add ParallelDo example for benchmark/fluid

[luotao1]

• add ParallelDo example for multi-thread CPU training in benchmark/fluid. For four models in benchmark/fluid, only add ParallelDo example for mnist and resnet now.
• add --cpus arguments to specify the number of cpus. The usage likes:

python fluid_benchmark.py --model resnet --batch_size=64 --pass_num 5 --device CPU --data_set cifar10 --no_test --cpus=20

• remove unused batch_size_tensor = fluid.layers.create_tensor(dtype='int64'), since fluid.layers.accuracy(input=predict, label=label) will auto create it.
  

  Paddle/python/paddle/fluid/layers/metric.py

  Lines 28 to 39 in ee9832a

  	def accuracy(input, label, k=1, correct=None, total=None):
  	"""
  	This function computes the accuracy using the input and label.
  	The output is the top k inputs and their indices.
  	"""
  	helper = LayerHelper("accuracy", **locals())
  	topk_out, topk_indices = nn.topk(input, k=k)
  	acc_out = helper.create_tmp_variable(dtype="float32")
  	if correct is None:
  	correct = helper.create_tmp_variable(dtype="int64")
  	if total is None:
  	total = helper.create_tmp_variable(dtype="int64")

[tonyyang-svail]

Looks good to me.