TileGrad implements only one very narrow special case, and it implements
it wrongly. Raise NotImplementedError instead of using this botched
implementation.

Raise a proper error when reps is not a constant (since a tensor value
is not supported by this implementation, as it is) and also force
x.ndim == len(reps), since bad things happen if it's not the case.

Hide warnings for a past bug in the buildbot

Fix subtensor crash

Fix runtime crash with the local_subtensor_of_alloc optimization when the input have a broadcastable dimensions.

Edit NEWS.txt

Doc op

Also merge duplicate documentation of grad_sources_inputs.

PEP8 and docstring fixes.

Fix test in TrueDivTester

Since dtype mismatch is one reason the error is raised, it makes
interpreting it easier.

Fix Ger C code when output is a row or column.

Some BLAS versions did not accept (1,1) strides in that case.

Update tests, gpu_gemv is used now, not gemm/dot22

Fix test of elemwise_fusion, now intdiv is fused.

more helpful error to catch symbolic shape vecs in infer_shape

Add warning for bug in Numpy <= 1.6.1.

When using numpy's advanced indexing on huge arrays (2**32 elements or
more), Numpy <= 1.6.1 has a bug where the resulting array is not
properly filled. We emit a warning in those cases.