Change some texts to prepare theano 1.0.

Refactor dnn conv functions

Cite the papers about MRG31k3p in the code. Not just in the acknowled…

Add a Python script to help run more exhaustive cuDNN algorithms tests

Prepare release 0.10.0beta4

Don't initialize the GPU in MPI subprocesses.

Implement truncated normal with box-muller

It isn't duplicated

This reverts commit 949d78c6.

Add good user error in corner case that are hard to debug.

Minimum maximum gradient fix

Fix MRG with 0 output size. This fix this test: THEANO_FLAGS=device=cuda,floatX=float32 theano-nose theano/sandbox/tests/test_rng_mrg.py:test_normal_truncation

Allow Gpu multinomial to have different input dtype for pvals and unis as the CPU code. This fix a test crash with: THEANO_FLAGS=device=cuda,floatX=float32 theano-nose theano/sandbox/tests/test_rng_mrg.py:test_multinomial

Change the build image to avoid maintaining our own.