版本、环境信息：
1）PaddlePaddle版本：1.5
2）CPU：
3）GPU：None
4）系统环境：centos 6.3, python 2.7.13

训练信息
1）单机，3cpu

复现信息：
逐步执行时，发现是这一步出错：optimizer.minimize(train_loss)。

问题描述：请详细描述您的问题，同步贴出报错信息、日志、可复现的代码片段
希望根据输入（cnt_in）的值判别，当该输入<0时，将embedding出的向量各分量乘以-1
模型建立和编译都能通过，但是开始训练时会报错：按位乘法维数不匹配。
（如果乘子fct不是通过IfElse给出，而是直接通过fill_constant定义的话，就不会出错）

模型网格：
self.lr_dat["usr_in"] = fluid.layers.data(name='usr_in', shape=[1], dtype='int64')
self.lr_dat["cnt_in"] = fluid.layers.data(name='cnt_in', shape=[1], dtype='int64')
self.lr_dat["label_in"] = fluid.layers.data(name='label_in', shape=[1], dtype='float32')

self.lr["usr_emb"] = fluid.layers.embedding(input=self.lr_dat["usr_in"], size=[7, 32], is_sparse=False)
    self.lr["cnt_emb"] = fluid.layers.embedding(input=self.lr_dat["cnt_in"], size=[14, 32], is_sparse=False)

    # cnt_fct = fluid.layers.fill_constant(shape=[32], value=-1.0, dtype='float32')

    cond = self.lr_dat["cnt_in"] > fluid.layers.zeros_like(self.lr_dat["cnt_in"])
    ifel = fluid.layers.IfElse(cond)
    with ifel.true_block():
        fct = fluid.layers.fill_constant(shape=[32], value=1.0, dtype='float32')
        ifel.output(fct)
    with ifel.false_block():
        fct = fluid.layers.fill_constant(shape=[32], value=-1.0, dtype='float32')
        ifel.output(fct)

    cnt_fct = ifel()
    self.lr["cnt_emb"] = fluid.layers.elementwise_mul(self.lr["cnt_emb"], cnt_fct)

    self.lr["inner_prod"] = fluid.layers.cos_sim(self.lr["usr_emb"], self.lr["cnt_emb"])
    self.lr["pred"] = fluid.layers.sigmoid(self.lr["inner_prod"])
    self.lr["raw_loss"] = fluid.layers.square_error_cost(input=self.lr["pred"], label=self.lr_dat["label_in"])
    self.lr["train_loss"] = fluid.layers.mean(self.lr["raw_loss"])

错误信息：
WARNING:root:2019-09-11 11:04:40.428257: start training
I0911 11:04:40.473984 65127 parallel_executor.cc:329] The number of CPUPlace, which is used in ParallelExecutor, is 3. And the Program will be copied 3 copies
I0911 11:04:40.478063 65127 build_strategy.cc:340] SeqOnlyAllReduceOps:0, num_trainers:1
Traceback (most recent call last):
File "test_ifelse.py", line 115, in
train_resu = exe.run(train_prog_comp, feed=data, fetch_list=[network.lr["train_loss"]])
File "/home/paddle/tools/paddle_release_home/python/lib/python2.7/site-packages/paddle/fluid/executor.py", line 666, in run
return_numpy=return_numpy)
File "/home/paddle/tools/paddle_release_home/python/lib/python2.7/site-packages/paddle/fluid/executor.py", line 528, in _run_parallel
exe.run(fetch_var_names, fetch_var_name)
paddle.fluid.core_avx.EnforceNotMet: Invoke operator elementwise_mul error.
Python Callstacks:
File "/home/paddle/tools/paddle_release_home/python/lib/python2.7/site-packages/paddle/fluid/framework.py", line 1771, in append_op
attrs=kwargs.get("attrs", None))
File "/home/paddle/tools/paddle_release_home/python/lib/python2.7/site-packages/paddle/fluid/layer_helper.py", line 43, in append_op
return self.main_program.current_block().append_op(*args,**kwargs)
File "/home/paddle/tools/paddle_release_home/python/lib/python2.7/site-packages/paddle/fluid/layers/nn.py", line 9904, in _elementwise_op
'use_mkldnn': use_mkldnn})
File "/home/paddle/tools/paddle_release_home/python/lib/python2.7/site-packages/paddle/fluid/layers/nn.py", line 9965, in elementwise_mul
return _elementwise_op(LayerHelper('elementwise_mul',**locals()))
File "test_ifelse.py", line 64, in construct_network
self.lr["cnt_emb"] = fluid.layers.elementwise_mul(self.lr["cnt_emb"], cnt_fct)
File "test_ifelse.py", line 80, in
network.construct_network()
C++ Callstacks:
Enforce failed. Expected x_dims[i + axis] == y_dims[i], but received x_dims[i + axis]:32 != y_dims[i]:64.
Broadcast dimension mismatch. at [/paddle/paddle/fluid/operators/elementwise/elementwise_op_function.h:63]
PaddlePaddle Call Stacks:
0 0x7f0786f9f808p void paddle::platform::EnforceNotMet::Initstd::string(std::string, char const*, int) + 360
1 0x7f0786f9fb57p paddle::platform::EnforceNotMet::EnforceNotMet(std::string const&, char const*, int) + 87
2 0x7f07876b6b5cp paddle::operators::get_mid_dims(paddle::framework::DDim const&, paddle::framework::DDim const&, int, int*, int*, int*) + 364
3 0x7f07876baba0p void paddle::operators::ElementwiseComputeEx<paddle::operators::MulFunctor, paddle::platform::CPUDeviceContext, float, float>(paddle::framework::ExecutionContext const&, paddle::framework::Tensor const*, paddle::framework::Tensor const*, int, paddle::operators::MulFunctor, paddle::framework::Tensor*) + 816
4 0x7f0787c26aa3p void paddle::operators::default_elementwise_mul<paddle::platform::CPUDeviceContext, float>(paddle::framework::ExecutionContext const&, paddle::framework::Tensor const*, paddle::framework::Tensor const*, paddle::framework::Tensor*) + 115