Add a method to properly convert the graph of Composite to float32.

Gpuarray pool grad grad

Fix some failing tests in debugmode

[REG,BUG] Use _props_dict() from the right Op.

CudaNdarray.filter rejects broadcastable dimensions with non-zero
strides

Add Max3dPoolGradGrad which accept 5d input and pools over last 3
dimensions. Also add padding parameter and now accept tensor variables
to make_node which removes necessity to check for constant args. Local
optimization are updated accordingly which now work for 2d and 3d
pooling.

It was using the one from GpuFromHost instead of AdvancedIncSubtensor1.

Fix some problems in float16.

[ENH] fix opt error/skip and better error printing.

Correct blocks/threads for gpuarray CorrMM and Corr3DMM

we do it the normal way now.

Doc: Fix library path. Closes #5156.

Number of blocks and number of threads were swapped.

use floatX in gpuarray dnn tests

Pool 2d rename

Add profiling of which node make_thunk take times.

Issue 5008 fixed