In decision tree predict, each sample read \(i\) layer to select a branch to \(i+1\) layer. The max loop number of each sample is \(MAX_DEPTH\) Based on the algorithm, the description of decision tree predict if provided by the following figure.
To eliminate the loop-carried dependency for memory port deficiency, we partition different layers of decision tree in differen URAM. In batch mode predict, each sample reads various uram, each round has no dependence so that the initiation interval (II) could achieve to 1. It is present as follows:
