[NNRT-1111] add memory layout for doc

docs/Operators.md.template

@@ -1,3 +1,23 @@
+# Memory Layout
+
+There are two different memory layout mode:
+
+- Row-Major. Also called C contiguous, which is firstly introduced in C by AT&T Bell Lab(1960s). Caffe/TensorFlow/Pytorch follow the Row-Major layout mode.
+
+- Column-Major. Also called Fortran contiguous, which is firstly introduced in Fortran by IBM(1958). Matlab/OpenGL/OpenCL/OpenVX follow the Column-Major layout mode.
+
+See also :
+
+http://eigen.tuxfamily.org/dox/group__TopicStorageOrders.html
+
+On the other hand, a memory layout can be described by both Row-Major and Column-Major mode.
+
+It easily tranlate the layout mode from one to another: just reverse the dimesions of the tensor.
+
+For example, TensorFlow usually describe the tensor as NHWC format. This is equal the CWHN format in OpenVX.
+
+Likewise, the WHCN format in OpenVX is equal the NCHW format in Caffe.
+
 # Operators
 
 {DOCS}

include/tim/vx/ops/conv2d.h

@@ -37,11 +37,17 @@ namespace ops {
  *
  * Performs a 2-D convolution operation, include classic Conv2D /
  * Depthwise Conv2D / Group Conv2D / Dilation Conv2D.
+ * 
+ * Input:
+ * - input [WHCN or CWHN].
+ * - kernel [ WHIcOc ] (Ic: Input Channels. Oc: Output Channels).
+ * - bias [ O ]. Optional.
  *
+ * Attribute:
  * - weights : the output channel number for weight tensor.
  * - ksize : the height and width for weight tensor.
  * - padding : AUTO, VALID or SAME.
- * - pad : pad value for each spatial axis. 
+ * - pad : pad value for each spatial axis.
  * - stride : stride along each spatial axis.
  * - dilation : dilation value along each spatial axis of the filter.
  * - multiplier: function similar to group attribute on other framework,