Half precision convolution not working

to reproduce
```python
import torch
import torch.nn as nn
from torch.autograd import Variable

x = Variable(torch.rand(1,3,100,100).cuda().half())
m = nn.Conv2d(3,16,5).cuda().half()

y = m(x)
```

Will result in:
`RuntimeError: CUDNN_STATUS_ARCH_MISMATCH`