composite to float32.

Comes with a test.

fix tests in mode=FAST_COMPILE

GpuCorrMM and GpuCorr3dMM in new backend

Fix recursion limit that happen in the new gpu back-end

 Removing _op_use_c_code attribute

For empty list/tuple, use floatX for the dtype instead of the numpy d…

clean following the fix

Replace Conv3D references in tests with Corr3dMM

Fix maxargmax zero shape

Regression FIX

This one would have failed in DebugMode and FAST_COMPILE before the fix.

GpuDnnBatchNorm with 5d inputs

buildbot explicit cuda path

theano.tensor.signal.Pool with 3D support

Make sure fgraph attribute are always removed from apply node and variables when they aren't in an fgraph anymore.

Adding an AbstractConv3d interface