[NNVM] Fix softmax gradient

* rename FInferLayout -> FCorrectLayout

* correct stupid IDE

* update submodule tvm

* initial NHWC layout support

* remove layout param from softmax

* more nhwc support

* fix typo

* add nhwc layout test

* fix lint

* update tvm

* update for c++ topi

* fix lint

* update tvm

* Updated TVM version. Implemented fix for nnvm_compiler crash on exit on windows. Changed TOPI ops from using python to using C++ where applicable.

* Fix lint

* Fix lint

* Fix macro

* Fix reshape

* Update TVM to fix test fails

* Add conv2d max_pool backward op

* Added tests

* Fix testing

* Address comments

* Change dot to matmul

* Address comments

* Break down indicator function

* Make greater, less numpy compatible

* make shape inference of BatchNorm layout neutral

* refactor to use the axis variable to do BatchNorm shape inference

* refactor to use the axis variable to do BatchNorm shape inference

* add unittest to the axis param for batch norm shape inference

* keras frontend tutorial

* fix

* vgg16 success

* remove six.PY2, use sys.version_info;
convert_activation() accepts activation type name(str, e.g. 'relu') as input;

* add convert_merge

* fix convert_batchnorm;
improve tests

* fix lint

* add numpy-style pad operator

* deal with asymmetry padding

* resnet50 success

* fix pool_convert; xception passes test

* update tvm

* fix bias error; all tests pass

* use > >, not >>

* [TOP] level4 except argmax/min, correct split

* [DOCS] Add doc generator for top

* [TOP] complete level2

* [TOP] add split

* improve infer shape/type error message

* fix dense infer shape