Merge pull request #31511 from tensorflow/ggadde-cp3

Cherrypicks to fix the disconnected graph issue, and missing CUDA compute capabilities.

tensorflow/python/keras/engine/base_layer_utils.py

@@ -225,7 +225,8 @@ def _create_keras_history_helper(tensors, processed_ops, created_layers):
             # configured improperly.
             constants[i] = op_input
           else:
-            constants[i] = backend.function([], op_input)([])
+            with ops.init_scope():
+              constants[i] = backend.function([], op_input)([])
       processed_ops, created_layers = _create_keras_history_helper(
           layer_inputs, processed_ops, created_layers)
       name = op.name
@@ -239,7 +240,7 @@ def _create_keras_history_helper(tensors, processed_ops, created_layers):
   return processed_ops, created_layers
 
 
-def needs_keras_history(tensors):
+def needs_keras_history(tensors, ignore_call_context=False):
   """Check if any Tensors need to be wrapped in TensorFlowOpLayers.
 
   This will never return True inside a sublayer, because sublayers
@@ -249,12 +250,18 @@ def needs_keras_history(tensors):
 
   Arguments:
     tensors: An arbitrary nested structure of Tensors.
+    ignore_call_context: Whether to ignore the check of if currently
+      outside of a `call` context. This is `True` when creating
+      KerasHistory inside `Node`, where we always know that Tensors
+      are being used with the Functional API.
 
   Returns:
     Bool, whether at least one Tensor needs to be wrapped.
   """
   input_tensors = nest.flatten(tensors)
-  if call_context().in_call or all(
+  if call_context().in_call and not ignore_call_context:
+    return False
+  if all(
       getattr(tensor, '_keras_history', None) is not None
       for tensor in input_tensors):
     # KerasHistory already set.

tensorflow/python/keras/engine/network.py

@@ -1547,7 +1547,7 @@ class Network(base_layer.Layer):
     def _get_min_depth(node):
       """Gets the minimum depth at which node can be computed."""
       min_depth = 0
-      for layer, node_id, _, _ in node.iterate_inbound():
+      for layer, node_id, _, _ in node.iterate_inbound(include_arguments=True):
         inbound_node = layer._inbound_nodes[node_id]
         if inbound_node in node_to_depth:
           min_depth = min(min_depth, node_to_depth[inbound_node])
@@ -1720,7 +1720,8 @@ def _map_graph_network(inputs, outputs):
     nodes_in_progress.add(node)
 
     # Propagate to all previous tensors connected to this node.
-    for layer, node_index, tensor_index, tensor in node.iterate_inbound():
+    for layer, node_index, tensor_index, tensor in node.iterate_inbound(
+        include_arguments=True):
       build_map(tensor, finished_nodes, nodes_in_progress, layer, node_index,
                 tensor_index)
 

tensorflow/python/keras/engine/node.py

@@ -20,6 +20,7 @@ from __future__ import print_function
 
 from tensorflow.python.framework import ops
 from tensorflow.python.keras import backend
+from tensorflow.python.keras.engine import base_layer_utils
 from tensorflow.python.util import nest
 
 
@@ -111,6 +112,15 @@ class Node(object):
     # Optional keyword arguments to layer's `call`.
     self.arguments = arguments
 
+    # Create Keras History for any Keras Tensors in `arguments`.
+    tensor_arguments = [
+        t for t in nest.flatten(self.arguments) if isinstance(t, ops.Tensor)
+    ]
+    for tensor_argument in tensor_arguments:
+      if base_layer_utils.needs_keras_history(
+          tensor_argument, ignore_call_context=True):
+        base_layer_utils.create_keras_history(tensor_argument)
+
     # Add nodes to all layers involved.
     for layer in nest.flatten(inbound_layers):
       if layer is not None:
@@ -121,15 +131,39 @@ class Node(object):
     # accessor here.
     outbound_layer.inbound_nodes.append(self)
 
-  def iterate_inbound(self):
+  def iterate_inbound(self, include_arguments=False):
     """Returns a list of tuples representing the inbound data.
 
+    Arguments:
+      include_arguments: Whether to also iterate over any Keras Tensors
+        passed as args, kwargs.
+
     Returns:
       List of tuples like: (inbound_layer, node_index, tensor_index, tensor).
     """
-    return zip(
-        nest.flatten(self.inbound_layers), nest.flatten(self.node_indices),
-        nest.flatten(self.tensor_indices), nest.flatten(self.input_tensors))
+    inputs_inbound = list(
+        zip(
+            nest.flatten(self.inbound_layers),
+            nest.flatten(self.node_indices),
+            nest.flatten(self.tensor_indices),
+            nest.flatten(self.input_tensors)))
+
+    if include_arguments:
+      keras_tensor_arguments = [
+          kt for kt in nest.flatten(self.arguments)
+          if hasattr(kt, '_keras_history')
+      ]
+
+      def _get_inbound(keras_tensor):
+        kh = keras_tensor._keras_history
+        return kh.layer, kh.node_index, kh.tensor_index, keras_tensor
+
+      arguments_inbound = nest.map_structure(_get_inbound,
+                                             keras_tensor_arguments)
+
+      return inputs_inbound + arguments_inbound
+    else:
+      return inputs_inbound
 
   def _get_all_node_dependencies(self):
     """Returns all of the nodes this node immediately depends on."""

tensorflow/python/keras/layers/tensorflow_op_layer_test.py

@@ -135,6 +135,19 @@ def _float64_op():
   return keras.Model(inputs, outputs)
 
 
+class MyAdd(keras.layers.Layer):
+
+  def call(self, x, y):
+    return x + y
+
+
+def _layer_with_tensor_arg():
+  inputs = keras.Input(shape=(10,))
+  x = inputs * 2
+  outputs = MyAdd()(inputs, x)
+  return keras.Model(inputs, outputs)
+
+
 class LayerWithLayer(keras.layers.Layer):
 
   def build(self, input_shape):
@@ -191,6 +204,7 @@ class AutoLambdaTest(keras_parameterized.TestCase):
       ('_float64_op', _float64_op),
       ('_inner_layer', _inner_layer),
       ('_reuse_ancillary_layer', _reuse_ancillary_layer),
+      ('_layer_with_tensor_arg', _layer_with_tensor_arg),
   )
   def test_autolambda(self, model_fn):
     model = model_fn()
@@ -208,7 +222,11 @@ class AutoLambdaTest(keras_parameterized.TestCase):
     model(np_inputs)  # Test calling the model directly on inputs.
 
     new_model = keras.Model.from_config(
-        model.get_config(), custom_objects={'LayerWithLayer': LayerWithLayer})
+        model.get_config(),
+        custom_objects={
+            'LayerWithLayer': LayerWithLayer,
+            'MyAdd': MyAdd
+        })
     new_model.compile(
         adam.Adam(0.001),
         'mse',

third_party/toolchains/preconfig/ubuntu16.04/gcc7_manylinux2010-nvcc-cuda10.0/clang/bin/crosstool_wrapper_driver_is_not_gcc

@@ -53,6 +53,11 @@ NVCC_PATH = '/usr/local/cuda-10.0/bin/nvcc'
 PREFIX_DIR = os.path.dirname(GCC_HOST_COMPILER_PATH)
 NVCC_VERSION = '10.0'
 
+# Environment variable for supported TF CUDA Compute Capabilities
+# eg. export TF_CUDA_COMPUTE_CAPABILITIES=3.5,3.7,5.2,6.0,6.1,7.0
+CUDA_COMPUTE_ENV_VAR = 'TF_CUDA_COMPUTE_CAPABILITIES'
+DEFAULT_CUDA_COMPUTE_CAPABILITIES = '3.5,6.0'
+
 def Log(s):
   print('gpus/crosstool: {0}'.format(s))
 
@@ -202,7 +207,7 @@ def InvokeNvcc(argv, log=False):
   srcs = ' '.join(src_files)
   out = ' -o ' + out_file[0]
 
-  supported_cuda_compute_capabilities = [ "3.0", "6.0" ]
+  supported_cuda_compute_capabilities = os.environ.get(CUDA_COMPUTE_ENV_VAR, DEFAULT_CUDA_COMPUTE_CAPABILITIES).split(',')
   nvccopts = '-D_FORCE_INLINES '
   for capability in supported_cuda_compute_capabilities:
     capability = capability.replace('.', '')

third_party/toolchains/preconfig/ubuntu16.04/gcc7_manylinux2010-nvcc-cuda10.0/windows/msvc_wrapper_for_nvcc.py

@@ -36,7 +36,14 @@ GCC_HOST_COMPILER_PATH = ('/dt7/usr/bin/gcc')
 NVCC_PATH = '/usr/local/cuda-10.0/bin/nvcc'
 NVCC_VERSION = '10.0'
 NVCC_TEMP_DIR = "C:\\Windows\\Temp\\nvcc_inter_files_tmp_dir"
-supported_cuda_compute_capabilities = [ "3.0", "6.0" ]
+DEFAULT_CUDA_COMPUTE_CAPABILITIES = '3.5,6.0'
+
+# Taken from environment variable for supported TF CUDA Compute Capabilities
+# eg. export TF_CUDA_COMPUTE_CAPABILITIES=3.5,3.7,5.2,6.0,6.1,7.0
+supported_cuda_compute_capabilities = os.environ.get(
+    'TF_CUDA_COMPUTE_CAPABILITIES',
+    DEFAULT_CUDA_COMPUTE_CAPABILITIES).split(',')
+
 
 def Log(s):
   print('gpus/crosstool: {0}'.format(s))