Note: the setup develop call is required IFF there is a change
to the set of command-line (or other) entry points for the package,
but it is not required for *most* changes.

abergeron requests removal of it, so I'm removing it.

Instead of putting the "if you work harder you can get better performance"
note before the code, make the code the first thing you see and only
note that you could do it another way in the performance testing section.

Also switches to code-blocks throughout.

Adds note that the openblas thread limit note seems to be outdated,
the section above notes that we should be linked to multithreaded
blas already, and testing seems to suggest we are.

Includes link in case users aren't familiar with the tool.

fix test in FAST_COMPILE

Crash fix and change scan.allow_gc default.

Small crash fix when cxx=icpc

Fix tests that cause error in the daily buildbot.

Change the ModuleCache to only take the lock for actual compilation.

Fix tests

Added missing import for ShapeError exception.

disable MRG_RandomStreams warning that it can't guess the number of stream. There is no explaination how to do this and confuse people.

AdvancedIncSubtensor1.c_code

Parameterized local_elemwise_alloc_opt for GPU support

Cudnn2

Conflicts:
	theano/sandbox/cuda/tests/test_dnn.py

Conflicts:
	theano/sandbox/cuda/dnn.py

Fixed #2071: Moving test_record.py from Pylearn2 to Theano.

timing information added in debugprint