All buffers have to be of the same dtype

[untom]

Currently, all buffers (parameters, internals, gradients, ...) are assumed to have the dtype (typically either float or double). This is a bit restrictive: For example, in a max-pooling operation, one would like to store which cell in the current window has the maximum value (as discussed in #29). Something similar would happen in a Maxout-Layer, or when implementing a Top-K Autoencoder. I can work around this for the max-pooling OP, but in general it would be nice to be able to specify an optional dtype for each ShapeTemplate.

[flukeskywalker]

Indeed, this is essentially since the layer has a handler and each handler currently only supports one dtype. This is more for simplicity than anything else -- we don't need separate kernels for inputs being int or floats, for example. We can transition to a default_dtype for each handler along with other supported dtypes in the future.

[flukeskywalker]

Let's set this aside as on the first TODO item after 0.5? This will require some changes in buffer allocation, and probably several additions to PyCudaHandler, but no major reorg.

[Qwlouse]

Yes! Let's definitely not tackle that before the release.