Add missing documentation for fused kernels (#744)

Author: xadupre

operators/cuda/negxplus1.h

@@ -8,6 +8,9 @@
 
 namespace contrib {
 
+/**
+* NegXPlus1(X) = 1 - X
+*/
 template <typename T>
 struct NegXPlus1 {
   template <typename TDict>

operators/cuda/scatter_nd_of_shape.h

@@ -8,6 +8,9 @@
 
 namespace contrib {
 
+/**
+* ScatterNDOfShape(shape, indices, updates) = ScatterND(ConstantOfShape(shape, value=0), indices, updates)
+*/
 template <typename T>
 struct ScatterNDOfShape {
   OrtStatusPtr OnModelAttach(const OrtApi& api, const OrtKernelInfo& info) {
@@ -71,6 +74,12 @@ struct ScatterNDOfShape {
 };
 
 
+/**
+* MaskedScatterNDOfShape(shape, indices, updates) = ScatterND(ConstantOfShape(shape, value=0),
+*                                                             indices[indices != maskedValue],
+*                                                             updates[indices != maskedValue])
+*
+*/
 template <typename T>
 struct MaskedScatterNDOfShape {
   OrtStatusPtr OnModelAttach(const OrtApi& api, const OrtKernelInfo& info) {

operators/cuda/transpose_cast.h

@@ -8,6 +8,9 @@
 
 namespace contrib {
 
+/**
+*  Transpose2DCast(X, to=to) = Cast(Transpose(X, perm=[1, 0]), to=to)
+*/
 template <typename TIN, typename TOUT>
 struct Transpose2DCast {
   template <typename TDict>