RuntimeError: cudnnConvolutionBackwardFilter

I was running code (https://github.com/neubig/nn4nlp-code/blob/master/05-cnn/cnn-class.py) with gpu.

Then I got follow error:

``` python
The dy.parameter(...) call is now DEPRECATED.
There is no longer need to explicitly add parameters to the computation graph.
Any used parameter will be added automatically.
---------------------------------------------------------------------------
RuntimeError Traceback (most recent call last)
in ()
80 my_loss = dy.pickneglogsoftmax(scores, tag)
81 train_loss += my_loss.value()
---> 82 my_loss.backward()
83 trainer.update()
84 logger.debug("iter %r: train loss/sent=%.4f, acc=%.4f, time=%.2fs" % (ITER, train_loss/len(train), train_correct/len(train), time.time()-start))

_dynet.pyx in _dynet.Expression.backward()

_dynet.pyx in _dynet.Expression.backward()

RuntimeError: cudnnConvolutionBackwardFilter(dev.cudnnHandle, &alpha, x_desc_, x->v, y_desc_, dy->v, conv_desc_, bwd_f_algo_, bwd_filter_workspace, workspace_bwd_filter_size_, &beta, filter_desc_, dxi)
```

Start by running 05-cnn/cnn-class.py from the linked neubig/nn4nlp-code repository with GPU support and inspect the failure at my_loss.backward(). Compare the runtime environment and convolution backward path around the reported cudnnConvolutionBackwardFilter call. Done means reproducing the error and establishing a verified resolution or a clear diagnosis.

Written by the indexing model from the issue text.