Modeling static-order schedules in synchronous dataflow graphs

Abstract

Abstract-Synchronous dataflow graphs (SDFGs) are used extensively to model streaming applications. An SDFG can be extended with scheduling decisions, allowing SDFG analysis to obtain properties like throughput or buffer sizes for the scheduled graphs. Analysis times depend strongly on the size of the SDFG. SDFGs can be statically scheduled using static-order schedules. The only generally applicable technique to model a staticorder schedule in an SDFG is to convert it to a homogeneous SDFG (HSDFG). This conversion may lead to an exponential increase in the size of the graph and to sub-optimal analysis results (e.g., for buffer sizes in multi-processors). We present a technique to model periodic static-order schedules directly in an SDFG. Experiments show that our technique produces more compact graphs compared to the technique that relies on a conversion to an HSDFG. This results in reduced analysis times for performance properties and tighter resource requirements.

Modeling pipelined application with Synchronous Data Flow graphs

“…In this section some notations about SDF graphs which will be used in Section IV are introduced, then the Decision State Modeling technique [2], which is extended in the first part of the proposed methodology, is briefly described.…”

“…Some techniques have been proposed to overcome these expressiveness limits by means of new types of data flow graphs or new programming languages. Damavandpeyma et al [2] proposed to enrich a SDF graph with new actors and channels to include the scheduling solution directly in the graph. The added actors force the serialization of activation of actors which can run in parallel but that have been assigned to the same processing element.…”

“…Decision State Modeling (DSM) [2] has been proposed to include PSOSs directly in SDF graphs. Note that all the activations of an actor must be part of the same PSOS.…”

“…With respect to the Decision State Modeling presented in [2], the methodology does not model only the partial scheduling of the application, but also the mapping and the pipeline stage assignment. The modeling is obtained by enriching the initial SDF graph with new actors and channels representing the particular design solution.…”

“…For this reason, several solutions to formally describe this information by enriching SDF graphs have been proposed. These solutions however include only part of the missing information (e.g., [2]) or describe it by means of new representations which have to be coupled with application SDF graphs (e.g., [3], [4]). While these are a well known representation used in several methodologies and tools, none of the representations proposed to integrate them has had the same success and has been commonly accepted, potentially preventing integration of different analyses and design algorithms in the same design flow.…”