Issue description

It seems torch.relu() when given nan as input, it produces 0 instead of nan. I am wondering if this is the intended behavior. This behavior actually hides code bugs, making troubleshooting harder.

Code example

a = torch.Tensor(1).fill_(float('nan'))
torch.relu(a)  # => 0

System Info

Collecting environment information...
PyTorch version: 0.4.0
Is debug build: No
CUDA used to build PyTorch: 8.0.61

OS: Ubuntu 14.04.5 LTS
GCC version: (Ubuntu 4.8.5-4ubuntu8~14.04.2) 4.8.5
CMake version: version 3.2.2

Python version: 3.5
Is CUDA available: Yes
CUDA runtime version: 8.0.61
GPU models and configuration: GPU 0: GeForce GTX 1080
Nvidia driver version: 384.111
cuDNN version: Probably one of the following:
/usr/lib/x86_64-linux-gnu/libcudnn.so.7.0.5
/usr/local/cuda-8.0/lib64/libcudnn.so
/usr/local/cuda-8.0/lib64/libcudnn.so.5
/usr/local/cuda-8.0/lib64/libcudnn.so.5.1.10
/usr/local/cuda-8.0/lib64/libcudnn.so.7
/usr/local/cuda-8.0/lib64/libcudnn.so.7.0.5
/usr/local/cuda-8.0/lib64/libcudnn_static.a

Versions of relevant libraries:
[pip] numpy (1.13.3)
[pip] numpydoc (0.7.0)
[pip] torch (0.4.0)
[pip] torchvision (0.2.0)
[conda] pytorch 0.4.0 py35_cuda8.0.61_cudnn7.1.2_1 pytorch
[conda] torchvision 0.2.0 py35_0