Support reshape for dynamic shape in tf converter (#4185)

* Support reshape for dynamic shape in tf converter

* Only allow reshape directly after shape function for symbolic input shape

* Fix lint

python/tvm/relay/frontend/tensorflow.py

@@ -612,6 +612,16 @@ def _slice():
 def _reshape():
     def _impl(inputs, attr, params):
         pop_node = inputs.pop(1)
+
+        # We use reshape_like directly to deal with dynamic shape.
+        if isinstance(pop_node, tvm.relay.expr.Call):
+            if "shape_of" not in str(pop_node.op):
+                raise RuntimeError("If shape operator is used in reshape to "
+                                   "express reshape_like, shape_of must be "
+                                   "the direct ancestor of reshape when input "
+                                   "shape is symbolic.")
+            return _op.reshape_like(inputs[0], pop_node.args[0])
+
         try:
             shape_arg = _get_tuple_param(params, pop_node)
         except AttributeError:
@@ -788,7 +798,18 @@ def _relu6():
 
 def _shape():
     def _impl(inputs, attr, params):
-        return np.array(attr['_input_shapes'][inputs[0]], dtype='int32')
+        is_symbolic_shape = False
+        for axis in attr['_input_shapes'][inputs[0]]:
+            if not isinstance(axis, (int, tvm.expr.IntImm, tvm.expr.UIntImm)):
+                is_symbolic_shape = True
+                break
+
+        if is_symbolic_shape:
+            ret = _op.shape_of(inputs[0], dtype='int32')
+        else:
+            ret = np.array(attr['_input_shapes'][inputs[0]], dtype='int32')
+        return ret
+
     return _impl
 
 def _fill():

tests/python/frontend/tensorflow/test_forward.py

@@ -543,12 +543,23 @@ def _test_reshape(data, out_shape):
 
         compare_tf_with_tvm(data, 'Placeholder:0', 'Reshape:0')
 
+def _test_reshape_like(data, shape_like):
+    """ A special case for reshape. """
+
+    with tf.Graph().as_default():
+        in_data = array_ops.placeholder(shape=data.shape, dtype=data.dtype)
+        in_shape_like = array_ops.placeholder(shape=shape_like.shape, dtype=data.dtype)
+        out_shape = array_ops.shape(in_shape_like)
+        array_ops.reshape(in_data, out_shape)
+
+        compare_tf_with_tvm(data, 'Placeholder:0', 'Reshape:0')
 
 def test_forward_reshape():
     _test_reshape(np.arange(6.0), [2, 3])
     _test_reshape(np.arange(6), [-1, 2])
     _test_reshape(np.arange(6), [3, -1])
     _test_reshape(np.arange(6), [-1])
+    _test_reshape_like(np.zeros((3, 6)), np.zeros((9, 2)))
 
 #######################################################################
 # DepthToSpace