* change debug enable from cmake flag to environment var so it can be changed without rebuild

* Doc for abs, update doxygen for ops with .rst doc

* Review

* WIP with test ops

* update WIP

* Add LLVM compilation detail from INSTALL to installation guide and cleanup todo comments on convolution.rst and remove deprecated file

* Clarify LLVM compilation flag on Ubuntu and delete WIP draft from earlier

fixed isnan issue on centos 7.2

MKLDNN Avg Pool Backprop kernel

* GPU kernels for reshape, GEMM, EW ADD/Mult, Maximum

(A + B) * C test now with cuBLAS
Additional gemm and gemv calls
cmake updates for cuDNN calls
memcpy wrappers in gpu_util

Additional passing tests:
aliased outputs, parameter, constant tensor memcopy

* compare derivatives from bprop and bprop with fprop cache

* code format

Fix pep8 warning in copyright

added CMAKE_OPTIONS_EXTRA

CPU: Kill warnings

* deserialize fix to return the right outermost function

* fix tests

* more switches to print out/visualize shapes

* revert to shared argon lib

* installs argon headers

* -DINSTALLED_HEADERS_PATH=${CMAKE_INSTALL_PREFIX}/include

* formatg

* remove debug info

* revert

* revert back to cpp-master

* Main chunk of style description

* Edit code guide for readability as raw rst. Update subheadings with clearer delineation

* Minor edits

* Typo

* Fix warnings in zero-sized tests; fix zero-sized tests return vector type for comparison ops

* Revert changes to cpu_emitter

* Enable CPU backprop tests

* Fix to dot codegen for cases where n_reduction_axes != 1

Dynamically create and destroy mkldnn engine objects to work around f…