def test_run_in_eager_and_graph_modes_skip_eager_runs_graph(self):
modes = []
def _test(self):
if context.executing_eagerly():
self.skipTest("Skipping in eager mode")
modes.append("eager" if context.executing_eagerly() else "graph")
test_util.run_in_graph_and_eager_modes(_test)(self)
self.assertEqual(modes, ["graph"])
def test_run_in_eager_and_graph_modes_skip_eager_runs_graph(self):
modes = []
def _test(self):
if context.executing_eagerly():
self.skipTest("Skipping in eager mode")
modes.append("eager" if context.executing_eagerly() else "graph")
test_util.run_in_graph_and_eager_modes(_test)(self)
self.assertEqual(modes, ["graph"])
def test_run_in_graph_and_eager_modes(self):
l = []
def inc(self, with_brackets):
del self # self argument is required by run_in_graph_and_eager_modes.
mode = "eager" if context.executing_eagerly() else "graph"
with_brackets = "with_brackets" if with_brackets else "without_brackets"
l.append((with_brackets, mode))
f = test_util.run_in_graph_and_eager_modes(inc)
f(self, with_brackets=False)
f = test_util.run_in_graph_and_eager_modes()(inc)
f(self, with_brackets=True)
self.assertEqual(len(l), 4)
self.assertEqual(set(l), {
("with_brackets", "graph"),
("with_brackets", "eager"),
("without_brackets", "graph"),
("without_brackets", "eager"),
})
def test_run_in_graph_and_eager_modes(self):
l = []
def inc(self, with_brackets):
del self # self argument is required by run_in_graph_and_eager_modes.
mode = "eager" if context.executing_eagerly() else "graph"
with_brackets = "with_brackets" if with_brackets else "without_brackets"
l.append((with_brackets, mode))
f = test_util.run_in_graph_and_eager_modes(inc)
f(self, with_brackets=False)
f = test_util.run_in_graph_and_eager_modes()(inc)
f(self, with_brackets=True)
self.assertEqual(len(l), 4)
self.assertEqual(set(l), {
("with_brackets", "graph"),
("with_brackets", "eager"),
("without_brackets", "graph"),
("without_brackets", "eager"),
})
return
print("Testing InceptionFwd %s", (input_size, filter_size, stride,
padding))
tf_logging.info("Testing InceptionFwd %s", (input_size, filter_size, stride,
padding))
self._AclCompareFwdValues(input_size, filter_size, [stride, stride], padding)
return Test
if __name__ == "__main__":
for index, (input_size_, filter_size_, output_size_, stride_,
padding_) in enumerate(GetShrunkInceptionShapes()):
print("testInceptionFwd_"+ str(index), input_size_, filter_size_, output_size_, stride_)
setattr(Conv2DTest, "testInceptionFwd_" + str(index),
test_util.run_in_graph_and_eager_modes()(
GetInceptionFwdTest(input_size_, filter_size_, stride_,
padding_)))
# TODO(b/35359731)
# Fwd, BckInput, and BackFilter to test that for certain input parameter
# set, winograd nonfused algorithm will be excluded from conv autotune. If
# in such case, winograd nonfused algorithm is added as one option of the
# conv autotune, and cuDNN version is smaller than 7, the following tests
# will fail.
ishape = [1, 400, 400, 1]
fshape = [1, 1, 1, 256]
oshape = [1, 400, 400, 256]
setattr(Conv2DTest, "testInceptionFwd_No_Winograd_Nonfused",
test_util.run_in_graph_and_eager_modes()(
GetInceptionFwdTest(ishape, fshape, 1, "SAME", gpu_only=True)))
test.main()
