Clarify error message related to expected kern_shape specs (1D, int64) in GpuDnnConvDesc

Clarify error message related to expected kern_shape in GpuDnnConvDesc make_node() method, and fix a related test in test_dnn.

Fix invalid pointer casts and negative stride handling in sparse

Fix crash due to excessive decref of 1

Fix compilation crash show by THEANO_FLAGS='floatX=float32,dnn.conv.a…

Fix `-Wl,rpath` parameter for Windows (fix #5463).

Do not decref borrowed references.

Fix batchnorm problems

Fix compilation crash show by THEANO_FLAGS='floatX=float32,dnn.conv.algo_bwd_data=time_once,dnn.conv.algo_bwd_filter=time_once' theano-nose theano/sandbox/cuda/tests/test_dnn.py:test_dnn_conv_grad. fix gh-5517

Cast npy_intp (usually int64) to int (not npy_intp* to int*, which is a
strict aliasing violation)

Use first-in-memory-order where the BLAS interface requires it,
not first-in-index-order; add negative stride test case

Consistently use '\n' in outputted code

All generated C code uses `\n` line separators, except some in `gof/op.py` when on Windows. This fixes it.

Strangely, this also fixes the issue I described in [#5464](/Theano/Theano/issues/5463#issuecomment-279826487).

Convert dos end of line to unix version

multiple data type support for GPUMultinomialFromUniform output

Improve the example for strict=True in the scan() doc part 2.

Fix test for Python 3 on 32-bit

Interface addition and cleanup

replace multinomial_wo_replacement by choice

Change MRG choice default of a to be None. We can do this as in the old back-end, when a wasn't provided and pvals was, a=2, but it was only valid when that match pvals shapes. So it was duplicate information. As MRG have pvals mandatory, to work it need a matching value for a.

Merge optimizer 5319

Add C code for CPU and GPU for tensor.Split op.

Fix gpu integers in scan.

validate numpy1.12 compatibility changes