I'm positive that it has never been used, because there's an error in
its definition, that would raise an error as soon as we try to call it.

Now:
    - we return the cached mode only if the default mode was asked
    - we chack that the cached mode is actually of the right class
      (because config.mode can change during the execution).

This allows to cache the default mode.

 - Compile a function, and call it on random values (so DebugMode can verify
   things)
 - Work on floatX, so gradients can be verified
 - Verify gradients
 - Check that some incorrect inputs lead to errors

start/stop/tep

into the graph

allows you to color the nodes with colors. You can optionally provide a
dictionary of colors to the function.

Fix two more problems in multi_dtype_cast_tests. Obviously not thinking straight today. Tests run now.

added python binding for cuda nd array dimshuffle
added test of the above
used the above to cover all four tranpose cases of test_cuda_ndarray.py:test_dot

fixed an issue where test_reshape in test_cuda_ndarray did not use the correct seed for random number generation

These tests were previously computing wrong values.

condition.

one that works on scalars.

previously introduced function for merging two slices.