* [VTA][Chisel] End-to-end Inference with Chisel VTA

* Update TensorAlu.scala

* [Relay] add max_pool3d in relay and TF converter

* fix comments

* [TEST] Remove nnvm related code in topi and test script

* Remove docs dep

dtype.h -> runtime/data_type.h

Changes:
- Rename all old reference of tvm::Type to DataType
- ExprNode.type -> ExprNode.dtype
- Expr.type() -> Expr.dtype()
- Change Expr related functions to expr_operator.
  - DataType::min() -> min_value(DataType)
  - DataType::max() -> max_value(DataType)
- Move type constructor Int, UInt, Float, Handle, Bool into DataType.
  - Int(bits) -> DataType::Int(bits)
  - UInt(bits) -> DataType::UInt(bits)

* [VTA] improved virtual memory mapping

* Update virtual_memory.cc

* [PYTHON][FFI] Cythonize NDArray.copyto

* Cythonize the shape property

* [DOCS] Mention Ninja build system in install/from_source.rst

* Address comments

One of the loops in hybrid_nms used for
performing the max_output_size reordering
was incorrectly designated as parallel
resulting in incorrect behaviour. This patch
changes that loop to a serial loop.

Change-Id: I97184f5887f5f028d8ab339fa2808eb7630a4017

* Add Expand to onnx.py

* add test function for expand

* Fix a onnx frontend test

* Add tests for the value itself instead of shape only on test_expand

* Cleaned up some unnecessary modifications.

* Update legacy places from nnvm to relay.

This PR prepares the current mainline to remove nnvm compiler dep.

* remove legacy stage

Change-Id: If2075df5475505f2da87dae7145af5a7ab83d8a4

* Use the best tuner possible

* Add comment denoting availability of better tuners

* Fix typos and wording

* Replace sigmoid() with tanh() in tests for TANH

* Fix bias_add gradient

A change caused collapse_sum_like to reject implicit dimension
broadcasting for bias_add gradient, so switch to explicit sum reduction
on the non-bias axis dimensions.

* Lint fix

* fix mul rewrite

* register Realize Rewrite for global avg pool and add test

* remove unnecessary check

* improve the test case

* [Fix][Hybrid] Fix hybrid script to support array of tensors

* add test case

* clean up

* trigger ci

* [TOPI] implement pool3d op

* use PoolInferCorrectLayout for both 2d and 3d pooling

* unify MakeMaxPool and MakeAvgPool

* fix serialization precision loss in float

When we want to serialize a tvm.tensor object(like pickle)， we will get a precision loss cause by std::to_string()。
For example, a2.value will be 0.0 while a.value=0.00000001 in the following: 
    import tvm
    import pickle
    a = tvm.const(0.00000001, 'float32')
    a2 = pickle.loads(pickle.dumps(a))

* remove line end spaces

This imports the CUDA schedules to rocm.