- fixing an issue introduced with CUDA-TK 7.0, when automatic selection
  of free gpu by os / library no longer works. Library always selects '0',
  which leads to crash in case of 'exclusive mode'.
- this fix is inspired by Dan Povey's fix for Kaldi:
  https://github.com/kaldi-asr/kaldi/commit/6548565445167e00125848f91d7da5f3f949b2a2
- it does a loop over gpus until a free gpu is taken.

Implement GpuAdvancedSubtensor1 for gpuarray

elemwise opt

Scan inplace opt

Use shape_of instead of explicitly inserting Shape_i

d3viz: Interactive visualization of Theano compute graphs

Add a 'name' argument to '.copy' method of Theano variables

These Shape_i nodes will not always be optimized away otherwise, and
they prevent local_shape_to_shape_i from being applied.
Instead, we manually apply the equivalent of local_shape_to_shape_i,
by using shape_of[...] directly.

Whitespace needed in theano_vs_c.txt

Fixed the docstring for the scan_opt.

Fix a non-ascii character in a comment in test_flake8.py.

Add some minimal tests for cudnn softmax.

Enable merging of GpuDnnPoolDesc

fix gh-3377 bug introduced by 41a8e89b merged by 7320e1b1. GpuElem…

fix gh-3377 bug introduced by 41a8e89b merged by 7320e1b1. GpuElemwise with non contiguous inputs give bad results

Change default value of reoptimize_unpickled_function

Remove the assert in PyGpuArray_DEV_DATA since it might cost a bit too much to verify that condition everytime.

Should fix bugs in current uses and makes porting easier too.

Update the subtensor test to have varying values in order to really check that it writes the right thing at the right spot.