+    name: "RunOptions"
+    mtype: "<type \'type\'>"
+  }
   member {
     name: "Server"
     mtype: "<type \'type\'>"

tensorflow/tools/api/golden/v1/tensorflow.tpu.-padding-spec.pbtxt

@@ -0,0 +1,12 @@
+path: "tensorflow.tpu.PaddingSpec"
+tf_class {
+  is_instance: "<enum \'PaddingSpec\'>"
+  member {
+    name: "AUTO"
+    mtype: "<enum \'PaddingSpec\'>"
+  }
+  member {
+    name: "POWER_OF_TWO"
+    mtype: "<enum \'PaddingSpec\'>"
+  }
+}

tensorflow/tools/api/golden/v1/tensorflow.tpu.pbtxt

@@ -4,6 +4,10 @@ tf_module {
     name: "CrossShardOptimizer"
     mtype: "<type \'type\'>"
   }
+  member {
+    name: "PaddingSpec"
+    mtype: "<class \'enum.EnumMeta\'>"
+  }
   member {
     name: "experimental"
     mtype: "<type \'module\'>"
@@ -34,7 +38,7 @@ tf_module {
   }
   member_method {
     name: "replicate"
-    argspec: "args=[\'computation\', \'inputs\', \'infeed_queue\', \'device_assignment\', \'name\', \'maximum_shapes\'], varargs=None, keywords=None, defaults=[\'None\', \'None\', \'None\', \'None\', \'None\'], "
+    argspec: "args=[\'computation\', \'inputs\', \'infeed_queue\', \'device_assignment\', \'name\', \'maximum_shapes\', \'padding_spec\'], varargs=None, keywords=None, defaults=[\'None\', \'None\', \'None\', \'None\', \'None\', \'None\'], "
   }
   member_method {
     name: "rewrite"

tensorflow/tools/api/golden/v2/tensorflow.distribute.-mirrored-strategy.pbtxt

@@ -54,7 +54,7 @@ tf_class {
   }
   member_method {
     name: "experimental_run_v2"
-    argspec: "args=[\'self\', \'fn\', \'args\', \'kwargs\'], varargs=None, keywords=None, defaults=[\'()\', \'None\'], "
+    argspec: "args=[\'self\', \'fn\', \'args\', \'kwargs\', \'options\'], varargs=None, keywords=None, defaults=[\'()\', \'None\', \'None\'], "
   }
   member_method {
     name: "experimental_split_to_logical_devices"

tensorflow/tools/api/golden/v2/tensorflow.distribute.-one-device-strategy.pbtxt

@@ -54,7 +54,7 @@ tf_class {
   }
   member_method {
     name: "experimental_run_v2"
-    argspec: "args=[\'self\', \'fn\', \'args\', \'kwargs\'], varargs=None, keywords=None, defaults=[\'()\', \'None\'], "
+    argspec: "args=[\'self\', \'fn\', \'args\', \'kwargs\', \'options\'], varargs=None, keywords=None, defaults=[\'()\', \'None\', \'None\'], "
   }
   member_method {
     name: "experimental_split_to_logical_devices"

tensorflow/tools/api/golden/v2/tensorflow.distribute.-run-options.pbtxt

@@ -0,0 +1,23 @@
+path: "tensorflow.distribute.RunOptions"
+tf_class {
+  is_instance: "<class \'tensorflow.python.distribute.distribute_lib.RunOptions\'>"
+  is_instance: "<class \'tensorflow.python.distribute.distribute_lib.RunOptions\'>"
+  is_instance: "<type \'tuple\'>"
+  member {
+    name: "experimental_bucketizing_dynamic_shape"
+    mtype: "<type \'property\'>"
+  }
+  member {
+    name: "experimental_enable_dynamic_batch_size"
+    mtype: "<type \'property\'>"
+  }
+  member_method {
+    name: "__init__"
+  }
+  member_method {
+    name: "count"
+  }
+  member_method {
+    name: "index"
+  }
+}

tensorflow/tools/api/golden/v2/tensorflow.distribute.-strategy.pbtxt

@@ -53,7 +53,7 @@ tf_class {
   }
   member_method {
     name: "experimental_run_v2"
-    argspec: "args=[\'self\', \'fn\', \'args\', \'kwargs\'], varargs=None, keywords=None, defaults=[\'()\', \'None\'], "
+    argspec: "args=[\'self\', \'fn\', \'args\', \'kwargs\', \'options\'], varargs=None, keywords=None, defaults=[\'()\', \'None\', \'None\'], "
   }
   member_method {
     name: "experimental_split_to_logical_devices"

tensorflow/tools/api/golden/v2/tensorflow.distribute.experimental.-central-storage-strategy.pbtxt

@@ -54,7 +54,7 @@ tf_class {
   }
   member_method {
     name: "experimental_run_v2"
-    argspec: "args=[\'self\', \'fn\', \'args\', \'kwargs\'], varargs=None, keywords=None, defaults=[\'()\', \'None\'], "
+    argspec: "args=[\'self\', \'fn\', \'args\', \'kwargs\', \'options\'], varargs=None, keywords=None, defaults=[\'()\', \'None\', \'None\'], "
   }
   member_method {
     name: "experimental_split_to_logical_devices"

tensorflow/tools/api/golden/v2/tensorflow.distribute.experimental.-multi-worker-mirrored-strategy.pbtxt

@@ -54,7 +54,7 @@ tf_class {
   }
   member_method {
     name: "experimental_run_v2"
-    argspec: "args=[\'self\', \'fn\', \'args\', \'kwargs\'], varargs=None, keywords=None, defaults=[\'()\', \'None\'], "
+    argspec: "args=[\'self\', \'fn\', \'args\', \'kwargs\', \'options\'], varargs=None, keywords=None, defaults=[\'()\', \'None\', \'None\'], "
   }
   member_method {
     name: "experimental_split_to_logical_devices"

tensorflow/tools/api/golden/v2/tensorflow.distribute.experimental.-parameter-server-strategy.pbtxt

@@ -54,7 +54,7 @@ tf_class {
   }
   member_method {
     name: "experimental_run_v2"
-    argspec: "args=[\'self\', \'fn\', \'args\', \'kwargs\'], varargs=None, keywords=None, defaults=[\'()\', \'None\'], "
+    argspec: "args=[\'self\', \'fn\', \'args\', \'kwargs\', \'options\'], varargs=None, keywords=None, defaults=[\'()\', \'None\', \'None\'], "
   }
   member_method {
     name: "experimental_split_to_logical_devices"

tensorflow/tools/api/golden/v2/tensorflow.distribute.experimental.-t-p-u-strategy.pbtxt

@@ -54,7 +54,7 @@ tf_class {
   }
   member_method {
     name: "experimental_run_v2"
-    argspec: "args=[\'self\', \'fn\', \'args\', \'kwargs\'], varargs=None, keywords=None, defaults=[\'()\', \'None\'], "
+    argspec: "args=[\'self\', \'fn\', \'args\', \'kwargs\', \'options\'], varargs=None, keywords=None, defaults=[\'()\', \'None\', \'None\'], "
   }
   member_method {
     name: "experimental_split_to_logical_devices"

tensorflow/tools/api/golden/v2/tensorflow.distribute.pbtxt

@@ -40,6 +40,10 @@ tf_module {
     name: "ReplicaContext"
     mtype: "<type \'type\'>"
   }
+  member {
+    name: "RunOptions"
+    mtype: "<type \'type\'>"
+  }
   member {
     name: "Server"
     mtype: "<type \'type\'>"

tensorflow/tools/compatibility/renames_v2.py

@@ -1328,6 +1328,8 @@ renames = {
         'tf.compat.v1.to_int64',
     'tf.tpu.CrossShardOptimizer':
         'tf.compat.v1.tpu.CrossShardOptimizer',
+    'tf.tpu.PaddingSpec':
+        'tf.compat.v1.tpu.PaddingSpec',
     'tf.tpu.batch_parallel':
         'tf.compat.v1.tpu.batch_parallel',
     'tf.tpu.bfloat16_scope':