DIVA: A Declarative and Reactive Language for in situ Visualization
Qi Wu, Tyson Neuroth, Oleg Igouchkine, Konduri Aditya, Jacqueline H. Chen, and Kwan-Liu Ma
IEEE Symposium on Large Data Analysis and Visualization (LDAV), 2020
Figure 1: A DIVA program is processed through three layers. Users typically specify their program using the declarative interface (left); then the language parser will translate it into an internal DAG representation; this representation will then be interpreted into a low-level dataflow API for execution. A) A DIVA program computes a volume rendering for every 5 timesteps, and saves the rendering on disk. B) The same program in the DAG representation. C) The same program in the low-level API. Because the C++ API is not declarative, in part C), statements have to be executed in order. Moreover, because C++ does not track data dependencies automatically, all variables declared in C) should be wrapped by lifting operators (e.g., divaCreateSource). D) The hierarchy of primitives defined by the low-level dataflow API: All values in DIVA are signals; values depending on external inputs are sources; values returning to the environment are actions (e.g., a saved image file); triggers are special primitives that decide which actions to compute based on predicates; rest values are internal to the workflow and are represented by either pure (i.e., DivaPureOp) or impure functions (i.e., DivaImpure).
