using GPU in theano operation

[Dominik Otto]

I would like to use multiple GPUs in my custom Theano operation as in http://deeplearning.net/software/theano_versions/dev/tutorial/using_gpu.html#id1

However, `theano.misc.pycuda_init` seems to be missing and initializing Pycuda after Theano produces memory issues unless I restrict them to there respective half.

What is the proper way to use the GPU in the `preform` method of my Theano operations?