[Doc] Add MKL-DNN operator list

docs/install/index.md

@@ -187,7 +187,12 @@ $ pip install mxnet --pre
 
 </div> <!-- End of master-->
 <hr> <!-- pip footer -->
-MXNet offers MKL pip packages that will be much faster when running on Intel hardware.
+MXNet offers MKL pip packages that will be much faster when running on Intel hardware. Try the following command line to install it and find performance numbers and tuning guide in [performance on Intel CPU](https://mxnet.incubator.apache.org/versions/master/faq/perf.html#intel-cpu).
+
+```
+$ pip install mxnet-mkl --pre
+```
+
 Check the chart below for other options, refer to <a href="https://pypi.org/project/mxnet/">PyPI for other MXNet pip packages</a>, or <a href="validate_mxnet.html">validate your MXNet installation</a>.
 
 <img src="https://raw.githubusercontent.com/dmlc/web-data/master/mxnet/install/pip-packages-1.4.0.png" alt="pip packages"/>

docs/tutorials/mkldnn/MKLDNN_README.md

@@ -19,7 +19,9 @@
 
 A better training and inference performance is expected to be achieved on Intel-Architecture CPUs with MXNet built with [Intel MKL-DNN](https://github.com/intel/mkl-dnn) on multiple operating system, including Linux, Windows and MacOS.
 In the following sections, you will find build instructions for MXNet with Intel MKL-DNN on Linux, MacOS and Windows.
-
+
+Please find MKL-DNN optimized operators and other features in [MKL-DNN operator list](http://mxnet.incubator.apache.org/tutorials/mkldnn/operator_list.html)
+
 The detailed performance data collected on Intel Xeon CPU with MXNet built with Intel MKL-DNN can be found [here](https://mxnet.incubator.apache.org/faq/perf.html#intel-cpu).
 
 

docs/tutorials/mkldnn/operator_list.md

@@ -0,0 +1,83 @@
+<!--- Licensed to the Apache Software Foundation (ASF) under one -->
+<!--- or more contributor license agreements.  See the NOTICE file -->
+<!--- distributed with this work for additional information -->
+<!--- regarding copyright ownership.  The ASF licenses this file -->
+<!--- to you under the Apache License, Version 2.0 (the -->
+<!--- "License"); you may not use this file except in compliance -->
+<!--- with the License.  You may obtain a copy of the License at -->
+
+<!---   http://www.apache.org/licenses/LICENSE-2.0 -->
+
+<!--- Unless required by applicable law or agreed to in writing, -->
+<!--- software distributed under the License is distributed on an -->
+<!--- "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY -->
+<!--- KIND, either express or implied.  See the License for the -->
+<!--- specific language governing permissions and limitations -->
+<!--- under the License. -->
+
+# MKL-DNN Operator list
+
+MXNet MKL-DNN backend provides optimized implementations for various opertors covering a broad range of applications including image classification, object detection, natural language processing. We also provide the lower precision version for part of these operators on CPU leveraging the DL Boost technology from Intel. On computation graph level, a set of graph fusion pass and quantization pass is implemneted based on the sugraph feature of MXNet. To help users understanding MKL-DNN backend better, the tables below summarize the list of supported operators, data types and functionalities. As the community keeps working on more new features for MKL-DNN backend, the tables will be updated continuously.
+
+| Operator           | Function                   | FP32 Training (backward) | FP32 Inference | INT8 Inference |
+| :--:               | :--:                       | :--:                     | :--:           | :--:           |
+| **Convolution**    | 1D Convolution             | Y                        | Y              | N              |
+|                    | 2D Convolution             | Y                        | Y              | Y              |
+|                    | 3D Convolution             | Y                        | Y              | N              |
+| **Deconvolution**  | 2D Deconvolution           | Y                        | Y              | N              |
+|                    | 3D Deconvolution           | Y                        | Y              | N              |
+| **FullyConnected** | 1D-4D input, flatten=True  | N                        | Y              | Y              |
+|                    | 1D-4D input, flatten=False | N                        | Y              | Y              |
+| **Pooling**        | 2D max Pooling             | Y                        | Y              | Y              |
+|                    | 2D avg pooling             | Y                        | Y              | Y              |
+| **BatchNorm**      | 2D BatchNorm               | Y                        | Y              | N              |
+| **LRN**            | 2D LRN                     | Y                        | Y              | N              |
+| **Activation**     | ReLU                       | Y                        | Y              | Y              |
+|                    | Tanh                       | Y                        | Y              | N              |
+|                    | SoftReLU                   | Y                        | Y              | N              |
+|                    | Sigmoid                    | Y                        | Y              | N              |
+| **softmax**        | 1D-4D input                | Y                        | Y              | N              |
+| **Softmax_output** | 1D-4D input                | N                        | Y              | N              |
+| **Transpose**      | 1D-4D input                | N                        | Y              | N              |
+| **elemwise_add**   | 1D-4D input                | Y                        | Y              | Y              |
+| **Concat**         | 1D-4D input                | Y                        | Y              | Y              |
+| **slice**          | 1D-4D input                | N                        | Y              | N              |
+| **Quantization**   | 1D-4D input                | N                        | N              | Y              |
+| **Dequantization** | 1D-4D input                | N                        | N              | Y              |
+| **Requantization** | 1D-4D input                | N                        | N              | Y              |
+
+Besides direct operator optimizations, we also provide graph fusion passes listed in the table below. Users can choose to enable or disable these fusion patterns through environmental variables.
+
+For example, you can enable all fusion passes by:
+
+```
+export MXNET_SUBGRAPH_BACKEND=MKLDNN
+```
+
+And disable `Convolution + Activation(ReLU)` fusion by:
+
+```
+export MXNET_DISABLE_MKLDNN_FUSE_CONV_RELU=1
+```
+
+| Fusion pattern                                            | Enable                  | Disable                             |
+| :--:                                                      | :--:                    | :--:                                |
+| Convolution + Activation(ReLU)                            | MXNET_SUBGRAPH_BACKEND  | MXNET_DISABLE_MKLDNN_FUSE_CONV_RELU |
+| Convolution + elemwise_add                                | MXNET_SUBGRAPH_BACKEND  | MXNET_DISABLE_MKLDNN_FUSE_CONV_SUM  |
+| Convolution + BatchNorm                                   | MXNET_SUBGRAPH_BACKEND  | MXNET_DISABLE_MKLDNN_FUSE_CONV_BN   |
+| Convolution + Activation(ReLu) + elemwise_add             | MXNET_SUBGRAPH_BACKEND  |                                     |
+| Convolution + BatchNorm + Activation(ReLu) + elemwise_add | MXNET_SUBGRAPH_BACKEND  |                                     |
+| FullyConnected + Activation(ReLU)                         | MXNET_SUBGRAPH_BACKEND  | MXNET_DISABLE_MKLDNN_FUSE_FC_RELU   |
+| Convolution (INT8) + re-quantization                      | MXNET_SUBGRAPH_BACKEND  |                                     |
+| FullyConnected (INT8) + re-quantization                   | MXNET_SUBGRAPH_BACKEND  |                                     |
+
+
+To try these features out, you can install MXNet MKL-DNN backend through pip:
+
+```
+pip install mxnet-mkl [--pre]
+```
+
+To build MXNet MKL-DNN backend from source code, please refer to [MKL-DNN backend readme](http://mxnet.incubator.apache.org/tutorials/mkldnn/MKLDNN_README.html)
+
+For performance numbers, please refer to [performance on Intel CPU](https://mxnet.incubator.apache.org/versions/master/faq/perf.html#intel-cpu)

tests/tutorials/test_sanity_tutorials.py

@@ -35,6 +35,7 @@
              'gluon/index.md',
              'mkldnn/index.md',
              'mkldnn/MKLDNN_README.md',
+             'mkldnn/operator_list.md',
              'nlp/index.md',
              'onnx/index.md',
              'python/index.md',