Refactor how the parameters are iterated over to reduce the total time of the test.  (This covers less cases, but should be almost equivalent.)

Force shape inputs to int64 in CorrMM to avoid having something that does not match the 'int' assumed type of before.

Also makes sure that the CPU versions are tested even if there are no GPUs.

Move the extending_theano documentation out of tutorial

Temp disable warn64

Really skip the flake8 part in "everything but tensor".

Fix last DLT problem(s) with the new backend.

A few small stuff

fixing make_node param passing for conv_grad (conv_transpose)

This was causing problems in the case of Pool, but I also fixed it elsewhere just to be sure.

Faster travis