by hardcoding the rounding instead of relying
on NumPy functions, so that this op should work
on CPU and on GPU.

Tests (CPU and GPU) passed on my computer, let's
see if jenkins/travis are ok with that.

(as I have changed the headers list).

Initialize the blas kernels when creating the contexts.

Add user warning to prevent issue #5143

mixed

removed pruner function of EquilibriumOptimizer class

While looking for subsamples I have detected duplicated values for test_shapes in `test_dnn.py` both in old and new backend. I remove the duplicate.

* Remove the "[new backend]" part of the message and idem in the old back-end.
* Move the check to the make_node of the ops.
I'll continue analyzing the subsamples.

Show user warning in old and new backend when Theano flags `dnn.conv.algo_bwd_filter` is not `none`, to help user prevent the issue #5143.

Fix theano scripts on windows

make tensor.join() return the input when there is only 1 variable to join.

Use Scan.L_op instead of Scan.grad() to help speed up the second deri…

Fixes

numpy 1.12.0rc2 compatibility fix related to variable typecasting

Switch gs and ls to follow libgpuarray.