Created by: Aurelius84

support Tensor.shape in control_flow_if

Example code:

def foo(x):
    batch_size = fluid.layers.shape(x)
    if batch_size[0] > 16:
        x = x + 1
    return x

def foo(x):
    if fluid.layers.shape(x)[0] > 16:
        x = x + 1
    return x

Currently, we can identify more cases that need to transform into layers.cond. But now the case with and / or is not supported and will be added in next PR.