Gpua rng opt

fixed a bug in tensor/opt.py exception

While there fix the inplace optimization so that it actually works with the GPU versions.

Fix tests with pycuda example that now fail in the buildbot.

Make the DotModule c_code being cached.

Kernel precomp

Fix the last remaining problems with the existing clients and make the kernel code for MRG more OpenCL-friendly.

Specify shape and print op to the new gpu back-end

Print current device when initializing gpuarray.

Fix tests crash for gpuarray back-end. Don't do constant folding with th...

Gpureduce: support multiple dtype, prod, max and min

Fix doc generation.

Renamed "copy_inputs" into "share_inputs" in clone

GpuArray join/split