making the subtensor merge optimization work

Fix for CUDA 4.1 on OS X, prevent -no_pie flag

Create the $TMPDIR directory if needed.

Always introduce Alloc node when calling alloc()

Otherwise, nvopencc complains.

Also use correct mode in test.

Minor stuff (PEP8 and typos mostly)

Gpu setsubtensor

That way, DebugMode can catch shape errors.

Do not wrap in as_tensor_variable unless needed

Do not test div on denominator too close to 0.

The numeric estimation of the gradient was sometimes quite off.

Remove new bugged version of GpuSum commited in a pep8 commits.

Crash fix gpu reshape

Blas

Remove AddSSData Optimization

Fix elusive output-corruption bug in Scan.

The problem was that when copying a part of an array onto an overlapping
part of the same array, sometimes elements were overwritten before being
copied, leading to silent data corruption.

Import 'any' for python 2.4

Fixed crash when cuda is not available