2. Modify the output of the old order (add max_node_memory_saved_by_view & max_node_memory_saved_by_inplace)

(last time Travis occurs an error about this, fix now)

2. add count_running_memory method to take places of 2 loops in ../compile/profiling.py
3. add new_order's memory value in Store the max stats

2. Create the node_cleared_order in ../gof/vm.py
3. Add loops for new orders in .. /compile/vm.py

added ptp() and associated case test

[ENH] Hermitian generalized eigenvalues and gradient

Added no axis test for matrices.
Test results are now compared against numpy result

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