The Faust compiler relies on several intermediate representations to translate a Faust program. One step, in particular, consists of moving from a functional representation of computations on infinite signals to an imperative (stateful) representation of computations on samples. This translation phase is complex, as it combines the recursive tree traversal, the division of the computations into instructions, the scheduling, and the code generation. As a result, the implementation of new code generation strategies is difficult to achieve.

In this paper, we propose a new intermediate representation in the form of a graph whose nodes represent computations on infinite signals and the edges time dependencies between these computations. The graph structure makes it much easier to handle recursive dependencies as well as to experiment with all kinds of scheduling strategies. We will present several of them whose performance have been tested with the examples of the Faust distribution. Performance gains can sometimes be quite significant compared to the current compiler.