],
)

tf_cc_test(
    name = "framework_gradients_test",
    srcs = ["framework/gradients_test.cc"],
    deps = [
        ":cc_ops",
        ":client_session",
        ":grad_op_registry",
        ":grad_ops",
        ":gradients",
        ":testutil",
        "//tensorflow/core:all_kernels",
        "//tensorflow/core:framework",
        "//tensorflow/core:framework_internal",
        "//tensorflow/core:protos_all_cc",

                     llvm::SmallVectorImpl<Operation*>& erase_list,
                     llvm::SmallPtrSetImpl<Operation*>& dead_ops) {
  for (mlir::Value res : op->getResults()) {
    for (Operation* user : res.getUsers()) {
      if (!dead_ops.insert(user).second) continue;
      RecursiveRemove(user, erase_list, dead_ops);
    }
  }

  erase_list.push_back(op);

  for (auto& use : op->getOpOperands()) {

def _composite_fully_connected(input_, filter_, bias, act):
  res = tf.raw_ops.MatMul(
      a=input_, b=filter_, transpose_a=False, transpose_b=True)
  res = tf.raw_ops.Add(x=res, y=bias)
  if act == 'RELU':
    return tf.raw_ops.Relu(features=res)
  elif act == 'RELU6':
    return tf.raw_ops.Relu6(features=res)
  elif act == 'TANH':
    return tf.raw_ops.Tanh(x=res)
  else:
    return res

                const std::vector<Output>& grad_inputs,
                std::vector<Output>* grad_outputs) {
  auto shift = op.input(1);
  auto axis = op.input(2);
  auto grad_op = Roll(scope, grad_inputs[0], Neg(scope, shift), axis);
  grad_outputs->push_back(grad_op);
  grad_outputs->push_back(NoGradient());
  grad_outputs->push_back(NoGradient());
  return scope.status();
}
REGISTER_GRADIENT_OP("Roll", RollGrad);

}  // namespace

  shape = input_.shape.as_list()
  for i in range(len(shape)):
    rdims = tf.raw_ops.OneHot(
        indices=i, depth=len(shape), on_value=True, off_value=False, axis=-1)
    rarray = tf.raw_ops.Reverse(tensor=input_, dims=rdims)

    left_padding_size = tf.raw_ops.GatherNd(params=paddings, indices=[i, 0])
    right_padding_size = tf.raw_ops.GatherNd(params=paddings, indices=[i, 1])

    if mode == 'REFLECT':

    ],
    licenses = ["notice"],
)

package_group(
    name = "friends",
    packages = [
        "//tensorflow/compiler/mlir/tfr/...",
    ],
)

gen_op_libraries(
    name = "pad_ops",
    src = "ops_defs.py",
    deps = [
        "//tensorflow:tensorflow_py",
        "//tensorflow/python/ops:array_ops_gen",
        "//tensorflow/python/platform:flags",
        "@absl_py//absl:app",
    ],
)

def _composite_fully_connected(input_, filter_, bias, act):
  res = tf.raw_ops.MatMul(
      a=input_, b=filter_, transpose_a=False, transpose_b=True)
  res = tf.raw_ops.Add(x=res, y=bias)
  if act == 'RELU':
    return tf.raw_ops.Relu(features=res)
  elif act == 'RELU6':
    return tf.raw_ops.Relu6(features=res)
  elif act == 'TANH':
    return tf.raw_ops.Tanh(x=res)
  else:
    return res

```

# Verify for functions with control return values, the island with only a
# consumed control return value has its control output added to the GraphOps
# FetchOp.

# Match the island containing the "tf.Neg", capture the output
# CHECK:          %[[ISLAND_0:[a-z_0-9]*]], %[[ISLAND_0_control:[a-z_0-9]*]] = tf_executor.island wraps "tf.Neg"

    }
    kwargs_ = {
        'input_': input_,
        'paddings': paddings,
        'mode': mode,
    }
    # Make sure the composition python function is correct
    self._assertOpAndComposite([input_], tf.raw_ops.MirrorPad,
                               ops_defs._composite_mirror_pad, kwargs_, kwargs)
    # Make sure the translation and decomposition is correct
    self._assertOpAndComposite([input_],

# Verify for functions with control return values, the island with a consumed
# data output and a consumed control has both its outputs added to the GraphOps
# FetchOp.

# Match the island containing the "tf.Neg", capture the output
# CHECK:          %[[ISLAND:[a-z_0-9]*]], %[[ISLAND_control:[a-z_0-9]*]] = tf_executor.island wraps "tf.Neg"