GpuArray BatchNorm

Add gpu_contiguous for batch normalization.

Make the old backend support as_input_f32 theano.config.dnn.precision.

Float16 convolutions

Multinomial Without Replacement

Small mixed changes.

first trial - midwork

start local memory

debugging

debugging

initial test

fixed bugs

modified test

moved test file

debugging

first trial - finished

* flake 8 changes

* flake 8 changes

* add missing imports

* code cleanup

* code clean up

* flake 8 changes

* code cleanup

* code cleanup

* code cleanup

* code cleanup

* add missing import

In GpuElemwise, don't call Elemwise.make_node, small speed up and help investigate. Also build directly a GpuDimShuffle instead of a DimShuffle that will be lifter later.

Use nvcc flags to compile with cuDNN

Fix for issue #4739

Fix for issue #4739

OpenCL support for Image2Neibs kernels

OpenCL support for Image2Neibs requires a bugfix for libgpuarray
in commit f8b8e156, otherwise python will crash in certain cases.

Use the six builtins

Speeding up local_gpua_careduce

Fix some problems in the buildbot

gpuarray tests should pass when floatX=float32 and add this to the daily buildbot