Fix cuda test on 32 bit python.

Softmax optimization

Make use of the verbosity parameter in numpy system_info.

is avaiable not just in the inner of the loop for the Cross entropy
fct. This also remove duplicate code.

bugfix, for a case that don't happen right now.

The preallocated output size for make_vector should ALWAYS be the good
shape right now as it only depend on the ndim and that is constant in
the graph and we only give preallocated output of the same node.

This fix will be useful in the futur if we make other type of
preallocated output that could be of the wrong size.

[BUG] in fast compile related to inplace opt.

Add subgraph_grad

Bug and crash fix in fast_compile

This fix bugs on the CPU in fast_compile introduced in:
commit 4f8f0da5
Date:   Thu Feb 27 09:56:12 2014 -0500

Add tile in the doc.

Update the c_code_cache_version that was forgotten.

Mixed