Add the test case for MaxPooling3DGrad in Eager mode

tensorflow/python/eager/BUILD

@@ -114,9 +114,11 @@ cuda_py_test(
         ":backprop",
         ":context",
         ":test",
+        "//third_party/py/numpy",
         "//tensorflow/python:embedding_ops",
         "//tensorflow/python:array_ops",
         "//tensorflow/python:control_flow_ops",
+        "//tensorflow/python:layers",
         "//tensorflow/python:math_ops",
         "//tensorflow/python:nn_ops",
         "//tensorflow/python:resource_variable_ops",

tensorflow/python/eager/backprop_test.py

@@ -30,6 +30,7 @@ from tensorflow.python.framework import dtypes
 from tensorflow.python.framework import ops
 from tensorflow.python.framework import tensor_shape
 from tensorflow.python.framework import test_util
+from tensorflow.python.layers.pooling import max_pooling3d
 from tensorflow.python.ops import array_ops
 from tensorflow.python.ops import control_flow_ops
 from tensorflow.python.ops import custom_gradient
@@ -1140,5 +1141,32 @@ class BackpropTest(test.TestCase):
       g = f(c)
     self.assertAllEqual(self.evaluate(t.gradient(g, c)), 4.0)
 
+  @test_util.run_in_graph_and_eager_modes
+  def testMaxPooling3DGradient(self):
+    def forward(a):
+      r = max_pooling3d(
+          a,
+          pool_size=pool_size,
+          strides=strides,
+          padding="SAME")
+      return r
+
+    input_sizes = [1, 3, 2, 4, 1]
+    pool_size = (2, 2, 1)
+    strides = (1, 1, 1)
+
+    total_size = np.prod(input_sizes)
+    x = np.arange(1, total_size + 1, dtype=np.float32)
+    aa = constant_op.constant(x, shape=input_sizes, dtype=dtypes.float32)
+    da = backprop.gradients_function(forward)(aa)
+
+    if not context.executing_eagerly():
+      tf_aa = constant_op.constant(x, shape=input_sizes, dtype=dtypes.float32)
+      tf_max = max_pooling3d(
+          tf_aa, pool_size=pool_size, strides=strides, padding="SAME")
+      tf_da = gradients.gradients(tf_max, [tf_aa])
+      self.assertAllEqual(da[0], tf_da[0].eval())
+
+
 if __name__ == '__main__':
   test.main()

tensorflow/python/eager/pywrap_tfe_src.cc

@@ -2020,6 +2020,7 @@ PyObject* RecordGradient(PyObject* op_name, PyObject* inputs, PyObject* attrs,
   PyObject* op_outputs;
   bool op_outputs_tuple_created = false;
   std::pair<bool, tensorflow::gtl::FlatSet<int>>* outputs_not_required;
+
   if (OpGradientDoesntRequireOutputIndices(c_op_name, &outputs_not_required)) {
     if (outputs_not_required->first) {
       op_outputs = Py_None;