Software Synthesis for DSP Using Ptolemy
In this section we will review how the various modules of the Ptolemy platform interact to generate code for a target application. The code generation procedure is detailed in figure 14. First, the setup() method is called for all blocks relevant to particular application. This allows the schedulers, target modules, and stars to initialize internal variables. Next, the schedule pass is done. The scheduler returns a list that details the firing order of the blocks in a particular application. Based on this schedule, the resources can be allocated. In the case of assembly code, the memory is allocated as well. Note, the resource allocation stage must follow the scheduling stage so that the buffer lengths are known. Now we are ready to generate the initialization code for the given application. At this point, the initCode() method of all the blocks are fired. Finally, we are ready to generate the main loop code.
First we initialize the main loop. Notice that the code generation algorithm forks into two different paths, one signifying that the code currently being generated is intended for a target on the inside of a wormhole, and the other for applications not running inside a wormhole. If we are inside of a wormhole, we generate code to read the data from the Ptolemy Universal construct. Then we generate the main loop code and finally generate code to write the data into the Ptolemy construct. The wormhole code is written is a way which automatically synchronizes the DSP system and the host workstation. If we are not inside a wormhole, we simply generate the main loop code. Finally, we close the main loop and then fire the wrapup() methods of all of the blocks relevant to a particular application.