* __init__ updated

* pull request updated

* build_module added

* typo fixed

* another typo fixed

* conv2d gpu scheduler for two layouts moved to tvm

* changes made according to CR

* conv2d_nchw formating updated, conv2d_hwcn tests updated

* lint error fixed

* element wise operator schedule fusing fixed for conv2d

* conv2d_nchw topi test added, all resnet workloads now pass

* conv compute lint error fixed

* fixed python 3 compatibility problem

* conv2d tensor input support added, test typo fixed, ir_pass.Simplify changed to util.get_const_int

* fixed channel numer < 4 error, also made sure other splitting factor woudn't be 0

* [TOPI] Add ops compute

Remove 'compute' and add assert for safety

Add document

fix lint

fix softmax

* fix batch norm

* [TOPI] Move ewise.h -> elemwise.h

* fix test

[TOPI] Add broadcast and reduce operators

* add dilation operators

* fix pylint

* dilate testcases success

* n-D tensor dilation

* support arbitrary dimension

* [DOC] Include TOPI in doxygen

* update

* __init__ updated

* pull request updated

* build_module added

* typo fixed

* another typo fixed

* conv2d gpu scheduler for two layouts moved to tvm

* changes made according to CR

* conv2d_nchw formating updated, conv2d_hwcn tests updated

* lint error fixed

* element wise operator schedule fusing fixed for conv2d

* conv2d_nchw topi test added, all resnet workloads now pass

* conv compute lint error fixed

* fixed python 3 compatibility problem

* conv2d tensor input support added, test typo fixed, ir_pass.Simplify changed to util.get_const_int

* [TOPI] Move topi.nn.util to topi.util

* update the path

* [WIP] C++ topi contributions

Summary:
This diff implements C++ topi contributions for:
  - relu with parametrix threshold
  - pad with generic padBefore / padAfter specification
  - matmult with transposes
  - conv2d_nchw, conv2d_hwcn with runtime constant padding and strides
  - depthwise_conv2d_nchw with runtime constant padding and strides
  - group_conv2d_ngchw with runtime constant padding and strides
  - broadcast_to a broadcastable shape
  - broadcast_bop where bop is an usual binary op (+ - * / %)

Convolution padding is implemented using the pad operation.
To avoid extra memory consumption, it is generally recommended to inline the padding with the autoinliner.
Unfortunately in its current form the elemwise checks are too restrictive to allow inlining.
So this diff also proposes an extension to LHS injective (i.e. no reduction axis in the current IR design)

Test Plan:
Tested in C++ testsuite in a separate repository, I am looking for suggestions to quickly spin up some tests for tvm.

Reviewers: tqchen

Subscribers:

Tasks:

Tags:

Blame Revision:

* Review + Lint + GSG C++

* [PASS][FIX] Fix LiftAttrScope with if

* [PASS] Fix on proc sync

* fix

* [TEST] Upgrade gpu docker to cudnn7

* fx

* [tvm4j] RPC Server

* [tvm4j] fix recursively function calling; connect to proxy server; osx rename .so to .dylib

* [tvm4j] test case for proxy connection; thread pool for serving

* [RUNTIME][PASS] Allow declare vector type array

* fix bcast

* [BUFFER] Enable vload/store function in buffer

* ok

* [NNPACK] Add nnpack.convolution

* Add instrinsic

* Fix lint

* [EXECUTOR] Split graph_executor to header file and (runtime) source file

* Fix

* [PASS] Enhance LayoutTransform pass

* Fix

* Fix Compilation

* Refactor

* Refactor

* doc

* fix

* add file

* [PASS] Refactor thread storage sync to a common visitor

* Fix the sync scope check behavior

* Typofix.

Signed-off-by: Edward Z. Yang <ezyang@fb.com>

* Probe for nvrtc in lib directory as well.

Signed-off-by: Edward Z. Yang <ezyang@fb.com>

* Conda build recipe for TVM.

Signed-off-by: Edward Z. Yang <ezyang@fb.com>

* [PASS] Improve graph fusion

* Change fusion center to segment head

* Use 'master' to identity the schedule node

* Make things compact

* Fix